Conformance checking using activity and trace embeddings

Abstract

Conformance checking describes process mining techniques used to compare an event log and a corresponding process model. In this paper, we propose an entirely new approach to conformance checking based on neural network-based embeddings. These embeddings are vector representations of every activity/task present in the model and log, obtained via act2vec, a Word2vec based model. Our novel conformance checking approach applies the Word Mover’s Distance to the activity embeddings of traces in order to measure fitness and precision. In addition, we investigate a more efficiently calculated lower bound of the former metric, i.e. the Iterative Constrained Transfers measure. An alternative method using trace2vec, a Doc2vec based model, to train and compare vector representations of the process instances themselves is also introduced. These methods are tested in different settings and compared to other conformance checking techniques, showing promising results.