目标函数与策略优化的文本情感分析研究

发布时间：2018-03-05 09:18

本文选题：情感分析　切入点：词向量　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：文本的情感分析又称为观点挖掘,是通过文字针对人对于实体的情绪的分析,主要关注人通过文字所表达的积极或消极的情绪。本课题研究采用统计语言模型,以基于机器学习的方法,以基于词向量的深度学习算法实现文本的特征提取,以分类器进行文本的情感分类,实现文本的自动情感分析。研究的主要工作包括文本特征提取算法的目标函数优化、参数寻优算法的仿生策略优化和文本情感分析的参数寻优,研究的主要创新点如下:(1)针对Doc2Vec算法的目标函数以余弦相似度表征向量差异性的不足,提出一种目标函数优化的文本特征提取算法——T-Doc2Vec算法。T-Doc2Vec算法以扩展的余弦相似度函数——Tonimoto系数作为向量相似度函数,在余弦相似度函数的基础上考虑了向量模的影响,能更细致的反映向量之间的差异程度。并通过IMDB数据集的测试实验验证了算法优化的有效性。(2)针对标准鲸鱼算法在收敛性和全局性方面的不足,提出一种仿生策略优化的混合鲸鱼算法(HBWOA),并通过基准测试函数集的对比实验证明了该算法优化的收敛性能。仿生策略优化的混合鲸鱼算法,通过混沌映射初始化种群和自适应调整搜索策略实现鲸鱼算法的仿生策略优化,结合粒子群算法"认知部分"的优点对鲸鱼算法收敛过程进行改进。(3)结合仿生策略优化的混合鲸鱼算法实现文本情感分析的参数寻优。首先以目标函数优化的T-Doc2Vec算法作为文本情感分析的特征提取算法,然后通过仿生策略优化的混合鲸鱼算法对文本情感分析进行参数寻优,优化文本情感分析的性能表现。
[Abstract]:Text emotional analysis, also known as viewpoint mining, is based on text analysis of people's emotions for entities, focusing on positive or negative emotions expressed by people through text. Based on machine learning, text feature extraction is realized by depth learning algorithm based on word vector, and text emotion classification is carried out by classifier. The main work of the research includes the optimization of the objective function of the text feature extraction algorithm, the bionic strategy optimization of the parameter optimization algorithm and the parameter optimization of the text emotion analysis. The main innovation of the study is as follows: 1) aiming at the deficiency of the objective function of the Doc2Vec algorithm, the cosine similarity is used to characterize the difference of the vector. A text feature extraction algorithm based on objective function optimization, T-Doc2Vec algorithm. T-Doc2Vec algorithm, using extended cosine similarity function, Tonimoto coefficient as vector similarity function, is proposed. The influence of vector modules is considered on the basis of cosine similarity function. Can reflect the degree of difference between vectors more detailedly. And through the test of IMDB data set, the validity of algorithm optimization is verified. 2) aiming at the shortage of convergence and globality of the standard whale algorithm, A hybrid whale algorithm for bionic strategy optimization (HBWOAA) is proposed, and the convergence performance of the algorithm is proved by the comparison of benchmark function sets. Using chaotic mapping to initialize the population and adjust the search strategy adaptively to optimize the bionic strategy of whale algorithm. Combined with the advantages of particle swarm optimization (PSO), the convergence process of whale algorithm is improved. 3) combined with bionic strategy optimization, hybrid whale algorithm is used to optimize the parameters of text emotional analysis. Firstly, T-Doc2Vec, which is optimized by objective function, is used to optimize the parameters of text emotion analysis. Algorithm as a feature extraction algorithm for text emotional analysis, Then the parameters of text emotion analysis are optimized by hybrid whale algorithm, which is optimized by bionic strategy, and the performance of text emotion analysis is optimized.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.1

【参考文献】