Welcome to Project SNEFT

Jan 13, 2013Posted by Harsh

 

In News :

Our research objective is to extract and utilize information from keyword-frequency data obtained from Online Social Networks (OSNs) such as Facebook and Twitter, in order to provide timely prediction of the emergence and spread of an influenza epidemic.

Reports of Influenza-Like Illness (ILI) cases by the Centers for Disease Control and Prevention (CDC), though authoritative, typically have a 1-2 weeks delay due to the largely manual process. Public health authorities need the earliest possible warning to ensure effective intervention, and therefore more efficient and timely methods of estimating influenza incidence are urgently needed.

More and more people are using OSNs everyday and talking about their daily activities and events. When they are sick, announcements such as "I'm coughing and sneezing and feeling sick" are often posted. Although these data would be "noisy" individually, in the sense that not everyone who sneezes and coughs is infected with an influenza virus, in aggregate, they provide a previously untapped data source that can be transformed into a large scale picture of the underlying epidemic pattern in time and space. Such a picture will have a very short time lag, since OSN data are almost concurrent with what's going on and can be obtained almost in real time.

Already, Google Flu Trends uses web search terms such as "influenza complication" and "cold remedy" collectively to predict the onset of the annual influenza season 1-2 weeks ahead of the CDC ILI data. It is thus expected that the "I have a cold" status and "get well soon" messages exchanged between OSN users and their friends may provide earlier and robust prediction. Such data will be available from scanning public OSN profiles.

This project will develop an automated system to aggregate and transform OSN data into information for the early detection and prediction of temporal and geographic influenza incidence. It will be capable of

  • Novelty Detection, to focus on detecting the transition from a "normal" baseline situation to a pandemic,
  • ILI prediction using (ARMA) time series models, to provide "preview" of possible scenarios
  • Nonlinear Filtering, enhances predictive power of mathematical models (influenza) with unknown params.

Such a system will become a valuable tool for public health authorities, for example by becoming part of the CDC's BioSense program.

Download : Poster