| The Current Issue |
| Volume 5, Fall, 2004 |
| Archives |
| Volume 4, Fall, 2003 |
| Volume 3, Fall, 2002 |
| Volume 2, Fall, 2001 |
| Volume 1, Fall, 2000 |
| |
| About JECDOL |
| Submission information |
| Editorial Board |
- Faculty Advisor:
Dr. Diane McGrath
|
| |
| ECDOL |
| College of Education |
| Kansas State University |
|
 |
Volume 5, Fall, 2004
Foundations of Case-Based Reasoning and Applications to Hypertext |
|
| |
By Cecil Schmidt |
| |
| Cecil Schmidt is a doctoral student in Curriculum and Instruction with a specialization in educational computing, design and online learning at Kansas State University. Currently Cecil is an Associate Professor of Computer Information Sciences at Washburn University. He teaches Computational Intelligence, Database Management Systems, Contemporary Programming Methods, Data Mining, and Systems Analysis and Design. He is also a computer consultant with past and present clients including several of the fortune 500 companies along with many of the local state agencies in Kansas. His areas of interest include Case Based Reasoning with applications in education. Cecil can be reached at cecil.Schmidt@washburn.edu. |
|
Abstract |
| |
| Case-Based Reasoning (CBR) is a generalization of multiple learning methods that uses remindings of similar situations in order to solve new problems. One of the key issues in CBR is the ability to identify these remindings through alternative indexes. Providing an adequate set of indexes provides a learner the random access required to support advanced knowledge acquisition in domains that are ill-structured. Combining CBR with hypertext can provide an environment that supports constructivism and the cognitive flexibility necessary for advanced learning. This paper starts with a background of CBR, its foundations, and methodology. It is followed by a discussion of its applications in hypertext and how it can support the cognitive flexibility theory (Spiro, Feltovich, Jacobson, & Coulson, 1992). Finally, several examples of CBR systems are provided illustrating the application of CBR in support of learning. |
| |
Introduction |
| |
| Case-Based Reasoning (CBR) is a problem-solving paradigm that utilizes the specific knowledge of previous experiences and concrete problems in order to solve new or similar situations (Aamodt & Plaza, 1994). It is a technique that supports incremental learning in that both our failures and our successes are cataloged for future use. As a cognitive model, CBR provides concrete knowledge versus abstract knowledge. In CBR the concrete knowledge is a set of examples based upon prior experience whereas abstract knowledge is often implemented as a set of rules. Experience is provided by means of a case library that has cause, effect, and lessons learned components. These cases have multiple indexes from which we may access this knowledge. Incorporating an appropriate set of indexes over these experiences provides the learner with alternative views into the same sets of cases. This technique provides support for learning in ill-structured domains, constructivism, and the cognitive flexibility required for advanced learning. This paper provides a brief literature review of CBR and its applications to hypertext . |
| |
CBR in Context |
| |
| Leake (1996) identifies two tenets of nature which underlie the CBR paradigm:
1. Similar problems have similar solutions.
2. The types of problems encountered by a learner tend to recur.
Thus, the primary source of knowledge is a set of cases in memory that can be adapted to new situations. A learner builds upon knowledge of both success and failure. CBR explicitly integrates memory, learning, and reasoning (Kolodner & Guzdial, 2000). This is motivated primarily from cognitive science and learning theory. As the knowledge base grows, the learner's intelligence grows as well. It is a natural evolution. In computer science this type of learning is referred to as lazy learning because the learning is delayed until a new situation is encountered and solved (Mitchell, 1997). Kolodner and Guzdial (2000) point out that CBR can (a) suggest resources through well indexed case libraries and lessons learned, (b) suggest activities through the writing of new cases, (c) suggest ways of moving the learning forward through sharing of the cases, and (d) suggest ways of creating useful case libraries by seeding the library with an initial set of cases and then let it grow.
|
| |
Examples of CBR |
| |
| Some examples of where CBR might be useful are illustrated in the following situations:
1. A teacher is trying to determine if she is breaking any copyright laws when she purchased a single copy of a workbook and wants to make copies of some of the pages for her students. She looks for similar situations and how they were approved or not approved by the authority at her school. Based upon these remindings she makes the decision not to copy the pages.
