A Comprehensive Approach and Application for Assessing the Richness of AI-Generated Articles

Abstract

This paper presents a comprehensive method for evaluating the richness of articles generated by Artificial Intelligence. This unique method carries out a multi-dimensional evaluation, comprising four aspects: Vocabulary, Syntax, Semantics, and Text information(VSST), and further employs techniques such as Gradient Boosting Trees in an attempt to mimic human scoring. For empirical validation of the proposed method, we used several large language models to construct a Chinese dataset for assessing the richness of texts, and conducted a series of experiments, including score fitting experiments, sample size adjustment experiments and ablation experiments. Finally, we showcased the applicability of the VSST method across diverse article genres, and confirmed the utility and effectiveness of the method by measures including Pearson correlation coefficient, mean squared error (MSE) and average bias.

Ähnliche Arbeiten