The Positive Role of Negative Ratings for Recommender
Ike Song, Su Chen, Peng Zhang · 2017
In this paper, inspired by the network-based inference, a new weighted approach is presented to experimentally assess the role of negative data. This weighted approach is conductive to distinguish the contributions from positive and negative ratings. By conducting the positive and negative data with twofold weights, the method relative to NBI and NBIS can obtain a bigger precision and a smaller ranking score, leading to a better recommendation quality. Via the further numerical tests on three benchmark datasets, the results show that the presented approach can better reveal the positive role of negative ratings for improving the recommendation quality. Moreover, by using some appropriate tools, the positive recommendation role of negative data will strengthen, and thoughtlessly removing negative data not only miss some valuable information, but also can weaken the quality of recommendation system.