Master's Thesis Tobias Dümmling

A Methodology to Evaluate the Feasibility of Automation with Natural Language Processing

Abstract

The industrial application of Natural Language Processing is in its infancy. Those that can utilize the technology will gain significant competitive advantage. This potential fuels the desire of companies to apply the technology and places pressure on decision-makers. They need to detect and leverage arising opportunities. However, they often lack the technical knowledge to do so. 

To bridge this gap, we propose a framework to access the feasibility of NLP based automation in a structured way. Therefore we perform grounded theory-based research based on six expert interviews to analyze the automation process with NLP. We name technical and economic characteristics to identify promising automation projects. We use our findings to propose a three-step methodology that guides the applicant along the evaluation process of a given use case. 

The methodology is evaluated using an industry use case. It is applied in a workshop like setting and the resulting recommendation to implement the project is validated with a prototype. In an online survey, business professionals are asked to apply the methodology by themselves. Participants consider the methodology valuable as it facilitates understanding and provides the right frame for discussion. Problems occur, especially with the evaluate the technical aspects. Thus experts from the technical domain are needed for the evaluation of a given use case. Our research sets the frame for this joint evaluation as it highlights the critical aspects of successful process automation with Natural Language Processing.

Research Questions

RQ 1: How does the automation process with NLP look today?

RQ 2: How can promising task candidates for automation with NLP be identified?

RQ 3: How should a framework look that accompanies the decision process of automation with NLP for a given use case?

RQ 1: How does such a framework perform when applied to an industry use case?