A Truth Discovery Approach with Theoretical Guarantee

摘要: In the information age, people can easily collect about same set of entities from multiple sources, among which conflicts are inevitable. This leads to an important task, truth discovery, i.e., identify true facts (truths) via iteratively updating truths and source reliability. However, convergence is never discussed in existing work, thus there no theoretical guarantee results these discovery approaches. contrast, this paper we propose a approach with guarantee. We randomized gaussian mixture model (RGMM) represent multi-source data, where parameters. incorporate bias captures its reliability degree into RGMM formulation. The task then modeled as seeking maximum likelihood estimate (MLE) truths. Based on expectation-maximization (EM) techniques, population-based (i.e., limit infinite data) sample-based finite samples) solutions for MLE. Theoretically, prove that both contractive e-ball around MLE, under certain conditions. Experimentally, evaluate our method simulated real-world datasets. Experimental show achieves high accuracy identifying