cascading behavior in large blog graphs patterns and a model

14
Cascading Behavior in Large Blog Graphs Patterns and a Model Leskovec et al. (SDM 2007)

Upload: taran

Post on 14-Jan-2016

34 views

Category:

Documents


0 download

DESCRIPTION

Cascading Behavior in Large Blog Graphs Patterns and a Model. Leskovec et al. (SDM 2007). Why?. Temporal Aspects How does information spread in Social Network? How does the popularity die? Linearly, exponentially, or …? Topological Aspects Do information cascades have common structures? - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Cascading Behavior in Large Blog Graphs Patterns and a Model

Cascading Behavior in Large Blog GraphsPatterns and a Model

Leskovec et al.(SDM 2007)

Page 2: Cascading Behavior in Large Blog Graphs Patterns and a Model

Why?

• Temporal Aspects– How does information spread in Social Network?– How does the popularity die? Linearly,

exponentially, or …?

• Topological Aspects– Do information cascades have common

structures?– Their properties like size distribution

Page 3: Cascading Behavior in Large Blog Graphs Patterns and a Model

Preliminaries

• Trivial vs. Non-trivial Cascades• Cascade Initiator• Stars and Chains• Connector nodes

Page 4: Cascading Behavior in Large Blog Graphs Patterns and a Model

Dataset• 21.3 million posts, 2.5 million blogs from Aug and Sep 2005• Start with most cited blog posts in Aug’05• Traversed conversations forward (inlinks) and backward

(outlinks)• Max depth = 100; max breadth = 500• Collected

– Unique post ID– Blog URL– Post Permalink– Post Date– Post Content– Post Links

Page 5: Cascading Behavior in Large Blog Graphs Patterns and a Model

Temporal Patterns

How Popularity dies?

Page 6: Cascading Behavior in Large Blog Graphs Patterns and a Model

Blog Network Topology

Popular blogs that receive lots of inlinksdoes not necessarily sprout many outlinks.

Page 7: Cascading Behavior in Large Blog Graphs Patterns and a Model

Post Network Topology

98% of the posts are isolated

Page 8: Cascading Behavior in Large Blog Graphs Patterns and a Model

Topological Patterns

Common Cascade Shapes (Gr has the frequency rank r)

97% are trivial cascades

Page 9: Cascading Behavior in Large Blog Graphs Patterns and a Model

Topological Patterns

Cascade Size Distribution

Page 10: Cascading Behavior in Large Blog Graphs Patterns and a Model

Observations

• Most cascades follow tree like structures.• Linear increase in diameter requires

exponential increase in the cascade size.• The probability that a node will be a part of a

cascade decreases with the number of cascades it is already a part of.

Page 11: Cascading Behavior in Large Blog Graphs Patterns and a Model

Generative Model

• Susceptible-Infected-Susceptible (SIS) Model

• β: “infection probability” of a post

• Blog can be either “infected” or “susceptible”

Page 12: Cascading Behavior in Large Blog Graphs Patterns and a Model

Summary

• Temporal patterns• Topological patterns• Generative model

Page 13: Cascading Behavior in Large Blog Graphs Patterns and a Model

Food for thought• Blogs are sparsely linked. Not many posts link to

the original post from which they got the content. How to study information diffusion in these scenarios?– Beyond link analysis

• Uniform infecting probability is an unrealistic assumption

• Multiple cascades initiating simultaneously• Not many study the “tipping point” in cascades• Does the cascade die its natural death or is there

some factor that affects the lifespan of a cascade

Page 14: Cascading Behavior in Large Blog Graphs Patterns and a Model

T-1 T T+1

Backward Forward

InlinkOutlink