An Intelligent Grading System for e-Learning
Environments
Shehab A. Gamalel-Din
Faculty of Computing & Information Technology, King Abdulaziz University, Jeddah, K.S.A.
Abstract. Many educators believe that among effective learning methods are
learning-by-explaining-experience and learning-by-correcting-mistakes through
one-on-one interactions with students. It is not surprising, then, that an effective
way of teaching is to give students immediate feedbacks on exercises and
problems that they have just solved. Unfortunately, such one-on-one teaching
scenarios are becoming increasingly difficult to arrange in e-learning, especially
web-based learning, where the number of participating students is continuously
increasing. Computerized empowered intelligent graders that provide students
with comprehensive explanations on their mistakes and what a correct answer
would be are of urgent need to overcome such difficulties. This research have
introduced the new concept of Intelligent Grading Systems (IGS), designed a
generic framework, and implemented Smart Grader (SG) – a simple proof of
concept prototype. SG is assumed to either work as a separate grading tool or as
an integrated component to e-Learning systems for grading programming and
mathematics problems.
1 Introduction
E-Learning technology offers various e-tools to support all types of education and
training systems. This paper presents one of such tools, the Smart Grader, that
supports one critical activity for all education systems, namely, student assessment.
Due to the lack of enough human resources, full independence on a human grader was
a target, which limits the types of questions to those that could be machine marked. In
such types of questions, the final result is what matters, which means that the
approach/strategy the student followed to reach a specific selection doesn’t matter and
is not assessed. This is just enough for the first three levels of Bloom’s taxonomy, but
doesn’t generally suit the higher levels, e.g., application, synthesizing, analyzing, and
evaluation, especially in those subjects, such as programming and mathematics, where
the assessment of the problem-solving skill is what really matters.
Studies [1,2] have revealed that immediate feedback of tutors in problem-solving
yields the most efficient learning model than those that do not provide such a
feedback. Accordingly, these researches suggested that explicit guidance is required
to improve the efficiency of learning. This observation had motivated the idea behind
this research.
Therefore, this research tries to assess the above hypothesis by introducing a new
concept that we analogously call it Intelligent Grading Systems (IGS). In contrast
A. Gamalel-Din S..
An Intelligent Grading System for e-Learning Environments.
DOI: 10.5220/0004087300030013
In Proceedings of the 1st International Workshop on Interaction Design in Educational Environments (IDEE-2012), pages 3-13
ISBN: 978-989-8565-17-4
Copyright
c
2012 SCITEPRESS (Science and Technology Publications, Lda.)
with current grading systems, an IGS does not only evaluate students based on their
final answers to quizzes and exercises but also on the steps they followed to reach
those answers. A student might reach a correct problem solution; however, the grader
might not accept the followed approach, as it doesn’t match the taught concepts being
tested; or on the other hand, the grader might suggest a better approach, which adds to
the effectiveness of the learning process.
Unfortunately, current e-Learning systems do not usually interfere with the
students during their exercise/test sessions. Therefore, integrating Interactive IGS
with e-Learning environments would overcome such a weakness and improve its
learning effectiveness. An interactive online IGS may interfere with the student in a
testing/tutorial session for several reasons among them are correcting misconceptions,
and highlighting better or more advanced approaches [3], which makes the
assessment process an integral part of the learning process and which adds to its
efficiency. Therefore, an interactive grader should be active, alert, and ignited all the
time during problem solving sessions.
On the other hand, offline IGS would allow a totally new dimension of questions
that had never been thought in distance learning tests. Such questions would usually
require human graders, which in most cases is either practically impossible or very
expensive, especially in case of international exams that are conducted worldwide
simultaneously on the Internet with a huge number of students.
In summary, the Intelligent Graders (IGS) are computerized empowered intelligent
graders. This research introduced the new concept of Intelligent Grading Systems
(IGS), developed a model of IGSs, designed a generic framework, and implemented
Smart Grader (SG)—a prototype for grading Lisp programming problems. SG can
work both offline or as integrated to Intelligent Tutoring Systems (ITS), to provide
more effective learning through grading student tests, correcting mistakes, and
providing advices on better ways of problem solving.
