Authors:
Bingnan Li
;
Zi Chen
and
Samsung Lim
Affiliation:
School of Civil and Environmental Engineering, University of New South Wales, Sydney, Australia
Keyword(s):
Social Media, Geolocation Prediction, Tweets, Influenza-like Illness, Data Mining.
Abstract:
Twitter has become an effective platform for gathering massive event-related data from growing popularity. It provides an approach to monitoring and analysis of the emergence and devolvement of events. In the field of data mining and social media analysis, geographic information is an important element to be factored in. However, only nearly 2% of tweets contain accurate geographic information because of various concerns e.g. complexity and privacy. In order to overcome this restriction, devising methods of geolocation prediction has become the main topic in this filed. Geographic information plays a valuable role in responding to the control and surveillance of epidemic diseases. In this study, we constructed a geolocation prediction method based on potential location-related tweet metadata. Coordinate information can be calculated from the bounding box, while location information can be extracted from the text content, the user’s location at the time of use and the labelled place n
ames using the Named Entity Recognition technique. Three types of coordinate sets of Australian suburbs are defined and used to construct coordinates references from the place names. Models with different parameters have been applied to predict geolocations of influenza-like illness from the tweets of the 2019 flu season in Australia. The results show that the proposed models with four parameters perform better than the existing models. When the area threshold is set to 4,500 km2, the best model can successfully predict influenza-like illness with the mean error distance of 4.65 km and the median error distance of 2.57 km. Hence the proposed method is shown to enhance the geographic information associated with the tweets and make the emergency response to influenza-like illness more effective and efficient.
(More)