Research On Internet Of Short-text Classification Based On Convolution Neural Networks

With the rapid development of the Internet and the using of modern social medias such as WeChat,QQ,Baidu Tieba,BBS,blog,Weibo to name a few,the human activities become inseparable from the virtual world or internet.The main form generated by these social media on the Internet is Short-text.The focus of Natural Language Processing research is to obtain the valuable information and control the hottest information from these Short-tests.To do this,the main technology of information acquisition is text classification technology.It plays an important of role in text information processing.In the past few years,Deep-Learning has achieved a good result in image processing and speech recognition.However,it has not been fully adapted in the text information process.To this end,this thesis focuses on using the deep-learning method to classify the Internet of Short-text.(1)According to the characteristics of Chinese Internet of Short-text,this thesis proposes a method of Internet of Short-text classification based on convolution neural networks.This method consists of data preprocessing,feature processing and classification.Firstly,this method the optimization of the word segmentation and denoising,in addition,it constructs the text feature matrix used Word2 vec word vector and TF-IDF value in the data preprocessing module.Secondly,it used different-pool and different types of convolution neural networks to deal with the low-level text’s features in the feature processing module.Lastly,it used the softmax function to text classification operation.The experiment results show that the dynamic convolution neural networks under the maximum pool operation that has the good effect on the Internet of Short-text classification by using the Word2 vec character vector and the feature matrixsuperimposed at the end of the TF-IDF value.(2)This paper compares the above mentioned convolution neural network method with KNN,SVM and DBN text classification methods under the same experiment condition that carry out the two levels classification to Internet of short-text.After the establishment of an effective classification system,it collects the data that satisfying the experimental requirements.Through the Internet of short-text two levels classification experiments,it is concluded that the convolutional neural networks classification method proposed in this thesis is able to effectively classify the Internet short-text classification and achieve a better stability compared to other state-of-art classification methods.