Deep Learning for the Identification of Decision Modelling Components from Text.

Decision and process descriptions often find themselves encapsulated in long descriptions such as regulations or guidelines. Decision modelling aims at modelling the structure and logic of a decision. For decision modellers, analysing textual documents in search for relevant sentences is a time consuming activity. A promising research topic is to build decision models from text. In this paper, an automatic decision modelling component classifier using deep learning is proposed. Using a large dataset consisting of labeled sentences, the usability of deep learning techniques is investigated. In total three deep learning techniques are evaluated and compared to non-deep learning techniques using both Bag of Words and Term Frequency-Inverse Document Frequency. We conclude that classifying decision modelling components is possible and report that BERT for sequence classification is the best performing technique.