Computer scientists at the University of Guelph developed a Twitter surveillance system capable of tracking avian influenza
By Jackie Clark
Staff Writer
Farms.com
Researchers at the University of Guelph have used Twitter surveillance to track the spread of avian influenza. The computer scientists used machine learning to find relevant tweets and pinpoint locations to help track the outbreak and spread of the disease.
Avian influenza, sometimes called bird flu, is a virus that infects many bird species, including domestic poultry. “Among humans it could be quite deadly,” Dr. Rozita Dara, assistant professor in the school of computer science at the University of Guelph, told Farms.com.
Because of the virus’s significance for animal and human health, “farmers and farm organizations will tweet if there is an outbreak or if they feel there is an outbreak in the region,” Dara said.
Over the course of a year “we collected over 200,000 tweets globally,” she explained.
The computer scientists created a system that identified keywords from several languages. To filter out irrelevant content “we used a subject matter expert because machine learning (works with) highly accurate data,” Dara added.
The subject matter expert sorted through tweets with the keywords and marked them as either relevant or not relevant to the goal of tracking avian influenza.
The researchers developed a program that was “quite strong and capable of understanding the context of the topic. With that we found that with 80 per cent accuracy we can get rid of the tweets that were not related to the topic. So we used that as a filter to get rid of the tweets that were unrelated to make our predictions more accurate,” Dara explained.
The goal was to identify if there is a correlation between Twitter activity and actual outbreaks tracked by official monitoring from institutions like the World Health Organization. The researchers also wanted to see if Twitter activity could detect outbreaks before they were officially reported or confirmed.
The study was successful.
With “access to location data, we can have a rough idea that yes, probably something is happening in that region,” Dara said. The study results published in Nature Scientific Reports stated that 75 per cent of outbreaks could be identified using Twitter, and “one-third of outbreak notifications were reported on Twitter earlier than official reports.”
Tracking on Twitter does not replace official channels. “This (Twitter surveillance system) is just kind of a weak indicator that something is happening” and useful in combination with other data, Dara explained.
In the future, similar methods may be used to help detect the spread of other animal diseases.
“You can apply (the methodology) to other diseases as well,” Dara added. However, the severity of the disease and importance to the people tweeting about it will impact the success of tracking.
Avian influenza was a good candidate because the spread “will impact other farms, but also will impact humans,” she said.
Roman Stavila\iStock\Getty Images Plus photo