A Tensor Decomposition Based Approach for Context-Aware Recommender Systems (CARS)

Recommender Systemsare used to suggest items of interest to users so that their overall browsing experience of the website is enhanced as well as they are not overwhelmed with the abundance of available information. The benefit of incorporating context in recommender systems is evident as the preferences of the users are highly dependent of the context in which they are making the decision.In our proposed approach, we have used context as an explicit feature to improve the recommendations so that it can adapt to the user's needs according to different scenario.We have extended the traditional two dimensional matrix factorization used in collaborative filtering to N-dimensional tensor factorization. Tensor appropriately models the different ratings given by a user to the same item in different scenario. The experimental results obtained using contextual variables proved to be of higher accuracy.