Not stated in source 2021 Needs review

Natural language processing for textual process descriptions

Visit original source

Abstract

Textual descriptions of processes are ubiquitous in organizations as a way to store, maintain and manage procedures. Formal notations for processes exist in fields like business process management, for example BPMN. With the rise of Natural Language Processing there is a growing interest on using these techniques to bridge the gap between textual processes and formal tools to analyze them. Nevertheless, an overview of the literature reveals that often the process datasets used are either small or too simplistic. We compile processes from three different sources to train classifiers with self-learning techniques, in order to leverage readily available unlabeled data. We train the Universal Sentence Encoder on these datasets, where bootstrapping greatly outperforms fully-supervised training. Furthermore, we build a framework to train and use these classifiers to detect processes on PDF sources.

Keywords

Not available

Specialisations

Similar theses