Assessing domain adaptation in adverse drug event extraction on real-world breast cancer records

Gabriel Herman Bernardim Andrade, Tomohiro Nishiyama, Takako Fujimaki, Shuntaro Yada, Shoko Wakamiya, Mari Takagi, Mizuki Kato, Isao Miyashiro, Eiji Aramaki

Abstract: Highlights•Medical texts can have different characteristics from one to another.•Such differences can cause significant impact in the performance of information extraction models.•Fine-tuning a model with relatively low amount of the target domain data can notably improve model performance.•There are diminishing returns when fine-tuning the model with larger datasets.•The developed system can help uncover potential unknown ADEs during cancer treatment.