Research On Chinese Herbal Medicines News Recommendation Methods Based On Graph Embedding And Convolutional Neural Networks

Chinese herbal medicines news recommendation belongs to the vertical field news recommendation problem,but there are still few studies on Chinese herbal medicines news recommendation.Common news recommendation methods do not perform well in the field of Chinese herbal medicines news.On the one hand,news in vertical fields often contains information in a specific field,and general news recommendation methods cannot fully capture users' interests and preferences at the knowledge level.On the other hand,users will be interested in multiple concepts in news at the same time.The relationship between users and news is not only reflected in the level of domain knowledge but also in the level of textual semantic information.Using only a single level of information cannot fully capture the potential connection between users and news.Based on the above problems,this article starts research on how to use the knowledge of Chinese herbal medicines for news recommendation and how to combine the domain knowledge of Chinese herbal medicines with the semantic information of news for news recommendation.The details of the research are as follows.(1)Aiming at the problem of how to use the knowledge in the field of Chinese herbal medicines to recommend Chinese herbal medicines news,a method of Chinese herbal medicines news recommendation based on graph embedding is proposed.Firstly,a heterogeneous graph including users,news,varieties of Chinese herbal medicines,main producing areas of Chinese herbal medicines,month of production of Chinese herbal medicines,categories of Chinese herbal medicines and other concepts in the field of Chinese herbal medicines was constructed.Then,the embedded vector representation of nodes in heterogeneous graph is generated by using Deep Walk,a kind of graph embedding method.Then,the cosine similarity between user vector and news vector is calculated,and the Top-K news to be recommended is generated in descending order of similarity.In order to alleviate the problem of frequent updates of Chinese herbal medicines news in the process of recommendation,this article also proposes a cold start news representation method based on the concepts of the Chinese herbal medicines.Next,two Chinese herbal medicines news datasets are used to verify the proposed method.The results show that the proposed method performs well in user representation and news representation,and the recommendation performance in general news recommendation and news cold start recommendation scenarios is better than that of the baseline methods.Finally,specific examples are used for case study in order to show the recommendation results more vividly.(2)Aiming at the problem of how to combine the knowledge of Chinese herbal medicines with the semantic information of news to recommend Chinese herbal medicines news,a method of Chinese herbal medicines news recommendation based on convolutional neural networks is proposed.This method includes three parts: representation layer,learning layer and output layer.In the representation layer,user profiles are generated from the graph embedding level and the word embedding level,so as to capture user preferences more comprehensively.In the learning layer,a convolutional neural network structure is designed,which can integrate multiple levels of user features and news features,and mine the potential relationship between users and news.The output layer outputs the click probability of the user to the news.Two Chinese herbal medicines news datasets are used to verify the proposed method,and the results show that the proposed method is better than the baseline methods.Moreover,when using different data sets for verification,the proposed method has high stability while ensuring the accuracy of recommendation.However,the recommendation performance of baseline methods is quite different on different data sets.Finally,specific examples are used for case study in order to further demonstrate the rationality of the recommended results.