Research on personalized news recommendation based on dynamic knowledge subgraph perception
Guang Wang, Yue Sun, Danni Li · 2023
Aiming at the problem of how to effectively construct the interpretability of time-sensitive personalized news recommendation, a dynamic knowledge subgraph perception research method is proposed. Firstly, the dynamic knowledge subgraph is extracted by using the time neighborhood perception technology. For each news, a dynamic knowledge subgraph covering the core entities in the news and the neighbor entities related to the news in the knowledge graph is extracted. Secondly, the CNN and Transformer technology sets are used to form a semantic collaborative encoder. The user 's personalized features and news content are encoded by the user news collaborative encoder. Finally, the matching score is calculated by the correlation model and model training. The larger the score, the more interested in that type of news. Compared with the independent modeling of user interest and candidate news, experiments are carried out on the data sets MIMD and Feeds, and four indicators of AUC, MRR, nDGG @ 5 and nDGG @ 10 are used for evaluation. The experimental results show that the proposed network model has great advantages over other benchmark models, making the recommended results more interpretable.