A personalized authoritative user-based recommendation for social tagging

Abstract

Personalized information recommendation based on social tagging is a hot issue in academia nowadays, butthe concept of an authoritative user has not been emphasized in the existing literature. This paper first proposes a method to determine user authority in a social tagging system, in which the quality authority and quantity authority of users are calculated from a user co-occurrence network, which is derived from users’ participation in the social tagging system. Degree centrality is employed for the user authority calculations, which are taken as weights for tag voting. On this basis, a resource model is constructed by summing up the tags from each user and their corresponding weights to represent each resource in the collection. User models are then obtained based on the resource models, and cosine similarity is used for making resource recommendations to users. An experiment was conducted on a dataset crawled from Delicious.com. The results show that the average GP relevance of the authoritative user based algorithm reaches 0.6115 much better than two benchmark algorithms.