jalebi

THE ELEMENTS OF DREAM LIFE

At my startup, we’ve made a decision that may seem odd in this day and age. Especially for a technology company that uses data science to power its offering. We’re spending countless hours getting to know our customers on a personal level. And by “personal,” I mean . personal The reason? While aggregated customer data delivers important insights, I’ve seen companies rely increasingly — even exclusively — on numbers. If this trend continues, we’re at risk of forfeiting something vital: understanding customers as . people Data analysis is perfect for telling you WHAT customers are doing. It can even predict what they’ll do in the future. But it doesn’t tell you they’re doing it. WHY Heatmaps to show you where users linger on a webpage, A/B tests to compare conversion rates, and social media analytics to zero in on the ideal targeting. The numbers have a precision to them and it’s exciting when they tell a compelling story. Except that the story is not complete. Because human desires are not — and never will be — entirely quantifiable. So, my co-founder and I decided to talk to every customer willing to take the time. This would help us get inside her head — as well as guide what data we should be collecting, measuring and optimizing for. We’ve done in-depth interviews with more than 100 women in less than a year. Time consuming? Yes. Time well spent? Nothing could have been more important. Our decisions are profoundly influenced by a veritable mountain of learning in the form of opinions, feelings, concerns, and insights. Every company can gather “human-driven insights.” Here’s how: 1. Talk live to customers, or prospective customers, at least twice a week Phone interviews are just one format. You can, for example, conduct research in their workplaces, in their homes, in shopping malls, and by bringing them together in your home. Just be sure you’re in a position to listen and learn. 2. Identify WHAT will change your actions Before crafting your line of questioning, write down what will change based on what you learn. Will it affect your offering? Your positioning? Your pricing? Whether or not to introduce a new product? This upfront thinking can be hard work. But if you don’t do it, you run the risk of learning “nice-to-knows” that don’t inform next steps. 3. Clarify WHOSE feedback will drive your actions Talking to the right people is critical. And who is right depends on your business. Whom you talk to might turn on whether or not they have pets, are within a certain age range, influence technology decisions for their company — or any other criteria that separate the people you most need to understand from everyone else. Start with qualifying questions (example: “Do you have a pet?”) and politely end the interview if the person doesn’t meet your criteria. 4. Write down your hypotheses Articulate your current beliefs or hypotheses. This will help you design the learning to either challenge or validate these assumptions. Examples: “Women who consume organic food are predisposed to embrace non-toxic beauty products,” or “Prospective customers believe a higher price equates to higher quality.” 5. The one thing you should write out is your questions not Sound counterintuitive? Hang with me here. This is , where you’ll probe deeply into any nuggets that arise. qualitative research Of course you should outline the topics you’ll cover in advance. You do need to control the conversation, make sure you touch on key topics and gently reorient her if she’s going down a rathole. But don’t read out pre-canned questions like a pollster. Listen carefully, follow where she takes you and . What you get back will be richer. pull the learnings out of a conversation 6. Be dispassionate It’s insanely hard, but you have to do it. This is because the cards are stacked against the truth. People generally want to be nice and to tell you want you want to hear. So, for heaven’s sakes, don’t let them. Explain that you are, for now, the research department and nothing she says can hurt your feelings. That the biggest gift she can give you is honest feedback, including criticism and negative opinions. She will still fail to give you the full truth if you do any of the following: Appear happy with positive comments Seem disappointed with negative comments Indicate, however subtly, what you hope to hear Attempt to solve her problems Indicate that she’s misunderstood your company/offering Explain how things will be fixed in the future Just be deeply curious about what she really thinks. You’re her therapist and have no opinion. Her perception of reality is the only thing that matters. Make sure you fully understand it. You can celebrate, complain, or adopt any other manner of emotion AFTER she’s out of earshot. In the meantime, keep your eye on the prize: getting inside your customer’s DNA so you really know her. And so you can create those rare, loyalty-inducing experiences where she feels truly and deeply understood.

Your Customer Is Not a Statistic.

Jaleh Bisharat Instagram This post is the third in a series about co-founding a company. When we were raising money for our clean beauty startup, , we got plenty of advice from fellow entrepreneurs. NakedPoppy One said, “Whatever you do, do not send your deck in advance.” Her reasoning? Investors love to look smart. One way they achieve this is by shooting rapid-fire holes in your plan. This is especially easy when you are two women, two laptops and an idea. So, the best way to control the situation, according to my friend’s reasoning, is to not give them fodder they can mull over in advance. For us, there was a simpler reason we never sent our deck in advance. We didn’t have one. I had become a fan of the , for all the same reasons Amazon believes in them. When you write, you are forced to be crisp. Words are precise. You cannot fool yourself or dance around your thoughts. Amazon-style six-page memo In other words, no fuzzy thinking or in-the-moment interpretation, as is possible with slides. For readers, the format is usually magical. In 15 minutes they can inhale what might take four times as long to digest with a live presentation. Without interruption, they can see precisely how your logic flows. What’s the customer problem? What’s the solution? What’s your differentiation? How will you make it work? You can then have a fully informed discussion at the end. So, that’s what we did. We summarized our entire vision in less than six pages. The memo began as something we carefully crafted for ourselves. We genuinely wanted to be sure our dissection of the customer problem and our solution hung together. Where there were holes in our thinking, we talked to prospective customers — online and in shopping malls and farmers markets. We iterated, challenged ourselves and iterated some more. I dared to show the early draft to a mentor, Thomas Layton, for his feedback. Two reactions ensued: He became an angel investor. He advised us to not waste time translating the material to a pitch deck. “Go contrarian. It’s a better format anyway,” he said. Well, not everyone agreed. All of the following happened to us: A Palo Alto venture capitalist asked for our presentation before the meeting. We said we didn’t have one. Because we didn’t. . VC disappeared In Menlo Park, a venture capitalist said, “You are trying to change the way we operate, but let me tell you something: People don’t like change.” In Silicon Valley, investors eat disruption for breakfast, lunch and dinner so we were rather disappointed. VC did not invest. Another Menlo Park venture capitalist cringed. “I’d like to invest, but please, for the love of God, create a slide deck to share with my partners. I will even sit with you and help you craft it.” By then, we’d become convinced that decks were a waste of time and that VCs should like co-founders who don’t waste time. We said no and held our breath. And held it. And held it. We think of ourselves as pretty humble, but what if we’d overstepped? VC convinced his partners to invest. An East Coast venture capitalist said, “I wasn’t interested the first time I read your memo. But, I read it a second time, and a third time, and it grew on me. Anybody this precise in their writing deserves consideration.” VC invested. We know that taking a different approach was risky. But, it was what we believed in, and we didn’t want to change our tune. It’s true that a written document doesn’t work in every context (think: large audiences). But, in small group settings, where details matter, there’s really nothing like it. For more on our journey with NakedPoppy, read the first and second posts in this series. This post originally appeared on Entrepreneur .

