Improved Information Extraction from Bridge Inspection Reports using Fine-tuned Generative Pre-trained Transformers
Bibliographic record
Abstract
Bridge inspection reports contain a wealth of crucial data on bridge components and their related structural defects.This study introduces a novel method that harnesses the power of Generative Pre-trained Transformers (GPT) for improved information extraction from bridge inspection reports.While most studies in this domain focus solely on data extraction, this study transforms inspection data into a ready-to-use format for better utilization in condition assessment and predictive modeling, enabling better-informed budget allocation and decision-making.It employs (1) a baseline GPT model (BL-GPT), which leverages OpenAI's large language models to process textual inspection data through optimized prompt engineering, and (2) a finetuned GPT model (FT-GPT), which enhances the baseline by incorporating domain-specific training to improve performance.These models capture and evaluate the severity levels of reinforced concrete bridge defects based on textual inspection data.The models are validated on data extracted from 2,255 inspection reports-spanning a period of five years (2018-2022)for a set of bridges in Québec, Canada.The FT-GPT is found to significantly improve performance, stability, and reliability in detecting and standardizing the severity of different types of concrete defects in bridge decks.In specific, it achieves accuracy rates of 98.79%for rebar corrosion, 99.09% for delamination, and 98.64% for cracking, scaling, and spalling of concrete.This study demonstrates the potential integration of generative AI in asset management, an application that has yet to be realized.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.010 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.003 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".