Design Of Movie Recommender System Based On Big Data And Deep Learning

In the era of mobile Internet,while people enjoy rich network services,they are also troubled by redundant and inefficient information.The recommender system can generate accurate personalized recommendations for users by mining the relevant information of users and items,which can solve the above problems to a certain extent.In recent years,the development of deep learning technology has promoted the rapid evolution of recommendation algorithms.At the same time,it also puts forward higher requirements for the feature data of the recommender system.In order to meet the requirements of recommendation algorithms for massive features and real-time processing of data,it is necessary to use big data tools to process data and information.Based on big data and deep learning technology,this paper constructs recall algorithm and sorting algorithm and designs a complete recommender system with the theme of movie recommendation.Firstly,the recall algorithm uses item and user portrait information as the main feature data,and designs a multi-way recall strategy to make the recall results more comprehensive.The offline recall algorithm is mainly based on the Item2 vec model recall,which is used to ensure the recall rate requirements,and cooperates with the label inverted index recall and statistical recall to solve the cold start problem and increase the diversity of recommendations.The fusion ratio of the three algorithms is determined through fusion experiments,then another experiments verify the effectiveness of multi-channel fusion recall in terms of recall rate,coverage rate and other indicators compared with single model.In addition,a real-time recall module based on user instant behavior and new hot item recommendation is designed to quickly complete the recommendation.Secondly,in order to fully mine the behavior sequence features of users,a deep learning model is used to construct a ranking algorithm.The user behaviors are divided into long-term behaviors and short-term behaviors to be modeled respectively,and the LSIN model is proposed on this basis.The user's long-term behavior first uses the self-attention mechanism to extract features,and then they are input into the GRU unit together with the short-term behavior for sequence modeling.In addition,the user's long-term features and short-term behaviors also calculate the attention score through the basic attention mechanism.After splicing through the MLP layer,the final click-through rate is predicted.The effectiveness of the LSIN model compared to several mainstream ranking models is demonstrated on the Movie Lens and Netflix datasets.Finally,with the Movie Lens data set as the initial data,combined with big data processing technology and Web technology,this paper completes the construction of the recommender system based on the above recall and sorting algorithms.Using the Lambda architecture,a data flow framework for recommender systems is built through offline systems and real-time systems.From the perspective of engineering design,this system completes the entire recommendation business process,including data collection,feature processing,recall set construction,recall item sorting,and return of recommendation results,thus forming a complete closed loop of the recommendation business flow.