基于时空主题模型的微博主题提取  

Constructing Spatio-Temporal Topic Model for Microblog Topic Retrieving

作　　者：段炼[1,2,3] 呙维[1] 朱欣焰[4] 胡宝清[2,3] 

机构地区：[1]武汉大学测绘遥感信息工程国家重点实验室,湖北武汉430079 [2]广西师范学院北部湾环境演变与资源利用教育部重点实验室,广西南宁530001 [3]广西师范学院资源环境科学学院,广西南宁530001 [4]武汉大学空天信息安全与可信计算教育部重点实验室,湖北武汉430079

出　　处：《武汉大学学报（信息科学版）》2014年第2期210-213,243,共5页Geomatics and Information Science of Wuhan University

基　　金：国家863计划资助项目(2013AA12A203;2011AA010502);国家科技支撑计划资助项目(2012BAH35B03);广西北部湾重大基础研究专项基金资助项目(2011GXNSFE018003;2012GXNSFEA053001)~~

摘　　要：已有地理主题模型没有考虑不同区域对微博主题影响程度的差异性,同时他们将时间要素离散化,难以得到连续时间上的微博主题强度。提出了一种顾及连续时间及区域影响力因素的时空主题模型。该方法将城市划分为多个区域,依据各兴趣点类型及数量对区域赋予权重以表达区域社会功能对微博主题的影响程度,基于稀疏增量式生成模型表达微博主题分布,利用Beta分布描述主题在连续时间中的强度,最终通过Gibbs采样得到时空主题模型各参数。实验表明,本文方法能发现连续时间上微博主题的演变,与已有地理主题模型相比,能更加准确地提取微博主题。Existing geography topic models do not ence microblog topics. Meanwhile, these models which prevents the acquisition of topic intensities spatio-temporal topic model to discover microblog consider the degree to which different regions infludescribe the topic evolutions in a discrete manner over continuous time. This paper proposes a novel topics by introducing continuous time and region influences. A city was divided into multiple geographic regions. Region weights, expressing the region function influence degree on microblog topics, were allocated to regions based on the number of different POI （Point of Interest） types. Then a sparse additive generative model was applied to generate microblog topic distributions. Beta distributions were employed to depict topic evolution over continuous time. Finally, we use a Gibbs sampling method to estimate model parameters. Experimental results showed that not only does our model track the temporal distribution of microblog topics but also enhances topic extraction accuracy when compared with other geography topic models.

关 键 词：地理主题模型 微博主题挖掘 时空分布 时空推理中图法 

分 类 号：P208[天文地球—地图制图学与地理信息工程]