Research on Key Information Extraction Method of Navigational Warnings Based on Rank-Stable Llama 3 Fine-Tuning

Navigational warnings provide key information such as the type of danger, impact range, and effective time in the area where the ship is located, which is key information that needs to be processed during the navigation process of intelligent ships. This paper combines rank stability with the principal singular values and singular vectors adaptation (PiSSA) fine-tuning method to classify navigational warnings according to different danger properties and extract key information from them. This method has been tested on a large number of real navigational warnings, showing that it can effectively extract key information from navigational warnings, including time, geographic location, and danger type, providing an effective means for intelligent ships to process navigational warnings.