Learning common sense knowledge from user interaction and principal component analysis

Abstract

In this thesis, I present a system for reasoning with common sense knowledge in multiple natural languages, as part of the Open Mind Common Sense project. The knowledge that Open Mind collects from volunteer contributors is represented as a semantic network called ConceptNet. Using principal component analysis on the graph structure of ConceptNet yields AnalogySpace, a vector space representation of common sense knowledge. This representation reveals large-scale patterns in the data, while smoothing over noise, and predicts new knowledge that the database should contain. The inferred knowledge, which a user survey shows is often correct, is used as part of a feedback loop that shows contributors what the system is learning and guides them to contribute useful new knowledge.