SEFD: Semantic-Enhanced Framework for Detecting LLM-Generated Text

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Computational Linguistics

Background:

• Large language models (LLMs) are increasingly prevalent, necessitating robust detection methods.
• Paraphrasing techniques often evade current LLM-generated text detectors.
• Existing detection tools struggle with nuanced AI-generated content.

Purpose of the Study:

• To introduce a novel semantic-enhanced framework for detecting LLM-generated text (SEFD).
• To improve the accuracy and robustness of AI-generated text detection, particularly for paraphrased content.
• To address the challenge of identifying sophisticated AI-generated text in real-world applications.

Main Methods:

• Developed a semantic-enhanced framework (SEFD) integrating retrieval-based mechanisms with traditional detectors.
• Employed a curated retrieval strategy balancing comprehensive coverage and computational efficiency.
• Evaluated the framework's performance on sequential text scenarios and various LLM-generated content.

Main Results:

• The SEFD framework significantly enhances detection accuracy for paraphrased LLM-generated text.
• The framework demonstrates robustness in identifying standard LLM-generated content.
• Experiments confirmed improved detection capabilities in sequential text applications like forums and Q&A platforms.

Conclusions:

• The SEFD framework offers a substantial advancement in detecting sophisticated LLM-generated text.
• This approach contributes to safeguarding information integrity in the era of prevalent AI content.
• The semantic-enhanced, retrieval-based method proves effective against paraphrasing evasion tactics.