Considering Model-based Adaptivity for Learning Objects

Considering Model-based Adaptivity for Learning Objects

Adaptivity for Learning Objects: Adaptive Hypermedia and Simple Sequencing

Adaptive Hypermedia (AH) and IMS Simple Sequencing (SS) are different approaches but both intend to attain a similar goal: tailored content for learning, just as Abdullah et al. dis-cussed in [1]. However, these two distinct approaches have their own merits and defects. For SS, it takes the conformity with learning objects (LO) as the prime principle, and thus become the main approach to achieve dynamic presentation under the paradigm of using LOs to wrap up learning materials. But due to the absence of explicit domain and user models, SS cannot perform adaptivity in terms of learners’ cognition, such as prior knowledge, learning styles, etc. On the other hand, AH systems focus on constructing explicit models that represent vari-ous aspects of information related to decision making, such as user’s prior knowledge, pref-erences, learning domain, pedagogical knowledge, etc. Therefore, AH systems could perform elaborate decision making based on these models. However, issues like interoperability and reusability remain challenging to researchers in the AH field.

Here we discuss and present our observation on the issue of model-based adaptivity (i.e. AH-oriented approach) for learning objects. That is, we will consider how to bridge the gap between the AH and LO paradigms as described below.

Why Models?

Why should we consider domain and user models for the LO paradigm? We illustrate the need with two points: (1) to support in-depth adaptivity with respect to learners’ cognition (2) to apply technologies developed for intelligent tutoring.

First, an evident cultural difference should be noticed. AH systems take cognitive effect as the main concern. For example, adaptive navigation support aims to share learners’ cognitive load and prevent learners from disorientation. To handle cognitive issues, appropriate domain and user models are necessary. Dependency relations of domain concepts, users’ proficiencies on topics and users’ behavioral patterns are fundamental information to be modeled in general. On the other hand, a typical LO paradigm does not address these issues much1. The presence of SS gradually changes the scenario. But SS still cannot perform adaptivity related to subtle cognitive effects due to the lack of corresponding models.

Second, model-based approach could benefit by various intelligent technologies. Some could be applied to LO paradigm seamlessly. A promising instance is to adopt course se-quencing techniques to generate adaptive presentation in a systematic manner [4]. We will discuss this approach in the next section. Besides, it is also possible to apply machine learning techniques, e.g. theory refinement methods, Bayesian methods, etc. to automate the process of decision making [2]. Especially, when the scale of web-based learning is quite large, “hard-wired” sequencing rules employed by the SS approach might become inflexible and unmanageable. Instead of relying on the laborious process of authoring hand-coded sequenc-ing rules, AI tools based on machine learnsequenc-ing or data minsequenc-ing techniques would be hopeful to induce required knowledge as models.

Technical Issues

Though the areas of AH and LO are developed in parallel most of time, LO paradigm is not totally incompatible with AH systems. To achieve the integration, we notice two different situations: LOs with or without SS.

1 _{More precisely, SCORM focus on “ilities” not QoL (quality of learning) [7]}

If we purely consider LOs without SS, the integration is easier. It is possible to package learning items up as LOs in an AH system. Brusilovsky et al. illustrated the separation of do-main model and learning items in Figure 1 [4]. Nodes of the dodo-main model represent abstract pieces of domain knowledge, i.e. concepts, or topics. Materials to be taught and learned reside in the pool of learning items. By using this scheme, learning items could be wrapped as LOs without doubt. This approach will not change the characteristics that an LO should have, such as reusability and interoperability. In other words, we can still exchange and reuse individual LOs.

