An Improved Collaborative Filtering Algorithms Based on Maximum Entropy Model and Markov Process to Mine User Behavior

Abstract

Aiming at problems of data sparsity and poor interpretability of recommendation reasons in user-based collaborative filtering algorithm, an improved collaborative filtering algorithm based on maximum entropy model and markov process is proposed to mine user behavior habits. Firstly, the maximum entropy model is established through the user-behavior matrix of collaborative filtering algorithm, and the user behavior is fuzzily and preliminarily mined. Secondly, fuzzy behavior information obtained by the maximum entropy model is introduced as the initial state of markov process, and an optimization model is set up according to the internal relation of user-behavior matrix, thus reaching the state probability transition matrix. Once the markov process reaches a stable state, the user’s final behavior habit is excavated. Finally, this user behavior is integrated into the recommendation process, so that the interpretability of the recommendation is enhanced. The experimental results show that the improved collaborative filtering algorithm can promote the recommendation accuracy.