2. A speech pathologist is trying to evaluate a child's speech disorder that only occurs in high stress situations. He is reminded of several other children that had a similar disorder and reviews how they were treated and their success and failures. Based upon these remindings the speech pathologist recommends a treatment. The success and failure is recorded for future reference.
|
| |
Types of CBR |
| |
| CBR is, in essence, a paradigm covering a set of alternative methods that are based upon the search for and use of remindings. A reminding is nothing more than a past experience. This experience could be a previous conversation, cause-effect remembrance, or a situation in which the learner found herself before. These remindings are stored in a database of cases that can be accessed by the learner in order to solve a new or similar problem. The following is a list of some of the primary types of CBR from Aamodt and Plaza (1994) including a brief discussion of each.
Exemplar-Based Reasoning
Exemplar-based reasoning (EBR) employs nothing more than classification of a new case based on how previous cases were classified. For example, suppose a learner was trying to determine how to classify a movie as G, PG, PG-13, or R. Using the case base the learner could look for similar movies with similar attributes and see how they were classified. Based upon how these were classified, the learner could then classify the new movie similarly. If not all of the movies with similar attributes were perhaps classified the same, for example, some may have been G and some PG, the learner could look at how the majority of the similar movies were classified and use that as its classification scheme. The learner has the final say in the classification process and thus can override the solution. An alternative to this category is in instance-based learning that does not allow for the learner to override the solution.
Memory-Based Reasoning
In memory-based reasoning (MBR) decisions are based upon memories of specific events versus that of using relationships or rules built up from experience (Stanfill & Waltz, 1988). For example, we remember a situation but we have no knowledge of how we got there or its relationship to anything else. An MBR system looks for those cases that match on indexes that are a close to the current situation as possible. The remindings are syntactic in nature versus the rich semantic nature of alternative CBR methods. Thus, computational processes that get the learner the most relevant cases based upon a matching of the index attributes are the ones employed in this method. Providing the cases in a relevance order allows the learner to explore the related cases to determine if they fit. How best to organize the memory separates this from other CBR methods.
Analogy-Based Reasoning
In analogy-based reasoning (ABR) the learner solves the new situations with past experiences from a different domain whereas in CBR the past experiences are from the same domain. For example, in solving new computing applications we often look to biology and human cognition to help create the new solutions. In education we often provide analogies between disciplines in order to support learner understanding. Providing the learner with these cross-domain analogies in a computer program is a quite complex task, but it offers a tremendous area of growth in the field of the CBR paradigm.
Case-Based Reasoning
The generic term case-based reasoning (CBR) used to refer to the case-based reasoning paradigm is often confused with the actual case-based reasoning method. The case-based reasoning method, however, does have a specific meaning as well (Aamodt & Plaza, 1994). The CBR method distinguishes itself in that it employs more than an index structure to the case base. The cases are more complex and are supported by background knowledge. In a computer implementation this background knowledge could be a set of if-then rules or decision trees. Additionally, not all cases in the case base are complete but may be sub-cases that can be rolled together to solve the problem. In Mitchell (1997) several generic properties are listed that distinguish case-based reasoning from exemplar-based reasoning. Summarized from Mitchell (1997), CBR is an instance-based method in which the cases may be rich relational descriptions, the retrieval and recombination process relies on both the case base and general knowledge, and there tends to be a tight coupling in the computer implementation between the retrieval, the reasoning from the general knowledge, and the problem solving.
|
| |
The Learning Cycle and Its Implementation |
| |
The learning cycle in CBR can be generalized to have the following four steps (Aamodt & Plaza, 1994).
