| In recent years.social media such as Microblog has provided a good platform for the public to make their voices heard,and has also had a wide influence in public opinion supervision.The network has the advantages of timeliness and convenience.For the situation after the implementation of the " double reduction " policy,the public can legitimately express their opinions on social media,express their feelings and show their emotional tendencies.In the face of extensive text review data in social media.this thesis hopes to grasp the future emotional trend of the public and predict the information heat through algorithm model.This thesis crawls the "double reduction " policy comment data(including comment text,number of retweets,number of likes,number of comments,etc.)from August 1.2021 to October 31,2021 on the Microblog platform,carries out sentiment classification and heat prediction tasks for the comment data,and understands the sentiment classification and heat trend of the comment information through the algorithm model.The main innovation of this thesis is to propose a sentiment classification method based on ERNIE-BiLSTM model and a heat prediction method based on EEMD-GRU-XGBoost model,which are applied to the analysis of this thesis.The main research results and conclusions are as follows:1.The information of microblog text is fully mined.Considering that most microblog text mining studies only consider sentiment analysis or hotness prediction.this thesis combines these two aspects to analyze the microblog text comprehensively and maximize the use of existing text data to mine information.2.ERNIE-BiLSTM model and EEMD-GRU-XGB model are proposed.The ERNIE-BiLSTM model uses the ERNIE model to embed text sequence words and encode the word embedding matrix.The generated text vector representation is used as the input of the BiLSTM model to obtain long-distance dependencies from both positive and negative directions.The Softmax classifier performs decoding to achieve sentiment analysis.The EEMD-GRU-XGB model uses the EEMD model to decompose the heat value.The decomposed IMF is fitted for the first time using the GRU model to obtain the fitted value of each IMF,and the error between the first fitted value and the original value is fitted for the second time using the XGBoost model.The two fitting values are added to the predicted value of each IMF component,and the predicted value of all IMFs is the predicted value of heat,so as to achieve the purpose of heat prediction.3.The ERNIE-BiLSTM model and EEMD-GRU-XGB model are used to analyze the pre-processed data based on the "double reduction" policy comments,and the analysis process and experimental results are presented.A variety of mainstream models are used for performance comparison experiments to evaluate the effects of different models.Finally,it is found that the model proposed in this thesis is indeed superior to other models and has achieved good analysis results. |
