Jay Hartzell, president of the University of Texas at Austin, explains the handling of pro-Palestinian protests — including why Texas DPS was called.

UT-Austin president: Protesters put campus safety at risk

Gov. Greg Abbott has issued a disaster declaration for dozens of counties in Texas on Thursday as severe weather and flooding threaten the state.

Houston area facing 'life-threatening' flood conditions as severe weather pummels Texas - ABC News

The Metropolitan Police has launched a murder probe in south London after the death of the 26-year-old died of stab wounds.Separate stabbing incidents » · Conservative leadership hopeful Boris Johnson has defended his infamous description of black people in Africa bearing “watermelon smiles” ...

Heavy Rain Triggers Flooding in Southeast Texas

According to a county spokesperson, toxicology reports suggest that the nine people who died in Austin, Texas, this week had ingested fentanyl.

Officials say opioid 'outbreak' in Austin, Texas, linked to 9 deaths and 75 overdoses

Neighbors in the Las Vegas Trail neighborhood said they&#x27;re shaken up and worried about everyone&#x27;s safety following the shooting Wednesday night.

Outrage after drive-by shooting in North Texas injures 4 children - CBS Texas

An outbreak of opioid overdoses in Austin, Texas, led to at least eight deaths and over 50 emergency calls. The state experienced a 1000% spike in overdose 911 calls likely caused by drugs laced with fentanyl.

Austin, Texas sees 1000% spike in overdose 911 calls likely caused by fentanyl

For over a week, relentless severe weather has been torturing states across America. That trend will continue Thursday, with the bulk of the central U.S. facing another day of dangerous storms.

Relentless severe storms continue to torture areas from Texas to Upper Midwest | Fox Weather

Mandatory evacuations were ordered in parts of Texas, and residents in Harris County, home to Houston, were told to be ready to stay put for days after heavy rain caused flooding.

Mandatory evacuations ordered in Texas after heavy rain and floods

The superintendent’s resignation follows an investigation into the student’s removal, which sparked nationwide backlash.

Texas superintendent resigns after trans student’s removal from ‘Oklahoma!’

The group set up an encampment Thursday outside the library at the University of Texas at Arlington.

Group of pro-Palestinian protestors set up informal encampment at the University of Texas-Arlington

TextUs

textus

Table of Links Abstract and Introduction


Domain and Task
2.1. Data sources and complexity
2.2. Task definition


Related Work
3.1. Text mining and NLP research overview
3.2. Text mining and NLP in industry use
3.3. Text mining and NLP for procurement
3.4. Conclusion from literature review


Proposed Methodology
4.1. Domain knowledge
4.2. Content extraction
4.3. Lot zoning
4.4. Lot item detection
4.5. Lot parsing
4.6. XML parsing, data joining, and risk indices development


Experiment and Demonstration
5.1. Component evaluation
5.2. System demonstration


Discussion
6.1. The ‘industry’ focus of the project
6.2. Data heterogeneity, multilingual and multi-task nature
6.3. The dilemma of algorithmic choices
6.4. The cost of training data


Conclusion, Acknowledgements, and References 4.3. Lot zoning The text content extracted above may or may not contain relevant information for lots. Therefore, this component deals with the filtering of the text content to identify the relevant content areas for further processing (i.e,. ‘zoning’ to the right content). We focus on two tasks: 1) selecting the relevant pages; 2) selecting the relevant tables. We define ‘relevance’ as containing a description of a lot (lot reference) and/or its items (lot items). And for both, we cast them as a text classification task. For example, in Figure 2, a valid ‘lot reference’ is ‘Lot Number 1’, and it contains two items (row 2 and 3); in Figure 6, the first bullet point in Section 2.1 has a lot reference ‘Lot 1’, with the following text being its only item. 4.3.1. Page selection Given a page 𝑝, we define 𝑠 ∈ 𝑝 as a sentence from 𝑝, and 𝐵𝑂𝑊 as a function that can be applied to 𝑠 or 𝑝 to convert it into a simple BOW representation applying stopwords removal and lowercasing (lightweight NLP that is language dependent). We create several language-independent features for learning a relevant page classifier such that it can be used for content in any language. For each word 𝑤 in the domain specificity lexicon 𝑤 ∈ 𝐿, we search the word in 𝐵𝑂𝑊(𝑝) and count its frequency as 𝑐𝑜𝑢𝑛𝑡(𝑤, 𝑝). Then for each of the four metrics introduced above, we create four features (so a total of 16 features): the sum, average, minimum and maximum of the scores, as follows, where ‘metric’ generalises the four metrics explained before (ntf, ndf, ntf-ndf, weirdness). Further, for each of the two domain dictionaries (form and measure), we create four features (so a total of 8 features) using the same equations below, by replacing 𝐿 with a dictionary 𝐷, and setting 𝑚𝑒𝑡𝑟𝑖𝑐(𝑤) = 1 always. Together this gives us a total of 24 ‘page-level’ features. Finally, we calculate the percentage of sentences that contain at least a number (e.g., indicating a volume/quantity or measurement although can be noise sometimes), and evaluate the percentage of sentences that contain at least one matched word from the lexicon/dictionaries and are calculated as follows (where 𝐿 indicates the domain specificity lexicon but can be replaced by either of the two dictionaries): In total, we obtain 52 features for a page. Our feature extraction process is arguably language independent, as it depends on counting word frequencies using lexicons/dictionaries and pre-computed scores assigned to the entries in these lexicons/dictionaries. The language dependent processes are the translation of the domain lexicons/dictionaries, and the lightweight NLP processes for which models for different languages are more available. For businesses, these help create more affordable multilingual NLP solutions than other choices such as translating input documents, or using more language dependent models. We will discuss this further in later sections. In order to learn a classifier that determines if a page is relevant or irrelevant, we asked domain experts to annotate a random collection of pages extracted from the raw TED datasets. For the machine learning algorithms, we compare linear SVM, logistic regression, and Random Forest. 4.3.2. Table selection Similar to page selection, the goal here is to filter tables that are irrelevant. This is common because tables are used in tender notices to describe not only the lots, but oftentimes award criteria, product specifications, and formatting purposes. In this work, we assume that all tables are ‘horizontal’, that is, their columns define attributes of data instances and their rows contain individual data instances. We define 𝑡 as a table and 𝑟 ∈ 𝑡 and 𝑐 ∈ 𝑡 represents rows and columns in the table. If we consider content in a table as a structure-less page, and concatenate each of its rows or columns into a sequence of words equivalent to a sentence potentially describing an instance, then we can apply the same feature extraction methods explained above. In cases of a row or column span, we duplicate the value across all the rows/columns covered by the span. Therefore, we obtain a total of 80 features as follows: ● 24 ‘page’ level features if we treat the table as a simple flat textual page ● 28 ‘sentence’ level features if we treat the table as a simple flat textual page, and its rows as sentences ● 28 ‘sentence’ level features if we treat the table as a simple flat textual page, and its columns as sentences For training, our domain experts annotated a collection of tables extracted from the raw TED dataset (to be detailed later) and we compared the same group of machine learning algorithms. Authors:
(1) Ziqi Zhang*, Information School, the University of Sheffield, Regent Court, Sheffield, UKS1 4DP (Ziqi.Zhang@sheffield.ac.uk);
(2) Tomas Jasaitis, Vamstar Ltd., London (Tomas.Jasaitis@vamstar.io);
(3) Richard Freeman, Vamstar Ltd., London (Richard.Freeman@vamstar.io);
(4) Rowida Alfrjani, Information School, the University of Sheffield, Regent Court, Sheffield, UKS1 4DP (Rowida.Alfrjani@sheffield.ac.uk);
(5) Adam Funk, Information School, the University of Sheffield, Regent Court, Sheffield, UKS1 4DP (Adam.Funk@sheffield.ac.uk). This paper is available on arxiv under CC BY 4.0 license. Table of Links Abstract and Introduction Domain and Task
2.1. Data sources and complexity
2.2. Task definition Related Work
3.1. Text mining and NLP research overview
3.2. Text mining and NLP in industry use
3.3. Text mining and NLP for procurement
3.4. Conclusion from literature review Proposed Methodology
4.1. Domain knowledge
4.2. Content extraction
4.3. Lot zoning
4.4. Lot item detection
4.5. Lot parsing
4.6. XML parsing, data joining, and risk indices development Experiment and Demonstration
5.1. Component evaluation
5.2. System demonstration Discussion
6.1. The ‘industry’ focus of the project
6.2. Data heterogeneity, multilingual and multi-task nature
6.3. The dilemma of algorithmic choices
6.4. The cost of training data Conclusion, Acknowledgements, and References Abstract and Introduction Abstract and Introduction Abstract and Introduction Abstract and Introduction Domain and Task 2.1. Data sources and complexity 2.2. Task definition Domain and Task Domain and Task 2.1. Data sources and complexity 2.1. Data sources and complexity 2.2. Task definition 2.2. Task definition Related Work 3.1. Text mining and NLP research overview 3.2. Text mining and NLP in industry use 3.3. Text mining and NLP for procurement 3.4. Conclusion from literature review Related Work Related Work 3.1. Text mining and NLP research overview 3.1. Text mining and NLP research overview 3.2. Text mining and NLP in industry use 3.2. Text mining and NLP in industry use 3.3. Text mining and NLP for procurement 3.3. Text mining and NLP for procurement 3.4. Conclusion from literature review 3.4. Conclusion from literature review Proposed Methodology 4.1. Domain knowledge 4.2. Content extraction 4.3. Lot zoning 4.4. Lot item detection 4.5. Lot parsing 4.6. XML parsing, data joining, and risk indices development Proposed Methodology Proposed Methodology Proposed Methodology 4.1. Domain knowledge 4.1. Domain knowledge 4.2. Content extraction 4.2. Content extraction 4.3. Lot zoning 4.3. Lot zoning 4.4. Lot item detection 4.4. Lot item detection 4.5. Lot parsing 4.5. Lot parsing 4.6. XML parsing, data joining, and risk indices development 4.6. XML parsing, data joining, and risk indices development Experiment and Demonstration 5.1. Component evaluation 5.2. System demonstration Experiment and Demonstration Experiment and Demonstration 5.1. Component evaluation 5.1. Component evaluation 5.2. System demonstration 5.2. System demonstration Discussion 6.1. The ‘industry’ focus of the project 6.2. Data heterogeneity, multilingual and multi-task nature 6.3. The dilemma of algorithmic choices 6.4. The cost of training data Discussion Discussion 6.1. The ‘industry’ focus of the project 6.1. The ‘industry’ focus of the project 6.2. Data heterogeneity, multilingual and multi-task nature 6.2. Data heterogeneity, multilingual and multi-task nature 6.3. The dilemma of algorithmic choices 6.3. The dilemma of algorithmic choices 6.4. The cost of training data 6.4. The cost of training data Conclusion, Acknowledgements, and References Conclusion, Acknowledgements, and References Conclusion, Acknowledgements, and References Conclusion, Acknowledgements, and References 4.3. Lot zoning The text content extracted above may or may not contain relevant information for lots. Therefore, this component deals with the filtering of the text content to identify the relevant content areas for further processing (i.e,. ‘zoning’ to the right content). We focus on two tasks: 1) selecting the relevant pages; 2) selecting the relevant tables. We define ‘relevance’ as containing a description of a lot (lot reference) and/or its items (lot items). And for both, we cast them as a text classification task. For example, in Figure 2, a valid ‘lot reference’ is ‘Lot Number 1’, and it contains two items (row 2 and 3); in Figure 6, the first bullet point in Section 2.1 has a lot reference ‘Lot 1’, with the following text being its only item. 4.3.1. Page selection 4.3.1. Page selection Given a page 𝑝, we define 𝑠 ∈ 𝑝 as a sentence from 𝑝, and 𝐵𝑂𝑊 as a function that can be applied to 𝑠 or 𝑝 to convert it into a simple BOW representation applying stopwords removal and lowercasing (lightweight NLP that is language dependent). We create several language-independent features for learning a relevant page classifier such that it can be used for content in any language. For each word 𝑤 in the domain specificity lexicon 𝑤 ∈ 𝐿, we search the word in 𝐵𝑂𝑊(𝑝) and count its frequency as 𝑐𝑜𝑢𝑛𝑡(𝑤, 𝑝). Then for each of the four metrics introduced above, we create four features (so a total of 16 features): the sum, average, minimum and maximum of the scores, as follows, where ‘metric’ generalises the four metrics explained before (ntf, ndf, ntf-ndf, weirdness). Further, for each of the two domain dictionaries (form and measure), we create four features (so a total of 8 features) using the same equations below, by replacing 𝐿 with a dictionary 𝐷, and setting 𝑚𝑒𝑡𝑟𝑖𝑐(𝑤) = 1 always. Together this gives us a total of 24 ‘page-level’ features. Finally, we calculate the percentage of sentences that contain at least a number (e.g., indicating a volume/quantity or measurement although can be noise sometimes), and evaluate the percentage of sentences that contain at least one matched word from the lexicon/dictionaries and are calculated as follows (where 𝐿 indicates the domain specificity lexicon but can be replaced by either of the two dictionaries): In total, we obtain 52 features for a page. Our feature extraction process is arguably language independent, as it depends on counting word frequencies using lexicons/dictionaries and pre-computed scores assigned to the entries in these lexicons/dictionaries. The language dependent processes are the translation of the domain lexicons/dictionaries, and the lightweight NLP processes for which models for different languages are more available. For businesses, these help create more affordable multilingual NLP solutions than other choices such as translating input documents, or using more language dependent models. We will discuss this further in later sections. In order to learn a classifier that determines if a page is relevant or irrelevant, we asked domain experts to annotate a random collection of pages extracted from the raw TED datasets. For the machine learning algorithms, we compare linear SVM, logistic regression, and Random Forest. 4.3.2. Table selection 4.3.2. Table selection Similar to page selection, the goal here is to filter tables that are irrelevant. This is common because tables are used in tender notices to describe not only the lots, but oftentimes award criteria, product specifications, and formatting purposes. In this work, we assume that all tables are ‘horizontal’, that is, their columns define attributes of data instances and their rows contain individual data instances. We define 𝑡 as a table and 𝑟 ∈ 𝑡 and 𝑐 ∈ 𝑡 represents rows and columns in the table. If we consider content in a table as a structure-less page, and concatenate each of its rows or columns into a sequence of words equivalent to a sentence potentially describing an instance, then we can apply the same feature extraction methods explained above. In cases of a row or column span, we duplicate the value across all the rows/columns covered by the span. Therefore, we obtain a total of 80 features as follows: ● 24 ‘page’ level features if we treat the table as a simple flat textual page ● 28 ‘sentence’ level features if we treat the table as a simple flat textual page, and its rows as sentences ● 28 ‘sentence’ level features if we treat the table as a simple flat textual page, and its columns as sentences For training, our domain experts annotated a collection of tables extracted from the raw TED dataset (to be detailed later) and we compared the same group of machine learning algorithms. Authors: (1) Ziqi Zhang*, Information School, the University of Sheffield, Regent Court, Sheffield, UKS1 4DP (Ziqi.Zhang@sheffield.ac.uk); (2) Tomas Jasaitis, Vamstar Ltd., London (Tomas.Jasaitis@vamstar.io); (3) Richard Freeman, Vamstar Ltd., London (Richard.Freeman@vamstar.io); (4) Rowida Alfrjani, Information School, the University of Sheffield, Regent Court, Sheffield, UKS1 4DP (Rowida.Alfrjani@sheffield.ac.uk); (5) Adam Funk, Information School, the University of Sheffield, Regent Court, Sheffield, UKS1 4DP (Adam.Funk@sheffield.ac.uk). Authors: Authors: (1) Ziqi Zhang*, Information School, the University of Sheffield, Regent Court, Sheffield, UKS1 4DP (Ziqi.Zhang@sheffield.ac.uk); (2) Tomas Jasaitis, Vamstar Ltd., London (Tomas.Jasaitis@vamstar.io); (3) Richard Freeman, Vamstar Ltd., London (Richard.Freeman@vamstar.io); (4) Rowida Alfrjani, Information School, the University of Sheffield, Regent Court, Sheffield, UKS1 4DP (Rowida.Alfrjani@sheffield.ac.uk); (5) Adam Funk, Information School, the University of Sheffield, Regent Court, Sheffield, UKS1 4DP (Adam.Funk@sheffield.ac.uk). This paper is available on arxiv under CC BY 4.0 license. This paper is available on arxiv under CC BY 4.0 license. available on arxiv available on arxiv

