A paper accepted in Knowledge-Based Systems 2021

Title: “How to Get Consensus with Neighbors?”: Rating Standardization for Accurate Collaborative Filtering
Author: Hong-Kyun Bae, Hyung-Ook Kim, Won-Yong Shin, and Sang-Wook Kim
Abstract
In this paper, we present neighbor-aided rating standardization (NARS), a new framework for rating standardization that leverages the ratings of users’ neighbors for more accurate collaborative filtering. Our approach is motivated by the insight that users tend to give ratings to items according to different criteria of their own, which causes the accuracy degradation in item recommendation. Our NARS framework intelligently alleviates the difference in rating criteria among all users through rating standardization in the context of consensus with neighbors. Consensus, referred to as the process of reducing disagreement in rating criteria among users, is facilitated by effectively aggregating the ratings of all users. Consequently, the ratings adjusted with the unified rating criterion among all users (i.e., standardized ratings) can be found via an iterative consensus process and are used as input of CF for top-N recommendation. Experimental results show that our proposed NARS framework consistently improves the accuracy of recommendation in terms of several accuracy metrics compared with various competing CF methods.