基于链接关系主题模型.pdfVIP

Computer Science and Application 计算机科学与应用, 2017, 7(3), 232-239 Published Online March 2017 in Hans. /journal/csa /10.12677/csa.2017.73029 Relation Topic Model Based on Links Quanmin Wang, Yanfeng Sun, Zhenguo Li, Shi Gu, Kaiyang Wang Faculty of Information Technology, Beijing University of Technology, BJUT, Beijing th th th Received: Mar. 7 , 2017; accepted: Mar. 25 , 2017; published: Mar. 28 , 2017 Abstract LDA model is a complete model of probability topic generation, which can use effective probability algorithm to train and use model. However, LDA model does not consider the effect of link be- tween documents on topic generation in training process. In the RTM model of this paper, the link between the documents is added to the calculation, and the calculation process uses the EM algo- rithm to calculate the potential variables. Because it cannot be accurately calculated, the varia- tional distribution algorithm is used. Finally, we predict the data without training, according to the document prediction link and the link prediction document. We can see that the two RTM models in the dataset are all good. Keywords LDA Model, RTM Model, Links, Variational Distribution 基于链接的关系主题模型 王全民，孙艳峰，李振国，谷 实，王开阳 北京工业大学信息学部计算机学院，北京 收稿日期：2017年3月7 日；录用日期：2017年3月25 日；发布日期：2017年3月28 日 摘 要 LDA模型是一种完全的概率主题生成模型，可以利用有效的概率算法来训练和使用模型，但是LDA模型 在训练过程中并没有考虑文档之间的链接对主题生成的影响，而在本文提出的RTM模型中，就加入对文 档之间链接的计算，计算过程中使用EM算法来对潜在变量进行计算，因为无法准确计算，所以采用变分 分布算法。最后我们对没有训练的数据进行预测，分别为根据文档预测链接和根据链接预测文档，在试 验结果中可以看到在数据集中两个RTM模型的变型都表现良好。 文章引用: 王全民, 孙艳峰, 李振国, 谷实, 王开阳. 基于链接的关系主题模型[J]. 计算机科学与应用, 2017, 7(3): 232-239. /10.12677/csa.2017.73029