Collaborative Topic Regression Recommendation Algorithm With LF-LDA And Social Relationships

Since entering the era of big data,the recommendation system has become an important tool to solve information overload.The core technology of the recommendation system is to model information such as user history,item content,and social relationships to infer user interest,and recommend items of interest to the user.After researching the traditional recommendation algorithms such as collaborative filtering algorithm and recommendation algorithm based on the topic model,two improvements are introduced to overcome the traditional disadvantages.Firstly,faced with the disadvantages of the traditional probability matrix decomposing model relying too much on the scoring matrix,a new method using the LF-LDA(Latent FeatureLatent Dirichlet Allocation)Extracting topic distributions is introduced,which integrates them into the item preferences extracted from the probability matrix decomposition to obtain new recommendation results and optimize the cold start effects.Second,a new method putting social relationships into collaborative topics regression,which uses the trust weight coefficient of a multi-layered circle of friends is established as the social factors containing the trust weights introduced into the calculation of the maximum likelihood function as penalty weights,optimizes the recommendation effect for new users.Based on the above two improvements,a new collaborative topic regression recommendation model and algorithm based on LF-LDA and social relationships is proposed,and we call it Social Latent Feature-Collaborative Topic Regression,abbreviated as SLF-CTR.Experiments were conducted on the lastfm and delicious datasets with strong sparseness and lots of phrases.Experiments show that the SLFCTR algorithm is obviously superior to traditional probability matrix factorization algorithm and collaborative topic regression algorithm.The improvement of the SLFCTR algorithm recommendation is proved.In the practical application of the current recommendation field,data sets are often very sparse,and along with the cold start of articles and users,the items content of which is often short,it can be seen that the SLF-CTR algorithm can be well applied to the actual application of the current recommendation system and has good research value.