Enhancing recommendation competence in nearest neighbour models

Collaborative Filtering approaches are viewed as essential tools to suggest products to users based on historical knowledge. Widely adopted user based Collaborative Filtering approaches rely on ratings provided by similar users of the target user to generate appropriate recommendations. The pair-wise user similarity is based on set of common rated items between users and identifying similar users is a challenging task when the set is small. To improve the recommendation quality in low correlated data, User Trait Model and Bayesian Global Agreement model are suggested in this work. The proposed models assign global agreement score to users. A linear function of any baseline user similarity and global agreement score of users is defined as a new similarity measure. From inception, recommendation approaches are keen on improving accuracy of recommended items and downplays the diversity of recommendations, which results in poor user satisfaction. The proposed models focus on improving accuracy and diversity of recommendations. Experiments are conducted on three benchmark data sets and the results are compared with other user based CF approaches suggested in the literature. The experimental results indicate that the proposed approaches outperform other user based CF approaches based on the evaluation metrics namely, MAE, RMSE, F1 for prediction accuracy, MN, ILD for recommendation diversity and Coverage for the extent of recommendations.