A0267
Title: Joint latent space models for ranking data and social network
Authors: Jiaqi Gu - Stanford University (United States) [presenting]
Philip Yu - The Education University of Hong Kong (Hong Kong)
Abstract: Human interaction and communication has become one of the essential features of social life. Individuals' preferences may be influenced by those of their peers or friends in a social network. In the literature, individuals' rank-order preferences and their social networks are often modelled separately. A new joint probabilistic model is proposed for ranking data and social networks. With a latent space for all the individuals and items, the proposed model assumes that the social network and rankings of items are governed by the locations of individuals and items. Based on an efficient MCMC algorithm, a set of Bayesian inference approaches is developed for the proposed model, including procedures of model selection, criteria to evaluate model fitness and a test for conditional independence between individuals' rankings and their social network given their positions in the latent space. Simulation studies reveal the usefulness of the proposed methods for parameter estimation, model fitness evaluation, model selection and conditional independence testing. Finally, the model is applied to the CiaoDVD dataset, consisting of users' trust relations and implicit preferences on DVD categories.
