Your SlideShare is downloading. ×
The Design of Cognitive Social Simulation Framework using Statistical Methodology in the Domain of Academic Science
The Design of Cognitive Social Simulation Framework using Statistical Methodology in the Domain of Academic Science
The Design of Cognitive Social Simulation Framework using Statistical Methodology in the Domain of Academic Science
The Design of Cognitive Social Simulation Framework using Statistical Methodology in the Domain of Academic Science
The Design of Cognitive Social Simulation Framework using Statistical Methodology in the Domain of Academic Science
The Design of Cognitive Social Simulation Framework using Statistical Methodology in the Domain of Academic Science
The Design of Cognitive Social Simulation Framework using Statistical Methodology in the Domain of Academic Science
Upcoming SlideShare
Loading in...5

Thanks for flagging this SlideShare!

Oops! An error has occurred.

Saving this for later? Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime – even offline.
Text the download link to your phone
Standard text messaging rates apply

The Design of Cognitive Social Simulation Framework using Statistical Methodology in the Domain of Academic Science


Published on

Modeling the behavior of the cognitive architecture in the context of social simulation using statistical methodologies is currently a growing research area. Normally, a cognitive architecture for an …

Modeling the behavior of the cognitive architecture in the context of social simulation using statistical methodologies is currently a growing research area. Normally, a cognitive architecture for an intelligent agent involves artificial computational process which exemplifies theories of cognition in computer algorithms under the consideration of state space. More specifically, for such cognitive system with large state space the problem like large tables and data sparsity are faced. Hence in this paper, we have proposed a method using a value iterative approach based on Q-learning algorithm, with function approximation technique to handle the cognitive systems with large state space. From the experimental results in the application domain of academic science it has been verified that the proposed approach has better performance compared to its existing approaches.

Published in: Education, Technology
  • Be the first to comment

  • Be the first to like this

No Downloads
Total Views
On Slideshare
From Embeds
Number of Embeds
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

