Digitale Epidemiebekämpfung

During the COVID-19 pandemic and previous epidemics, such as the 2013 Ebola outbreak, social media provided valuable real-time data: Millions of tweets were posted – with useful information on symptoms, transmission routes, and prevention measures, but also with irrelevant or misleading content. How can this flood of data be used efficiently? A recent study by L3S shows how AI-powered models can filter and summarise relevant content from thousands of tweets to deliver key insights through a trustworthy approach.

The method presented in the study uses advanced machine learning techniques to automatically classify and concisely summarise tweets. The focus is not only on the model accuracy but also on the interpretability of the results. This is because many current AI models operate as a black box, users cannot understand how the AI arrives at its decisions.

Our model extracts key information, so-called rationales, from the tweets to explain its decisions,” says Thi Huyen Nguyen, first author of the study. Moreover, rationales capture essential content in tweets. These rationales can be employed to create concise summaries of the situation, making them especially valuable for decision-makers who need to respond promptly to evolving circumstances.

The results of the study are promising: the developed model achieved a classification accuracy of 82 per cent, outperforming conventional methods. In addition, the proposed simple graph-based ranking method helped to filter out the most important information and avoid redundancy. The generated short updates provided a comprehensive picture of the situation during an outbreak.

Social networks are therefore not only a place for communication but also a valuable source of data for crisis management – provided you have the right tools to manage the flood of data.

Thi Huyen Nguyen, Marco Fisichella, Koustav Rudra: A Trustworthy Approach to Classify and Analyze Epidemic-Related Information From Microblogs. IEEE Trans. Comput. Soc. Syst. 11(5): 6229-6241 (2024) ieeexplore.ieee.org/abstract/document/10530086