Research on Railway Official Document Text Correction Algorithm Based on GhatGLM Model and Domain Knowledge Integration

The accuracy and standardization of railway official documents directly influence the efficiency of railway operations and the effective implementation of policies and regulations. However, the mainstream approach to correcting railway official documents still relies on manual correction, which is time-consuming and labor-intensive. Moreover, the current mainstream text correction methods are not tailored for the railway document field. To address the correction tasks for railway official document texts, this paper proposes a railway document text correction algorithm based on the integration of the GhatGLM model and domain knowledge. In response to the issue that current correction models are not targeted at the railway document domain, this paper independently constructs a railway document correction dataset and utilizes LLaMA-Factory for fine-tuning. Considering the abundance of proper nouns in railway document texts and the need for real-time updates, this paper builds a proper noun database specific to the railway document domain and employs edit distance matching and cosine similarity algorithms for accurate corrections. Experimental results demonstrate that the proposed algorithm exhibits excellent performance in handling actual railway document texts, validating the model’s effectiveness and practicality. It provides robust technical support for improving the quality and efficiency of railway document text correction.