A Novel Approach to Define and Model Contextual Features in Recommender Systems

Recommender Systems(RS) provide more accurate and more relevant recommendations using contextual feature(s). This accuracy improvement is at the cost of computational expenses. Therefore, finding and selecting the most relevant contextual features is an important problem. Moreover, modeling and incorporating the selected contextual features in RS algorithms has an impact on both the accuracy and computational cost. We are conducting a series of studies to detect, define, select, model and incorporate the most relevant contextual features for RS algorithms. The feature detection, definition and selection approach involves the evaluation of features derived from implicit and explicit information. The selected features from this approach can be modeled and incorporated in any selected RS algorithm. In our recent works, we also propose a series of algorithms that incorporates multiple contextual features in the baseline matrix factorization (MF) algorithm. We use the selected contextual features to modify user biases and item biases in the baseline MF.