No notes for slide


  • 1. International Journal of Research in Computer ScienceeISSN 2249-8265 Volume 3 Issue 3 (2013) pp., A Unit of White Globe Publicationsdoi: 10.7815/ijorcs. 33.2013.062www.ijorcs.orgTHE DESIGN OF COGNITIVE SOCIAL SIMULATIONFRAMEWORK USING STATISTICAL METHODOLOGYIN THE DOMAIN OF ACADEMIC SCIENCEV. Maniraj1, R. Sivakumar2*Associate Professor, Computer Science, A.V.V.M Sri Pushpam College (Autonomous), Poondi, Thanjavur, INDIAE-mail:, 2rskumar.avvmspc@gmail.comAbstract: Modeling the behavior of the cognitivearchitecture in the context of social simulation usingstatistical methodologies is currently a growingresearch area. Normally, a cognitive architecture foran intelligent agent involves artificial computationalprocess which exemplifies theories of cognition incomputer algorithms under the consideration of statespace. More specifically, for such cognitive systemwith large state space the problem like large tablesand data sparsity are faced. Hence in this paper, wehave proposed a method using a value iterativeapproach based on Q-learning algorithm, withfunction approximation technique to handle thecognitive systems with large state space. From theexperimental results in the application domain ofacademic science it has been verified that theproposed approach has better performance comparedto its existing approaches.Keywords: Cognitive architecture, Social Simulation,Reinforcement learning, Function approximation.I. INTRODUCTIONA cognitive architecture specifies the underlyinginfrastructure for an intelligent system. Briefly,architecture includes those aspects of a cognitive agentthat are constant over time and across differentapplication domains. These typically include:1. The short-term and long-term memories that storecontent about the agent’s beliefs, goals, andknowledge;2. The representation of elements that are contained inthese memories and their organization into larger-scale mental structures;3. The functional processes that operate on thesestructures, including the performance mechanismsthat utilize them and the learning mechanisms thatalter them.Because the contents of an agent’s memories canchange over time, one would not consider theknowledge and beliefs encoded therein to be part ofthat agent’s architecture. As different programs can runon the same computer architecture, so the differentknowledge bases and beliefs can be interpreted by thesame cognitive architecture. There is also a directanalogy with a building’s architecture, which consistsof permanent features like its foundation, roof, androoms, rather than its furniture and appliances, whichone can move or replace.II. EXAMPLES OF COGNITIVE ARCHITECTURESA. ACT -RACT-R [7][8] is the latest in a family of cognitivearchitectures, concerned primarily with modelinghuman behavior, that has seen continuous developmentsince the late 1970s. ACT-R is organized into a set ofmodules, each of which processes a different type ofinformation. These include sensory modules for visualprocessing, motor modules for action, an intentionalmodule for goals, and a declarative module for long-term declarative knowledge. Each module has anassociated buffer that holds a relational declarativestructure (often called ‘chunks’, but different fromthose in Soar). Taken together, these buffers compriseACT-R’s short-term memory.B. SOARSoar [4][5][1] is a cognitive architecture that hasbeen under continuous development since the early1980s. Procedural long-term knowledge in Soar takesthe form of production rules, which are in turnorganized in terms of operators associated withproblem spaces. Some operators describe simple,primitive actions that modify the agent’s internal stateor generate primitive external actions, whereas othersdescribe more abstract activities. For many years, Soarrepresented all long-term knowledge in this form, butrecently separate episodic and semantic memorieshave been added. The episodic memory [2] holds ahistory of previous states, while semantic memorycontains previously known facts.C. ICARUSICARUS is a more recent architecture [14] thatstores two distinct forms of knowledge. Concepts
  • 2. 2 V. Maniraj, R. Sivakumarwww.ijorcs.orgdescribe classes of environmental situations in terms ofother concepts and percepts, whereas skills specify howto achieve goals by decomposing them into ordered subgoals. Both concept and skills involve relations amongobjects, and both impose a hierarchical organization onlong-term memory, with the former grounded inperceptions and the latter in executable actions.Moreover, skills refer to concepts in their heads, theirinitiation conditions, and their continuation conditions.D. PRODIGYPRODIGY [6] is another cognitive architecture thatsaw extensive development from the middle 1980s tothe late 1990s. This framework incorporates two mainkinds of knowledge. Domain rules encode theconditions under which actions have certain effects,where the latter are described as the addition or deletionof first-order expressions. These refer both to physicalactions that affect the environment and to inferencerules, which are purely cognitive. In contrast, controlrules specify the conditions under which thearchitecture should select, reject, or prefer a givenoperator, set of operator bindings, problem state, orgoal during the search process.PRODIGY’s explanation-based learning moduleconstructs control rules based