Transcript Learning SQL with a Computerized Tutor (Centered on SQL

Learning SQL with a Computerized
Tutor (Centered on SQL-Tutor)
Antonija Mitrovic
(University of Canterbury)
Presented by Danielle H. Lee
Agenda




Problem regarding to learning SQL
Purpose of SQL-Tutor System
Architecture of SQL-Tutor
Evaluation of SQL-Tutor
Problem regarding to learning SQL
 Burden of having to memorize database schemas
(incorrect table or attribute names)
 Misconceptions in student’s understanding of the
elements of SQL and the relational data model in
general
 Not easy to learn SQL directly by working with a
DBMS
 Inadequacy of feedback from a RDBMS
 Example (in Ingres): E_USOB63 line 1, the
columns in the SELECT clause must be contained
in the GROUP BY clause.
 Inability of a RDBMS to deal with semantic errors
Research By the Univ. of Canterbury
 DatabasePlace
 Web portal for database related lectures.
 SQL-tutor: teaches the SQL database query
language
 NORMIT: data normalization tutor
 ER-tutor: teaches database design using the
Entity-Relationship data model
 Constraint-based tutors
Automated Tutoring System
 The School of Computing, Dublin City University
 Developed for an online course name ‘the introduction to
databases’
 To Provide a certain level advice and guide by using
feedback, assessment, and personalized guidance
 Limited the contents to the SQL SELECT sentence.
 The most fundamental of the SQL
 Simple but having the capacity to become quite complex
 There are correction model and pedagogical model.
 Correction model: Multi-level error categorization scheme
according to three aspects (from, where, select)
 Pedagogical model: analyses the information stored by the
student’s answers, it provides feedback, assessment, and
guidance
Purpose of Project
 Personalized ITS for Database Courses
 Personalized tutoring system for learning SQL
 To adapt SQL-tutor technology for use with a
different audience and to explore some ways
to maximize the educational value for every
student.
 Exploration of personalized guidance
technology based on the ideas of adaptive
hypermedia
Purpose of SQL-Tutor system
 To explore and extend constraint based modeling
 Problem-solving environment intended to
complement classroom instruction.
 Problem sets with nine levels of complexity
defined by a human expert
 Students have a assigned educational level and
the level is updated by observing the student’s
behavior.
 Novice, intermediate, or experienced
System Demo
 http://ictg.cosc.canterbury.ac.nz:8000/sqltutor/login
Architecture of SQL-Tutor
Student
Modules
CBM
Constraints
Pedagogical
module
Databases,
Problems,
Solutions
Interface
Student
Constraint-based model (contd.)
 Ohlsson’s theory of learning from errors (1996)
 Error recognition
 Error correction
 Conceptual domain knowledge is represented in terms of over
500 constraints
 Constraints define equivalence classes of problem states
 Equivalence class triggers the same instructional action
 A student’s solution is matched to constraints to identify any
that are violated.
 Neutral with respect to the pedagogy and knowledge domain
Constraint-based model
 Example: specifying the SELECT clause of
a SQL query cannot be empty
Unique No.
(p 2
“The SELECT clause is a mandatory one.
Specify the attributes/expressions to retrieve
from the database.”
(not (null (select-clause ss)))
“SELECT”)
Instructional Message
Part of the constraint
Evaluation
 Computer Science students, Univ. of Canterbury
 Three experiments for evaluation
 First (April 1998): to evaluation how well CBM supports
student learning and to evaluate the interface and
constraint base of SQL-Tutor
 Subject No: 20
 Second (May 1999): to evaluate the effectiveness of
various types of feedback in the system
 Subject No: 33
 Third (October 1999): to evaluate the advanced
pedagogical agent (no explanation)
Results of subjective evaluation
Mastery of constraints
 The degree of mastery of a given constraint
is a function of the amount of practice on
that constraint
 Measured the number of occasions
relevant to each constraint and calculate
the probability of violating a given
constraint.
Evaluation results of learning effects
Result of first experiment
Group
Mean
Std Dev.
Experimental
82.75
8.76
Control
71.23
17.56
Total
76.24
15.39
Kinds of feedback






Positive/negative feedback
Error flag
Hint
All errors
Partial solution
Complete solution
Result of second experiment (contd.)
Result of second experiment
 CBM-based general feedback is superior to
offering a correct solution.
 Among six feedbacks, the initial learning
rate is highest for all errors (0.44) and error
flag (0.40), closely followed by
positive/negative (0.29) and hint (0.26).
The learning rate for partial (0.15) and full
solution (0.13) are low.
Thank you