A Bayesian Model for Aggregating Rank Data with Covariates

Dingdong Yi, Xinran Li, Jun S. LiuarXiv 2016 · arXiv

Abstract

Rank aggregation is the combining of ranking results from different sources to generate a better ranking list. Existing rank aggregation methods utilize only rank data. In some real applications, however, rank data come with covariates information as well. Moreover, some rankers may be more informative and reliable than others. Therefore, it is desirable for a rank aggregation method to incorporate covariates information and to distinguish high quality rankers from spam rankers. Here, we propose the Bayesian aggregation of rank-data with covariates (BARC) and its weighted version, BARCW, to achieve the aforementioned goals. Both BARC and BARCW employ latent variable models\textemdash which can incorporate covariates information\textemdash for ranked entities. Furthermore, BARCW quantitatively measures the difference in qualities of rankers. We generate an aggregated ranking and its probability interval based on posterior distributions of latent variables derived from a parameter-expanded Gibbs sampler. Simulation studies show that our methods outperform competing methods in a variety of scenarios. Real-data applications also validate our estimation of reliability parameters and effects of covaraiates.
BibTeX
@misc{yi2016bayesian,
  author = {Dingdong Yi and Xinran Li and Jun S. Liu},
  title = {{A Bayesian Model for Aggregating Rank Data with Covariates}},
  year = {2016},
  eprint = {1607.06051},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1607.06051},
}

Loading related content...

Figure