Entity Linking is a very effective NLP technique that allows free form text to be associated with entities in a knowledge base. Entity linking is a much more granular task than named-entity recognition where only the type of entity is detected. Many entity linkers that that leverage Wikipedia are used in many informational tasks including educational recommendation [1]. 

With the advent of large language models, we have an opportunity to leverage the LLM's understanding of the world to perform entity linking. While LLM-based entity linking has been attempted in the past, they usually involve finetuning an LLM to predict entities that are output classes. This approach suffers from massive computational cost involved in having to train the entire network when new entities need to be added to the entity linker [2,3]. 

This project aims to leverage a novel method that entails using descriptions of entities to build entity embeddings that can be compared with text phrases to check the alignment [4]. This approach is significantly computationally efficient compared to conventional methods. Their utility to healthcare will be explored during the project. 

References

[1] Brank, J., Leban, G. and Grobelnik, M., 2018. Semantic annotation of documents based on wikipedia concepts. Informatics , 42 (1). 

[2] Loureiro, D. and Jorge, A.M., 2020, April. Medlinker: Medical entity linking with neural representations and dictionary matching. In European Conference on Information Retrieval (pp. 230-237). Cham: Springer International Publishing. 

[3] Mohan, S., Angell, R., Monath, N. and McCallum, A., 2021, August. Low resource recognition and linking of biomedical concepts from a large ontology. In Proceedings of the 12th ACM conference on bioinformatics, computational biology, and health informatics (pp. 1-10). 

[4] Wu, L., Petroni, F., Josifoski, M., Riedel, S. and Zettlemoyer, L., 2019. S