How Vamstar Identifies Relevant Content for Lots in Tender Documents

Cover photo: Daniel Cheung, Unsplash In the last couple of years, when being initiated into a new design project, I was told a few times that my work should be aimed at Gen Z or Millennials. Vaguely enough, often there was no substantial reasoning for choosing a particular demographic cohort or any other details about the target audience, leading to long conversations attempting to understand the client's thinking. Mostly, I hear something about Zoomers being better at adapting to new technologies or the desire to serve people with shared values often expressed by individuals older than the target audience that they envision for their product. Also, the idea that Gen Zers have better visual taste pops up from time to time. I see no problem in startup founders and product managers limiting their target audience by age when it's with good reason. I frequently find, though, that thinking about age in regard to designing a product comes based on the wrong premises while the right premises get ignored. The question of age and generational theories is a predominantly complex matter. In the rather narrow field of design it's multidimensional too. The Birth of Generational Labels Headlines like "Generation Z this" and "Boomers that" can be seen all over mass media. I also frequently encounter memes about generational features and differences on TikTok. It seems that we are experiencing a great deal of brainwashing about generational labels on a daily basis. No wonder, the conversation about generations occasionally emerges. The buzz around generations is rooted in works of Hungarian sociologist Karl Mannheim and American authors William Strauss and Neil Howe. Mannheim his highly influential theory of generations in 1928, stating that people are considerably impacted by the socio-historical circumstances of their youth. Strauss and Howe, working as a duo, applied Mannheim's ideas to American society in their book . formulated "Generations: The History of America's Future" When we hear about boomers, zoomers, millennials, and alphas, these words are referencing the Strauss-Howe classification, hence the terminology is not universal and can hardly be employed outside of the American market anyway. It's not the main problem, though. Generational Divide Is Misleading I believe that socio-historical circumstances impact people indeed. Nonetheless, I don’t think this idea is enough to put individuals into boxes based on age and pronounce them to have similar features and behaviors because of that. The main problem with the generational theory is that it has no hypotheses that can be tested, thus it’s just an appealing speculation. If we look at thoroughly done research, there's more evidence that generational segregation is meaningless than any verification of it. For instance, according to , less than 20% of individuals within the same generation agree on their values, and this is true across all parts of the world. Researchers, who tried to assess differences in workplace attitude among generations, no substantial differences. Valuegraphics found In 2021 Professor Bobby Duffy, a renowned British social researcher, published the book debunking a lot of common stereotypes about generations. In May 2023 the famous Pew Research Center that it would stop using generational labels due to their meaninglessness. "The Generation Myth", announced I sometimes encounter these labels when scrolling through memes. Funnily enough, in the memes stereotypes are being attributed all over the place, making it hard to comprehend how exactly zoomers, millennials, and boomers are different. Moreover, in the comment section people often write that if the meme is correct, they don't belong to their generation. TikTok My hypothesis is that the hype around generations is artificially blown up by mass media and overshadows the real problem — insufficiency of communication. It looks like people find it easier to label others in an attempt to understand their behaviour rather than learning to communicate and be a good listener. I bet that we can overcome a lot of "generational differences" if we all learn to communicate better. Approach to Age in Design Designing a product and all the experience around it is a complex task. One needs to think about how the product is going to be experienced by customers on multiple levels, including the logic behind the interface, the look of the interface, and the marketing materials. The conversation about the age of the target audience may pop up at any of these levels. It makes sense to aim at a particular generation, when the product is indeed intended for people within a narrow age bracket. However, even within the chosen cohort, there are going to be very different people with diverse preferences, tastes, and values. Based on my experience, most often, when a client says that they need design appealing to Gen Z, they mean that they want something trendy regarding the visual side of the project. For example, a font that is currently overexploited by all edgy fashion brands or is popular on Dribbble. Trends are fine, but they are tricky. Today they are in, tomorrow they are out, and you are paying for one redesign after another. It makes more sense to define values and ideas that you want to channel through the visuals and how it can differentiate you from competitors. These things may have very little to do with customers' age or with trends. Conversely, I feel that we talk about age insufficiently when it comes to usability and accessibility. If you are creating a product for a broad audience, it makes sense to care about users of all ages. Otherwise, you have a chance to never reach some of your potential customers or scare them away. to the Pew Research Center, the number of senior US citizens owning a smartphone and using social media is constantly growing. Seniors are gradually adapting to the digital age, but they still perceive interfaces differently in comparison to younger generations. According For instance, a recent shows that while younger users easily understand the functionality behind a button consisting of an icon only, a text button or a button combining an icon with text is the best option if you want to make an interface easy to comprehend and use for people of all ages between 16 and 65. Markswebb's research Don Norman, the co-founder of the legendary Nielsen Norman Group, is almost 90 years old himself, and two years ago he us that the world designed with no consideration of the needs of the seniors may feel isolating for this demographic. For instance, they may need a bigger font size to be able to read it from the screen with comfort and communicate via a messenger with their grandkids. reminded To be honest, I feel a bit of anxiety because of the media's mindless overexploitation of generational labels. It leads to nothing but ageism which may already be working against our parents and will work against us in the future. We all – startup founders, product owners, developers, designers, marketers, etc. – have our share of impact on the world, in our power to make it a better place.

