Adaptive Residual Attention Recommendation Model Based on Interest Social Influence

Existing social recommendation models mostly directly use original social data in the social space. However, original social data may contain a large amount of redundant and noisy social relationships. Additionally, existing feature fusion methods struggle to adaptively fuse features between nodes deeply, which can degrade the recommendation performance of the model. Addressing these issues, this paper proposes an Adaptive Residual Attention Recommendation Model based on Interest Social Influence. Firstly, we construct a novel Interest Social Mapping Module to model the confidence of social relationships based on user interests and map original social data to interest social space, thereby gaining a deeper understanding of user interest relationships in social networks. Secondly, we introduce a unique Social Selection Mechanism that dynamically filters and removes meaningless social interactions in the interest social space using social confidence scores, effectively filtering out social information that may interfere with or mislead users. Finally, we design an Adaptive Residual Attention Mechanism to flexibly adjust the feature fusion method of nodes, thereby obtaining more effective node information to improve recommendation accuracy. Experimental results show that compared to several state-of-the-art methods, the proposed model exhibits significant improvements on the Ciao and Epinions datasets.

@article{Fang2024,
  title     = {Adaptive Residual Attention Recommendation Model Based on Interest Social Influence},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {6},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {Sheng Fang and Xiaodong Cai and Yun Xue and Wei Lu},
  doi       = {10.14569/IJACSA.2024.0150671},
  url       = {https://doi.org/10.14569/IJACSA.2024.0150671}
}