Analogy and Deduction for Knowledge Discovery

Analogy-based hypothesis generation is a promising technique for knowledge discovery. However, some hypotheses generated are nonsensical. This paper describes a two-phased method to increase the quality of analogy reasoning. The first phase employs an established approach to generate hypotheses through similarity matching. The second phase utilizes deductive reasoning to eliminate hypotheses that are clearly false or absurd. The basis for elimination is violation of common sense or domain knowledge, which is represented in a suite of ontologies. We describe a set of preliminary experiments conducted to validate this two-phased approach. The experiments involved much larger test cases than reported by any other analogy researchers, and the results are very encouraging.

• Type: Conference
• Year: 2004
• Where: IKE
• Authors: Jim Reynolds, Adam Pease, John Li
• Keywords: Analogy Reasoning, Analogy-based Hypothesis Generation, Hypotheses, IKE 2004, IKE 2007