Research And Implementation Of Movie Sequential Recommendation System Using User Feature

In recent years,the rapid development of information technology has provided people with many convenient and fast ways to obtain information,but also brought the problem of information overload.Therefore,people proposed personalized recommendation technology.In order to make better use of the information contained in the item interaction sequence,some scholars further proposed sequential recommendation models.However,some sequential recommendation models only used the id information of users and items when modeling user embedding vectors and sequence embedding matrices,so that the input of the recommendation model was not rich enough,which affected the performance of the model.In response to the above problems,this thesis proposes a Hierarchical Gating Recommendation Model Using User Features(HGRF),which attempts to incorporate user feature information into the user embedding vector and item position index information into the sequence embedding matrix.The gating layer utilizes the user embedding vector to preserve useful item information in the sequence.Experimental results show that richer user and item feature representations can indeed improve the performance of the model in terms of HR@10 and NDCG@10.The main work of this thesis is as follows:(1)A Hierarchical Gating Recommendation Model Using User Features is proposed.Besides user id,more user features such as age feature and gender feature are added to the user input of the model to describe the user portrait more accurately.After converting different features into different embedding vectors through different weight matrices,HGRF concatenate feature embedding vectors as the embedding representation of the user.Besides item id,the position index of the item in the sequence is also added to the item input of the model.After converting into different embedding matrices through different weight matrices,the two embedding matrices are added as a item sequence embedding matrix with position information.The gating layer utilizes the user embedding vector to preserve useful item information in the sequence.Then the short-term user interests are extracted from it by average aggregation.(2)Experiments are performed on the movie datasets.In this thesis,hyper-parameters tuning and ablation analysis of the HGRF model are carried out.The experimental results show that incorporating the user’s gender information,age information or occupational information can effectively improve the performance of the model,but incorporating the user’s geographic location information can’t improve the performance of the model.The combination of gender information and age information can bring more improvement.The user embedding vector needs a longer length to fully express richer user features.The ablation analysis shows that incorporating the position index into the item embedding representation can indeed improve the performance of the model.Finally,this thesis also conducts comparative experiments with several recommendation models.The experimental results show that the HGRF model outperforms other comparative models.(3)An online movie recommendation system is designed and implemented based on the HGRF model.The front-end and back-end of the system are separated from each other.The front-end adopts the React framework and the Ant Design UI component library,the back-end adopts the Java-based Spring Boot framework,and the recommended model adopts the Python-based Pytorch framework.Ordinary users can browse or search movies and rate them.They can also use the system’s various movie recommendation functions,which can recommend not only popular movies,but also highly rated and praised movies,and can also provide personalized recommendation results based on user historical rating records;administrators can manage movies and users separately,can remove movies in the system,can freeze,unfreeze or log out of ordinary users.In the end,this thesis also conducts various tests on each module of the system and records the test results.