ARTH - Automated Rail Traveller inquires Handling

ARTH outlines an end-to-end product to automate and scale the processing of train passenger requests received by the ÖBB. The proposal is composed of modules that implement different processing stages. An implementation of each module is sketched below. Each of them can be improved to refine its functionality or to expand its capabilities.
1. Data gathering: The currently existing method must be replaced by an user-friendly online form that is filled via web to digitalize the process and enforce data validation. The submission of forms filled in paper should still be accepted to avoid any kind of discrimination; these requests will be pre-processed using a specially designed computer vision technology process.
2. Data extraction and early request filtering: Data is inserted into a secured database under the ÖBB domain. Data is structured using natural language processing (NLP) to extract relevant names, validate specific fields of the form and transform their open text fields into numeric values suitable to be fed into an algorithm. Such an algorithm filters the enquiries that do not correspond to passenger rights’ requests. On those discarded, a different NLP algorithm is used to summarize the nature of the inquiry and assign a category to it. The inquiry and its summary is sent to their pre-specified departments for processing.
3. Request validation: The details of the request are validated by comparing the provided data with the ÖBB database; e.g. it is checked whether the delay/cancellation being claimed matches with the ÖBB database, if the provided alternative route/train existed and is valid, if the ticket provided in the enquiry is valid and has not claimed another compensation before, etc. Requests that do not pass this validation are automatically handled: an email is sent to the provided address with an automatic explanation of the validation failures.
4. Request categorization: Based on the validated data the system assigns a category to the request. Delay time is automatically calculated for those requests that only involve delays, splitting them between “normal” (60-120 min) or “abnormal” (>120min). Inquiries that involve train cancellation affecting the customer’s journey are categorized accordingly and alternatives taken to mitigate the cancellations are also noted, together with their cost.
5. Compensation proposal: A different compensation procedure is applied depending on the assigned category. The degree of autonomy of each procedure depends on its complexity and the risk of the involved compensation; e.g. inquiries labeled as “normal delays” might be (almost) fully automated, posting the corresponding wire transfer request. In more involved categories, a compensation proposal will be made by an algorithm, created on the basis of predefined rules or trained with examples of previous enquiries. In all cases, the system will be overseen by human analysts that will be able to iterate with the proposed solution.