The Role of Taxonomy and Ontology in Semantic Layers - Heather Hedden.pdf
Text Virality in Social Networks - ICWSM 2011
1. Exploring Text Virality in Social Networks
Marco Guerini, Carlo Strapparava, Gözde Özbal {marco.guerini@trentorise.eu, strappa@fbk.eu, ozbal@fbk.eu}
Virality: tendency of a content to spread quickly in a community by word-of-mouth...
Virality is a phenomenon related Virality is a social phenomenon in
which there are no “immune carriers”
to the magnitude of the social connections
Virality is a phenomenon related to the Social network connections
Social network graphs,
characteristics of the content being spread accounts for HOW content
Opinion Leaders, etc.
spreads, rather than WHY
Virality as a single-
faceted phenomenon number of people who accessed a
content in a given time interval
e.g. number of comments, the ability
number of “I_like” to induce discussion among users e.g.
number of “I_like”
polarize
the audience (pro and against the
given content).
Virality has how much
many facets! people share content by
forwarding it
positive comments.
how much
“The best product I have ever bought”
Dataset and Metrics people comment a content
negative comments.
“Do not buy this product, it is a rip-off”
Digg dataset as a unique framework.
Text-based contents. Define metrics
to formalize every viral phenomenon.
Rais = (NCL /NCT ) ∗ NUC
Cont = min(A,B)/max(A,B)
Buzz = …
Prediction ! Good prediction of the viral phenomena
Class Overlapping
Experiments and results F1 using just the wording of the content!
App Buzz Cont Rais 25 words are enough to predict viral phenomena with a good F1
App 0.78
!
Machine learning framework: SVM light. App - 15.1% 4.2% 14.8%
Features: title & snippet words. PoS-tagged to Buzz 0.81
Buzz 77.0% - 3.1% 51.7% These viral phenomena are
reduce data sparseness.
Specialized datasets: each viral phenomenon, Cont 21.4% 3.0% - 48.6%
Cont 0.70 quite independent
binary classification: 50/50 pos. neg. examples. Rais 65.0% 44.6% 42.5% - Rais 0.68 It is possible to separately predict them