Research On Sentiment Analysis Of Chinese Text-Oriented Neural Network Models

With the advent of the 5G era,data resources will see explosive growth.In terms of the Natural Language Processing,the soaring subjective texts provide sufficient corpus for sentiment analysis.Traditional manual feature selection is likely to cause inaccurate segmentation semantics.Compared with deep learning neural network models,shallow machine learning classifiers have limited ranges for improvement in classification accuracy.Different from English,Chinese is a very large language system composed of Chinese characters.Its diverse expression methods and complex semantics have brought huge challenges to Chinese sentiment analysis.In order to extract the semantic features of sentences,remove irrelevant noise,reduce vector dimensions,and improve the accuracy of sentiment discrimination,the following research works have been done:(1)High-quality preprocessing results have great significance for subsequent neural network model learning.Therefore,an algorithm module(SL-W2V-Plus)based on sentiment lexicon and Word2 vec incremental training has been proposed for feature selection and word vector training.First and foremost,Chinese comment set was executed data cleaning operation.Secondly,the data set was segmented by loading customized sentiment lexicon.Then,the Word2 vec model was obtained through the Word2 vec algorithm with the preprocessing results.Finally,the Word2 vec model was done incremental training by adding the Sogou CA corpus.Here we got the relationship between words.Based on the SL-W2V-Plus feature extraction method,the experimental results show that the F1 value and accuracy have improved by 1%-2%,proved the effectiveness of the proposed method.(2)In order to solve the problems that a single Convolutional Neural Network lacks information exchange in one layer and a simple Recurrent Neural Network cannot solve the problem of long-term dependence,a sentiment classification model with hierarchical network CNN-Bi LSTM introducing attention mechanism has been proposed.Firstly,the Convolutional Neural Network was used to extract phrase features deeply.Secondly,the Bi-directional Long Short Term Memory network was used to serialize the sentence information and to obtain the holistic characteristics of the sentence.Finally,an attention layer was added to select the effective features by weighted summation of the sentence,then the result was classified by the binary classification function.Under the data set,several comparative experiments have been done.The results show that the proposed HCBLA model get a higher F1 value and accuracy.It shows that the model has good application ability in dealing with the classification of Chinese comment text.