The Web Comment Text Sentiment Analysis Based On Deep Learning

In the age of information,various kinds of network reviews have increased dramatically.People are used to express opinions and express emotions on the network.These texts,which contain a large number of opinion attitudes and trends of opinion,contain important commercial and research values.Therefore,text sentiment analysis has become an important research field in natural language processing.In recent years,with the development of deep learning,deep learning technologies have made significant progress in the field of image,speech and natural language processing and demonstrate the potential values.Therefore,this paper carries out a study on text sentiment analysis based on the deep learning models.Firstly,a text sentiment analysis method based on LSTM network is established under the word vector model,and its improvement is made.Due to word vectors are learned from contexts and lack emotional information,in order to further improve the classification accuracy,for the characteristics of irregular text grammar,sparse features,etc.,this paper combines the deep learning technology with the emotion rule method to propose a text sentiment analysis method based on multi-feature fusion deep belief network.It can introduce the expression of semantic information,improve the model's ability of learning and understanding the meaning of texts.The main tasks of this paper are: establish a text sentiment analysis mothod based on LSTM network;Collect and arrange major emotional dictionaries and establish a complete emotional dictionary;Improve the classic emotional dictionary classification method,and extract effective emotional information from it to expand into emotional features.Then combine the three features processed by word segmentation and StanfordNlp processing to form a multi-feature fusion machine learning feature template,and use it as the input of the deep belief network.This paper used the Professor Tan Songbo's hotel review data sets for experiment.Theresults show the following results: 1.The two methods can better complete the text sentiment analysis task.2.The DBN can learn and express emotion features better than SVM.3.The deep belief network method combined with the emotion rule method and deep learning technology has better performance than LSTM network method using word vector model.