Na?ve Random Neighbor Selection for memory based Collaborative Filtering

摘要

Collaborative Filtering (CF) is one challenging problem in information retrieval, with memory based become popular among other applicable methods. Memory based CF measure distance/similarity between users by calculating their rating to several items. In the next step system will predict user rating with specific algorithm e.g. Weight Sum. One similarity measurement that often used is Pearson correlation. Since CF used many (almost all) users and items, Pearson correlation suffer on time and space complexity. To overcome this problem, CF that used Pearson correlation often selects some user to be used as neighbor. The mechanism itself, never mention clearly. In this paper, we introduce Nai?ve Random Neighbor Selection mechanism. Our research show that best performance achieve at parameter combination of Pearson Correlation Threshold = 0.1 and Number of Neighbor = 21 that shows MAE = 0.791 that placed on the third position among other algorithm.