Classification of Unstructured Text for RPA: A Comparative Machine Learning Analysis

Resumo

Effective human-robot collaboration in digital work environments depends on automated systems correctly interpreting human-generated free-text. Unstructured text types such as bug reports, customer requests, call center records, emails, and free comment fields cannot be processed directly by robotic process automation (RPA) due to the contextual expressions, typos, and stylistic inconsistencies they contain. RPA processes operate rule based. Making sense of this unstructured data, which does not conform to the rules, will strengthen human-robot interaction. In this study, it is proposed to use machine learning methods to provide automatic classification of unstructured texts. In this study, five basic classification algorithms (Logistic Regression, Naive Bayes, Support Vector Machines, Decision Trees, and Random Forest) frequently used in the literature were compared in terms of their features and computational methods. The findings reveal that classifying unstructured text has the potential to significantly increase the accuracy of RPA-based workflows. In this way, it is anticipated that improvements can be made by shortening process times through the interpretation of texts in RPA processes. As a result, it is envisaged that robot-human interaction can become more efficient by making sense of the unstructured data in RPA processes.