on its problem-solvingexperience [20]. Successful achievement of a goal aftersearch leads to creation of selection or preference rulesrelated to that goal and to the operators whoseapplication achieved it. In addition, PRODIGYincludes separate modules for controlling search byanalogy with earlier solutions [11] learning operatordescriptions from observed solutions orexperimentation [21], and improving the quality ofsolutions [12]. Although most research in thisframework has dealt exclusively with planning andproblem solving, PRODIGY also formed the basis foran impressive system that interleaved planning andexecution for a mobile robot that acceptedasynchronous requests from users [10].E. THE CLARIONCLARION is an integrative cognitive architectureconsisting of several distinct sub systems [16][18][15].These include the action-centred subsystem (ACS), thenon-action-centred subsystem (NACS), themotivational subsystem (MS), and the metacognitivesubsystem (MCS). The ACS controls actions, whetherfor external physical movements or internal mentaloperation. The NACS maintains general knowledge,either implicit or explicit. The MS provides under-lyingmotivations for perception, action, and cognition interms of impetus and feedback (for example, indicatingwhether outcomes are satisfactory or not). The MCSmonitors, directs, and modifies the operations of theACS dynamically, as well as the operations of all theother subsystems.Modeling the behavior of the cognitive architecturein the context of social simulation using statisticalmethodologies is currently a growing research area.Normally, a cognitive architecture for an intelligentagent involves artificial computational process whichexemplifies theories of cognition in computeralgorithms under the consideration of state space.More specifically, for such cognitive system with largestate space the problem like large tables and datasparsity are faced. Hence in this paper we haveproposed a method using a value iterative approachbased on Q-learning method with functionapproximation to handle the cognitive systems withhuge state space. From the experimental results in thedomain of academic science it has been proceeded thatto proposed approach results better performancecompared to its existing approach.III. CAPABILITIES OF COGNITIVEARCHITECTURESAny intelligent system is designed to engage incertain activities that, taken, together, constitute itsfunctional capabilities. This section, discusses thevaried capabilities that a cognitive architecture cansupport. Although only a few abilities, such asrecognition and decision making, are strictly requiredto support a well-defined architecture, the entire setseems required to cover the full range of human-levelintelligent activities.• Recognition and Categorization• Decision Making and Choice• Perception and Situation Assessment• Prediction and Monitoring• Reasoning and Belief Maintenance• Execution and Action• Interaction and Communication• Remembering, Reflection and Learning• Problem solving and PlanningIV. COGNITIVE MODELSCognitive models attempt to represent the users asthey interact with a system, modeling aspects of theirunderstanding, knowledge, intentions or processing.Cognitive models are divided into three categories.• The first described the hierarchical structuring ofthe user’s task and goal structure. Eg. GOMS(Goals, Operators, Method, Selection) CCT(Cognitive Complexity Theory).• The second model concerned with Linguistic andgrammatical models, which emphasized the user’sunderstanding of the user system – dialog. Eg. BNF(Backus Naur Form), TAG (Task ActionGrammar).
  • 3. The Design of Cognitive Social Simulation Framework using Statistical Methodology in the Domain of Academic Science• Third model based on the more solid understandingof the Human Motor System. Eg. KLM (Keystroke– Level Model).V. EXISTING A COGNITIVE ARCHITECTURE INSOCIAL SIMULATIONOne application of CLARION to social simulationis in understanding organizational decision making andthe interaction between organizational structures andcognitive factors in affecting organizational decisionmaking [Sun, 04].In terms of organizational structures, there are twomajor types: (1) teams, in which agents actautonomously, individual decisions are treated asvotes, and the organizational decision is the majoritydecision; and (2) hierarchies, which are characterizedby agents organized in a chain of command, such thatinformation is passed from subordinates to superiors,and the decision of a superior is based solely on therecommendations of his/her subordinates. In addition,organizations are distinguished by the structure ofinformation accessible by each agent. Two varieties ofinformation access are: (1) distributed access, in whicheach agent see a different subset of attributes (no twoagents see the same sub-set of attributes), and (2)blocked access, in which several agents see exactly thesame subset of attributes.Several simulation models were considered in [9].The experiments by [9] were done in a 2 x 2 fashion(organization x information access). In addition,human data for the experiment were compared to theresults of the four models [9].In their work, the agent models used were verysimple, and the “intelligence” level in these modelswas low. Moreover, learning in these simulations wasrudimentary: there was no complex learning process asone might observe