Movie Recommendation System Based On Information Fusion Of Multiple Knowledge Graphs

In recent years,due to the explosive growth of online content and services,recommendation systems have been applied to various scenes in life,such as film and television recommendation,book recommendation,song recommendation,route introduction,and so on.In order to improve the quality of recommendations,a key factor is the ability to accurately characterize and understand user preferences.However,these preferences are dynamic and constantly changing in nature.Therefore,it is essential to model sequential interaction sequences in the recommendation.With the popularity of neural networks,many studies have tried to use sequential neural models,and sequential recommendations have made significant progress.However,the existing sequential recommendation models have the following problems: First,traditional Markov chains(Markov Chain,MC)and recent recurrent neural networks(Recurrent Neural Network,RNN)and self-attention mechanisms(Self-Attention,SA)can capture the preference of sequential patterns.However,these recommendation methods discard the timestamps of the items and only retain the order relationship of the interaction sequence,but different time intervals between items have different effects on the recommendation results.Second,the sequential recommendation model can capture the order relationship,but they are in the ability to capture complex user preferences is still limited,and it is difficult to capture fine-grained user preferences from the interaction sequence.Third,these sequential recommendation models will be subject to the sparseness and sparseness of user-item interactions like the classic model of collaborative filtering.The impact of cold start.Fourth,most sequential recommendation systems lack interpretability,and interpretability is also important in recommender systems.If a reason for recommending movies is given in the recommendation,it may increase user acceptance and recognition.To cope with these challenges,this article has done the following research:(1)To capture the dynamic and changing user preferences and learn the impact of different time intervals on the next project prediction,this article model the user interaction sequence as a sequence with different time intervals,and adds the time interval information to the self-attention mechanism model to form Sequential recommender to mine user’s historical preferences;(2)To enrich the multi-faceted information of item characteristics,solve the problems of cold start and matrix sparseness of the recommendation system,and enhance the ability to model fine-grained user preferences in an interpretable manner.The idea of this article is to integrate external knowledge into the sequential recommender.The incorporated knowledge should be rich and applicable to different scenarios to represent contextual information in different fields.This article will recommend interactive items in the system and existing knowledge graph entities Link to expand item attributes and use structured entity information to improve sequence recommendation.Therefore,to obtain the user’s preference representation more accurately,this article uses two knowledge graphs and movie tag information to enhance the item representation.However,the two knowledge graphs correspond to two different semantic spaces,and the use of data from the two parts of the knowledge graph may be restricted.In this article,the method of maximizing mutual information is used to bridge the semantic gap between the two knowledge graphs;(3)This paper studies the effect of adding auxiliary information on the performance of the model and shows that the two indicators are better than the baseline model.