Ranking-Based Multi-source Rating Aggregation with Social Contexts

Rating aggregation is critical to the quality control of recommendation systems and its effectiveness is a deep concern of all users. However, there are some problems in existing recommendation systems. For example, some of the raters from certain source are much more stringent than others, leading the phenomena that some entities with better quality are rejected. In this paper, we propose a novel raNking-based multI-source ratiNg Aggregation (NINA) approach. In this approach, based on the collected social context of recommenders and recommendees, the credibility of rankings of ratings from multiple sources can be estimated, and it can be used to deal with the disagreement during ranking-based rating aggregation from multiple sources. Hence, the proposed approach can effectively estimate the 'true' rating during aggregation based on the ratings from multiple sources, even though no prior knowledge exists about the distribution of stringent raters and lenient raters in different sources. We have studied the properties of NINA empirically. In particular, the experiments illustrate that compared with existing approaches, our proposed NINA can significantly reduce the influence of ratings from stringent raters and lenient raters, leading to trust enhanced rating aggregation, no matter what kind of the distributions of stringent raters and lenient raters are.