Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Sequence-aware Large Language Models for Explainable Recommendation

Gangyi Zhang, Runzhe Teng, Chongming Gao

Abstract

Large Language Models (LLMs) have shown strong potential in generating natural language explanations for recommender systems. However, existing methods often overlook the sequential dynamics of user behavior and rely on evaluation metrics misaligned with practical utility. We propose SELLER (SEquence-aware LLM-based framework for Explainable Recommendation), which integrates explanation generation with utility-aware evaluation. SELLER combines a dual-path encoder-capturing both user behavior and item semantics with a Mixture-of-Experts adapter to align these signals with LLMs. A unified evaluation framework assesses explanations via both textual quality and their effect on recommendation outcomes. Experiments on public benchmarks show that SELLER consistently outperforms prior methods in explanation quality and real-world utility.

Sequence-aware Large Language Models for Explainable Recommendation

Abstract

Large Language Models (LLMs) have shown strong potential in generating natural language explanations for recommender systems. However, existing methods often overlook the sequential dynamics of user behavior and rely on evaluation metrics misaligned with practical utility. We propose SELLER (SEquence-aware LLM-based framework for Explainable Recommendation), which integrates explanation generation with utility-aware evaluation. SELLER combines a dual-path encoder-capturing both user behavior and item semantics with a Mixture-of-Experts adapter to align these signals with LLMs. A unified evaluation framework assesses explanations via both textual quality and their effect on recommendation outcomes. Experiments on public benchmarks show that SELLER consistently outperforms prior methods in explanation quality and real-world utility.
Paper Structure (51 sections, 16 equations, 6 figures, 5 tables)

This paper contains 51 sections, 16 equations, 6 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Notations and Definitions
  4. Task Definition: Sequence-Aware Explainable Recommendation
  5. Unified Evaluation Framework: Explanation Utility Assessment
  6. Methodology
  7. Sequence-Aware Explanation Generator
  8. Dual-Path Sequence Encoding
  9. MoE-based Sequence Adaptation
  10. Sequence-Aware LLM Fine-tuning
  11. Training Objective
  12. Explanation-Enhanced Recommender
  13. Explanation Encoding
  14. Hypernetwork-Generated Adapter
  15. Explanation-Aware SASRec Fine-tuning
  16. ...and 36 more sections

Figures (6)

  • Figure 1: Overall architecture of the SELLER framework. The Sequence-Aware Explanation Generator (SEG) captures user behavioral patterns through dual-path encoding and generates personalized explanations via MoE-based adaptation. The Explanation-Enhanced Recommender (EER) serves as our unified evaluation framework to assess explanation utility through downstream recommendation performance.
  • Figure 2: Ablation study on the sequence-aware components of SELLER. BE: MoE-adapted behavioral embedding, SE: MoE-adapted semantic embedding, CT: category texts, DE: BE+SE.
  • Figure 3: Structured meta-information filtering process for the Yelp dataset. The raw metadata is filtered to retain relevant fields that capture business characteristics while removing non-essential information.
  • Figure 4: Input structure for the Sequence-Aware Explanation Generator, including the system prompt, input fields, and expected output format with examples.
  • Figure 5: Ground truth explanation generation process for the KuaiRec dataset, showing the task specification, prompt design, and sample output format.
  • ...and 1 more figures