Frustratingly Easy Truth Discovery
This addresses the problem of extracting correct answers from noisy crowdsourced data for applications like data labeling and quality assessment, though it appears incremental as it builds on existing truth discovery methods with a simpler heuristic.
The paper tackles the problem of truth discovery from noisy crowdsourced answers by proposing an extremely simple heuristic that estimates worker competence using average proximity to other workers. The result shows this approach enables effective separation of high and low quality workers and yields substantial practical improvements over unweighted aggregation and other algorithms.
Truth discovery is a general name for a broad range of statistical methods aimed to extract the correct answers to questions, based on multiple answers coming from noisy sources. For example, workers in a crowdsourcing platform. In this paper, we consider an extremely simple heuristic for estimating workers' competence using average proximity to other workers. We prove that this estimates well the actual competence level and enables separating high and low quality workers in a wide spectrum of domains and statistical models. Under Gaussian noise, this simple estimate is the unique solution to the MLE with a constant regularization factor. Finally, weighing workers according to their average proximity in a crowdsourcing setting, results in substantial improvement over unweighted aggregation and other truth discovery algorithms in practice.