2 Intelligent Grading Systems—The Model
Active help systems (AHS) [4,5], intelligent assistants (IAS) [6,7], and intelligent
coaching systems (ICS) [8,9] monitor their users, predict and evaluate their action
plans, and give them advices on correct or more optimum directions. IGSs have many
things in common with such systems; however, they also differ in many ways. The
following discusses some of the major unique characteristics of an intelligent grader
in contrast to help, assistance, and coaching systems. This discussion is aimed at
highlighting the design directions of an IGS framework.
2.1 Action Plan Recognition in IGS
Generally, students during a problem solving session follow steps of a certain Action
Plan leading to the final answer. Grading systems need to analyze such steps, evaluate
them, and assess the student’s grasping of the concepts being tested for assigning
grades. Generally, those action plans have some unique characteristics that motivated
this research:
4
The goal is already known in advance; it is the solution to the given problem.
Therefore, the grader is not puzzled with what the student is trying to achieve.
Therefore, the space of solutions, and hence action plans, is limited and known.
The student previous knowledge, with regard to those subjects under current
training, will further narrow down the space to a subset of the solution space.
In some cases, students may skip some steps to shorten their answer text. This
will require the grader to try to infer how to relate the given high level steps to
the detailed ones that are not explicitly stated in the answer and that lead to such
results.
Intelligent Graders are either synchronous (online) or asynchronous (offline) as
opposed to the other systems that have to be active and interfering with the user
actions all the times.
Because of these similarities and dissimilarities with AHS, IAS, and ICS systems,
similar techniques are adopted with appropriate adaptations in this research to suit
those IGSs’ unique characteristics.
2.2 IGS as Integrated to an e-Learning Environment
Fig. 1 depicts the different types of interactions of the Intelligent Grader with the
different components of an e-learning environment. Based on whether the Grader is
working under synchronous mode (online problem solving sessions simulating oral
exams) or asynchronous mode (off/online assessment exams), IGS responses will
differ.
Fig. 1. IGS in the Context of e-Learning (A Use Case).
An Asynchronous Grading Scenario. The following is one possible scenario for
Smart Grader interaction with the student:
1 A problem is drawn from a question bank and is given to the student to answer.
IGS
(Smart Grader)
LMS/ITS
Q-Bank
Add Question
Instructor
Define test
Criteria
Examiner
Assessment
Report
Evaluation
Student
Model
Update
Take test
Generate test
Student
5
2 The student writes down the answer in a format that is acceptable and
understandable by the Intelligent Grader. An intelligent editor for the problem
domain is helpful here.
3 IG, would then, syntactically and semantically analyzes the answer trying to
recognize the action plan behind it. The answer is compared to the
knowledgebase of expert’s solutions. Inconsistencies are analyzed and reasons
for errors are identified by the Executable Action Plan Recognition Machine.
Hence, comments and feedbacks are compiled for reporting back to the student
with proper explanations and suggestions.
4 The Smart Grader, then, gives appropriate grade according to its assessment of
how much the student have grasped the concepts being tested in the given
question. This step will be guided by grading framework as described in the
expert’s knowledgebase.
A Synchronous Grading Scenario. Another possible scenario might look like the
asynchronous scenario except that the system is kept active during the answer session.
Although the grader is kept silent with no interference with the student actions, it
keeps analyzing his actions in the background and tries to recognize the strategy
behind them through matching them with previously fed instructor’s solution(s)
strategies. Interference is possible only upon a help request from student. IGS allows
student to request a support: help, hint, or next answer step; the impact of each on
marking will be weighed differently.
Assessment Reports and Evaluation. Different types of information could be
reported:
Annotation report in which explanations, suggestions, correct answers are
annotated to each wrong step.
The instructor’s strategy is reported as a model solution’s strategy.
Overall assessment of the achievement level with reference to course objectives,
knowledge, and other soft skills, e.g., good at solving problems, grasped and
missed knowledge, … etc.
