Improved CF based prediction technique for recommendation systems

Any eCommerce websites use recommendation system to recommend items to users. Collaborative filtering is a technique to recommend an item to the customers by understanding the past behavior of the same user and other similar users. The accuracy of the recommendation system is a major issue while recommending items to users. In this paper a research is done to propose a new system to predict ratings for items for the users by finding similarity between the users. Similarity between the users is found by analyzing the previous history of the users for rating items. A similarity matrix is created that store a similar weight between users. Similar users are selected if the similarity weight between the users is found greater than a similarity threshold. The proposed system is implemented on a data set and the quality of the proposed system is analyzed by comparing the value of mean absolute error (MAE). The experimental results are found better than some other existing techniques. The value of MAE is approximate 11% better and value of RMSE is 15% better as compared to existing algorithms.