Key points are not available for this paper at this time.
The anonymization of unstructured texts has become a very popular and widely researched topic. This is due not only to the latest GDPR regulation, but also due to the development of state-of-the-art models in the field of natural language processing. The texts required for building such models have to be anonymized before and very often have to be anonymized on the premises of data providers, not the machine learning teams. In this work, we present the use of machine learning models such as part-of-speech tagger or named entity recognizer and their integration with regular expressions for anonymization of unstructured texts in Polish. The goal is to create a system that recognizes many types of sensitive data and can remove, tag, and pseudo-anonymize (replace with words from the same category in an appropriate form) the detected tokens. To test the performance of this system, we prepared a manually annotated dataset containing different categories of sensitive data. The paper presents a detailed analysis of the proposed method’s performance. Moreover, a deployment architecture is discussed in the paper, that results in the creation of a scalable tool capable of processing a large amount of data that can be easily used.
Oleksy et al. (Fri,) studied this question.