This project examines how public feedback can strengthen transportation asset management, a field traditionally driven by physical condition metrics such as pavement age and roughness. Using 925 public comments collected by the Southwestern Pennsylvania Commission, it applies a semi-automated workflow combining generative AI, natural language processing, and BERTopic topic modeling to convert unstructured feedback into actionable themes. Comments were geolocated, linked to census tract data, and grouped by tract for analysis.
Project Outputs
A reproducible NLP and topic-modeling pipeline, with documented validation safeguards (manual review of a sampled comment set, cross-checks against traditional topic modeling, standardized neutral prompts, and word-cloud verification). The analysis produced 14 thematic clusters across three categories: Physical Infrastructure, Traffic Flow and Transit, and Non-Motorized User Safety. Of 925 comments, 864 were geolocated and analyzed at the tract level; roughly 30% addressed physical infrastructure directly. Deliverables include the report, a TRB paper, and a journal manuscript in progress.
Outcomes and Impact
The study demonstrates that AI-assisted analysis can help agencies process public input at scale and integrate community priorities into infrastructure decisions. Comparison with International Roughness Index data shows community sentiment complements engineering metrics by surfacing concerns invisible to condition measures alone, giving agencies an evidence-based method to align maintenance and safety investment with documented public need across census tract groups.
Universities Involved
University of Pittsburgh
Principal Investigators
Lev Khazanovich
Julie Vandenbossche
Subject Areas
Infrastructure Design and Planning, Public Policy