• Retrieve the most similar case or cases
• Reuse the information and knowledge in that case to solve the problem
• Revise the proposed solution
• Retain the parts of the experience likely to be useful for future problem solving
Although Aamodt and Plaza (1994) are speaking in computational terms, this process is very analogous to a human's cognitive process for problem solving. Remindings are retrieved to solve new problems. These remindings could cross subject domains and could be used in an analogy process. These retrieved experiences are then reused to solve the new problem. It may include a revision to an existing case or a combination of solutions from more than one case. The solution is then tested by the learner and some level of success or failure is recorded. Based upon this feedback, the case and its solution are tagged appropriately for further reference. This process is additionally supported by the learner’s general knowledge of the problem domain.
Computer implementation of the CBR cycle requires that a knowledge acquisition and engineering process take place first. This includes defining the cases electronically and providing the proper index and search capability. It often requires that experts be interviewed and their solutions to problems be recorded. Additionally, background knowledge is implemented in some fashion such as if-then rules, decision trees, or any number of mechanisms. An interface is developed that allows a learner to query the case base and retrieve cases that are similar to the one that needs to be solved. If similar cases are not found, then others may be generated that are either derived or are a combination of more than one case. This intelligent system may need to access the knowledge base to assist in the derivation. The learner has the option to create whole new cases if necessary. Finally, the computer implementation must provide the mechanism to remember the new situation and how it was eventually solve. This may include both success and failure information.
|
| |
Applications to Cognitive Flexibility Theory |
| |
| Spiro, Feltovich, Jacobson, and Coulson (1992) describe a constructivist theory of learning, Cognitive Flexibility Theory (CFT), which suggests that advanced learning in ill-structured domains must be supported through alternative cases and paths through a set of knowledge content. They suggest that the complexity of these types of domains cannot simply be learned in a single pass. Instead, the learning environment must be flexible with the ability to present the same knowledge of items in a variety of different ways. Their argument entails the notion that simply providing a set of situations and their solutions in order to illustrate a concept is not sufficient to provide support for advanced learning. Instead, the learner must have the ability to construct solutions to new situations from the knowledge content. This ability to adapt and form new content and knowledge can be implemented using theory-based hypertext systems, which incorporate this flexibility. One of the theory-based alternatives is a hypertext system with a built-in case-based reasoner.
Cognitive Flexibility Theory is actually a theory of case-based learning. As such, a case-based learning environment should incorporate features that support the five principles of a CFT as identified by Spiro and Jacobson (1995). These include:
• Use multiple conceptual representations of knowledge
• Link and tailor abstract concepts to different case examples
• Introduce domain complexity early
• Stress interrelated and web-like nature of knowledge
• Encourage knowledge assembly
By the very nature of a system built around case-based reasoning, these principles are upheld. For example, use of analogies to identify similar cases is an example of using multiple conceptual representations of knowledge. An alternative to this might be to use alternate points of view (for example student, instructor, or administrator) as indexes into a case-base. Again this supports the multiple conceptual representation of knowledge. CBR systems provide real-life case examples with their causes and their solutions as opposed to abstract concepts. This technique supports the second principle of linking and tailoring abstract concepts to different case examples. CBR systems provide cases that entail relationships that are not learned in isolation, but rather packaged together. This feature supports the third principle that domain complexity should be introduced early. With multiple indexes into its case-base, a CBR system provides for a web-like nature of knowledge and stresses interrelationships between the cases that support the fourth principle. Finally, a CBR system, by providing multiple alternative cases, supports the fifth principle of encouraging knowledge assembly.
