ML-Net: multi-label classification of biomedical texts with deep neural networks
share
Background: Multi-label text classification is one type of text classification where each text can be assigned with one or more labels. Multi-label text classification, which has broad applications in biomedical domain, is often considered harder than other types of text classification, as each textual document can be assigned with indeterminate number of labels. Methods: In this work, we propose ML-Net, a novel end-to-end deep learning framework, for multi-label classification of biomedical tasks. ML-Net combines the label prediction network with a label count prediction network, which can determine the output labels based on both label confidence scores and document context in an end-to-end manner. We evaluated the ML-Net on publicly available multi-label biomedical text classification tasks from both biomedical literature domain and clinical domain. Example-based metrics including precision, recall and f-measure were calculated. We compared the ML-NET with both traditional machine learning baseline models as well as classic deep learning models. The codes are available at: https://github.com/jingcheng-du/ML_Net Results & Conclusions: The deep neural network based approaches showed superiority compared to traditional machine learning algorithms in multi-label classification of biomedical texts. Deep neural networks alleviate human efforts for feature engineering, and avoid building individual classifiers for each label. ML-NET is able to dynamically estimate the label count based on the document context in a more accurate manner.
READ FULL TEXT