Capturing Contextual Influence in Context Aware Recommender Systems

In the present evolving phase of information technology, Recommender systems (RSs) have been established as widely accepted platform for handling & managing the information overload problem. In order to facilitate an individual user in decision making for selection of any product or service, user preferences are first captured implicitly or explicitly and a predictive model is built to derive personalized recommendations. Data Sparsity, malicious / biased data by users, diversity in recommendations, temporal dynamics, etc are few of the key challenges experienced by new age recommendation systems. Dimensionality reduction techniques help to decompose the rating matrix in the form of latent factors with lower ranks and attempts to solve the data sparsity problem. Matrix Factorization (MF) is the well-accepted technique in this aspect. In order to generate more accurate and meaningful recommendations, Context aware RS (CARS) is the new emerged technique in recommender systems. Tensor Factorization or generalized matrix factorization facilitates the most suitable and generic way of integrating contextual information into RSs. The relevant contextual information will always improvise the performance of the recommender system but irrelevant contextual information could degrade it drastically. The work presented here, provides the discussion over comparison of Recommender System's performance at various scenarios such as generating recommendations without considering any contextual information and with considering the contextual information. There are few techniques available to find out the context relevancy in CARS. Their impact on the performance of RS is also discussed.