Recommender System using Collaborative Filtering and Demographic Features

Abstract

Recommender systems use variety of data mining techniques and algorithms to identify appropriate preferences of items for users in a system out of available millions of choices. Recommender systems are classified into Collaborative filtering (CF), Content-Based filtering (CBF) and Knowledge-Based filtering (KBF) and Hybrid filtering (HF) systems. In present scenario recommender systems are facing many challenges like data sparsity, cold start problem, scalability, synonymy, shilling attacks, gray sheep and black sheep problems etc. These problems and challenges consequently affect the performance of recommender systems to a great extent. Among these cold start problem is one of the challenges which comes into scene when either a new user enters into a system or a new product arrives in catalogue. Both situations lead to difficulty in predicting user preference in the absence of availability of sufficient user rating history. The research work in this thesis is based upon exploiting user demographic characteristics for finding similarity between new user and already existing users in the system. The efficiency of recommender systems can be improved by proposed Hybrid recommender system which calculates recommendations for new user on the basis of his / her demographic attributes like age, gender, occupation similarity of existing users in the system. The proposed approach results in generation of more relevant and accurate recommendations as compared to traditional methods of finding recommendations. The work is based upon problem domain related to online movie recommendation using MovieLens dataset.