用户生成内容情感分析方法研究

发布时间：2018-05-09 04:13

本文选题：用户生成内容 + 情感分析　；参考：《安徽财经大学》2016年硕士论文

【摘要】：用户生成内容情感分析一般是指通过基于词典语义规则或者基于机器学习的方法对Web用户主动生成的内容进行挖掘和分析,并且从中获得表达用户态度和观点的一些主观信息,比如用户针对于某个事件的观点、建议、情绪、情感倾向等。近两年,随着“互联网+”行动不断助力电子商务企业的发展,针对电子商务企业评价信息情感分析的研究得到了研究者们越来越多的重视。本研究主要是以淘宝某商品文本评价信息为研究对象,通过对某商品评价信息的分析,使得消费者和商家从不同的角度了解商品的真实信息,为后来消费者购买该商品提供决策依据,也可以使得商家不断地提高商品品质,改进商品服务以及制定合适的销售方案。论文主要通过PMI和《知网》相似度计算的方法给情感词语赋予相应的权值,采用基于语义规则,多分类机器学习以及模糊综合的情感分析方法从某商品每个评价对象的分值,好评率,商品的综合值等方面对商品评价信息进行多角度的分析,并验证了三种方法的有效性。具体的主要工作如下:1基于带权值的情感词典的评论信息情感分析。首先在基础情感词典的基础上,结合《知网》相似性计算和统计的PMI方法给词语赋予一定的权值,构建带权值的领域情感词典,除此之外,构建否定词,程度副词,转折词语情感词典并赋予不同的权值。然后,获取评价信息四元组,利用相应的情感极性计算规则获得评论句子的情感极值,并根据情感极值对评论句子进行情感分类。最后,验证了带权值领域情感词典和情感计算规则用于情感分析的有效性。2基于多分类的机器学习方法的评论信息情感分析。论文对文本表示,特征选择和特征权重的计算方法进行了介绍,并针对短文本评价信息的特性改进了布尔函数特征权值的计算方法,用词语权值代替“1”,“0”进行文本表示,并且引入参数α来缩小权值相近的评价信息的距离,提高分类的准确率。并用χ统计方法进行特征选择,通过KNN,SVM和朴素贝叶斯方法三种分类模型验证了改进的布尔函数特征权重计算方法的有效性。3基于模糊综合评价方法的评论信息情感分析。论文针对模糊综合评价的需要,构建了指标体系,确定了评价对象应该归属的指标,引入了情感分析的方法。然后利用模糊统计的方法确定了模糊综合评价矩阵和指标的相应权重,最后获得商品的综合评价值。并利用模糊综合评价方法对某商品的评价信息进行了评价。
[Abstract]:Generally speaking, user-generated content emotional analysis refers to mining and analyzing the content generated by Web users based on lexicon semantic rules or machine learning, and obtaining some subjective information to express users' attitudes and opinions. Such as the user's point of view on an event, advice, emotions, emotional tendencies and so on. In the past two years, with the development of e-commerce enterprises, more and more researchers have paid more and more attention to the research of evaluation information emotion analysis of e-commerce enterprises. This research mainly takes Taobao commodity text evaluation information as the research object, through the analysis to the certain commodity appraisal information, causes the consumer and the merchant to understand the commodity real information from the different angle, To provide decision basis for later consumers to buy the goods, it can also make merchants continuously improve the quality of goods, improve goods and services, and formulate appropriate sales plans. In this paper, the corresponding weights are given to the affective words by the methods of PMI and similarity calculation. The emotion analysis method based on semantic rules, multi-classification machine learning and fuzzy synthesis is used to evaluate the value of each object of a commodity. The evaluation information is analyzed from different angles in the aspects of high praise rate and comprehensive value of commodity, and the validity of the three methods is verified. The main work is as follows: 1. Emotion Analysis of comment Information based on weighted emotion Dictionary. First of all, on the basis of the basic emotion dictionary, combining the PMI method of similarity calculation and statistics, this paper gives a certain weight to the words and constructs a domain emotion dictionary with weighted value. In addition, the negative words and degree adverbs are constructed. A lexicon of words and expressions with different weights. Then, the quaternion of evaluation information is obtained, and the emotional extremum of the comment sentence is obtained by using the corresponding rules of emotional polarity calculation, and the emotion classification of the comment sentence is carried out according to the emotional extremum. Finally, the validity of weighted domain emotion dictionary and emotion computing rules for emotion analysis is verified. This paper introduces the methods of text representation, feature selection and calculation of feature weights, and improves the calculation method of Boolean function feature weights according to the characteristics of the evaluation information in this paper, and uses the word weight instead of "1". "0" is used to represent the text, and the parameter 伪 is introduced to reduce the distance of the evaluation information with similar weights and to improve the accuracy of classification. The feature selection is carried out by 蠂 -statistic method. The validity of the improved Boolean function feature weight calculation method based on the fuzzy comprehensive evaluation method is verified by three classification models, KNN SVM and naive Bayes method. 3. The emotion analysis of comment information is based on fuzzy comprehensive evaluation method. In order to meet the needs of fuzzy comprehensive evaluation, the paper constructs an index system, determines the index that the evaluation object should belong to, and introduces the method of emotion analysis. Then the corresponding weights of fuzzy comprehensive evaluation matrix and index are determined by the method of fuzzy statistics and finally the comprehensive evaluation value of commodities is obtained. The fuzzy comprehensive evaluation method is used to evaluate the evaluation information of a commodity.
【学位授予单位】：安徽财经大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：F724.6;F274

【相似文献】