ANALYSIS AND EMPIRICAL-STUDIES OF DERIVATIONAL ANALOGY

Derivational analogy is a technique for reusing problem solving experi ence to improve problem solving performance. This research addresses a n issue common to all problem solvers that use derivational analogy: o vercoming the mismatches between past experiences and new problems tha t impede reuse. First, this research describes the variety of mismatch es that can arise and proposes a new approach to derivational analogy that uses appropriate adaptation strategies for each. Second, it compa res this approach with seven others in a common domain. This empirical study shows that derivational analogy is almost always more efficient than problem solving from scratch, but the amount it contributes depe nds on its ability to overcome mismatches and to usefully interleave r euse with from-scratch problem solving. Finally, this research describ es a fundamental tradeoff between efficiency and solution quality, and proposes a derivational analogy algorithm that can improve its adapta tion strategy with experience.