Research On Movie Recommendation Algorithm Based On LDA Topic Model

The popularity of the Internet makes it very convenient for people to obtain movie resources.However,with the increasing richness of movie resources,it is difficult for people to find a movie they like conveniently and quickly.Therefore,for a video website,a good recommendation system is needed to recommend the movie to the user that may be interested to the movie.In order to allow users find their favorite movies accurately and timely,this article introduces the LDA topic model.The emergence of LDA was originally designed to solve the problem of text mining.The process of analog text mining,this paper treats users as documents,movies as words,and ratings of users to movies as word frequencies,in this way,the modeling process of document-word is migrated to the modeling process of user-movie in the movie recommendation system.The user-movie modeling process can be viewed as the user selecting a topic with a certain probability,and then selecting a movie based on the topic with a certain probability.After the model is built,this article proposes two improved strategies:Firstly,in order to further improve the recommendation quality of the LDA topic model recommendation algorithm,this paper proposes a learning method named negLDA based on negative samples.By learning the positive samples,LDA can extract the potential topic elements that the user likes,and then make positive predictive scores on the movies.By creating negative samples,the algorithm proposed in this paper extracts topic elements that the user does not like from the negative sample learning,and then make negative prediction scores on the movies,the algorithm not only considers the elements that the user likes,but also considers elements that the user does not like,so that the degree of user's likeness to the movie can be more comprehensively evaluated,and the effectiveness of the improved algorithm is also proved through experiments.Secondly,many researchers have proposed a probabilistic graph model based on the development and improvement of LDA.This paper focuses on two of them and has been inspired by them,this paper also proposes an improved probability graph model—dual topic LDA model(dtLDA),and applying it to the movie recommendation system.The dual topic LDA model adds the concept of user topic based on the original LDA model.The user-movie modeling process will become that a user selecting a user topic(or a user group,a user community)with a certain probability,and then choose a movie topic with a certain probability,and finally select a movie with a certain probability,this modeling process takes into account that a single user's interest will be affected by the small group,experimental results prove that the improved algorithm can improve the accuracy of the recommendation.