We Ditched the Investor Deck. And It Worked.

A good product or service is wonderful, but great talent is your life force. You’ve taken the plunge and started a business. You’ve never done anything so exciting — or difficult — but customers are rolling in, and the business is starting to grow. You’re now building out what’s arguably the most important asset you’ll ever have: your team. After all, a good product or service is wonderful, but great talent is your life force. show that an engaged workforce leads to increased productivity and decreased turnover — and can spell the difference between a business’s success or failure. Every minute your people spend feeling unclear, unfocused or unhappy is a moment that could otherwise be devoted to customers. Studies It doesn’t happen by accident. I know this from having built and nurtured teams (sometimes from scratch) at many companies, including several startups as well as OpenTable, Upwork and Amazon. Here are some lessons I learned about how to craft high-performance teams: 1. Know what you stand for. Articulate it often. If you were to ask everyone at your company what word they’d use to describe the culture, would they all choose the word? Is it the word you want? It’s never too late to proactively define your culture and give it purpose. same At my startup, NakedPoppy, we’ve explicitly chosen to build a culture of joy. Part of this is because, in my long career, I’ve consistently seen that when team members feel upbeat and cared for, they dig deep and do their best work. The other part is because it’s authentic; my co-founder and I both love to be happy. And our brand is all about injecting joy into customers’ lives. It isn’t easy to serve customers so well that they’d credit us with bringing them joy. So, this cultural value plays another important role: It sets a high bar for attitude and for performance. 2. Master the art of motivational goal setting. Start by painting an inspiring vision and then communicate how each person has an impact. For example, if you are running a chain of blow-dry bars, you might tell your team that they’re empowering women and giving them confidence, one beautiful hairstyle at a time. Follow up with written goals for each team member. The act of writing forces you to work out any kinks in your thinking and to be crisp. It leaves no room for ambiguity, so they can focus on results, not on second-guessing what you mean. Finally, don’t forget to spell out the behaviors you value. When people see concrete guidelines like “exudes positive energy,” “goes the extra mile for customers” or “is respected as a team player,” they have a way of adopting those behaviors. Working on set as a team. From my Instagram 3. Model your obsession with customers. This one doesn’t need much elaboration. People follow their leaders. If you’re constantly listening to your customers, learning from them and going to the ends of the earth for them, your team will too. 4. Turn mistakes into opportunities. It goes without saying that you should celebrate exceptional work. If the appreciation is genuine, the team can tell, and it begets even more good work. But what about those mistakes? Those terrible mistakes that make people swing between feeling upset, defensive and demoralized? One way to make the situation is worse is to tell them what they did wrong, especially when you’re frustrated. Instead, collect yourself and ask, “What did we learn?” You might offer time to reflect before responding. By putting your team members in charge of their own learning, you have empowered them, respected them and freed them to get better. It’s a rare person who will not feel grateful and work harder as a result. 5. Do the tough stuff swiftly. You can create a wonderful culture, but not everyone will shine. Preserve it by nipping problems in the bud. If someone tends to dominate, take credit for other people’s work or behave in other ways that damage the culture, for example, confront them privately. Give them a chance to improve, but keep a close eye. Trust me, the team senses it all. And if, despite every effort, someone doesn’t improve their performance, don’t delay. The kindest thing you can do is to gently move them along. It’s better for everyone — and your team will also understand that their excellent performance is truly valued. Letting go of a poor match is also often better for the team member in question. A few years ago, I had a talented employee who seemed incapable of bringing her full self to work every day. After a couple of weeks that began with asking her how she was feeling about her job, it was no surprise when I asked her to move on. Her small team was pleased to get a new supervisor. And, she got a clean start at another job where she’s now very successful. 6. Make use of the two most empowering words in a manager’s vocabulary: “You decide.” This does not mean you never step in, especially when stakes are high. In general, though, if you err on the side of giving people your trust and confidence, they will feel empowered, learn more rapidly and do their best work. 7. Stay calm. This matters. Every business has those moments. A competitor makes a threatening announcement. A product falls short. You’ve been there, right? Well, keep your doubts to yourself. uh oh — we’re dead Productivity can be lost during these storms. Either you can join the distraction, or you can show a steady hand. This is a marathon and not a sprint. Nothing is ever as bad as it seems. A great team can turn a challenge into an even better business — and go on to win. How can you be sure you’ve created a high-performing team? One litmus test is what happens when you’re not there. Do they slow down? Await your instructions? Or do they power ahead, and perhaps do an even better job than you would? The reality is when people know what they’re striving for, feel motivated, empowered, respected and valued, you free them up to do the best work of their lives. And you know you’ve created something truly extraordinary if you ever hear them say, “It doesn’t feel like work!” This post originally appeared in Entrepreneur .

Does Your Team Love Coming to Work? Follow These 7 Strategies to Make Sure.

