76.031| Volume 14, SUPPLEMENT 1, e384-e385, March 2010

The potential of social networks for early warning nad outbreak detection systems: the swine flu Twitter study

Open ArchivePublished:March 08, 2010DOI:
      Background: The recent swine flu outbreak in April-May 2009 truly demonstrated the potential of these media for early warning systems. Web 2.0 has generated a great interest recently as a possible media for early warning system for outbreak detection and epidemic intelligence (EI). Traditional systems such as GPHIN, Medisys are well established and used by ECDC/WHO on a daily bases, however, there has been recent interest in the ability to estimate flu activity via aggregating online search queries for keywords relating to flu and its symptoms by commercial companies like Google. However, the search data remain proprietary and therefore not useful for research. However, content of social networks such as twitter are in public domain and therefore available.
      Methods: Twitter, a micro-blogging service that allows people to post and read other users’ 140 character messages currently has over 15 million unique users per month. Twitter allow third parties to search user messages using an open source API and return the text along with information from the poster's profile, such as their location, in a format that can be easily stored and analysed. We searched and collected over 1 million tweets in the period from May until August 2009 and carry on collecting them on a minute bases to understand public concerns, keywords used and the profile of users who discuss these topics on the web.
      Frequencies of the most popular words found in all tweets.
      Results: We found over 1 million tweets reporting flu related illnesses and symptoms via Twitter in this period. Most popular words in tweets were these (frequency in brackets): flu (138, 260), Swine (99, 179), Have (13, 534), Cases (13, 300), H1N1 (9, 134), Has (8, 010) and others. The actual sentence “I have swine flu” appeared 2, 888 times and “I have flu” 1,530 times. Further evaluation of the collected tweets, semantic relationship of keywords, geo-location of posters is underway and will be presented at the conference.
      Collocation of words, one word to the right and the left of the word “flu”.
      Conclusion: The potential of social networking system for early warning systems and for better understanding public concerns about their health is enormous, however, further research is required to reveal the underlying principles and implement adequate integration with existing healthcare services.

      Linked Article