When knowledge-driven beats data-driven: achieving sub-city level geolocation without prior training



Laura Di Rocco

When knowledge-driven beats data-driven: achieving sub-city level geolocation without prior training
Speaker:
Laura Di Rocco
Abstract:
Knowing the location a microblog message, such as a tweet, originates from enables relevant social analyses. Since, however, onlys a smallsubset (around 1%) of microblog messages come alreadyg geolocated automatic geolocation methods have been developed. State of the art data-driven methods do not perform well at subcitye level as they are prone to overfitting noisy data. In this paper, wepropose an alternative knowledge-base method, called Sherloc, toaccurately geolocate messages at sub-city level, exploiting the resence of toponyms referring to the area in the message. Sherloc exploits the semantics associated with toponyms contained in gazetteersand embeds them into a metric space that captures the semanticdistance among them. This allows toponyms to be represented as points and indexed by a spatial access method, allowing us to identify the semantically closest terms to a microblog message, thatalso form a cluster with respect to their spatial locations. Incontrast to state of the art methods, Sherloc requires no priorr training it is not limited to geolocating on a fixed spatial grid and it is experimentally demonstrated able to infer the location at sub-city levelwith higher accuracy.
Short Bio:
Laura Di Rocco is a PhD Student at University of Genova, Italy in computer science. Her research focus is in Geographic Information Retrieval from microblog messages. During her PhD, she collaborate with Prof. Michela Bertolotto in UCD, Dublin, Ireland.
Date:
Thursday, June 28th, 2018 - 14:30
Location:
DIBRIS, Valletta Puggia, Conference Room 3rd floor - Via Dodecaneso 35, Genova 

Sei qui: Home Ricerca Seminari di ricerca When knowledge-driven beats data-driven: achieving sub-city level geolocation without prior training

Questo sito usa i cookie.