Search papers, labs, and topics across Lattice.
This paper investigates the efficacy of LLMs as detectors of AI-generated modern Chinese poetry, an area previously unexplored. They introduce an image-semantic guided detection method that incorporates images reflecting the poetry's content to complement the textual analysis. Experiments show that LLM detectors using this image-semantic approach outperform text-only baselines, with a Gemini-based detector achieving a state-of-the-art Macro-F1 score of 85.65%.
Injecting image-derived semantics into LLMs dramatically improves their ability to detect AI-generated Chinese poetry, surpassing even specialized text-based detectors.
Previous detection studies have shown that LLMs cannot be effectively used as detectors, but these studies have not addressed modern Chinese poetry. Moreover, no relevant research has explored the performance of LLMs in detecting modern Chinese poetry. This paper evaluates and enhances the performance of LLMs as detectors for modern Chinese poetry, and proposes an image-semantic guided poetry detection method. Compared with traditional detection approaches, our method innovatively incorporates images that reflect the content of the poetry. Through example-driven approaches, our method effectively integrates information such as meaning, imagery, and feeling from the image, then forms a complementary judgment with the poem text. Experimental results demonstrate that the LLM detectors based on our method outperform baseline detectors based on plain text, and even surpass the best-performing traditional detector, RoBERTa. The Gemini detector using our method achieves a Macro-F1 score of 85.65%, reaching the state-of-the-art level. The performance improvements of different LLM detectors on multiple LLMs-generated data prove the effectiveness of our method.
