New framework tracks public opinion shifts during a Chikungunya outbreak on Weibo

Researchers have introduced a dual-level framework that couples event knowledge graphs with a topic modeling approach to follow how public opinion evolves on social media during public health emergencies. The work, published in EPJ Data Science, was applied to 22,744 Weibo posts related to the Chikungunya fever outbreak.

The framework addresses two persistent problems in social media monitoring: the semantic noise that arises when analyzing very short texts, and the fragmented way in which discourse structures and user behavior have typically been studied separately. The authors combine Event Knowledge Graphs (EKG) with user behavior profiling, both built on a Centroid-Alignment Factor (C-AF) embedding-enhanced BERTopic model.

When applied to the Weibo corpus, the C-AF model improved Topic Diversity from 0.5725 to 0.6480. According to the study, this optimization enabled the construction of a high-precision event knowledge graph that reveals core discourse evolutionary pathways, showing how conversation themes shifted as the outbreak unfolded.

For user behavior segmentation, the framework achieved a peak Silhouette Coefficient of 0.909, which the authors report as outperforming the 0.670 baseline by 35.7 percent. An external semantic indicator, Topic Coherence, also improved from 0.550 to 0.783, providing a second line of validation for the clustering results.

This dual validation allowed the researchers to identify five distinct user profiles among the accounts posting about the outbreak: Early Cognitive, Knowledge Dissemination, Risk-Attention, Prevention-Focused, and Preventive Alert types. Each profile represents a different pattern of information behavior, from early awareness of the event to active dissemination of knowledge and attention to preventive measures.

The authors position the framework as a scalable, data-driven tool for precise public health crisis monitoring and targeted digital governance intervention. By linking what people are talking about with how different groups of users behave, the approach aims to give authorities a more granular picture of public sentiment during an emergency than topic modeling alone can provide.

The study is published as open access research in EPJ Data Science, with the article shared early to provide faster access to peer-reviewed, accepted research; the published version is subject to further edits and will be replaced by the final Version of Record. The article was published on 9 October 2026 and is authored by Juan Wang, Xianghui Kong and Xiaojing Zhang.

Because the framework was developed and tested on a single dataset of Weibo posts about one outbreak, its performance on other platforms, languages, and public health events remains to be demonstrated. The reported metrics describe model performance on this corpus rather than outcomes of any intervention deployed in the field, and the identified user profiles characterize posting behavior within this specific dataset.

Subject of Research: A topic modeling and event knowledge graph framework for tracking public opinion and user behavior on social media during a public health emergency.

Article Title: Quantifying public opinion evolution and information behavior dynamics: a framework coupling event knowledge graphs with centroid-aligned topic modeling

Article References: Wang, J., Kong, X., & Zhang, X. (2026). Quantifying public opinion evolution and information behavior dynamics: a framework coupling event knowledge graphs with centroid-aligned topic modeling. EPJ Data Science. https://doi.org/10.1140/epjds/s13688-026-00709-3

Image Credits: AI Generated

DOI: 10.1140/epjds/s13688-026-00709-3

Keywords: public opinion, topic modeling, event knowledge graphs, Weibo, Chikungunya, social media, public health, user behavior profiling, BERTopic, crisis monitoring, text mining, Quantifying