The Dynamic Vectors-Based Attention Model for Chinese Mathematical Term Extraction

Term extraction aims to extract domain-specific terminology from documents. Secondary school mathematics terminology plays a guiding role in the learning and teaching of secondary school mathematics. We found that there needs to be more work in this field and that there are two main issues in extracting the terms from the middle school mathematics text: the polysemy of one character problem and nested terms problem. Accordingly, we have constructed a secondary school mathematics corpus and proposed a model to resolve these problems, termed BERT-LLA-CRF. Using pretraining to obtain dynamic word vectors to improve the model’s understanding of text semantics and solve the problem of multiple meanings of one word, the proposed model integrates attention mechanisms and Chinese character information to solve the nested terminology problem. The experiment results show that the proposed model effectively solves those problems and achieves good middle-school mathematics term extraction results.