A Mixed Recommended Algorithm Combining User Properties and Item Features
Ning Xia, Shaofei Wu, Shenkuo Wang, Huajie Zhang · IOP Conference Series Earth and Environmental Science · 2020
Abstract In response to the problems of traditional recommended algorithm, such as data sparsity, changing of users’ interest, etc., this thesis proposes a mixed recommended algorithm blending User Properties and item features. First, propose a recommended algorithm based on User Properties after leading in users’ interaction factor and property weight in traditional similarity computation. Second, decompose the user-item matrix, and build the preference matrix and rating frequency matrix of user-item features. And then it proposes a recommended algorithm based on item features through leading in interest’s stable phase and improve the time attenuation function. Lastly, it generates a new mixed recommended algorithm after linearly combining these two algorithms. The Movielens dataset is used in this experiment, which turns out that the mean absolute error of the recommended algorithm proposed in this thesis significantly decreases comparing to traditional recommended algorithm.