Refining Bridge Engineering-based Construction Scheme Compliance Review with Advanced Large Language Model Integration

The review of bridge construction schemes necessitates the accurate matching with relevant code provisions and a fine-grained comparative analysis. Traditional manual-based review methods are time-consuming and susceptible to inaccuracies. With the advent of large language models (LLMs), there exists a potential to improve the efficiency and accuracy of these reviews by using the vast informational content of code provisions. This study proposed a novely collaborative methodology combining a pre-trained model with a LLM to optimize engineering construction scheme reviews. First, the semantic representation capabilities of the pre-trained model are utilized to identify the most pertinent code provisions at the sentence level. Second, the generalization ability of the LLM is harnessed to fine-tuning itself for precise knowledge extraction and evaluation specific to construction scheme texts. To verify the method effectiveness, an actual bridge engineering construction scheme was used for validation. The validation results demonstrate that this method achieves a sentence-level semantic matching accuracy of 91.79% and a fine-grained knowledge extraction review accuracy score of 0.881. These findings not only verify the method's applicability to bridge construction schemes but also highlight its potential to enhance the standardization and intelligence of bridge construction processes.