izpis_h1_title_alt

Anonymization of Slovenian texts
ID Petreski, Viktor (Author), ID Bajec, Marko (Mentor) More about this mentor... This link opens in a new window

.pdfPDF - Presentation file, Download (719,07 KB)
MD5: F75B6FABD5731A95E39D1D2CB95448F5

Abstract
Law and court documents may contain sensitive information that should not be accessible to everyone. Most of the publicly available documents are only partially anonymized or use a general anonymization system that makes the once anonymized data unsuitable for further research. We focus on developing several types of anonymization systems, with implementing NER, POS tagging, coreference resolution as well as specific anonymization rules that target Slovene entities. The anonymization models are integrated into a publicly available web tool, which allows convenient and fast anonymization in four different modes, depending on the sensitivity of the documents. The obtained results prove that the developed models can successfully anonymize Slovene documents. The anonymized documents can be distributed without violating someone's privacy and can be used in various research in the future.

Language:English
Keywords:text anonymization, Slovenian, de-identification, deep learning, transformer models
Work type:Master's thesis/paper
Typology:2.09 - Master's Thesis
Organization:FRI - Faculty of Computer and Information Science
Year:2022
PID:20.500.12556/RUL-142138 This link opens in a new window
COBISS.SI-ID:128899331 This link opens in a new window
Publication date in RUL:21.10.2022
Views:1229
Downloads:84
Metadata:XML DC-XML DC-RDF
:
Copy citation
Share:Bookmark and Share

Secondary language

Language:Slovenian
Title:Anonimizacija slovenskih besedil
Abstract:
Pravni in sodni dokumenti lahko vsebujejo občutljive podatke, ki ne bi smeli biti dostopni vsem. Večina javno dostopnih dokumentov je le delno anonimiziranih ali pa uporablja splošni sistem anonimizacije, zaradi česar so enkrat anonimizirani podatki neprimerni za nadaljnje raziskovanje. Osredotočeni smo na razvoj več tipov anonimizacijskih sistemov, z implementacijo NER, POS označevanja, koreferenčne ločljivosti ter specifičnih anonimizacijskih pravil, ki ciljajo na slovenske subjekte. Modeli anonimizacije so integrirani v javno dostopno spletno orodje, ki omogoča priročno in hitro anonimizacijo v štirih različnih načinih, odvisno od občutljivosti dokumentov. Dobljeni rezultati dokazujejo, da lahko razviti modeli uspešno anonimizirajo slovenske dokumente. Anonimizirane dokumente je mogoče distribuirati brez poseganja v zasebnost nekoga in jih je mogoče uporabiti v različnih raziskavah v prihodnosti.

Keywords:anonimizacija besedila, slovenščina, deidentifikacija, globoko učenje, transformer model

Similar documents

Similar works from RUL:
Similar works from other Slovenian collections:

Back