基于文本分类的微博情感倾向研究

发布时间：2018-03-28 22:18

本文选题：微博　切入点：情感倾向　出处：《重庆师范大学》2016年硕士论文

【摘要】：近年来,人们越来越依赖于社交软件来表达自己的观点,如使用微博、论坛、贴吧等渠道发表意见,表达情感。而人们对于热点事件和话题的各种各样的情感信息,则通过上述方式直观的反映了出来。因此,通过挖掘和研究大众对各类事件的情感倾向,可以达到对热点舆论的监控和大众看法的感知,为国家、企业等作出决策提供着重要依据。以往的信息检索及采集技术,多数是以关键词为基础,难以支持情感倾向的挖掘,信息提取和文本分类通常没有进行深层次的语义挖掘,因而对文本中所表达出的情感倾向无法进行深层次的挖掘。因此在大数据时代,有效地利用数据挖掘及文本挖掘的相关知识,挖掘微博热点信息以及评论中的情感倾向,将在商品智能推荐、政府舆情监控、文本自动分类等方面有广阔的发展前景。本论文的主要研究内容如下:(1)对情感倾向分析和文本分类的相关概念和技术做了分析,包括微博短文本情感倾向分类以及文本分类的国内外研究现状、文本分类常用的几种算法:支持向量机(SVM),最大熵、决策树、人工神经网络算法等,并重点分析了传统的k最近邻算法,研究算法的基本思路以及该算法在文本分类当中的应用。(2)基于传统的k最近邻算法,并结合一定的关联规则,本文提出了一种对于k最近邻算法的改进:基于主题--情感相关联的k最近邻算法,在文本分类时主题与情感相关联,进行特征值的组合,能够更好的确定k值,提高分类效率。(3)将改进后的基于主题--情感的k最近邻算法在微博热点信息数据集中加以应用,进行情感倾向正向、负向、中性三种分类实验,同时也利用传统的k最近邻算法进行对比实验。在实验前完成微博短文本预处理、特征频繁项集建立等工作。实验结果表明,较传统的k最近邻算法,本文改进的算法在查准率、查准率以及F1值等各项测评指标中都有所提高,降低了算法的时间复杂度,提高了情感分类的效率,有利于更加准确的挖掘微博热点信息中的情感倾向,在处理海量包含主题的中文热门微博时具有一定的可行性。
[Abstract]:In recent years, people have become increasingly dependent on social software to express their views, such as using Weibo, forums, posts and other channels to express their opinions and feelings. And people on hot events and topics of all kinds of emotional information, Thus, by digging and studying the emotional tendencies of the public on various events, we can achieve the monitoring of hot public opinion and the perception of public opinion, for the country, In the past, most of the information retrieval and collection techniques are based on keywords, so it is difficult to support the mining of emotional tendency, and the information extraction and text classification usually do not carry out deep semantic mining. Therefore, in big data's time, the relevant knowledge of data mining and text mining can be effectively used to mine the hot information and the emotional tendency in commentaries. There will be broad prospects in the fields of intelligent recommendation of commodities, monitoring of government public opinion, automatic text classification, etc. The main contents of this paper are as follows: 1) the related concepts and techniques of affective tendency analysis and text categorization are analyzed. Including the current research situation of Weibo short text text classification and text classification at home and abroad, and several commonly used text classification algorithms: support vector machine (SVM), maximum entropy, decision tree, artificial neural network algorithm, etc. The basic idea of the algorithm and its application in text categorization are studied. (2) based on the traditional k-nearest neighbor algorithm and some association rules. In this paper, an improved k-nearest neighbor algorithm is proposed, which is based on subject-affective association algorithm. When text is classified, the topic is associated with emotion, and the combination of feature values can better determine the value of k. To improve the classification efficiency. (3) the improved subject-based k-nearest neighbor algorithm is applied to Weibo hot spot information data set to carry out three classification experiments: positive, negative and neutral emotional tendency. At the same time, the traditional k-nearest neighbor algorithm is used to carry out comparative experiments. Before the experiment, Weibo short text preprocessing and feature frequent itemsets are completed. The experimental results show that the k-nearest neighbor algorithm is more efficient than the traditional k-nearest neighbor algorithm. In this paper, the improved algorithm improves the precision rate, precision rate and F1 value, which reduces the time complexity of the algorithm and improves the efficiency of emotion classification. It is helpful to excavate the emotion tendency of Weibo hot spot information more accurately, and has certain feasibility when dealing with the massive Chinese hot topic Weibo.
【学位授予单位】：重庆师范大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.1

【相似文献】