Scholarlink

In this study, we investigate the incorporation of anomaly detection into session-based recommender systems using Graph Neural Networks (GNNs) to enhance recommendation accuracy. We explore various integration stages, including item embedding, fusion layers, and session representation. Our methodology involves initially zeroing out anomalous embeddings and subsequently replacing them with what we consider to be the best embeddings based on anomaly scores derived from the Isolation Forest algorithm. Through extensive experimentation on a real-world dataset, we observed that anomaly detection at the session representation layer yields significant improvements in hit rate (HR) and mean reciprocal rank (MRR). Our findings demonstrate that strategically integrating anomaly detection can effectively improve recommendation quality and system performance.

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 Anomaly Detection: A Graph Neural Network Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Bahareh Rahmatikargar,
  • Pooya Moradian Zadeh,
  • Ziad Kobti

摘要

In this study, we investigate the incorporation of anomaly detection into session-based recommender systems using Graph Neural Networks (GNNs) to enhance recommendation accuracy. We explore various integration stages, including item embedding, fusion layers, and session representation. Our methodology involves initially zeroing out anomalous embeddings and subsequently replacing them with what we consider to be the best embeddings based on anomaly scores derived from the Isolation Forest algorithm. Through extensive experimentation on a real-world dataset, we observed that anomaly detection at the session representation layer yields significant improvements in hit rate (HR) and mean reciprocal rank (MRR). Our findings demonstrate that strategically integrating anomaly detection can effectively improve recommendation quality and system performance.