Master thesis : Integration of a chatbot in Industry 4.0

Abstract

Nowadays, many tasks can be automated moreover the number of interactions necessary for a user to find the information he is looking for can also be decreased. It is on these two foundations that this thesis was based to create a chatbot that could be integrated into Technifutur environment but which would be based on a global architecture that could be integrated into other companies with the same objectives. We decided on two main objectives which were to answer the questions of the user on a particular domain, and also to answer and advise the user regarding the training offered by Technifutur while keeping in mind that the chatbot should respond as much as possible like a human, allowing maximum flexibility and control. An important fact that is very challenging for this thesis is that we assume that all chatbot users speak French while the large majority of natural language processing (NLP) works are in English.

To develop our thesis, we had first to explore the different state-of-the-art (SOTA) models and tasks in NLP which are constantly and rapidly evolving during the last years, in order to define which techniques were the most suitable for creating the chatbot. We then developed an architecture based on two distinct models on two different tasks. These models were trained using publicly available data as well as custom data. The database of training offered by Technifutur has also been connected to the architecture in order to provide the most up-to-date information possible and a very simple interface has also been created so that the user is in a simplified environment when interacting with the chatbot.

The final result meets Technifutur’s expectations because the chatbot already allows to answer about one hundred questions covering a large majority of the questions usually asked by users. Additionally it also allows some flexibility in the form of the question asked by the user, and it gives access to a tool allowing its improvement on two axes according to user feedback. This will allow it to be more and more complete over time while maintaining strict control and interesting customization. It also makes it easier to look after and register for a training compared to the current Technifutur website by offering a new search system that greatly surpasses the one used previously.