基于神经网络的微博情感分析

Sentiment Analysis of Micro-blog Based on Neural Networks

微博情感分析的目的是发现用户对热点事件的态度及观点,目前已有的相关研究大部分是使用SVM根据手工标注的微博情感特征对微博进行情感分析,然而由于微博文本通常含有有限的上下文信息,因而对其进行情感分析是具有挑战性的。为了能有效地解决这一任务,文中提出基于卷积的神经网络结构模型,被命名为汉字到句子卷积神经网络。该网络使用两个卷积层从任何规模的字和句子中抽取相关特征,从字到句子级层面卷积神经网络信息来完善短文本的信息进行微博情感分析。通过实验来证明卷积神经网络对于微博情感分析的有效性,并与基于层次结构的多策略方法和基于词典与机器学习的方法进行对比,结果表明文中提出的方法对于微博短文本情感分析更有效。

The purpose of the micro-blog sentiment analysis is to identify the users＇ attitudes and opinions of the hot events. At present, most of the existing researches did the micro-blog sentiment analysis by using SVM, which is based on the manual annotation of micro- blog sentimental features. However, micro-blog always contains the limited contextual information that leads to the challenge of sentiment analysis of micro-blog messages. In order to deal with it efficiently,propose the sentiment analysis of micro-blog based on convolutional neural networks method. It was named as the sentence level to word level convolution neural network, which uses two layers to extract rel- evant features from the convolution of any size words and sentences. It performs sentiment analysis of micro-blog from word to sentence level information and uses convolution to improve the short text messages of the neural network information. Conduct experiments to prove the effectiveness of the convolution neural network for micro-blog sentiment analysis,and take the experiments contrasts with the other two methods, which are hierarchical structure based hybrid approach and lexicon and machine learning based methods on the senti- ment analysis of micro-biog. The experiments results show that this method has better performance in sentiment analysis of micro-biog.