17
Measuring the Transfer of Knowledge Skills Constrained- student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information Systems,Japan Safia Belkada Toshio Okamoto {safia,okamoto}@ai.is.uec.ac.jp

Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Embed Size (px)

Citation preview

Page 1: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Measuring the Transfer of Knowledge Skills Constrained-student Modeler

Autonomous Agent

University of Electro-Communications Graduate School of Information Systems,Japan

Safia Belkada & Toshio Okamoto

{safia,okamoto}@ai.is.uec.ac.jp

Page 2: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Outline of the Presentation

Problem state Framework Research Purpose The domain knowledge model The tutor agent model Comments about current and future research

Page 3: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Problem State

Defect of previous methods in building a learner models for ITSs– Learner model revision

• Changes of the learner’s knowledge are not represented.

• Information about previous learner’s reasoning are lost

Page 4: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Framework

We propose an approach of learner modeling that focuses on instruction purpose only.

Page 5: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Research purpose

Domain knowledge representation• The components’ behavior process level of the domain

knowledge, which is described by relevant/ satisfaction constraints.

The help system build as reactive tutor agent • The controller of the learning process level that manages

the feedback in regard to the learner's goals   and the transfer of knowledge skill acquisition.

Page 6: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Student GU

ISym

bolic and procedural know

ledge

Objects’ Library+domain application

constraints initialization of the problem state

KBNND

ata R

ules

Constraints verification pattern matcher

Hints’ rules

learner modeler Agent

System Control Flow and Building Blocks

Model generator

Problem solver

Knowledge evaluator of CK, MK

Student DB

Page 7: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

The Transfer of Knowledge Skill Acquisition

Model knowledge(Generality)

Concept Knowledge

proceduralKnowledge

Learning/discovery

Learning/discovery

Data acquisition

Data acquisition

exploration

Domain application +Generic tools = symbolic knowledge

Help activation

Page 8: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Domain Knowledge Representation Components of the domain knowledge = collection of

constraints. State constraint => unit Each state => ordered pair <Cr, Cs>. Cr is a cluster

relevant constraints and Cs is cluster of satisfaction constraints.

Problem state Pi = constraint is relevant. Learning objective Loi = learning goals of the learner Subset of domain knowledge=> DK = <Lo1, P1> ,<Lo2,

P2>, <Lo3, P3> <Lo4, P4>

Page 9: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Description:

Lo=<Sk, Pk, Ck> Learning objective

Ck ::= Sk symbolic knowledge

(Ck1 …Ckn) conjunction of concepts

Pk(Sk1….Skn) functional dependency between

procedural and conceptual

knowledge

Page 10: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Description (continue)

Pi can be defined inductively as following:    All elements in Cr are action types. If Lo1…Lon are distinct objects in Dk and A1…An

are the appropriate instruction set during the design steps, then every expression which conforms to one of the following is a problem state of the form:

– [Lo1:Cr1=>A1…. Lon:Crn=>An]; sub-expressions

Loi:Cri 1<i< n are constrained components.

– {Cr,Cs} values are agent’s inputs.

Page 11: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Initialize

Constraints analysis

Update commitments

Tasks

Commitments

Communication

Rulebase

React

Tutor Agent Model

Cr patterns Matches the problemState that matches Cs

no

yes

Page 12: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Modeling the agent’s task

The agent’s task is a method that is described as, a.taskModel(<Loi, Pi>,p, <Cr, Cs>), where:

– a is the instance of the agent

– p the purpose of the model (generate appropriate feedback, measurement of student's knowledge).

– <Cr, Cs> corresponds to the computational constraints.

Page 13: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Modeling Agent's Tasks

The agent accomplishes two types of tasks:

- Measurement of the student's knowledge.

- Difficulty for a particular learning objective.

- Dependency between learning objective

- Constraints violated

- Hint Taken

- The generation of the feedback related to the measured knowledge.

Page 14: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Communication Skills

The event modeling consists on: identifying an event

library for the whole system, determining an object model

encapsulating the learner action and problem state.

class Method selector

constraints

Component Reference to method

Reference to <Cr,Cs>

Send-Msg(class, instruction_selector) Initial message receiver object

Page 15: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Commitment Rule

Each commitment rule contains a message condition and an action.

In order to determine whether such a rule fires, the message

condition is matched against the current tasks of the agent. If the

rule fires, then the agent becomes committed to the action.

The operation of the agent is described by the following loop

1) Read all current messages, updating commitments where necessary;

2) Execute all commitments for the current cycle where the capability condition of the associated action is satisfied;

3) Goto(1)

Page 16: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Update the Commitment Rule

– determining which methods are applicable to which components,

– defining what effects these methods have on the objects and the corresponding pair of relevant/satisfaction constraint clusters <Cr,Cs>.

– identifying the clusters Cr1…Crn of learner’s problem states in which instructional A1…An. are appropriate, and retrieve hints associated to them. These allow the agent to update its commitment rules.

Page 17: Measuring the Transfer of Knowledge Skills Constrained-student Modeler Autonomous Agent University of Electro-Communications Graduate School of Information

Conclusion

System that need adaptivity without having a runnable learner or the expert models.

Focus on computation of knowledge as well as understanding level of the learners, rather that traditional focus on diagnosis and assessment.