Download presentation
Presentation is loading. Please wait.
Published byDuane Ledington Modified over 9 years ago
1
Modeling Blog Dynamics Speaker: Michaela Götz Joint work with: Jure Leskovec, Mary McGlohon, Christos Faloutsos Cornell University Carnegie Mellon University Stanford University
2
Modeling the Blogosphere Blogosphere System of interactions: ▫Entities: Bloggers, Posts, Topics, … ▫Relations: Blogger publishes post Post is about a topic Post links to post Model Simple set of rules (followed by a blogger) that creates these interactions Evaluation Creating a synthetic blogosphere and comparing it to real blogosphere Motivation Forecasting, advertising Michaela GötzModeling Blog DynamicsICWSM 2009
3
How is this different from modeling social networks? ▫2 networks combined: Blog vs. Post network B1 B4 B3 B2 B6 B7 B5 Blogosphere B1 B4 B3 B2 B6 B7 B5 2 1 1 1 1 1 1 Postnetwork Blognetwork Cascade Michaela GötzModeling Blog DynamicsICWSM 2009
4
▫2 networks combined: Blog vs. Post network ▫Complex temporal dynamics B4 B3 B2 B6 B7 B5 B1 How is this different from modeling social networks? Michaela GötzModeling Blog DynamicsICWSM 2009
5
▫2 networks combined: Blog vs. Post network ▫Complex temporal dynamics B1 B4 B3 B2 B6 B7 B5 B8 How is this different from modeling social networks? Our goal: Model micro-level interactions to create the macro-level patterns (both the temporal and topological) of the blogosphere Michaela GötzModeling Blog DynamicsICWSM 2009
6
Models – Related Work Michaela GötzModeling Blog DynamicsICWSM 2009 Realistic Topological PatternsRealistic Dynamical Patterns Links (SVM) [Adar and Adamic] Blog Mortality [Venolia] Links (Epidemiological Model) [Gruhl et al.] Links (Epidemiological Model) [Leskovec et al.] Inter-arrival time of emails (Markov Chain) [Kleinberg] Links (Preferential Attachment) [Karandikar et al.] Time between answering two consecutive emails (PA) [Barabasi; Vazquez] Blog Email
7
What are the properties of the blogosphere? Topological properties ▫Blog network ▫Post network Temporal properties ▫User posting activity ▫Popularity of posts over time (link creation) Michaela GötzModeling Blog DynamicsICWSM 2009
8
Some Topological Patterns Michaela GötzModeling Blog DynamicsICWSM 2009 Power law: [Leskovec et al.] Blog In-Degree Post In-Degree Cascade Size
9
Michaela GötzModeling Blog DynamicsICWSM 2009 Burstiness & Self-Similarity: Measure: Entropy at various resolutions Plot: Burstiness & Self-Similarity: Measure: Entropy at various resolutions Plot: Self-Similarity -> Linearity Burstiness -> Slope < 1 Self-Similarity -> Linearity Burstiness -> Slope < 1 Blogosphere – Some Temporal Patterns Time vs. Count Example 1 Aggregation Level vs. Entropy Example 2
10
Blogosphere – Some Temporal Patterns Michaela GötzModeling Blog DynamicsICWSM 2009 [McGlohon et al.] Inter-Posting Time Popularity: time t vs. #in-links received t days after publishing Burstiness of Posting Activity Burstiness & Self-Similarity: Measure: Entropy at various resolutions Plot: Burstiness & Self-Similarity: Measure: Entropy at various resolutions Plot:
11
Blogosphere - Dynamics B1 B4 B3 B2 B6 B7 B5 Michaela GötzModeling Blog DynamicsICWSM 2009 B8 WANTED: Model - Simple (no parameters) - Intuitive (local rules) - Creating Realistic Topology and Dynamics WANTED: Model - Simple (no parameters) - Intuitive (local rules) - Creating Realistic Topology and Dynamics
12
First try solution Let’s assume: ▫ Inter-posting times are sampled from exponential distribution ▫Links are created using Preferential Attachment But it won’t work: Inter-Posting Time Burstiness of Posting Activity Blog In-Degree Michaela GötzModeling Blog DynamicsICWSM 2009
13
Michaela GötzModeling Blog DynamicsICWSM 2009 Our Model ZC In Every Round For Every Blog In Every Round For Every Blog post P no link create link choose neighbor B choose neighbor B choose non-neighbor B choose non-neighbor B choose post Q of B choose post Q of B link to random posts upward in the cascade link to random posts upward in the cascade “random walk” …… “explore”“exploit” “link expansion” 012-2 1/2
14
Michaela GötzModeling Blog DynamicsICWSM 2009 Our Model ZC In Every Round For Every Blog In Every Round For Every Blog post P no link create link choose neighbor B choose neighbor B choose non-neighbor B choose non-neighbor B choose post Q of B choose post Q of B link to random posts upward in the cascade link to random posts upward in the cascade “random walk” …… “explore”“exploit” “link expansion” 012-2 1/2 B3 B2 B1 0 0 10 -5 -6
15
Michaela GötzModeling Blog DynamicsICWSM 2009 Our Model ZC In Every Round For Every Blog In Every Round For Every Blog post P no link create link choose neighbor B choose neighbor B choose non-neighbor B choose non-neighbor B choose post Q of B choose post Q of B link to random posts upward in the cascade link to random posts upward in the cascade “random walk” …… “explore”“exploit” “link expansion” 012-2 1/2 B3 B2 B1 0 0 10 -6
16
Michaela GötzModeling Blog DynamicsICWSM 2009 Our Model ZC In Every Round For Every Blog In Every Round For Every Blog post P no link create link choose neighbor B choose neighbor B choose non-neighbor B choose non-neighbor B choose post Q of B choose post Q of B link to random posts upward in the cascade link to random posts upward in the cascade “random walk” …… “explore”“exploit” “link expansion” 012-2 1/2 B3 B2 B1 0 0 10 -6 ?
