Download presentation
Presentation is loading. Please wait.
Published byCory Williams Modified over 6 years ago
1
Extract Agent-based Model from Communication Network
Hung-Ching (Justin) Chen Matthew Francisco Malik Magdon-Ismail Mark Goldberg William Wallance RPI
2
Given a society’s communication history,
Goal Given a society’s communication history, can we: Deduce something about “nature” of the society: e.g., Do actors generally have a propensity to join small groups or large groups? Predict the society’s future: e.g., How many social groups are there after 3 months? e.g., What is the distribution of group size?
3
General Approach Individual “Learn” Behavior Society’s (Micro-Laws)
History “Predict” (Simulate) Society’s Future
4
General Approach Individual “Learn” Behavior Society’s (Micro-Laws)
History “Learn” Individual Behavior (Micro-Laws) “Predict” (Simulate) Society’s Future
5
Social Networks Individuals (Actors) 1 2 3 Groups
6
Social Networks Individuals (Actors) 1 2 - Join - Leave Groups 3
7
Social Networks Individuals (Actors) - Join Groups - Leave - Disappear
4 1 - Join - Leave 2 Groups - Disappear - Appear - Re-appear 3
8
Society’s History
9
General Approach Individual “Learn” Behavior Society’s (Micro-Laws)
History “Predict” (Simulate) Society’s Future
10
Join / Leave / Do Nothing
Modeling of Dynamics Parameters History Groups & Individuals Join / Leave / Do Nothing Micro-Law # 1 # 2 # N … Actions
11
Actor X likes to join groups.
Example of Micro-Law Actor X likes to join groups. SMALL LARGE Parameter
12
ViSAGE Virtual Simulation and Analysis of Group Evolution
State: Properties of Actors and Groups State Decide Actors’ Action Normative Action State State update Actor Choice State Feedback to Actors Process Actors’ Action Real Action
13
General Approach Individual “Learn” Behavior Society’s (Micro-Laws)
History “Predict” (Simulate) Society’s Future
14
Learning ? ? Parameters #1 in Micro-Laws Learn Parameters #2 in
Communications Parameters #1 in Micro-Laws ? Learn Parameters #2 in Micro-Laws ?
15
Groups & Group Evolution
Communications Groups Evolution Groups: Overlapping clustering Group evolution: Matching
16
Actor’s Types Leader: prefer small group size and is most ambitious
Socialite: prefer medium group size and is medium ambitious Follower: prefer large group size and is least ambitious
17
Learning Actors’ Type Maximum log-likelihood learning algorithm
Cluster algorithm EM algorithm
18
Testing Simulation Data
19
Testing Real Data Cluster Algorithm EM Algorithm Learned Actors’ Types
Leader Socialite Follower Number of Actor 822 550 156 Percentage 53.8% 36.0% 10.2% EM Algorithm Learned Actors’ Types Leader Socialite Follower Number of Actor 532 368 628 Percentage 34.8% 24.1% 41.1%
20
General Approach Individual “Learn” Behavior Society’s (Micro-Laws)
History “Predict” (Simulate) Society’s Future
21
Testing & Simulations Micro-Laws & Parameters # 1 Simulate # 2
22
Prediction
23
Prediction
24
Future Work Test Other Predictions
e.g., membership in a particular group Learn from Other Real Data e.g., s and blogs
25
Questions?
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.