Scholarlink

<p>There is a growing emphasis in research on multi-criteria recommender systems, aiming to deliver personalized recommendations aligned with users’ diverse preferences. Simultaneously, session-based recommender systems (RSs) have gained prominence for providing immediate, context-aware suggestions based on users’ ongoing session behavior. Despite the attention directed at session-based RSs, existing research predominantly fixates on historical data or user reviews, neglecting dynamic multi-criteria considerations and evolving user intentions, posing a notable challenge. An additional complexity arises when trying to balance personalized recommendations with user privacy concerns in the handling of log files or historical data. Therefore, predicting overall product rating based on a sequence of multi-criteria ratings has become a new state-of-the-art for RSs. In this paper, we introduce a novel Recurrent Neural Network based model, CCSD-Init-ALSTM, constructed on a Long-Short Term Memory recurrent neural network architecture. This model incorporates an attention layer utilizing a Correlation Coefficient and Standard Deviation (CCSD) schema to calculate weights for each criterion within the sequence of multi-criteria ratings. Our methodology is designed to operate in two scenarios: firstly, in situations where exclusively explicit multi-criteria ratings are available, and secondly, in cases where users provide general permissible metadata within a recommendation context. Specifically, the ALSTM model is employed when metadata is accessible, while the CCSD-Init-ALSTM or CCSD-ALSTM models are utilized in scenarios involving explicit ratings. Extensive experiments on three real datasets were conducted for evaluation, revealing that the proposed method outperforms state-of-the-art session-based models.</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

A multi-criteria attention-LSTM approach for enhancing privacy and accuracy in recommender systems

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年4月11日

查看全文
  • Yahya Bougteb,
  • Brahim Ouhbi,
  • Bouchra Frikh,
  • El Moukhtar Zemmouri

摘要

There is a growing emphasis in research on multi-criteria recommender systems, aiming to deliver personalized recommendations aligned with users’ diverse preferences. Simultaneously, session-based recommender systems (RSs) have gained prominence for providing immediate, context-aware suggestions based on users’ ongoing session behavior. Despite the attention directed at session-based RSs, existing research predominantly fixates on historical data or user reviews, neglecting dynamic multi-criteria considerations and evolving user intentions, posing a notable challenge. An additional complexity arises when trying to balance personalized recommendations with user privacy concerns in the handling of log files or historical data. Therefore, predicting overall product rating based on a sequence of multi-criteria ratings has become a new state-of-the-art for RSs. In this paper, we introduce a novel Recurrent Neural Network based model, CCSD-Init-ALSTM, constructed on a Long-Short Term Memory recurrent neural network architecture. This model incorporates an attention layer utilizing a Correlation Coefficient and Standard Deviation (CCSD) schema to calculate weights for each criterion within the sequence of multi-criteria ratings. Our methodology is designed to operate in two scenarios: firstly, in situations where exclusively explicit multi-criteria ratings are available, and secondly, in cases where users provide general permissible metadata within a recommendation context. Specifically, the ALSTM model is employed when metadata is accessible, while the CCSD-Init-ALSTM or CCSD-ALSTM models are utilized in scenarios involving explicit ratings. Extensive experiments on three real datasets were conducted for evaluation, revealing that the proposed method outperforms state-of-the-art session-based models.