ABSTRACT
Recommendation system is the need of users to avoid dilemma while purchasing the products. Collaborative filtering, one of the filtering method of recommendation system provides, personalised help to users by analysing their purchase history. Collaborative filtering has traditional similarity correlation measures like pearson, cosine, adjusted cosine, jaccard, mean squared difference and constrained pearson. These measures give unsatisfactory performance for user with less and no purchase history i.e issues like sparse data and new user or item. This paper introduces new measure for user or item similarity. Performance analysis is done on standard datasets like movielense used by many collaborative filtering researchers. The experimental analysis shows proposed method outperforms existing traditional similarity measures. The proposed method efficiently handles cold start and data sparsity problems.
