基于循环神经网络的网络舆情文本情感分析技术研究

发布时间：2018-09-12 06:40

【摘要】：文本情感分析作为自然语言处理领域的研究热点之一,近年来受到越来越多的关注,其目的在于自动地从文本中提取和归纳主观情感信息。现有的文本情感分析方法主要可以分为基于情感词典的方法和基于机器学习的方法。基于情感词典的方法严重依赖于情感词典的质量,而构建一个覆盖范围广且精确的情感词典是非常困难的。基于机器学习的文本情感分析方法依赖于构建和抽取的特征,但传统的特征表示方法都不能很好的保留文本的语义信息。随着深度学习技术在计算机视觉领域的广泛应用,深度学习模型被证明在特征提取方面有着很大优势,其中循环神经网络由于其特殊的循环链式结构,非常适合处理文本等序列数据,已被广泛应用到自然语言处理领域。本文主要研究基于循环神经网络的文本情感分析方法。本文的工作内容可以分为如下两块:1)循环神经网络在传统的情感分析任务上的应用。考虑到文本的情感极性很大程度上取决于句子中带有情感倾向性的词语,对文本中的情感词加以关注将有助于提高情感分类的效果。针对目前情感分析中的循环神经网络模型缺乏对情感词的关注,本文提出了一种基于循环神经网络的情感分析注意力模型(RNN-Attention),通过引入注意力机制,在情感分类时着重考虑文本中的情感词的影响。在NLPCC 2014情感分析数据集以及IMDB影评数据集上进行的实验结果表明,该模型能够提高情感分析的效果。2)针对目标依赖情感分析任务的循环神经网络模型。目标依赖情感分析针对给定的一个目标对象,判断句子关于该对象的情感极性。目前,大多数的情感分析方法都是针对传统的情感分析任务,即对给定的一段文本直接分析它所表达的情感倾向。使用传统情感分析的方法进行目标依赖的情感分析时,由于没有考虑目标对象的信息,会导致作出错误的判断。为了解决这个问题,本文在RNN-Attention模型的基础上进行改进,提出RNN-Attention-T模型,该模型在对文本进行建模的同时引入目标对象的信息。此外,考虑到目标对象的上文和下文对句子情感倾向的影响程度往往是不同的,本文接着提出了一种对目标对象的上文和下文分别建模的RNN-Attention-C模型。实验结果表明,与现有的目标依赖情感分析方法相比,本文所提出的改进模型能够在不使用句法解析器和外部情感词典的情况下获得较好的分类效果。
[Abstract]:As one of the research hotspots in the field of natural language processing, text emotional analysis has attracted more and more attention in recent years. Its purpose is to automatically extract and induce subjective emotional information from text. The existing text emotion analysis methods can be divided into affective dictionary based method and machine learning method. The method based on emotion dictionary depends heavily on the quality of emotion dictionary, and it is very difficult to construct an emotion dictionary with wide coverage and accuracy. The text emotion analysis method based on machine learning depends on the feature of construction and extraction, but the traditional feature representation method can not preserve the semantic information of the text. With the wide application of depth learning technology in computer vision field, depth learning model has been proved to have great advantages in feature extraction. It is very suitable for processing sequential data such as text, and has been widely used in the field of natural language processing. This paper mainly studies the text emotion analysis method based on cyclic neural network. The work of this paper can be divided into two parts: 1) the application of cyclic neural network in traditional affective analysis task. Considering that the emotional polarity of the text depends to a great extent on the words with emotional orientation in the sentence, paying attention to the emotional words in the text will help to improve the effect of emotion classification. In view of the lack of attention to affective words in the current cyclic neural network model, this paper proposes an affective analysis attention model (RNN-Attention) based on cyclic neural network, which introduces attention mechanism. In affective classification, the influence of affective words in the text is taken into account. The experimental results on the NLPCC 2014 emotional analysis data set and the IMDB image review dataset show that the model can improve the effect of emotional analysis. 2) the circulatory neural network model for the goal dependent affective analysis task. Goal-dependent affective analysis determines the emotional polarity of a sentence about a given object. At present, most of the affective analysis methods are aimed at the traditional task of affective analysis, that is, to analyze the emotional tendency of a given text directly. When the traditional affective analysis method is used to analyze the goal-dependent emotion, the wrong judgment will be made because the information of the target object is not considered. In order to solve this problem, this paper improves on the RNN-Attention model and proposes a RNN-Attention-T model, which not only models the text but also introduces the information of the target object. In addition, considering that the influence of the above and the following on the affective tendency of the sentence is often different, this paper then proposes a RNN-Attention-C model of the target object, which is modeled separately above and below. The experimental results show that the proposed improved model can achieve better classification results without using syntactic parsers and external affective dictionaries compared with existing object-dependent affective analysis methods.
【学位授予单位】：华侨大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.1;TP183

【相似文献】