融合情感分析与多元时间序列的区块链产业舆情监测研究

Research on Public Opinion Monitoring of Blockchain Industry That Integrates Sentiment Analysis and Multiple Time Series

[目的/意义]区块链技术被纳入“新基建”范畴后,其产业发展演进快、舆情热度高。本研究将情感因素纳入新兴产业网络舆情热度预测,探究区块链产业关注主题及发展态势。[方法/过程]论文融合情感分析与多元时间序列特征提出舆情热度预测模型,采用BERT-BiLSTM(Bi-directional Long Short-Term Memory,BiLSTM)方法对舆情文本分类并赋值,挖掘情感极性类别的主题,将不同情感倾向的情感值分别取绝对值累加,构建基于情感因素的多元时间序列特征体系,并输入LSTM(Long Short Term Memory,LSTM)模型进行区块链产业舆情热度预测。[结果/结论]BERT-BiLSTM在情感分类任务中准确率为84%,其中消极和中性情感类属文本的成因主要为“对于区块链技术的不信任”和“缺乏区块链相关概念的了解”。在热度预测模型中,模型均方根误差(Root Mean Square Error,RMSE)降低17.67,平均绝对误差(Mean Absolute Error,MAE)降低15.14,决定系数(R-Square,R2)提升11%,模型总体性能良好。

[Objective/Significance]The blockchain technology has developed fast and attracted a lot of attentions after being included in the scope of“new capital construction”.The research integrated the emotional factor into the prediction for the degree of discussion of online public opinion about the emerging industry and explored the hot topics and development trends of the blockchain industry.[Method/Process]The research proposed the prediction model for the degree of discussion of online public opinion through combining the emotional analysis and the multivariate time series characteristic;made classification and assignment for the text of public opinions by the BiLSTM(Bi-directional Long Short-Term Memory)method,explored the topic under the category of sentiment polarity,cumulated the absolute values of the emotional value with different emotional tendency,created the multivariate time series characteristic system based on the emotional factor,and input the LSTM(Long Short Term Memory)module for predicting the degree of discussion of online public opinion about the blockchain industry.[Results/Conclusions]The empirical results revealed that the BERT-BiLSTM method could reach the accuracy rate of 84%in the emotion classification task,where the factors for negative and neutral emotion type text included“not trust in the blockchain technology”and“insufficient understanding for the blockchain related concept”.In the prediction model,the RMSE(Root Mean Square Error)decreased 17.67 and the MAE(Mean Absolute Error)decreased 15.14,while the R2(R-Square)increased 11%,indicating good overall performance of the model.