Weighted Percentile-Based Context-Aware Recommender System

Context-Aware Recommender Systems (CARS) focus on the improvement of accuracy and user’s satisfaction by incorporating context features while making recommendations. Using too many context features aggravate the data sparsity problem and may impair predictive performance while few context features fail to capture the contextual effects. Though genre-based ratings called implicit ratings play an important role while making recommendations, despite that most of the studies have focused on utilizing explicit ratings. Addressing these issues, we propose a novel framework that demonstrates multiple rating prediction algorithms based on user neighborhood and item neighborhood approaches exploiting explicit and implicit ratings. The algorithms incorporate context communities to alleviate the data sparsity problem. We have also used weighted percentile method to increase the precision. Furthermore, we extended our research to Group Recommendations to see the effectiveness of the proposed algorithms. Finally, the results using two datasets indicate that the proposed context-aware weighted percentile algorithms are superior than the baseline approaches. The item neighborhood-based approaches are more accurate than user neighborhood-based approaches and the performance of explicit and implicit ratings are dataset dependent. The results obtained also prove the effectiveness of the algorithms for Group Recommendations.