In a business process, we expect multiple related entities to interact. For example, in the Order-to-Cash process, one can observe the events that are recorded with respect to the orders, to the packages, or to the items that are included in the order. Each case notion carries only a part of the aspects of the overall situation, thus, reducing the process to a single case notion means to deliberately neglect certain facets of reality, and moreover, it conceals a major risk to present features of one particular case notion as the global truth of the process. The goal of this work is to structure a problem and suggest a solution for discovering patterns when a business process involves multiple entities. We propose an embedding representation that captures simultaneously the similarity of traces within the objects of the same object type, as well as the relationships between the objects of different types. We formulate an optimization problem that involves the similarity matrices, the cross-objects types relationships matrices, and the embeddings. Then, we follow an iterative algorithm to optimize it and deliver the embedding representation, and eventually the cluster memberships for each object type.
P. Delias, L. Moussiades, V. G. Kaburlasos, “Potentials for decision support in business processes through a multi-layer network embeddings approach”, 32nd EURO Conference, Espoo, Finland, 3-6 July 2022, p. 191.