Angular 8 is finally here (in May)! Software developers and Angular community were eagerly waiting for the new release of Angular to happen with quite high expectations. The new version looks promising and is majorly focused on making it easier for developers to build different kinds of apps, as well as improving the performance. It is a major release that will bring changes to the entire platform, including the material, command line interface (CLI), and framework. Angular 8 is much faster and lighter than its previous versions, delivers new APIs, and aligns better with modern web standards and ecosystem. Before getting deep into the new features and improvements, let’s have a look at its previous versions. Angular at a glance Angular is one of the most used JavaScript-based front-end web frameworks in the market. Maintained by Google and a community of organizations, it was first launched in 2009. It wasn’t that innovative in the beginning because of large size, lack of faster performance, and other technical issues. However, the team behind it worked hard and revamped the framework completely. They used powerful libraries and added features like tree-shaking, and ahead-of-time compiling. featured Typescript which enabled the framework to display the errors in HTML templates. brought view-engine improvements and reduced code generation for creating apps. Angular 2 Angular 4 And then there was no stopping for the Angular. It kept on evolving with time, introducing new features, and making things easier and simpler for developers. Last year in October, the was released. And now, the latest version — Angular v8 is available. Angular 7 Features of Angular 8 and improvements 1. Differential Loading by default Angular will now be able to take advantage of differential loading by default. This will enable the browsers to select the modern or traditional JavaScript bundles of the app on the basis of the capabilities of the app. What this means that the bundle of JS will be loaded accordingly when the app is used. This has been made possible because the new version will feature separate JS bundles for modern (es2015) and traditional (es5) builds. As a result, the loading time and speed of the app will be faster. 2. Improved support from code editors The route configurations in Angular will now use dynamic imports, which will improve the support from code editors such as Visual Studio Code and WebStorm. These editors will be able to understand and validate the imports automatically. What Angular team has done is updated the loadChildren key in the route configuration. Previously, this key was custom to Angular, and now it has been moved to industry standard dynamic imports. 3. New Builder APIs The latest version of Angular comes with new Builder APIs. Using these APIs, the developers will be able to go into the ng build, ng test, and ng run for development processes, such as deployment and building. Angular has worked closely with cloud service providers for the new APIs. The AngularFire is now available for use. Developers can install it to add deploy commands. Following installation, this command will build and deploy the app as suggested by . AngularFire 4. Opt-in telemetry in CLI Angular 8 will collect anonymous data such as commands and build speed to keep everything aligned with the community needs. This will also help the team behind Angular to learn how developers use the platform so that they can improve the things in future. The data will be collected by Angular only when the developers allow it to. 5. New API for easier workspace configuration There is a new API with Angular 8, aimed to make things easier for developers while making updates to the workspace configuration. Previously, they needed to manually open and make changes to the angular.json files in order to update the workspace configuration. 6. Support for web workers For the developers whose apps perform CPU-intensive processing, Angular is now going to have web worker support. This will accelerate the performance of apps despite the heavy workloads. What web workers do is allow developers to take some of the burden of the app to a background thread, like the image or video manipulation. 7. Backwards compatibility mode To make it easier to upgrade large-sized apps, the new Angular version will have a backwards compatibility mode. Enhancements have been made across lazy loading parts of the AngularJS app from Angular. Thus, the developers will now be able to move the main features of the app first, while loading only the AngularJS for the subsets. 8. Ivy & Bazel At Google, I/O 2018, the search engine giant had announced these features for . Since then, the users are eagerly waiting for them. Angular development Ivy is basically a new rendering pipeline and view engine that will translate the templates and components into regular HTML and JavaScript. There are several benefits of Ivy. For instance, it can reduce payload size, accelerate re-build time, eliminate the need for metadata.json, etc. Whereas, the Bazel is an open source tool for orchestration of build processes. This can help in optimization of the build flow and speed. Both these features are currently in the experimental phase and will be available in future. Updating to Angular v8 To update to the latest version of Angular, visit . Alternately, the developers can use this command for the update: update.angular.io ng update @angular/cli @angular/core Wrapping up In this article, we have mentioned all the new features and improvements in Angular 8. These features are very likely to make the lives of developers easier and boost their productivity. When Ivy and Bazel will join the Angular, things will become all the more innovative. What are your favorite features in Angular 8? Let us know via the comments section below.