in humans. With theseshortcomings in mind, it is worthwhile to undertake asimulation that involves more complex agent modelsthat more accurately capture human performance.Moreover, with the use of more cognitively realisticagent models, one may investigate individually theimportance of different cognitive capacities andprocess details in affecting organizational performance[19].Hence, a simulation with CLARION used formodelling individual agents in an organization wasconducted [19]. The results closely accord with thepatterns of the human data, with teams outperforminghierarchical structures, and distributed access provingsuperior to blocked access.Also, as in humans, performance is not grosslyskewed towards one condition or the other, but isroughly comparable across all conditions, unlike someof the simulation results from [9]. The match with thehuman data is far better than in the simulationsconducted in [9]. The better match is due, at least inpart, to a higher degree of cognitive realism in thissimulation. See [19] for further details, including theinteresting effects of varying cognitive parameters.Another application of CLARION to socialsimulation is in capturing and explaining the essentialprocess of publication in academic science and itsrelation to cognitive processes [3]. Science develops incertain ways. In particular, it has been observed thatthe number of authors contributing a certain number ofarticles to a scientific journal follows a highly skeweddistribution, corresponding to an inverse power law. Inthe case of scientific publication, the tendency ofauthorship to follow such a distribution was known asLotka’s law.Herbert Simon developed a simple stochasticprocess for approximating Lotka’s law. One of theassumptions underlying this process is that theprobability that a paper will be published by an authorwho has published i articles is equal to a/ik, where a isthe constant of proportionality. Using Simon’s work asa starting point, [13] attempted to model Lotka’s law.He obtains his simulation data based on some verysimplified assumptions and a set of mathematicalequations. To a significant extent, Gilbert’s model isnot cognitively realistic. The model assumes thatauthors are non-cognitive and interchangeable; ittherefore neglects a host of cognitive phenomena thatcharacterize scientific inquiry (e.g., learning,creativity, evolution of field expertise, etc.).Using a more cognitively realistic model, one canaddress some of these omissions, as well as exploringother emergent properties of a cognitively based modeland their correspondence to real-world phenomena.The results of the simulation based on CLARION [3]are shown along with results (reported by [13]) forChemical Abstracts and Econometrica, and estimatesobtained from previous simulations by [3]. The tablesindicate number of authors contributing to eachjournal, by number of papers each has published. Morespecifically, for such cognitive system with large statespace the problem like large tables and data sparsityare faced. Hence in this paper we have proposed amethod using a value iterative approach based on Q-learning algorithm with function approximationtechnique to handle the cognitive systems with largestate space.VI. PROPOSED WORKFunction approximation and feature based methodA. Problem SpecificationWe consider the following form of RL task for thelearning agent.
  • 4. 4 V. Maniraj, R. Sivakumarwww.ijorcs.orgWhat is given?• Set S of possible states.• Set A of possible actions.• Unknown transition function δ: S X A  S• Reward function R, which is 1 in goal state and oelse where goal states are terminal.What to find?An optimal policy ∏ ∗ = 𝑆 → 𝐴 for selecting thenext action ai based on current observed states.It may be generally very difficult to learn a Q-Function perfectly. We often expect learningalgorithms to set only some approximation to thetarget function.B. Proposed ArchitectureFigure 1: Clarion Architecture with Function ApproximationWe will use function approximation here and learna representation of the Q-Function as a linear functionof combinations of features, where the featuresdescribe a state. In other words, we will translate astate S into the set of features f1, f2,….fn. Where n isthe number of features.We have set of Q- functions:𝑄 𝑎(𝑠, 𝑎) = 𝜃1 𝑎𝑓1 + ⋯ + 𝜃 𝑛 𝑎𝑓𝑛The update rule isθak= θak + α[r + γ max a’ Q a( s’ , a’) - Q a(s,a)] dQa(s , a )dθak(1)Algorithm The RL Algorithm with FunctionAlgorithminitialize all thetas to 0repeatGenerate a starting state s0i = 0repeatChoose an action ai, using the policy obtained fromThe current values of thetasExecute action ai, observe R and si+1i=i+1until si is terminal(i.e., a goal state)for j = i – 1 to 0 doUpdate the value of θak for all taken actions using (1)end foruntil no more episodesVII. EVALUATION AND RESULTSScience develops itself in particular ways. In particularthe number of authors contributing a certain number ofarticles to scientific journals tends to follow a highlyskewed distribution, corresponding to an inversepower law. In the case of scientific publications, thetendency of authorship to follow such a distribution isknown as Lotka’s law. Herbert Simon developed asimple stochastic process for approximating Lotka’sLaw. One of the assumptions underlying this process isthat the probability that a paper will be published by anauthor who has publiched i articles is alik, where a is aconstant of proportionality. Using Simon’s work as a
