Sequential Recommendation Model That Combines Attention Mechanism with User Scoring Information

Xiujin Shi, Peiyang Liu, Guohao Sun, Chenyun Tao, Jiulei JiangICCV 2022 · DOI

Abstract

Most commonly used sequential recommendation models are based on behavior sequence data. In these models, they seldom consider the user rating factors and the different extraction abilities of a single attention extraction module for different sequence lengths. To overcome this problem, we propose a method to calculate the propagation ability of items and to correct ratings by SI (Sensitive Infected) module. In addition, we also propose a new attention mechanism that combines CNN and Transformer. In this new model (SRec-AU), we first apply the information dissemination model to calculate the dissemination capability of goods, and apply the dissemination capability to correct user ratings. Then, we use the attention module to extract the hidden information. Our proposed model embeds the modified user rating sequence and the original user behavior sequence, and then uses the attention layer to extract the feature information, and finally predicts the next item in the sequence by matrix factorization. The experimental results show that our method outperforms the state-of-the-art sequential models on both sparse and dense datasets and different evaluation metric. Experimental results show that the proposed SRec-AU outperforms the state-of-the-arts approaches (+8.51% in Reca11@10, +6.2% in NDCG@10, compared to the maximum value in baseline models). We also validate the effectiveness of dissemination model in improving the recommendation performance.
BibTeX
@inproceedings{shi2022sequential,
  author = {Xiujin Shi and Peiyang Liu and Guohao Sun and Chenyun Tao and Jiulei Jiang},
  title = {{Sequential Recommendation Model That Combines Attention Mechanism with User Scoring Information}},
  booktitle = {IEEE/CVF International Conference on Computer Vision},
  year = {2022},
  doi = {10.1109/cvidliccea56201.2022.9825095},
}

Loading related content...

Figure