基于集成方法的情感分析

发布时间：2018-06-30 20:48

本文选题：集成方法 + 情感分析　；参考：《河北地质大学》2016年硕士论文

【摘要】：网络社交媒体为人们提供了快速便捷的交流环境,通过这些社交平台人们可以对社会舆论热点发表自己的观点和看法,情感分析是一种从观点和看法中分析人们所表达情感的技术,已成为当前自然语言处理的热点问题,在情感预测领域具有重大的应用前景。针对单分类算法模型的情感分析泛化能力较弱等问题,提出了集成方法的情感分析,泛化能力较好,在情感分析上有很好的适应性。本文研究的主要工作围绕以下两个方面展开。1.对目前比较流行的开放情感词典进行了分析,构建了情感分析中使用的情感词典,按照情感倾向划分了词汇库。根据微博短文本数据的特点,对数据预处理筛选出研究的语料库,并基于选取的特征,对语料库进行了向量表示。调查了情感分析的现有模型和常用的单分类算法模型分析优点和不足,并对单分类算法模型进行了实验验证。2.提出了一种异质集成方法的情感分析模型,解决了单分类算法模型在假设空间的训练集上进行学习时误选而导致泛化性能不佳等问题。基本思想是,首先,利用构建好的情感词典对微博短文本特征进行表示和选择。然后,根据不同的单分类算法模型特点将模型按照一定的策略进行集成,包括两种方式,一种是将两种或两种以上单分类算法模型的实验结果采用置信度加权平均的方法组合情感分析模型,选择置信度高的子分类器做分类决策;另一种是将分类器的分类结果和置信度作为更高一层集成分类器的特征的方法组合成情感分析模型。最后,设计实现舆情情感分析平台。实验结果表明了提出的集成方法对情感分析是有效和可行的。
[Abstract]:Internet social media provides a rapid and convenient environment for people to communicate. Through these social platforms, people can express their views and opinions to the hot spots of public opinion. Affective analysis is a technique to analyze the emotions expressed by people from the viewpoint and view. It has become a hot issue in natural language processing and has a great application prospect in the field of emotion prediction. Aiming at the problem that the generalization ability of emotion analysis of single classification algorithm model is weak, this paper puts forward the integration method of emotion analysis, which has good generalization ability and good adaptability in emotion analysis. The main work of this paper is focused on the following two aspects. 1. The open emotion dictionary which is popular at present is analyzed, the emotion dictionary used in emotion analysis is constructed, and the lexical database is divided according to emotion tendency. According to the characteristics of Weibo short text data, the corpus of the study is screened out by data preprocessing, and based on the selected features, the corpus is represented by vector. The advantages and disadvantages of the existing emotional analysis model and the commonly used single classification algorithm model are investigated, and the experimental verification of the single classification algorithm model is given. An affective analysis model of heterogeneous ensemble method is proposed to solve the problem of poor generalization performance caused by misselection of single classification algorithm model in training set in hypothesis space. The basic idea is: firstly, using the constructed affective dictionary to express and select the features of Weibo short text. Then, according to the characteristics of different single classification algorithm models, the model is integrated according to a certain strategy, including two ways, One is that the experimental results of two or more single classification algorithms are combined with the affective analysis model using the method of confidence weighted average, and the sub-classifier with high confidence level is selected to make classification decision. The other is to combine the classifier classification results with the confidence degree as the feature of a higher level integrated classifier to form an emotional analysis model. Finally, the design and implementation of public opinion emotional analysis platform. The experimental results show that the proposed integration method is effective and feasible for emotional analysis.
【学位授予单位】：河北地质大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.1

【参考文献】