Structured Information Extraction from Pharmaceutical Records

Publikace nespadá pod Filozofickou fakultu, ale pod Fakultu informatiky. Oficiální stránka publikace je na webu



Rok publikování 2019
Druh Článek ve sborníku
Konference Proceedings of the Thirteenth Workshop on Recent Advances in Slavonic Natural Language Processing, RASLAN 2019
Fakulta / Pracoviště MU

Fakulta informatiky

Klíčová slova structured information extraction; table understanding; entity recognition
Popis The paper presents an iterative approach to understanding semi-structured or unstructured tabular data with pharmaceutical records. Thetask is to split records with entities such as drug name, dosage strength,dosage form, and package size into the appropriate columns. The data isprovided by many suppliers, and so it is very diverse in terms of structure.Some of the records are easy to parse using regular expressions; othersare difficult and need advanced methods. We used regular expressionsfor the easy-to-parse data and conditional random fields for the morecomplex records. We iteratively extend the training data set using theabove methods together with manual corrections. Currently, the F1 scorefor correct classification into 5 classes is 95%.
Související projekty: