2015•IIASA PURE (International Institute of Applied Systems Analysis)Requires access

How to Increase the Accuracy of Crowdsourcing Campaigns

Oleg Nurmukhametov, Artem Baklanov, Steffen Fritz, Michael Khachay, Carl F. Salk, Linda See, Dmitry Shchepashchenko

Open publisher page 0 citations

Abstract

Crowdsourcing is a new approach to performing tasks, with a group of volunteers rather than experts. For example, the Geo-Wiki project [1] aims to improve the global land-cover map by crowdsourcing for image recognition. Though crowdsourcing gives a simple way to perform tasks that are hard to automate, analysis of data received from non-experts is a challenging problem that requires a holistic approach. Here we study in detail the dataset of the Cropland Capture game (part of Geo-Wiki project) to increase the accuracy of campaign’s results. Using this analysis, we developed a methodology for a generic type of crowdsourcing campaign similar to the Cropland Capture game. The proposed methodology relies on computer vision and machine learning techniques. Using the Cropland Capture dataset we showed that our methodology increases agreement between aggregated volunteers’ votes and experts’ decisions from 77% to 86%. \n \n[1] Fritz, Steffen, et al. “Geo-Wiki. Org: The use of crowdsourcing to improve global land cover.” Remote Sensing 1.3 (2009): 345-354.

Open-access reader

About this research paper

What this paper is about

Crowdsourcing is a new approach to performing tasks, with a group of volunteers rather than experts. For example, the Geo-Wiki project [1] aims to improve the global land-cover map by crowdsourcing for image recognition. Though crowdsourcing gives a simple way to perform tasks that are hard to automate, analysis of data received from non-experts is a challenging problem that requires a holistic approach. Here we study in detail the dataset of the Cropland Capture game (part of Geo-Wiki project) to increase the accuracy of campaign’s results. Using this analysis, we developed a methodology for a generic type of crowdsourcing campaign similar to the Cropland Capture game. The proposed methodology relies on computer vision and machine learning techniques. Using the Cropland Capture dataset we showed that our methodology increases agreement between aggregated volunteers’ votes and experts’ decisions from 77% to 86%. \n \n[1] Fritz, Steffen, et al. “Geo-Wiki. Org: The use of crowdsourcing to improve global land cover.” Remote Sensing 1.3 (2009): 345-354.

Why it matters

A significance statement is not available in the OpenAlex record.

Key contribution

A contribution statement is not available in the OpenAlex record.

Method / approach

Method details are not available in the OpenAlex metadata.

Main findings

Findings are not separately available in the OpenAlex metadata.

Limitations

Limitations are not available in the OpenAlex metadata.

Applications

Application details are not available in the OpenAlex metadata.

Available abstract

Crowdsourcing is a new approach to performing tasks, with a group of volunteers rather than experts. For example, the Geo-Wiki project [1] aims to improve the global land-cover map by crowdsourcing for image recognition. Though crowdsourcing gives a simple way to perform tasks that are hard to automate, analysis of data received from non-experts is a challenging problem that requires a holistic approach. Here we study in detail the dataset of the Cropland Capture game (part of Geo-Wiki project) to increase the accuracy of campaign’s results. Using this analysis, we developed a methodology for a generic type of crowdsourcing campaign similar to the Cropland Capture game. The proposed methodology relies on computer vision and machine learning techniques. Using the Cropland Capture dataset we showed that our methodology increases agreement between aggregated volunteers’ votes and experts’ decisions from 77% to 86%. \n \n[1] Fritz, Steffen, et al. “Geo-Wiki. Org: The use of crowdsourcing to improve global land cover.” Remote Sensing 1.3 (2009): 345-354.

Key concepts: Crowdsourcing, Computer science, Citizen science, Data science, Cover (algebra), Land cover, World Wide Web, Land use

Related papers

Back to paper searchBrowse research topicsOriginal source
How to Increase the Accuracy of Crowdsourcing Campaigns — Research Paper | ScholarLens