Veronika Strnadová – University of California, Santa Barbara
David Jurgens, Tsai-Ching Lu – HRL Laboratories, LLC
Supported by the Intelligence Advanced Research Projects Activity (IARPA) via Department of Interior National Business Center (DoI/NBC) contract number D12PC00285.
Goal – Analyze Microtext Characterize online discussions using a network
model of Twitter
Goal – Analyze Microtext Characterize online discussions using a network
model of Twitter
Identify different discussion types
Background
Correlating financial time series with micro-blogging activity. Ruiz E., Hristidis V., Castillo C., Gionis A., Jaimes A.
Keyword Categories
Musicians Sports Teams Companies Countries
Lady Gaga Dodgers Microsoft Great Britain
Jay-Z Yankees IBM United States
Red Hot Chili Peppers Cardinals Intel Uganda
Dave Matthews Band Diamondbacks Apple Greece
Kanye West Padres Sony Mexico
Dataset Twitter Decahose: a live feed of 10% of real tweets,
sampled at random
15,000 tweets per keyword graph per day
Network Construction: Graph Schema
Network Construction: Graph Schema
Network Construction: Example
Los Angeles, California, USA
Alice
Bob Clarissa
tweet1
#mlb
tweet2
tweet3 tweet4
#dodgers
tweet5
tweet6 #cardinals
St. Louis, Missouri, USA
Temporal Changes in Network Properties
Whole-Graph Metrics Graph Node Metrics
• Node Type Composition • Number of Connected Components • Diameter • Normalized Diameter
• Degree Distribution • Pagerank Distribution • Closeness Distribution
Results: Normalized Diameter
Results: Normalized Diameter “Yankees” Keyword Graph
Results: Normalized Diameter “Yankees” Keyword Graph
July 2012
Sunday Monday Tuesday Wednesday Thursday Friday
Results: Pagerank
Results: Pagerank “IBM” Keyword Graph
Results: Pagerank “Lady Gaga” Keyword Graph
Closeness
Closeness
Conclusion and Future Directions Modeling online discussions as relationships between people, locations, and topics has potential to reveal interesting properties of online discourse
Conclusion and Future Directions Modeling online discussions as relationships between people, locations, and topics has potential to reveal interesting properties of online discourse
Future work:
Move beyond basic graph metrics
Conclusion and Future Directions Modeling online discussions as relationships between people, locations, and topics has potential to reveal interesting properties of online discourse
Future work:
Move beyond basic graph metrics
Identify key elements in a discussion
Conclusion and Future Directions Modeling online discussions as relationships between people, locations, and topics has potential to reveal interesting properties of online discourse.
Future work:
Move beyond basic graph metrics
Identify key elements in a discussion
Track geographic shifts of public interest in a topic
Thank You!
Backup – Graph Construction