基于优化BP神经网络的微博舆情预测模型研究

发布时间：2018-09-14 19:22

【摘要】：随着互联网技术的飞速发展,微博已经成为人们生活中重要的一部分。由微博引发的舆情也越来越受到各界的关注。由于微博信息的传播速度快,传播范围广以及微博消息发布的任意性,使微博上的信息有真有假,有虚有实。正面的微博舆情和负面的微博舆情会给人们的生活带来截然不同的影响,有些负面的微博舆情甚至会构成危机,严重影响社会公共安全。因此,对微博舆情预测的研究具有现实意义。进行微博舆情的预测,首先要获取能够表示微博舆情的数据。本文采用离散的时间序列描述微博舆情的趋势。本文以新浪微博平台为背景,对微博文本中的热点话题进行提取、分析并对微博舆情进行预测。时间序列获取的步骤：一是用新浪微博的API接口,获取一段时间内的微博文本；二是根据微博文本的特点进行相应的预处理后,使用微博话题统计的方法,发现微博热点话题；三是统计一段时间内微博热点话题的回复数和转发数,组成舆情预测模型实验中的时间序列数据。 BP神经网络能够较好地拟合微博舆情时间序列的非线性变化的函数关系,能用其预测微博舆情,但是也存在着一定的局限性：BP神经网络的学习算法对已经学习的样本有遗忘。在样本中有噪声时,可能会使BP神经网络的性能变差；BP神经网络还存在收敛速度慢、容易陷入局部极小值的缺点。我们做了两个工作：一是对BP神经网络的网络结构进行改进。在BP神经网络的输入层后面添加一个神经元层—输入承接层,当样本中出现噪声时,能够延迟网络参数调整,从而提高BP神经网络的性能。二是用遗传模拟退火算法(GSA)对BP神经网络的网络参数进行优化。GSA具有收敛速度快,能够较好的避免出现局部极小值的问题,从而弥补BP神经网络的收敛速度慢,容易陷入局部极小值的不足。本文对从已有的微博中获取的舆情时间序列,分别用四种舆情预测模型进行舆情预测对比实验。实验结果表明,用GSA优化的改进BP神经网络的舆情预测模型能够取得较好的预测效果。
[Abstract]:With the rapid development of Internet technology, Weibo has become an important part of people's lives. The public opinion caused by Weibo also receives more and more attention from all walks of life. Due to the high speed of Weibo's information, the wide range of dissemination and the arbitrariness of Weibo's news release, the information on Weibo is true and false. The positive public opinion of Weibo and the negative public opinion of Weibo will bring different influence to people's life, and some negative public opinion will even form a crisis, which will seriously affect the social public safety. Therefore, the study of Weibo public opinion prediction has practical significance. Weibo's public opinion prediction, first of all, to obtain the data that can express Weibo public opinion. This paper uses discrete time series to describe the trend of Weibo's public opinion. Based on the platform of Sina Weibo, this paper extracts the hot topics in Weibo's text, analyzes and predicts the public opinion of Weibo. The steps of time series acquisition are as follows: first, using the API interface of Sina Weibo to obtain the Weibo text for a period of time; Third, count the number of replies and retweets of Weibo's hot topics for a period of time. The BP neural network can fit the nonlinear function relation of Weibo's public opinion time series, and can be used to predict the public opinion of Weibo, which is composed of the time series data of public opinion prediction model, the BP neural network can fit the function relation of the nonlinear change of Weibo's public opinion time series, However, there are some limitations in the learning algorithm of the BP neural network. When there is noise in the sample, it may make the performance of BP neural network worse and the BP neural network have the disadvantage of slow convergence rate and easy to fall into local minimum. We have done two works: one is to improve the network structure of BP neural network. After the input layer of BP neural network, we add a neuron layer-input accept layer, which can delay the adjustment of network parameters when there is noise in the sample, so as to improve the performance of BP neural network. Secondly, the genetic simulated annealing algorithm (GSA) is used to optimize the network parameters of BP neural network. The convergence speed of BP neural network is fast, and the problem of local minimum value can be avoided, which can make up for the slow convergence speed of BP neural network. It is easy to fall into the deficiency of local minima. In this paper, the time series of public opinion obtained from Weibo are compared with four models of public opinion prediction. The experimental results show that the improved BP neural network model of public opinion prediction with GSA optimization can achieve better prediction results.
【学位授予单位】：华中师范大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP183;TP393.092

【参考文献】