A method for linking the Agent-based Social Simulation to the... Hiroki Shima

From: AAAI Technical Report WS-02-10. Compilation copyright © 2002, AAAI (www.aaai.org). All rights reserved.
A method for linking the Agent-based Social Simulation to the Real World
Hiroki Shima
Graduate School of Media and Governance, Keio University
Endo 5322, Fujisawa-shi, Kanagawa, 252-8520 Japan
Fujita Institute of Future Management Research
Shinbashi 1-1-1, Minato-ku, Tokyo, 105-0004 Japan
sh@sfc.keio.ac.jp
Yoshiyasu Takefuji
Graduate School of Media and Governance, Keio University
Endo 5322, Fujisawa-shi, Kanagawa, 252-8520 Japan
Abstract
It is necessary to mature the method to improve a gap
between Agent-based Social Simulation (ABSS) and a real
world where we may utilize ABSS effectively. The
important point to apply ABSS to an actual scene in a real
society is, (1) to set a concrete specified purpose, (2) not to
depend on the abstract traditional model in modeling, (3) to
clarify the effective range of the constructed model, (4) to
use the investigation to the target society efficiently on
modeling and setting parameters. World Wide Web is
extremely effective for the investigation using ABSS where
the complicated architecture is often required.
Introduction
As the result of rapid spread of high performance
computers in 1990's, the computers have been used in the
area of a social research. Agent-based Social Simulation
(ABSS) is becoming very popular in the last few years.
However, ABSS has not arrived at the stage where it is
effectively used in a real society as a scheme for the
policy-formulating or the marketing of the enterprise etc.
Most of simulations that have been developed might be
difficult to apply to a concrete case and to verify the
validity of the model where the abstraction level is high. It
is certainly very meaningful to analyze social models and
to elucidate complex systems, because the possibility of
the social science has been extended remarkably by the
development of information technology. There are a lot of
fields which the social scientist should pioneer. However,
another way to develop ABSS is to seek a method to apply
on the actual scene of the society.
In this research, we aim to establish a method for the
practicable use of ABSS in a real society, especially which
is focused on the commodity market for the consumer. As
the current work, cafe market is handled as the sample case.
We are currently working on building the practicable
system as a tool to make marketing strategy by relating the
model to a concrete marketing operation. In this paper, we
would discuss a scheme for linking ABSS to the real world
thorough our case study of cafe market.
Viewpoint
While simulation should originally be an emulation of
reality, where simulation target is on society with abstract
models, a big dilemma would be caused that simulation
has a large gap from the reality after all. The current
ABSS has a problem in model verification and is not
effectively used in the scene of various social problems
and decision-making. Social simulation researchers might
have to build more realistic models and effective usages.
Purpose of Simulation
In general, the following four points are considered as the
main purpose of simulation.
1. To understand a specific phenomenon in the real
world logically and mathematically,
2. To predict the movement of the real world and find
measures,
3. To plan or to design something of the real world, and
4. To comprehend human behavior or to train human.
Current Position of ABSS
The conventional ABSS works have been largely studying
on the first point, while only a few ABSS study on the
other points. It is still extremely difficult to build an
accurate model to make everyone trust.
The main reason for this is that the degree of freedom in
the model is extremely large compared with other
approaches. This is an essential feature of ABSS that
cannot be evaded, which is not only an advantage but also
a disadvantage. 1 We should think how to reduce a gap
from the reality in this given restriction. When more
researchers engaged in ABSS, the schemes are
sophisticated, and the researched results come to be
organized, the simulation will be used for the second, third,
and fourth purpose in ABSS. And, it is expected that
industrial demand rise in ABSS.
Linkage with Real World
We would arrange the basic idea at the purpose of this
research on the linkage of ABSS with the real world.
Purpose-oriented Design
Simulation should be based on purpose-oriented design to
link with a real society. The real society is based on multi
dimension viewpoints and the angles. Every model is only
the one which has captured a part of society or human
features and has abstracted them. Therefore, to construct
the trusted model, we have to examine clipped aspects to
and the condition of the application after defining the
purpose of simulation. The simulation purpose should be
concrete.
Policy of Purpose Setting. It is preferable to choose the
one that can elicit the advantages of ABSS on setting the
purpose. There are the following advantages in ABSS.
