Agent-Based Social Modelling and Simulation with Fuzzy Sets Samer Hassan Collado Luis Garmendia Salvador Juan Pavón Mestras ESSA 2007 Dep. Ingeniería del Software e Inteligencia Artificial Acknowledgments. This work has been developed with support of the project TIN C03-01, funded by the Spanish Council for Science and Technology.
Samer Hassan HAIS Index Why can the fuzzy logic be useful for Agent- Based Social Simulation? The case under study is a complex sociological problem: the evolution of values in the Spanish post-modern society Fuzzification of ABSS, step by step Results a system that approaches more to reality
Samer Hassan HAIS Why Fuzzy Logic? The simulation of Multi-Agent Systems (MAS) is a powerful technique for studying complex systems behaviour Social Simulation allows the observation of emergent behaviour of a system of agents/individuals Limitation? when considering the evolution of complex mental entities, such as human believes and values Social sciences are characterized by uncertain and vague knowledge The fuzzy semantic predicates can determine this type of knowledge
Samer Hassan HAIS Why Fuzzy Logic? In the case study: European Value Survey, World Value Survey Questions about the degree of happiness, satisfaction in aspects of life, or trust in several institutions (“Very much” “Partially”…) Fuzzy logic can be applied to model different aspects of the MAS
Samer Hassan HAIS Case study Objective: to simulate the process of change in values in a period: in a society: Spanish A problem with many factors involved: Ideology, Economy, Demography, Values, Relationships, Inheritance… many of them uncertain or diffuse Far from the typical industrial applications of ABSS that require software engineers: task- driven agents, clear defined rules… Input Data: EVS
Samer Hassan HAIS Design of the MAS model Agent/Individual: From EVS Agent MS atts: ideology, religiosity, economic class, age, sex… Different behaviour while life cycle: youth, adult, old Demographic micro- evolution: couples, reproduction, inheritance World: Demographic model Network relationships: Friends groups Relatives
Samer Hassan HAIS MAS system Hundreds of agents in continuous interaction Real-time graphics that show system evolution
Samer Hassan HAIS Fuzzifying the MAS: Relationships Friendship: it’s unrealistic just “to be” or “not to be” friends. Friendships is defined as a fuzzy relationship with real values between 0 and 1: R friend : UxU [0,1] Immediate effect: distinguishing between “close friends” and “known people” The same process could be done to family
Samer Hassan HAIS Fuzzifying the MAS: fuzzy characteristics For fuzzy operations, it is needed to define fuzzy sets over the agents' characteristics/variables Defining fuzzy sets over these variables: i.e. religious : U [0,1] religious (ind)= 0.2 means that “ind” is mainly not religious For instance, for age can be defined several fuzzy sets: Old AdultYouthAge
Samer Hassan HAIS Fuzzifying the MAS: Similarity Similarity operation: rates how similar two agents are, based on their characteristics In the MAS is used for: Finding possible friends Choosing couple Fuzzified as OWA (weighted aggregation) of similarities of attribute fuzzy sets: R similarity (Ind, Ind2)= OWA ( att_i defined, N( att_i (Ind)- att_i (Ind2)))
Samer Hassan HAIS Fuzzifying the MAS: Couple Choosing couple is highly improved: Now, we can know how “compatible” are two agents: R compatible (Ind, Ind2) := OWA ( R friend (Ind, Ind2), R similarity (Ind, Ind2) ) R couple (Ind, Ind2) := Adult(Ind) AND Ind2 = Max R compatible ( Ind,{ Ind i Friends(Ind) where: R couple (Ind i ) == false AND Sex(Ind) Sex(Ind i ) AND Adult(Ind i ) } )
Samer Hassan HAIS Fuzzifying the MAS: other aspects Many other points where fuzzy logic can be applied Local influence is a “fuzzy concept”: how much an agent influences its friends and family Inheritance between generations: composition of parents variables (with random mutation factor): X attribute of Ind, x (Ind) = x (Father (Ind)) o x (Mother (Ind)) Fuzzy states can be implemented for smoother agents behaviour
Samer Hassan HAIS Extracting knowledge with fuzzy logic Fuzzy transitive property in friendship works: “the friend of my friend is somehow my friend” But how much is that “somehow”? Having friend(A,B)=0.4, friend(B,C)=0.6 friend(A,C)= Min(0.4, 0.6)= 0.4 friend(A,C)= Prod(0.4, 0.6)= 0.24 friend(A,C)= Lw(0.4, 0.6)= max(0, a+b-1)=0
Samer Hassan HAIS Extracting knowledge with fuzzy logic The T-transitive closure is a fuzzy operation that applies consecutively the transitive property In the case of friendship it can be applied to know how friends are all the non-connected agents. In friendship, T should be “Prod” Other powerful possibilities for extracting knowledge: inference with rules, fuzzy implications, or fuzzy compositions
Samer Hassan HAIS Application and Results Implementation of some of these fuzzy applications has been done over the MAS studied: Fuzzification of friendship Fuzzy sets over attributes New fuzzy similarity New matchmaking, that produced a great improvement in the micro aspect of finding couples T-transitive closure, with its consequent extraction of knowledge (agents know more people, with grading)
Samer Hassan HAIS For application in other contexts The example has shown how to fuzzify relations that determine agents’ interactions Agents’ attributes can be defined in terms of fuzzy sets Context-dependant functions, like inheritance, can be modelled as well as a typical fuzzy similarity operation Life states of agents are frequent in systems that evolve over time, especially in task solving environments A global fuzzy operation over all the agents was defined on a fuzzy relation to make inference with coherent results
Samer Hassan HAIS Thanks for your attention! Samer Hassan Collado Dep. Ingenieria del Software e Inteligencia Artificial Universidad Complutense de Madrid
Samer Hassan HAIS Contents License This presentation is licensed under a Creative Commons Attribution You are free to copy, modify and distribute it as long as the original work and author are cited