When Age Matters in Designing a Product and When it Doesn't

For most people, when you mention Search Engine Optimization, keywords are what come to mind. But there’s more to than keywords, content pillars and cluster topics. SEO First, let’s take a broad look at SEO What Are the Three Types of SEO? There are three types of SEO, all of which work together to improve a website's ranking. I like to look at them as the three main organs of a body. When one isn't working optimally, it’ll affect the others. Make sense? I thought so! Off-Page SEO Off-page or Off-site SEO refers to the actions performed outside your website to help it rank on the search engine results page(SERPs) This includes; Backlinks Guest Posting Listings PR On-Page SEO As the name suggests, On-page SEO refers to measures taken directly within the website to improve its position in SERPs. This includes; Content Quality Keyword HTML Tags Internal Linking Meta Data URL Image Optimization Technical SEO Technical SEO refers to server and website optimizations that make it easier for search engine spiders to effectively crawl and index your website to help improve organic ranking. These include: Security Schema Markup Site Speed Accessibility Canonical URLs User Experience (UX) Indexation As you can see, there’s a whole lot that goes into SEO, but in this article, we’ll take a deeper look at Image optimization. What is Image Optimization and Why is it Important? Simply put, image SEO is optimizing images on your site using contextual information so that both bots and humans can understand them. Why is image SEO crucial? Because images are an essential part of your website’s design and user experience. When correctly optimized, they significantly impact your website’s performance, helping it rank higher in SERPs. According to data from Jumpshot, over of all U.S. web searches happen on Google Images. 20% That’s huge! And Without properly optimizing images, you are wasting a valuable SEO asset. According to Google’s Gary Illyes, video and image search are often overlooked, and SEOs ought to be paying more attention. In fact, in 2019, tasked more engineers with working on media search. Here's what Gary Illyes had to say on . Google Reddit “We simply know that media search is way too ignored for what it’s capable of doing for publishers so we’re throwing more engineers at it as well as more outreach.” Optimized high-quality graphics, for example, stand a better chance of appearing in image searches. This consequently drives more traffic to your website. Image SEO also lowers bounce rates, improves UX, improves load time and increases engagement. Image from Witty Content Writers 4 Tips to Get You Started on Image SEO All high-performing websites you’ve come across have taken effort and time to optimize their images. There are various ways you can optimize images such as: Using High-Quality ImagesAdding Alt TextCustomizing Image File NamesInserting Title TextUsing Captions and the <figcaption> TagCreating Unique ImagesEnsuring Surrounding Text Matches Your ImagesAvoiding Keyword StuffingUsing Accurate Names and TextCreating an Image SitemapServe Responsive, Mobile-Friendly Images Let’s look at four great actionable practices for Image SEO. Optimize Image Sizes and Formats Choosing the right size and format for images is one of the ways to significantly impact image SEO. , JPEG, SVG, and PNG formats offer different advantages depending on your website's needs. WebP WebP images have small image sizes optimized for the web and lossless compression. For less complex images like logos, SVG images provide good quality for such small file sizes. Although they tend to be smaller, JPEGs are of low quality. In contrast, PNG files have higher image quality and larger file sizes. As such, they’re best suited for highly visual websites. Displaying images at their full size isn’t necessary for most websites. Overly large images increase load time and take up unnecessarily more server space. That’s why you need to compress and resize images before you upload them to your site. You can do this by: Using a free image compression plugin like ShortPixel Using a web-based solution like to compress WebP, PNG and JPEG graphics TinyPNG Add Alt Text Alternate text is a brief description of the content of an image. When the graphic can’t be displayed on a site, the alt text is what’s visible. It also helps those who are visually impaired understand visual content using screen readers. Image from Wo WordPress Wordpress.com Search engines use alt metadata to determine what type of content is in the image and page. That’s why including keywords in alt text enhances your image SEO. Just be sure to use keywords that match the content of your images. Create an Image Sitemap An image sitemap is a document carrying all graphic elements on your site. It helps search engines rank your content faster and more effectively. Google's shows you how to add image sitemaps to your site. It’s possible to integrate an image sitemap into a normal sitemap or have it completely separate. guide on image sitemaps Image from Image Maps An image on an isolated page may be overlooked in place of more accessible or popular content. But by using a sitemap, all your valuable visual content get’s accounted for. It also helps search engines discover it. With common WordPress SEO plugins like Yoast and Rank Math, you can include images in your sitemaps without worrying about code. Serve Responsive, Mobile-Friendly Images , as of June 2022, mobile device traffic worldwide was at 58.99%. This goes to show just how essential mobile-friendly responsive images are. They adapt easily to different mobile devices, allowing users to easily interact with them. According to Statista Offering various versions of an image helps web visitors’ browsers select the most appropriate graphic size for them, avoiding any odd placements. Key Takeaways Choose the correct file format Reduce file size to improve page load speed Match your on-page SEO elements, like metadata and structured data with your image For crawlability, create an image sitemap and ensure that your images are in your sitemap Optimizing images is no joke. Advancements in voice search technology have made media increasingly important and optimizing images is a great way to reap the benefits of this very often overlooked way to improve your content and help rank higher in SERPs.

4 Ways to Optimize Your Site Images and Drive Traffic to Your Site

Character Encoding Demystified: Everything you Need to Know About ASCII, Unicode, UTF-8

Does this sound familiar - you have a great idea, you build that idea, you start looking at the final outcome and think... This looks terrible. And the worst part, you don’t know how to fix it. We are developers after all, not designers. Luckily, you don’t have to be an amazing designer. You just have to be resourceful and understand a couple of simple design tips. The internet is full of free tools and resources to supplement your lack of design skills. I’m going to show you how to take advantage of those resources to create something that you can be proud of. This guide will cover the tools, techniques, and resources I use for designing websites. It is not the end all be all but it is a great guideline for people who struggle with design. We are going to build an application called DesignDo. It will be a collection of design tips and resources (so meta!). Everything we use will be To keep our focus on the point of this post, we won’t be using a front-end framework. Just static HTML. 100% free. Planning Use a CSS Framework Have you ever been bowling with someone who is terrible at it? They put those bumpers in the lane so they can't throw a gutter ball? The CSS framework is your bumper. It’s almost impossible to throw a “gutter ball” with them. They will give you a great starting point and also provide constraints that a dev who struggles with design needs. I would recommend Tailwind CSS. It uses classes that are abstractions of CSS properties. It serves as training wheels to learn CSS plus it’s easy to make great-looking sites with it. Read the docs . here Also, if you are not using a front-end framework like React or Vue, I would suggest using alpine js. It’s great for simple UI changes like turning your nav into a hamburger menu when on mobile devices. Colors Keep it simple. Use a white background and black or dark gray text. If you have black text in an area with a darker background color, like a button, change it to white. for your logo and call to action buttons. Use it sparingly for things you want to “stand out” in your design. 1 primary color should be a light color to make cards stand out from the white background. The simplest way is to just use various light grays for this. They go with everything. Secondary colors Determine the personality of your site Every site has a personality. They can be serious and business-like. They can also be fun and light-hearted. Figure out what you want the personality of your site to be. This will determine a couple of CSS properties you will be using. Figure it out? Good. Now, if you chose , I suggest using: serious : Use the default tailwind CSS font. You won’t have to configure anything. Font : Do not use border radius. Square buttons and boxes give off serious vibes. Border Radius : For your primary color, dark blues, greens, grays, and purples are all great choices. Color If you chose , I suggest using: fun : Tailwind Default or Font Outfit : Use a medium to large border radius. Pick 1 though. Do not mix different border-radius sizes. Keep it consistent. Border Radius For your primary color, you have a lot of options. I would suggest a lighter vibrant color though. Color: Use 1 font for your application. It’s too hard to find multiple fonts that complement each other, especially if you suck at design. Find Inspiration Alright, so you have an idea of what personality your site will take on. Now go find some inspiration to spark your creativity. DO NOT COPY someone's design to the T or their code. Use it as your north star. Web Design Inspiration Sites https://land-book.com/ https://www.awwwards.com https://www.siteinspire.com/ https://dribbble.com/ My choices DesignDo will be a fun/creative site. I’ll be using the default tailwind font and a large border radius on buttons and cards. I’ll also be using light violet as my primary color. I'll be using this website template as my inspiration... It has a nice hero section where I can put some copy about what my site is and a vertical list, which I’ll need to list my design tips and resources. Work Smarter, Not Harder Part of getting better at development and design is swallowing your ego. You aren’t a better developer because you build everything from scratch. You are just a slow developer. The same goes for building your site. Find free components or themes I never start my HTML from scratch. I always start with some pre-built components and go from there. Were going to use a great free option, . Wicked Blocks Using my inspiration site to guide me, I picked out the following blocks to use. My Header Hero Below is the fold content: Footer Put your blocks together in HTML Before we start customizing our site, let’s put our blocks together in HTML. It’s a good start but we have some work left before DesignDo is where we want it. Logo Again, keep it simple! I usually go with a text-based logo for my sites. The name of my site is DesignDo, so that is what my logo will be. I’ll typically use the primary color as a background and use white or black text. The text color is all about readability. So if the white text is hard to read on top of your primary color, use black text and vice versa. You can create your logo by either writing HTML or using an image editor. If you decide to use an image editor, I recommend . Photopea It’s a knock-off of photoshop, it’s free, and it’s amazing! You can also upload your own fonts for text. For the DesignDo logo, I’m just going to use the name as the text with extra bold font. The background will be my primary color. I’m also adding a slight rotation to the logo to give it a little more character. (Do not do this if you're going for a more serious design. Nav Links There shouldn’t be much to do with nav links. Make sure the text is gray or black. If you want to add icons along with the text you can. is a great free icon library that I use all of the time. I would suggest using this if you're making a more serious site. Heroicons You can also use Heroicons for a fun/creative site. I also like to use emojis for icons if my site's personality is fun and creative. For DesignDo I only have 1 nav link and I’ll be using the fire emoji to bring some more attention to it. Hero Because we used Wickedblocks to give us a headstart, the hero section doesn’t have a lot for us to do. We just need to update some text. When dealing with large headline tags like ours, structure it in a way where the first line is longer than the second. It’s more appealing to the eye. We also need to update the color of our Notify me button to our primary color Content Once again, we're not far off on how we want our below-the-fold content to look. I’m not feeling the icons. They don’t go with the personality of the site. So I’m going to reach out for Heroicons to get something more in line with our theme. That’s better! I’ll leave the background of the icons alone since it’s a light gray and is a nice contrast with our dark icon and white background. It also is rounded so it follows our theme of large border-radiuses. The “Short Length Headline” is an H2 but it’s using the same styles as its description below it. We want the headline of the tip to stand out and get the reader's attention. I’m going to increase its size and font-weight slightly (we don’t want it larger than the H1 tag above it). Finally, we will update the cookie-cutter text that came from wicked blocks with our own design tips! Summary A lot of people champion using their own handmade CSS and using design tools like Figma to create mockups. I agree, they are both great things. But the truth is, as a developer, I don’t have the time to become an expert in Figma or to write my own custom CSS library that I can carry from project to project. But I can be resourceful and use free tools and resources on the internet to create good-looking designs at a fast pace. If you want to check out the project I created in this post, follow these links: You can find DesignDo . here The repo for this project is . here If you enjoyed this article, feel free to join my newsletter @ thefreelancedev.com https:// Also published . here