What’s New in Angular V8 — All Powerful Features and Improvements

Enhance Sentiment Analysis with Role-Flipping Multi-LLM Negotiation

Authors:
(1) Xiaofei Sun, Zhejiang University;
(2) Xiaoya Li, Shannon.AI and Bytedance;
(3) Shengyu Zhang, Zhejiang University;
(4) Shuhe Wang, Peking University;
(5) Fei Wu, Zhejiang University;
(6) Jiwei Li, Zhejiang University;
(7) Tianwei Zhang, Nanyang Technological University;
(8) Guoyin Wang, Shannon.AI and Bytedance. Table of Links Abstract and Intro Related Work LLM Negotiation for Sentiment Analysis Experiments Ablation Studies Conclusion and References 2 Related Work 2.1 Sentiment Analysis Sentiment analysis (Pang and Lee, 2008; Go et al., 2009; Maas et al., 2011a; Zhang and Liu, 2012; Baccianella et al., 2010; Medhat et al., 2014; Bakshi et al., 2016; Zhang et al., 2018) is a task that aims to determine the overall sentiment polarity (e.g., positive, negative, neutral) of a given text. Earlier work often formalized the task as a two-step problem: (1) extract features using RNNs (Socher et al., 2013; Qian et al., 2016; Peled and Reichart, 2017; Wang et al., 2016b; Guggilla et al., 2016; Vo and Zhang, 2015), CNNs (Kalchbrenner et al., 2014; Wang et al., 2016a; Guan et al., 2016; Yu and Jiang, 2016; Mishra et al., 2017), pretrained language models (Lin et al., 2021; Sun et al., 2021; Phan and Ogunbona, 2020; Dai et al., 2021), etc; and (2) feed extracted features into a classifier for obtaining a pre-defined sentimental label. In recent years, in-context learning (ICL) has achieved great success and changed the paradigm of NLP tasks. Many works adapt ICL to the sentiment analysis task: Qin et al. (2023b); Sun et al. (2023a) propose a series of strategies to improve ChatGPT’s performance on the sentiment analysis task; Fei et al. (2023) propose a threehop reasoning framework, which induces the implicit aspect, opinion, and finally the sentiment polarity for the implicit sentiment analysis task;ƒ Zhang et al. (2023d) find that LLMs can achieve satisfactory performance on the binary sentiment classification task, but they underperform to the supervised baseline on more complex tasks (e.g., fine-grained sentiment analysis) that require deeper understanding or structured sentiment information. 2.2 Large Language Models and In-context Learning Large language models (LLMs) (Wang et al., 2022a; Zhang et al., 2023b) are models trained on massive unlabeled text corpora with selfsupervised learning techniques. Based on the model architecture, LLMs can be categorized into three types: (1) encoder-only models, which contain a text encoder and generate the input representations, such as BERT (Devlin et al., 2018) and its variants (Lan et al., 2019; Liu et al., 2019; Sun et al., 2020; Clark et al., 2020; Feng et al., 2020; Joshi et al., 2020; Sun et al., 2020, 2021); (2) decoder-only models, which have a decoder and generate text conditioned on the input text like GPT-series models (Radford et al., 2019; Brown et al., 2020; Keskar et al., 2019; Radford et al., 2019; Chowdhery et al., 2022; Ouyang et al., 2022; Zhang et al., 2022a; Scao et al., 2022; Zeng et al., 2022b; Touvron et al., 2023a; Peng et al., 2023; OpenAI, 2023); and (3) encoder-decoder models, which have a pair of encoder-decoder and generate text conditioned on the input representation, such as T5 (Raffel et al., 2020) and its variants (Lewis et al., 2019; Xue et al., 2020). Starting with GPT-3 (Brown et al., 2020), LLMs have shown emerging capabilities (Wei et al., 2022a) and completed NLP tasks through incontext learning (ICL), where LLMs generate labelintensive text conditioned on a few annotated examples without gradient updates. Many studies in the literature propose strategies for improving ICL performances on NLP tasks. Li and Liang (2021); Chevalier et al. (2023); Mu et al. (2023) optimize prompts in the continuous space. Liu et al. (2021a); Wan et al. (2023); Zhang et al. (2023a) search through the train set to retrieve k nearest neighbors of a test input as demonstrations. Zhang et al. (2022b); Sun et al. (2023b); Yao et al. (2023) decompose a task into a few subtasks and solve them step-by-step towards the final answer conditioned on LLM-generated reasoning chains. Sun et al. (2023a); Wang et al. (2023) propose to verify LLMs’ results by conducting a new round of prompting; Liu et al. (2021b); Feng et al. (2023) use LLMs to generate natural language knowledge statements and integrate external knowledge statements into prompts. 2.3 The LLM collaboration The LLM collaboration involves multiple LLMs working together to solve a given task. Specifically, the task is decomposed to several intermediate tasks, and each LLM is assigned to complete one intermediate task independently. The given task is solved after integrating or summarizing these intermediate results. The LLM collaboration approach can exploit the capabilities of LLMs, improve performances on complex tasks and enable to build complicated systems. Shinn et al. (2023); Sun et al. (2023a); Gero et al. (2023); Wang and Li (2023); Chen et al. (2023b) construct auxiliary tasks (e.g., reflection, verification tasks) and revise the response to the original task referring to the result of the auxiliary task. Talebirad and Nadiri (2023); Hong et al. (2023); Qian et al. (2023) assign characterize profiles (e.g., project manager, software engineer) to LLMs and gain performance boosts on character-specific tasks through behavior animations. Li et al. (2022); Zeng et al. (2022a); Chen et al. (2023a); Du et al. (2023); Liang et al. (2023) use a debate strategy in which multiple different LLMs propose their own responses to the given task and debate over multiple turns until getting a common final answer. Besides, Shen et al. (2023); Gao et al. (2023); Ge et al. (2023); Zhang et al. (2023c); Hao et al. (2023) employ one LLM as the task controller, which devises a plan for the given task, selects expert models for implementation and summarizes the responses of intermediate planned tasks. Other LLMs serve as task executors, completing intermediate tasks in their areas of expertise. This paper is available on arxiv under CC 4.0 license. Authors: (1) Xiaofei Sun, Zhejiang University; (2) Xiaoya Li, Shannon.AI and Bytedance; (3) Shengyu Zhang, Zhejiang University; (4) Shuhe Wang, Peking University; (5) Fei Wu, Zhejiang