・ The diversity of human can be considered by respecting
individuality of each subject as material of calculation.
・ The dynamic relationship (interaction) can be handled
which is seen between each subjects.
・ An emergent process can be handled by the bottom-up
simulation as if it were "In the beginning was the
Object".
・ The model can be easily exchanged by using the
approach of object-oriented simulation.
We should choose the one that utilize these advantages on
setting the purpose of the simulation.
Necessity of model building
A quite new model should be designed for ABSS to satisfy
concrete needs of a real society in numerous cases. Even
when using traditional models, the researcher should fill a
lot of logical vacancy on his (Her) own.
Limit of traditional model. Traditional model, a lot of
which is designed on the assumption of thinking by human
brain, go amiss on applying to ABSS with a computer.
Most of traditional models often bump to the logical
vacancy on applying them to a concrete case because the
1
In other words, ABSS (generally, Object-Oriented Social
Science) is more excellent than other mathematical models
in the point how complex logical structure can be
constructed. On the other hand, it is difficult to unite to
demand reality up to the reasonable level in the potentiality
of model building and to confirm the validity of the model.
abstraction level is too high. It is difficult to construct an
enough logical structure to handle a real society with
ABSS if only traditional models are used.
Moreover, if you execute the simulation in the frame of
the logical structure that traditional model can cover itself,
the simulation remain unreal to the end. While traditional
model have contributed to social science greatly, it
sometimes works like “Spell”, if expecting authorization of
the validity to traditional model is related to being arrested
to an antique frame. We have to be always aware of that
the research environment is much different from in the age
we didn't have any computer. Social science should move
the next stage as the development of IT (involving
computer hardware, programming technology), though
“Legacy models” should be respected.
Clarification of Range of Model
When the model is constructed, it is important to clarify
the range (condition) in which the model have
effectiveness. To begin with, it should be reminded that
every model depends on various environments like age,
region, culture, religion, law, technology, economic
condition, or infrastructure, etc.2 It is an illusion to think
that such universal model exists which is not depending on
any environment, and then we should better avoid an
ambitious attempt to make up a transcendental model. To
build an efficient model for the scene in the reality, it is
important to construct the model specialized in a particular
case and to define the condition in which it is effective.
Assessment of Validity of Model
Standing in the above-mentioned way, the validity
foundation of the model might be impeached. A basic idea
for this is that a new model doesn't have to be assessed at
the same time when it is constructed.
When the
researching style is generalized that the models are shared
not by natural-languages but by artificial language, each
model will be verified on referred and applied many times,
and then the reliability will be assessed. This will become
a new paradigm in the scheme of the social science.3
Modeling
The behavior of the constituent in real society cannot
always be comprehended by logical framework handled in
computer simulation. However, it is not avoidable to
reduce the behavior to a logical process to execute a
2
Even if there is a model which doesn't depend on any
environment, it is abstract so that almost useless on the
practical case. Or, it might be possible to construct a
model for general purpose to a degree if it is not an
inclusive model but a simple module (component) as
which a particular function is designed.
3
This idea is discussed as a possibility of "Object-Oriented
Social Science" (OOSS) in 'Innovation of social science'
(Quarterly Future Management, p.71-p.75 of 1st number).
simulation. It is needless to say that the work to observe
the real world carefully and patiently to find regularity in it
and to reduce it to a logical structure is needed in modeling.
It is also important to modulate the structure of the model
considering whether the model can be expected to behave
appropriately in the simulation world. We should not
spare imagination in this process. If the composer cannot
have a good image that the model takes the behavior in
conformity with reality, the model would already be equal
to the garbage of no practical use from this step.
Parameter setting
It is necessary to fix the parameters of each agent and the
environment in the preparation of simulation, after model
building is completed and the variation of action frame and
the action pattern of agent are determined. The setting of
the parameter is different by the target social system or the
target community, while a model is effective in some area.
As a matter of course, the parameter must be a reflection of
status of the target.
model might congeal to some narrow pattern influenced by
the world view of the researcher. Questionnaire of free
description or interview is also significant to comprehend
how exactly the constructed model catches the target.
