Scholarlink

<p>Session-based recommendation (SBR) predicts items a user may interact with in an anonymous session by analyzing short-term behavioral patterns. However, accidental clicks in user interaction sequences can distort model predictions from true intentions. Existing noise filtering methods remove noise items by extracting user intent but fail to eliminate noise item features and their connecting edges effectively. Additionally, while category information provides valuable clues to help identify user preferences, current methods still underutilize it. To address these issues, we propose CNF-MGNN, a Mixed graph neural network for session-based recommendation with category-aware noise filtering. Our method evaluates items by combining interaction frequency and category preference, then learns item score distributions via an autoencoder to reconstruct score errors and identify potential noise. We further model item embeddings and session representations using a hypergraph and heterogeneous graph mixed structure. Experiments on three real datasets demonstrate the superior recommendation performance of CNF-MGNN. The source code is available at: <a href="https://github.com/Typejunjie/code.">https://github.com/Typejunjie/code.</a></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

Mixed graph neural network for session-based recommendation with category-aware noise filtering

  • 期刊论文

  • 发表日期：

    2025年8月28日

查看全文
  • Hui Wang,
  • Junjie You,
  • Jianbing Yi,
  • Junyan Liu,
  • Runjie Huang

摘要

Session-based recommendation (SBR) predicts items a user may interact with in an anonymous session by analyzing short-term behavioral patterns. However, accidental clicks in user interaction sequences can distort model predictions from true intentions. Existing noise filtering methods remove noise items by extracting user intent but fail to eliminate noise item features and their connecting edges effectively. Additionally, while category information provides valuable clues to help identify user preferences, current methods still underutilize it. To address these issues, we propose CNF-MGNN, a Mixed graph neural network for session-based recommendation with category-aware noise filtering. Our method evaluates items by combining interaction frequency and category preference, then learns item score distributions via an autoencoder to reconstruct score errors and identify potential noise. We further model item embeddings and session representations using a hypergraph and heterogeneous graph mixed structure. Experiments on three real datasets demonstrate the superior recommendation performance of CNF-MGNN. The source code is available at: https://github.com/Typejunjie/code.