University; (6) Jiwei Li, Zhejiang University; (7) Tianwei Zhang, Nanyang Technological University; (8) Guoyin Wang, Shannon.AI and Bytedance. Authors: Authors: (1) Xiaofei Sun, Zhejiang University; (2) Xiaoya Li, Shannon.AI and Bytedance; (3) Shengyu Zhang, Zhejiang University; (4) Shuhe Wang, Peking University; (5) Fei Wu, Zhejiang University; (6) Jiwei Li, Zhejiang University; (7) Tianwei Zhang, Nanyang Technological University; (8) Guoyin Wang, Shannon.AI and Bytedance. Table of Links Abstract and Intro Abstract and Intro Related Work Related Work LLM Negotiation for Sentiment Analysis LLM Negotiation for Sentiment Analysis Experiments Experiments Ablation Studies Ablation Studies Conclusion and References Conclusion and References 2 Related Work 2.1 Sentiment Analysis Sentiment analysis (Pang and Lee, 2008; Go et al., 2009; Maas et al., 2011a; Zhang and Liu, 2012; Baccianella et al., 2010; Medhat et al., 2014; Bakshi et al., 2016; Zhang et al., 2018) is a task that aims to determine the overall sentiment polarity (e.g., positive, negative, neutral) of a given text. Earlier work often formalized the task as a two-step problem: (1) extract features using RNNs (Socher et al., 2013; Qian et al., 2016; Peled and Reichart, 2017; Wang et al., 2016b; Guggilla et al., 2016; Vo and Zhang, 2015), CNNs (Kalchbrenner et al., 2014; Wang et al., 2016a; Guan et al., 2016; Yu and Jiang, 2016; Mishra et al., 2017), pretrained language models (Lin et al., 2021; Sun et al., 2021; Phan and Ogunbona, 2020; Dai et al., 2021), etc; and (2) feed extracted features into a classifier for obtaining a pre-defined sentimental label. In recent years, in-context learning (ICL) has achieved great success and changed the paradigm of NLP tasks. Many works adapt ICL to the sentiment analysis task: Qin et al. (2023b); Sun et al. (2023a) propose a series of strategies to improve ChatGPT’s performance on the sentiment analysis task; Fei et al. (2023) propose a threehop reasoning framework, which induces the implicit aspect, opinion, and finally the sentiment polarity for the implicit sentiment analysis task;ƒ Zhang et al. (2023d) find that LLMs can achieve satisfactory performance on the binary sentiment classification task, but they underperform to the supervised baseline on more complex tasks (e.g., fine-grained sentiment analysis) that require deeper understanding or structured sentiment information. 2.2 Large Language Models and In-context Learning Large language models (LLMs) (Wang et al., 2022a; Zhang et al., 2023b) are models trained on massive unlabeled text corpora with selfsupervised learning techniques. Based on the model architecture, LLMs can be categorized into three types: (1) encoder-only models, which contain a text encoder and generate the input representations, such as BERT (Devlin et al., 2018) and its variants (Lan et al., 2019; Liu et al., 2019; Sun et al., 2020; Clark et al., 2020; Feng et al., 2020; Joshi et al., 2020; Sun et al., 2020, 2021); (2) decoder-only models, which have a decoder and generate text conditioned on the input text like GPT-series models (Radford et al., 2019; Brown et al., 2020; Keskar et al., 2019; Radford et al., 2019; Chowdhery et al., 2022; Ouyang et al., 2022; Zhang et al., 2022a; Scao et al., 2022; Zeng et al., 2022b; Touvron et al., 2023a; Peng et al., 2023; OpenAI, 2023); and (3) encoder-decoder models, which have a pair of encoder-decoder and generate text conditioned on the input representation, such as T5 (Raffel et al., 2020) and its variants (Lewis et al., 2019; Xue et al., 2020). Starting with GPT-3 (Brown et al., 2020), LLMs have shown emerging capabilities (Wei et al., 2022a) and completed NLP tasks through incontext learning (ICL), where LLMs generate labelintensive text conditioned on a few annotated examples without gradient updates. Many studies in the literature propose strategies for improving ICL performances on NLP tasks. Li and Liang (2021); Chevalier et al. (2023); Mu et al. (2023) optimize prompts in the continuous space. Liu et al. (2021a); Wan et al. (2023); Zhang et al. (2023a) search through the train set to retrieve k nearest neighbors of a test input as demonstrations. Zhang et al. (2022b); Sun et al. (2023b); Yao et al. (2023) decompose a task into a few subtasks and solve them step-by-step towards the final answer conditioned on LLM-generated reasoning chains. Sun et al. (2023a); Wang et al. (2023) propose to verify LLMs’ results by conducting a new round of prompting; Liu et al. (2021b); Feng et al. (2023) use LLMs to generate natural language knowledge statements and integrate external knowledge statements into prompts. 2.3 The LLM collaboration The LLM collaboration involves multiple LLMs working together to solve a given task. Specifically, the task is decomposed to several intermediate tasks, and each LLM is assigned to complete one intermediate task independently. The given task is solved after integrating or summarizing these intermediate results. The LLM collaboration approach can exploit the capabilities of LLMs, improve performances on complex tasks and enable to build complicated systems. Shinn et al. (2023); Sun et al. (2023a); Gero et al. (2023); Wang and Li (2023); Chen et al. (2023b) construct auxiliary tasks (e.g., reflection, verification tasks) and revise the response to the original task referring to the result of the auxiliary task. Talebirad and Nadiri (2023); Hong et al. (2023); Qian et al. (2023) assign characterize profiles (e.g., project manager, software engineer) to LLMs and gain performance boosts on character-specific tasks through behavior animations. Li et al. (2022); Zeng et al. (2022a); Chen et al. (2023a); Du et al. (2023); Liang et al. (2023) use a debate strategy in which multiple different LLMs propose their own responses to the given task and debate over multiple turns until getting a common final answer. Besides, Shen et al. (2023); Gao et al. (2023); Ge et al. (2023); Zhang et al. (2023c); Hao et al. (2023) employ one LLM as the task controller, which devises a plan for the given task, selects expert models for implementation and summarizes the responses of intermediate planned tasks. Other LLMs serve as task executors, completing intermediate tasks in their areas of expertise. This paper is available on arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. available on arxiv

