An Efficient Recommender System Technique in Social Networks Based on Association Rule Based Mining

Abstract

A recommender system is an information filtering system that has become a buzzword in various areas of marketing and research such as movies, music, books, products and research articles. The main role of recommender systems is to guide users on a personal level to provide an optimum set of suggestions based on the users’ taste, explicit rating of items, his/her demographic and other related valuable information. In the past decade, several approaches have been discussed for recommendation of items to online users keeping in mind the accuracy of prediction, the cold-start problem and the problem of sparsity. However, most of these existing recommender system techniques have failed to define a proper recommendation model that can be well suited for any online social network and can consider majority of the limitations when modeling real-market recommendations. In this paper, we present a novel efficient recommender system technique RecGyp and many other standard commonly used existing prediction techniques and also perform experimental evaluations to make a comparative analysis among each technique. The experiments carried out on the MovieLens 100K and the Yahoo! Webscope Movie datasets demonstrate the superior nature of the proposed RecGyp technique in solving the scalability issue and accuracy of results for recommending items to users of a social network. In addition to the traditional similarity measurements evaluation, results are also provided for three evaluation metrics, Precision, Recall and ROC, to evaluate the accuracy of all the recommender system techniques.