Predicting Typical User Preferences Using Entropy in Content based Collaborative filtering System

Abstract

Recommendation techniques using collaborative filtering have the sparsity problem because recommendations are made based on thousands of user preferences on items they click. Another problem is that these systems give recommendations based on profiles match between only two users. The collaborative filtering techniques using data accumulated from users’ mouse clicks contain inaccurately rated information. Therefore user preferences cannot be automatically regarded as accurate data, so users within the matrix in collaborative filtering system need to be optimized by using entropy. The proposed method is capable of optimizing collaborative users, in which users are grouped according to the vector space model and the K-means algorithm to solve the sparsity problem. To extract features of documents, the proposed method uses the association word mining method with essential word in content based filtering to solve the problem of disambiguation and the multidimensional problem. After grouping, the typical preference can be extracted by assigning typical user preferences in the form of weights. This method reduces the inaccuracy of recommendations based on preferences that unproven users rate. In addition, it saves the time for retrieving a similar user within a group. It thus enables dynamic recommendations.