Scholarlink

<p>Existing review-based recommender systems (RSs) are constrained by their shallow semantic representation capabilities, whereas large language models (LLMs) have garnered widespread attention due to their strong contextual understanding and semantic analysis abilities. Despite the significant performance achieved by LLM-powered review-based RSs, they face challenges in terms of inadequate inference efficiency, which hinders their ability to meet the low-latency requirements of real-time recommendations. To enhance the performance of RSs with LLMs while maintaining high efficiency, we propose AEFARec (<b>A</b>spect-aware <b>E</b>xplicit <b>F</b>eature <b>A</b>ugmentation for Recommendation). In AEFARec, we initially compute aspect-aware explicit features. This involves utilizing LLMs to extract aspects from unstructured reviews and calculate explicit attribute features of items and explicit preference features of users. Subsequently, we devise a hybrid feature fusion network that integrates implicit features with aspect-aware explicit features to generate final rating predictions. Extensive experiments demonstrate the effectiveness of the proposed AEFARec, and efficiency analysis experiments reveal that AEFARec possesses significantly superior inference efficiency compared to LLM-powered RSs, providing an effective solution for real-time recommendation scenarios.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing recommender systems with LLM-extracted explicit features

  • 期刊论文

  • 发表日期：

    2025年12月29日

查看全文
  • Shuai Zhao,
  • Guoquan Jiang

摘要

Existing review-based recommender systems (RSs) are constrained by their shallow semantic representation capabilities, whereas large language models (LLMs) have garnered widespread attention due to their strong contextual understanding and semantic analysis abilities. Despite the significant performance achieved by LLM-powered review-based RSs, they face challenges in terms of inadequate inference efficiency, which hinders their ability to meet the low-latency requirements of real-time recommendations. To enhance the performance of RSs with LLMs while maintaining high efficiency, we propose AEFARec (Aspect-aware Explicit Feature Augmentation for Recommendation). In AEFARec, we initially compute aspect-aware explicit features. This involves utilizing LLMs to extract aspects from unstructured reviews and calculate explicit attribute features of items and explicit preference features of users. Subsequently, we devise a hybrid feature fusion network that integrates implicit features with aspect-aware explicit features to generate final rating predictions. Extensive experiments demonstrate the effectiveness of the proposed AEFARec, and efficiency analysis experiments reveal that AEFARec possesses significantly superior inference efficiency compared to LLM-powered RSs, providing an effective solution for real-time recommendation scenarios.