Combining parametric and nonparametric topic model to discover microblog event
Combining parametric and nonparametric topic model to discover microblog event
复制标题
DOI：
10.1109/infoseee.2014.6946176
复制
发表时间：
2014-04
期刊：
影响因子：
--
通讯作者：
Shengbing Liu;Li Liu;Ruzhong Cheng
中科院分区：
文献类型：
--
作者：
Shengbing Liu;Li Liu;Ruzhong Cheng
与传统新闻媒体相比，微博在传播速度快、话题覆盖面广等方面具有压倒性优势。微博成为事件信息的有效、独特和重要的载体，基于微博的事件发现等文本分析任务具有特殊的意义。然而，常用的内容分析工具，如主题模型，由于微博长度较短，存在严重的数据稀疏问题。遵循前人的思想，比如将个人兴趣帖子和全局事件帖子分开，我们进一步区分了一般主题和事件主题，并采用非参数方法对事件的诞生和死亡进行建模。我们在Twitter数据集上进行了实验，实验结果表明，该方法不仅可以有效地发现事件，而且可以挖掘出更高质量的一般主题。
Compared with traditional news media, microblog holds overwhelming superiority in fast-diffusion and comprehensive coverage of topics. Microblog becomes an effective, particular and important carrier of affair information and many other text analysis tasks, e.g., event discovering based on microblog have special significance. Common tools of content analysis, such as topic model, however, experience severe data sparsity problems due to short length of microblog. Following previous researchers' idea, such as separating personal interest post from global event post, we further differentiate general topics from event topics and adopt nonparametric method to model the birth and death of event. We conduct experiments on Twitter data set, and the experimental results demonstrate that our method can not only discover event effectively, but also mine higher quality general topics.