融合人工特征和自注意力的RCNN模型的机器写作判别方法  

A Machine Writing Discrimination Method Based on RCNN Model with Artificial Features and Self-attention

作　　者：易黎 鲁彦禹 YI Li;LU Yanyu(Wuhan Research Institute of Posts and Telecommunications Co.,Ltd.,Wuhan 430074;Nanjing Fenghuo Tiandi Communication Technology Co.,Ltd.,Nanjing 210019)

机构地区：[1]武汉邮电科学研究院,武汉430074 [2]南京烽火天地通信科技有限公司,南京210019

出　　处：《计算机与数字工程》2025年第2期437-443,共7页Computer & Digital Engineering

摘　　要：随着互联网与深度学习技术的快速发展,利用人工智能技术撰写新闻已经不再是新鲜事,读者甚至很难判别文章究竟是出自于机器还是人类。与此同时,网络上利用人工智能技术编写假新闻与传播无用信息的网络乱象也层出不穷,急需通过有效方法来对机器写作进行甄别。针对以上问题,采用了基于人工特征与自注意力的循环卷积模型(Recur-rent Convolutional model based on Artificial Features and Self-Attention,RCAFSA)的方法来对机器写作内容进行判别,该模型基于改进的循环卷积神经网络(Recurrent Convolutional Neural Networks,RCNN)模型并融入SelfAttention自注意力机制与人工特征,最后通过A-Softmax层输出结果。该模型在进行机器写作判别任务中与传统机器学习算法、卷积神经网络和循环神经网络相比,在精确率、召回率和F1值均有一定的提升。With the rapid development of Internet and deep learning technology,it seems that using AI tech to write news is common.Readers cannot even tell whether the article was written by a machine or a human.Meanwhile,the chaos of using AI tech to write fake news and spread useless information on the Internet has emerged,and there is an urgent need for effective ways to iden⁃tify the authorship of articles.To solve this problem,the recurrent convolutional model based on artificial features and self-attention(RCFASA)is proposed.This model is based on the improved recurrent convolutional neural networks(RCNN)model and it incorpo⁃rates self-attention mechanism and artificial features.In the end,results will be output through the A-Softmax layer.Experimental results show that compared with traditional machine learning algorithm,convolutional neural networks algorithm and recurrent neu⁃ral networks algorithm,this model can improve the value of accuracy,precision,recall rate and F1 score.

关 键 词：机器写作判别 循环卷积神经网络 人工特征融合 自注意力 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]