SLIDE 1
Towards a Data-driven Approach for Agent-Based Modelling
2
Towards a Data-driven Approach for Agent-Based Modelling: - - PowerPoint PPT Presentation
Towards a Data-driven Approach for Agent-Based Modelling: - - PowerPoint PPT Presentation
Towards a Data-driven Approach for Agent-Based Modelling: Simulating Spanish Postmodernisation Defended by Samer Hassan Collado Directed by Juan Pavn Mestras Milln Arroyo Menndez Contents Frame the problem Define a methodology
SLIDE 2
SLIDE 3
Towards a Data-driven Approach for Agent-Based Modelling
3
Introduction
SLIDE 4
Towards a Data-driven Approach for Agent-Based Modelling
4
Agent-Based Social Simulation
Simulation of Complex Systems
Nonlinearity: equations? Self-organisation Emergence: Total > ∑ parts Non-deterministic
Social Simulation
Simulation of social phenomena
Agent-Based Social Simulation
Agent ~ Individual Different from classical MAS Simple micro rules => Complex macro behaviour
Artificial Intelligence, Engineering, Physics, Mathematics, Philosophy, Ecology, Economics, Sociology, Anthropology, Political Science, Psychology, Cognitive Science
SLIDE 5
Towards a Data-driven Approach for Agent-Based Modelling
5
Agent-Based Social Simulation
SLIDE 6
Towards a Data-driven Approach for Agent-Based Modelling
6
Agent-Based Social Simulation
SLIDE 7
Towards a Data-driven Approach for Agent-Based Modelling
7
Objectives
Build a methodology for DDABM Design an agent framework Develop a data-driven case study
Validates framework & methodology Supports theoretical hypotheses
Applications of AI in ABM
Agents + Fuzzy Logic + NLP + DM
SLIDE 8
Towards a Data-driven Approach for Agent-Based Modelling
8
Methodological Approach
SLIDE 9
Towards a Data-driven Approach for Agent-Based Modelling
9
Methodological Approach
Review Research Aim
Theoretical Data-driven
Proposed Data-driven Approach
Stress on Data Deepening KISS Data-driven Cycle
SLIDE 10
Towards a Data-driven Approach for Agent-Based Modelling
10
Classical Logic of Simulation
Gilbert (1999)
SLIDE 11
11
Methodological Approaches
Positivism
Goldspink (2002) McKelvey (1999)
SLIDE 12
12
Methodological Approaches
Positivism Social Sciences
Goldspink (2002) McKelvey (1999)
SLIDE 13
13
Methodological Approaches
Positivism Social Sciences Social Simulation
Goldspink (2002) McKelvey (1999)
SLIDE 14
Towards a Data-driven Approach for Agent-Based Modelling
14
Research Aim
Theoretical KISS Structural Validation Abstract General
SLIDE 15
Towards a Data-driven Approach for Agent-Based Modelling
15
Research Aim
Data-driven Non-KISS Empirical Validation Specific (case study) Expressive Theoretical KISS Structural Validation Abstract General
SLIDE 16
Towards a Data-driven Approach for Agent-Based Modelling
16
Research Aim
Data-driven Non-KISS Empirical Validation Specific (case study) Expressive Theoretical KISS Structural Validation Abstract General
SLIDE 17
Towards a Data-driven Approach for Agent-Based Modelling
17
Classical Logic of Simulation
SLIDE 18
Towards a Data-driven Approach for Agent-Based Modelling
18
Proposed Data-Driven Logic
SLIDE 19
Towards a Data-driven Approach for Agent-Based Modelling
19
Data-driven Approach
Empowering Empirical Data
Data Collection => Exhaustive Random Initialisation => Empirical Initialisation Abstraction => Data-driven Design Structural Validation => Empirical Validation
SLIDE 20
Towards a Data-driven Approach for Agent-Based Modelling
20
Data-driven Approach
Deepening KISS
Gradually increasing complexity Exploration of the model space Antunes et al (2006)
SLIDE 21
21
Proposed Data-driven Cycle
SLIDE 22
22
Proposed Data-driven Cycle
SLIDE 23
23
Data Collection
SLIDE 24
24
Proposed Data-driven Cycle
SLIDE 25
25
Data-driven Design
SLIDE 26
Towards a Data-driven Approach for Agent-Based Modelling
26
Methodology Highlights
Stress on Data
How to handle it
Deepening KISS
Guides agent framework architecture Guides model extension
Data-driven Cycle
Guides ABM building
SLIDE 27
Towards a Data-driven Approach for Agent-Based Modelling
27
The Sociological Problem
SLIDE 28
Towards a Data-driven Approach for Agent-Based Modelling
