%0 Conference Paper %1 falis-etal-2022-horses %A Falis, Matús %A Dong, Hang %A Birch, Alexandra %A Alex, Beatrice %B Proceedings of the 21st Workshop on Biomedical Language Processing %C Dublin, Ireland %D 2022 %I Association for Computational Linguistics %K clinical_coding data_augmentation evaluation icd icd-9 multi-label-classification myown ontologies zero-shot zsl %P 389--401 %T Horses to Zebras: Ontology-Guided Data Augmentation and Synthesis for ICD-9 Coding %U https://aclanthology.org/2022.bionlp-1.39 %X Medical document coding is the process of assigning labels from a structured label space (ontology -- e.g., ICD-9) to medical documents. This process is laborious, costly, and error-prone. In recent years, efforts have been made to automate this process with neural models. The label spaces are large (in the order of thousands of labels) and follow a big-head long-tail label distribution, giving rise to few-shot and zero-shot scenarios. Previous efforts tried to address these scenarios within the model, leading to improvements on rare labels, but worse results on frequent ones. We propose data augmentation and synthesis techniques in order to address these scenarios. We further introduce an analysis technique for this setting inspired by confusion matrices. This analysis technique points to the positive impact of data augmentation and synthesis, but also highlights more general issues of confusion within families of codes, and underprediction.

@inproceedings{falis-etal-2022-horses, abstract = {Medical document coding is the process of assigning labels from a structured label space (ontology {--} e.g., ICD-9) to medical documents. This process is laborious, costly, and error-prone. In recent years, efforts have been made to automate this process with neural models. The label spaces are large (in the order of thousands of labels) and follow a big-head long-tail label distribution, giving rise to few-shot and zero-shot scenarios. Previous efforts tried to address these scenarios within the model, leading to improvements on rare labels, but worse results on frequent ones. We propose data augmentation and synthesis techniques in order to address these scenarios. We further introduce an analysis technique for this setting inspired by confusion matrices. This analysis technique points to the positive impact of data augmentation and synthesis, but also highlights more general issues of confusion within families of codes, and underprediction.}, added-at = {2022-05-18T21:18:35.000+0200}, address = {Dublin, Ireland}, author = {Falis, Mat{\'u}{\v{s}} and Dong, Hang and Birch, Alexandra and Alex, Beatrice}, biburl = {https://www.bibsonomy.org/bibtex/2e07aaa826b42890aad83ed46baf05d96/hangdong}, booktitle = {Proceedings of the 21st Workshop on Biomedical Language Processing}, description = {Horses to Zebras: Ontology-Guided Data Augmentation and Synthesis for ICD-9 Coding - ACL Anthology}, interhash = {00d2e2104f11197f1f17462cb47dbf32}, intrahash = {e07aaa826b42890aad83ed46baf05d96}, keywords = {clinical_coding data_augmentation evaluation icd icd-9 multi-label-classification myown ontologies zero-shot zsl}, month = may, pages = {389--401}, publisher = {Association for Computational Linguistics}, timestamp = {2022-05-18T21:18:35.000+0200}, title = {Horses to Zebras: Ontology-Guided Data Augmentation and Synthesis for {ICD}-9 Coding}, url = {https://aclanthology.org/2022.bionlp-1.39}, year = 2022 }