How To Make A Great Looking Website Even If You Can't Design

Cómo hacer un sitio web atractivo incluso si no puedes diseñar

Comment créer un site Web attrayant même si vous ne pouvez pas concevoir

अगर आप डिज़ाइन नहीं कर सकते हैं तो भी एक शानदार दिखने वाली वेबसाइट कैसे बनाएं

デザインができなくても見栄えのするウェブサイトを作成する方法

Como fazer um site bonito, mesmo que você não saiba criar

Cách tạo một trang web đẹp ngay cả khi bạn không thể thiết kế

即使你不会设计，如何制作一个漂亮的网站

Natural Language Processing (NLP) is the ability of a computer system to understand human language. Natural Language Processing is a subset of Artificial Intelligence (AI). There are multiple resources available online which can help you develop expertise in Natural Language Processing. In this blog post, we list resources for the beginners and intermediate level learners. Natural Language Resources for Beginners A beginner can follow two methods i.e. Traditional Machine Learning and Deep Learning to get started with Natural language processing. These two methods are very different from each other. For the inquisitive, the difference between these two. here’s Traditional Machine Learning Traditional machine learning algorithms are complex and often not easy to understand. Here are a few resources which will help you get started in learning NLP using machine learning: Speech and Language Processing by Jurafsky and Martin is the popularly acclaimed bible for traditional Natural Language Processing. You can access it . here For a more practical approach, you can try out Natural Language Toolkit. Deep Learning Deep learning is a subfield of machine learning and is far better than traditional machine learning due to the introduction of Artificial Neural Networks. Beginners can start with the following resources: CS 224n: This is the best course to get started with using Deep Learning for Natural Language Processing. This course is hosted by Stanford and can be accessed . here Yoav Golberg’s free and paid books are great resources to get started with Deep Learning in Natural Language Processing. The free version can be accessed and the full version is available . here here A very thorough coverage of all algorithms can be found in Jacob Einsenstein’s notes from GATECH’s NLP class which deals in almost all NLP methods. You can access the notes on GitHub . here Natural Language Resources for Practitioners If you are a practicing Data Scientist, you will need three types of resources: Quick Getting Started guides / Knowing about what is hot and new Problem-Specific Surveys of Methods Blogs to follow regularly Quick Getting Started guides / Knowing about what is hot and new One can start with Otter et al.’s Deep Learning for Natural Language Processing survey. You can access it . here A survey paper by Young et al tries to summarize everything hip in Deep Learning based Natural Language Processing, and is recommended to get started with Natural Language Processing for practitioners. You can access the paper . here You can refer to article to understand the basics of LSTMs and RNNs, which are used in Natural Language Processing a lot. Another much more cited (and highly reputed) survey of LSTMs is . This is an interesting paper to understand how the hidden states of RNNs work. It is an enjoyable read and can be accessed . I always recommend the following two blog posts anyone who hasn’t read them: this here here http://colah.github.io/posts/2015-08-Understanding-LSTMs https://distill.pub/2016/augmented-rnns/ Convolutional Neural Networks (Convnets) can be used to make sense of Natural Language. You can visualize how Convnets work in NLP by reading this paper . here How Convnets and RNNs compare with each other has been highlighted in paper by Bai et al.. All its pytorch (I have stopped or reduced, to a large extent, reading deep learning code not written in pytorch this ) code is open sourced and gives you a feel of Godzilla v/s King Kong or Ford Mustang vs Chevy Camaro(if you enjoy(ed) that type of thing). Who will win! . here Problem-specific Surveys of Methods Another type of resources practitioners need is answers to questions of the type: “I have to train an algorithm to do X, what is the coolest (and easily accessible) thing I can apply?”. Here’s what you will need for this: TEXT CLASSIFICATION What’s the first problem people solve? Text Classification, mostly. Text Classification can be in the form of categorizing text into different categories or detecting sentiment/emotion within the text. I would like to highlight an easy read about different surveys of Sentiment Analysis described in this ParallelDots . Though the survey is for sentiment analysis technologies, it can be extended to most text classification problems. blog Our (ParallelDots) surveys are slightly less technical and aim to direct you to cool resources to understand a concept. The Arxiv survey papers I point you to will be very technical and will need you to read other important papers to deeply understand a topic. Our suggested way is to use our links to get familiar and have fun with a topic but then to be sure to read the thorough guides we point to. (Dr. Oakley’s talks about chunking, where you first try to get small bits here and there before you jump deep). Remember, it is great to have fun but unless you understand the techniques in detail, it will be hard to apply concepts in a new situation. course Another survey of Sentiment Analysis algorithms (by people at Linked University and UIUC) is . here The Transfer Learning revolution has already hit Deep Learning. Just like in images where a model trained on ImageNet classification can be fine-tuned for any classification task, NLP models trained for language modeling on Wikipedia can now transfer learn text classification on a relatively lesser amount of data. Here are two papers from OpenAI and Ruder, and Howard which deal with these techniques. https://arxiv.org/abs/1801.06146 . https://s3-us-west-2.amazonaws.com/openai-assets/research-covers/language-unsupervised/language_understanding_paper.pdf Fast.ai has a more friendly documentation to apply these methods . here If you are transfer learning two different tasks (not transferring from Wikipedia language modeling task), tricks to use Convnets are mentioned . here IMHO, such approaches will slowly take up on all other classification methods (simple extrapolation from what has happened in vision). We also released our work on which gets good accuracy without any training on a dataset and are working on its next generation. We have built our custom text classification API commonly called the Custom Classifier in which you can define your own categories. You can check out the free . Zero Shot Text classification here SEQUENCE LABELING Sequence Labeling is a task which labels words with different attributes. These include part-of-speech tagging, Named Entity Recognition, Keyword Tagging etc. We wrote a fun review of methods to tasks like these . here An excellent resource for such problems is the research paper from this year’s COLING which gives optimal guidelines to train Sequence labeling algorithms. You can access it . here MACHINE TRANSLATION One of the biggest advances in NLP in recent days has been the discovery of algorithms that can translate text from one language to another. Google’s system is an insane 16 layered LSTM (which requires no dropout because they have tons of data to train on) and gives state-of-the-art translation results. Media experts blew the hype out of proportion with hyperbole reports claiming “Facebook had to shut down AI which invented its own language”. Here are some of these. https://gadgets.ndtv.com/social-networking/news/facebook-shuts-ai-system-after-bots-create-own-language-1731309 https://www.forbes.com/sites/tonybradley/2017/07/31/facebook-ai-creates-its-own-language-in-creepy-preview-of-our-potential-future/#1d1ca041292c For an extensive tutorial on Machine Translation, refer to Philip Koehn’s research paper . A specific review to use Deep Learning for Machine Translation (which we call NMT or Neural Machine Translation) is . here here A couple of my favorite papers are here – This tells you how to solve a problem end-to-end when you have a lot of money and data. paper by Google Facebook’s (just because of its cool convolutional approach) and its code is released as a library . Convolutional NMT system here is a framework for fast translation in C++ https://marian-nmt.github.io/ http://www.aclweb.org/anthology/P18-4020 enables everyone to train their NMT systems. http://opennmt.net/ QUESTION ANSWERING IMHO this is going to be the next “Machine Translation”. There are many different types of Question Answering tasks. Choosing from options, selecting answers from a paragraph or a knowledge graph and answering questions based on an image (also known as Visual Question Answering) and there are different datasets to get to know the state of the art method. is a question answering datasets which tests an algorithm’s ability to read comprehensions and answer questions. Microsoft published a paper earlier this year claiming they have reached human-level accuracy for the . The paper can be found . Another important algorithm (which I feel is the coolest) is Allen AI’s and its improvements. SQuAD dataset task here BIDAF Another important set of algorithms is Visual Question Answering which answers questions about images. Teney et al.’s from VQA 2017 challenge is an excellent resource to get started. You can also find its implementations on Github . paper here Extractive Question Answering on large documents (like how Google Highlights answer to your queries in first few results) in real life can be done using Transfer Learning (thus with few annotations) as shown in this ETH paper . A very good paper criticizing the “understanding” of Question Answering algorithms is . Must read if you are working in this field. here here PARAPHRASE, SENTENCE SIMILARITY OR INFERENCE The task of comparing sentences. NLP has three different tasks: Sentence Similarity, Paraphrase detection and Natural Language Inference (NLI) for this, each requiring more semantic understanding than the last. and its subset Stanford NLI are the most well-known benchmarks datasets for NLI and have become the focus of research lately. There are also MS Paraphrase Corpus and Quora Corpus for paraphrase detection, and a SemEval Dataset for STS (Semantic Text Similarity). A good survey for advanced models in this domain can be found . Applied NLI in the clinical domain is very important. (Finding out about right medical procedures, side effects and cross effects of drugs etc. ). from applied NLI in the medical domain is a good read if you are looking to apply the tech in a specific domain. MultiNLI here This tutorial Here is a list of my favorite papers in this domain Natural Language Inference over Interaction Space — It highlights a very clever approach for putting a DenseNet (Convolutional Neural Network on Sentence representations). The fact that this was the outcome of an internship project makes it even cooler! You can read the paper . here This research from Omar Levy’s group shows that even simple algorithms can perform the task. This is because algorithms are still not learning “inference”. paper BiMPM is a cool model to predict paraphrases and can be accessed . here We have a new work for Paraphrase detection too which applies Relation Networks on top of sentence representations and has been accepted at this year’s AINL conference. You can read it . here OTHER FIELDS Here are some of the more detailed survey papers to get information about research for other tasks you might encounter making an NLP system. Language Modelling is the task of learning an unsupervised representation of a language. This is done by predicting the (n+1)th word of a sentence given the first N words. These models have two important real-world uses, autocomplete and acting as a base model for transfer learning for text classification as mentioned above. A detailed survey is . If you are interested in learning how to autocomplete LSTMs in cellphones/search engines work based on search history, is a cool paper you should read. Language Modelling(LM) — here here Relation extraction is the task of extracting relations between entities present in a sentence. A given sentence “A is related as r to B”, gives the triplet (A,r, B). A survey of the research work in the field is . is a research paper that I found to be really interesting. It uses BIDAFs for Zero Shot Relation extraction (that is, it can recognize relations it was not even trained to recognize). Relation Extraction — here Here With the onset of the chatbot revolution, Dialog systems are now the rage. Many people (including us) make dialog systems as a combination of models such as intent detection, keyword detection, question answering etc, while others try to model it end-to-end. A detailed survey of dialog system models by the team at JD.com is . I would also like to mention Parl.ai, a framework by Facebook AI for the purpose. Dialog Systems — here Text Summarization is used to get a condensed text from a document (paragraph/news article etc.). There are two ways to do this: extractive and abstractive summarization. While extractive summarization gives out sentences from the article with the highest information content (and what has been available for decades), abstractive summarization aims to write a summary just like a human would. This from Eintein AI brought abstractive summarization into mainstream research. There is an extensive survey of techniques . Text Summarization — demo here Natural Language Generation is the research where the computer aims to write like a human would. This could be stories, poetries, image captions etc. Out of these, current research has been able to do very well on image captions where LSTMs and attention mechanism combined has given outputs usable in real life. A survey of techniques is available . Natural Language Generation (NLG) — here Blogs to follow Here’s a list of blogs which we highly recommend for anyone interested in keeping track of what’s new in NLP research. Einstein AI — https://einstein.ai/research Google AI blog — https://ai.googleblog.com/ WildML — http://www.wildml.com/ DistillPub — (distillpub is unique, blog and publication both) https://distill.pub/ Sebastian Ruder — http://ruder.io/ If you liked this article, you must follow our blog. We come up with resource lists quite frequently here. That’s all folks! Enjoy making neural nets understand language. You can also read about Machine Learning algorithms you should know to become a Data Scientist . here We hope you liked the article. Please for a free ParallelDots account to start your AI journey. You can also check out free demos of ParallelDots AI APIs . Sign Up here You can read the original article . here

