Who moves with the robotaxi? A role-based NLP sentiment analysis of acceptance, trust, and action in autonomous mobility

This study investigates public engagement with SAE Level 4 robotaxi services based on more than 170,000 publicly available comments about Apollo Go in Wuhan, China. We use a role-sensitive acceptance–trust–action lens, where role is defined at the comment level (i.e., passenger, pedestrian, motorist) to reflect how the same person may adopt different positions across posts. Role stances are inferred at scale using a RoBERTa classifier fine-tuned on a hand-labeled subset, and Sentence-BERT embeddings are used for semantic clustering to distinguish acceptance evaluations, trust assessments, and action orientations. The analysis examines role-specific discourse patterns and how evaluative language aligns with perceived feasibility and constraints. Results reveal distinct engagement configurations. Pedestrians focus on safety, accountability, and intent legibility, with resistance outweighing trial-oriented language. Passengers express positive ride experiences and trust, but routine use is constrained by service usability frictions such as access rules, operating boundaries, and first–last-mile coordination. Motorists adopt a more defensive stance, citing mixed-traffic unpredictability, congestion externalities, and distributive or occupational concerns. Our study offers a role-structured account of robotaxi engagement and a scalable digital-trace pipeline that complements survey evidence by locating integration bottlenecks under real operations, informing governance and service design as robotaxi deployment moves toward city scale.