基于情感词典与规则结合的微博情感分析模型研究

发布时间：2018-05-03 04:14

本文选题：微博 + 情感分析　；参考：《安徽大学》2014年硕士论文

【摘要】：二十一世纪以来,中国互联网行业得到了蓬勃的发展,网民规模也逐年攀升。微博是近年来互联网上越来越流行的消遣方式,上到政商名流,下至普通百姓,皆乐在其中,微博已逐渐变成了许多人生活中不可缺少的元素。新浪微博平台每天都产生了数以亿计的微博来分享内容、传播信息,这庞大的用户量和数据量背后伴随而来的则是潜藏的商业、社会等多方面价值。对微博进行情感分析的研究,就是发掘微博潜藏的商业、社会等多方面价值的过程,研究微博情感分析能应用于舆情发现及监控、信息预测、产品评价及改进等领域。深入研究微博内容、获取微博情感倾向是非常有必要的。目前的微博情感极性分类方法存在着准确率较低、依赖领域知识、较少考虑句内句间关系等缺点,我们的研究希望找到一种方法使分类准确率能得到提高,方法的普适性能得到加强。基于此出发点,本文对结合情感词典与规则的微博情感分析方法进行了研究,主要内容包括以下两个部分： (一)本文通过构建情感词典,获取语义规则,以情感词为中心,归纳了6种情感词组合,兼顾情感词、否定词、程度副词之间的相互作用,结合情感词典与规则,运用微博子句情感值、整句情感值计算方法,最终实现了微博情感极性分类。实验表明,本文提出的方法比表情符号判别法、情感词典判别法、SVM判别法等方法的微博情感极性分类效果都好。 (二)本文在(一)的基础上,研究转折连词对微博情感表达的影响,从转折连词的4种一般使用情形,考虑微博的句内关系、句间关系,引入转折连词权重系数来改进(一)的微博子句情感值、整句情感值计算方法,提升微博情感极性分类效果。实验表明,考虑转折连词的方法比之前方法分类效果得到了提升。整体实验对比验证了本文所提出的方法不依赖领域知识,普适性较强,准确率较高。
[Abstract]:Since the 21 century, the Internet industry in China has been booming, and the scale of Internet users has been rising year by year. Weibo is a more and more popular pastime on the Internet in recent years, from the political and commercial celebrities to the ordinary people, they all enjoy it. Weibo has gradually become an indispensable element in the life of many people. Sina Weibo platform produces hundreds of millions of Weibo every day to share content and spread information, this huge number of users and data is accompanied by hidden commercial, social and other values. The research on Weibo's emotion analysis is the process of discovering the commercial and social value hidden by Weibo, and studying the affective analysis of Weibo can be applied in the fields of public opinion discovery and monitoring, information prediction, product evaluation and improvement and so on. It is very necessary to study Weibo's content in depth and to obtain Weibo's emotional tendency. The current classification methods of Weibo's affective polarity have some disadvantages, such as low accuracy, dependence on domain knowledge, less consideration of the relationship between sentences, etc. Our research hopes to find a method to improve the classification accuracy. The universality of the method is enhanced. Based on this starting point, this paper studies Weibo's affective analysis method combined with emotion dictionary and rules. The main content includes the following two parts: (1) by constructing the emotion dictionary, obtaining the semantic rules, taking the emotion word as the center, this paper sums up six kinds of emotion words combination, which takes into account the interaction between emotion words, negative words and degree adverbs, and combines the emotion dictionary with the rules. By using Weibo clause emotion value and the whole sentence emotion value calculation method, we have finally realized Weibo emotion polarity classification. The experimental results show that the proposed method is better than the emoji discriminant method, the emotion dictionary discriminant method and SVM discriminant method in the classification of Weibo's affective polarity. (2) on the basis of (1), this paper studies the influence of turning conjunctions on Weibo's emotional expression. From the four general usage situations of turning conjunctions, we consider the intra-sentence and inter-sentence relations of Weibo. The weight coefficient of turning conjunction is introduced to improve (1) the calculation method of Weibo clause emotion value and the whole sentence emotion value to improve the effect of Weibo affective polarity classification. The experimental results show that the classification effect of the method is better than that of the previous method. The overall experimental results show that the proposed method does not rely on domain knowledge, and is more general and accurate.
【学位授予单位】：安徽大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.092;TP391.1

【参考文献】