Transferability of General Polish NER to Electronic Health Records

Varování

Publikace nespadá pod Ekonomicko-správní fakultu, ale pod Fakultu informatiky. Oficiální stránka publikace je na webu muni.cz.
Autoři

ANETTA Krištof ARSLAN Mahmut

Rok publikování 2021
Druh Článek ve sborníku
Konference Recent Advances in Slavonic Natural Language Processing (RASLAN 2021)
Fakulta / Pracoviště MU

Fakulta informatiky

Citace
www
Klíčová slova EHR; Electronic health records; Healthcare texts; NER; Named entity recognition; NLP; Natural language processing; Slavic languages; Polish; PolDeepNer2; spaCy; Spark NLP
Popis This paper investigates the transferability of general Polish named entity recognition tools to the analysis of Polish health records. The tools, namely PolDeepNer2, spaCy’s pl_core_news_lg pipeline and Spark NLP’s entity_recognizer_md pipeline for Polish, were run on the pl_ehr_cardio corpus and their results were analyzed, paying special atten- tion to their performance when processing these highly specific texts and to the applicability of the results in the healthcare domain. Even though the precision of PolDeepNer2 proved to be superior to both spaCy and Spark NLP, the paper concludes that without additional training, general named entity recognition tools for Polish have very limited use in the medi- cal analysis of electronic health records. However, they could be helpful in partial tasks ranging from de-identification to entity disambiguation and discovery of mistyped entities or candidate entities that are not present in medical dictionaries.
Související projekty:

Používáte starou verzi internetového prohlížeče. Doporučujeme aktualizovat Váš prohlížeč na nejnovější verzi.