Real-Time FRA Form 57 Population from News

This paper addresses a practical problem: the delay in obtaining information about railway incidents. It proposes a real-time system to extract data from news articles and populate the FRA Form 57, which is crucial for situational awareness. The use of vision language models and grouped question answering to handle the form's complexity and noisy news data is a significant contribution. The creation of an evaluation dataset is also important for assessing the system's performance.

• •Addresses the problem of delayed information in railway incident investigations.
• •Uses a pipeline involving vision language models and grouped question answering.
• •Creates an evaluation dataset for assessing system performance.