The Power of Multi-LLM Frameworks in Overcoming Sentiment Analysis Challenges

সেন্টিমেন্ট এনালাইসিস চ্যালেঞ্জ কাটিয়ে উঠতে মাল্টি-এলএলএম ফ্রেমওয়ার্কের শক্তি

Die Leistungsfähigkeit von Multi-LLM-Frameworks bei der Bewältigung von Herausforderungen in der Sentimentanalyse

El poder de los marcos de múltiples LLM para superar los desafíos del análisis de sentimientos

La puissance des cadres multi-LLM pour surmonter les défis de l'analyse des sentiments

भावना विश्लेषण चुनौतियों पर काबू पाने में मल्टी-एलएलएम फ्रेमवर्क की शक्ति

感情分析の課題を克服するマルチ LLM フレームワークの力

감정 분석 과제를 극복하는 다중 LLM 프레임워크의 힘

O poder das estruturas Multi-LLM na superação dos desafios da análise de sentimento

Сила рамок Multi-LLM в преодолении проблем анализа настроений

Duygu Analizi Zorluklarının Aşılmasında Çoklu Yüksek Lisans Çerçevelerinin Gücü

Sức mạnh của các khung Multi-LLM trong việc vượt qua các thách thức về phân tích tình cảm

多 LLM 框架在克服情感分析挑战中的力量

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Xiaofei Sun, Zhejiang University & xiaofei_sun@zju.edu.cn;
(2) Xiaoya Li, Shannon.AI, Bytedance & xiaoya_li@shannonai.com;
(3) Shengyu Zhang, Zhejiang University & sy_zhang@zju.edu.cn;
(4) Shuhe Wang, Peking University & wangshuhe@stu.pku.edu.cn;
(5) Fei Wu, Zhejiang University & wufei@zju.edu.cn;
(6) Jiwei Li, Zhejiang University &  jiwei_li@zju.edu.cn;
(7) Tianwei Zhang, Nanyang Technological University & tianwei.zhang@ntu.edu.sg;
(8) Guoyin Wang, Bytedance & guoyin.wang@bytedance.com. Table of Links Abstract and Introduction
Related Work
LLM Negotiation for Sentiment Analysis
Experiments
Ablation Studies
Conclusion and References 2 Related Work 2.1 Sentiment Analysis Sentiment analysis (Pang and Lee, 2008; Go et al., 2009; Maas et al., 2011a; Zhang and Liu, 2012; Baccianella et al., 2010; Medhat et al., 2014; Bakshi et al., 2016; Zhang et al., 2018) is a task that aims to determine the overall sentiment polarity (e.g., positive, negative, neutral) of a given text. Earlier work often formalized the task as a two-step problem: (1) extract features using RNNs (Socher et al., 2013; Qian et al., 2016; Peled and Reichart, 2017; Wang et al., 2016b; Guggilla et al., 2016; Vo and Zhang, 2015), CNNs (Kalchbrenner et al., 2014; Wang et al., 2016a; Guan et al., 2016; Yu and Jiang, 2016; Mishra et al., 2017), pretrained language models (Lin et al., 2021; Sun et al., 2021; Phan and Ogunbona, 2020; Dai et al., 2021), etc; and (2) feed extracted features into a classifier for obtaining a pre-defined sentimental label. In recent years, in-context learning (ICL) has achieved great success and changed the paradigm of NLP tasks. Many works adapt ICL to the sentiment analysis task: Qin et al. (2023b); Sun et al.(2023a) propose a series of strategies to improve ChatGPT’s performance on the sentiment analysis task; Fei et al. (2023) propose a threehop reasoning framework, which induces the implicit aspect, opinion, and finally the sentiment polarity for the implicit sentiment analysis task;ƒ Zhang et al. (2023d) find that LLMs can achieve satisfactory performance on the binary sentiment classification task, but they underperform to the supervised baseline on more complex tasks (e.g., fine-grained sentiment analysis) that require deeper understanding or structured sentiment information. 2.2 Large Language Models and In-context Learning Large language models (LLMs) (Wang et al., 2022a; Zhang et al., 2023b) are models trained on massive unlabeled text corpora with self supervised learning techniques. Based on the model architecture, LLMs can be categorized into three types: (1) encoder-only models, which contain a text encoder and generate the input representations, such as BERT (Devlin et al., 2018) and its variants (Lan et al., 2019; Liu et al., 2019; Sun et al., 2020; Clark et al., 2020; Feng et al., 2020; Joshi et al., 2020; Sun et al., 2020, 2021); (2) decoder-only models, which have a decoder and generate text conditioned on the input text like GPT-series models (Radford et al., 2019; Brown et al., 2020; Keskar et al., 2019; Radford et al., 2019; Chowdhery et al., 2022; Ouyang et al., 2022; Zhang et al., 2022a; Scao et al., 2022; Zeng et al., 2022b; Touvron et al., 2023a; Peng et al., 2023; OpenAI, 2023); and (3) encoder-decoder models, which have a pair of encoder-decoder and generate text conditioned on the input representation, such as T5 (Raffel et al., 2020) and its variants (Lewis et al., 2019; Xue et al., 2020). Starting with GPT-3 (Brown et al., 2020), LLMs have shown emerging capabilities (Wei et al., 2022a) and completed NLP tasks through incontext learning (ICL), where LLMs generate labelintensive text conditioned on a few annotated examples without gradient updates. Many studies in the literature propose strategies for improving ICL performances on NLP tasks. Li and Liang (2021); Chevalier et al. (2023); Mu et al. (2023) optimize prompts in the continuous space. Liu et al. (2021a); Wan et al. (2023); Zhang et al. (2023a) search through the train set to retrieve k nearest neighbors of a test input as demonstrations. Zhang et al. (2022b); Sun et al. (2023b); Yao et al (2023) decompose a task into a few subtasks and solve them step-by-step towards the final answer conditioned on LLM-generated reasoning chains. Sun et al. (2023a); Wang et al. (2023) propose to verify LLMs’ results by conducting a new round of prompting; Liu et al. (2021b); Feng et al. (2023) use LLMs to generate natural language knowledge statements and integrate external knowledge statements into prompts. 2.3 The LLM collaboration The LLM collaboration involves multiple LLMs working together to solve a given task. Specifically, the task is decomposed to several intermediate tasks, and each LLM is assigned to complete one intermediate task independently. The given task is solved after integrating or summarizing these intermediate results. The LLM collaboration approach can exploit the capabilities of LLMs, improve performances on complex tasks and enable to build complicated systems. Shinn et al. (2023); Sun et al. (2023a); Gero et al. (2023); Wang and Li (2023); Chen et al. (2023b) construct auxiliary tasks (e.g., reflection, verification tasks) and revise the response to the original task referring to the result of the auxiliary task. Talebirad and Nadiri (2023); Hong et al. (2023); Qian et al. (2023) assign characterize profiles (e.g., project manager, software engineer) to LLMs and gain performance boosts on character-specific tasks through behavior animations. Li et al. (2022); Zeng et al. (2022a); Chen et al. (2023a); Du et al. (2023); Liang et al. (2023) use a debate strategy in which multiple different LLMs propose their own responses to the given task and debate over multiple turns until getting a common final answer. Besides, Shen et al. (2023); Gao et al. (2023); Ge et al. (2023); Zhang et al. (2023c); Hao et al. (2023) employone LLM as the task controller, which devises a  plan for the given task, selects expert models for implementation and summarizes the responses of intermediate planned tasks. Other LLMs serve as task executors, completing