Transfer Learning in NLP for classification of chat messages

As businesses continue to expand their digital presence, customer support interactions are increasingly taking place through chat-like interfaces. This choice has created a demand for more sophisticated NLP technologies that can help optimize the support process. This thesis presents the work done during an internship, focused on the development of a classification algorithm for chat messages with the goal of optimizing the customer service process of Remedee Labs, a biotechnology and digital health startup. The proposed algorithm utilizes a pre-trained state-of-the-art french language model to leverage text embeddings, enabling effective classification through traditional machine learning methods in a transfer learning fashion. In addition, an innovative approach is presented. It aims to enhance classification by integrating external information to the standard methods.