TransCRF—Hybrid Approach for Adverse Event Extraction

In recent times, with the immense availability and usability of Internet, the revealing of adverse drug reaction (ADR) in pharmacovigilance has seen a surge in use of social media and online platform for reporting of ADR. This results in a huge volume of data that is generated in various human languages and need to be processed in order to derive meaningful insight regarding behavior of drugs. The extraction of relevant ADR information from the unstructured has become a very important NLP problem. Because of the tremendous volume of the information produced, it is preposterous to expect to handle the information utilizing conventional strategies. We propose an approach for identification of the ADR phrase using a combination of two different phrase extraction approaches and then collating the output of both the approaches to extract the adverse drug reaction phrases from the natural language text. The proposed method partial F1-score 82.4 for CADEC and 72.8 for SMM4H.