社交媒体文本的价值测度——基于深度学习的网络消费舆情指数构建及其作用研究

Value Measurement of Social Media Texts—Research on the Construction and Function of Online Consumer Public Opinion Index Based on Deep Learning

摘要: 随着互联网和社交媒体的快速普及，社交媒体文本因其实时性、互动性和无偏性而受到关注。挖掘量化社媒消费舆情，并进一步研究其对消费者信心的影响成为有意义的探索。文章首先构建了6个不同主题维度的消费热点关键词库，爬取了社交平台约178.2万相关热门微博。进一步地，基于Bert预训练模型挖掘情感倾向，并引入微博热度得分和小波变换，创建了一个用来测度消费舆情的网络消费舆情指数。研究发现，网络消费舆情指数对消费者信心具有显著正向影响且两者存在长期稳定的均衡关系；进一步将其加入消费者信心预测模型，所有评价指标均得到显著提升，且LSTM在不同预测期限内的预测效果和鲁棒性都优于GRU和Random Forest。

Abstract: With the rapid popularization of the Internet and social media, social media texts have attracted attention because of their real-time, interactive and unbiased nature. It has become a meaningful exploration to explore the quantitative public opinion of social media consumption and further study its impact on consumer confidence. The paper first constructed a consumer keyword database with 6 different topic dimensions, and crawled about 1.782 million related popular microblogs on the social platform. Further, based on Bert pre-training model, emotional tendency is mined, and an online Consumer Confidence Index is created to comprehensively quantify consumer confidence by introducing Weibo heat score and wavelet transform. It is found that the online CPOI has a significant positive impact on consumer confidence and there is a long-term stable equilibrium relationship between the two. When it is further added into the consumer confidence forecasting model, all evaluation indicators are significantly improved, and the forecasting effect and robustness of LSTM are better than that of GRU and Random Forest in different forecasting periods.