Authors:
(1) Saketh Reddy Karra, University of Illinois Chicago, 601 S Morgan St, Chicago, IL 60607, United States;
(2) Theja Tulabandhula, University of Illinois Chicago, 601 S Morgan St, Chicago, IL 60607, United States. Authors: Authors: (1) Saketh Reddy Karra, University of Illinois Chicago, 601 S Morgan St, Chicago, IL 60607, United States; (2) Theja Tulabandhula, University of Illinois Chicago, 601 S Morgan St, Chicago, IL 60607, United States. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2. Related Work 2. Related Work 3. Preliminaries 3. Preliminaries 4. The InteraSSort Framework 4. The InteraSSort Framework 5. Illustration 5. Illustration 6. Conclusion and References 6. Conclusion and References 3. Preliminaries In this section, we discuss the assortment planning problem in detail and key questions that can be answered through interactivity. 3.1. Assortment planning 3.1. Assortment planning The complex nature of the assortment planning problem requires the development of robust optimization methodologies that can work well with different types of constraints and produce viable solutions within reasonable time frames. In this study, we adopt a series of scalable efficient algorithms (Tulabandhula et al., 2022) addressing the assortment optimization problem. 3.2. Key questions answered through interactivity 3.2. Key questions answered through interactivity As previously highlighted, store planners with their deep domain knowledge, often need to generate insights for questions that involve variations of the assortment optimization problem. Accordingly, our framework InteraSSort needs to interactively address the key questions outlined below. InteraSSort • What would be the optimal assortment when constrained by a specific limit on the assortment size? • What constitutes the optimal assortment when a product cannot be included? • What is the expected revenue of the assortment if a product is to be part of the selection? 4. The InteraSSort Framework Solving an assortment planning problem in real-world scenarios involves several crucial steps. The process begins with exhaustive data collection and analysis, followed by selecting a suitable choice model and estimating its parameters. Subsequently, the relevant optimization algorithm is executed to determine the optimal assortment. The process concludes with communication and implementation of derived decisions among various stakeholders. Our framework, InteraSSort, as shown in Figure 2 takes input via user prompts. The LLM with function calling ability translates these prompts into the desired format and executes the optimization tools following a validation check. The generated solutions are relayed back to the user via the LLM. This interactive process repeats as the user provides additional prompts, fostering a dynamic exchange of information. The process discussed above is structured into multiple stages: 1) prompt design, 2) prompt decomposition, and 3) tool execution & response generation. InteraSSort 4.1. Prompt design 4.1. Prompt design InteraSSort uses an LLM to perform detailed analyses of user requests, which are submitted as text prompts. Therefore, the design of the prompts is crucial for accurately capturing and utilizing user requests in later stages. To facilitate this, the framework requires a standardized template for input prompts to systematically extract constraints and other relevant information i.e., the dataset and choice model to be used. InteraSSort 4.2. Prompt decompostion 4.2. Prompt decompostion In this stage, InteraSSort leverages the function-calling capabilities of the LLM to break down the standardized prompts. This feature empowers the LLM to generate JSON objects containing arguments for calling functions that conform to the predefined specifications required for solving the optimization problem. The function calling template incorporates multiple slots, such as ‘model’, ‘dataset’, and ‘cardinality’, to represent various variables and constraints as shown in Figure. By adhering to these task specifications, InteraSSort efficiently utilizes the LLM to analyze user requests and accurately parse them. InteraSSort InteraSSort To facilitate interactive multi-turn conversations, InteraSSort has the capability to append chat history to the follow-up prompts. This is crucial, as these prompts may lack the entire context required to generate a solution. Consequently, whenever a user poses a follow-up question, InteraSSort can reference past interactions and trace prior user responses to answer subsequent questions. This functionality enables InteraSSort to more effectively manage context and respond to user requests in multiturn dialogues. InteraSSort InteraSSort InteraSSort 4.3. Tool execution & response generation 4.3. Tool execution & response generation InteraSSort effectively manages and processes the output received from the prompt decomposition stage. This involves conducting thorough validation checks, such as InteraSSort range and consistency assessments, to ensure the accuracy and reliability of the decomposed prompts. InteraSSort maintains a comprehensive database for parameters of choice models across multiple datasets. Upon successful validation, it retrieves corresponding parameters based on the choice model identified in the decomposed prompt. Utilizing the choice model parameters and any other constraints as arguments, InteraSSort executes the optimization scripts using tools like optimization solvers to achieve the best possible results. Finally, InteraSSort enables the LLM to receive these results as input and generate responses in user-friendly language. InteraSSort InteraSSort InteraSSort This paper is available on arxiv under CC by 4.0 Deed (Attribution 4.0 International) license. This paper is available on arxiv under CC by 4.0 Deed (Attribution 4.0 International) license. available on arxiv available on arxiv

Part of HackerNoon's growing list of open-source research papers, promoting free access to academic material.

The AI Framework That Makes Optimization as Easy as Chatting

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

AI Transforms 800K+ Grocery Transactions into Smart Insights

Adopting AI in Sales and Marketing

Redefining the Digital Age: The AI-Driven Evolution in Marketing

Six Emerging Applications of AI in Marketing

Inside the Web3 Marketing Revolution: Addressable's $13.5M Milestone with Tomer Sharoni

Attention-grabbing PR Tactics for the Digital Age: How to Boost Organic Reach

AI Transforms 800K+ Grocery Transactions into Smart Insights

Adopting AI in Sales and Marketing

Redefining the Digital Age: The AI-Driven Evolution in Marketing

Six Emerging Applications of AI in Marketing

Inside the Web3 Marketing Revolution: Addressable's $13.5M Milestone with Tomer Sharoni

Attention-grabbing PR Tactics for the Digital Age: How to Boost Organic Reach

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps