An approach for knowledge extraction from source code (KNESC) of typed programming languages

Abstract

Knowledge extraction is the discovery of knowledge from structured and/or unstructured sources. This knowledge can be used to build or enrich a domain ontology. Source code is rarely used. But implementation platforms evolve faster than business logic and these evolutions are usually integrated directly into source code without updating the conceptual model. In this paper, we present a generic approach for knowledge extraction from source code of typed programming languages using Hidden Markov Models. This approach consist of the definition of the HMM so that it can be used to extract any type of knowledge from the source code. The method is experimented on EPICAM and GeoServer developed in Java and on MapServer developed in C/C++. Structural evaluation shows that source code contains a structure that permit to build a domain ontology and functional evaluation shows that source code contains more knowledge than those contained in both databases and meta-models.