Institutional Repository of Vilnius University
Didelių kalbos modelių adaptavimas nustatant plazminių ląstelių skaičių nestruktūrizuotuose medicininiuose įrašuose /
Abstract
dc:descriptionThe work applied natural language processing technologies of the artificial intelligence branch to a relevant task in a medical context. A dataset containing unstructured medical records and information about a relevant numerical value mentioned in some records was used. Given this mechanically prepared dataset, efforts were made to train and refine a large question answering language model (“BERT-Small fine-tuned on CORD-19 QA dataset”) achieving good accuracy in predicting the placement of the required number in the text. Starting with a baseline model with rather poor accuracy (32.65% on the test set), experimentation and multiple significant modifications led to significantly better results (89.27% average accuracy with cross-validation). As the results were achieved with a small dataset (479 rows), the applicability of the model to other domains and tasks is very feasible, and even better accuracy can be expected with larger amounts of data.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Braun, Ugnius Byron,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD210643362&prefLang=en_US