Research On Movie Recommendation System Based On Heterogeneous Graph Neural Network

With the rapid development of the internet and the continuous growth of video users,movies have become one of the indispensable information media in people’s life.How to help users pick out the movies they are interested in through the recommended algorithm has attracted extensive attention.Currently,graph neural network has been found to be powerful in graph data learning and has been applied to recommended research work.Many researchers integrate user or project assistance information into large-scale network modeling to form heterogeneous graph networks containing various kinds of information.Most algorithms acquire similar users based on meta-path and extract effective information from network representation learning.However,there are still several problems as below.Current studies focus more on the interaction information between users and movies,but the modeling of preference influence relationship between individual users and attribute groups is insufficient.Users are affected differently by preferences of different aspects,and existing studies are not personalized enough for the fusion of multiple relationships in heterogeneous graphs.The existing heterogeneous graph structure models mainly focus on the interaction behavior between users and movies,and the interaction modeling between users and movies is incomplete.All these affect the accuracy of the recommendation system.So,the following studies are carried out in this thesis:1.To the problem of inadequate modeling of the relationship between users and attribute group preferences.In the heterogeneous graph neural network recommendation algorithm based on attribute information,attributes are abstracted as a group of people,and the relation between users and attribute groups is expressed by graph structure.Firstly,correlation between user and attribute group is calculated by using explicit feedback information,distinguishing the influence of different attributes group on the users,and extracting attribute preference.Then,the preferences under different relation in the heterogeneous graph are fine-grainedly fused through the attention mechanism.Finally,the inner product is used to get the user’s movie recommendation list,and it is verified that the method can effectively improve the recommendation performance on real movie data sets.2.This thesis improves a heterogeneous graph neural network recommendation algorithm that combines user opinions.It is mainly aimed at the incompleteness of interaction modeling between users and movies.Firstly,to construct multiple interaction subgraphs according to the time stamp fragments,and then model the interaction time and score information based on the interaction relationship of each subgraph to build the interaction time-effectiveness opinion.Secondly,to capture the collaborative signals under different time-effectiveness opinions by attention network,and transmit the information by graph neural network so that the collaborative signals can be fully combined with timeeffectiveness opinions.Finally,the embedded representation of each subgraph is fused to obtain a complete representation of behavior preference.The validity of the improved model for film recommendation is verified in real data sets.3.This thesis designs a movie recommendation system based on heterogeneous graph neural network.A movie recommendation system according to heterogeneous graph neural network is designed and implemented based on the above model.