Study on Decision Tree-Based Recognition of AI-Generated Texts

Abstract

This paper investigates AI writing. By inferring the basic rules of text generation, 12 indicators are constructed, and a decision tree model is established to determine whether the text is written by AI. Through randomly extracting paragraphs from articles and rewriting them with AI, the analysis reveals that AI-generated texts have simpler structures and convey more singular expressions, resulting in the identification of 12 distinct features. For AI-generated articles, a decision tree model is trained and tested using constructed data and indicators based on basic rules, facilitating the assessment of whether an article is AI-generated. In determining the origin of articles, a classification is made between Chinese and English articles, and a trained decision tree model is employed for classification. The results indicate that 8 Chinese articles are human-authored, while 2 English articles are AI-generated. Finally, regarding the issue of plagiarism in mathematical models, images, and formulas, data from both human and AI-generated sources are collected. Relevant indicators are extracted, and a decision tree model is established to classify examples. The research results are expected to deepen our understanding of AI text generation, holding significant implications for the study of AI writing.

Ähnliche Arbeiten