An official website of the United States government
As AI-based face recognition technologies are increasingly adopted for high-stakes applications like locating suspected criminals, public concerns about the accuracy of these technologies have grown as well. These technologies often present a human expert with a shortlist of high-confidence candidate faces from which the expert must select correct match(es) while avoiding false positives, which we term the “last-mile problem.” We propose Second Opinion, a web-based software tool that employs a novel crowdsourcing workflow inspired by cognitive psychology, seed-gather-analyze, to assist experts in solving the last-mile problem. We evaluated Second Opinion with a mixed-methods lab study involving 10 experts and 300 crowd workers who collaborate to identify people in historical photos. We found that crowds can eliminate 75% of false positives from the highest-confidence candidates suggested by face recognition, and that experts were enthusiastic about using Second Opinion in their work. We also discuss broader implications for crowd–AI interaction and crowdsourced person identification.
more »
« less
From Crowd Ratings to Predictive Models of Newsworthiness to Support Science Journalism
The scale of scientific publishing continues to grow, creating overload on science journalists who are inundated with choices for what would be most interesting, important, and newsworthy to cover in their reporting. Our work addresses this problem by considering the viability of creating a predictive model of newsworthiness of scientific articles that is trained using crowdsourced evaluations of newsworthiness. We proceed by first evaluating the potential of crowd-sourced evaluations of newsworthiness by assessing their alignment with expert ratings of newsworthiness, analyzing both quantitative correlations and qualitative rating rationale to understand limitations. We then demonstrate and evaluate a predictive model trained on these crowd ratings together with arXiv article metadata, text, and other computed features. Based on the crowdsourcing protocol we developed, we find that while crowdsourced ratings of newsworthiness often align moderately with expert ratings, there are also notable differences and divergences which limit the approach. Yet despite these limitations we also find that the predictive model we built provides a reasonably precise set of rankings when validated against expert evaluations (P@10 = 0.8, P@15 = 0.67), suggesting that a viable signal can be learned from crowdsourced evaluations of newsworthiness. Based on these findings we discuss opportunities for future work to leverage crowdsourcing and predictive approaches to support journalistic work in discovering and filtering newsworthy information.
more »
« less
- Award ID(s):
- 1845460
- PAR ID:
- 10386489
- Date Published:
- Journal Name:
- Proceedings of the ACM on Human-Computer Interaction
- Volume:
- 6
- Issue:
- CSCW2
- ISSN:
- 2573-0142
- Page Range / eLocation ID:
- 1 to 28
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract Wetland loss is increasing rapidly, and there are gaps in public awareness of the problem. By crowdsourcing image analysis of wetland morphology, academic and government studies could be supplemented and accelerated while engaging and educating the public. The Land Loss Lookout (LLL) project crowdsourced mapping of wetland morphology associated with wetland loss and restoration. We demonstrate that volunteers can be trained relatively easily online to identify characteristic wetland morphologies, or patterns present on the landscape that suggest a specific geomorphological process. Results from a case study in coastal Louisiana revealed strong agreement among nonexpert and expert assessments who agreed on classifications at least 83% and at most 94% of the time. Participants self‐reported increased knowledge of wetland loss after participating in the project. Crowd‐identified morphologies are consistent with expectations, although more work is needed to directly compare LLL results with previous studies. This work provides a foundation for using crowd‐based wetland loss analysis to increase public awareness of the issue, and to contribute to land surveys or train machine learning algorithms.more » « less
-
null (Ed.)Crowdsourcing is widely used to create data for common natural language understanding tasks. Despite the importance of these datasets for measuring and refining model understanding of language, there has been little focus on the crowdsourcing methods used for collecting the datasets. In this paper, we compare the efficacy of interventions that have been proposed in prior work as ways of improving data quality. We use multiple-choice question answering as a testbed and run a randomized trial by assigning crowdworkers to write questions under one of four different data collection protocols. We find that asking workers to write explanations for their examples is an ineffective stand-alone strategy for boosting NLU example difficulty. However, we find that training crowdworkers, and then using an iterative process of collecting data, sending feedback, and qualifying workers based on expert judgments is an effective means of collecting challenging data. But using crowdsourced, instead of expert judgments, to qualify workers and send feedback does not prove to be effective. We observe that the data from the iterative protocol with expert assessments is more challenging by several measures. Notably, the human--model gap on the unanimous agreement portion of this data is, on average, twice as large as the gap for the baseline protocol data.more » « less
-
null (Ed.)Concerns about the spread of misinformation online via news articles have led to the development of many tools and processes involving human annotation of their credibility. However, much is still unknown about how different people judge news credibility or the quality or reliability of news credibility ratings from populations of varying expertise. In this work, we consider credibility ratings from two “crowd” populations: 1) students within journalism or media programs, and 2) crowd workers on UpWork, and compare them with the ratings of two sets of experts: journalists and climate scientists, on a set of 50 climate-science articles. We find that both groups’ credibility ratings have higher correlation to journalism experts compared to the science experts, with 10-15 raters to achieve convergence. We also find that raters’ gender and political leaning impact their ratings. Among article genre of news/opinion/analysis and article source leaning of left/center/right, crowd ratings were more similar to experts respectively with opinion and strong left sources.more » « less
-
Concerns about the spread of misinformation online via news articles have led to the development of many tools and processes involving human annotation of their credibility. However, much is still unknown about how different people judge news credibility or the quality or reliability of news credibility ratings from populations of varying expertise. In this work, we consider credibility ratings from two “crowd” populations: 1) students within journalism or media programs, and 2) crowd workers on UpWork, and compare them with the ratings of two sets of experts: journalists and climate scientists, on a set of 50 climate-science articles. We find that both groups’ credibility ratings have higher correlation to journalism experts compared to the science experts, with 10-15 raters to achieve convergence. We also find that raters’ gender and political leaning impact their ratings. Among article genre of news/opinion/analysis and article source leaning of left/center/right, crowd ratings were more similar to experts respectively with opinion and strong left sources.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1145/3555542
