Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Millennium Bismay, Xiangjue Dong, James Caverlee

TL;DR

ReasoningRec is presented, a reasoning-based recommendation framework that leverages Large Language Models (LLMs) to bridge the gap between recommendations and human-interpretable explanations, demonstrating that ReasoningRec surpasses state-of-the-art methods by up to 12.5\% in recommendation prediction while concurrently providing human-intelligible explanations.

Abstract

This paper presents ReasoningRec, a reasoning-based recommendation framework that leverages Large Language Models (LLMs) to bridge the gap between recommendations and human-interpretable explanations. In contrast to conventional recommendation systems that rely on implicit user-item interactions, ReasoningRec employs LLMs to model users and items, focusing on preferences, aversions, and explanatory reasoning. The framework utilizes a larger LLM to generate synthetic explanations for user preferences, subsequently used to fine-tune a smaller LLM for enhanced recommendation accuracy and human-interpretable explanation. Our experimental study investigates the impact of reasoning and contextual information on personalized recommendations, revealing that the quality of contextual and personalized data significantly influences the LLM's capacity to generate plausible explanations. Empirical evaluations demonstrate that ReasoningRec surpasses state-of-the-art methods by up to 12.5\% in recommendation prediction while concurrently providing human-intelligible explanations. The code is available here: https://github.com/millenniumbismay/reasoningrec.

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

TL;DR

ReasoningRec is presented, a reasoning-based recommendation framework that leverages Large Language Models (LLMs) to bridge the gap between recommendations and human-interpretable explanations, demonstrating that ReasoningRec surpasses state-of-the-art methods by up to 12.5\% in recommendation prediction while concurrently providing human-intelligible explanations.

Abstract

This paper presents ReasoningRec, a reasoning-based recommendation framework that leverages Large Language Models (LLMs) to bridge the gap between recommendations and human-interpretable explanations. In contrast to conventional recommendation systems that rely on implicit user-item interactions, ReasoningRec employs LLMs to model users and items, focusing on preferences, aversions, and explanatory reasoning. The framework utilizes a larger LLM to generate synthetic explanations for user preferences, subsequently used to fine-tune a smaller LLM for enhanced recommendation accuracy and human-interpretable explanation. Our experimental study investigates the impact of reasoning and contextual information on personalized recommendations, revealing that the quality of contextual and personalized data significantly influences the LLM's capacity to generate plausible explanations. Empirical evaluations demonstrate that ReasoningRec surpasses state-of-the-art methods by up to 12.5\% in recommendation prediction while concurrently providing human-intelligible explanations. The code is available here: https://github.com/millenniumbismay/reasoningrec.

Paper Structure

This paper contains 38 sections, 5 equations, 2 figures, 12 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLMs for Recommender Systems.
  4. Explanations in Recommendation.
  5. LLM Reasoning.
  6. ReasoningRec
  7. Preliminaries
  8. Item Description Generation
  9. User Profile Creation
  10. Explanatory Reasoning Generation
  11. Finetuning with Personalized Reasoning
  12. Experimental Setup
  13. Datasets
  14. Baselines and Evaluation Metrics
  15. Implementation Details
  16. ...and 23 more sections

Figures (2)

  • Figure 1: Overview of ReasoningRec framework. (a) We use an LLM, Mixtral-8x7B-Instruct-v0.1, to generate item description and user profile. We use Chain-of-Thought prompting with the most recent $k$ items and leverage the rich semantic information to generate reasoning ground truth, conditioned on the target item ground truth rating. (b) We propose a lightweight instruction finetuning framework, utilizing LoRA architecture, with very limited training samples to instruction finetune an SLM, Llama-2-7b-chat-hf. (c) We use the instruction finetuned model to perform recommendation prediction and reasoning generation.
  • Figure 2: The multi-line graph represents the unique number of users with at least $k$ user-item interactions in each dataset. ML1M drops beyond $k=20$ suggesting very dense dataset. However, Beauty and Fashion drops drastically from $k=5$ to $k=7$, highlighting the sparsity in the dataset.