Investigation for Parameter Setting. It is necessary to
institute the survey of the target social system for setting
the parameters in order to duplicate the feature of the real
society appropriately. It is very important process for
linking the simulation with the reality that we try to reflect
the results of the survey targeted at the constituent member
of the real society, on fixing the combination, the
proportion and the dispersion of the logical structures
which the agents behaving in the simulation world have. It
is necessary at least to institute a survey to the extent that
we comprehend the numerical dispersion. While there are
various constituent members in a real society, the diversity
itself is sometimes essentially important to grasp the social
dynamics. The meaning to use ABSS weakens if such a
factor is missed, considering that one of the main
advantages of ABSS is that the individuality of each agent
can be respected as a calculation element.
Investigation on the Target Society
Case Study of Cafe Market
One of effectual measures to reduce the gap between the
real world and the simulation world in ABSS is that we
carry out social survey to observe the principle of human
behavior and use them to decide the models and the
parameters of the simulation world.4
This survey is an extremely important process when
executing the simulation linked with the real world, though
it is unnecessary when executing the simulation used
abstract models. While it is difficult to clip data which
faithfully express the feature of the real world, the
practicability of the simulation in concrete scene of the real
society largely depends on what appropriate survey is
executed.
Investigation for Model Building. On building a model,
it is effective work that we institute a survey which have
high degrees of freedom, like questionnaire of free
description or interview, and extract the logical structure
from thence by using flowchart or state transition diagram.
We often find the limitation of the way to fix the logical
structure just from the observation of real society when
modeling. The reason is that there are a lot of factors
which do not surface like the process generated in human
inside such as decision making process. An effective way
to glean them is that we institute a survey which has high
degrees of freedom and obtain the materials there.
Moreover, it is requested to prepare the various patterns
of logical structures because each human individual has
different principle of behavior. However, work to observe
individual behavior carefully is not only a burden, but the
In modeling, we have to consider carefully how the real
world is clipped with due regard to the purpose to use as a
tool to make a brand strategy of the cafe. We referred for
modeling to observation of cafe and cafe user, documents
about cafe (including magazines etc.) and the results of
informal hearing and questionnaire. 6
Market. In this case study, we regard the cafe market as
matching of cafes and consumers who use the cafes, that
structure is modeled. (Fig. 1) It is just applied the essence
of the general market model that is caught as matching of
demand and supply. In the simulating market, there are
two kinds of agents, consumer agents who try to get higher
satisfaction and cafe agents who try to increase sales.
4
5
It is a matter of course that it takes some cost to carry out
the survey. Unless standing in the lucky circumstances to
have enough budgets for the experimental investigation,
we would have to arrange a profitable research plan on the
assumption that it can provide an industrial use.
We would introduce the abstract of the model and the
investigation in our cafe market simulation that is a sample
case to construct the method that we use ABSS as a tool to
make marketing strategy in the real market by keeping
little the gap between simulation and reality.5
Purpose of the Simulation
First of all, the purpose of the simulation is to use it as a
tool to make a brand strategy of cafe. While there are
various elements in brand strategy of cafe, we focus on the
aspect how to control the value provided by shop.
Model
Furthermore, it is scheduled to report again about the
detail of the investigation and the simulation handled by
this case study, after obtaining the results of the simulation.
6
This time we didn't conduct a survey for modeling on a
large scale.
Matching
【Café User】
Sence of value
Café User
Café
(Demand)
(Supply)
Fig. 1 Architecture of Market
Value (Key of Matching). It is "Value" of service that is
focused as the key to understand the correspondence of
consumer and shop in matching. The consumer tries to
choose the shop where he (she) can get higher satisfaction
respectively in his (her) own sense of value. On the other
hand, each shop tries to increase sales attracting more
consumers by coordinating the value that it provides.
Then, the values which the shop provides and the values
which the consumer enjoys are treated assuming that they
are positioned in the space that is weaved by the same axes
associated with each other respectively. (Fig. 2) While the
values cafe provides are various, 41 value axes were
prepared in this simulation, and each cafe is featured by
combining these values.7 Besides, each value is specified
in several formats of measure of value in the model.
Consumer Agent. It is necessary to design the model
which grasps the feature of cafe use in the real society to
duplicate the cafe market. The following features are
given to the model of consumer agent in this simulation.
