AI-driven passenger experience monitoring in railway operations: a Reddit-based NLP analysis of service failures, sentiment and operational priorities

Abstract

This study develops an artificial intelligence (AI)-driven framework for using passenger-generated social media discourse as a complementary source of operational intelligence in railway service quality management. It examines whether Reddit discussions can reveal dimensions of passenger experience that are not adequately captured by official performance indicators. Using the official Reddit API, the study collected 61,328 documents published between July 2024 and July 2026 across five railway communities. The framework combines lexicon-based classification into seven theory-informed service failure categories with non-negative matrix factorization topic modelling, VADER sentiment analysis, lexicon-based emotion detection and robust regression. Delay-and-disruption discourse showed the highest proportion of negative documents, while safety discourse generated the greatest negativity intensity and visibility; comfort and cleanliness and ticketing and refunds were the most frequently discussed dimensions. Affective intensity and public engagement proved partially independent: delay and ticketing posts attracted significantly less engagement than average, safety, comfort posts significantly more. The index ranked comfort, cleanliness, staff, communication, ticketing and refunds as leading improvement priorities, with ticketing, delay dominating British, comfort dominating US discourse. Findings rest on self-selected Reddit communities and rule-based instruments validated internally rather than against human gold labels; they indicate digitally expressed passenger concerns, not representative estimates of the passenger population. Operators and regulators can use the framework as a low-cost, continuously updated instrument complementing official performance indicators. Its extension into an early-warning system is set out as a design proposition requiring prospective validation against operational incident data, not as a capability demonstrated here. This study highlights social media as a democratic space for public accountability in transit management. By amplifying organic passenger discourse, the framework empowers commuters who's daily lived experiences, such as comfort, cleanliness and communication issues, are often overlooked by rigid official KPIs. Recognizing these grievances underscores how transit quality directly impacts public trust, equity and socio-economic well-being. Furthermore, addressing regional variations in passenger concerns helps operators design more inclusive services. Ultimately, integrating this digital feedback into policy ensures that vital public infrastructure better aligns with the diverse safety, accessibility and comfort needs of the communities it serves. The study demonstrates how large-scale passenger discourse can be transformed into actionable railway service intelligence by integrating theory-informed classification, affective analysis, engagement modelling and a multidimensional operational priority measure.