Improved Pearson Similarity for Collaborative Filtering Recommendation System
Sravan Kumar Gadekula, Udai Pratap Rao, Ram Krishan Vyas, Amrutha Lakshmi Dontula, Suraj V Gaikwad · International Conference on Computing for Sustainable Global Development · 2019
Recommendation systems are used in many applications such as social media, e-commerce web sites, movies, books, search engines. Recommendation system filters out the information in which user is interested from a huge amount of data. For example, Amazon recommends items to the user according to their interest, from millions of users and items. Various recommendation system techniques have been proposed since mid-1990s, some of them are content based, collaborative, demographic, and hybrid. Among these techniques, collaborative filtering is popular due to its effectiveness (recommend similar items to the user according to its own interest). User based collaborative filtering and item based collaborative filtering are two generalized approaches under collaborative filtering. In this report, we propose a Weighted Pearson similarity metric using mahout that increases the similarity of items or users. This similarity is used for predicting the unknown ratings of users. Improved similarity of items or users helps in increasing the prediction (finding unknown ratings) accuracy. Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) are used to evaluate prediction accuracy. The proposed Weighted Pearson similarity gives less MAE and RMSE error rate than the traditional Pearson similarity.