Concept for the (semi-)automated generation of knowledge resources using unqualified documents as a basis for interactive assistance systems

Universität Ulm

MA Abschlussvortrag, Emre Inanc, Ort: O27/5202, Datum: 08.11.2018, Zeit: 18:00 Uhr

This master thesis deals with the application of machine learning algorithms on description data, which consists of short description texts of Mercedes-Benz vehicle components. Particularly, this work examines the usage of neural networks in order to build a model that is able to predict vehicle component labels by using natural language text input. Two models are implemented and evaluated. The Sequence-to-Sequence model shows poor results due to the lack of training data. Therefore, a custom model that fits the comparatively small number of samples is proposed. With a validation accuracy of around 60% using only around 3000 to 4000 samples, the model shows promising results. In future work, the number of training samples can be increased and the model could be incorporated into a chatbot, which can answer marketing questions about Mercedes-Benz vehicles.