28
The Sociological Problem
SLIDE 29
Towards a Data-driven Approach for Agent-Based Modelling
29
Social Phenomena
SLIDE 30
Towards a Data-driven Approach for Agent-Based Modelling
30
Case Study
Sociological problem
Change in social values
- Mainly: political & religious
Specific context
- Time: period 1980-2000
- Space: Spanish society
Validating
Theory from Sociology The Data-driven methodology The Agent framework Arroyo (2004, 2005)
SLIDE 31
Towards a Data-driven Approach for Agent-Based Modelling
31
Sociological Theory
SLIDE 32
Towards a Data-driven Approach for Agent-Based Modelling
32
The Change of Values
Cultural Modernisation
Individualisation
- Discourages Authorities
- Encourages Self-expression
Post-Materialism
- Based on Maslow hierarchy
- Material welfare => Post-Mat. Priorities
Dynamics
Inter-generational Intra-generational Inglehart (1997) Halman (1994) Maslow (1987)
SLIDE 33
Towards a Data-driven Approach for Agent-Based Modelling
33
The Change of Values
Inter-generational dynamics
Change across generations Demography
Intra-generational dynamics
Change within a generation Internal evolution in a person course of life
Inglehart
Socialisation in youth Stability over course of life Inter-gen >> Intra-gen Inglehart (1997) Halman (1994) Maslow (1987)
SLIDE 34
Towards a Data-driven Approach for Agent-Based Modelling
34
Inglehart’s hypothesis
Inter-generational dynamics
Change across generations Demography
Intra-generational dynamics
Change within a generation Internal evolution in a person course of life
Inglehart
Socialisation in youth Stability over course of life Inter-gen >> Intra-gen
SLIDE 35
Towards a Data-driven Approach for Agent-Based Modelling
35
Axiomatic Base
Model for the change of values
Focus: political & religious values Demography
- Intergenerational dynamics
Socialisation in youth
- Stability over course of life
SLIDE 36
Towards a Data-driven Approach for Agent-Based Modelling
36
Modelling Decisions
SLIDE 37
Towards a Data-driven Approach for Agent-Based Modelling
37
Theoretical Modelling Decisions
Demography
Static:
- Population pyramids
Dynamics:
- Reproduction <-- Partner
- Partner <-- Friendship
- Friendship <-- Social Network
Socialisation
Inheritance of crossed values Stability in course of life
SLIDE 38
Towards a Data-driven Approach for Agent-Based Modelling
38
Methodological Modelling Decisions
Empowering Empirical Data
Exhaustive Data Collection Empirical Initialisation Data-driven Design Empirical Validation
SLIDE 39
Towards a Data-driven Approach for Agent-Based Modelling
39
Methodological Modelling Decisions
Empowering Empirical Data
Exhaustive Data Collection
- Multiple data sources & waves
Empirical Initialisation
- European Values Study
- Representative sample: EVS-1980
Data-driven Design
- Demography
- Equations (e.g. birth rate)
- Qualitative info from domain expert
- Life cycle, Social relations
Empirical Validation
- EVS-1990, EVS-1999
SLIDE 40
Towards a Data-driven Approach for Agent-Based Modelling
40
Model Insight
SLIDE 41
Towards a Data-driven Approach for Agent-Based Modelling
41
Mentat: Architecture
SLIDE 42
Towards a Data-driven Approach for Agent-Based Modelling
42
Mentat: Architecture
Methodological implications Agent Framework design
Global pattern: MVC Agent pattern: Layers
Model description
World Agent
SLIDE 43
Towards a Data-driven Approach for Agent-Based Modelling
43
Mentat: Architecture
‘Deepening KISS’ approach
Gradually increasing complexity
- Modularity & Flexibility
Exploration of the model space
- Activating independent modules
- Facilitate extension
- Architectural patterns
Architectural pattern
Fundamental structural organization schema Subsystems, responsibilities, interrelations
SLIDE 44
44
Mentat Architecture: MVC
SLIDE 45
Towards a Data-driven Approach for Agent-Based Modelling
45
Agent Architecture: Layers
SLIDE 46
Towards a Data-driven Approach for Agent-Based Modelling
46
Agent Architecture: Layers
SLIDE 47
Towards a Data-driven Approach for Agent-Based Modelling
47
Agent Architecture: Layers
SLIDE 48
