Ranking Based Approach for Noise Handling in Recommender Systems

Abstract

Collaborative filtering (CF), one of the successful social recommendation approaches, makes use of history of user preferences in order to make predictions. Even though this characteristic of Recommender System has attracted many applications, the quality of recommendations is still inclined by the unreliability of user provided data. In most Recommendation Systems (RS), users are asked to rate items explicitly or their behavior is monitored to collect their preferences. But in the real scenario, users may not provide genuine rating for all items of the data set. A genuine user must be knowing about the highly popular items of the domain. So the proposed approach assigns lesser popularity score to users who are not giving good ratings for highly popular items. Users with a popularity score less than a threshold are identified as noisy users. The experiments are conducted on real world data sets Movielens and Jester. The results claim that the proposed approach is effective in identifying and handling noisy users in the rating database.