Comparative study of similarity measures for item based top n recommendation

Abstract

The objective of the present work is to evaluate and analyse the total execution time of the generation of Top-N recommendation list by using the item based collaborative filtering(CF) approach. Different similarity measures are the key for the analysis. The user based collaborative filtering approach has some flaws, so item based collaborative filtering approach is taken into consideration. Different similarity measure like cosine based similarity, adjusted cosine similarity, extended jaccard co-efficient and correlation based similarity has been used to compare the total execution time of the item based collaborative filtering approaches. Behaviour of both the item based CF approach is analysed taking different similarities measures into consideration. For generation of Top N recommendation, dataset has been taken from the jester online joke recommender system. These datasets contains many users and about hundreds of jokes. This approach will predict the jokes (Prediction Problem (PP)) that the user is most likely that he/she may like. For prediction of jokes for the user the recommender system will look into the jokes that the users have previously rated or liked. By this recommendation it will be easier for user to choose the jokes which they may prefer to read. Recommender system (RS) is a personalized information filtering technology. Different similarity measure has been used so as to see how the algorithm behaves with differently. Main aim to find out the advantages and disadvantages of the algorithm and which approach takes the least time to generate the Top-N recommendation list using which similarity measure.