论文部分内容阅读
【目的】在LDA模型基础上融合时间和作者特征,提出动态作者主题(DAT)模型,更好地揭示文本内容、主题和作者之间的关系。【应用背景】从海量文本中实现特征抽取和语义挖掘已经成为情报研究人员的重要工作。【方法】获取NIPS会议论文作为数据集并进行预处理,按发表年份划分到每个时间片形成一阶马尔科夫链,使用困惑度确定最优主题数,并在每个时间片内通过吉布斯采样估算作者主题概率分布和主题词项概率分布。【结果】实验结果表明,该模型将文档表示为作者主题概率分布和主题词项概率分布,时间维度上可观测主题强度变化和作者兴趣变化。【结论】DAT模型能够有效地融合文档内容与外部特征,实现文本挖掘。
【Objective】 Based on LDA model, time and author characteristics are fused, and a dynamic author topic (DAT) model is proposed to better reveal the relationship between text content, topic and author. 【Application Background】 Feature extraction and semantic mining from massive texts have become important work for intelligence researchers. 【Method】 Obtaining NIPS conference papers as data sets and preprocessing, forming first-order Markov chain according to the publication year and forming the first-order Markov chain according to the publication year, using the confusion to determine the optimal topic number, Booth sampling estimates the authors’ probability distributions and topic probability distributions. 【Result】 The experimental results show that the model represents the document as the author’s topic probability distribution and the topic item’s probability distribution, the observable topic’s intensity change on the time dimension and the author’s interest change. 【Conclusion】 The DAT model can effectively combine the content of documents and external features to achieve text mining.