3 The Intelligent Grading Systems (IGS)—A Conceptual
Framework
The Smart Grader uses the AI technique of Action Plan Recognition for its grading
technique. The unique characteristics of action plans as mandated by IGS, simplifies
the job of the automated IGS as it will not be puzzled with what the student is trying
to achieve. This would also limit the space of solution strategies and action plans.
Accordingly, the instructor can design few strategies and detailed action plans for
possible solutions as model answers.
The following sections discuss how IGS utilizes action plan recognition through
introducing what we called it the Executable Action Plan Machine (EAPM). The next
section discusses the structure of APM in terms of the description of the Plan
Hierarchy knowledge-base and how the different APMs might interact.
6
3.1 The Action Plan Hierarchy Knowledgebase
It should be noted that most problems might be solved by several correct ways. The
student needs to follow only one single approach of them. Accordingly, the “Action
Plan Hierarchy” (APH), a modified AND/OR tree, is a hierarchical structure in which
the root node is the goal and all intermediate nodes are the sub-goals and the terminal
leaves are the primitive actions or events. A sub-goal node can also be viewed as a
root of a sub-tree for solving a reduced problem, the sub-goal.
IGS requires the instructor to define an APH for each problem. The root of the tree
represents the problem’s main plan. The next level below the root is either OR or
AND branches but not both: AND branches indicate that this plan requires all the sub-
plans to be done to achieve its goal, and OR branches indicate that there are several
possible correct sub-plans. Similarly, all intermediate tree nodes are recursively
defined. The leaves of the tree are the student’s steps that are required to achieve the
solution. To explain, let us consider the following example problem:
“The student is asked to write down a Lisp program that takes three lists A,
B, C, and manipulates them to return a new list that is composed of the first
element of A and the last two elements of C embedded with the reverse of B
in between. In other words, Write a Lisp function that takes the lists (a b c),
(d e f), and (g h i j) to return (a f e d i j).”
This problem can be solved in different ways. Two solutions with their strategies are
presented in Fig. 2, which presents a portion of the frame structure used to represent
the knowledge regarding the Action plans and Strategies for each problem. The
instructor will be requested, with the aid of tools, to define one such a frame for each
problem.
Noteworthy, the Action Plan Hierarchy is a top-down breakdown structure while
student actions lie at the bottom of the hierarchy and hence, plan recognition goes
bottom-up, which adds to the complexity of the recognition job.
3.2 The Executable Action Plan Machine
The plans of Fig 2 indicate that the solution strategies follow several interacting
threads. Accordingly, students also might follow multiple threads during answering
the question. Some of those threads might be in the correct direction and are actual
sub-plans towards the final solution, while others have nothing to do with the solution
and are only investigative trials in the answering process. Multiple threads are usually
active concurrently and the student swaps from one thread to another. Therefore, this
research investigated the use of Dataflow graphs (DFG) as an active pictorial means
for representing such concurrency behavior. Action Plan Machine (APM) is an
adaptation of DFG to get an executable effect for simulating and, hence, identifying
the actual student behavior during the answering session. The graphical depiction of
action plans by APM is automatically constructed from the APH representing the
problem solutions. Fig 3 depicts the APM for the APH of Fig 2.
The construction process of APM starts from the Action Plan Hierarchy. The
constructs of the APM are only action nodes and directed links. Each action node
represents one strategy step (or sub-strategy) of the problem solution. Those action
7
iv
d
a iii ii
b
c
i
P
S
2
S
1
e
Problem
Breakdown
Plan
Recognition
reverse append cddr nthcdr2 car replacd cons append
Elementary Actions
The Problem:
Write a Lisp function that takes the lists
(a b c),
(d e f), and (g h i j) to return (a f e d i j).
AND/OR Action Plan Hierarchical Structure
Two Solutions
Solution 1:
(cons (car ‘(a b c)) (append (reverse ‘(d e
f)) (cddr ‘(g h i j)) ) )
Solution 2:
(append (rplacd ‘( a b c) (reverse ‘(d e f)))
(nthcdr 2 ‘(g h i j)) )
Strategies
Strategy S1:
a. Extract a from (a b c).
b. Reverse the list (d e f).
c. Extract the sub-list (i j) from (g h i j).
d. Combine 2 lists into ( f e d i j).
e. Combine into the final list (a f e d i
j).
