Event detection algorithm based on label semantic encoding

One major challenge in event detection tasks is the lack of a large amount of annotated data. In a low-sample learning environment, effectively utilizing label semantic information can mitigate the impact of limited samples on model training. Therefore, this chapter proposes the SALM-Net (Semantic Attention Labeling & Matching Network) model. Firstly, a Label Semantic Encoding (LSE) module is designed to obtain semantic encodings for labels. Next, a contrastive learning fine-tuning module is introduced to fine-tune the label semantic encodings produced by the LSE module. Finally, an attention module is used to match text encodings with label semantic encodings of events and arguments, thus obtaining event detection results. Experiments are conducted on the publicly available ACE2004 dataset, and the algorithm’s effectiveness is validated through an analysis of experimental results, comparing them with state-of-the-art algorithms. Article Highlights Innovative Event Detection: Introduces SALM-Net, an advanced model for event detection in texts, improving accuracy by focusing on semantic meanings of labels rather than just surface-level data. Semantic Encoding Advancement: Utilizes Label Semantic Encoding (LSE) for deeper understanding of text, enhancing the model's ability to interpret and classify events accurately in various contexts. Enhanced Learning with Limited Data: Demonstrates effective learning in environments with limited samples, using contrastive learning and attention mechanisms for better model training and event detection performance.