An Interdisciplinary Literature Classifier Based on Multi-task Multi-label Learning

Interdisciplinary integration is one of the motive power of scientific innovation and development. In order to improve the classification effect of interdisciplinary literature, this paper adopts multi-task learning method to learn interdisciplinary literature categories with similar research topic. Aiming at the imbalance and intersectionality of the distribution of the categories of the literature in the field of Library and Information Science, this paper proposes a classification framework for interdisciplinary literature based on multi-task learning. The framework is based on BERT and improves the classification effect of the model in minority categories by introducing the machine reading comprehension task, which predicts the position of keywords in titles and abstracts. The results show that the multi-task learning method is more effective than decision tree, support vector machine, convolutional neural network, recurrent neural network and pre-trained models. In addition, compared with cost-sensitive method, the proposed method is more helpful for the minority class, and its Macro-F1 value has reached 74.84%.