from local patterns to global models: towards domain driven educational process mining nikola trčka...

16
From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Tr č ka Mykola Pechenizkiy

Upload: abigail-mcbride

Post on 30-Mar-2015

213 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining

Nikola TrčkaMykola Pechenizkiy

Page 2: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

ISDA’09 - EDM

Motivation

What is the real curriculum (study program)?

How do students really study?Is there a typical/best way to

study?Do current prerequisites make

sense?

What is my expected time to finish?Should I take course A or course B

now?…

Student database withexam records

?

YES/NO 80%

Process:

Standard techniques

Page 3: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

Proposed approach and architecture

ISDA’09 - EDM

Approach: Isolate a set of standard curriculum patterns and based on this patterns1. mine the curriculum as an executable quantified formal model and

analyze it, or2. (first) manually devise a formal model of the assumed curriculum and

test it against the data.

Data log

Pattern mining Pattern set

Pre-authored pattern templates

Process model

Process assembling

Conformance checking Model extension

Online monitoring

Educators

Event Log -MXML format

supported by ProM

Typical forms of requirements in the curriculum

Colored Petri net

Page 4: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

ISDA’09 - EDM

Colored Petri nets

Page 5: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

ISDA’09 - EDM

Classical Petri nets

• Well known and established formalism• Supports all routing constructs (choice, parallelism,

sequence, etc.)• No explicit support for data• Example - Complaints handling workflow:

recordcomplaint

start

contact deprtmt

collectinfo

contact client

decide positive

decide negative

pay

sendletter

filecase

end

place transition(task)

token

arc

Page 6: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

ISDA’09 - EDM

Colored Petri nets

• Extend Petri nets with data information• Data in tokens - Places typed

Decidenegative

INFO

INFO

DECISION

info_c

make_decision(info_c,info_d)Collect

info

Decidepositive

dec

dec

[positive(dec)]

[not positive(dec)]

info_d

Page 7: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

ISDA’09 - EDM

Modeling Academic Curriculum Patterns

Page 8: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

Course - Exam construct

• Models an exam for course C for every student id

• Firing of C adds a new grade to the grade list

• There is a maximum number of attempts

ISDA’09 - EDM

g

(id,g::grds)(id,grds)

C

[length(grds)<MAX_NUM_ATTEMPTS]

[0,1,2,3,4,5,6,7,8,9,10]

GRADE

grades_of_C

[]

GRADES

Page 9: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

Start and End pattern

• Models courses that must be taken first• Starting place of the model• Graduation is always the last course• Example: Start with either C1 or C2

ISDA’09 - EDM

id

idid

GRAD

C2C1 end

STUD_ID

start

STUD_ID

Page 10: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

M-out-of-N pattern

• M courses out of a group of N courses must be passed before some other course can be taken

• Example: Two from {C1,C2,C3} before D

ISDA’09 - EDM

(id,grds3)

(id,grds2)

(id,grds1)D

[check2out3(grds1,grds2,grds3)]

grades_of_C3

[]

GRADES

grades_of_C2

[]

GRADES

grades_of_C1

[]

GRADES

Page 11: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

Dependency pattern

• Result of C is automatically also a result for some other (weaker) course D

• Firing of D adds a grade to the list for course C

ISDA’09 - EDM

grades_of_C

GRADES

[](id,g::grdsc)

(id,grdsc)

D

Page 12: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

Expiration pattern

• Grades stay valid only for some time, i.e. they can expire

• Expiration condition arbitrary

• Firing of GradesExpire remove all the grades of id for course C

ISDA’09 - EDM

(id,[]) (id,grds)

GradesExpire

[expireCond(grds)]

grades_of_C

[]

GRADES

Page 13: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

ISDA’09 - EDM

Application: Conformance checking

Page 14: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

Conformance checking

• Check whether the (manually constructed) model complies with the log (observed behavior)

• Has a curriculum pattern always been respected?

• Possible use: Fraud detection• Supported in ProM for classical Petri nets

ISDA’09 - EDM

(id,grds3)

(id,grds2)

(id,grds1)D

[check2out3(grds1,grds2,grds3)]

grades_of_C3

[]

GRADES

grades_of_C2

[]

GRADES

grades_of_C1

[]

GRADES

80%

Page 15: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

Example 2-out-of-3 pattern check

• At least 2 courses from { 2Y420,2F725,2IH20 } must be taken before graduation.

ISDA’09 - EDM

Page 16: From Local Patterns to Global Models: Towards Domain Driven Educational Process Mining Nikola Trčka Mykola Pechenizkiy

Conclusions

• A framework for mining and analysis of educational data is proposed.

• Main idea: Model/Mine a curriculum as a Colored Petri net using some standard (predefined) patterns

• Applied in a real-world case study using ProM.• Future work:

• Implement the actual mining algorithm, and • enable online monitoring support.

ISDA’09 - EDM