Towards Building Ontologies with the Wisdom of the Crowd

Crowdsourcing provides a valuable source of input that reflects the human diversity of domain knowledge. It has increasingly been used in ontology engineering and evaluation, however, few approaches consider different types of crowdsourcing for data acquisition. In this paper, we compare two crowdsourcing techniques - a mechanized labor-based task and a game-based approach - to acquire shared knowledge from which we semi-automatically build an ontology. This paper focuses on the first two steps of ontology engineering, the forming of concepts and their hierarchical relations. To this end, we adapt a distributional semantic and class-based word sense disambiguation approach and a knowledge-intensive tree traversal algorithm. Each step along the process and the final resources are evaluated manually and by a gold standard created from Wikipedia data. Our results show that the ontology resulting from data obtained with the mechanized labor-based approach provides a higher level of granularity than the game-based one. However, the latter is faster and seems more enticing to participants