Research On Film Recommendation Method Integrating User Profile

With the rapid development of information technology and the wide application of the Internet,the amount of worldwide information has risen sharply,resulting in the phenomenon of information overload becoming more and more serious.Movie recommendation system is widely used in various movie platforms because it can help users quickly retrieve the resources users are interested in.However,the traditional movie recommendation methods solely rely on the user’s interactive information is not accurate enough in representing user preference and movie themes.It results in the final recommendation results can’t meet the user’s needs very well.To solve the above problems,this thesis uses user profiles to obtain users’ preferences,and integrating user profiles with existing movie recommendation methods to improve the accuracy of existing movie recommendation results.The main research contents of this thesis are as follows:(1)To solve the problem that the exist movie recommendation methods only use user ratings to model user preferences which are not accurate enough.In this thesis,user preferences are represented by building user profiles.We design a user profiles modeling method in the field of movie recommendation by using user interaction movie attributes,descriptions and comments information,and we design a user similarity calculation method based on the user profile.Based on this,we propose a collaborative filtering recommendation model IUPCF.This model first extracts the profile tags from the comments,descriptions and attribute information of the movies that the user interacts with,and then builds the user profile model based on these tags.Secondly,vectoring the user profiles using natural language processing technology,and calculating the similarity between users based on the vectorized user profiles.Then fusing the user similarity based on user profiles and traditional collaborative filtering based on ratings matrix.Finally,the fused user similarity is used for recommendation.Four datasets are used to validate the proposed model.The experimental results verify the validity of the proposed user profile modeling method and the positive impact of integrating user profile on model recommendation.(2)To solve the problem that existing graph neural recommendation methods usually ignore the importance of user and item initial embedding representation and the latent connection of side text information,this thesis proposes a light relational graph model Light RGAN that integrates the user profiles,item profiles and various text information.The model first uses the vectorized profiles model of users and items as the initial embedding of them.Secondly,user comments,item attributes and item descriptions are introduced as text side information,and a text embedding network based on multi-head self-attention network is designed to mine the latent connection of the texts that user interacted.Then,the light relational GCN(Graph Convolutional Network)is used to learn the embedding of users and items,and an attention network is added to calculate the weights of neighboring nodes.Finally,using a prediction network based on multi-layer perceptron to predict the interaction scores between users and items.We test our model on four datasets,and the experimental results show that Light RGAN has superior recommendation performance,and the recommendation effect is better than the baseline recommendation models.