Towards a Data-driven Approach for Agent-Based Modelling
48
Agent Architecture: Layers
SLIDE 49
49
Agent Architecture: Layers
SLIDE 50
Towards a Data-driven Approach for Agent-Based Modelling
50
Mentat Model
World:
3000 agents Grid 100x100 6 indep. parameters
Social Network:
Communication with Neighbourhood Friend Network Family Network
SLIDE 51
Towards a Data-driven Approach for Agent-Based Modelling
51
Mentat Agent
General characteristics
Gender, Age Education, Economy Religiosity, Political Ideology
Tolerance levels
Euthanasia, Suicide, Homosexuality, Abortion,
Divorce, etc
Life cycle
Child, Adult, Elder
SLIDE 52
Towards a Data-driven Approach for Agent-Based Modelling
52
Mentat Agent: Life Cycle
SLIDE 53
Towards a Data-driven Approach for Agent-Based Modelling
53
Mentat: Social Dynamics
SLIDE 54
Towards a Data-driven Approach for Agent-Based Modelling
54
Mentat: Social Dynamics
Neighbourhood Social Network Friendship Dynamics
Meeting & Mating Fuzzification
SLIDE 55
55
Neighbourhood
Von Neumann Moore Extended Moore
SLIDE 56
Towards a Data-driven Approach for Agent-Based Modelling
56
Social Network
SLIDE 57
Towards a Data-driven Approach for Agent-Based Modelling
57
Social Network
SLIDE 58
Towards a Data-driven Approach for Agent-Based Modelling
58
Social Network
SLIDE 59
Towards a Data-driven Approach for Agent-Based Modelling
59
Social Network
SLIDE 60
Towards a Data-driven Approach for Agent-Based Modelling
60
Social Network
SLIDE 61
Towards a Data-driven Approach for Agent-Based Modelling
61
Social Network
SLIDE 62
Towards a Data-driven Approach for Agent-Based Modelling
62
Social Network
SLIDE 63
Towards a Data-driven Approach for Agent-Based Modelling
63
Social Network
SLIDE 64
Towards a Data-driven Approach for Agent-Based Modelling
64
Understanding Friendship Dynamics
“Meeting” & “Mating”: strangers => acquaintances => friends => partner “Meeting”: depends on opportunities
space & time
“Mating”: depends on opportunities & attraction Proximity principle: ‘the more similar two individuals
are, the stronger their chances of becoming friends’
Features channel individual preferences Homogeneous friendship choices Huckfeldt (1983) Verbrugge (1977)
SLIDE 65
Towards a Data-driven Approach for Agent-Based Modelling
65
Mentat Social Dynamics
Meeting
Agents randomly distributed in space
Mating
Similarity operator => Friendship
Matchmaking
Partner chosen among “candidates” Couples should be similar
SLIDE 66
Towards a Data-driven Approach for Agent-Based Modelling
66
Be Fuzzy, my Friend
Similar, Friend: fuzzy concepts Social sciences
Uncertain & vague knowledge Qualitative concepts
- Natural Language
Fuzzification improves:
Accuracy of similarity Behaviour of friendship Quality of couples
SLIDE 67
Towards a Data-driven Approach for Agent-Based Modelling
67
Be Fuzzy, my Friend
Attributes Similarity Emergence of friendship
SLIDE 68
68
Be Fuzzy, my Friend
Boolean Friendship? =>Fuzzy relationship Static Friendship? =>Evolution function
Hedstrom (2005)
SLIDE 69
Towards a Data-driven Approach for Agent-Based Modelling
69
Mentat: Results
SLIDE 70
Towards a Data-driven Approach for Agent-Based Modelling
70
Mentat: Results
Review Results
Graphical Quantitative Theoretical
Conclusions
SLIDE 71
Towards a Data-driven Approach for Agent-Based Modelling
71
Reviewing Mentat
Agent
Attributes Life Cycle
Social Behaviour
Communication Similarity Acquaintances Friends
Reproduction
Partner Finding (Matchmaking) Children Inheritance (& Socialisation)
Social Dynamics
Friendship network Family network
Demography Dynamics Demographic Equations
Life expectancy Birth rate
SLIDE 72
72
Graphical Output
SLIDE 73
73
Results
EVS Mentat 1981 1990 1999 1981 1990 1999 Gender Men 49% 47% 49% 49% 48% 49% Women 51% 53% 51% 51% 52% 51% Age (mean) 45 43 46 45 47 49 %65+ 16% 18% 21% 15% 19% 24% %Single (100w.up) 28% 29% 29% 100% 42% 35% %Single (500w.up) 100% 34% 30% %Single (1000w.up) 100% 29% 28% Population Growth (100w.up) +8% +7.2% Population Growth (500w.up) +8.6% Population Growth (1000w.up) +10% Ideology Left 29% 33% 31% 29% 33% 36% Centre 18% 19% 23% 18% 18% 17% N/A 30% 25% 24% 30% 29% 27% Right 22% 23% 21% 23% 22% 20% Religiosity Ecclesiastical 33% 25% 22% 33% 29% 25% Low-Intensity 22% 26% 23% 22% 23% 22% Alternative 14% 17% 19% 14% 16% 16% Non-religious 31% 32% 35% 31% 34% 37% Tolerance Abortion 2,89 4,13 4,34 2,89 3,08 3,3 Divorce 4,79 5,51 6,1 4,79 5,13 5,4 Euthanasia 3,18 3,97 4,73 3,18 3,43 3,6 Suicide 2,26 2,25 2,77 2,26 2,36 2,5
