A framework for knowledge base refinement through multistrategy learning and knowledge acquisition

This paper presents a general approach to knowledge base refinement which integrates multistrategy learning, active experimentation and guided knowledge elicitation. Three main features characterize this approach. First, knowledge base refinement is based on a multistrategy learning method that dynamically integrates the elementary inferences (such as deduction, analogy, abduction, generalization, specialization, abstraction and concretion) that are employed by the single-strategy learning methods. Second, much of the knowledge needed by the system to refine its knowledge base is generated by the system itself. Therefore, most of the time, the human expert will need only to confirm or reject system-generated hypotheses. Third, the knowledge base refinement process is efficient due to the ability of the multistrategy learner to reuse its reasoning process. The paper illustrates a cooperation between a learning system and a human expert in which the learner performs most of the tasks and the expert helps it in solving the problems that are intrinsically difficult for a learner and relatively easy for an expert.