intermediate tasks in their areas of expertise. This paper is available on arxiv under CC 4.0 license. Authors: (1) Xiaofei Sun, Zhejiang University & xiaofei_sun@zju.edu.cn; (2) Xiaoya Li, Shannon.AI, Bytedance & xiaoya_li@shannonai.com; (3) Shengyu Zhang, Zhejiang University & sy_zhang@zju.edu.cn; (4) Shuhe Wang, Peking University & wangshuhe@stu.pku.edu.cn; (5) Fei Wu, Zhejiang University & wufei@zju.edu.cn; (6) Jiwei Li, Zhejiang University &  jiwei_li@zju.edu.cn; (7) Tianwei Zhang, Nanyang Technological University & tianwei.zhang@ntu.edu.sg; (8) Guoyin Wang, Bytedance & guoyin.wang@bytedance.com. This paper is available on arxiv under CC 4.0 license. Authors: (1) Xiaofei Sun, Zhejiang University & xiaofei_sun@zju.edu.cn; (2) Xiaoya Li, Shannon.AI, Bytedance & xiaoya_li@shannonai.com; (3) Shengyu Zhang, Zhejiang University & sy_zhang@zju.edu.cn; (4) Shuhe Wang, Peking University & wangshuhe@stu.pku.edu.cn; (5) Fei Wu, Zhejiang University & wufei@zju.edu.cn; (6) Jiwei Li, Zhejiang University &  jiwei_li@zju.edu.cn; (7) Tianwei Zhang, Nanyang Technological University & tianwei.zhang@ntu.edu.sg; (8) Guoyin Wang, Bytedance & guoyin.wang@bytedance.com. Table of Links Abstract and Introduction Related Work LLM Negotiation for Sentiment Analysis Experiments Ablation Studies Conclusion and References Abstract and Introduction Abstract and Introduction Related Work Related Work LLM Negotiation for Sentiment Analysis LLM Negotiation for Sentiment Analysis Experiments Experiments Ablation Studies Ablation Studies Conclusion and References Conclusion and References 2 Related Work 2.1 Sentiment Analysis Sentiment analysis (Pang and Lee, 2008; Go et al., 2009; Maas et al., 2011a; Zhang and Liu, 2012; Baccianella et al., 2010; Medhat et al., 2014; Bakshi et al., 2016; Zhang et al., 2018) is a task that aims to determine the overall sentiment polarity (e.g., positive, negative, neutral) of a given text. Earlier work often formalized the task as a two-step problem: (1) extract features using RNNs (Socher et al., 2013; Qian et al., 2016; Peled and Reichart, 2017; Wang et al., 2016b; Guggilla et al., 2016; Vo and Zhang, 2015), CNNs (Kalchbrenner et al., 2014; Wang et al., 2016a; Guan et al., 2016; Yu and Jiang, 2016; Mishra et al., 2017), pretrained language models (Lin et al., 2021; Sun et al., 2021; Phan and Ogunbona, 2020; Dai et al., 2021), etc; and (2) feed extracted features into a classifier for obtaining a pre-defined sentimental label. In recent years, in-context learning (ICL) has achieved great success and changed the paradigm of NLP tasks. Many works adapt ICL to the sentiment analysis task: Qin et al. (2023b); Sun et al.(2023a) propose a series of strategies to improve ChatGPT’s performance on the sentiment analysis task; Fei et al. (2023) propose a threehop reasoning framework, which induces the implicit aspect, opinion, and finally the sentiment polarity for the implicit sentiment analysis task;ƒ Zhang et al. (2023d) find that LLMs can achieve satisfactory performance on the binary sentiment classification task, but they underperform to the supervised baseline on more complex tasks (e.g., fine-grained sentiment analysis) that require deeper understanding or structured sentiment information. 2.2 Large Language Models and In-context Learning Large language models (LLMs) (Wang et al., 2022a; Zhang et al., 2023b) are models trained on massive unlabeled text corpora with self supervised learning techniques. Based on the model architecture, LLMs can be categorized into three types: (1) encoder-only models, which contain a text encoder and generate the input representations, such as BERT (Devlin et al., 2018) and its variants (Lan et al., 2019; Liu et al., 2019; Sun et al., 2020; Clark et al., 2020; Feng et al., 2020; Joshi et al., 2020; Sun et al., 2020, 2021); (2) decoder-only models, which have a decoder and generate text conditioned on the input text like GPT-series models (Radford et al., 2019; Brown et al., 2020; Keskar et al., 2019; Radford et al., 2019; Chowdhery et al., 2022; Ouyang et al., 2022; Zhang et al., 2022a; Scao et al., 2022; Zeng et al., 2022b; Touvron et al., 2023a; Peng et al., 2023; OpenAI, 2023); and (3) encoder-decoder models, which have a pair of encoder-decoder and generate text conditioned on the input representation, such as T5 (Raffel et al., 2020) and its variants (Lewis et al., 2019; Xue et al., 2020). Starting with GPT-3 (Brown et al., 2020), LLMs have shown emerging capabilities (Wei et al., 2022a) and completed NLP tasks through incontext learning (ICL), where LLMs generate labelintensive text conditioned on a few annotated examples without gradient updates. Many studies in the literature propose strategies for improving ICL performances on NLP tasks. Li and Liang (2021); Chevalier et al. (2023); Mu et al. (2023) optimize prompts in the continuous space. Liu et al. (2021a); Wan et al. (2023); Zhang et al. (2023a) search through the train set to retrieve k nearest neighbors of a test input as demonstrations. Zhang et al. (2022b); Sun et al. (2023b); Yao et al (2023) decompose a task into a few subtasks and solve them step-by-step towards the final answer conditioned on LLM-generated reasoning chains. Sun et al. (2023a); Wang et al. (2023) propose to verify LLMs’ results by conducting a new round of prompting; Liu et al. (2021b); Feng et al. (2023) use LLMs to generate natural language knowledge statements and integrate external knowledge statements into prompts. 2.3 The LLM collaboration The LLM collaboration involves multiple LLMs working together to solve a given task. Specifically, the task is decomposed to several intermediate tasks, and each LLM is assigned to complete one intermediate task independently. The given task is solved after integrating or summarizing these intermediate results. The LLM collaboration approach can exploit the capabilities of LLMs, improve performances on complex tasks and enable to build complicated systems. Shinn et al. (2023); Sun et al. (2023a); Gero et al. (2023); Wang and Li (2023); Chen et al. (2023b) construct auxiliary tasks (e.g., reflection, verification tasks) and revise the response to the original task referring to the result of the auxiliary task. Talebirad and Nadiri (2023); Hong et al. (2023); Qian et al. (2023) assign characterize profiles (e.g., project manager, software engineer) to LLMs and gain performance boosts on character-specific tasks through behavior animations. Li et al. (2022); Zeng et al. (2022a); Chen et al. (2023a); Du et al. (2023); Liang et al. (2023) use a debate strategy in which multiple different LLMs propose their own responses to the given task and debate over multiple turns until getting a common final answer. Besides, Shen et al. (2023); Gao et al. (2023); Ge et al. (2023); Zhang et al. (2023c); Hao et al. (2023) employone LLM as the task controller, which devises a  plan for the given task, selects expert models for implementation and summarizes the responses of intermediate planned tasks. Other LLMs serve as task executors, completing intermediate tasks in their areas of expertise.