  • 5. The Design of Cognitive Social Simulation Framework using Statistical Methodology in the Domain of Academic Science 5www.ijorcs.orgstarting point, Gilbert attempted to model Lotka’sLaw. He obtained his simulation data on the basis ofsome simplified assumptions and a set of mathematicalequations. To a significant extent, Gilbert’s model wasnot cognitively realistic. The model assumed thatauthors were non-cognitive and interchangeable; ittherefore neglected a host of cognitive phenomena thatcharacterized scientific inquiry.In clarion simulation they have introduced acognitive approach by comparing the previous results.In our model the number of articles per authorreflected the cognitive ability of an author, as opposedto being based on auxiliary assumptions such as thosemade by Gilbert.Our model is using learning and functionapproximation is able to give a good match to theresults of clarion simulation. Using functionapproximation we have got better output than theprevious results. That is, we put more distancebetween mechanisms and outcomes which made itmore difficult to obtain a match with the human data.Even though, being able to match the human datareasonably well shows the power of our approach.The proposed method data for the two journalscould be fit to the power curve f(i) = a/ik, resulting inan excellent match. Table 1 and Table 2 clearly showthe good results achieved from the proposed method.Table 1: Number of Authors Contributing to Chemical Abstracts.No. of Papers Actual Simon’s Estimate Gilbert’s simulation CLARION simulation Proposed Method1 3991 4050 4066 3803 38502 1059 1160 1175 1228 12383 493 522 526 637 6464 287 288 302 436 4445 184 179 176 245 2526 131 120 122 200 2067 113 86 93 154 1608 85 64 63 163 1689 64 49 50 55 6010 65 38 45 18 2011 or more 419 335 273 145 150Table 2: Number of authors contributing to EconometricaNo. of Papers Actual Simon’s Estimate Gilbert’s simulation CLARION simulation Proposed approach1 436 453 458 418 4302 107 119 120 135 1383 61 51 51 70 724 40 27 27 48 505 14 16 17 27 286 23 11 9 22 237 6 7 7 17 198 11 5 6 18 199 1 4 4 6 910 0 3 2 2 411 or more 22 25 18 16 20
  • 6. 6 V. Maniraj, R. Sivakumarwww.ijorcs.orgFigure 2: Proposed method result for different settingFigure 3: Proposed method with CLARIONFigure 4: Proposed method result for different settingFigure 5: Proposed method with clarion.Note that, in this simulation, the number of papersper author reflected the cognitive ability of an author,as opposed to methods based on auxiliary assumptionssuch as those made by [13]. This explains, in part, thegreater divergence of these results from the humandata: whereas Gilbert’s simulation consists of equationsselected to match the human data. This approach relieson much more detailed and lower-level mechanismsnamely, a cognitive agent model that is generic ratherthan being task-specific. The result of the CLARIONbased simulation is therefore emergent, and not a resultof specific and direct attempts to match the human data.That is, one put more distance between mechanismsand outcomes, which makes it harder to obtain a matchwith the human data. Thus, the fact that it was able tomatch the human data reasonably well shows the powerof this cognitive architecture based approach.VIII. CHALLENGES FACING COGNITIVESOCIAL SIMULATIONAn important development in the social science hasbeen that of agent-based social simulation (ABSS).This approach consists of instantiating a population ofagents, allowing the agents to run, and observing theinteractions between them. The use of agent-basedsocial simulation as a means for computational study ofsocieties mirrors the development of cognitivearchitectures in cognitive science.The two fields of social simulation and cognitivearchitectures can be profitably integrated. This is animportant challenge. As it has been argued before,social processes ultimately rest on the choices anddecisions of individuals, and thus understanding themechanisms of individual cognition can lead to bettertheories describing the behavior of aggregates ofindividuals.At the same time, by integrating social simulationand cognitive modeling, one can arrive at a betterunderstanding of individual cognition. Traditionalapproaches to cognitive modeling have largely ignoredthe potentially decisive effects of socially acquired anddisseminated knowledge (including language, norms,and so on). By modeling cognitive agents in a socialcontext, one can learn more about the socio culturalprocesses that influence individual cognition.The most fundamental challenge in this regard is todevelop better ways of conducting detailed socialsimulation based on cognitive architectures as basics.Although some cognitive details may ultimately proveto be irrelevant, this cannot be determined a priori, andthus simulations are useful in determining whichaspects of cognition can be safely abstracted awaybuilding blocks. This is not an easy task. Althoughsome initial work has been done (e.g., [17][19] muchmore work is needed.There is also the challenge of computationalcomplexity. Social simulation could involve a largenumber of agents, up to thousands. Computationalcomplexity is thus already high, even without involvingcognitive architectures as agent models. To incorporatecognitive architectures into social simulation, one hasto deal with a great deal of added complexity.010002000300040005000Simon’sEstimateGilbert’ssimulationCLARIONsimulationFunction Approx010002000300040005000CLARIONsimulationFunction Approx0100200300400500Simon’sEstimateGilbert’ssimulationCLARIONsimulationFunction Approx0100200300400500CLARIONsimulationFunction Approx
