Text2Dec: Extracting Decision Dependencies from Natural Language Text for Automated DMN Decision Modelling

Abstract

Decisions are of significant value to organisations. Business decisions are often written down in textual documents, and modelling them is a tedious and time-consuming task. Although decision modelling has seen a surge of interest since the introduction of the Decision Model and Notation (DMN) standard, limited research has been conducted regarding automatically extracting decision models from the text. In this paper, we propose a text mining technique to automatically extract the decisions and their dependencies from natural language text to build the decision requirements diagram. A case-based evaluation is shown for the proposed mining approach with promising results. This approach can serve as a groundwork for further research in the field of decision automation.

Identification of Decision Rules from Legislative Documents Using Machine Learning and Natural Language Processing

“…The paper by Etikala et al [4], presents a text mining technique that is capable of automatic extraction of decisions and their dependencies from natural language text and building the decision requirements diagram (DRD). This approach labelled as Text2Dec uses open-source tool kits such as Stanford's core NLP 3 , NLTK 4 , neuralcoref 5 , and SpaCy 6 libraries to build the NLP pipeline which enables decision logic extraction. This approach was evaluated using a real-life use case of Employee Health Assessment.…”

“…The approach proposed by Rozinat and van der Aalst [1] present first efforts in mining decisions from transaction logs or audit trails through the use of machine learning techniques. More recently, work by van der Aa et al [2] focused on extracting decision logic and visualizing it as declarative process models from the natural language, while Arco et al [3], and Etikala et al [4] focused on decision rule extraction and the decision table and decision requirements diagram generation. Research on rule-based approaches often suffers from a lack of annotated source data, and the inherent complexity of extracting decision logic from natural texts [5,6,7,8].…”

Context-Aware Verification of DMN

“…For example, consider the model in Fig. 3 reproduced from [15]: if we want to perform table verification on the "Risk Level" table, its context would consist of the "BMI" and the "BMILevel" tables.…”

An Overview of Methods for Acquiring and Generating Decision Models

“…From textual descriptions As discussed in earlier sections, mining decision rules from text, using text mining, and transforming these rules into a decision table is one thing [4]. It is even more challenging to mine dependencies between decisions, and other elements of the requirements level [13].…”