A Recommender Systems Based on Similarity Networks: MovieLens Case Study

Recommender systems help consumers to have proper choices regarding personal preferences. Well-known recommender approaches generally benefit from profile similarity, items ratings, behavioral profile or other Hybrid filtering techniques. One of the important drawbacks of previous works is a lack of understanding similarity between users' opinion and items' taste beyond considering feature based similarity measure. In this paper, a novel hybrid recommender system proposed by applying network science techniques to overcome those shortages. We build weighted networks corresponding to users' opinion and items taste for tagging their related groups in order to measure similarity weights between groups of users and items. In the next step, users and items are clustered to similar groups based on individual features. Linear combination of weights generated from the system is used to predict rates for recommended items to users. Experimental results on MovieLens datasets show that the proposed approach improves recommendation performance and agility considerably.