An item recommendation algorithm based on matrix decomposition and topic model

Abstract

Traditional collaborative filtering models are often faced with sparse user data, and the relationship between users and items is not clear enough so that the accuracy of user-item rating prediction is still not high. According to the idea that LDA model can mine hidden information in user comments and Probability Matrix Decomposition model can alleviate user data sparse, this paper proposes IRCMT algorithm. Firstly, IRCMT algorithm using the LDA model for the user(item) - topic distribution, thus obtains the user (item) theme similarity. Secondly, the user(item) topic similarity can be incorporated into probability matrix decomposition, resulting in improved user (item) feature matrix, and predict refactoring score matrix, finally combining items-theme similarity to calculate for the final prediction score, Office Products dataset experiments show that the algorithm in terms of rating prediction accuracy is superior to most of the traditional recommendation algorithm and prove that users' comments have great research value.