CBCARS: Content boosted context-aware recommendations using tensor factorization

Matrix Factorization (MF) is widely used by researchers in Collaborative Filtering (CF) technique to generate recommendations. In literature it is used for predicting the missing ratings by approximating the two dimensional rating matrix. These predictions do not analyze items (item content) and users (user context). Integrating item content and user context increases the quality of recommendation. Incorporating item content and user context in recommendations calls for a modified version of MF technique as it can no more be represented using a two dimensional matrix. This paper proposes a hybrid RS known as Content Boosted Context-Aware Recommender System (CBCARS), by incorporating item content and user context using the principle of Tensor Factorization. A tensor which is a generalization of matrix in numerous dimensions is used to accommodate item content and user contextual information. Recommendations are generated using Tensor Factorization (TF) which factorizes the sparse user-item-context tensor to fill in the missing values. The performance of CBCARS is evaluated using Mean Average Error (MAE) and Root Mean Square Error (RMSE) with experiments conducted on restaurant dataset. Results show that CBCARS outperforms the three approaches of traditional MF, content-boosted and context-aware MF by significantly decreasing both RMSE and MAE.