Recommendation based on frequent N-adic concepts

Abstract

In social networks, many users tend to share items such as movies, books, songs and images by rating them with a series of discrete numbers or annotating them with a set of tags. Clearly, there are some semantic relationships among the users, items, ratings, tags and other information. Most of the past works only focused on some ternary relationships such as users-items-ratings or users-items-tags to make recommendations. But the ternary relationships which do not make good use of the given information are insufficient to provide accurate recommendations. In this paper, we propose a novel recommendation method based on frequent n-adic concepts which can mine the hidden conceptualization in the relationships. If there are tags, we model the relationships into the quadruples and if there are no tags, we also have some other information and model the relationships into the quintuples. Experimental results on MovieLens dataset demonstrate that our method has shown a significant improvement over the state-of-the-art recommendation approaches in terms of precision. © 2014 Springer International Publishing Switzerland.