Sentiment Analysis through LLM Negotiations: Related Work

Sentiment Analysis through LLM Negotiations: Conclusion and References

turned 1 year old this month :) Thanks for joining this journey. In our 6th month we surpassed 1 million minutes reading per month, and over the last 6 months, we’ve accumulated 18,801,558 minutes reading! That’s the equivalent of 35 years, 9 months and 12 days reading. Hacker Noon This week — we’re bringing you the stories, mindsets, moves, tools, management, series’ and the-why-it’s-built that makes a hacker a hacker. 4 Hacker Stories by The Sandnigger Programmer from Mississippi Nader Dabit by . The … Donald J. Trump’s $30M Testing Team Michael Babyak GOP tweeted this by I’m Sorry, Surface, But I Just Can’t Anymore. Jennifer Hoelzer by . I used LAMP to make a SaaS with $3700/mo profit. Here’s how. Alex Moskovski 10 Hacker Mindsets by . Very funny. The 9 Types of Programmers and Why Some are Better than Others Dominik T by . Learn To Code In 2017, Get Hired, And Have Fun Along The Way Andrei Neagoie by The best way to learn new technologies is by doing. Here’s some inspiration André Pena by 5 Tips for Shipping Your App Pedro Silva Moreira by The advantages of building your website from scratch Owen Far by Practical scaling techniques for web sites Adam Derewecki by Want to become a Software Developer? Here’s how to get there André Pena … by Documentation is a bit like people dying William Belk by It’s All Fun & Games until AI wins them all Matthew Biggins by If we build it, why would anyone care? Christian Sepulveda 4 Hacker Moves by . AI can help you choose where to live Silver Keskkula by . The Programmer’s Guide to Booking a Plane Zeke Gabrielse by After 6 incredible years at Buffer I’m moving on to something else Leo Widrich by . Is your move to listen to blog posts instead of reading them? We got you covered. No time to read? Just listen! Mahmoud Felfel 10 Hacker Tools by . Launch here! React Studio Public Beta 2 React Studio by Polyfills: everything you ever wanted to know, or maybe a bit less David Gilbertson by Appointment Web Forms Matthew Talebi How do GANs intuitively work? by Chanchana Sornsoontorn by From “What is a Markov Model” to “Here is how Markov Models Work” Alexander Dejeu by Top 11 JavaScript Libraries That Are Worth Your Attention Eugeniya Korotya by Offline First React Native + Meteor Apps Spencer Carli by Straightforward code splitting with React and Webpack Didier FRANC by Exclude files from git without committing changes to .gitignore 🙅 Dave Lunny by Build an Alexa Skill in 7 Minutes Flat with Node.js and StdLib Keith Horwood Contribute to Hacker Noon today ! 4 Hacker Management Tips by 10 Big Software PM Time Wasters John Cutler by Taking the Associate Android Developer Exam Josh Burton by Seniority Krzysztof Zabłocki by Building an app? Do these things. Paul Keck A Couple of Hacker Series’ by The Learn AI If You Suck at Math Series (4 Parts) Daniel Jeffries by How I Hacked Google DayDream Controller (5 Parts) Matteo Pisani And a Why it’s Built The Way It’s Built Story: by . Apple’s Icons Have That Shape for a Very Good Reason Mark Stanton “A lot of hacking is playing with other people, you know, getting them to do strange things,” said Steve Wozniak. Until next time — enjoy , the power of your digital sabbatical , . David Smooke Hacker Noon your favorite hacks! P.S. Tweet @HackerNoon Check out what our community is reading, as filtered in , , & . P.P.S. real-time top tweets verified accounts

Stories, News & Mentions about JALEBI

Jalebi’s Stories on HackerNoon