Exploring Transformer-Based Learning for Negation Detection in Biomedical Texts

NLP techniques have been widely adopted in the biomedical domain to perform various text-analytics tasks, such as searching literature and extracting deriving new knowledge from data. One type of data is clinical texts (e.g., cases medical records), which typically contain physicians’ notes about a patient’s health, including previous history (symptoms, diseases, lab exams, treatments, etc.), every visit hospital leads addition more information record. Another biological articles, discuss explore certain phenomenon, behavior entities genetic relations interactions among them) roles specific processes causing diseases how amplification can cause tumorous diseases). For both types data, negation detection an essential analytics task that be applied identify negated contexts text detecting presence statement establishing patient does not have/fit condition or statements indicate nonexistence entities). This has addressed prior work by considering variety approaches rule-based systems, conventional machine-learning classifiers, deep learning approaches. In this work, we propose applying transformer-based for texts. We use pre-trained BERT other similar models (such ALBERT, XLNet, ELECTRA) address two negation-detection subtasks: sentence identification scope recognition. evaluated our approach using BioScope corpus relying on measures accuracy, precision, recall, F1, percentage correct scopes (PCS). Our findings show potential detection, reaching accuracy 99% PCS 95%