A CBR Hypertext Example Without an Automated Reasoner
An interesting example of how a hypertext environment can be implemented as a case-base reasoning system whose search for and use of examples is driven by the learner and not automated in an intelligent computer program (an automated reasoner) is clearly illustrated by the hyper-book Engines for Educations (Schank & Cleary, 1995). The hyper-book describes a learning theory that suggests learners build knowledge through reasoning on what they remember or their past experiences. What makes this hyper-book unique is the ability to crisscross the material through multiple paths that is a requirement of the cognitive flexibility theory. By providing alternate indexes into the same body of material, Schank and Cleary have cleverly incorporated the techniques described by Spiro, et al. (1992). Additionally, examples or cases have been provided as a support mechanism throughout the text. They are provided as remindings for the learner to use in their construction of new knowledge. This has all been accomplished with no automated reasoner that makes it all the more intriguing.
A CBR Hypermedia Example with an Automated Reasoner
Creanimate is a case-based teaching system that uses hypermedia and an automated reasoner to teach children about biology (Edelson, 1998). It was designed to help elementary school-aged children learn about how animals adapt with particular emphasis on physical features and how the physical features enable them to survive. The case-based reasoning system asks the students questions about creating new animals based upon the remindings. These remindings might suggest why some animals have wings and how it helps them adapt and survive in their environment. It is left up to the learner to create a new animal with the appropriate features that will allow them to survive. The learner can crisscross through the case-base creating new animals and learn biology in the process. Although this is not an implementation in hypertext, it provides us with an alternative that illustrates how a case-based reasoning system can help students learn.
A Proposed CBR Hypertext Example with an Automated Reasoner
A final CBR system example which uses a hypertext interface but also includes some built-in reasoning is an application that is currently in the "proof of concept" stage of development called CBRubric (Unpublished work by Schmidt and Lalonde, 2004). This application is being developed to support the development of assessment rubrics for education. Assessment rubrics are used to measure outcomes of student performance in relationship to goals and objectives of a particular unit or course. Typically these rubrics can be broken down and indexed by a skill, dimension of the skill, a ranking of the skill, and a related textual description of the rank. Below is a hypothetical example:
|
|
UNSATISFACTORY |
BASIC |
PROFICIENT |
Writing |
Responsive To Article |
Presents a response to the ideas presented in the article that is surface and/or lacks in-depth engagement of the ideas presented. Weak presentation of the relationship between the ideas presented in the article and an education related issue. |
Presents a logical response to the ideas presented in the article, and an exploration of the impact of these ideas on an education related issue that is fairly well supported. |
Presents an insightful, logical, and compelling response to the ideas presented in the article and a well-supported exploration of the impact of these ideas on an education related issue. |
|
| |
| In this example there is one skill listed, Writing, with one dimension called Responsive to Article. It has three rankings including unsatisfactory, basic, and proficient. The relationship between the ranks, skills, and dimensions are described in the text. These are referred to as benchmarks.
Assessment of students and the use of the assessment knowledge are very complex and ill-structured. To get an idea of this, just ask a couple of different people in education about their thoughts on it, and you will get many different answers. Thus, the building of a rubric and its use for assessment is a great opportunity for us to use a case-based reasoning system to support this process.
To get an idea of how we might implement such a system and conform to the CFT principles set forth by Spiro and Jacobson (1995), we need to consider the indexing mechanisms, accessing strategies, alternative points of view, and the crisscrossing mechanisms for this application. Much of that work has yet to be done, but here might be some alternatives. For the indexing mechanism it is probably appropriate to provide at a minimum indexes on the skill, dimension, rank, and benchmark. Additionally since rubrics are tied to units or courses, these might be indexes as well. Sample queries into the case-base could provide lists such as giving me a list of the rubrics that test writing skills and are math based. Additionally alternate points of view could be implemented to support the student, teacher, and the administrator. A student might have a different view point of assessment of writing in math versus that of a teacher. Thus the system should provide for that. Accessing and implementation will be through a hyperlinked document with an underlying relational database. Finally, A CBR system should be able to adapt and incorporate new knowledge. Thus, as new rubrics are generated, the system will need to allow for the storage and retrieval of the new knowledge.