List of free resources to learn Natural Language Processing

Online surveys show user’s feedback on your products. Here are 20+ useful user experience survey questions (UEQ) for you. Also with 10+ user survey templates and tools. is undoubtedly essential for companies to find out user’s feedback on a large scale. Especially when releasing a brand new product or the next iteration, survey plays a big role and save a big budget for it’s lower costs of money, time or sites fee. What is the key to conducting a successful user experience ? The questions that you ask. I’ve selected for your inspiration. User experience survey user experience survey 20 most useful user experience survey questions (UEQ) The whole article contains 4 main parts: Part 1: How to ask user experience survey questions properly？ Part 2: 20+ most useful user experience survey questions. Part 3: 10 Best user experience survey templates. Part 4: UX survey tools and websites recommendation. Part 1: How to ask user experience survey questions properly？ It’s not easy to ask good questions, and it’s actually harder to ask right questions to get useful results. Here are some suggestions when you are asking user survey questions: 1) The questions should all be focusing on UX issues. you’d also better keep your online surveys short. 2) As gathering user feedback, It depends on your research goals. For example, some UEQs aim at your product, some for your website, and even for both. Some questions are for your existing users while some for the newcomers. 3) No necessary to ask all the questions. It’s true you may ask multiple-choice questions for better response rate, but also try to ask “why” or “why not” to dig deeper. It’s a good way to get more information of your audience which can provide you critical insights and ideas. 4) Don’t forget to ask open-ended questions. It’s a good thing you get visitors to answer your survey. But how do you grab the chance to get them register as your potential user? You may lead the visitors to a landing page of your site, or offer a form or CTA to allow them to sign up directly in your survey. 5) Try to turn “visitors ” to “registered users”. Part 2 20+ Most Useful Questions for User Experience Surveys 1. How did you first learn about our product/website? This question can evaluate your marketing channels and point out where people meet your brand, so you can finger out what to focus on in your advertising efforts. It’s true Google Analytics can show the visitors come to you from certain channels like AdWords or Facebook campaigns, organic search, etc. But it can not tell the direct traffic. Suggestion: Target new visitors especially. They can remember how to found out about you more clearly. 2. How likely is it that you could recommend our product/website to a friend or colleague? This question shows user feedback directly, and it’s also a popular way of measuring customer satisfaction. If a user won’t recommend you to their friend, they may feel disappointed at some aspects. That’s where you should improve. Suggestions: Target your existing users or visitors who visit your website multiple times (based on cookies or history of browsed pages). If you got the negative answer, try to ask “why not?” you can use free text or list several options like ABCDE. 3. What comes to your mind when thinking about our product/website? This is a good way to capture user’s perceptions of a website or product. But you need to make sure that visitors or users can show their views clearly. Then you may find out what your customers think of you. Suggestion: Allow users to enter their own words (as low as 1 word and as high as you think people need) or a user canchoose 1 or more options from a list. It’s hard to analyse random data like this but maybe create a word cloudcan be helpful. 4. If review our product/website with a score out of 10, what score would you give us? It’s the same as online shoppers do customer reviews and ratings. You can ask users to review your website or product. Suggestions: Ask for an overall rating or ask them to rate specific areas. Use a 10 point score rather than 5 points because it gives participants more flexibility. You might consider a follow-up question: “why have you given this score?” 5. Which of our competitors did you consider before choosing us? It’s a good question to get to know your competitors. What you can learn from your competitors is beyond your imagination. Sometimes you might be surprised who is considered as your competitor. This may bring you different inspirations Suggestions: Use this question in a post-purchase survey. You’d better know who you compete with in advance. 6. Why choose us over competitors? It’s a good question to know your advantages over competitors. There are dozens of possible reasons why user chose you, it might be the price, the brand, the functionality, etc. This question helps you know what makes you really stand out. You’ll know how to acquire even more customers with those features. Suggestions: Use this question in a post-purchase survey. Ask this question to existing customers. 7. How does our product/website compare to X product? It helps you to investigate the strengths and weaknesses of your website or product. pay attention to thecompetitor you chose. Suggestions: The competitor you chose should better be known to the customers. Consider follow up questions: “Which do you prefer?” This makes you know more about to handle competitions. 8. What do you like least/most about our product/website? This to know what makes your user sad and what makes them happy. It’s not right to add more and more new features and neglect to improve the existing features. This question helps you know how to solve the pain point and improve the existing features. : Suggestions You’d better offer free text rather than just providing a list. Make sure they can express their pain points clearly. The question is especially for planning any changes to your website or product. 9. How easy is our product/website to use? This question is to investigate the usability of your product or service. Offer some options here ranging from Very Easy to Impossible, like “Very easy to use”, “Easy to use”, “Neither easy nor difficult to use”, “Difficult to use” and“Very difficult to use”. Suggestions: Don’t the options balanced, don’t provide answers with 4 positive choices and 1 negative one. Also consider asking “Why? if a user selects the negative answers. It helps you investigate the reasons behind and improve your service. 10. Which feature of our product/website are most important to you? You know the important features of the website/product very well, but for the customers, they may choose you for a tiny feature you’ll never think about. This question helps you concentrate on the features that matter most to your users. You’ll probably capitalize on the features if they are not easy to find. Suggestion: Target existing users. 11. Which feature of our product/website are least important to you? You cannot deny that there are features nobody wants or finds out useful. This question helps to point out those unnecessary parts and make you concentrate on fixing those that everybody really needs. Suggestion: Target existing users. 12. What is the most important feature you think we should add? We hear the words that you should add on certain features a lot from our users. This question informs you the weakness and shortage. Moreover, it can help you build a good product roadmap. Suggestion: Target returning registered visitors. Provide free text options. 13. Why started using our product/website? This is the same to what was your goal when using our product. Visitors have different reasons for visiting your website. Some want to check what you can do, some want to buy, and some want to read an article on your blog, and some even form your competitor’s side. It’s better to know what draws their attention. Suggestions: Target all visitors Ask the follow-up questions “Did you meet the goal of your visit?”. 14. What stopped you from buying? It’s a question to optimize your purchase path and increase sales. There are dozens of possible reasons that may stop users from buying. Is it the price? The features? You have no clear ideas actually. Well, this is why you may consider asking this question. Suggestions: Consider this as a post-purchase survey question. Target the registered visitors especially. 15. What is your preferred payment/delivery method? You may think it’s ridiculous that some people quit buying from you just for the payment they don’t use, but it is true. Make sure you offer all methods popular with your visitors. Suggestions: Target all visitors who browsed you buying page. Use sampling. 16. Is our pricing clear to you? Usually, we offer different subscriptions packed with different features. But is the distinctions between plans clear to visitors? Suggestions: Target visitors who spend more than average on the pricing page. Ask a follow-up question: “How can we improve it?” when someone answered no. 17. What was your first impression when you entered the website? The first impression matters a lot. People still charge a book by its cover. Your website can influence users mood magically. For example, if a website of a bank looks lousy and fashion, what will you think? Suggestion: Ask this question on the main page, target only new visitors. 18. Is there anything missing on this page? Pages of your website should provide visitors with all information they need. But unfortunately, it’s so hard to make all visitors satisfied. Sometimes the information may get you overwhelmed. This question helps you figure out what are the necessary ones that shouldn’t miss. suggestions: Set the question to appear after 75% of scroll or more. Target new visitors especially. 19. What are your favourite blogs? Guest blogging is now a common and effective way of expanding your reach. How to choose from the crowd? Go ask your users. Suggestion: Target readers of your blog and active users. 20. If you were unhappy with us what can we do to get you to come back? This questions may recall uses’ affection and also save a chance to find any customer issues. Other questions you might consider: 1. What was the one time you found our product/website to be highly satisficing? 2. Have you paid for any of these other products or tools? 3. What do you like or dislike about these other products or tools? 4. How upset would you be if [your product or website] was no longer around? 5. What other products do you use to accomplish similar tasks and why? 6. If you could change one thing about website/product what would it be and why? 7. How often do you use our product/website? 8. Was there anything surprising or unexpected about this product? 9. What are the most frequent tasks you do when using our product? Part 3: 10 Best User Experience Survey Templates 1. Customer Satisfaction Survey Template Customer satisfaction surveys can help you find out what people think of your company, get feedback on customer service, and more. You can also customize the templates and add questions if there’s a part of the customer experience you want to explore more fully. 2. Software and App Customer Feedback with NPS® This template is tested, certified, and built by survey research experts to ask the questions that are most associated with raising the NPS of software companies and products. Send it to your users to uncover what’s driving their rating of your product or service. The feedback you get can be invaluable to improving your product or business. 3. The Net Promoter® Score Survey Template Net Promoter Score sample survey template is designed to make it easy to understand how satisfied your customers are. Net promoter scores are produced by asking respondents to rate the likelihood they would refer your business to someone else, using a scale from 0 to 10. You can also customize our NPS template to help it fit your product offering, or to ask additional questions to understand your customers even better. 4. Customer Service Survey Template Use the results from this survey template to provide support and training to your customer service team. The questions can help to measure customer satisfaction and improve how your team works with your customers. Customize the survey if there is more you’d like to ask. 5. Customer Comments Survey Template You can just send out this quick survey to get a sense of how customers view your company. The survey asks customers to explain their responses, so you can have a better understanding of their answers. Customize the template if you want to ask additional questions. 6. Product Feedback Survey Template This template helps you gather user’s feedback and provides solutions to solve it. 7. Product Survey Form Get a quick feedback on your products from your customers using this Product Survey Form. 8. Patient Satisfaction Survey Get patient feedback with this online satisfaction survey and improve your service. 9. Customer Satisfaction Survey Great tool to capture customers concerns on products and services used 10. Feedback Form See your business or product through the eyes of your customers to help you stay ahead of the curve. 11. Client Satisfaction Form A survey form for my freelance clients. 12. Buyers Questionnaire Crg Form used to contact buyers. Useful for real estate management. Realtor friendly. 13. User Experience Survey Gather direct feedback to gain insight inon your website’s user experience. Part 4: UX Tools and Websites Recommendation. Mockplus — A prototyping and wireframe tools How do you ensure a product/website obtains great user experience at the very beginning? How do fully display those good ideas and inspirations? I suggest you start with prototyping. Mockplus is known for quick prototyping design. With more than 200 highly-packaged components and over 3000 vector icons available, you can quickly finish the layout. Moreover, Mockplus adds interactive settings with a quick and easy drag-and-drop. You can generate clickable, dynamic product prototypes in just minutes. It supports 8 ways to preview and you can test on native device instantly. You can review the prototype with commenting and marking up the design easily. On the other hand, before sending out a survey to your users, you’d better do a prototype with Mockplus to test inside. A: Mockplus More prototyping tools: The Best Mockup & Wireframing Design Tools & Apps for UI/UX Designers in 2017 2. UserTesting — User surveys and user testing tools Do you want to know when someone visited your site? Where did the visitor click? Have they found what they need? These questions can be answered by using user surveys and user testing tools. You might consider UserTesting, a tool that can provide a variety of ways to test the user experience of your site. You can freely choose any one of user to investigate. Other tools/websites: Optimizely — A / B test UEQ-Online — User experience question online. PollDaddy — Create stunning surveys, polls, and quizzes in minutes. SurveyGizmo — Advanced online survey software tool that comes with amazing customer service. SurveyMonkey — Free online survey tool to capture the voices and opinions of your user. Lookback — A good tool to record users. Wufoo — Online survey builder. If you like this article, please don't hesitate to leave a comment. Thanks.

