Upload
lu-chen
View
392
Download
0
Embed Size (px)
DESCRIPTION
Citation preview
Lu Chen @ http://knoesis.wright.edu/
Understanding User’s Geographic Context in Twitter
What are people thinking and doing in a given time and place?
Sentiment
Event
NOW?
User’s Geographic Context in Twitter
User location in profile Where is the user?
Geo-tags attached with the tweets Where is the user now (when the tweet is created)?
Places mentioned in the tweets ?
How are they related to each other? How can they be helpful to disambiguate or even predict each other?
Identifying and Disambiguating the Place Mentioned in Tweets
How to identify? Linked Geo Data
How to disambiguate? Other place names mentioned in tweets
In the same tweet In the tweets sent by the same user
Geo-tag of the tweet User location
Any doubt?
It’s not as easy as I was thinking… In fact, it’s pretty hard…
Start from collecting data Key words? Geo tags? User s?
Wikipedia: Top 10 most common U.S. place namesTracked 5000 usersCollected 2,187,205 tweets in total
Some Facts about the Data7.12% (155,705) tweets mention place names (according to
LGD)1.97% (42,988) tweets have geo-tags57.36% (2,868) users provide location information (might
be invalid)
Some Facts about Linked Geo DataBased on the Open Street Map.org planet file from 6th
April 201166 million triples, 10GBVirtuoso SPARQL endpoint:
http://knoesis-twit.cs.wright.edu:8890/sparql
Location DisambiguationKnowledge from LGD
Label, type, is_in, population, latitude, longitude For each place with its disambiguator
Is_in relations Minimize the minimum bounding box Minimize the distances
Type and populationhttp://www.movable-type.co.uk/scripts/latlong.html
Next StepRefine the algorithmEvaluationVisualization
What I LearnedFrom reading papers in this areaFrom using Linked Geo DataFrom coding, trying different tools and packages…From all these difficultiesFrom thinking, including thinking about what I learned
When I am thinking hard…