Research On Sentiment Classification Based On Emotional Dictionary And Deep Learning Technology

With the development of mobile 4G technology and the popularization of smart mobile phones,users around the world can evaluate movies?products on the internet at anytime and anywhere.now,more and more researchers use natural language processing technology to deal with increase per second GB information of text,explore its inner sentiment tendency and its successfully applied in the fields of social public opinion supervision,equity investment etc.Sentiment analysis are based on the methods of emotional dictionary and machine learning in the past.However,these methods are based on artificial construction rules and artificial extraction features,and it waste high human cost.in addition,the vector representation based on the one-hot word representation are faces the challenges of "high latitude"and“semantic independence".With the development of deep learning technology,deep learning is emerging in the field of natural language processing,and solved the problems.therefore,this paper mainly studies the word vector representation and sentiment classification technology based on deep learning.The details are as follows:(1)Aim at the problems about high dimension and word semantic independence caused by one-hot vector representation.this article trained word vector by word2vec for text representation.this method not only greatly reduces the dimension of text expression,and also can conveys the semantic information of the words.by this way provides more semantic information for the neural network model.In view of the accuracy of the extraction of emotional words can affect the quality of word vector,this article take advantage of combine the user-defined dictionary when using jieba to participle,Integrate the emotional resources as a user-defined dictionary libraiy.by this method,we can improve the accuracy participle and the accuracy of sentiment words extract,in deal with the new emotional words,this article training model of synonyms by word2vec to handle it,and extend the user-defined dictionary.By training the different word vectors by the different participle as the input of the Convolutional neural netwok calssification model,the different classification results are analyzed to verify the feasibility of this method.(2)The traditional sentiment classification method requires a large amount of manpower and material resources,but the deep learning model like convolutional neural network can automatically extract the features of different dimensions without the help of manpower.The GRU neural network can store more information and study context dependence with simple construction.in this article,the two nerual networks are combined to construct an new sentiment classification model of hybrid convolutional neural network and GRU neural network.by the experiments,the models based on deep learning can effectively improve the classification effect compared to the traditional machine learning.and the proposed neural neural network model has achieved good results when compared to the Recurrent neural network and long short term memory.