Research And Application Of Recommendation Algorithm Based On Knowledge Embedding And Attention Mechanism

The recommendation system can provide users with recommended items that meet their hobbies,and has a wide range of applications in the Internet,such as product recommendation,book recommendation and so on.However,the problems of data sparseness and poor interpretability are common in existing recommender systems.To solve these problems,it has been proposed to introduce knowledge as a kind of auxiliary information into the recommender system,which can not only alleviate the above problems,but also provide a certain degree of interpretability for the recommendation results.Knowledgebased recommendation is to transfer the relevant information aggregated in the knowledge graph to the recommendation system,so as to achieve the purpose of using knowledge to enhance the recommendation.Although the knowledge-based recommendation model has achieved good results,there are still some problems in the existing methods: First,the user preferences in the recommendation system are diverse and changeable,and the current mainstream models do not make full use of the relationship in the knowledge graph in the process of modeling user preferences,which affects the accuracy of the recommendation.Second,when using the graph neural network to embed the knowledge graph,it does not take into account that each edge has a corresponding type,and it is limited to learning the representation of nodes,and the degree of knowledge utilization is not enough.In order to solve the above problems,thesis has completed the following work:1.A preference modeling model is proposed,which combines knowledge with attention mechanism for modeling.The user's preference is associated with the distribution of relations in the knowledge graph to further abstract the user's implicit purpose.The importance of different relations in the implicit purpose is calculated by the attention mechanism,and the more important relations will be assigned a larger attention score.At the same time,independence constraints are introduced to encourage greater differentiation between different implicit purposes.The introduction of relations also provides some interpretability for the recommendation results.Compared with the baseline model that does not utilize the relationship,this model has achieved certain improvements in all five evaluation indicators.2.A new graph convolutional network model based on multiple relation types is proposed.In the process of aggregating knowledge using graph convolutional networks,corresponding operations will be performed for different relation types,so that the knowledge aggregation process can better collect information from knowledge graphs.Then combined with the implicit purpose of the user,the generated high-quality knowledge embedding is integrated into the recommendation system,so as to obtain a better representation of users and items,and improve the accuracy of the recommendation system.Thesis conducts a series of experiments on two real public datasets.The experimental results show that the method in thesis has reached the current leading level in the evaluation indicators.3.The model proposed in thesis is applied to the recommendation system to realize the recommendation of books.First collect the relevant information of the book,and then preprocess the book data to build the book knowledge graph.Then,according to the training model,a book recommendation system is implemented,and a rating-based recommendation system environment is built.Finally,the scoring prediction is realized,and several books are recommended for the corresponding users according to the predicted scoring results.