Download files
Abstract
The overwhelming reach of the Internet has seen the emergence of a new wave of digital platforms in the form of knowledge sharing communities, social networks and crowdsourcing as a quintessential mode for collaborations and knowledge spreading. Specifically, for the last decade, the advancements in these technologies have enabled crowdsourcing to be utilized as a means of exploiting the intelligence of people (the crowd) for the purpose of solving simple to complex tasks. Organizations have started leveraging the collective intelligence on crowdsourcing platforms (e.g., Amazon Mechanical Turk) to create innovations in product development, enhance marketing, and improve customer services.
Nevertheless, despite the wide adoption of crowdsourcing practices, we face several challenges ranging from the quality of the output to the selection of the right worker (crowd), given the variations in workers' skills representations. More specifically, the lack of rich semantics and latent knowledge representations of workers' skills lead to poor matching between the skills required for a task and the skills possessed by a worker, making it increasingly difficult to find the right set of workers with suitable skills and expertise to undertake a task successfully.
In this thesis, we propose several computational techniques to give a better understanding and semantic representations of workers' skills and expertise from the perspective of jointly modelling the tasks and workers' skills in the same semantic vector space. Specifically, inspired by the deep learning embedding models, we develop novel vector space models and techniques to capture and represent the workers' skills, textual artefacts (e.g., answers, vulnerability discovery reports) and rich latent interactions between different communities to support the selection of the most appropriate workers for specific tasks. We also develop algorithms to capture the influence of users in these communities. Moreover, we validate and evaluate the proposed models and techniques through different experimental scenarios.