Scholarlink

<p>Signed network is an important class of social networks, where users can mark their neighbors as trusted or untrusted. In such signed social networks, users are more likely to accept the recommendation from their trusted friends and reject the recommendation from users they do not trust. Therefore, how to utilize the trust and distrust relationships between users in the recommendation system becomes very important. Existing recommendation methods for signed social networks face the problem that the aggregation process needs to reconstruct the original graph structure when embedding trust and distrust relationships, which inevitably introduces noise and affects the quality of embeddings. In addition, these methods treat neighbors with different signs equally when aggregating information from them, lacking a specific mechanism to deal with the different signs of connections, which further affects the quality of embeddings and limits the final recommendation performance. To address these problems, we propose a <b>T</b>rust and <b>D</b>istrust <b>E</b>nhanced <b>R</b>ecommendation (TDER) algorithm for signed social networks, which captures higher quality trust and distrust embeddings and improves recommendation performance in signed social networks. Our proposed approach updates trust and distrust embeddings of the target user by aggregating from the positive, negative and neutral neighborhoods without reconstruction operations. TDER utilizes signed attention mechanism and mutual information maximization to bring the potentially trusting users closer and push the potentially distrusting users further apart in the embedding space. Through the collaborative deep learning strategy, trust and distrust embeddings and rating information are fused for rating prediction. Comprehensive experiments demonstrate that the proposed approach can generate high-quality trust and distrust embeddings between users without changing the original graph structure, and its recommendation performance outperforms the state-of-the-art methods.</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

Trust and distrust enhanced recommendation algorithm for signed social networks

  • 期刊论文

  • 发表日期：

    2026年7月22日

查看全文
  • Yongyin Han,
  • Rui Wang,
  • Renhui Wu,
  • Xiaobin Rui

摘要

Signed network is an important class of social networks, where users can mark their neighbors as trusted or untrusted. In such signed social networks, users are more likely to accept the recommendation from their trusted friends and reject the recommendation from users they do not trust. Therefore, how to utilize the trust and distrust relationships between users in the recommendation system becomes very important. Existing recommendation methods for signed social networks face the problem that the aggregation process needs to reconstruct the original graph structure when embedding trust and distrust relationships, which inevitably introduces noise and affects the quality of embeddings. In addition, these methods treat neighbors with different signs equally when aggregating information from them, lacking a specific mechanism to deal with the different signs of connections, which further affects the quality of embeddings and limits the final recommendation performance. To address these problems, we propose a Trust and Distrust Enhanced Recommendation (TDER) algorithm for signed social networks, which captures higher quality trust and distrust embeddings and improves recommendation performance in signed social networks. Our proposed approach updates trust and distrust embeddings of the target user by aggregating from the positive, negative and neutral neighborhoods without reconstruction operations. TDER utilizes signed attention mechanism and mutual information maximization to bring the potentially trusting users closer and push the potentially distrusting users further apart in the embedding space. Through the collaborative deep learning strategy, trust and distrust embeddings and rating information are fused for rating prediction. Comprehensive experiments demonstrate that the proposed approach can generate high-quality trust and distrust embeddings between users without changing the original graph structure, and its recommendation performance outperforms the state-of-the-art methods.