Contrast Learning for Conceptual Proximity Matching

Availability of general knowledge is considered essential in intelligent systems design to avoid brittle behavior. However, knowledge is long and tedious to acquire. This paper proposes a knowledge acquisition method that allows for the acquisition of useful general knowledge for semantic matching. The approach is based on the idea that processing example cases that contrast with existing knowledge but are conceptually close provide a learning opportunity. There are many possible applications for the proposed knowledge acquisition approach including eliciting knowledge for the semantic web and semantic bridging of heterogeneous databases. We present experimental results with a set of real life examples and demonstrate that the newly acquired knowledge facilitates processing of novel cases.