SLIDE 74
Towards a Data-driven Approach for Agent-Based Modelling
74
Quantitative Results
Warming-up stage
Increases network cohesion (% singles) Increases population growth
Ideology
1980-1990 correct trend 1990-2000 does not twist
Religiosity
Similar trends in all indicators More affected by inter-generational dynamics
Tolerance
Too moderated evolution More affected by intra-generational dynamics
SLIDE 75
Towards a Data-driven Approach for Agent-Based Modelling
75
Theoretical Results
Support for Inglehart’s Axiomatic Base
Socialisation in youth Stability over course of life Demographic evolution Inter-gen >> Intra-gen
Key importance of Demographic Dynamics
Change across generations Based on the social network
SLIDE 76
Towards a Data-driven Approach for Agent-Based Modelling
76
Conclusions
Defined methodology
Data-driven ABM Contexts of abundant data Design, Initialisation, Validation Gradual increase of complexity
Applied in agent framework
Architecture following methodology Facilitates exploration Applicable in a family of cases
SLIDE 77
Towards a Data-driven Approach for Agent-Based Modelling
77
Conclusions
Sociological case study
Data-driven model
- Surveys
- Complementary Q&Q data
Validating
- methodology & framework
Support for theory
- Inglehart’s hypothesis
AI technologies
Agents, Fuzzy Logic Natural Language Processing, Data Mining
SLIDE 78
Towards a Data-driven Approach for Agent-Based Modelling
78
Open Research Lines
SLIDE 79
Towards a Data-driven Approach for Agent-Based Modelling
79
Open Research Lines
Modular framework
Activating modules for exploration
Exploring application of AI:
Fuzzy logic: qualitative input NLP: qualitative output DM: clustering input & output
SLIDE 80
80
Biography of representative individuals
Complementary output in natural language Events tracing -> XML -> NL Life-story of agents
- e.g. hyper-inflation
Introducing NLP
SLIDE 81
81
Quantitative & Qualitative Output Generation
Life-Story of Representative Individual (ideal type) Analysis , Filtering and NLG Macro Trends Micro processes : interactions Quantitative Statistics and Graphs European Values Survey Simple Filtering
Content Determination: Complex filtering based
- n rules
Discourse Planning : Ordering for a coherent story Surface Realization : Natural Language Generation Events log (XML)
SLIDE 82
Towards a Data-driven Approach for Agent-Based Modelling
82
Introducing Data Mining
Extracting patterns from large amounts of data Pre-processing of empirical data:
Selection of representative clusters
Post-processing of simulation output:
Show non-visible patterns Cluster Validation
SLIDE 83
Towards a Data-driven Approach for Agent-Based Modelling
83
Publications
SLIDE 84
Towards a Data-driven Approach for Agent-Based Modelling
84
Publications describing the ABM
Samer Hassan, Luis Antunes, and Juan Pavón. Mentat: A Data-Driven
Agent-Based simulation of social values evolution. In: Multi-Agent- Based Simulation X, Revised selected papers, Lecture Notes in Artificial Intelligence, Springer-Verlag (2009). ISSN 0302-9743 (To appear)
SLIDE 85
Towards a Data-driven Approach for Agent-Based Modelling
85
Publications about Methodological issues
Samer Hassan, Luis Antunes, and Millan Arroyo. Deepening the demographic mechanisms In: a Data-Driven social simulation of moral values evolution. Multi- Agent-Based simulation IX, 5269:167–182, 2009. ISSN 0302-9743.
Samer Hassan, Rubén Fuentes-Fernández, José M. Galán, and Adolfo López-Paredes. Reducing the modeling gap: On the use of metamodels in Agent-Based simulation. In: Proceedings of the Sixth Conference of the European Social Simulation Association (ESSA09). Guildford, UK, 2009. (To appear).
