Research On Personalized News Recommendation Model Based On Knowledge Graph

The development of online news platforms makes it easy for users to get news information,but it also brings the problem of information overload.In this context,personalized news recommendation emerged and received wide attention.In recent years,as auxiliary information on the item side,knowledge graph has been widely combined with academic research and industrial applications in the field of personalized recommendation.News language is concise and contains part of knowledge entities,so knowledge graph can be used as auxiliary information to introduce news recommendations and enrich news representations.However,the combination of knowledge graph and news recommendation mainly relies on single information of news(such as titles),while the method of using multiple types of text information often ignores the association of news at the knowledge level.To solve these problems,this paper builds a personalized news recommendation model based on knowledge graph.Text is the main part of news,so this paper starts from the text content,analyses the characteristics of each part of news.Considering that the title and abstract cover the most critical information and contain the main knowledge entities,so through the word vector representation technology and knowledge graph embedding to vectorize it preliminarily,including word vector representation,entity vector representation,entity context vector representation.Secondly,it builds a personalized news recommendation model based on text features,which mainly includes three sub-modules:multi-view news feature representation,user click behavior representation and click probability prediction.In the multi-view news feature representation sub-module,the convolutional neural network is used to extract the features of the news title and abstract respectively,and the LDA topic model is used to obtain the potential topics in the news body.After that,the fully connected neural network is used to extract the news features from the perspectives of news categories,sub-categories and bodies.The attention mechanism is selected to fuse the features of each perspective of the news to obtain a unified news representation.In the user click behavior representation sub-module,the attention mechanism is used to fuse the user’s news Click history to obtain the corresponding user representation.In the click probability prediction sub-module,the click probability of the user to the candidate news is obtained by dot product operation,and the corresponding TOP-N recommendation list of the user is generated in descending order.Finally,four groups of control experiments are constructed to verify the overall performance of the model and the effectiveness of each main part on the news data set.The experimental results show that the personalized news recommendation model based on knowledge graph proposed in this paper can better complete the task of news recommendation.Compared with the baseline model,the effect of the model is improved.AUC,MRR,nDCG@5,nDCG@10 are increased by 3%、2.6%、2.5%、3%on average.In addition,the ablation experimental results show that the rational use of knowledge graph,multi-view news features and attention mechanism can improve the effect of the model to a certain extent.This paper provides ideas for the combination of knowledge graph and personalized news recommendation,provides accurate news recommendation for users,improves user satisfaction and improves platform service ability.