Personalized Movie Recommendation Algorithm Based On Knowledge Graph

The development speed of human science and technology is increasing geometrically.Every technological upgrading brings great changes to the society.There are lots of messages on mobile devices makes it difficult for us to choose.The recommendation system has been in-vented to help customers better choose their favorite information according to their data,which makes customers satisfied.However,the sparseness and cold boot troubles of tra-ditional recommendation technique limit the effectiveness of some recommendation tech-niques to some extent.This thesis first proposes a collaborative filtering algorithm based on knowledge graph based on the traditional item-based collaborative filtering algorithm;and also proposes a deep autoencoder recommendation algorithm based on knowledge graph,which introduces knowledge graph information and improves it through auxiliary informa-tion.Performance and final performance of the recommended system:(1)This thesis proposes a collaborative filtering recommendation algorithm based on Knowl-edge graph,which is called KGCF method.It gives its complete framework and related pro-cesses.The core thought in this method combines the vectorization representation method of project information in knowledge graph with the CF algorithm,and make up for the problem of data sparsity by using the auxiliary information from knowledge graph that CF algorithm based on item does not need to consider its own relevant messages.The purpose of this action is to increase the effects of personalized recommendation.(2)A deep autoencoder recommendation model based on knowledge graph is proposed,which is called KG-DAE.This method combines deep autoencoder,movie knowledge graph and at-tention mechanism together to obtain better recommendation results.Experiments on movie-lens 1m data set show that KG-DAE algorithm can improve the recommendation effect In addition,knowledge graph auxiliary information can be used as input vector for initial score in the era of data sparsity,which can alleviate the problem of insufficient recommendation accuracy caused by data sparsity to a certain extent.(3)The experiment uses movielens data set,and compares the technique which is applied in this thesis with other experiments.The first experiment shows that the technique which is applied in this thesis is superior to other techniques in F1,which shows that this technique has some improvement in movie recommendation.The second experiment shows the AUC and MSE of KG-DAE algorithm is better than baseline method.