Constructing core user set for recommendation based on relations between users and attributes of individuals
Gaofeng Cao, Li Kuang · 2016
With the rapid explosion of information, more and more people find it difficult to get the information that they really require and are interested in quickly and effectively.Recommender systems have been developed to effectively help users to make decisions to a certain degree.However, existing researches on recommender system mainly focus on improvingthe accuracy of recommendation by designingnew algorithms.In this paper, we propose a different idea to improve the efficiency, accuracy and novelty of recommendation by extracting core users who carry most of the information for recommendation.Specifically speaking, we propose new approaches to identifying core users based on interest similarity between users and users' own attributes.First, the interest similarity between all user pairs are calculated and sorted.The users' attributes are quantified and integrated.And then three strategies are used to select core users, which are frequency-based, rank-based, and fusion-sorting-based.In the experiment part, we compare our proposed methods with other existing methods from various perspectives, including accuracy, novelty, and long-tail distribution.Experiments show the effectiveness of our core user extraction method and prove that 20% of core users enable recommender systems to achieve more than 90% of the accuracy of the top-Nrecommendation.