Mix and match: Collaborative expert-crowd judging for building test collections accurately and affordably

Abstract

Crowdsourcing offers an affordable and scalable means to collect relevance judgments for information retrieval test collections. However, crowd assessors may showhigher variance in judgment quality than trusted assessors. In this paper, we investigate how to effectively utilize both groups of assessors in partnership. We study how agreement in judging is correlated with three factors: relevance category, document rankings, and topical variance. Based on this, we then propose two collaborative judging methods in which some document-topic pairs are assigned to in-house assessors for relevance judging while the rest are assessed by crowd workers. Results on two TREC collections show encouraging results when we distribute work intelligently between our two groups of assessors.