Strategy S2:
i. Reverse the list (d e f).
ii. Combine 2 lists into (a f e d).
iii. Extract the sublist (i j) from (g h i j).
iv. Combine into the final list (a f e d i
j).
Indexing and Referencing Information
Supported Topic: T9
Tested Student Level: intermediate
Node S1.b: (a section for each node)
Reference Material: T5, T2.
Messages: expert: xx, intermediate: yy, …….
……etc….. etc…. etc
Fig. 2. A Problem Case-Frame for a Lisp Example.
nodes are connected together through the directed links to compose an Action Plan
Network (APN). The action nodes in an APN are organized in such a way to reflect
the order of execution for a correct plan. Therefore, both sequences and concurrencies
are natively represented. The inputs to each action node are the parameters (or data)
required for that sub-strategy to be achieved; hence, it might be an external input data
or an intermediate result from another interlinked intermediate sub-strategy.
In fact, a node is an active actor that contains all necessary information for the
grader to perform its job, e.g., messages, responses … etc. Therefore, each action
node is an active mini-grader for the sub-goal it represents.
On the other hand, alternative strategies (OR branches in the Action Plan
Hierarchy) are represented by separate nodes internal to the mother strategy node. The
inputs to the mother strategy node are also inputs to each alternative sub-strategy, and
so are the outputs. Fig
. 4 depicts an Action node for the sub-strategy S1.c or S2.iii of
the example of Fig. 3. Those internal alternative nodes are mutually exclusive.
Noteworthy, the constructed machine can be verified for consistency by verifying
that all inputs to the modeled sub-strategy are consumed by the different action nodes.
Note that some nodes exist in both machines expressing common sub-strategies,
which adds another complexity dimension to the plan recognition process since such
actions will confusingly activate both plans simultaneously.
8
Fig. 3. Example APM. Fig. 4. Nested Alternative APM fo
r
Strategy S1.c and S2.iii.
4 Prototype Implementation—Smart Grader
The Smart Grader (SG) is the prototype implemented to test the IGS framework. It is
partially developed using the .NET C# development environment. The core of SG is
the Smart Assessor responsible for implementing the IGS model.
4.1 The Implementation of the Smart Assessor
Fig. 5 simplifies the overall process followed by the Smart Assessor to fulfill its
function. The Lisp machine is an APM whose nodes are active nodes corresponsing to
Lisp constructs and hence can execute. Student’s answer is first parsed to construct
the Lisp APN. However, a Smart Lisp editor, through which the student writes down
the answer, can also be used and hence, parsing is embedded and the machine would
be implemented incrementally.
Fig 5. The Components of Smart Assessor.
Therefore, the parsed student’s answer is first projected onto the APM structure. A
time stamped token is generated to represent each terminal in the parse tree. The
terminals of the parse tree are considered the input literals of the APM. Therefore, the
algorithm matches those tokens with those literal inputs and, hence places them at the
Instructor‘s
Strategy
Student’s
Strategy
Instructor ‘s
Lisp APN
Student’s
Lisp APM
Testing
Results
Lisp Machine
Construction
Lisp Machine
Construction
EAPM
(g h i j)
cdr
cdr
cddr
nthcdr
(a b c)
(g h i j)
(a b c)
(d e f)
(d e f)
(g h i j)
S1.c
S1.
S1.a
S1.d
S1.e
S2.i
S2.ii
S2.ii
S2.i
9
appropriate places in the APM. The internal nodes of the parse tree represent the
student actions corresponding to the APM’s active nodes. In fact, the projection
process uses existing nodes in the model APM, if they exist; otherwise it generates
new ones and superimposes them on the same APM. In summary, the parse tree is
projected onto the APM machine by inserting the appropriate time stamped tokens to
the appropriate APM nodes.
Once the APM inputs are properly placed and the active nodes are properly
colored, the APM is activated and tokens propagate through the APM network. Once
the inputs of any node of the APM are complete and the node is colored, that node is
activated to generate an output token that propagates through the corresponding
output pipeline (link) to the next node. Token propagation indicates the plan followed
by the student to answer the problem. If that plan followed those nodes of the original
APM till the final goal node, then the answer is correct. Using newly inserted nodes
that don’t finally merge with the original APM would be reported and will require
human evaluation.
