Основные понятия
AI-Catcher is a novel method to detect ChatGPT-generated scientific text, integrating MLP and CNN models to improve accuracy significantly.
Аннотация
The research introduces AI-Catcher, a model to distinguish between human-written and ChatGPT-generated scientific text. A new dataset, AIGTxt, enhances AI-generated text detection tools. Experiments show AI-Catcher outperforms baseline methods and commercial tools in accuracy, precision, recall, and F1-score.
Several plagiarism detection methods are reviewed along with machine-generated text detection approaches. The study emphasizes the importance of maintaining integrity in academic literature amidst AI advancements.
Key features of AI-Catcher include linguistic/statistical feature extraction by MLP, sequential pattern extraction by CNN, and fusion of hidden patterns. Results demonstrate the effectiveness of AI-Catcher in detecting ChatGPT-generated content accurately.
Статистика
On average, AI-Catcher improved accuracy by 37.4%.
AIGTxt contains 3000 records across ten domains.
The average paragraph length in AIGTxt is approximately 194 words.
The dataset encompasses a rich vocabulary of 29,226 unique words.
Accuracy results achieved by AI-Catcher range from 2.65% to 89.19% improvement over baseline methods.
Цитаты
"AI-powered tools have the potential to generate scientific content indistinguishable from human-written work."
"AI-Catcher significantly improves accuracy in distinguishing between human-written and ChatGPT-generated texts."