Rule Induction for Knowledge Acquisition in RKF

The RKF Year 2 Challenge problem highlighted the difficulty that Subject Matter Experts (SMEs) have in writing inference rules. During the evaluation period, the KRAKEN SMEs entered about five to seven rules each [Pool et al., 2003]. One solution to this problem is to provide support for learning inference rules from examples rather than requiring domain experts to write the rules themselves. We believe that this will provide another tool to accelerate the knowledge entry and knowledge base development. Learning from examples can help solve this particular problem because it is easier to describe examples than to write rules. There may also be independent sources of examples, perhaps in databases that don’t support the expression of rules. Those examples could be imported and used as the input data to a rule induction algorithm. This report describes the refinement and extension of a rule induction algorithm [Moriarty, 2000] originally implemented in PowerLoomTM [PowerLoom] as part of the High Performance Knowledge Base (HPKB) project.