Building a commonsense knowledge base for context-awareness inference

Abstract

Current context-aware systems often model rigid inference rules for limited user contexts, which constrain their effectiveness in real world usage. In order to achieve flexible context-aware inference, a commonsense knowledge base is essential. But such knowledge base is hard to construct manually. This paper proposes some automatic algorithms to extract commonsense directly from text corpuses. We evaluate the extraction algorithms with comparison with human annotators. We also evaluate the effectiveness of the knowledge base in practical context-awareness inference.