Keyword-based Data Augmentation Guided Chinese Medical Questions Classification

Abstract

For the existing data of medical and health questions, the majority of them are so inarticulate short texts with few terms that the text features are sparse, posing a daunting challenge to relevant classification effort. Against this background, to enlarge the terms and datasets of short tests, this paper proposes a keyword-based data augmentation algorithm, which can be used in two ways: (1) With regard to short texts featuring few terms, for the purpose of keyword expansion, keywords are extracted by topic model and trained through domain knowledge-assisted word vector model to obtain synonyms of expanded keywords, so as to expand the original keywords; (2) with regard to incomplete health questions, the synonyms are used to replace original keywords. Then the augmented samples obtained by the above two methods are sent to the classifier. As a result, the algorithm in this paper significantly improves recall, precision and macro value compared to those without data augmentation.