SBB has the busiest network in Europe. This also includes deviations in operation (malfunctions, train cancellations, irregularities) which are often reported as free texts in form of a report to a central coordination office. This coordination center codes the message based on the text: each message is assigned some information for statics purpose. In the case of redundant messages, this process should be automated or semi-automated: by using modern NLP techniques, the Informations are automatically coded based on the text. If the prediction of the algorithm reaches a certain threshold, the Informations are assigned, otherwise the human being takes over the coding as usual.
As a result, an algorithm was developed in a productive environment including an inferencing pipeline, training pipeline and a python library, which can encode automatically the messages saving time for humans. Further, it has been shown how Deep Learning can be used in a real problem with the texts commonly used in Switzerland and how it can take over human tasks