4.2 SG Interaction with the Student
Student interactions with SG may take many forms: solution actions and event-
request actions. This is done via a set of interface buttons: Hint, Help, or even go back
to ITS for reviewing certain material. SG working in a synchronous mode (e.g.,
simulating oral exams) detects when the students take longer than expected for the
next move and accordingly issue a time-out event. In addition, the student can request
a report on his/her performance to which the Grader reviews the Working APN and
the token time stamps, and then reports a full detailed history of the student actions
and events during solving a specific problem.
5 Model Validation and Evaluation
To validate the mode of IGS, other examples were tried. A Java programming
problem for calculating Factorial(n) is used. Recursive and non-recursive strategies
with many alternative sub-strategies are considered. Fig. 6 depicts the alternative
hierarchical solution strategies, which are used as the basis for building the APH
Frame.
Factorial (n)
Non-Recursive
Forward multiplication
Two Loop control strategies
Backward multiplication
Two Loop control strategies
Recursive
Two termination condition strategies
Fig. 6. Hierarchical strategic plans for Factorial(n) Java Programming Problem.
10
6 Related Work
Gamalel-Din [9] presented the Smart Coach (SC). He introduced an approach for
action plan recognition, which is more suitable for coaching systems. This approach is
based on architecture of an executable Action Plan Machine (APM) that arranges
planned actions into a DFD network of active nodes. This architecture made the plan
recognition process as well as student advising possibly automated. SC intrudes with
the student work to offer more appropriate solution strategies. SC reports graphical
depictions of both expert’s and student’s solution strategies for more powerful and
effective coaching. SG research is the most inspiring for this IGS research.
On the other hand, many researches focused on supporting the grading process by
providing human graders with supportive tools. For instance, Eftimie et al. [10] had
designed Korect that is a solution for automatic test generation, processing and
grading, with inline answer marks, and answer sheet processing using auto-feed
scanners and OCR detection with an objective of optimizing the process workflow.
Wu et al [11] designed a computer-aided grading support system for spoken English
tests. Answer data are divided into numerous independent data cells. Human graders
are then automatically allocated by the system to do grading jobs and independently
report grading results to the system. Rules are defined to obtain a final result.
Bloomfield [12] designed a system that allows digital grading, by a human, of
traditional paper-based exams or homework assignments, which is intended to reduce
the grading time due to the automation of: flipping to the correct page, summing up
the pages, recording the grades, returning the exams, etc.
Furthermore, other researchers worked on semi-automating the grading. For
instance, Juedes [13] presented a semi-automated Web-based grading system for data
structure courses in which the student is requested to write Java codes that are then
assessed partially automatic through running the code with predefined test cases and
with providing aids for the human grader for evaluating the program design and
documentation.
On the other hand, other researchers focused on student assessment rather than test
grading. For instance, Zhang [14] designed an indicator system for evaluating college
student performance based on student characteristics. They used the analytic
hierarchy process (AHP) and case based reasoning (CBR). Antal and Koncz [15]
discussed the need for computer-based test systems according to the student model,
and hence, proposed a method for the graphical representation of student knowledge.
7 Conclusions and Future Work
This research introduced the new concept of Intelligent Grading Systems (IGS). IGS
monitors students, analyzes their steps in solving problems, predicts and evaluates
their action plans, and gives them advices on correct or more optimum solutions. In
contrast with ITS systems, IGS does not only evaluate students based on their final
responses to quizzes and exercises but also on the plans they had considered in
reaching those responses. IGS’s feedback considers course objectives, e.g., a student
might reach a problem solution correctly; however, the grader might not accept
11
his/her followed approach, as it doesn’t match the taught concepts that were the
subject of the current training being tested.
