Named Entity Recognition from Table Headers in Randomized Controlled Trial Articles

Tables in biomedical articles often contain important information of research findings. However, they are often not available for direct uses by downstream computational applications due to its unstructured nature, with both structural and semantic complexity. In this study, we developed a deep learning-based approach that takes contextual information into consideration to recognize biomedical entities in tables headers in Randomized Controlled Trial (RCT) articles, using a manually annotated corpus. Our evaluation shows that it achieved good performance with an F1 score of 92.60% for entity recognition in headers. We believe the proposed approach for table information extraction, as well as the developed annotated corpus, would be great resources for biomedical text mining, thus facilitating other biomedical research and applications.