17
Michaela GötzModeling Blog DynamicsICWSM 2009 Our Model ZC In Every Round For Every Blog In Every Round For Every Blog post P no link create link choose neighbor B choose neighbor B choose non-neighbor B choose non-neighbor B choose post Q of B choose post Q of B link to random posts upward in the cascade link to random posts upward in the cascade “random walk” …… “explore”“exploit” “link expansion” 012-2 1/2 B3 B2 B1 0 0 10 -6
18
Michaela GötzModeling Blog DynamicsICWSM 2009 Our Model ZC In Every Round For Every Blog In Every Round For Every Blog post P no link create link choose neighbor B choose neighbor B choose non-neighbor B choose non-neighbor B choose post Q of B choose post Q of B link to random posts upward in the cascade link to random posts upward in the cascade “random walk” …… “explore”“exploit” “link expansion” 012-2 1/2 B3 B2 B1 0 0 10 -6
19
Michaela GötzModeling Blog DynamicsICWSM 2009 Our Model ZC In Every Round For Every Blog In Every Round For Every Blog post P no link create link choose neighbor B choose neighbor B choose non-neighbor B choose non-neighbor B choose post Q of B choose post Q of B link to random posts upward in the cascade link to random posts upward in the cascade “random walk” …… “explore”“exploit” “link expansion” 012-2 1/2 B3 B2 B1 0 0 10 -6
20
Michaela GötzModeling Blog DynamicsICWSM 2009 post P no link create link choose neighbor B choose neighbor B choose non-neighbor B choose non-neighbor B choose post Q of B choose post Q of B link to random posts upward in the cascade link to random posts upward in the cascade “random walk” …… “explore”“exploit” “link expansion” 012-2 1/2 Our Model ZC Properties - Simple (no parameters) - Intuitive (local rules) Properties - Simple (no parameters) - Intuitive (local rules) Theorem: The interposting time of ZC follows a power- law with exponent -1.5 Theorem: The interposting time of ZC follows a power- law with exponent -1.5 Theorem: The posting activity is self-similar and bursty.
21
Michaela GötzModeling Blog DynamicsICWSM 2009 Our Model ZC – Experimental Evaluation Inter-Posting TimeBurstiness of Posting Activity Real Data Simulation Popularity over time 45k blogs, 2.2 million posts [Glance et al.] 45k blogs, 2.2 million posts
22
Michaela GötzModeling Blog DynamicsICWSM 2009 Post In-Degree Cascade Size Our Model ZC – Experimental Evaluation Blog In-Degree Real Data Simulation
23
Michaela GötzModeling Blog DynamicsICWSM 2009 Conclusions Model ZC of the Blogosphere - Simple (no parameters) - Intuitive (local rules) - Creating Realistic Blogosphere: - evaluated on 3 topological patterns - evaluated on 3 temporal patterns - Useful for forecasting and to explore dynamics for advertising Model ZC of the Blogosphere - Simple (no parameters) - Intuitive (local rules) - Creating Realistic Blogosphere: - evaluated on 3 topological patterns - evaluated on 3 temporal patterns - Useful for forecasting and to explore dynamics for advertising post P no link create link choose neighbor B choose neighbor B choose non-neighbor B choose non-neighbor B choose post Q of B choose post Q of B link to random posts upward in the cascade link to random posts upward in the cascade “random walk” …… “explore”“exploit” “link expansion” 012-2 1/2
24
Michaela GötzModeling Blog DynamicsICWSM 2009 Thanks! post P no link create link choose neighbor B choose neighbor B choose non-neighbor B choose non-neighbor B choose post Q of B choose post Q of B link to random posts upward in the cascade link to random posts upward in the cascade “random walk” …… “explore”“exploit” “link expansion” 012-2 1/2
Similar presentations
© 2024 SlidePlayer.com Inc.
All rights reserved.