孤立点预处理和Single-Pass聚类结合的微博话题检测方法  

Topic detection method of outlier pretreatment combining with Singel-Pass for Chinese microblog

作　　者：叶施仁[1] 杨英[1] 杨长春[1] 朱明峰[1] 

机构地区：[1]常州大学信息科学与工程学院,江苏常州213164

出　　处：《计算机应用研究》2016年第8期2294-2297,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(61272367)

摘　　要：针对如何快速发现微博中的热点话题,提出了一种具有更高准确率的中文微博话题检测方案。首先,优化了微博文本的特征选择,经过分析获得的这些博文特征有助于不同词性对话题表达的重要性;其次,在此基础上,提出了通过计算博文阈值的方法,将零散主题的博文作为噪声过滤,并用来降低博文集的维度;在现有Single-Pass聚类算法的基础上,引入了主题词的概念,根据中心向量的特征权重选择主题词,最终形成一种孤立点预处理与Single-Pass相结合的微博话题检测方法。实验结果表明,相比传统的Single-Pass算法,该方法有效去除了数据集的大部分孤立点,不仅具有较低的漏检率和误检率,而且在时间损耗方面表现更佳。This paper attempted to explore a high accuracy topic delection method for Chinese microblog for quickly detecting hot topics in Chinese microblog. Firstly, It took optimizing the text feature selection into consideration. The different parts of speech of these features obtained from analysis were very important to topic expression. Secondly, on this basis, it proposed a scoring mehod to filter out those topic-unrelated tweets. It used the method to dimension reduction by high-dimensional data. Fi- nally, it introduced the idea of keywords on the basis of Single-Pass clustering algorithm. It selected the keywords according to the feature weight of center vector. Then it used the method to detect topics by combining the outlier pretreatment with Single- Pass. Compared with the typical Single-Pass clustering algorithm, it shows that this method effectively remove the outliers ＂of the dataset in the experiment. What＇ s more,it not only exhibits lower missing rate and lower false detection rate on the test corpus from Sina Weibo, but also performed better in terms of time loss.

关 键 词：微博 热点话题 增量聚类 孤立点 话题检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]