MGTBench
收藏资源简介:
MGTBench是由CISPA亥姆霍兹信息安全中心创建的一个用于检测机器生成文本(MGT)的基准框架。该数据集包含13种不同的检测方法,旨在评估和比较各种方法在检测由强大语言模型(如ChatGPT)生成的文本方面的效果。MGTBench通过广泛的评估,展示了不同检测方法在公共数据集上的表现,并揭示了它们在面对不同语言模型和数据集时的性能和鲁棒性。该数据集的应用领域包括自然语言处理、信息安全和人工智能伦理,旨在解决机器生成文本的识别和归属问题,以防止虚假信息和提高文本内容的透明度。
MGTBench is a benchmark framework for detecting machine-generated text (MGT) developed by CISPA Helmholtz Center for Information Security. This dataset includes 13 distinct detection methods, designed to evaluate and compare the effectiveness of various approaches in detecting text generated by advanced language models such as ChatGPT. Through comprehensive evaluations, MGTBench showcases the performance of different detection methods on public datasets, and reveals their performance and robustness across diverse language models and datasets. Its application domains cover natural language processing, information security, and AI ethics, with the goal of addressing the identification and attribution of machine-generated text to combat disinformation and enhance the transparency of textual content.

- 1MGTBench: Benchmarking Machine-Generated Text DetectionCISPA亥姆霍兹信息安全中心 · 2024年



