Identify User Intention for Recommendation using Chain-of-Thought Prompting in LLM

It is crucial for a recommendation system to accurately identify a user’s in-tention since it dictates the user’s selection once facing multiple candidates. Traditional user intention identification uses the user’s selection among mul-tiple items. This technique relies primarily on historical behavioral data and results in problems such as the cold start, unintended choice, and failure when items are new. Motivated by recent advancements in Large Language Models (LLMs) like ChatGPT, we present an approach for user intention identification by embracing LLMs with Chain-of-Thought (CoT) prompting. We use the initial user profile as inputs to LLMs and design a collection of prompts to align the LLM’s response through various recommendation tasks encompassing rating prediction, search and browse history, user clarification, etc. to identify the user’s intention and then the same input with generated user intention feed to the LLM to produce recommendations. We tested our approach with real-world datasets to demonstrate the improvements in the recommendation comparing the recommendation without the intention merged into a user model. The results indicate the clear improvements.