Research On Movie Recommendation System Based On Deep Learning

With the rapid development of Internet technology,while bringing convenience to people,it has also generated a huge amount of data,making people face the dilemma of difficult choices.Recommendation systems are able to filter out content of interest to users from the vast amount of information based on user profiles and generate personalised recommendation lists to users,thus effectively solving the problem of information overload.In recent years,recommendation algorithms based on deep learning have become dominant.In movie recommendation,the introduction of deep learning models to extract user behavioural sequence features and accurately portray user profiles can help improve the recommendation effect.Recommendation systems are designed to balance accuracy as well as efficiency,so a hierarchical filtering model is usually used to divide the recommendation process into two steps:recall and sorting.In this paper,recommendation algorithms for the recall and sorting layers are investigated separately for these two steps.The main work is as follows:(1)Recall layer model research: The goal of the recall layer is to quickly recall items that users may be interested in from a large set of items.Obtaining users’ precise preferences is the main problem faced at the recall layer,and this paper proposes a recall model that fuses users’ long-term and short-term preferences.The model introduces a self-attentive mechanism to learn users’ short-term interest preferences,uses a gated recurrent neural network to learn their long-term interest preferences,and finally fuses long-term and short-term interests through a gated unit module.The model is able to accurately and comprehensively characterise the user,resulting in a more valuable list of recalled items.Experiments on publicly available datasets show that the model achieves better results in two evaluation metrics,Hit Rate(HR)and Normalized Discounted Cumulative Gain(NDCG),compared with current similar models.(2)Research on the ranking layer model: The goal of the ranking layer is to fine-rank the candidate items initially screened by the recall layer.In this paper,an interest extraction ranking model based on the user’s historical behaviour sequence is used to fully exploit the dynamic change information present in the user’s behaviour sequence using deep neural networks.Firstly,the behavioural sequence features of users are extracted using the multi-headed self-attentive mechanism,then the sequence interaction is modelled by GRU network to learn the dynamic changes of user interests,and finally,the degree of influence of user behavioural sequence features and user interest interaction features on the films to be recommended is calculated by the interest activation layer.In this paper,the AUC value is used to evaluate the model,and experiments on public datasets prove the effectiveness of the ranking model.(3)Movie recommendation system design: Based on the algorithm research of this paper,a movie recommendation system is designed.The recall layer model and the ranking layer model were first deployed to the server side,then the Flask framework was used to encapsulate the model as a backend service and provide a service interface for the application layer to call,and finally the front-end application was designed to be able to call the algorithms studied in this paper and generate personalised movie recommendations for users.The application layer first uses a recall model to recall the top K movies of interest to the user,then uses a sorting model to sort these K movies according to the predicted click-through rate values,and finally selects the top N movies as a recommendation list for the user.In this paper,the corresponding models are proposed for the task requirements faced by the recall and sorting layers.And the proposed model is applied to design a movie recommendation system.