Network and Cluster Analysis on Bridge Inspection Reports Using Text Mining Algorithms

According to 2021 ASCE Infrastructure Report Card, 7.5% of the nation’s more than 617,000 bridges were found to be structurally deficient. About 42% of all bridges across the country need to be replaced, widened, or rehabilitated. Although bridges are regularly inspected every two years, the current bridge program views a bridge as an isolated entity for the condition measurements in a state. While the National Bridge Inventory has been operated for over 25 years, the database relies mostly on the inspection report data. Currently, no integrated database system is available to embrace the extensive information of a bridge from historical and transactional data in economic, physical, and social sources. This paper introduces new approaches of network and clustering analyses using text mining algorithms to gather multi-source heterogeneous data generated by a specific event (e.g., planning, design, inspection, and repair) and to interpret the complex data. Specifically, we performed a text mining tool to extract features (e.g., girder beam and bearing devices) from bridge inspection reports. The features were cleaned out by imputation and filtering out features with large missing values. Given the semi-structured features, further analyses, such as network analysis and clustering, were performed. This automatic data processing system shows the potential to extract features of interest and efficiently generate big data from historical and transactional data to establish the federated data analysis for bridge management. The federated data makes a framework from similar types of bridges by identifying interactions, interdependences, and interrelationships between them as pathognomonic signs or symptoms practiced in medical practice.