Regularized Minimax Conditional Entropy for Crowdsourcing

TL;DR

This paper introduces a novel minimax conditional entropy approach to accurately infer true labels from noisy crowdsourced data, accounting for worker ability and item difficulty.

Contribution

It proposes a unique probabilistic model and objective measurement principle for crowdsourcing label aggregation, validated on diverse real datasets.

Findings

Effective in handling noisy labels from crowdsourcing.

Outperforms existing methods on multiple datasets.

Provides a principled way to measure label quality.

Abstract

There is a rapidly increasing interest in crowdsourcing for data labeling. By crowdsourcing, a large number of labels can be often quickly gathered at low cost. However, the labels provided by the crowdsourcing workers are usually not of high quality. In this paper, we propose a minimax conditional entropy principle to infer ground truth from noisy crowdsourced labels. Under this principle, we derive a unique probabilistic labeling model jointly parameterized by worker ability and item difficulty. We also propose an objective measurement principle, and show that our method is the only method which satisfies this objective measurement principle. We validate our method through a variety of real crowdsourcing datasets with binary, multiclass or ordinal labels.