In this paper a distributed connectionist-based algorithm for analogical mapping is introduced, where the Labeling Recursive Auto-Associative Memory (LRAAM) (Sperduti 1993) network is used for computing "distributed representations" of object-oriented specification components. Two types of constraints are considered in this work: semantic and structural. Semantic constraints impose the restriction that the elements, one from a previous solved solution and another from a new problem, must share some common meaning to be mapped, whereas structural constraints impose the restriction that the elements must exist in similar structural relationships to be mapped. The proposed mapping algorithm is tested in the domain of software engineering, where the aim is to use analogical reasoning in the reuse process of object-oriented specifications. This paper demonstrates that under certain conditions the LRAAM network is capable of computing similar "distributed representations" for similar object-...

Mapping is an important process of analogical reasoning, by establishing correspondences between elements of previous solutions and new problem. After recalling potential analogs from a long-term memory based on surface similarities, a subsequent mapping process captures deeper similarities by establishing correspondences between elements based on common relational structures (Falkenhainer et al. 1989). An analogical mapping process is also constrained semantically, where the probability of mapping elements in the two analogs increases if they have similar meaning. In this paper we describe a technique for applying neural networks in the analogical mapping process that includes both structural and semantic constraints. Directed graphs related to reuse of object-oriented specifications are used in a set of experiments and report promising results.