An evolutional and conceptual structure based method of analogical reasoning

A computational method of analogical reasoning is developed based on a conceptual network which evolves as a series of reasoning tasks are processed. More specifically, given a pair of concepts whose logical relationship is unknown, the method first retrieves pairs of concepts which are similar to the given one and whose logical relationships are well-known, then infers the unknown relationship from the similarities among the retrieved and given pairs of concepts, and finally adds the inferred relationship to the original conceptual network. The retrieval of the similar pairs turns out to require the identification of isomorphic substructures of the network, which is a combinatorial and time-consuming task. Two strategies are presented to address this problem. First, the conceptual network is structured to facilitate the combinatorial substructure matching, and secondly an evolutional approach is incorporated into the method that minimizes the computational efforts in the long series of reasoning tasks. Our method is applied to a large chemical database system comprised of instances of organic compounds and reactions