Study on the evolution of online public opinion and the correlation between event propagation based on ELM model and emotion measurement

In the context of the deep penetration of digital technology into social governance, this study proposes a framework for analysing the evolution of public opinion, integrating Extreme Learning Machine and Sentiment Computing. The study employs web crawler technology to procure public opinion data from open web data sources, and inputs this data into the Extreme Learning Machine model following text cleaning, word segmentation and labelling, and other pre-processing steps. The ELM algorithm, underpinned by the random initialization mechanism of the hidden weights, facilitates the rapid recognition of public opinion evolution patterns in high-dimensional feature space. The emotion computation model, constructed by combining the HowNet emotion dictionary and the BiLSTM-CRF algorithm, quantifies public emotion in multiple dimensions. The experimental results demonstrate a significant correlation between the key indicators of public opinion evolution and the elements of event communication. The quantitative analysis model proposed in this study provides decision support for cyberspace governance. The model's application to a regional online public opinion monitoring platform has yielded positive outcomes, including enhanced response time and improved precision in public opinion guidance.