Research Of Online Comment Text Sentiment Classification Based On Long-short Term Memory Network

With the sustainable development of Internet technology,various kinds of information are developing towards the direction of diversity and enormous.The need of collect and analyze all kinds of information has exploded.As an important task in data mining,document sentiment classification has gradually became a study hotspot in relational fields.The traditional document sentiment classification methods are mostly based on the methods of features designed by hand and extract,but because of the abstractness of semantic information,the deep semantic information cannot be captured accurately just with statistical information.In order to solve this problem,we design a automatically method to extract features of document based on the Long Short Term Memory network(LSTM),and carry out related experiment.Focus the problem of long sentences and large vocabulary in the document,which is not conducive to the extraction of semantic information,this paper adopts the idea of hierarchical network structure and combines the Long Short-Term Memory network for feature extraction.Based on the characteristics of document that words make sentence and sentences make document,break up the structure of document to have a second time to extract features.Based on the minimum unit,the word,to extract features and get feature vectors of sentence in word encoding level,then extract features from sentense and get feature vectors of document in sentence encoding level.And finish the feature extraction of document sentiment classification.Aiming at the problem of selecting the features extracted at different time by the Long Short-Term Memory network,attention mechanism is introduced in the network encoding process,and H-LSTM-AM model is constructed by combining the idea of hierarchical network structure.Based on the characteristics that Convolutional Neural Network is sensitive to local information,the Convolutional Neural Network(CNN)is introduced into the sentence encoding level to build the LSTM-CNN network model.Proved the effectiveness of hierarchical network structure and attention mechanism for classification model.The visual processing of the classification model is completed by combining the attention mechanism to extract the words and sentences which is more important in the document.Aiming at the problem of unbalanced sample distribution in data,we improve the loss function structure of the network to improve the degree of attenti on to data with small mount.Finally,by combining the bagging algorithm of ensemble learning,a strong classifier is constructed by integrating multiple weak classifiers,and the improvement of classification performance is proved by experiments.