An object-extracting approach based on non-negative matrix factorization

Previous object-extracting approaches, which are based on subgraph merging and partition, need to compute excessive modules tightness and modules overlap. An object-extracting approach based on nonnegative matrix factorization is presented. First, in order to design effective similarity measure between subprogram vectors, nonnegative matrix factorization is applied to dimensionality reduction of the usage matrix. Secondly, fuzzy clustering algorithm is used to partition subprogram set and type set respectively, and several modules are generated accordingly. Last, objects are extracted by computing the modules cohesion. This object-extracting approach cannot only reduce the time for computing the modules cohesion by reducing the number of generated modules, but also extract several object sets from legacy systems for the user.