The investigation of vehicle defects, which is generally led by the National Highway Traffic Safety Administration (NHTSA) in the U.S., is critical to the continued trust of the general public in the safety of vehicles. NHTSA routinely receives millions of reports of potential defects, complaints, recalls, and manufacturer communications, which may provide evidence of a new vehicle defect. However, the large quantity and text-based communication make efficiently identifying defect trends difficult for analysts. To accelerate the investigation of defect reports, we introduce a natural language processing (NLP) application that identifies key topics and similar defect reports to assist analysts and investigators. Further, our application is built to provide users with a web interface for interacting with the NLP models. The integration of NLP with current NHTSA datasets provides a method for quickly identifying defect trends in large text-based datasets. To demonstrate the effectiveness of our method, we apply our approach to two publicly available NHTSA datasets, namely the Technical Service Bulletins and Recalls Dataset.

No abstract