基于LDA模型的移动投诉文本热点话题识别

发布时间：2018-03-03 23:04

本文选题：移动投诉　切入点：k-means　出处：《数据分析与知识发现》2017年02期 　论文类型：期刊论文

【摘要】：【目的】运用中文信息处理和话题识别与追踪的方法,从大量移动投诉文本中找出有价值的信息。【方法】从分析投诉文本的特点入手,使用k-means先对文本聚类。利用LDA对每个类进行建模,提取话题,并从词频、词跨度和词长三方面计算每个话题中词的权值,把权重最大的词作为该话题的标签,并计算每个话题的文档分布概率均值。对具有相同标签的话题,先按照均值最大的原则去掉重复标签话题,再对所有话题计算文档支持率,并将文档支持率作为话题的热度,通过热度区分热点话题和一般话题。【结果】对投诉文本进行时间上的建模,通过对比一般话题和热点话题,得出热点话题的支持文档率至少是一般话题的3倍,支持文档率变化趋势也比一般话题高,说明本文算法是有效的。【局限】没有考虑到话题之间的语义关系。【结论】利用LDA模型对移动投诉话题检测初探的方法是比较合理和有效的,对今后此领域的研究具有一定的借鉴意义。
[Abstract]:[objective] to find out valuable information from a large number of mobile complaint texts by means of Chinese information processing and topic recognition and tracking. [methods] starting with the analysis of the characteristics of complaint texts, Using k-means to cluster the text first. Using LDA to model each class, extract the topic, and calculate the word weight of each topic from three aspects of word frequency, word span and word length, take the words with the largest weight as the label of the topic. The document distribution probability mean of each topic is calculated. For the topic with the same label, the repetitive tagged topic is removed according to the principle of maximum mean value, then the document approval rate is calculated for all topics, and the document approval rating is regarded as the hot spot of the topic. Using heat to distinguish hot topic from general topic. [results] the time model of complaint text is established. By comparing general topic with hot topic, the supporting document rate of hot topic is at least three times as high as that of general topic. The trend of supporting document rate changes is also higher than that of general topics, It shows that this algorithm is effective. [limitation] does not take into account the semantic relationship between topics. [conclusion] the method of using LDA model to detect mobile complaint topics is reasonable and effective. It has certain reference significance to the future research in this field.
【作者单位】：杭州电子科技大学计算机学院;中国计量大学;
【基金】：国家自然科学基金青年基金项目“引入涉身认知机制的汉语隐喻计算模型及其实现”(项目编号:61103101);国家自然科学基金青年基金项目“基于马尔科夫树与DRT的汉语句群自动划分算法研究”(项目编号:61202281) 教育部人文社会科学研究青年基金项目“面向信息处理的汉语隐喻计算研究”(项目编号:10YJCZH052)的研究成果之一
【分类号】：TP391.1

【参考文献】