[Basic Action] Each consumer agent chooses one shop
where is expected to get higher satisfaction with reference
to the past experience and obtained information on every
opportunity to use a cafe. 8 The frequency of the
opportunity to use cafe is set for every agent.
[Individuality] Each agent has his (her) sense of value
respectively to evaluate cafe.
[Sense of Value] Sense of value is specified by the set of
the degrees by which each value is weighed and the
content of various measures of value.
[Situation] Each agent is given the structure to choose the
shop by different measure in different situation. This is
because, when asking by what measure to choose the shop
when using a cafe actually, a lot of people in real society
answer that it depends on the time and the circumstance.
[Communication]
Each agent obtains information
through media and exchanges the opinion with friends of
each for the selection to be able to get a higher satisfaction.
7
It is said that the biggest point for cafe users to choose
cafe is the location in many cases. This simulation treat
the location just as one of a lot of elements to specify the
brand, which is extremely simplified, so that it doesn't
work very well as a tool to make the strategy to set up new
shops. It is effective for making the strategy to set up new
shops to execute the simulation that the geographic
element is fore grounded.
8
The detail of the structure of the decision making process
at this time is omitted here.
・Value 1
・Value 2
・Value 3
・Value 4
・Value 5
Café User Agent
【Café】
Correspondence
Providing value
Simulation
Café Agent
・Value 1
・Value 2
・Value 3
・Value 4
・Value 5
Fig. 2 Correspondence between cafe
and cafe user by Value
Cafe Agent. In the model, the cafe agent is handled as a
subject that tries to increase sales by controlling the value
which itself provides. But, the cafe agent is not given so
dynamic action in this simulation. The major identity of
the cafe agent is the list of the value which oneself
provides, that is defined before simulation start.
Investigation
We conducted the survey to adjust the parameters of each
agent for executing the simulation in which the realities of
the target market is reflected. The survey outline is as
follows. Besides, the investigation architecture is shown in
Fig. 3. This investigation has a complicated architecture to
be designed for defining the action principle of each agent.
Survey Outline
Subject: Survey of buymanship of consumers in cafe market.
(The priority is given especially to the measurement of
sense of value in this time.)
Survey Item: Situation to use cafe, Frequency and average budget
of each situation, Points made much of when
choosing cafe in each situation, and Custom of using
cafe.
Target Sample: Men and women, from high school student to thirtysomething, living in the metropolitan area of Japan
(Tokyo, Kanagawa, Chiba, Saitama).
Sample Number: 2000 (collected).
Research tool: Web is used.
Application to Marketing
This simulation is designed for practical use on the actual
marketing scene. For example, when a certain marketing
strategy is planned, the prospect whether the operation is
appropriate can be set up beforehand by executing test
marketing in the simulation. Most of measures of values
set in the model are controllable by the operation to each
shop in the real world. Therefore, it is possible to perform
an experiment in the simulation beforehand to estimate
what influence is brought to use state and sales when
changing the value which the shop provide.
(For each situation)
1. To enumerate the situation to use a
cafe. (Free description)
3. To adjust the scene of value by the
situations that is enumerated in 1.
1-2. To fill the frequency and the expense of each
situation.
3-1. To adjust the degrees by which each value is
weighed and the content of various measures
of value as necessary from the standard
scene of value specified in 2.
Investigation Flow →
2. To specify the scene of value on choosing a cafe
in general case. (Standard scene of value)
2-1. To place checks in the item which is valued as criteria
from all the measures of value. (41 items)
4. To answer some simple questions
concerning the custom of using cafe
additionally.
2-2. To adjust the degree by which each value is weighed about
the items checked in 2-1.
2-3. To specify the taste about the items that separate by
individual in the items checked in 2-1.
1.
3.
2.
(2-1)
(2-2)
(2-3)
4.
Fig. 3 Architecture of Investigation (Sample pages are in Japanese)
Moreover, it is possible to examine what plan should be
adopted in the publicity campaign or media strategy by
calculating the effect of communications of agents.9
9
Besides, a lot of various uses are possible other than what
is mentioned here by build in other models or
investigations that complement the basic architecture of
this model according as the marketing operation targeted.
