A novel approach for classifying customer complaints through graphs similarities in argumentative dialogues

Automating customer complaints processing is a major issue in the context of knowledge management technologies for most companies nowadays. Automated decision-support systems are important for complaint processing, integrating human experience in understanding complaints and the application of machine learning techniques. In this context, a major challenge in complaint processing involves assessing the validity of a customer complaint on the basis of the emerging dialogue between a customer and a company representative. This paper presents a novel approach for modelling and classifying complaint scenarios associated with customer-company dialogues. Such dialogues are formalized as labelled graphs, in which both company and customer interact through communicative actions, providing arguments that support their points. We show that such argumentation provides a complement to perform machine learning reasoning on communicative actions, improving the resulting classification accuracy.