Imprecise Label Aggregation Approach Under the Belief Function Theory

Abstract

Crowdsourcing has become a phenomenon of increasing interest in several research fields such as artificial intelligence. It typically uses human cognitive ability in order to effectively solve tasks that can hardly be addressed by automated computation. The major problem however is that so far studies could not completely control the quality of obtained data since contributors are uncertainly reliable. In this work, we propose an approach that aggregates labels using the belief function theory under the assumption that these labels could be partial hence imprecise. Simulated data demonstrate that our method produces more reliable aggregation results.