A much more complex situation is to consider LOs associated with SS. This type of inte-gration of AH and SS is a relatively new direction. In [5], Chen et al. propose a global domain model to organize a group of activity trees. By analyzing prerequisites and contributed capa-bilities of each activity tree, the system intends to import more cognitive and pedagogical concerns into the SS approach. We observe that this approach could also be reduced to the scheme illustrated in Figure 1 conceptually, in which each “learning item” in the pool refers to an activity tree instead of an LO. However, constructing a global domain model to organize a crowd of activity trees may fall short in practice. Though there is no strict definition of what should be an “activity” of the activity tree, but the granularity of an activity tree is likely to be large enough as a complete course in general. Under this scenario, the power of superposing a global model on-top of activity trees seems relatively weak, because learners may not receive needed adaptation at the best time. The system could detect and give adaptation only after the learner finish a learning session—a whole activity tree. Before that, static sequencing rules determine every thing. In other words, the sensitivity of adaptation is related to the size of chunks (i.e. the granularity of activity trees).

The comparison of these two kinds of integration is summarized in Table 1. No matter which situation is considered, once the linkage between the domain model and learning items is established, we could use overlay modeling technique to represent learners’ prior knowl-edge as a subset of the domain model. Furthermore, various techniques could be used to gen-erate adaptive presentations adapted to learners’ prior knowledge and goal.

Reusability of Models

Figure 1. Bridging the gap between the domain model and learning items, adopted from [4]. Global domain model & LO (without SS) Global domain model & SS Methods employed Domain model

superposed Domain model superposed

Primitive

learning item LO Activity Tree

Sensitivity High Low

Complexity of

implementation Simple Complex

It is obvious that without a suitable standard, or if standardization is not feasible, we cannot exchange domain or user models properly. For domain-independent information about learner traits, the standard is available, such as [6]. But in current AH systems, some modeled rela-tions and features highly depend on the application context or the learning domain. So it is inherently difficult to exchange every aspects of information. Nevertheless, reusing such do-main-dependant information still could be made possible via other means if we can properly define relations between different domains, e.g. the method of distributed user modeling [3].

Conclusions

LO and AH address issues on computer-based learning from different aspects respectively. The LO paradigm successfully sets up a standard to exchange learning materials, and lower the cost of developing courseware, while AH develops substantial techniques to enhance learning.

The integration of these two areas would be quite beneficial for disseminating effective learning experiences. Since LO paradigm is very young, the exploration of the integration of AH and LO paradigm is still at the beginning. To bridge the gap, more efforts from both tech-nical and educational aspects are required further.

Acknowledgement

The first author would like to thank Prof. Jin-Tan Yang at National Kaohsiung Normal Uni-versity for offering discussion and interaction on learning objects issues.

References

[1] N. A. Abdullah, H. Davis, “Is Simple Sequencing Simple Adaptive Hypermedia?” in Proc. of ACM Hypertext Conference 2003 (HT‘03), Nottingham, U.K., 2003.

[2] J. E. Beck and M. K. Stern, “Bringing back the AI to AI&ED,” in Proc. of Ninth Interna-tional Conference on Artificial Intelligence in Education, 233-240, 1999.

[3] P. Brusilovsky, S. Ritter, E. Schwarz, “Distributed Intelligent Tutoring on the Web”, in Proc. of the 8th International Conference on Artificial Intelligence in Education, Kobe Japan, August 1997.

[4] P. Brusilovsky and J. Vassileva, “Course Sequencing Techniques for Large-scale Web-based Education,” International Journal of Cont. Engineering Education and Life-long Learning, Vol. 13(1-2), pp. 75-94, 2003.

[5] C-T. Chen, J-M. Su, S-S. Tseng, H-Y. Lin, C-Z. Chen, and Y-L. Liu, “Adaptive Learning Environment for Pedagogical Needs,” in Proc. of National Computer Symposium 2003, Taichung, Taiwan, 2003.

[6] IMS Learning Information Package Specification 1.0,

http://www.imsglobal.org/profiles/lipinfo01.html

[7] D. R. Rehak, “ADL/SCORM—What Does it Mean for Developers of ICT Projects,”

http://www.lsal.cmu.edu/lsal/expertise/papers/presentations/rldoz05122002/rld05122002. pdf

Hao-Chuan Wang

Computer Science Department National Chengchi University

g9101@cs.nccu.edu.tw

Tsai-Yen Li

Computer Science Department National Chengchi University