Jesus Perez - Academia.edu (original) (raw)

Uploads

Papers by Jesus Perez

Research paper thumbnail of University of glasgow (qirdcsuog) at TREC crowdsourcing 2011: TurkRank - Network-based worker ranking in crowdsourcing

NIST Special Publication, 2011

For TREC Crowdsourcing 2011 (Stage 2) we propose a networkbased approach for assigning an indicat... more For TREC Crowdsourcing 2011 (Stage 2) we propose a networkbased approach for assigning an indicative measure of worker trustworthiness in crowdsourced labelling tasks. Workers, the gold standard and worker/gold standard agreements are modelled as a network. For the purpose of worker trustworthiness assignment, a variant of the PageRank algorithm, named TurkRank, is used to adaptively combine evidence that suggests worker trustworthiness, i.e., agreement with other trustworthy co-workers and agreement with the gold standard. A single parameter controls the importance of co-worker agreement versus gold standard agreement. The TurkRank score calculated for each worker is incorporated with a worker-weighted mean label aggregation.

Research paper thumbnail of University of glasgow (qirdcsuog) at TREC crowdsourcing 2011: TurkRank - Network-based worker ranking in crowdsourcing

NIST Special Publication, 2011

For TREC Crowdsourcing 2011 (Stage 2) we propose a networkbased approach for assigning an indicat... more For TREC Crowdsourcing 2011 (Stage 2) we propose a networkbased approach for assigning an indicative measure of worker trustworthiness in crowdsourced labelling tasks. Workers, the gold standard and worker/gold standard agreements are modelled as a network. For the purpose of worker trustworthiness assignment, a variant of the PageRank algorithm, named TurkRank, is used to adaptively combine evidence that suggests worker trustworthiness, i.e., agreement with other trustworthy co-workers and agreement with the gold standard. A single parameter controls the importance of co-worker agreement versus gold standard agreement. The TurkRank score calculated for each worker is incorporated with a worker-weighted mean label aggregation.

Log In