|
| |
Concluding Remarks |
| |
| Case-based reasoning is a principle of constructivism that can be used to support advanced learning in an ill-structured domain. Several alternative methods such as exemplar-based reasoning, memory-based reasoning, analogy-based reasoning, and case-based reasoning are based on the case-based reasoning paradigm. Case-based reasoning provides an environment that allows for the crisscrossing of the knowledge content, and it underlies Cognitive Flexibility Theory. As such, hypertext learning environments can be established that incorporate case-based reasoning in a variety of ways. Examples can be provided in an automated way through an advanced search, reuse, and revise mechanism, or they can be provided with clever hypertext indexing schemes. CBRubric (Unpublished work by Schmidt and Lalonde, 2004) is a case-based reasoning application that exploits Cognitive Flexibility Theory providing scaffolding for the development of assessment rubrics. It provides us with a simple foundation to build alternative case-based reasoning systems that can support learning in other educational domains and suggests further research in this area. |
|
| 1 The literature review provided in this paper is part of the dissertation by Cecil Schmidt expected to finish in 2005 (Schmidt, 2005). |
| |
References |
| |
| Aamodt, A., & Plaza, E. (1994). Case-based reasoning: foundational issues, methodological variations, and system approaches. AI Communications, 7(1), 39-59.
Edelson, D., (1998), Learning from stories: an architecture for Socratic case based teaching in R. Schank (Ed.), Inside multi-media case based instruction. Mahwah, New Jersey: Erlbaum.
Jacobson, M. J., Maouri, C., Mishra, P., & Kolar, C. (1996). Learning with hypertext learning environments: theory, design, and research. Journal of Educational Multimedia and Hypermedia, 5(3/4), 239-281.
Kolodner, J. L. & Guzdial, M. (2000). Theory and practice of case-based learning aids. In D. H. Jonassen & S. M. Land (Eds.), Theoretical foundations of learning environment (pp. 215-242 ). Mahwah, NJ: Lawrence Erlbaum Associates
Leake, D. (1996). CBR in context: the present and future. In Leake, D. (Ed.), Case-based reasoning: experiences, lessons & future directions (pp. 3-30). Cambridge, MA: AAAI Press/MIT Press.
Mitchell, T. M. (1997). Machine learning. New York: McGraw-Hill.
Nelson, W. A. (1994). Efforts to improve computer-based instruction: the role of knowledge representation and knowledge construction in hypermedia systems. Computers in the Schools,10(3/4), 371-400.
Schank, R. (1982). Reminding and memory. In Dynamic memory: a theory of reminding and learning in computers and people (Chap. 2). Cambridge: Cambridge University Press.
Schank, R., & Cleary, C. (1995). Engines for education. Retrieved October 18, 2003 from http://engines4ed.org/hyperbook/index.html
Schmidt, C. P. (2005, expected). Hypertext videos and cognitively flexible hypertext in an object oriented programming course: Effects of instructional mode, content, motivation, and background knowledge on student learning. Unpublished doctoral dissertation, Kansas State University, Manhattan, Kansas.
Schmidt, C. P., & Lalonde, D. (2004). CBRubric. Retrieved June 20, 2004 from http://192.104.1.46/cbr/index.html
Spiro, R. J., Feltovich, R. P., Jacobson, M. J., & Coulson, R. L. (1992). Cognitive flexibility, constructivism, and hypertext: random access instruction for advanced knowledge acquisition in ill-structured domains. In T. M. Duffy & D. H. Jonassen (Eds.). Constructivism and the technology of instruction: a conversation (pp. 57-76). Hillsdale, NJ: Erlbaum.
Spiro, R. J., & Jacobson, M. J. (1995). Cognitive flexibility, and the transfer of complex knowledge: an empirical investigation. Journal of Educational Computing Research,12(4), 301-333.
Stanfill, C., & Waltz, D. L. (1988). The memory-based reasoning paradigm. In J. L. Kolodner (Ed.), Proceedings of the DARPA case-based reasoning workshop. CA: Morgan Kaufmann.
|
|
| |
|