Efficient crowdsourcing of crowd-generated microtasks.

Area of Science:

• Computer Science
• Human-Computer Interaction
• Artificial Intelligence

Background:

• Crowdsourcing microtasks combines efficiency with human creativity.
• Growing task sets in crowdsourcing can overwhelm resources.
• Existing algorithms often require a fixed task set size.

Purpose of the Study:

• Introduce cost forecasting for efficient crowdsourcing with dynamic task sets.
• Enable crowdsourcers to balance resource allocation between new and existing tasks.
• Improve accuracy and efficiency in crowdsourcing novel microtasks.

Main Methods:

• Developed a cost forecasting approach for microtask proposal systems.
• Implemented a decision mechanism based on predicted costs of new versus existing tasks.
• Conducted experiments using real and synthetic crowdsourcing data.

Main Results:

• Cost forecasting demonstrated improved accuracy in microtask completion.
• The method efficiently balances crowdsourcer resources as the task set grows.
• Achieved accuracy and efficiency gains for crowd-generated microtasks.

Conclusions:

• Cost forecasting is a viable strategy for managing dynamic crowdsourcing environments.
• This approach enhances the utility of collective intelligence for various applications.
• Potential applications include improving machine learning datasets and content platforms.