Biomedical Engineering Reference
In-Depth Information
Fig. 1. The system architecture of SNEFT
3.1
SNEFT Architecture
We propose the Social Network Enabled Flu Trends ( SNEFT ) architecture along with
its crawler, predictor and detector components, as our solution to predict flu activity
ahead of time with a certain accuracy.
CDC ILI reports and other influenza related data are downloaded into the “ILI Data”
database from their corresponding websites (e.g., CDC [3]). A list of flu related key-
words (“Flu” , “H1N1” and “Swine Flu”) that are likely to be of significance are used
by the OSN Crawler as inputs into public search interfaces to retrieve publicly avail-
able posts mentioning those keywords. Relevant information about the posts such as
time,location and other demographic information is collected along with the relative
keyword frequency and stored in a spatio-temporal “OSN Data” database for further
data analysis.
An Autoregressive with Exogenous input (ARX) model is used to predict ILI inci-
dence as a linear function of current and past OSN data and past ILI data thus providing
a valuable “preview” of ILI cases well ahead of CDC reports. Novelty detection tech-
niques can be used to continuously monitor OSN data, and detect transition in real time
from a “normal” baseline situation to a pandemic using the volume and content of OSN
data enabling SNEFT to provide a timely warning to public health authorities for further
investigation and response.
3.2
OSN Crawler
Based on the search API provided by Twitter and Facebook, we have developed crawlers
to fetch data at regular time intervals.
The Twitter search service accepts single or multiple keywords using conjunctions
(“flu” OR “h1n1” OR “#swineflu”) to search for relevant tweets. Search results are typ-
ically 15 tweets (maximum 50) per page up to 1,500 tweets arranged in chronologically
decreasing order, obtained from a real time stream known as the public timeline. The
tweet has the User Name, the Post with status id and the Timestamp attached with each
post. From the twitter username, we can get the number of followers, number of friends,
his/her profile creation date, location and status update count for every user.
The “Post by everyone” option allows us to search public posts for given keywords
in Facebook. All results that show up are available to the public for a limited time
period. We are interested in getting useful information (profile ID, time stamp of the
post, and the post content) out of posts. Given a profile ID, we will retrieve the detailed
Search WWH ::




Custom Search