基于态势感知模型的文本情感分析研究

发布时间：2018-08-04 17:50

【摘要】：随着互联网的快速发展,人工智能逐渐获得了人们更广泛的关注,其研究也在人们的不断努力下得到突破,进入了新的发展阶段。在此背景下,与其息息相关的情感分析相关研究工作也纷纷展开。本文的核心是:①通过分析现有文本情感分析方法,对比多种传统机器学习模型的情感分类效果;②引入集成学习方法,提出“多特征多分类器的元集成情感分析模型(MFMB-ME,Multi-Features Multi-Base-Classifiers Meta Ensemble Learning Sentiment Analysis Model)”,通过选用不同特征集与基分类器组合进行集成学习模型训练,分析其情感分类效果,进行情感分析实验并得出结论:通过使用MFMB-ME相比基于单特征集的集成分类模型,分类正确率能获得明显提升;③结合文本情感分析和态势感知理论研究,总结出“文本情感分析的态势感知模型(SA-SA,Sentiment Analysis Based On Situational Awareness Model)”。本文首先介绍了人工智能的发展并引出文本情感分析的意义与价值,通过对国内外研究现状的分析了解当前文本情感分析的发展情况以及存在的问题。详细介绍了文本情感分析相关内容及其现存流行技术。结合态势感知方法论分析“文本情感分析的态势感知模型”。其次,采用三次实验分别对比了基于传统机器学习算法,单特征集成学习算法,“多特征集多基分类器的元学习集成学习”算法的文本情感分类效果。基于传统机器学习方法实验中,分别采用决策树,支持向量机,逻辑回归算法进行分类建模,对比分析不同模型的情感分类结果;基于单特征集成学习算法实验中,采用随机森林对词特征集进行模型训练,对比分析其与传统机器学习算法的分类效果差异;基于多特征集多分类器的元学习集成学习实验中,组合不同的文本特征集(包括词,词干,词性,语法,n-gram等)与不同的基分类器(包括逻辑回归,语言模型等),通过以随机森林为元学习器的集成学习方法,对比分析不同组合策略的分类效果。最后综合实验结果可分析:对于实验语料,与单特征集成学习分类模型和传统机器学习分类模型相比较,本文提出的FMB-ME模型对测试集的分类正确率更高,具有更优的分类效果,且分类性能提升较明显。
[Abstract]:With the rapid development of the Internet, artificial intelligence has gradually gained more and more attention, and its research has also been broken through with the continuous efforts of people, and entered a new stage of development. Under this background, the related research work of affective analysis which is closely related to it is also carried out one after another. The core of this paper is the introduction of integrated learning method by analyzing existing text affective analysis methods and comparing the affective classification effects of many traditional machine learning models. This paper presents a multi-feature and multi-classifier Multi-Base-Classifiers Meta Ensemble Learning Sentiment Analysis Model) (MFMB-MEM Multi-Features Multi-Base-Classifiers Meta Ensemble Learning Sentiment Analysis Model). By using different feature sets and base classifiers to train the integrated learning model, the effect of emotion classification is analyzed. Through the experiment of emotion analysis, it is concluded that compared with the ensemble classification model based on single feature set with MFMB-ME, the classification accuracy can be significantly improved by combining text emotion analysis and situational awareness theory research. The SA-SAN sentiment Analysis Based On Situational Awareness Model) is summarized. This paper first introduces the development of artificial intelligence and elicits the significance and value of text emotional analysis, through the analysis of the current research situation at home and abroad to understand the current development of text emotional analysis and the existing problems. This paper introduces the related contents of text emotion analysis and the existing popular technology in detail. The situational perception model of text affective analysis combined with situational perception methodology. Secondly, three experiments are used to compare the effect of text emotion classification based on traditional machine learning algorithm, single feature ensemble learning algorithm and meta-learning ensemble learning algorithm based on multi-basis classifier. In the experiment based on traditional machine learning method, decision tree, support vector machine and logical regression algorithm are used to classify and model, and the results of emotion classification of different models are compared and analyzed. The model training of the word feature set is carried out by using random forest, and the classification effect is compared with the traditional machine learning algorithm. In the meta-learning ensemble learning experiment based on multi-classifier, different text feature sets (including words) are combined. Stem, part of speech, grammatical n-gram, etc.) and different base classifiers (including logical regression, language model, etc.). The classification effects of different combination strategies are compared and analyzed by using the integrated learning method using random forest as meta-learning device. Finally, the experimental results can be analyzed: for the experimental corpus, compared with the single feature integrated learning classification model and the traditional machine learning classification model, the FMB-ME model proposed in this paper has higher classification accuracy and better classification effect on the test set. And the classification performance is improved obviously.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.1

【参考文献】