An Improved Automatic Extraction of Chinese Mathematical Terminology with Iterated Dilated Residual Gated Convolutions

摘要

Automatic term extraction (ATE) is the primary work in ontology construction, text summarization, and knowledge graph. For automatic extraction of Chinese mathematical terminology, we propose an improved model base on Robustly Optimized BERT Pre-training Approach (RoBERTa), Iterated Dilated Residual Gated CNN (IDRG-CNN), and Bidirectional LSTM with a CRF layer (BiLSTM-CRF). To evaluate the model, we annotate a corpus of Chinese terms for probability theory and mathematical statistics. We divide the annotated corpus into a training set, verification set, and test set and ensure that the terms of the three data sets are not repeated to evaluate the model's generalization ability. Empirical results show that this model can effectively extract terms and reveal that the model has strong generalization ability, particularly which improves the ability to recognize long terms.