Research And Implementation Of Time-aware Movie Recommendation Algorithm

With the rapid development of the Internet and information technology,the existing movies,videos and other resources on the network are still growing exponentially,and people have entered the era of "information overload".In this era,how to make users find the items they are interested in from the massive resources,at the same time,also make the products produced by producers stand out and are loved by the majority of users.Therefore,the research of recommendation algorithm came into being.In the current recommendation algorithms,the collaborative filtering algorithm as the most classical personalized recommendation algorithm has been concerned by the majority of scholars,while deep learning algorithm as the most popular research direction in recent years has been widely studied in various fields.However,the two recommendation algorithms almost do not consider how the time context information affects the recommendation,and do not take the time stamp information of user's behavior as the reference factor of recommendation.However,in real life,users' preferences are likely to change over time.Therefore,the research on the time context information of user behavior has an important impact on improving the performance of the recommendation algorithm.In this paper,based on the classical collaborative filtering algorithm,firstly,the time stamp of user's behavior in the data set is transformed into specific time information,and the time context information is taken into consideration as an important feature.Secondly,considering the sparsity of collaborative filtering algorithm,the matrix factorization model method is used to reduce the matrix dimension.Discrete time parameters are introduced into the matrix factorization model to recommend users for their time context to improve the accuracy.The parallel computing of collaborative filtering algorithm is realized by Spark cluster,which can alleviate the scalability problem caused by the increase of data volume.Considering that the data volume is still growing,the performance of traditional collaborative filtering algorithm may be limited,and for convolutional neural network algorithms,the more data they obtain,the better the performance they get from training.Therefore,considering the time context information,the convolutional neural network algorithm implemented by Tensorflow framework is used to recommend for users aiming at the larger data set,It not only improves the accuracy of recommendation and improves the efficiency of recommendation.In this paper,the collaborative filtering algorithm based on spark and matrix factorization,and the convolutional neural network algorithm based on tensorflowframework are used to recommend for users with considering time context information.The two data sets of movielens-100 k and movielens-1m are used as the data sources of the two recommendation algorithms respectively,and experiments are designed to realize the movie recommendation for users.By comparing the MAE,RMSE and MSE evaluation metrics of proposed algorithms with other recommendation algorithms,it illustrates that the algorithms proposed in this paper are actually effective,and have certain reference value for future research.