Text Sentiment Analysis Based On Second Order Hidden Markov Models And Its Application In The "Double Reduction" Policy

With the development of the Internet,people pay more and more attention to social hot topics.Comments on social media platforms often directly reflect people’s views and emotional tendencies towards events.This paper aims to explore appropriate text sentiment analysis methods,explore the classification model based on second-order hidden Markov word segmentation model,and use the new model to mine and analyze the potential semantic sentiment information of the text related to the "double reduction" policy.The main work of this paper is as follows:Firstly,a text emotion classification method based on second-order HMM is proposed.This method is based on second-order hidden Markov word segmentation model,combined with character annotation and sentiment dictionary to analyze text sentiment.Baum-Welch algorithm was used to deduce the model parameters and determine the optimal path of the text sequence combined with Viterbi algorithm.Finally,labeled data sets were used for modeling and prediction.Practice has proved that emotion classification based on second-order HMM model can more fully identify and relate contextual information,effectively make up for the traditional first-order HMM is not suitable for real text word segmentation,and improve the accuracy of emotion analysis.Secondly,the text emotion analysis is carried out for the "double reduction" policy.Based on Python crawler technology to obtain effective commentary short text related to "double reduction" policy,combined with Boson sentiment dictionary to determine the text emotional tendency.The text data is vectorized after word segmentation by using Jieba word segmentation technology.Combined with Text Rank algorithm and word cloud map,descriptive statistical analysis of "double-minus" text is conducted to explore the concerns of the public on this policy.The text was divided into training set and test set,which were modeled and predicted by using naive Bayes,support vector machine,Long short term memory(LSTM)and Bi-directional long short-term memory(Bi LSTM).Empirical evidence shows that Bi LSTM has an accuracy rate of 89.8% for the classification of "double reduction" texts,and F1 scores of positive emotion and negative emotion texts are 88% and 91%,respectively,which is suitable for the emotional classification of "double reduction " policy.Thirdly,a text emotion classification method combining the second-order HMM and Bi LSTM is proposed,and public opinion emotion analysis is carried out on the "double reduction " text,to explore the influence of different word segmentation methods combined with different deep learning models on the accuracy of text emotion classification.The empirical results show that the emotion classification accuracy of this method is up to 92.4%,and the F1 scores of positive emotion and negative emotion classification are up to 91% and93%,respectively.Compared with other models,the performance of this method is superior,which provides certain practical significance for the subsequent research on this policy.Finally,the CNN-Bi LSTM classification model based on HA is proposed.Based on the second-order hidden Markov word segmentation model,the HA-CBLNet model is proposed on the basis of effectively preserving the potential emotional information in the preprocessing stage.This model effectively combines the advantages of CNN and Bi LSTM models for feature extraction,and makes up for the network forgetting some important information through the attention mechanism.The empirical results show that the classification accuracy of this model reaches 93.75%,which is improved to some extent compared with other models in the evaluation indexes of various models,and proves the effectiveness of feature fusion.