Details

Primerjalna analiza učinkovitosti detektorjev pri prepoznavanju sintetično ustvarjenih besedil : magistrsko delo
ID Shebunchak, Nazariy (Author), ID Pisanski, Jan (Mentor) More about this mentor... This link opens in a new window

.pdfPDF - Presentation file, Download (3,44 MB)
MD5: 063B8DA5404D3620C0D33A4EFA3E6DF3
.docxDOCX - Appendix, Download (65,06 KB)
MD5: 42B680B1DE7AF263399DE97615BC2471

Abstract
Uporaba orodij umetne inteligence za ustvarjenje sintetičnih besedil odpira vprašanje zanesljivosti njihovega avtomatskega prepoznavanja. Namen raziskave je bil ugotoviti, kako učinkovito izbrani detektorji razlikujejo med avtentičnimi izvlečki s področja bibliotekarstva, sintetično ustvarjenimi izvlečki in počlovečenimi različicami sintetično ustvarjenih besedil. Analizirani so bili trije detektorji, ki omogočajo detekcijo slovenskih besedil: AIDetector.com, ZeroGPT.com in MyDetector.ai. V analizo je bilo vključenih 44 izvirnih izvlečkov znanstvenih člankov v slovenskem jeziku, objavljenih v reviji Knjižnica v obdobju 2012–2017. Naslovi teh člankov so bili uporabljeni kot vsebinsko izhodišče za pozive v orodju ChatGPT, na podlagi katerih so nastali sintetični izvlečki. Tretjo skupino so predstavljala počlovečena besedila, dodatno preoblikovana v orodju ChatGPT iz sintetičnih izvlečkov tako, da bi se približala naravnemu človeškemu pisanju in zmanjšala oceno sintetičnosti pri detekciji. Rezultati so pokazali, da detektorji niso povsem zanesljivo razlikovali med človeško pisanimi in sintetično tvorjenimi besedili. Sintetične izvlečke so praviloma pogosteje prepoznali kot umetno ustvarjene, vendar so se pojavljali tako lažno pozitivni kot lažno negativni rezultati. Po počlovečenju so bila besedila ocenjena kot manj verjetno umetno ustvarjena, v nekaterih primerih celo manj kot izvirni človeško pisani izvlečki. Ugotovitve poudarjajo potrebo po previdni in kritični uporabi takšnih orodij, zlasti v akademskih postopkih, pri katerih imajo lahko napačne ocene resne negativne posledice.

Language:Slovenian
Keywords:primerjalna analiza, učinkovitost detektorjev, sintetična besedila, umetna inteligenca, MyDetector.ai, ZeroGPT.com, AiDetector.com, ChatGPT
Work type:Master's thesis/paper
Typology:2.09 - Master's Thesis
Organization:FF - Faculty of Arts
Publication status:Published
Place of publishing:Ljubljana
Publisher:N. Shebunchak
Year:2026
Number of pages:97 str.
PID:20.500.12556/RUL-189337 This link opens in a new window
UDC:004.912.05:004.8(043.2)
COBISS.SI-ID:301821443 This link opens in a new window
Publication date in RUL:04.10.2026
Views:157
Downloads:89
Metadata:XML DC-XML DC-RDF
:
Copy citation
Share:Bookmark and Share

Secondary language

Language:English
Title:Comparative Analysis of the Performance of Detectors in the Recognition of Synthetically Generated Texts
Abstract:
The use of artificial intelligence tools to create synthetic texts raises questions about the reliability of their automated detection. The aim of the study was to determine how effectively the selected detectors distinguish between authentic abstracts in the field of librarianship, synthetically generated abstracts, and humanized versions of synthetically generated texts. Three detectors capable of detecting Slovenian-language texts were analyzed: AIDetector.com, ZeroGPT.com, and MyDetector.ai. The analysis included 44 original abstracts of scientific articles written in Slovenian and published in the journal Knjižnica between 2012 and 2017. The titles of these articles were used as the thematic basis for prompts entered into ChatGPT, on the basis of which synthetic abstracts were generated. The third group consisted of humanized texts, additionally transformed in ChatGPT from the synthetic abstracts in order to make them more closely resemble natural human writing and to reduce their synthetic-content detection scores. The results showed that the detectors did not distinguish entirely reliably between human-written and synthetically generated texts. Synthetic abstracts were generally more frequently identified as artificially generated; however, both false-positive and false-negative results occurred. After humanization, the texts were assessed as less likely to have been artificially generated, in some cases even less so than the original human-written abstracts. The findings underscore the need for cautious and critical use of such tools, particularly in academic procedures where incorrect assessments can have serious negative consequences.

Keywords:comparative analysis, detector efficiency, synthetic texts, artificial intelligence, MyDetector.ai, ZeroGPT.com, AiDetector.com, ChatGPT

Similar documents

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

Back