Biomedical Event Detection Based on Dependency Analysis and Graph Convolution Network

Biomedical event detection is one of the most important tasks in biomedical event extraction, providing an important basis for disease prevention and drug development. The existing methods treat event detection tasks as multi-classification or sequence annotation tasks, only considering the sequence representation of sentences and striving to obtain more contextual information in sequence models. However, they overlook the shortcomings of sequence modeling methods in capturing long-distance dependency problems and the impact of syntactic structure dependencies on event detection performance. Therefore, the paper proposes a biomedical event detection model based on dependency analysis and graph convolutional neural networks. Firstly, we constructed a feature extraction framework based on BioBERT word embedding combined with entity type embedding and dependency parsing, effectively extracting sentence level features from natural language texts. In addition, dependency analysis is used to perform syntactic analysis on sentences, identify the grammatical dependencies between words in the sentence, and construct a dependency syntactic structure graph. Finally, a graph convolutional neural network is used to perform convolution operations on the dependency syntax graph, and the dependency relationships between various nodes in the dependency structure graph are dynamically updated during the training process, more fully capturing long-distance dependency relationships in sentences, effectively identifying and classifying the event trigger words in the sentences. The experimental results show that the proposed method achieves better performance on the MLEE dataset.