Research On Movie Recommendation Algorithm Integrating Knowledge Graph

With the rapid development of the information age,a huge amount of data is generated on the Internet,and users cannot select their favorite items in a timely and effective manner.Therefore,the timely emergence of the recommendation system solves the problem of information overload and shines in various fields.However,the traditional recommendation system has data sparseness and cold start problems.The recommendation integrated with knowledge graph solves this problem to a certain extent,because the knowledge graph contains the information of the item itself and the connection between the items,which improves the recommendation performance.This thesis studies the recommendation algorithm based on knowledge graph fusion in the film field.The main research contents are as follows:（1）Constructing the knowledge graph of the film domain,including the establishment of the film domain model layer and its instantiated data layer.Among them,when establishing the data layer,the required data is extracted from the data source through knowledge extraction,and stored in the database using appropriate knowledge storage technology.（2）A new distance-based knowledge representation model PTrans D is proposed.Aiming at the problems that the Trans D model has many model parameters and the relationship between the two representations of entities is not constrained,the number of parameters is reduced by reducing the number of entity projections and clustering entities,and at the same time,the KL divergence is used to limit the entity projection and the corresponding entity class probability.The distribution is the same,and an improved PTrans D knowledge representation model is established.Model training adopts a dual-objective alternate optimization method of triplet loss and K-L loss.The experimental results in triplet classification and link prediction show that the recommendation performance is significantly improved in various indicators.In this thesis,the PTrans D model is used in the recommendation algorithm.（3）A movie recommendation algorithm PTrans D-CF,which integrates knowledge graphs,is proposed.The movie entity similarity obtained by the PTrans D model based on the knowledge graph and the movie similarity obtained in the collaborative filtering based on the user’s rating of the movie are weighted and fused to predict the rating.Generate recommendation lists to implement personalized recommendations.In the latter similarity calculation,user ratings are considered,and the same ratings for movies by users indicate higher similarity between movies.Compared with the traditional collaborative filtering recommendation algorithm based on the PTrans D model,the experimental results of the algorithm in this thesis have improved in the two evaluation indicators of accuracy and recall,which confirms the advanced nature of the algorithm in this thesis.