Categorisation of Post-Sale Automotive Customer Feedback Using Natural Language Processing
Diego Vicente-Estévez, Ángel Dacal-Nieto, Andrés Paradela, Víctor Alonso-Ramos, Juan José Areal, Alberto Riomao · 2024
Automotive manufacturing is permanently searching new methods to improve its efficiency and quality. A relevant data source to do this is post-sale customer feedback. They can highlight situations in manufacturing processes not detected in the factory, that may repeat in the future. The analysis of these data is currently performed manually by experts, which try to label each new occurrence with known potential causes, in a long and inefficient process. This paper proposes a new system to analyse customer feedback in order to categorise it from their raw description, aiming to facilitate and boost problem solving in the related manufacturing process. It is based in the usage of Natural Language Processing and machine learning algorithms, since most of its description is based on non-standardized text fields, and in different languages. The solution has been validated with Stellantis automotive group data, obtaining that a combination of Decision Trees and Gradient Boosting classifiers provides an acceptable performance of 85% of accuracy, acting as a first solution to a very complex problem, with high potential of impact in the manufacturing efficiency.