Sentiment Word Vectors Generation Generation Model Research Based On Deep Learning

Internet as an information carrier keeps the whole world connected,which allows people to access and share information anytime and anywhere.Based on the micro-blog text,this article makes micro-blog sentiment analysis from a new point of view--sentiment word vectors based on deep learning methods.Sentiment word vector is one of the by-products of deep learning model after complete sentiment analysis tasks,which at the same time can mutually reinforce the sentiment classification accuracy.High-quality word vectors inputs is the key of efficient sentiment classification of micro-blog,while more accurate supervised classification results will feedback to generation of high-quality word vectors representation.This study was carried out from three aspects.First,it proposed a new convolution neural network(HV CNN)which contacts different dimensions of vector representation and combines the advantage of dynamic convolution neural network(DCNN)which can automatically extract different characteristics of different sample length.Respectively using word vectors and character vector as network input,experiment with different model on sentiment classification tasks.Second,determining the quality of word vectors generated from different models,respectively,from the aspect of using four similarity algorithm to calculate polarity relations between similar words and upgrading the existing model of sentiment classification effect.Third,it proposed two kinds of application of high-quality word vectors,one is applied to efficient shallow machine learning models on sentence level sentiment analysis which can avoid complex syntactic analysis and feature extraction.Another is to determine word's polarity according to the distribution of words in the semantic space.By using micro-blog corpus and mobile phone reviews corpus we do comparison test,and the result show that word vectors generated by HV CNN model are of high quality,and effect of applying the model to sentiment classification is prominent.