Social media is one of the most accurate representers of how we lead our lives in real time. Particularly so, tweets tell us what's happening and where it is happening right as it happens. Furthermore, the data which can be extracted from tweets is much more granular than what is available from any other sources such as official diagnoses, which often lag behind the actual spread of an infectious disease. We wanted to use data from tweets to be able to predict the outbreak of diseases before they become endemic.
We used the 2009-2010 Swine Flu outbreak to test our hypothesis that Twitter mentions of symptoms of a disease correlate with the spread of the actual disease. We confirmed our hypothesis using a visualization of the Swine Flu pandemic in the US alongside a spatiotemporal visualization of Twitter mentions of Swine Flu symptoms in the US.
After confirming our hypothesis, we built a tool that lets users search for potential outbreaks happening in their region and telling them the precautions they can take to keep themselves safe. We define a disease as a collection of its constituent symptoms, and the tool looks for tuples of disease-related words in recent tweets which are occurring in the user's region at a frequency 2 standard deviations higher than the baseline frequency of the tuple. It then attempts to map the symptoms to a disease. If no disease is found correlated to a tuple of symptoms, it detects a potential new infectious disease being spread.
In the future, we seek to use machine learning to more accurately detect disease-related tweets so that we can have a larger and more accurate dataset. We also hope to get access to more data from other social media platforms like Snapchat, Instagram, Facebook, and Whatsapp. World Bank parameters suggests that 21st century global pandemics could cost in excess of $6 trillion, with an expected loss of more than $60 billion per year. With more accurate predictions, this model could potentially provide huge monetary benefits to governments which can identify and contain outbreaks of infectious diseases earlier than they normally would.
Log in or sign up for Devpost to join the conversation.