A collaborative filtering recommendation based on users' interest and correlation of items

Abstract

Collaborative filtering (CF) is one of the most commonly used recommendation technologies in the recommender systems of e-commerce. However, due to the sparsity of users' rating data and the single ratings similarity, traditional CF algorithms show certain shortcomings. Aiming at these problems, a CF recommendation algorithm based on users' interests and the correlation of items is proposed. By using the algorithm, the similarity of users is measured according to users' interests based on the categorical attributes of items, while that of items is computed by introducing the association rules of data mining. The results of the tests on Movielens dataset manifest that the modified algorithm presents higher recommendation accuracy than the traditional CF algorithms.