Scholarlink

<p>Session-based recommendation systems (SRS) are essential in dynamic environments where user preferences evolve over time. This paper proposes a Hybrid session-based recommendation (HSRec) model that integrates "Convolutional Neural Networks" (CNN) with "Long Short-Term Memory" (LSTM) networks to enhance recommendations within session contexts. The CNN is employed to capture local interaction patterns between items within the session, whereas the LSTM component captures the sequential flow of user behaviour across items. In addition, the model introduces a dual attention mechanism, comprising item-level attention to focus on key interactions within the session and session-level attention to capture long-term preferences by weighting past sessions based on relevance. The proposed hybrid model utilizes item embeddings to represent items in dense vectors and applies CNN filters to capture short-term patterns in sessions, such as frequent item co-occurrences. The LSTM processes the output of the CNN to model long-term dependencies across the entire session. A final prediction layer with softmax activation predicts the next item in the session. The results indicate that HSRec outperforms traditional models by effectively learning both local and global dependencies in session data. The model’s effectiveness is assessed on different datasets, and the findings indicate that HSRec surpasses traditional methods, delivering greater efficiency in offering personalized recommendations to users.</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

Dual attention-based CNN-LSTM framework for session-based recommendations

  • 期刊论文

  • 发表日期：

    2025年7月5日

查看全文
  • Manisha Jangid,
  • Rakesh Kumar

摘要

Session-based recommendation systems (SRS) are essential in dynamic environments where user preferences evolve over time. This paper proposes a Hybrid session-based recommendation (HSRec) model that integrates "Convolutional Neural Networks" (CNN) with "Long Short-Term Memory" (LSTM) networks to enhance recommendations within session contexts. The CNN is employed to capture local interaction patterns between items within the session, whereas the LSTM component captures the sequential flow of user behaviour across items. In addition, the model introduces a dual attention mechanism, comprising item-level attention to focus on key interactions within the session and session-level attention to capture long-term preferences by weighting past sessions based on relevance. The proposed hybrid model utilizes item embeddings to represent items in dense vectors and applies CNN filters to capture short-term patterns in sessions, such as frequent item co-occurrences. The LSTM processes the output of the CNN to model long-term dependencies across the entire session. A final prediction layer with softmax activation predicts the next item in the session. The results indicate that HSRec outperforms traditional models by effectively learning both local and global dependencies in session data. The model’s effectiveness is assessed on different datasets, and the findings indicate that HSRec surpasses traditional methods, delivering greater efficiency in offering personalized recommendations to users.