A Novel Low-Resource Graph Convolutional Neural Network Approach for Entity Recognition

With the rise of large language models in natural language processing, the use of various generative large language models for entity recognition has become a cutting-edge research approach. However, such models generally necessitate hundreds of millions of parameters and tens of billions of datasets for training, leading to models that are exceedingly demanding in terms of hardware resources and necessitate costly computational resources. Meanwhile, Bert-like pre-trained language models rely on a large amount of textual data for pre-training and on a certain amount of labeled data for fine-tuning, which has low usability in the biomedical domain. In this paper, we propose a novel low resource graph convolutional neural network entity recognition method BGBGCN. Specifically, by overlaying a bidirectional graph convolutional network on top of a bidirectional long short-term memory network, we jointly model the dependencies and topological structures within a sentence sequence, taking into account both the linear characteristics and internal dependency features of entities. In addition, a new feature integration unit RAFI is designed, which can explicitly and interactively learn two different types of features by combining a gating mechanism with a graph encoder. Furthermore, considering the influence of out-of-vocabulary words on the model, we use convolutional neural networks to improve the character-level embedding process. The experimental results show that our proposed model achieves results comparable to those of large models such as PubMedBERT.