Goldilocks: Consistent Crowdsourced Scalar Annotations with Relative Uncertainty
Human ratings have become a crucial resource for training and evaluating machine learning systems. However, traditional elicitation methods for absolute and comparative rating suffer from issues with consistency and often do not distinguish between uncertainty due to disagreement between annotators...
Gespeichert in:
Veröffentlicht in: | Proceedings of the ACM on human-computer interaction 2021-10, Vol.5 (CSCW2), p.1-25, Article 335 |
---|---|
Hauptverfasser: | , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
container_end_page | 25 |
---|---|
container_issue | CSCW2 |
container_start_page | 1 |
container_title | Proceedings of the ACM on human-computer interaction |
container_volume | 5 |
creator | Chen, Quan Ze Weld, Daniel S. Zhang, Amy X. |
description | Human ratings have become a crucial resource for training and evaluating machine learning systems. However, traditional elicitation methods for absolute and comparative rating suffer from issues with consistency and often do not distinguish between uncertainty due to disagreement between annotators and ambiguity inherent to the item being rated. In this work, we present Goldilocks, a novel crowd rating elicitation technique for collecting calibrated scalar annotations that also distinguishes inherent ambiguity from inter-annotator disagreement. We introduce two main ideas: grounding absolute rating scales with examples and using a two-step bounding process to establish a range for an item's placement. We test our designs in three domains: judging toxicity of online comments, estimating satiety of food depicted in images, and estimating age based on portraits. We show that (1) Goldilocks can improve consistency in domains where interpretation of the scale is not universal, and that (2) representing items with ranges lets us simultaneously capture different sources of uncertainty leading to better estimates of pairwise relationship distributions. |
doi_str_mv | 10.1145/3476076 |
format | Article |
fullrecord | <record><control><sourceid>acm_cross</sourceid><recordid>TN_cdi_crossref_primary_10_1145_3476076</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>3476076</sourcerecordid><originalsourceid>FETCH-LOGICAL-a159t-780b6b832d198ca91b4f773b7aa7f5b8836a32c4669672eabdbaf0dadaf1e503</originalsourceid><addsrcrecordid>eNpNkE1LAzEYhIMoWGrx7ik3T6v52CS73sqiVSgoWs_Lm4_F6HYjSbT033elVTzNDPMwh0HonJIrSktxzUsliZJHaMKE4gWhJTv-50_RLKV3QgitBBE1m6CnReit74P5SDe4CUPyKbsh4yaGjU3hKxpn8YuBHiKeD0PIkP1I4Y3Pb_jZ9WP8dvh1MC5m8EPenqGTDvrkZgedotXd7aq5L5aPi4dmviyAijoXqiJa6oozS-vKQE112SnFtQJQndBVxSVwZkopa6mYA201dMSChY46QfgUXe5nTQwpRde1n9GvIW5bStqfK9rDFSN5sSfBrP-g33IHamlZ-w</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Goldilocks: Consistent Crowdsourced Scalar Annotations with Relative Uncertainty</title><source>ACM Digital Library Complete</source><creator>Chen, Quan Ze ; Weld, Daniel S. ; Zhang, Amy X.</creator><creatorcontrib>Chen, Quan Ze ; Weld, Daniel S. ; Zhang, Amy X.</creatorcontrib><description>Human ratings have become a crucial resource for training and evaluating machine learning systems. However, traditional elicitation methods for absolute and comparative rating suffer from issues with consistency and often do not distinguish between uncertainty due to disagreement between annotators and ambiguity inherent to the item being rated. In this work, we present Goldilocks, a novel crowd rating elicitation technique for collecting calibrated scalar annotations that also distinguishes inherent ambiguity from inter-annotator disagreement. We introduce two main ideas: grounding absolute rating scales with examples and using a two-step bounding process to establish a range for an item's placement. We test our designs in three domains: judging toxicity of online comments, estimating satiety of food depicted in images, and estimating age based on portraits. We show that (1) Goldilocks can improve consistency in domains where interpretation of the scale is not universal, and that (2) representing items with ranges lets us simultaneously capture different sources of uncertainty leading to better estimates of pairwise relationship distributions.