A Vector Space Approach to Process Model Matching using Insights from Natural Language Processing

Tim Niesen; Sharam Dadashnia; Peter Fettke; Peter Loos
In: Tagungsband Multikonferenz Wirtschaftsinformatik. Multikonferenz Wirtschaftsinformatik (MKWI-16), March 9-11, Illmenau, Germany, Pages 93-104, No. 1, Technische Universität Ilmenau, 2016.


Business process models have been widely adopted to support a variety of tasks regarding organizing, structuring and communicating business activities for many enterprises. As the number of models continually grows, methods to identify correspondences within existing model collection gain in importance. In this paper, we adopt the vector space model approach to the field of business process matching in order to leverage information from text statistics and, thus, to identify corresponding model elements. By using techniques from natural language processing, we seek to abstract from specific word forms or grammatical aspects like inflection to provide a higher level of abstraction. Furthermore, we use k-nearest neighbor classification of models to determine the significance of model elements per application domain, which is particular useful for cross-domain model collections. As a result, we present a semantic multi-phase matching algorithm as well as a prototypical implementation. The matching results produced by the algorithm have been evaluated in different domain settings to demonstrate the potential of the vector space approach in cross-domain model collections.

Weitere Links