摘要
为提高中文文本的分类效果,提出基于深度置信网络的中文文本分类模型,分别以文本的TF-IDF和LSI特征作为输入,利用深度置信网络强大的特征学习能力获取深层次特征,提高最终的分类效果。实验结果表明,LSI特征更适合作为深度置信网络文本分类模型的输入,相比SVM等浅层模型,深度置信网络在中文文本分类任务中更加有效,经过合理的训练和参数设置可以取得比SVM模型更好的分类效果,分类准确率提高了3.4%。
To improve the performance of Chinese text categorization,a Chinese text classification model based on deep belief networks was proposed.The TF-IDF and LSI features of the text were taken as input,and the powerful feature learning ability of deep belief networks was used to acquire deep level features to improve the final classification performance.Experimental results show that LSI feature is more suitable for DBN Chinese text classification model,and compared with SVM and other shallow models,deep belief networks is more effective.With reasonable training and parameter settings,deep belief networks can achieve better classification results than the SVM model and the classification accuracy is improved by 3.4%.
作者
蔡利忠
蔡晓晨
CAI Li-zhong;CAI Xiao-chen(Xilingol Electric Power Bureau Substation Management Office,Xilinhot 026000,China)
出处
《计算机工程与设计》
北大核心
2018年第9期2974-2978,2991,共6页
Computer Engineering and Design