摘要
LDA可以实现大量数据集合中潜在主题的挖掘与文本信息的分类,模型假设,如果文档与某主题相关,那么文档中的所有单词都与该主题相关.然而,在面对实际环境中大规模的数据,这会导致主题范围的扩大,不能对主题单词的潜在语义进行准确定位,限制了模型的鲁棒性和有效性.本文针对LDA的这一弊端提出了新的文档主题分类算法gLDA,该模型通过增加主题类别分布参数确定主题的产生范围,提高分类的准确性.Reuters-21578数据集与复旦大学文本语料库中的数据结果证明,相对于传统的主题分类模型,该模型的分类效果得到了一定程度的提高.
Latent Dirichlet Allocation is a classic topic model which can extract latent topic from large data corpus. Model assumes that if a document is relevant to a topic, then all tokens in the document are relevant to that topic. Through narrowing the generate scope that each document generated from, in this paper, we present an improved text classification algorithm for adding topic-category distribution parameter to Latent Dirichlet Allocation. Documents in this model are generated from the category they most relevant. Gibbs sampling is employed to conduct approximate inference. And preliminary experiment is presented at the end of this paper.
出处
《天津理工大学学报》
2014年第4期28-31,共4页
Journal of Tianjin University of Technology
基金
国家自然科学基金(61202169
61170027)