CNN算法改进及在舆情分析中的应用

Improvement of CNN Algorithm and Its Application in Public Opinion Analysis

作为深度学习图像识别的开创性复杂算法,卷积神经网络(CNN)在图像处理中有着其他机器学习算法所不具备的高精度的优点,同时小波神经网络(WNN)在训练中有着跳出局部极小值的特点,因此可达到的最小误差精度是大部分网络难以达到的。结合CNN与WNN各自的优点,本文提出了CNN与WNN相结合的两种网络:小波卷积小波神经网络(wCwNN)和小波卷积神经网络(wCNN)。基于wCwNN网络以及wCNN网络对文本分析问题进行探索,尝试用两种网络处理经由词向量模型(word2vec)处理后的文本信息,发现相比于传统的卷积神经网络,针对经word2vec处理后的文本,改进后的网络仍然具有一定的优势。本文最后针对经典的神经网络对处理文本类数据问题提出研究方向,并对神经网络未来发展提出想法。

As deep learning seminal complex algorithm of image recognition,convolution neural network (CNN) with other machine learning algorithms in image processing does not have the advantages of high precision,at the same time,the wavelet neural network (WNN) in training has the characteristics of the local minimum value,therefore can achieve the minimum error of accuracy is hard to achieve most of the network. Combining the advantages of CNN and WNN,this paper proposes two kinds of networks:wavelet convolution wavelet neural network (wCwNN) and wavelet convolution neural network (wCNN). Based on wCwNN network and wCNN network,this paper explores the problem of text analysis. Two kinds of networks are used to process text information processed by word vector model (word2vec). It is found that compared with traditional convolutional neural network,the improved network still has certain advantages for text processed by word2vec. In the end,this paper puts forward the research direction of classical neural network in dealing with text data,and puts forward some ideas for the future development of neural network.