An Improved Similarity Measure to Alleviate Sparsity Problem in Context-Aware Recommender Systems

Context-aware recommender systems (CARS) tend to incorporate contextual information while making recommendations, thereby enhancing accuracy and satisfaction. Similarity-based collaborative filtering is a very satisfactory and popular approach in this area. Typically, data sparsity problem in CARS becomes more severe when preferences are diluted with context factors in user-item preference matrix. Moreover, most of the researches have focused on utilizing traditional similarity measures to compute user/item similarity which is not suitable to sparse data and even did not utilize contextual conditions of the users. Traditional similarity measures suffer from co-rated item problem and do not consider global preferences. Therefore, this paper presents a new similarity measure and its variant which is suitable for CARS. Proposed measure utilizes contextual conditions, global preferences of the user behavior, and proportion of the common ratings. Subsequently, we applied them in similarity-based algorithms where each component is contextually weighted. The proposed algorithms are also analyzed for group of users. Recommendation results using two global context-aware datasets show that the proposed similarity measure-based algorithms outperform.