20+ User Experience Survey Questions and Templates for Inspiration

Más de 20 preguntas y plantillas de encuestas sobre la experiencia del usuario para inspirarte

Welcome, fellow traveller of the uncharted wastelands between the realms of programming and UX! Hopefully your journeys have been safe from dragons and giant mutant project managers. Here, have a cup of warm tea, sit by the fireplace, and tell us your tale. We at Neonto have been hard at work on improving React Studio, in the ardent hope it could be your new companion in your travels through these hazardous lands of adventure. For this 1.4 release, we’ve been gathering feedback from users and focusing on improvements for real-world projects. There are important new organization tools, new ways to model standard components like Login, new tools to handle data and query parameters for back-ends, more plugin APIs so that you can make the generated React projects work exactly as you need… And a ton of fixes and small improvements. To install, just launch React Studio and the update should be offered automatically. Or, you can download the app from . reactstudio.com There’s also a new example project (in File > ) that shows off some of these new organizational features in a simple way. And, if you’re interested in being a part of React Studio’s future, at the very end of this post we have a surprise. Let’s go! New From Template Project Map Sections Until now, it could be challenging to organize your React Studio project when it gets larger than one or two screenfuls. On a large project with a dozen screens and their associated components, the Project Map could stretch so large that it was difficult to find your way around. The new Map Sections feature solves this problem. Sections are a lot like folders: you can move screens and their related components into sections to keep them together. On the top level of the Project Map, the section is shown as a large block with a preview of its contents (see screenshot above). You can also add a note to each section to better keep track of what each section contains. To create a map section, select any blocks in the Project Map and choose “Make map section from selected” in the command menu that appears in the bottom-right corner of the Project Map. To edit a map section’s contents, double-click on it. To move blocks out of a section back to the top level, or into another section, select them and use the command menu in the bottom-right corner. You can also “unpack” all blocks from a section by selecting the section and using the command in the inspector. Sections are also going to be a useful tool for organizing React Studio’s code output to your liking. In the near future, we’ll add a feature that lets you decide how sections should be represented in the generated code — for example, you could have each section’s contents written into a separate source code folder, and top-level components placed into a folder named “shared”. New example project: “FauxAir Check-in” We’ve added a new project template that shows how map sections can be used in practice. It’s an app for those airport check-in consoles, made for a very fake airline called FauxAir. You can see the demo app here: https://neonto.cloud/u/checkindemo (Note that this demo doesn’t work on a phone. It’s meant for a check-in console that has a tablet-sized screen. With some extra effort, we could of course make it fit on a phone too, but it wouldn’t really make sense for an app that has features like “passport scanner”…) To explore this project, look in File menu > “New From Template”. Sticky Notes The FauxAir Check-in example project shows another new feature of the Project Map. You can now add little yellow “sticky notes” on the map. They’re pretty useful for keeping track of the intended use of your components, for example. To create a sticky note, drag from the Components box on the right onto the Project Map. Login Gate Take a glance at the Components box and you’ll see another new addition, the block. This is a special kind of screen that provides additional functionality for implementing logins. Login Gate Why add a new screen type for this? It’s because all logins tend to be very similar in function. Ideally there’s only one kind of login in the app, and the user shouldn’t even see a login screen if valid credentials are already available (e.g. with a browser cookie). Similarly, there should be only one implementation of the “log out” function. It’s much easier for React Studio to generate the right kind of code to handle all this logic when there’s a component that explicitly provides the visual part. When you add a Login Gate into your project, you’ll see a group of settings in the Inspector: The Login Gate has two modes, “Simple password” and “Web service”. The idea here is that you can start out prototyping the login using the simple password mode, and later seamlessly switch to a real implementation using a web service plugin. For either mode, you can specify two paths within the app. “When unlocked” lets you pick the screen that will be shown if login is successful, and the “If unlock fails” of course lets you provide a screen for the other case. For simple password mode, you can just write in a password here in the Inspector. You also need to do two things to make the login work: Place a text field in the screen and select it as the “Password field” in the Login Gate’s inspector. Place a button in the screen. In the Interaction tap, make it trigger the “Unlock gate” in its containing component, like this: That’s all! Now you have a minimally functional login that checks the password. It’s very important to note that you shouldn’t use the “Simple password” mode for production apps. The password is written into the web app’s JavaScript code in plain text, so it’s easy to hack: someone could just read the source code loaded in the browser to find the password. (The JS code that gets loaded in the browser is “minified” which makes it harder to read, but that doesn’t provide enough protection because the password is still out in the open.) So, if you need a real secure login, you probably already have a back-end system that implements one. At that point you can implement a web service plugin for React Studio that implements the login check by calling to your back-end API. The plugin implementation is actually quite simple. React Studio plugins are just JavaScript programs with “hooks” that let you write whatever code you need in a specific context. In this case, your plugin needs to implement the “write code for login” hook, and it will get called to write all the necessary code for login. Your code can do whatever it likes — HTTP requests, state updates — as long as it finishes by calling either one of the “success” or “failure” callbacks that are provided by React Studio. There’s more information about plugins in the section “ below. Plugin API improvements” Update and delete within data sheets (a.k.a. full “CRUD”) Data sheets in React Studio are like mini-databases that are easy to create, modify and link to a real back-end using plugins. Previously they were missing some methods that would allow a plugin to do all the necessary actions for a typical app. Now you can create UI interactions for the full range of essential database actions: Create, Read, Update, Delete (“CRUD”). Let’s take a quick look at how these work: means adding a row to a data sheet. You can do this using the interaction. The values for the row can be collected either from elements within the component or from data slots. The “Update data” UI lets you pick this for each column: “Create” Update data happens automatically when a data sheet is accessed. When the data sheet is connected to a web service plugin, the plugin can also do regular “live” updates to get the latest data in realtime.Reads also happen if your data sheet is using a dynamic query parameter that depends on a data slot, and that data slot’s value changes. See below in the section for further explanation. “Read” “Query setting for data sheets” works the same way as Create: you just do an Update Data interaction. How does React Studio determine whether it should be a create or update? If there is an , the action will be an update. In practice this happens when a component is used within a list, or a screen acts as a “details view” (properties were carried from a list item). If there isn’t a data sheet row context, the interaction will add a row instead.This is admittedly confusing because you need to know about the context of the interaction to understand what it actually does. We’re going to rework this part of React Studio to make it easier to understand and control. “Update” active data sheet row context is similar to Update in that it requires an active data sheet row context. If there is one, you can use the Delete action on the containing component to delete that data sheet row: “Delete” If your data sheet isn’t linked to a plugin, these actions will simply work on temporary data in the browser’s memory. You can add rows, modify them, delete them, but these changes don’t get registered anywhere. Once the browser window closes, the data is gone. To get persistence, you need to use a web service plugin. We provide a few baselines: can be used with any server that provides a plain “REST-style” web API, and can be used with the spreadsheet app in Google Docs. Generic JSON Google Sheets The latter is actually a really convenient way to create simple back-ends for apps that don’t need a full-fledged database. For example, if you have a bunch of items in a dynamic list that needs to be updated by someone with no tech skills, the Google Sheets plugin is often a great choice — very simple to set up and update. Similarly, if you need to collect data and store the responses in a spreadsheet for easy access, the Sheets plugin works great for that. An example use case might be something like a feedback/interview app that’s deployed to a limited audience: whe nyou know there are only going to be something like a few hundred respondents at maximum, it’s not worth building a real back-end to collect that data. Query setting for data sheets This new feature was briefly mentioned above in the context of data sheet “read” actions. Queries offer a way to make a data sheet depend on the contents of one or more data slots. That sounds rather abstract and complicated, so let’s look at a practical example. Imagine your web app is localized in multiple languages. Thanks to React Studio’s built-in localization support, you’re able to easily create the various language versions for the texts within the UI. But what about content that comes from an external source? Let’s say your app is a city guide. There is a list of recommended restaurants that is loaded from a web API provided by a third party. This API offers its content in several languages, but to load the correct content, you need to pass the selected language to the API somehow. Data sheet queries to the rescue! The current language is stored in a data slot named “ds_activeLang”. (This is the default setup in React Studio.) To pass it to a web service plugin that loads the contents of a data sheet, you’d enter the following in the “Query” field for that data sheet: language=$slot(‘ds_activeLang’) This value will be passed to the plugin at runtime by React Studio, and the query expression will be parsed so that the end result will be: (assuming the current language is English). The plugin can then send this value as part of its data requests. For a typical API, the query is included in the URL, although it’s really up to the plugin — it can use the query value in whatever way it sees fit. React Studio places very few limitations to what plugins can do in the generated projects. language=en Data linkage improvements There are now more content types that can be linked to properties. For example, the FauxAir Check-in example project mentioned above ( ) has a button that is enabled based on the value of a data slot: link to demo There’s a contract text that the user must accept to proceed. When the user clicks the “I agree” checkbox, the value of the data slot gets set to true, and in turn this button gets enabled. This is a very typical case where you require the user to perform one action before she can perform another. “ds_userAcceptedContract” List/grid using component states (= making a grid without a data sheet) Typically you want to use a data sheet to provide the content for a list or grid. However, there’s another common use case where the list/grid doesn’t need dynamic content, and it would be easier to just create different versions of content for each “cell” of the grid. A new setting available in the List/Grid element lets you do that. It’s called . That name is a real mouthful, but what it does is simple enough. “Take item values from nested component’s states” Imagine your app has six different sections, and a main menu that lets the user navigate to those sections. You’d like for the main menu to be a responsive grid of icons — sort of like a phone home screen. Making a data sheet to provide content for six unchanging icons feels like overkill. Instead, you can now make a component with six states. (The component’s size should match one icon, and then you’ll have each icon as a separate state.) In the main menu screen, add a List/Grid element, set your six-state component as its list item component, and toggle the “Item component’s states” setting. You’ll now have a grid of six items easily editable as states within one component. Plugin API improvements The plugin interfaces have seen a lot of updates in this release. Here’s an overview of new plugin hooks and methods. If you need help writing plugins, we’re happy to help you out! Just write to or reach out on the React Studio Slack channel (instructions for joining are in the app’s Welcome screen available through the Help menu). hello@neonto.com Web service plugins / handling login.New hook functions: writeReactWebCodeForLogin, writeReactWebCodeForStoredLoginCheck, writeReactWebCodeForLogout Web service plugins / handling data sheet add/update/delete actions.New hook functions:_writeReactWebCodeForDataSheetInit, writeReactWebCodeForDataSheetWriteOverrides_(See the “Generic JSON” plugin for a practical example) App service plugins / writing custom data into the generated project’s main HTML file.New hook function: getReactWebHtmlHeadElements All plugins / writing custom assets into the generated project’s “public” folder.New method on the object passed to hook functions: exporter exporter.writeCustomAsset() CSS fixes Generating correct CSS is a real bugbear. We’ve fixed a number of bugs related to centering, proportional positioning, etc. If you find anything that doesn’t display correctly, please let us know at hello@neonto.com Miscellaneous fixes This release again includes a lot of smaller fixes to exporter bugs and UI inconveniences. We’ve tweaked the Project Map, fixed issues with Columns element’s weight values not saving correctly, resolved many instances where wrong characters in e.g. element names might cause invalid code to be generated, fixed issues with line breaks not displaying correctly within text elements, and more. End surprise… Thanks for reading all this way! If you’re interested in React Studio, we have a genuinely unique opportunity right now. Neonto, the company that develops React Studio, is raising a small amount of funds with a stock offering this week. It’s primarily directed at existing stockholders, but we would very much like to have some React Studio users participating as well. So, would you like to ? Now is the chance. The company is based in Finland, so you don’t need to be an accredited investor (however you do need to provide a photo of your government-issued personal identification to be registered as a stockholder). The minimum investment is 390 euros, or about 480 USD. Binding commitments are required by .Please contact for further details. become an owner in React Studio Thursday 1st March 2018 hello@neonto.com

