Browsing by Subject "pseudonymisointi"

Deep neural networks are widely used in natural language processing. Large language models, trained with large corpora, enable improved information extraction from data that is too large for human processing. This thesis reviews the performance of a deep learning natural language processing pipeline in detecting and removing (anonymising) personal information. Methods for fast and accurate ano- or pseudonymisation of data containing sensitive information are vital to research and development in science and industry, as legislation demands extensive procedures concerning handling of data with direct or indirect personal information. We propose a method that achieves state of the art results on noisy data, and good performance on a contemporary benchmark. Our comparison of anonymisation performance is one of the first for Finnish free texts.