Research On Short Text Sentiment Analysis Based On Deep Learning

With the popularization and rapid development of the Internet and e-commerce,Internet users can easily express their opinions and feelings on hot events through social networks,express their experiences and opinions on the purchased products on the e-commerce website.Meanwhile,these contents contain many texts in the form of short texts.Analyzing and excavating information from these short texts have great commercial value and research significance in grasping the social public opinion and the potential improvement of products.In this paper,deep neural network method is used to study the sentiment analysis of short texts from the internet.Combined with two typical deep neural network structures,RNN and CNN,and the recent popular GAN were researched on represent short texts with word embedding,machine translation of short texts in bilingual environment and sentiment analysis.The main contributions are as follow:First of all,preprocessing the experimental dataset for sentiment analysis,including the stop words and word embedding training.After that,the texts information are represented by a vector matrix through training word embedding,while the vector matrix would be used as the input of the neural network model.Secondly,a Bidirectional Round LSTM neural network attribute-level sentiment analysis method is proposed in this paper.This method takes the property target words as the center and constructs a Bidirectional Round LSTM from both side of the target words.This kind of structure takes full consideration of the contextual information of the target words,and obtains higher performance of the short texts sentiment classification.Finally,we employ the generative adversarial network model,which is used for machine translation,and propose an adversarial network translation model,which mainly consists of generator model and discriminator model.Generator model is a machine translation model and discriminator model using convolution neural network as a classification.Generator model and discriminator model learn and combat from each other,while constantly improving their ability and the purpose is to higher quality of translation results.