Discovery of user-item subgroups via genetic algorithm for effective prediction of ratings in collaborative filtering

Collaborative filtering (CF) is the most widely used technique in recommender systems for predicting the missing ratings that a user might have given to an item. In traditional CF all items are considered in the prediction process, which may include items irrelevant to the active user (the user for whom the prediction is for). Recently, subgroup based methods have emerged which take into account correlation of users and a set of items to rule out consideration of superfluous items with the assumption that two users with similar interests on a set of items need not have similar interests on other set of items. In this paper we propose four novel techniques through which subgroups of correlated items based on a set of similar users are formed so as to get predictions for only relevant items. With the contention that users in each subgroup have similar preferences on the subset of items consisting in the subgroup, we explore different methods in selecting highly correlated user-item subgroups to predict the ratings of the user/s for unseen items. The results thus obtained are analysed and the algorithm with the best accuracy is compared with state-of-the-art algorithms. Extensive experiments are performed on benchmark datasets like Movielens to analyze the quality of the proposed model. Popular accuracy metrics such as RMSE, MAE, MAP and F1-score are used to evaluate the proposed algorithms for both prediction of missing ratings as well as top N recommendation of items.