Samer Hassan, Juan Pavón, Luis Antunes, and Nigel Gilbert. Injecting data into Agent- Based simulation. In: Keiki Takadama, Guillaume Deffuant, and Claudio Cioffi-Revilla, editors, The Second World Congress on Social Simulation, Selected papers from the 2nd World Congress on Social Simulation. Springer Series on Agent Based Social Systems. Springer, Washington, D.C., 2009. (To appear)
Juan Pavón, Millán Arroyo, Samer Hassan, and Candelaria Sansores. Agent-based modelling and simulation for the analysis of social patterns. Pattern Recognition Letters, 29(8):1039–1048, 2008. ISSN 0167-8655.
Samer Hassan, Luis Antunes, Juan Pavón, and Nigel Gilbert. Stepping on earth: A roadmap for data-driven Agent-Based modelling. In: Proceedings of the Fifth Conference of the European Social Simulation Association (ESSA08). Brescia, Italy, 2008.
Samer Hassan, Juan Pavón, and Nigel Gilbert. Injecting data into simulation: Can agent- based modelling learn from microsimulation? In: Proceedings of the 2nd World Congress
- f Social Simulation (WCSS 2008). Washington, D.C., 2008.
SLIDE 86
Towards a Data-driven Approach for Agent-Based Modelling
86
Publications about Fuzzy Agents
Samer Hassan, Luis Garmendia, and Juan Pavón. Introducing
uncertainty into social simulation: Using fuzzy logic for Agent-Based
- modelling. International Journal of Reasoning-based Intelligent Systems,
- 2009. ISSN 1755-0556 (To appear).
Samer Hassan, Mauricio Salgado, and Juan Pavón. Friends forever:
Social relationships with a fuzzy Agent–Based model. Hybrid Artificial Intelligence Systems, Selection from the Third International Workshop, HAIS 2008, 5271:523–532, 2008. ISSN 0302-9743.
Samer Hassan, Luis Garmendia, and Juan Pavón. Agent-Based social
modeling and simulation with fuzzy sets. Advances In: Soft Computing (Springer-Verlag), Selection from the conference 2nd International Workshop of Hybrid Artificial Intelligence Systems (HAIS07), (44):40–47, 2007. ISSN 1615-3871.
SLIDE 87
Towards a Data-driven Approach for Agent-Based Modelling
87
Publications about Sociology
Millán Arroyo and Samer Hassan. Marco teórico-sociológico y
- perativización para modelar un sistema multi-agente sobre la
evolución de la religiosidad española. In: Francisco J. Miguel, editor, Proceedings of the 2nd Workshop on Social Simulation and Artificial Societies Analysis (SSASA’08), volume 442, page 8. CEUR Workshop Proceedings, Barcelona, 2009. ISSN 1613-0073.
Millán Arroyo-Menéndez and Samer Hassan Collado. Simulación de
procesos sociales basada en agentes software. Empiria - Revista de metodología de ciencias sociales, (14):139–161, 2007. ISSN 1139-5737.
Millán Arroyo and Samer Hassan. Simulación de procesos sociales
basada en agentes software. In: Actas del IX Congreso Español de Sociología. Barcelona, 2007. Grupo de trabajo I: Metodología.
SLIDE 88
Towards a Data-driven Approach for Agent-Based Modelling
88
Publications about NLP & DM
Samer Hassan, Celia Gutiérrez, and Javier Arroyo. Re-thinking modelling: a call for the use of data mining in data-driven social simulation. In: Proceedings of the 1st Workshop on Social Simulation at IJCAI 2009. Pasadena, CA, 2009. (To appear).
Samer Hassan, Carlos León, Pablo Gervás, and Raquel Hervas. A computer model that generates biography-like narratives. In: A. Cardoso and G. A. Wiggins, editors, Proceedings of the 4th International Joint Workshop on Computational Creativity, pages 5–12. London, 2007.
Samer Hassan, Juan Pavón, Millán Arroyo, and Carlos León. Agent based simulation framework for quantitative and qualitative social research: Statistics and natural language generation. In: F. Amblard, editor, Proceedings
- f the Fourth Conference of the European Social Simulation Association
(ESSA07), pages 697—-707. Toulouse, France, 2007. ISBN 978-2-9520326-7-4.
Carlos León, Samer Hassan, Pablo Gervás, and Juan Pavón. Mixed narrative and dialog content planning based on BDI agents. Lecture Notes in Artificial Intelligence, from the Lecture Notes In: Computer Science Series (Springer- Verlag), (4788):150–159, 2007. ISSN 0302-9743. Selected papers from 12th Conference of the Spanish Association for Atificial Intelligence, CAEPIA 2007.
SLIDE 89
89