Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing Recommendation Explanations through User-Centric Refinement

Jingsen Zhang, Zihang Tian, Xueyang Feng, Xu Chen

TL;DR

NLER explanations often fail to satisfy user-centric requirements such as factuality, personalization, and sentiment coherence. RefineX introduces an LLM-powered, multi-agent system that refines initial explanations during inference through a plan-then-refine pipeline and hierarchical reflection, guided by a memory module and an aspect library. The approach identifies limitations of existing NLER methods, presents a Planner-Refiner-Reflector framework, and demonstrates improvements across three real-world datasets and human evaluations. The work shows that targeted, reusable refinement can produce more accurate, personalized, and sentiment-coherent explanations and adapt to diverse user goals in recommendation scenarios.

Abstract

Generating natural language explanations for recommendations has become increasingly important in recommender systems. Traditional approaches typically treat user reviews as ground truth for explanations and focus on improving review prediction accuracy by designing various model architectures. However, due to limitations in data scale and model capability, these explanations often fail to meet key user-centric aspects such as factuality, personalization, and sentiment coherence, significantly reducing their overall helpfulness to users. In this paper, we propose a novel paradigm that refines initial explanations generated by existing explainable recommender models during the inference stage to enhance their quality in multiple aspects. Specifically, we introduce a multi-agent collaborative refinement framework based on large language models. To ensure alignment between the refinement process and user demands, we employ a plan-then-refine pattern to perform targeted modifications. To enable continuous improvements, we design a hierarchical reflection mechanism that provides feedback on the refinement process from both strategic and content perspectives. Extensive experiments on three datasets demonstrate the effectiveness of our framework.

Enhancing Recommendation Explanations through User-Centric Refinement

TL;DR

NLER explanations often fail to satisfy user-centric requirements such as factuality, personalization, and sentiment coherence. RefineX introduces an LLM-powered, multi-agent system that refines initial explanations during inference through a plan-then-refine pipeline and hierarchical reflection, guided by a memory module and an aspect library. The approach identifies limitations of existing NLER methods, presents a Planner-Refiner-Reflector framework, and demonstrates improvements across three real-world datasets and human evaluations. The work shows that targeted, reusable refinement can produce more accurate, personalized, and sentiment-coherent explanations and adapt to diverse user goals in recommendation scenarios.

Abstract

Generating natural language explanations for recommendations has become increasingly important in recommender systems. Traditional approaches typically treat user reviews as ground truth for explanations and focus on improving review prediction accuracy by designing various model architectures. However, due to limitations in data scale and model capability, these explanations often fail to meet key user-centric aspects such as factuality, personalization, and sentiment coherence, significantly reducing their overall helpfulness to users. In this paper, we propose a novel paradigm that refines initial explanations generated by existing explainable recommender models during the inference stage to enhance their quality in multiple aspects. Specifically, we introduce a multi-agent collaborative refinement framework based on large language models. To ensure alignment between the refinement process and user demands, we employ a plan-then-refine pattern to perform targeted modifications. To enable continuous improvements, we design a hierarchical reflection mechanism that provides feedback on the refinement process from both strategic and content perspectives. Extensive experiments on three datasets demonstrate the effectiveness of our framework.

Paper Structure

This paper contains 23 sections, 9 equations, 4 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminary
  3. Approach
  4. Framework Overview
  5. The Refinement Phase
  6. The Reflection Phase
  7. Discussion
  8. Experiments
  9. Experiment Setup
  10. Overall Performance
  11. Ablation Study of the Reflection
  12. Adaptability Analysis
  13. Human Evaluation
  14. Case Study
  15. Related Work
  16. ...and 8 more sections

Figures (4)

  • Figure 1: Illustration of inadequate user-centric quality in explanations produced by the PETER model.
  • Figure 2: The overview of our RefineX framework, where "Exp" denotes "Explanation".
  • Figure 3: Human evaluation results of PETER and RefineX on three datasets across three aspects.
  • Figure 4: An example of the refinement process. The font colors in the explanations correspond to specific content in the existing reviews.