Amer Goli, Niklas Tscheuschner, Guo Xu, Juan Erazo, Kunal Udawat, Olga Otinova - Data Science (FB5) Semesterprojekt

Schaeffler Chatbot

Due to resource management, demands, and safety issues in industrial environments there is the mandatory requirement for delivery of information that has high accuracy, that is not ambiguous, and that offers consistency in interactive Question & Answer (Q&A) platforms such as Digital Assistants (here: chatbot functionality). This requires special knowledge not generally available outside, e.g., terminology. 

Although there is significant progress in the capabilities of Machine Learning, today’s NLP technology requires extensive training using manual supervision. Measuring and improving the Q&A process by using human reasoning feedback loops can lead to inconsistency, subjectivity, and randomness of the training process. Therefore, intent recognition often fails and cannot be explained properly. At Schaeffler we need to use natural language to easily dive into the documentation of our company portfolio of products and services offered to our customers and support our engineers. For that we already employ an extensible chatbot framework called eLISA.

We want to discover new relations in between several domains and topics that might be related due to their nature, which is currently not yet within the scope of eLISA.

As a result, the students of the team implemented a chatbot with dialogue functionality that is capable of answering questions about the particular Schaeffler products over structured and unstructured data sources.