Transportation Sentiment Analysis for Safety Enhancement

The main objective of this project was to develop a real-time Twitter monitoring system to automatically retrieve tweets related to transportation safety, extract the potential safety topics (e.g., traffic accidents, road flooding), calculate public sentiments, and finally visualize the topics and sentiments using word clouds, Open StreetMap, and other graphic tools. The potential users include transportation engineers (e.g., early identification of safety bottlenecks), transportation planners (e.g., adjustment of transportation policies in response to public sentiments and opinions), and public users (e.g., improved routing to avoid potentially unsafe regions). This objective was accomplished through four tasks, which are described in detail in this report. Task 1 (chapter 1) introduces related work for automatic Twitter data crawling and domain-specific content filtering, and presents improved and customized algorithms for transportation safety analysis. Task 2 (chapter 2) provides a literature review of related geocoding techniques for Twitter data and presents a hybrid approach that integrates location information from user profile, user social relationships, and location mentions in tweet texts. The geocoding was implemented at multi-resolutions, including state, city, and street levels, and estimated the quality score for each resolution. Task 3 (chapter 3) presents the procedures that were implemented to automatically discover safety related topics from the collected Twitter data, study the temporal evolution of these topics, and calculate their sentiment scores. The safety related topics were discovered at multiple geographic (e.g., city, state) and temporal (e.g., hourly, daily) resolutions. Task 4 (chapter 4) presents the implementation of a web-based interactive prototype system that allows users to query and visualize safety information and patterns that were discovered by previous tasks. Particularly, this chapter first presents the design of an optimized file management system that pre-computes and optimizes the storage of intermediate results for support of real-time performance at the client-side interface. It then presents technical details for the implementation of advanced visualization functions, such as work clouds, Open StreetMap, and dynamic query-based charting functions. Finally, it presents the graphic design of major web interfaces. The overall results show that there are a significant number of tweets discussing or reporting information related to transportation safety. The prototype system was able to retrieve high quality tweets in real time, and to geocode them to streets or even latitude/longitude locations. The web-based interactive interface allows users to quickly view the summary statistics of raw tweets, and to identify potential safety bottlenecks using the advanced topic discovery and sentiment analysis functions. Based on this prototype system, the second phase of the project for the next year period will focus on the development of more advanced machine learning functions for safety enhancement, including traffic accident detection, traffic congestion detection, and safety constrained routing for bikers and pedestrians.

Language

  • English

Media Info

  • Media Type: Digital/other
  • Edition: Final Report
  • Features: Figures; References;
  • Pagination: 40p

Subject/Index Terms

Filing Info

  • Accession Number: 01674442
  • Record Type: Publication
  • Contract Numbers: DTRT12GUTG11
  • Files: UTC, TRIS, RITA, ATRI, USDOT
  • Created Date: Jun 21 2018 12:30PM