Model and Algorithms for Optimizing a Human Computing System Oriented to Knowledge Extraction by Use of Crowdsourcing

The paper is addressing the actual context of Data Deluge, where the need and also premises to extract more knowledge are increasing, along with the increase of our expectations about performances. Besides, improving artificial intelligence (AI), by machine learning (ML), deep learning (DL) or cognitive learning (CL) performance/potential, when adding human contributions where necessary, is an important and promising research area. Consequently, our model, algorithms (ALG1; ALG2) and soft programs provide useful new instruments for implementing and optimizing the workflow based on crowdsourcing, when using human potential in a human computing system. We aim to increase AI quality adding multiple human outputs for every AI task and leveraging learning rules to be then extended to larger sets of tasks. This way, such hybrid system could be oriented to more knowledge extraction, by the generalization of images/ captions/labels toward more complex tasks, like providing content essential or question answering. Our instruments include features of ranking workers and tasks profiles, which will support the main original process of knowledge extraction, but also the inference elements, by small amounts of learning data (regarding the workers skills and tasks efficiency) to be transferred to AI/ML/DL/CL, which then could be used for processing larger volumes of similar data. Among the results conclusions is that using progressive optimization, structuring the data/tasks in variable (progressive) sets and potential (skill/number) of workers, is both efficacious and efficient, allowing a flexible control of the system and workflow for matching a diversity of tasks complexity/ difficulty/volume and leveraging knowledge extraction.