摘要
针对现实文本分类环境下通常仅有少量标记样本而影响分类精度的问题,提出了一种基于概率主题模型潜在Dirichlet分配的分类算法。借助标准词频逆文档频率函数将每个文档表示成术语权重向量;利用概率主题模型预处理以简化文档,并从文档中提取术语;再利用潜在Dirichlet分配模型进行关系学习,构建基于图的分类器完成分类。在公开的Reuters-21578资源库上的分类实验评估了该方法的有效性,相比分类效果较好的支持向量机,该方法在大部分情况下能够取得更高的分类精度。
For the issue that it is only a few labeled samples in really text classification environment which will affect the classification accuracy,this paper proposed a classification algorithm based on latent Dirichlet allocation using probabilistic topic model. Firstly,it used standard term frequency-inverse document frequency function to represent each document into term weight vector. Then,it used probabilistic topic model as pretreatment to simplify the document,and done term extraction from document. Finally,it used latent Dirichlet allocation model to do relational learning and used classification based on graph to finish classification. The effectiveness of proposed method has been verified by experiments on common resource library Reuters-21578. Experimental results show that proposed method has higher classification accuracy than support vector machine which has well classification effect in most cases.
出处
《计算机应用研究》
CSCD
北大核心
2015年第5期1428-1432,1444,共6页
Application Research of Computers
基金
国家自然科学基金资助项目(61202163)
山西省自然科学基金资助项目(2013011017-2)
山西省科技攻关项目(20130313015-1)