NER Based on Dependency Structure Feature Fusion

In this paper, we propose a named entity recognition model, KansNN-SynLSTM-CRF, based on the fusion of dependency structure features, which aims to combine contextual and dependency structure information to improve the accuracy of named entity recognition. The model captures complex dependencies by introducing a Hierarchical Graph Convolutional Architecture (KansNN), models linear context and structured information independently using SynLSTM cells, and optimises label sequences through a CRF layer. We conduct experimrnts on Catalan, Spanish, English, Chinese and EduNER datasets, and the results show that the model outperforms existing methods in both multilingual and cross-domain scenarios, while providing better generalisation ability and robustness. Further ablation experiments show that modules such as Attention, KAN and GCN are indispensable in improving performance. The research in this paper provides an efficient and scalable solution for the named entity recognition task.