React Studio 1.4: Tackling large projects with Map Sections, logins, “CRUD” data, and more

Authors:
(1) Florian Cafiero (ORCID 0000-0002-1951-6942), Sciences Po, Medialab;
(2) Jean-Baptiste Camps (ORCID 0000-0003-0385-7037), Ecole nationale des chartes, Universite Paris, Sciences & Lettres. Table of Links Abstract and Introduction Why work on QAnon? Specificities and social impact Who is Q? The theories put to test Authorship attribution Results Discussion Corpus constitution Quotes of authors outside of the corpus have been Definition of two subcorpus: dealing with generic difference and an imbalanced dataset The genre of “Q drops”: a methodological challenge Detecting style changes: rolling stylometry Ethical statement, Acknowledgements, and References Detecting style changes: rolling stylometry Collaborative writing is not necessarily easy to handle. The scenario in which authors simply took turns, and divided the work between themselves is already complicated to address. But when the collaboration is more complex, especially when the various authors contribute together to the same passages, the style of the original authors can be hard to recognize. The collaboration then results in a new style, that does not match the style of one of the authors (Kestemont et al., 2015). The principle of rolling stylometry (Eder, 2016) is simple: rather than attributing a whole text, we arbitrarily decompose it in a series of overlapping smaller parts: from the 1st word to the 1001th word, from the 2nd word to the 1002nd word etc. Then, we attribute each of these parts to a certain author. We only have to define the length of these parts, and by how much they overlap. Rolling stylometry has been successfully implemented in a wide variety of settings. With Burrows’ delta, it has for instance been used to assess Ford’s claims about his implications in collaborations worth Joseph Conrad (Rybicki et al., 2014), to determine the beginning of Vostaert’s intervention on Dutch Arthurian novel Roman van Walewein (van Dalen-Oskam and Van Zundert, 2007), or to understand Lovecraft’s and Eddy’s implication in The Loved Dead (Gladwin et al., 2017). Using support-vector machines, rolling stylometry more recently helped to confirm Fletcher and Shakespeare’s collaboration for Henry VIII (Plecha´ˇc, 2020) or Moli`ere and Corneille’s collaboration for Psych´e (Cafiero and Camps, 2021). Support Vector Machine We choose to train linear Support Vector Classifiers (SVC) to identify the style of each potential candidate. The family of Support Vector Machines algorithm has been widely and successfully used for authorship attribution in a variety of settings and languages, and for very diverse sources, ranging from e-mails or blogs to Shakespeare plays (De Vel et al., 2001; Diederich et al., 2003; Mikros, 2012; Ouamour and Sayoud, 2012; Marukatat et al., 2014; Plecha´ˇc, 2020). At the PAN competition, a reference for digital text forensics and stylometry, it also served as a baseline for the “cross-domain authorship attribution” tasks the last time they were proposed in 2018 (Stamatatos et al., 2018) and 2019 (Kestemont et al., 2019). The Q drops being a sort of “domain” in their own, our own task can be considered as a crossdomain authorship attribution task. Other classifiers could have been used, but would not offer the interpretability we need for that kind of task. In such a delicate context, being able to get a simple and clear intuition on which features the classifier relies is simply crucial. If the features selected were to be too much related to the content of the texts, rather than to properly linguistic properties of the person’s discourse, we need to be aware of it, and train more properly a new classifier. To determine the choice of features and the size of the training samples, we are constrained by two antagonistic goals: the shorter the samples, the more detailed and precise the results that we will get in terms of attribution, yet the longer, the more statistically reliable. Particularly, authorship attribution has proven to require relatively high amounts of data, with a floor for reliable authorship attribution between 1000 and 3000 words, depending on genre and language (Eder, 2015, 2017). The question of sample length can also be linked to the difficulty of the attribution task; cross-domain attribution with multiple candidates presents a challenge in this regard. On the other hand, the features we retained, character 3-grams, could increase robustness, ad they are known to reduce sparsity and perform well in attribution studies (Kestemont, 2014; Sapkota et al., 2015). While punctuation can strongly reflect authorial signature (Sapkota et al., 2015), we removed led to remove it because of variety of platforms from where the data were recovered could cause inconsistencies in the use of signs that can be encoded in different fashions, e.g., apostrophes. For these reasons, we retain a setup that is a compromise between reliability and finer grain analysis: Sample length 1000 words; features character 3-grams (all, including punctuation). To evaluate our setups, we opt for a leave-one-out cross evaluation on the training corpus (Table 1), for a combination of reasons. First, some of the samples are of relatively small size in the context of training an efficient classifier. A leave-one-out procedure helps us using the maximal amount of data. The constitution of which relevant and coherent training sets is also a question in our case. In that context, leave-one-out evaluation can provide a more robust estimate of the model’s performance, as it accounts for the potential variability caused by the specific sample chosen for testing. Finally, leave-one-out can help us avoid overfitting (Ng et al., 1997; Ghojogh and Crowley, 2019), which is a danger in our study - we do not want the classification to be based on some specific piece of news of particular interest to one of the candidates, but to have reliable classification based on linguistic features appearing in a large range of contexts. The confusion matrix gives more information on the nature of the small number of classification errors (Table 2). As can be expected, performance is slightly lower for authors for which training material is very limited (ColemanR, CourtneyT, RogerS). For the others, it is above 95%, if we except a few confusions between Michael F. and Roger S. (on the large corpus only). These can be explained by the limited size of Roger S. training data, thematic attractions, the fact that he talks about Flynn more or less directly, but also probably by a generational (age) bias. We then apply our models to all successive overlapping slices of Q drops, arranged in chronological order, with window size of length 1000 words and step 200 words. We then plot the resulting decision functions for each classifier. The higher the value, the more likely the attribution of a sample to a given author. All analyses are implemented in Python, inside the SuperStyl package (Camps et al., 2021), and use internally the SVM and pipeline facilities provided by Sklearn (Pedregosa et al., 2011). Plots are created using R and Python (matplotlib). Correspondence analysis Correspondence analysis (Benz´ecri, 1973) has been performed on a contingency table of the 1000 words samples by RonW and PaulF based on 4105 character 3-grams selected for statistically reliability based on a procedure previously described (Cafiero and Camps, 2019). The QDrops have then been inserted as supplementary rows, using the implementation in the R package FactoMineR (Lˆe et al., 2008). The significance of the two first axes remains relatively low, in part because of the high dimensionality of the input table, but the data clouds for PaulF and RonW are clearly separated, and the QDrops appear in an intermediary position. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Florian Cafiero (ORCID 0000-0002-1951-6942), Sciences Po, Medialab; (2) Jean-Baptiste Camps (ORCID 0000-0003-0385-7037), Ecole nationale des chartes, Universite Paris, Sciences & Lettres. Authors: Authors: (1) Florian Cafiero (ORCID 0000-0002-1951-6942), Sciences Po, Medialab; (2) Jean-Baptiste Camps (ORCID 0000-0003-0385-7037), Ecole nationale des chartes, Universite Paris, Sciences & Lettres. Table of Links Abstract and Introduction Abstract and Introduction Why work on QAnon? Specificities and social impact Why work on QAnon? Specificities and social impact Who is Q? The theories put to test Who is Q? The theories put to test Authorship attribution Authorship attribution Results Results Discussion Discussion Corpus constitution Corpus constitution Quotes of authors outside of the corpus have been Quotes of authors outside of the corpus have been Definition of two subcorpus: dealing with generic difference and an imbalanced dataset Definition of two subcorpus: dealing with generic difference and an imbalanced dataset The genre of “Q drops”: a methodological challenge The genre of “Q drops”: a methodological challenge Detecting style changes: rolling stylometry Detecting style changes: rolling stylometry Ethical statement, Acknowledgements, and References Ethical statement, Acknowledgements, and References Detecting style changes: rolling stylometry Collaborative writing is not necessarily easy to handle. The scenario in which authors simply took turns, and divided the work between themselves is already complicated to address. But when the collaboration is more complex, especially when the various authors contribute together to the same passages, the style of the original authors can be hard to recognize. The collaboration then results in a new style, that does not match the style of one of the authors (Kestemont et al., 2015). The principle of rolling stylometry (Eder, 2016) is simple: rather than attributing a whole text, we arbitrarily decompose it in a series of overlapping smaller parts: from the 1st word to the 1001th word, from the 2nd word to the 1002nd word etc. Then, we attribute each of these parts to a certain author. We only have to define the length of these parts, and by how much they overlap. Rolling stylometry has been successfully implemented in a wide variety of settings. With Burrows’ delta, it has for instance been used to assess Ford’s claims about his implications in collaborations worth Joseph Conrad (Rybicki et al., 2014), to determine the beginning of Vostaert’s intervention on Dutch Arthurian novel Roman van Walewein (van Dalen-Oskam and Van Zundert, 2007), or to understand Lovecraft’s and Eddy’s implication in The Loved Dead (Gladwin et al., 2017). Using support-vector machines, rolling stylometry more recently helped to confirm Fletcher and Shakespeare’s collaboration for Henry VIII (Plecha´ˇc, 2020) or Moli`ere and Corneille’s collaboration for Psych´e (Cafiero and Camps, 2021). Support Vector Machine We choose to train linear Support Vector Classifiers (SVC) to identify the style of each potential candidate. The family of Support Vector Machines algorithm has been widely and successfully used for authorship attribution in a variety of settings and languages, and for very diverse sources, ranging from e-mails or blogs to Shakespeare plays (De Vel et al., 2001; Diederich et al., 2003; Mikros, 2012; Ouamour and Sayoud, 2012; Marukatat et al., 2014; Plecha´ˇc, 2020). At the PAN competition, a reference for digital text forensics and stylometry, it also served as a baseline for the “cross-domain authorship attribution” tasks the last time they were proposed in 2018 (Stamatatos et al., 2018) and 2019 (Kestemont et al., 2019). The Q drops being a sort of “domain” in their own, our own task can be considered as a crossdomain authorship attribution task. Other classifiers could have been used, but would not offer the interpretability we need for that kind of task. In such a delicate context, being able to get a simple and clear intuition on which features the classifier relies is simply crucial. If the features selected were to be too much related to the content of the texts, rather than to properly linguistic properties of the person’s discourse, we need to be aware of it, and train more properly a new classifier. To determine the choice of features and the size of the training samples, we are constrained by two antagonistic goals: the shorter the samples, the more detailed and precise the results that we will get in terms of attribution, yet the longer, the more statistically reliable. Particularly, authorship attribution has proven to require relatively high amounts of data, with a floor for reliable authorship attribution between 1000 and 3000 words, depending on genre and language (Eder, 2015, 2017). The question of sample length can also be linked to the difficulty of the attribution task; cross-domain attribution with multiple candidates presents a challenge in this regard. On the other hand, the features we retained, character 3-grams, could increase robustness, ad they are known to reduce sparsity and perform well in attribution studies (Kestemont, 2014; Sapkota et al., 2015). While punctuation can strongly reflect authorial signature (Sapkota et al., 2015), we removed led to remove it because of variety of platforms from where the data were recovered could cause inconsistencies in the use of signs that can be encoded in different fashions, e.g., apostrophes. For these reasons, we retain a setup that is a compromise between reliability and finer grain analysis: Sample length 1000 words; Sample length features character 3-grams (all, including punctuation). features To evaluate our setups, we opt for a leave-one-out cross evaluation on the training corpus (Table 1), for a combination of reasons. First, some of the samples are of relatively small size in the context of training an efficient classifier. A leave-one-out procedure helps us using the maximal amount of data. The constitution of which relevant and coherent training sets is also a question in our case. In that context, leave-one-out evaluation can provide a more robust estimate of the model’s performance, as it accounts for the potential variability caused by the specific sample chosen for testing. Finally, leave-one-out can help us avoid overfitting (Ng et al., 1997; Ghojogh and Crowley, 2019), which is a danger in our study - we do not want the classification to be based on some specific piece of news of particular interest to one of the candidates, but to have reliable classification based on linguistic features appearing in a large range of contexts. The confusion matrix gives more information on the nature of the small number of classification errors (Table 2). As can be expected, performance is slightly lower for authors for which training material is very limited (ColemanR, CourtneyT, RogerS). For the others, it is above 95%, if we except a few confusions between Michael F. and Roger S. (on the large corpus only). These can be explained by the limited size of Roger S. training data, thematic attractions, the fact that he talks about Flynn more or less directly, but also probably by a generational (age) bias. We then apply our models to all successive overlapping slices of Q drops, arranged in chronological order, with window size of length 1000 words and step 200 words. We then plot the resulting decision functions for each classifier. The higher the value, the more likely the attribution of a sample to a given author. All analyses are implemented in Python, inside the SuperStyl package (Camps et al., 2021), and use internally the SVM and pipeline facilities provided by Sklearn (Pedregosa et al., 2011). Plots are created using R and Python (matplotlib). Correspondence analysis Correspondence analysis (Benz´ecri, 1973) has been performed on a contingency table of the 1000 words samples by RonW and PaulF based on 4105 character 3-grams selected for statistically reliability based on a procedure previously described (Cafiero and Camps, 2019). The QDrops have then been inserted as supplementary rows, using the implementation in the R package FactoMineR (Lˆe et al., 2008). The significance of the two first axes remains relatively low, in part because of the high dimensionality of the input table, but the data clouds for PaulF and RonW are clearly separated, and the QDrops appear in an intermediary position. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Rolling Stylometry and Machine Learning Analyzes  QAnon Texts Patterns

