FreqMark: Frequency-Based Watermark for Sentence-Level Detection of LLM-Generated Text
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866910649422970880 |
|---|---|
| author | Xu, Zhenyu Zhang, Kun Sheng, Victor S. |
| author_facet | Xu, Zhenyu Zhang, Kun Sheng, Victor S. |
| contents | The increasing use of Large Language Models (LLMs) for generating highly coherent and contextually relevant text introduces new risks, including misuse for unethical purposes such as disinformation or academic dishonesty. To address these challenges, we propose FreqMark, a novel watermarking technique that embeds detectable frequency-based watermarks in LLM-generated text during the token sampling process. The method leverages periodic signals to guide token selection, creating a watermark that can be detected with Short-Time Fourier Transform (STFT) analysis. This approach enables accurate identification of LLM-generated content, even in mixed-text scenarios with both human-authored and LLM-generated segments. Our experiments demonstrate the robustness and precision of FreqMark, showing strong detection capabilities against various attack scenarios such as paraphrasing and token substitution. Results show that FreqMark achieves an AUC improvement of up to 0.98, significantly outperforming existing detection methods. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2410_10876 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | FreqMark: Frequency-Based Watermark for Sentence-Level Detection of LLM-Generated Text Xu, Zhenyu Zhang, Kun Sheng, Victor S. Computation and Language Cryptography and Security Machine Learning The increasing use of Large Language Models (LLMs) for generating highly coherent and contextually relevant text introduces new risks, including misuse for unethical purposes such as disinformation or academic dishonesty. To address these challenges, we propose FreqMark, a novel watermarking technique that embeds detectable frequency-based watermarks in LLM-generated text during the token sampling process. The method leverages periodic signals to guide token selection, creating a watermark that can be detected with Short-Time Fourier Transform (STFT) analysis. This approach enables accurate identification of LLM-generated content, even in mixed-text scenarios with both human-authored and LLM-generated segments. Our experiments demonstrate the robustness and precision of FreqMark, showing strong detection capabilities against various attack scenarios such as paraphrasing and token substitution. Results show that FreqMark achieves an AUC improvement of up to 0.98, significantly outperforming existing detection methods. |
| title | FreqMark: Frequency-Based Watermark for Sentence-Level Detection of LLM-Generated Text |
| topic | Computation and Language Cryptography and Security Machine Learning |
| url | https://arxiv.org/abs/2410.10876 |