Accordingly, this research have studied and identified the characteristics of IGS as
opposed to other active support systems. Achieved are models for testing, questions,
and grading, which led to the design of a framework for IGS systems. Finally a
prototype (Smart Grader—SG) was then implemented. SG is supposed to integrate to
ITS systems to provide more effective learning through grading student tests,
correcting mistakes, and providing, on the spot, advices on better solution strategies.
A List programming problem was used for experimentation.
In general, the results achieved by the Smart Grader research project were
promising and encouraging. However, further investigations are still required along
several directions: first, assessing SG’s practical viability through measuring the cost
and time efficiency of the whole process of creating a question bank, preparing a test,
and holding a testing session, and hence, improving the framework design; second,
adding accumulative experience and machine learning component to the framework
so as to detect new correct answers and hence enlarging the knowledgebase; third,
adding dynamic adaptation of the Grader’s behavior; finally, Implementing a more
comprehensive environment with practical considerations, e.g., using the mobile and
cooperative agent technology to give more flexibility and power.
References
1. Corbett and H. Trask, “Instructional Interventions in Computer-Based Tutoring:
Differential Impact on Learning Time and Accuracy”, in the ACM CHI Letter, the
Proceedings of the CHI 2000 Conference on Human Factors in Computing Systems, Vol. 2,
issue 1, April 2000.
2. Corbett, J. Anderson, “Locus of Feedback Control on Computer-Based Tutoring: Impact on
Learning Rate, Achievement, and Attitudes”, Proceedings of the SIGCHI Conference on
Human Factors in Computing Systems, 2001.
3. Y. Shang, H. Shi and S. Chen, “An Intelligent Distributed Environment for Active
Learning”, Journal of Education Resources in Computing, ACM, Vol. 1, No. 2, Summer
2001.
4. W. Hefley, “Helping Users Help Themselves”, IEEE Software magazine, Vol. 11, No. 2,
March/April 1995.
5. G. Knight, D. Kilis, and P. Cheng, “An Architecture for an integrated Active Help System”,
in the Proceedings of the 1997 ACM Symposium on Applied Computing, 1997.
6. C. Gutierrez and J. Hidalgo, “Suggesting What to Do Next”, in the Proceedings of the 1988
ACM SIGSMALL/PC Symposium on ACTES, 1988.
7. J. Davies, A. Gertner, N. Lesh, C. Rich, C. Sidner, and J. Rickel, “Incorporating Tutorial
Strategies Into an Intelligent Assistant”, in the Proceedings of the 6th International
Conference on Intelligent User Interfaces, 2001.
8. H. Shah and A. Kumar, “A Tutoring System for Parameter Passing in Programming
Languages”, in the Proceedings of the 7th Annual Conference on Innovation and
Technology in Computer Science Education, 2002.
9. Shehab Gamalel-Din, “Intelligent Active Coaching – An Executable Plan Approach”,
AUEJ, Vol. 5, No. 4, Sept. 2002.
10. Eftimie, M. Bardac, R. Rughinis, “Optimizing the workflow for automatic quiz evaluation”,
in the Proceedings of the 10th Roedunet International Conference (RoEduNet), 2011, pp.1-
4.
12
11. Yijian Wu; Wenyun Zhao; Xin Peng; Yunjiao Xue, “A Computer Aided Grading System
for Subjective Tests”, In the proceedings of the International Conference on Internet and
Web Applications and Services/Advanced, AICT-ICIW '06, 2006, pp.2.
12. A. Bloomfield, “Evolution of a digital paper exam grading system”, Frontiers in Education
Conference (FIE), IEEE, 2010, pp. T1G-1 - T1G-6.
13. D.W. Juedes, “Experiences in Web-based grading”, 33rd Annual Frontiers in Education,
IEEE, 2003, Vol. 3, pp. S3F: 27-32.
14. Yucai Zhang, “Research on intelligent college student evaluation system based on CBR and
AHP”, In the Proceedings of IEEE International Conference on Service Operations and
Logistics, and Informatics, 2008. IEEE/SOLI 2008, Vol. 1, pp. 39 - 43
15. Margit Antal, Szabolcs Koncz, “Student modeling for a web-based self-assessment
system”, Expert Systems with Applications, Vol. 38, Issue 6, June 2011, Pages 6492-6497.
13