A Hybrid Approach to Commonsense Knowledge Acquisition [Best Presentation Award]
Published in Proceedings of the 8th European Starting AI Researcher Symposium, 2016
This work presents a knowledge acquisition platform and a certain game developed on that platform for endowing machines with common sense, by following a hybrid approach that combines crowdsourcing techniques, knowledge engineering, and automated reasoning. Short narratives are presented to players, who are asked to combine fragments of text into rules that would correctly answer a given question, to evaluate the appropriateness of gathered rules, and to resolve conflicts between them by assigning priorities. The text fragments that are used are a priori translated by a knowledge engineer into a machine-readable predicate form. Players are rewarded based not only on their inter-agreement (as in most games with a purpose) but also based on the objective ability of the rules to answer questions correctly, as determined by an underlying reasoning engine. Beyond discussing the knowledge acquisition platform and the game design, we analyze the common sense that has been gathered during the deployment of the game over a five-month period and we use the acquired knowledge to answer questions on unknown stories.
Cite this work: Christos Rodosthenous, Loizos Michael, "A Hybrid Approach to Commonsense Knowledge Acquisition [Best Presentation Award]." Proceedings of the 8th European Starting AI Researcher Symposium, 2016.
Download Paper