Research and Projects
My main research interest is to understand and improve crowdsourced development. I believe crowdsourcing can be used to develop a large variety of things, including complex, data-intensive systems.
I both build and study systems. My toolbox includes performance indicators, behavior studies, network analysis, pricing, evolutionary computation, machine learning, and natural language processing. Below is a selection of recent research topics.
Crowdsourced Task Decomposition
Project decomposition is a labor-intensive process that is holding back the adoption of crowdsourcing. Our current approach to this is to use modern NLP methods, to learn the relationship between task statements and project descriptions. These learned models will then be combined with evolutionary methods to create an automatic decomposition tool.
Crowdsourced Task Scheduling
We develop methods for dynamic task scheduling in crowdsourcing.
We use a deep learning-based predictor of task success given a worker and a temporal task context, then combine the task success predictor with an evolutionary algorithm. This method can be used to schedule tasks from several projects utilizing the same worker pool concurrently.
Crowd's Relation and Behavior
I investigate data from TopCoder, the software development crowdsourcing platform, and used ideas from psychology, software engineering, human factor engineering, and project management to develop appropriate measures for worker behavior and preferences. Using these measures, I build different models such as team elasticity, worker preferences, task completion patterns, and other relevant measures.
Collaborators: Hamid Shams
MiPasa: Co-Opetition crowdsourced platform for analyzed and visualized data related to COVID-19. The platform provides both collaboration and competition activities in different levels of data science analysis.