融合跨平台用户偏好与异质信息网络的推荐算法研究OA北大核心CHSSCDCSSCICSTPCD

[Purpose/Significance]This paper proposes a recommendation algorithm that integrates cross-platform us-er preferences and heterogeneous information networks,based on the heterogeneous big data of cross-platform users.It plays a significant role in alleviating the sparsity and cold start problems of personalized recommendation.[Method/Process]Initially,the paper constructed a user core interest social circle based on cross-platform heterogeneous informa-tion,captured user information preference features in both the source and target platforms through convolutional neural net-works and self-attention mechanisms.Subsequently,it built a heterogeneous information network based on the core interest network and the relationships among recommended items,and it employed a heterogeneous graph attention network model for feature aggregation.Finally,the study integrated the above feature embeddings into an improved matrix factorization model to compute recommendation scores.[Results/Conclusion]The model demonstrates superior performance across four independently constructed cross-platform datasets.This study not only fills the gap in cross-platform,multi-attribute,and fine-grained datasets in the field of recommendation but also enhances the theoretical and methodological system related to recommendation by introducing cross-platform features.