Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Explainable Recommendation with Personalized Review Retrieval and Aspect Learning

Hao Cheng, Shuo Wang, Wensheng Lu, Wei Zhang, Mingyang Zhou, Kezhong Lu, Hao Liao

TL;DR

ERRA addresses explainable recommendation under data sparsity by integrating retrieval-enhanced knowledge and aspect-based personalization within a multi-task Transformer framework. It retrieves sentence-level context from the training corpus with Sentence-BERT and Faiss ANN, and selects top-n aspects per user-item to tailor representations, forming the Joint Enhancement Transformer. The model optimizes a joint objective combining rating and explanation losses, showing consistent improvements over baselines on Yelp, Amazon, and TripAdvisor datasets, including up to 3.4% in prediction and 15.8% in explanations on TripAdvisor. This approach improves both predictive accuracy and the quality of natural language explanations, with practical implications for more persuasive and personalized recommender systems.

Abstract

Explainable recommendation is a technique that combines prediction and generation tasks to produce more persuasive results. Among these tasks, textual generation demands large amounts of data to achieve satisfactory accuracy. However, historical user reviews of items are often insufficient, making it challenging to ensure the precision of generated explanation text. To address this issue, we propose a novel model, ERRA (Explainable Recommendation by personalized Review retrieval and Aspect learning). With retrieval enhancement, ERRA can obtain additional information from the training sets. With this additional information, we can generate more accurate and informative explanations. Furthermore, to better capture users' preferences, we incorporate an aspect enhancement component into our model. By selecting the top-n aspects that users are most concerned about for different items, we can model user representation with more relevant details, making the explanation more persuasive. To verify the effectiveness of our model, extensive experiments on three datasets show that our model outperforms state-of-the-art baselines (for example, 3.4% improvement in prediction and 15.8% improvement in explanation for TripAdvisor).

Explainable Recommendation with Personalized Review Retrieval and Aspect Learning

TL;DR

ERRA addresses explainable recommendation under data sparsity by integrating retrieval-enhanced knowledge and aspect-based personalization within a multi-task Transformer framework. It retrieves sentence-level context from the training corpus with Sentence-BERT and Faiss ANN, and selects top-n aspects per user-item to tailor representations, forming the Joint Enhancement Transformer. The model optimizes a joint objective combining rating and explanation losses, showing consistent improvements over baselines on Yelp, Amazon, and TripAdvisor datasets, including up to 3.4% in prediction and 15.8% in explanations on TripAdvisor. This approach improves both predictive accuracy and the quality of natural language explanations, with practical implications for more persuasive and personalized recommender systems.

Abstract

Explainable recommendation is a technique that combines prediction and generation tasks to produce more persuasive results. Among these tasks, textual generation demands large amounts of data to achieve satisfactory accuracy. However, historical user reviews of items are often insufficient, making it challenging to ensure the precision of generated explanation text. To address this issue, we propose a novel model, ERRA (Explainable Recommendation by personalized Review retrieval and Aspect learning). With retrieval enhancement, ERRA can obtain additional information from the training sets. With this additional information, we can generate more accurate and informative explanations. Furthermore, to better capture users' preferences, we incorporate an aspect enhancement component into our model. By selecting the top-n aspects that users are most concerned about for different items, we can model user representation with more relevant details, making the explanation more persuasive. To verify the effectiveness of our model, extensive experiments on three datasets show that our model outperforms state-of-the-art baselines (for example, 3.4% improvement in prediction and 15.8% improvement in explanation for TripAdvisor).
Paper Structure (27 sections, 10 equations, 3 figures, 5 tables)

This paper contains 27 sections, 10 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Explainable Recommendation with Generation
  4. Problem Statement
  5. Methodology
  6. Overview of Model
  7. Retrieval Enhancement
  8. Retrieval Encode
  9. Retrieval Method
  10. Aspect Enhancement
  11. Joint Enhancement Transformers
  12. Rating Prediction
  13. Explanation Generation
  14. Aspect Discriminator
  15. Text Generation
  16. ...and 12 more sections

Figures (3)

  • Figure 1: A user's reviews of different items and selected reviews by different models. Specifically, (a) a CNN-based method, by which the review selected is too general, (b) a user-id attention-based query method ref29, by which the review selected is not specific, (c) a Co-attention based method ref5, by which the review selected contain some details, (d) our model: retrieval-based method generates informative and personalized reviews that are relevant to the hotel.
  • Figure 2: An overview of the ERRA framework.
  • Figure 3: Ablation analysis of prediction and explanation tasks