</description><identifier>ISSN: 2573-0142</identifier><identifier>EISSN: 2573-0142</identifier><identifier>DOI: 10.1145/3476076</identifier><language>eng</language><publisher>New York, NY, USA: ACM</publisher><subject>Collaborative interaction ; Human computer interaction (HCI) ; Human-centered computing ; Interaction paradigms</subject><ispartof>Proceedings of the ACM on human-computer interaction, 2021-10, Vol.5 (CSCW2), p.1-25, Article 335</ispartof><rights>ACM</rights><lds50>peer_reviewed</lds50><woscitedreferencessubscribed>false</woscitedreferencessubscribed><citedby>FETCH-LOGICAL-a159t-780b6b832d198ca91b4f773b7aa7f5b8836a32c4669672eabdbaf0dadaf1e503</citedby><cites>FETCH-LOGICAL-a159t-780b6b832d198ca91b4f773b7aa7f5b8836a32c4669672eabdbaf0dadaf1e503</cites></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><linktopdf>$$Uhttps://dl.acm.org/doi/pdf/10.1145/3476076$$EPDF$$P50$$Gacm$$H</linktopdf><link.rule.ids>314,776,780,2276,27901,27902,40172,76197</link.rule.ids></links><search><creatorcontrib>Chen, Quan Ze</creatorcontrib><creatorcontrib>Weld, Daniel S.</creatorcontrib><creatorcontrib>Zhang, Amy X.</creatorcontrib><title>Goldilocks: Consistent Crowdsourced Scalar Annotations with Relative Uncertainty</title><title>Proceedings of the ACM on human-computer interaction</title><addtitle>ACM PACMHCI</addtitle><description>Human ratings have become a crucial resource for training and evaluating machine learning systems. However, traditional elicitation methods for absolute and comparative rating suffer from issues with consistency and often do not distinguish between uncertainty due to disagreement between annotators and ambiguity inherent to the item being rated. In this work, we present Goldilocks, a novel crowd rating elicitation technique for collecting calibrated scalar annotations that also distinguishes inherent ambiguity from inter-annotator disagreement. We introduce two main ideas: grounding absolute rating scales with examples and using a two-step bounding process to establish a range for an item's placement. We test our designs in three domains: judging toxicity of online comments, estimating satiety of food depicted in images, and estimating age based on portraits. We show that (1) Goldilocks can improve consistency in domains where interpretation of the scale is not universal, and that (2) representing items with ranges lets us simultaneously capture different sources of uncertainty leading to better estimates of pairwise relationship distributions.</description><subject>Collaborative interaction</subject><subject>Human computer interaction (HCI)</subject><subject>Human-centered computing</subject><subject>Interaction paradigms</subject><issn>2573-0142</issn><issn>2573-0142</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2021</creationdate><recordtype>article</recordtype><recordid>eNpNkE1LAzEYhIMoWGrx7ik3T6v52CS73sqiVSgoWs_Lm4_F6HYjSbT033elVTzNDPMwh0HonJIrSktxzUsliZJHaMKE4gWhJTv-50_RLKV3QgitBBE1m6CnReit74P5SDe4CUPyKbsh4yaGjU3hKxpn8YuBHiKeD0PIkP1I4Y3Pb_jZ9WP8dvh1MC5m8EPenqGTDvrkZgedotXd7aq5L5aPi4dmviyAijoXqiJa6oozS-vKQE112SnFtQJQndBVxSVwZkopa6mYA201dMSChY46QfgUXe5nTQwpRde1n9GvIW5bStqfK9rDFSN5sSfBrP-g33IHamlZ-w</recordid><startdate>20211018</startdate><enddate>20211018</enddate><creator>Chen, Quan Ze</creator><creator>Weld, Daniel S.</creator><creator>Zhang, Amy X.