Authors:
(1) Florian Cafiero (ORCID 0000-0002-1951-6942), Sciences Po, Medialab;
(2) Jean-Baptiste Camps (ORCID 0000-0003-0385-7037), Ecole nationale des chartes, Universite Paris, Sciences & Lettres. Table of Links Abstract and Introduction Why work on QAnon? Specificities and social impact Who is Q? The theories put to test Authorship attribution Results Discussion Corpus constitution Quotes of authors outside of the corpus have been Definition of two subcorpus: dealing with generic difference and an imbalanced dataset The genre of “Q drops”: a methodological challenge Detecting style changes: rolling stylometry Ethical statement, Acknowledgements, and References The genre of “Q drops”: a methodological challenge The study of the Q drops raises a number of specific challenges. First of all, the Q drops constitute per se a kind of a genre. It follows specific rules that most people would not use in another context: they do not look like a regular blog, media or social media post, nor do they belong to any specific literary genre etc. This forces us to consider our attribution problem as a cross-topic attribution problem. This specific genre has consequences on many linguistic properties of the Q drops. The structural brevity of the sentences for instance prevents us from taking the sentence lengths as a clue of who wrote what. The overwhelming proportion of interrogative forms, especially in the first Q drops, makes it difficult to reason on morpho-syntactic sequences, as they are often extremely and artificially stereotyped. Part-of-Speech n-grams such as “Interrogative pronoun - conjugated verb - common noun” would for instance emerge as a signature of the Q drops, and could derail our analysis, by pointing to any of the suspects using interrogative forms the most in other contexts. Finally, the elliptic style of the Q drops, written almost as if they were a telegram, distorts the use of function words, less frequent than expected in the Q drops corpus. Approaches relying only on function words could be made less robust by this distortion. We thus chose to work on character trigrams, the most flexible and reliable feature we could use in the very specific context of this study, and a widely acknowledge feature in stylometry, in particulary for its supposed capacity to capture grammatical morphemes (Kestemont, 2014; Sapkota et al., 2015) while bearing in mind potential greater sensibility to thematic attractions in comparison to function words. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Florian Cafiero (ORCID 0000-0002-1951-6942), Sciences Po, Medialab; (2) Jean-Baptiste Camps (ORCID 0000-0003-0385-7037), Ecole nationale des chartes, Universite Paris, Sciences & Lettres. Authors: Authors: (1) Florian Cafiero (ORCID 0000-0002-1951-6942), Sciences Po, Medialab; (2) Jean-Baptiste Camps (ORCID 0000-0003-0385-7037), Ecole nationale des chartes, Universite Paris, Sciences & Lettres. Table of Links Abstract and Introduction Abstract and Introduction Why work on QAnon? Specificities and social impact Why work on QAnon? Specificities and social impact Who is Q? The theories put to test Who is Q? The theories put to test Authorship attribution Authorship attribution Results Results Discussion Discussion Corpus constitution Corpus constitution Quotes of authors outside of the corpus have been Quotes of authors outside of the corpus have been Definition of two subcorpus: dealing with generic difference and an imbalanced dataset Definition of two subcorpus: dealing with generic difference and an imbalanced dataset The genre of “Q drops”: a methodological challenge The genre of “Q drops”: a methodological challenge Detecting style changes: rolling stylometry Detecting style changes: rolling stylometry Ethical statement, Acknowledgements, and References Ethical statement, Acknowledgements, and References The genre of “Q drops”: a methodological challenge The study of the Q drops raises a number of specific challenges. First of all, the Q drops constitute per se a kind of a genre. It follows specific rules that most people would not use in another context: they do not look like a regular blog, media or social media post, nor do they belong to any specific literary genre etc. This forces us to consider our attribution problem as a cross-topic attribution problem. This specific genre has consequences on many linguistic properties of the Q drops. The structural brevity of the sentences for instance prevents us from taking the sentence lengths as a clue of who wrote what. The overwhelming proportion of interrogative forms, especially in the first Q drops, makes it difficult to reason on morpho-syntactic sequences, as they are often extremely and artificially stereotyped. Part-of-Speech n-grams such as “Interrogative pronoun - conjugated verb - common noun” would for instance emerge as a signature of the Q drops, and could derail our analysis, by pointing to any of the suspects using interrogative forms the most in other contexts. Finally, the elliptic style of the Q drops, written almost as if they were a telegram, distorts the use of function words, less frequent than expected in the Q drops corpus. Approaches relying only on function words could be made less robust by this distortion. We thus chose to work on character trigrams, the most flexible and reliable feature we could use in the very specific context of this study, and a widely acknowledge feature in stylometry, in particulary for its supposed capacity to capture grammatical morphemes (Kestemont, 2014; Sapkota et al., 2015) while bearing in mind potential greater sensibility to thematic attractions in comparison to function words. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Stories, News & Mentions about TEXTUS

Textus’s Stories on HackerNoon

Textus's News & Mentions