  • 7. The Design of Cognitive Social Simulation Framework using Statistical Methodology in the Domain of Academic Science 7www.ijorcs.orgIX. CONCLUSIONWe have proposed here a method using avalue iterative approach based on Q-learning algorithmwith function approximation to handle the cognitivesystems with large state space. From the experimentalresults in the domain of academic science it has beenproved that the results of proposed approach are betteras compared to its existing approach. These results andalgorithms can be very useful to the researchers thosewho are in academic science and cognitivearchitecture.X. FUTURE WORKEmotions play a central role in humanbehavior, yet few systems offer any account of theirpurpose or mechanisms. There is a need of newarchitectures that exhibit emotion in ways that linkdirectly to other cognitive processes and that modulatean intelligent behavior. Most architecture incorporatesome form of learning, but none have shown therichness of improvement that humans demonstrate.There is a need of more robust and flexible learningmechanisms that are designed for extended operationin complex, unfamiliar domains.XI. REFERENCES[1] Newell, “The Knowledge level”, Artificial Intelligence,18, 87 – 127, 1982. doi: 10.1016/0004-3702(82)90012-1[2] A.M. Nuxoll and J.E. Laird, “Extending cognitivearchitecture with episodic memory”, Proceedings of theTwenty-Second AAAI Conference on ArtificialIntelligence, Vancouver, BC: AAAI Press, 2007.[3] I. Naveh and R.Sun, “A cognitively based simulation ofacademic science”, Computational and MathematicalOrganization Theory, 12, 4, 313-337, 2006. doi:10.1007/s10588-006-8872-z[4] J.E. Larid, “Extending the Soar cognitive architecture”,Proceedings of the Artificial General IntelligenceConference. Memphis, TN: IOS Press, 2008.[5] J.E. Larid, A. Newell and P.S.Rosenbloom, “Soar: Anarchitecture for general intelligence”, ArtificialIntelligence, 33, 1- 64, 1987. doi: 10.1016/0004-3702(87)90050-6[6] J.G. Carbonell, C.A. Knoblock, and S. Minton,“PRODIGY: An integrated architecture for planningand learning”, In K. Van Lehn (Ed.), Architectures forintelligence, Hillsdale, NJ: Lawrence Erlbaum, 1990.[7] J.R. Anderson, “How can the human mind exist in thephysical universe?”, New York: Oxford UniversityPress, 2007.[8] J.R. Anderson, and C. Lebiere, “The atomiccomponents of thought”, Mahwah, NJ: LawrenceErlbaum ,1998.[9] K. Carley, M. Prietula, and Z. Lin, “Design versuscognition: The interaction of agent cognition andorganizational design on organizational performance”,Journal of Artificial Societies and Social Simulation, 13, 1998.[10] K. Haigh and M. Veloso, “Interleaving planning androbot execution for asynchronous user requests”,Proceedings of the International Conference onIntelligent Robots and Systems, Osaka, Japan: IEEEPress, 148-155, 1996. doi: 10.1109/IROS.1996.570649[11] M. Velso and J.G. Carbonell, “Derivational analogy inPRODIGY: Automating case acquisition, storage”, andutilization, Machine Learning, 10, 249-278, 1993.[12] M.A. Perez and J.G. Carbonell, “Control knowledge toimprove plan quality”, Proceedings of the SecondInternational Conference on AI Planning Systems,Chicago: AAAI Press, 323-328, 1994.[13] N. Gilbert, “A simulation of the structure of academicscience”, Sociological Research Online, 2,2, Availableonline at, 1997.[14] P. Langley, K. Cummings, K and D. Shapiro,“Hierarchical skills and cognitive architectures”,Proceedings of the Twenty-Sixth Annual Conference ofthe Cognitive Science Society, 779 – 784, Chicago, IL,2004.[15] R. Sun “A Tutorial on CLARION 5.0”, tech report,Cognitive Science Dept., rensselaer Polytechnic Inst.,22,, July,2003.[16] R. Sun, “Duality of the Mind”, Mahwah, NJ: LawrenceErlbanm Associates, 2002.[17] R. Sun, “Prolegomena to integrating cognitivemodelling and social simulation”, R. Sun (ed.),Cognition and Multi-Agent Interaction: From CognitiveModelling to Social Simulation. Cambridge UniversityPress, New York, 2006. doi:10.1017/CBO9780511610721.002[18] R. Sun.R, E. Merrill and T. Peterson, “From ImplicitSkills to Explicit Knowledge: A Bottom-up Model ofSkill Learning”, Cognitive science, 25, 2, 203-244,2001. doi: 10.1207/s15516709cog2502_2[19] R.I. Sun and I. Naveh, “Simulating organizationdecision-making using a cognitively realistic agentmodel”, Journal of Articificial Societies and SocialSimulation, Vol.7, No.3, June, 2004.[20] S.N. Minton, “Quantitative results concerning the utilityof explanation-based learning”, Artificial Intelligence,42, 363-391, 1990. doi: 10.1016/0004-3702(90)90059-9[21] X. Wang, “Learning by observation and practice: Anincremental approach for planning operatoracquisition”, Proceedings of the twelfth InternationalConference on Machine Learning , Lake Tahoe, CA:Morgan Kaufmann, 549-557, 1995.How to citeV. Maniraj, R. Sivakumar, "The Design of Cognitive Social Simulation Framework using Statistical Methodology inthe Domain of Academic Science". International Journal of Research in Computer Science, 3 (3): pp. 1-7, May 2013.doi: 10.7815/ijorcs. 33.2013.062