Research And Application Of Recommendation Algorithm Based On Recurrent Neural Network And Weighted Knowledge Graph

Aiming at the problem of information overload caused by data explosion,recommendation algorithm can find out the information that meets the user's interest from massive data according to the user's personal interest,so the research of recommendation algorithm is particularly important.Due to the sparsity problem and cold start problem of traditional recommendation algorithm,the recommendation effect is not good.As a kind of semantic network,knowledge graph contains the background information of items and the relationship between items.Therefore,the combination of knowledge graph and recommendation algorithm can effectively solve the sparse problem and cold start problem,so as to improve the recommendation effect.In recent years,researchers have found that deep learning can automatically learn effective feature representation from data,which makes deep learning play an increasingly important role in recommendation.Therefore,the text focuses on the combination of knowledge graph and deep learning to carry out recommendation algorithm research.In previous studies,when knowledge graph was used as a heterogeneous information network for recommendation,the relationship between entities and different attributes uses the same weight,which leads to the recommendation results easily affected by irrelevant attributes.Therefore,this paper proposes a weight knowledge graph method based on collaborative filtering.On this basis,the deep learning recurrent neural network(RNN)is used to mine user preferences,and a recommendation algorithm based on recurrent neural network and weighted knowledge graph(RNWKG)is proposed.The specific research contents are as follows:(1)This paper proposes a weighted knowledge graph method based on collaborative filtering.Firstly,the interaction matrix between users and items is learned by collaborative filtering algorithm based on items,and the set of similar pairs of items is obtained;secondly,the interpretable path is obtained by constructing a path in the knowledge graph to explain the similar information of items;finally,the interpretable path is integrated into the knowledge graph as a priori information,and the weighted knowledge graph is constructed.At the same time,experiments are designed to verify the effectiveness of the method.(2)In order to solve the problems of data sparsity,cold start and too much noise in the process of knowledge graph propagation caused by user preference,this paper proposes a recommendation algorithm based on recurrent neural network and weighted knowledge graph(RNWKG)by combining the advantages of embedded and path based knowledge graph recommendation methods.Firstly,the algorithm controls the propagation direction of user preferences in the knowledge graph by weight to remove the noise and avoid the huge training input samples,so as to improve the training effect and efficiency;secondly,the entity is embedded into the low dimensional vector by using the representation learning of knowledge graph,so as to fully retain the original semantic structure information of knowledge graph;finally,the preference data is analyzed by using the RNN deep level mining can predict the click probability of the user to the item,so as to generate the recommendation results.At the same time,through the comparative experiments on movie and book datasets,the effectiveness of the algorithm is proved.(3)Based on the above research,this paper designs and implements a movie recommendation system,applies RNWKG algorithm to the recommendation system,and provides personalized movie recommendation service for users.The system includes user module,administrator module and recommendation module.