So what we are after is like a cognitive solution and by the way is actually a full implemented demonstrator for that

Gordon Pipa, Professor for Cognitive Science , University of Osnabrück

Business Challenge story

Cognitive science is an academic field that studies mental processes. It examines language, perception, memory, attention, reasoning, emotion and more, across many other academic fields. Cognitive science research has led to key findings about artificial intelligence and cognitive disorders and has even furthered our understanding of human persuasion, coercion, bias and other behaviors. Osnabrück University in Germany is home to four research institutes producing promising work that spans academic disciplines, provides unprecedented opportunities for research students and successfully attracts third-party funding. The university’s Institute of Cognitive Science is focused on studying higher cognitive functions, both to acquire scientific knowledge and to develop technological applications of that knowledge. The institute traditionally seeks out important projects that can serve society while reinforcing the university’s reputation among scholars and supporters. In 2015, the Institute of Cognitive Science conceived of a research project to predict and manage flu outbreaks. It saw the difficulty in predicting outbreaks and untangling the real reasons behind them. The institute knew it would need to consider a wealth of data that was usually only captured by medical facilities. These facilities might take two weeks or more to provide the data, and they could only report on limited parameters. The institute wanted a better source of data to help it study, track and ultimately help predict flu outbreaks. It wanted a system that could find clues within free-form discussions on social media and even interact to advise people about the precautions they needed to take.


The Institute of Cognitive Science at Osnabrück University wanted to study and try to predict the timing and root causes of flu outbreaks. But the only data researchers could analyze for this complex question lay in hospital reports that were limited and always at least two weeks out of date. The researchers saw that social media held many potential clues, but they needed a way to effectively analyze the content and form predictions. In only six months, they were able to develop a natural language processing (NLP) system that analyzes Twitter feeds against a central body of knowledge to help predict flu outbreaks and even answer user questions. Now, the institute can generate predictions, analyze causes and suggest preventive actions for flu outbreaks based on instant analysis of social media and the latest research. Using IBM Cloud technology and the IBM Watson Developer Cloud to access the IBM Watson Natural Language Classifier service, the institute examines around 500 million English tweets worldwide every day. The institute is also developing a German system to use when the IBM Watson solution’s German language analysis becomes more established. The system studies nuances in the content and context of tweets, discerning whether someone is discussing the flu vaccine or flu-like symptoms. During analysis, the solution consults a central body of knowledge that includes more than 3,000 complete research papers and data from the Centers for Disease Control and Prevention (CDC) in the US. Based on this analysis, the system finds correlations between Twitter discussions and the factors known to lead to outbreaks. It uses these correlations to help predict outbreaks before they happen. The system also lets users ask free-form questions, answering them with IBM Watson Engagement Advisor technology. For example, users can ask the system “Is there a flu threat in this neighborhood?” or “Do I need to be vaccinated?” or even “How do I tell whether I have a cold or the flu?” The system guides users through a natural conversation to provide meaningful answers that help them take preventive measures and save costs by reducing unnecessary doctor visits. The solution took only six months to develop and improves its responses to questions over time as researchers study user questions to determine which new documents to load into the body of knowledge. To help improve the system’s prediction rate, researchers are loading more data from CDC, state agencies and hospitals. The system’s predictions also improve over time as it uses proprietary algorithms to learn from tweets. Once distributed beyond the Institute of Cognitive Science at Osnabrück University, the system has significant potential to impact healthcare and health insurance. For example, healthcare facilities can use the solution’s outbreak predictions to stock medications or make other preparations in advance. Better predictions can also reduce the severity and spread of outbreaks. In the future, the institute is looking at training the system to answer questions and make predictions about other diseases too. Researchers have already discovered that Twitter discussions hold clues about other highly infectious diseases that could be studied in subsequent work. You to the power of IBM The influenza virus infects millions of people each year and, sadly, many thousands die from it. If public health professionals had earlier warnings of flu outbreaks, they could target educational efforts and medical resources to those most at risk. Unfortunately, flu-case statistics from hospitals and physicians have minimal value for early warning because of a two-week delay in data collection. Now, thanks to a cognitive solution from the University of Osnabrück based on Twitter, healthcare workers can pinpoint outbreaks in near-real-time. When someone thinks they have the flu, they are likely to alert their friends on social media. The solution capitalizes on this by analyzing some 500 million tweets daily for references to the flu and flu-like symptoms, then correlates the analysis against 3,000 research papers to identify actual cases. Empowered by the ability to discover outbreaks as they happen, health workers can allocate resources accordingly, potentially saving many lives.


Saves 60 percent of the time researchers spent on coding and implementation, from 90 percent to only 30 percent, leaving more time for conceptual work; Virtually eliminates two-week analysis because the system provides instant analysis of tweets and core data rather than waiting for hospitals to report data; Generates a commercially relevant system that can help reduce healthcare demands and health insurance claims; Builds predictive capabilities using a cognitive analysis of discussions and data to help predict outbreaks and provide relevant precautions

Solution Category

  • AI/Watson