Discussion
Variation of investigation for Parameter Setting
The investigation for the parameter adjustment can be
thought separately to the following two features. That is,
the investigation to fix the logical structure of agent and
the investigation to adjust numerical measure of agent.
Investigation to Fix Logical Structure. Investigation to
fix the logical structure of agent extracts the information
what logical architecture is given as the action principle of
each consumer. A lot of consumer's action principle is not
possible to be comprehended only by adjusting numerical
parameters and handling by simple mathematical functions.
The action principles that are structurally different should
be handled by respectively different logical structures. 10
This investigation is for the comprehension, what logical
structure is appropriate to be built into each agent.11
Investigation to Adjust Numerical Measures. In general,
it is easy to handle the investigation to adjust numerical
measures compared with the one to coordinate the logical
structure on the simulation. The reason is that there is no
necessary to prepare different logical structures or to
consider the consistency between each structure because it
can be handled in the same logical structure.
For instance, the following items are enumerated with
this case; the frequency and the budget of each situation to
use a cafe, which value is how valued. Such numerical
values can be easily reflected on the parameter of each
agent.
Use of Web to Investigation
While the models handled in ABSS often have the
complicated structure in general, it is effective to use the
questionnaires on Web as the method to fill such a
complicated structure. 12 Also in this case, the sense of
value of agent has the complicated logical structure. It is
difficult to grasp the information to specify the values of
various measures which lie dispersed in the compound
structure by using the questionnaire of the paper media.
The consideration to ask questions efficiently is required
because the load of answerer can often be a bottleneck in
10
If the action principles that are structurally different are
put in one logical structure coercively, it will bring a
remarkably distorted consequence.
11
Several schemes are possible at the mounting level as
the method to embed different logical structures in each
agent respectively. One scheme is to switch whether each
function appears or not by on/off where all the logical
structures are built in the program of every agent
beforehand. In this research, we adopted the method of
designing each function as an object and coordinating the
modules (objects) built in when the agent is given.
Therefore, different agent might have different modules in
it even when they behave in the same environment.
12
Notice a lot of investigation instituted on Web, where
low-cost is made a point of, is not considered enough to
add credence of the outcome. While more examination is
requested about how to use the investigation on Web for
ABSS, it will surely become a promising tool in applying
ABSS to a concrete aspect and putting it to practical use in
the near future.
the actual investigation when complicated questionnaire is
designed. For example it is waste process to present all the
questions to those who answer. It is necessary to consider
for the answerer to answer just to the necessary questions
omitting needless questions in the prepared questionnaire
according as the content of the previous answer.13
In this simulation, it is requested to extract selection
criterion on each situation as information to set the
parameters of each agent. It is demanded for specifying
the selection criterion to pick up the important measure of
value from more than 40 items, to adjust the valued degree
and even to assign the measure of value if necessary.
Though it is almost impossible to execute such a thing in
questionnaire of the paper media, it is not too difficult to
do that if Web is used.
Conclusion
It goes without saying that the structure introduced here is
no more than one sample. We have to consider what
architecture of investigation should be executed in every
case according to the object and the use purpose of the
simulation.
We should mind that the model handled in ABSS often
have a complex structure, especially when it is intended for
a concrete aspect. And we will find much individual
difference between living people in the real world in
various respects corresponding to the modules in the
structure. It is necessary to construct the complicated
observation system so much to reflect the individual
variation in the simulation world. It is an effective way to
institute the questionnaire survey on Web like the case
shown in this paper for that. This is because it is possible
to institute the survey that is given a much complicated
architecture more efficiently on web than on paper.
Reference
[1] Hiroki Shima, "Innovation of Social Science - An
Exhortation of Object-oriented Social Science
(JAPANESE)", Quarterly Future Management, No.1,
2001 Spring (initial number), pp.71-75,Apr.2001.
13
While even the questionnaire in paper media can be
more efficient to some extent by attending to the
navigation, it is bounded. For instance, though it is
possible to navigate to advance ahead flying some needless
questions, it is not realistic to plan the following
operations; (a) to present the several questions which is
most appropriate that is picked up from 100 questions, (b)
to process an answer to a certain question and reflect it in
the latter question, (c) to permute the question order
according to the content of the previous answer for making
easy to answer.