Skip to content

Latest commit

 

History

History
51 lines (37 loc) · 4.92 KB

README.md

File metadata and controls

51 lines (37 loc) · 4.92 KB

Impact-Data-Mining

This research develops a framework using BART-large MNLI and BART-large CNN models to extract structural and community damage information from wind disaster reconnaissance reports. Training the model with these datasets enables it to perform 0SHOT-TC and summary generation, thereby mitigating the need for specialized training datasets related to wind hazard impacts. The accompanying diagram illustrates the overall framework of the project. Included are an evaluation from the Hurricane Ian Preliminary Virtual Reconnaissance Report (PVRR) by Structural Engineering Reconnaissance (StEER), a verification of the BART-large MNLI model to extract impact sentences from the Hurricane Otis PVRR, and the application of the framework to the Mw 7.8 Kaharamanmaraş - Türkiye Earthquake, demonstrating its versatility to various disaster contexts.

Framework Diagram

Project Contributors:

  • Huy Pham, Ph.D. Student, Wind & Society Laboratory, Department of Civil and Environmental Engineering, Virginia Tech, Blacksburg, VA, United States. Email: phamanhuy@vt.edu

  • Monica Arul, Assistant Professor, Wind & Society Laboratory, Department of Civil and Environmental Engineering, Virginia Tech, Blacksburg, VA, United States. Google Scholar Profile . Email: marul@vt.edu

Related Publications:

  • This project is based on the research paper titled "Automated Extraction and Summarization of Wind Disaster Data Using Deep Learning Models, with Extended Applications to Seismic Events" by H. Pham and M. Arul. The link to the full paper will be provided soon.
  • Pham, H., M. Arul. (2024) "Natural Hazards Research Summit 2024: Unstructured to Actionable: Extracting Wind Event Impact Data for Enhanced Infrastructure Resilience." DesignSafe-CI. https://doi.org/10.17603/ds2-r46y-g112 v1

Main Inputs

The primary inputs for this project are reconnaissance reports in PDF format and relevant keywords tailored to the areas of interest in the study.

Usage

Workflow Diagram

The diagram illustrates the project's implementation across three primary phases: Data Input, Execution, and Evaluation.

  • Data Input Phase: This initial phase involves preparing inputs for the subsequent stages. Start by converting the PDF report into a .docx format to maintain the text structure accurately (Adobe Acrobat Pro is suggested). If certain sections are deemed unnecessary, utilize the optional 00_optional_customize_scope_by_toc.ipynb file to review and possibly exclude them from the table of contents, which might differ from the document's original layout. Update the sections to be excluded and other inputs in 01_input.ipynb before proceeding to preprocessing.

  • Execution Phase: Begin processing with 1_preprocessing. Continue with notebooks 2a and 2b to extract impact sentences, followed by 3_generating_comprehensive_summaries.ipynb to generate impact summaries.

  • Evaluation Phase: Assess the quality of results by first manually labeling impact sentences in 4_data_labeling.ipynb. Then, use 5_analyzing_the_results.ipynb to determine confusion matrices and F1 scores, thereby evaluating the effectiveness of the BART-large MNLI model.

For detailed guidance on the usage and input requirements of each notebook, please refer to the information provided within the respective files. A comprehensive discussion of the code and methodologies used in this project is available in H. Pham's master's thesis, accessible here.

Important notes:

  • All notebooks are designed and should be run on Google Colab.
  • Users are advised to maintain the folder structure as in the repository to simplify navigation and access to files within the notebooks.
  • For 2a_extracting_impact_sentences.ipynb and 3_generating_comprehensive_summaries.ipynb, which utilize BART large MNLI and BART large CNN models, the use of GPU A100 is recommended. This ensures efficient handling of the extensive machine learning parameters and reduces processing time.

Report's links and models: