notesum.ai

Published at October 30

ReasoningRec: Bridging Personalized Recommendations and Human-Interpretable Explanations through LLM Reasoning

cs.IR

cs.AI

Released Date: October 30, 2024

Authors: Millennium Bismay¹, Xiangjue Dong¹, James Caverlee¹

Aff.: ¹Texas A&M University

Arxiv: http://arxiv.org/abs/2410.23180v1

Refer to caption

	Prompt Template and Output for ReasoningRec
	You are an expert <assign_role_for_llm>.You are provided with the user profile and list of recent <items> that the user has <user_activity> and whether the user likes it or not.
	User Profile – < $P_{u}$ >
	List of recent <items> and their description – Liked/Disliked $<i_{,}D_{i}>_{i=t-k-1}^{t-1}$
	Analyze all the information given in order. Do not use any information not mentioned above.
Prompt	Predict whether the user will like the target <item> – < $i_{t},D_{i_{t}}$ > or not. Answer with a Yes or No in the following format – Prediction: Yes or No by analyzing the user’s behavior from the given list of <items> and identify the <item> characteristics that the user likes and dislikes in at most 100 words. Explain with reasoning whether the user will like or dislike the target <item> – < $i_{t}$ > in at most 100 words.
	Prediction: Yes
	The user has shown a preference for cute and comfortable accessories …
Output	They have disliked products that appear cheap or break easily …
	… the bracelet is described as high quality and nicely polished, suggesting it will not have these issues … the user is likely to appreciate the bracelet’s adjustable and high-quality features …