An Efficient Graph Based Trust-Aware Recommendation System

With the increase in amount of information, it becomes important to build recommendation systems which can map and provide the relevant information based on the preferences, tastes and trust of users. The data clustering is applied in recommendation system to reduce the computational overhead. It has been shown empirically that with the increase in number of clusters, the rating coverage decreases monotonically. To reduce the impact of clustering, the rating prediction is computed in terms of the user similarity, trust and Jaccard similarity with each term having some coefficient to give them weights. The optimal weights are decided for each clusters which are finally used to make the recommendation. Calculation of optimal parameters is one of the expensive steps and they are fixed for each users of the clusters. In this paper, we dynamically compute the optimal parameters for each pair of users instead of using static optimal parameters for each clusters. The optimal parameters in the proposed approach are individually calculated for two users according to the ratio of Pearson, trust and Jaccard similarity between them. It helps us to reduce the complexity of the system as well as it results into increasing the accuracy of overall recommendations. Experiment results on real datasets illustrate that the our improved 2D-Graph method defeats the competing approaches based on accuracy and rating coverage.