Additional Operations of Simple HITs on Microtask Crowdsourcing for Worker Quality Prediction

• Item type: Journal(1)
• 公開日: 2019-01-15
• タイトル: Additional Operations of Simple HITs on Microtask Crowdsourcing for Worker Quality Prediction
• タイトル / 言語: en
• タイトル: Additional Operations of Simple HITs on Microtask Crowdsourcing for Worker Quality Prediction
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: [特集：全ての人とモノがつながる社会に向けたコラボレーション技術とネットワークサービス] crowdsourcing, worker quality, machine learning, random forest, microtask, active intervention
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_6501
• 資源タイプ: journal article
• 著者所属: Nara Institute of Science and Technology
• 著者所属: Nara Institute of Science and Technology
• 著者所属: Nara Institute of Science and Technology
• 著者所属(英): en
• 著者所属(英): Nara Institute of Science and Technology
• 著者所属(英): en
• 著者所属(英): Nara Institute of Science and Technology
• 著者所属(英): en
• 著者所属(英): Nara Institute of Science and Technology
• 著者名: Yu, Suzuki; Yoshitaka, Matsuda; Satoshi, Nakamura
• 著者名(英): Yu, Suzuki; Yoshitaka, Matsuda; Satoshi, Nakamura
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: In microtask crowdsourcing, low-quality workers damage the quality of work results. Therefore, if a system automatically eliminates the low-quality workers, the requesters will obtain high-quality work results with low wages. When we consider simple Human Intelligent Tasks (HITs), such as yes-no questions of a labeling task, the requesters have difficulty assessing the worker quality only from the work results. Therefore, we need a method to accurately predict the worker quality automatically from the behaviors of workers, such as working time and the number of clicks. When we accurately predict the worker quality, we are able to prepare many features from the worker behaviors. However, when we submit simple HITs, we can capture only a small number of behaviors of workers, then the accuracy of predicted worker quality will be low. To solve this issue, we propose a method to insert into the simple task of obtaining many features of worker behaviors. We prepared a classification task of tweets as simple HITs. We added a button to the work screen. The workers can browse the target tweets on the work screen during the time the workers are pressing the button, but the workers cannot browse the target tweets when the workers have released the button. Using this button, we can obtain six more kinds of features of worker behaviors. Using our method, we can improve the recall ratio 12% of identifying low-quality workers. However, as the load of workers increases, then the processing time becomes longer, and the motivation of workers decreases. From this result, we also discovered that there is a trade-off between the number of obtained behaviors and the load of workers.; ------------------------------; This is a preprint of an article intended for publication Journal of; Information Processing(JIP). This preprint should not be cited. This; article should be cited as: Journal of Information Processing Vol.27(2019) (online); DOI http://dx.doi.org/10.2197/ipsjjip.27.51; ------------------------------
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: In microtask crowdsourcing, low-quality workers damage the quality of work results. Therefore, if a system automatically eliminates the low-quality workers, the requesters will obtain high-quality work results with low wages. When we consider simple Human Intelligent Tasks (HITs), such as yes-no questions of a labeling task, the requesters have difficulty assessing the worker quality only from the work results. Therefore, we need a method to accurately predict the worker quality automatically from the behaviors of workers, such as working time and the number of clicks. When we accurately predict the worker quality, we are able to prepare many features from the worker behaviors. However, when we submit simple HITs, we can capture only a small number of behaviors of workers, then the accuracy of predicted worker quality will be low. To solve this issue, we propose a method to insert into the simple task of obtaining many features of worker behaviors. We prepared a classification task of tweets as simple HITs. We added a button to the work screen. The workers can browse the target tweets on the work screen during the time the workers are pressing the button, but the workers cannot browse the target tweets when the workers have released the button. Using this button, we can obtain six more kinds of features of worker behaviors. Using our method, we can improve the recall ratio 12% of identifying low-quality workers. However, as the load of workers increases, then the processing time becomes longer, and the motivation of workers decreases. From this result, we also discovered that there is a trade-off between the number of obtained behaviors and the load of workers.; ------------------------------; This is a preprint of an article intended for publication Journal of; Information Processing(JIP). This preprint should not be cited. This; article should be cited as: Journal of Information Processing Vol.27(2019) (online); DOI http://dx.doi.org/10.2197/ipsjjip.27.51; ------------------------------
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AN00116647
• 書誌情報: 情報処理学会論文誌; 巻 60, 号 1, 発行日 2019-01-15
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 1882-7764