</creator><general>ACM</general><scope>AAYXX</scope><scope>CITATION</scope></search><sort><creationdate>20211018</creationdate><title>Goldilocks: Consistent Crowdsourced Scalar Annotations with Relative Uncertainty</title><author>Chen, Quan Ze ; Weld, Daniel S. ; Zhang, Amy X.</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a159t-780b6b832d198ca91b4f773b7aa7f5b8836a32c4669672eabdbaf0dadaf1e503</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2021</creationdate><topic>Collaborative interaction</topic><topic>Human computer interaction (HCI)</topic><topic>Human-centered computing</topic><topic>Interaction paradigms</topic><toplevel>peer_reviewed</toplevel><toplevel>online_resources</toplevel><creatorcontrib>Chen, Quan Ze</creatorcontrib><creatorcontrib>Weld, Daniel S.</creatorcontrib><creatorcontrib>Zhang, Amy X.</creatorcontrib><collection>CrossRef</collection><jtitle>Proceedings of the ACM on human-computer interaction</jtitle></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Chen, Quan Ze</au><au>Weld, Daniel S.</au><au>Zhang, Amy X.</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Goldilocks: Consistent Crowdsourced Scalar Annotations with Relative Uncertainty</atitle><jtitle>Proceedings of the ACM on human-computer interaction</jtitle><stitle>ACM PACMHCI</stitle><date>2021-10-18</date><risdate>2021</risdate><volume>5</volume><issue>CSCW2</issue><spage>1</spage><epage>25</epage><pages>1-25</pages><artnum>335</artnum><issn>2573-0142</issn><eissn>2573-0142</eissn><abstract>Human ratings have become a crucial resource for training and evaluating machine learning systems. However, traditional elicitation methods for absolute and comparative rating suffer from issues with consistency and often do not distinguish between uncertainty due to disagreement between annotators and ambiguity inherent to the item being rated. In this work, we present Goldilocks, a novel crowd rating elicitation technique for collecting calibrated scalar annotations that also distinguishes inherent ambiguity from inter-annotator disagreement. We introduce two main ideas: grounding absolute rating scales with examples and using a two-step bounding process to establish a range for an item's placement. We test our designs in three domains: judging toxicity of online comments, estimating satiety of food depicted in images, and estimating age based on portraits. We show that (1) Goldilocks can improve consistency in domains where interpretation of the scale is not universal, and that (2) representing items with ranges lets us simultaneously capture different sources of uncertainty leading to better estimates of pairwise relationship distributions.</abstract><cop>New York, NY, USA</cop><pub>ACM</pub><doi>10.1145/3476076</doi><tpages>25</tpages></addata></record> |
fulltext | fulltext |
identifier | ISSN: 2573-0142 |
ispartof | Proceedings of the ACM on human-computer interaction, 2021-10, Vol.5 (CSCW2), p.1-25, Article 335 |
issn | 2573-0142 2573-0142 |
language | eng |
recordid | cdi_crossref_primary_10_1145_3476076 |
source | ACM Digital Library Complete |
subjects | Collaborative interaction Human computer interaction (HCI) Human-centered computing Interaction paradigms |
title | Goldilocks: Consistent Crowdsourced Scalar Annotations with Relative Uncertainty |
url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-02-14T07%3A33%3A00IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-acm_cross&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Goldilocks:%20Consistent%20Crowdsourced%20Scalar%20Annotations%20with%20Relative%20Uncertainty&rft.jtitle=Proceedings%20of%20the%20ACM%20on%20human-computer%20interaction&rft.au=Chen,%20Quan%20Ze&rft.date=2021-10-18&rft.volume=5&rft.issue=CSCW2&rft.spage=1&rft.epage=25&rft.pages=1-25&rft.artnum=335&rft.issn=2573-0142&rft.eissn=2573-0142&rft_id=info:doi/10.1145/3476076&rft_dat=%3Cacm_cross%3E3476076%3C/acm_cross%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true |