Published: 
By  Sensing and Evaluation Laboratory (I-S2EE)

Congratulations to Tianshu Li for getting her second paper “Mapping Textual Descriptions to Condition Ratings to Assist Bridge Inspection and Condition Assessment Using Hierarchical Attention ” accepted for publication in Automation in Construction (Elsevier). This paper seeks to improve the accuracy and consistency of manually assigned condition ratings of bridges, by leveraging the narrative descriptions from bridge inspection reports as an untapped data source and proposes a data-driven framework to map natural language descriptions to quantitative ratings. A hierarchical architecture employing recurrent neural network encoders with an attention mechanism was developed using a collection of reports from the Virginia Department of Transportation, which outperformed a variety of baseline systems. Visualization of the resulting attention patterns was shown to provide interpretable insights which highlight potentially-overlooked indicators embedded in the narrative descriptions.
Reference:
Li, T., Alipour, M., and Harris, D. K. (2021 accepted). “Mapping Textual Descriptions to Condition Ratings to Assist Bridge Inspection and Condition Assessment Using Hierarchical Attention”. Elsevier Automation in Construction.