An Enhanced Similarity Measure for Collaborative Filtering-based Recommender Systems

Abstract

In this era of the Internet and ubiquitous computing, huge information is being generated every moment. Obtaining useful information from the World Wide Web (WWW) has become too difficult. Recommender Systems appear to handle this problem of information overload to save user effort and time by recommending items of potential interest based on other similar users' ratings of the relevant items. Numerous algorithms have been proposed to recommend the potential items of interest. The most widely used methods include collaborative filtering (CF), content-based filtering and combination (hybrid) of two or more methods to get advantages of them. A CF-based recommender system method follows two main steps: computation of the similarity between two users/items and prediction of the unknown ratings to recommend items to a user. Several algorithms have been proposed for each of the steps. However, these methods may not be accurate in some situations, and hence the accuracies of predictions in CF-based RSs can be improved by overcoming those drawbacks. In this work, we propose a new method to compute the similarity between two users/items to overcome the shortcomings of the existing measures, and thereby improve the accuracy of prediction in the CF-based RSs. We evaluate our work, in comparison with other existing methods, based on recommendation accuracies in terms of different performance metrics such as Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE).