NLM IRP Seminar Schedule

Abstract:

Biomedical relation extraction (RE) aims to develop computational methods to extract the associations between biomedical entities from unstructured texts automatically. This task is crucial in various biomedical research topics such as biological knowledge/drug discovery. Most existing RE approaches formulate this task as a classification problem to categorize the entity pairs with relation or not. This type of methods is required to process all the pairs between two entities one by one, which is very time-consuming and not able to handle large-scale data using advanced deep learning techniques. Moreover, these methods ignore the dependency between multiple relations since they deconstructed RE into multiple independent relation classification subtasks. To address these problems, we propose a novel sequence labeling framework for the biomedical RE task. Our proposed framework has been evaluated on two independent applications. 1) Drug-protein interaction extraction, 2) Recognizing the corresponding species of gene names in the literature. Taken together, our proposed framework is more efficient and is able to fully exploit the dependencies of relations for improved performance on biomedical RE tasks.