부산대학교 도서관

Recommendation systems (RS) are very popular for recommending items to its users on internet. Collaborative Filtering (CF) is a technique to recommend items to users on the basis of similarity between the users. Finding similarity between the users is very important challenge for recommendation systems. Two users are considering similar if the similarity between them is greater to a threshold value. The similarity threshold is minimum value of similarity between the users in CF based recommendation system to decide whether two users are considered to be similar or not. In this paper the performance of CF based recommendation system is analyzed on the basis of the similarity threshold. The performance is measured in terms of mean absolute error [MAE] which is calculated on a sample data set. MAE is calculated and compared at different levels of similarity threshold. The proposed work experimentally concludes the implications and benefits of deciding the similarity threshold and concludes that the high value of similarity threshold gives good results.