Biomedical Entity Linking As Multiple Choice Question Answering

Although biomedical entity linking (BioEL) has made significant progress withpre-trained language models, challenges still exist for fine-grained andlong-tailed entities. To address these challenges, we present BioELQA, a novelmodel that treats Biomedical Entity Linking as Multiple Choice QuestionAnswering. BioELQA first obtains candidate entities with a fast retriever,jointly presents the mention and candidate entities to a generator, and thenoutputs the predicted symbol associated with its chosen entity. Thisformulation enables explicit comparison of different candidate entities, thuscapturing fine-grained interactions between mentions and entities, as well asamong entities themselves. To improve generalization for long-tailed entities,we retrieve similar labeled training instances as clues and concatenate theinput with retrieved instances for the generator. Extensive experimentalresults show that BioELQA outperforms state-of-the-art baselines on severaldatasets.