A Jaccard base similarity measure to improve performance of CF based recommender systems

Abstract

Listen

Translate article

Revolution in social computing has resulted in the wonderful evolution of recommender systems. Recommender systems maintain a repository of user profiles, created by a community of users, for generating personalized recommendations aimed at individual users. One of the approaches used in recommender systems is collaborative filtering (CF) which has become one of the most famous approaches for providing personalized recommendations to users. Nearest neighbors based methods used in CF are being widely used by many online stores to enhance users shopping experience. Nearest neighbors-based CF methods use some similarity measure techniques to find similar users/items for an active user/item. Almost all similarity measurement methods use ratings of commonly rated items while calculating similarity between a pair of users/items. Our approach works in the same manner as Jaccard similarity works. But Jaccard similarity does not consider the absolute value of rating and only considers the ratio of co-rated items. We take into account the ratio of absolute rating values which are equal in value, to the total no of co-rated items. An additional argument we take into account is the average rating value of users. We compared performance of our proposed method with many state-of-the-art similarity measures. Recommendation results from a set of real data sets show that our proposed measure has some performance improvement in terms of Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).