Learner modelling in existing standards is addressed at a rather coarse-grained level, although all related specifications have explicit provisions for the evolution of a learner’s model, or profile, over time. An example of specifications in this strand is the IMS Learner Information Package specification, which incorporates the results of “top-level” educational activities, in addition to relatively static information about the user (e.g., demographic).
Although this information is of paramount importance for e-Learning systems, the coarse-grained level of detail renders them of limited use in the context of ALEs. The main underlying problem is that ALEs require a “history” of the user’s interactions, in order to be able to tailor themselves to the particular needs of the individual user. Furthermore, this “history” is more often than not closely associated with the domain model itself (e.g., the course model). Consider, for instance, the very common desideratum (in ALEs) of basing adaptations no the user’s familiarity with a given concept. This requires the establishment of a new set of relationships, which codify a learner’s “status” with respect to a learning entity or concept. Such relationships may refer to directly observable learner behavior (e.g., whether a learner has read, or has not read a node in the learning material), or to inferred status drawn from multiple sources, including results of exercises, etc. (e.g., knows, does not know, or is ready for).
Arguably, the only standard available today that has extended provisions for modelling fine-grained user activities is PAPI. The PAPI standard reflects ideas from intelligent tutoring systems where the performance information is considered as the most important information about a learner, and also stresses the importance of inter-personal relationships (Vassileva et al., 2003). The strengths of PAPI in relation to ALEs stem from its support for representing learner activities in quite structured manner and in as great detail as necessary. Further to the above, PAPI provides a variety of bindings (multiple codings, APIs and protocols), which facilitate its employment in different scenarios within ALEs.
Although PAPI might be more appropriate for modelling users in the context of adaptive ALEs (as compared, for example, to IMS LIP), it is far from being adequate in all its dimensions. Dolog and Nejdl (2003), for example, report on recent work carried out in the context of the EU/IST Elena project, towards the development of the first version of a learner profile to support simple personalization techniques. To cater for omissions or weaknesses in each individual standard (as identified through scenario development and analysis), the RDF-based learner profile they propose is based on subsets of both IMS LIP and PAPI.
Whichever standard (or combination of standards) one might use as a basis for standardisation in ALEs, there exists an additional issue that needs to be addressed. Specifically, it would be necessary to agree upon ways of deriving portions of the learner model from the domain / course model (at least for as long as the learner is “taking” a course), as well as upon when and how such detailed information gets “summarised” into the more coarse-grained models that exist today. This is of particular importance in the case of ALEs that employ what are known as “overlay” models, to relate the learner’s current progress in a course, with the course model itself.
A newly created course is characterized by its authors as “fast” and “introductory”. Nevertheless, in practice, students need to spend three times the anticipated time and effort before they can get an acceptable level of familiarity with the material; additionally, upon completion, students are capable of solving problems from an associated repository at all levels of difficulty. It should be clear that selecting this course purely on the basis of its associated metadata might lead to serious mistakes (e.g., in the process of content filtering). Adding information from its actual use provides a more “informed” view of the course and has the potential to lead to better personalization as a direct consequence.
Maintaining detailed information about a user’s activities within an ALE also gives rise to a new opportunity in terms of group identification and modelling. Specifically, if one can refer to learner activities in a standardised way, then one can also identify dimensions of activities that should be used as predictors or measures for determining group membership. For example, one could identify that learners are to be grouped along the dimension “willingness to interact with peers”, which is to be inferred from (among other things) the user’s active participation in on-line discussion fora.
Unlike the case of learner modelling, group modelling as discussed in this paper is only cursorily covered by existing standards. In fact, PAPI seems to be the only specification that provides sufficient support for describing the characteristics of groups. However, the very features of PAPI that constitute its strengths in the case of learner modelling, turn into potential stumbling blocks in the case of group modelling: PAPI is mainly oriented towards the activities / performance of individuals or groups; however, it makes no explicit provisions for describing the characteristics / attributes that are shared between the group participants. As a result, semantic information over what actually qualifies a person as a member of a group can only be indirectly modeled.
Although this information is of paramount importance for e-Learning systems, the coarse-grained level of detail renders them of limited use in the context of ALEs. The main underlying problem is that ALEs require a “history” of the user’s interactions, in order to be able to tailor themselves to the particular needs of the individual user. Furthermore, this “history” is more often than not closely associated with the domain model itself (e.g., the course model). Consider, for instance, the very common desideratum (in ALEs) of basing adaptations no the user’s familiarity with a given concept. This requires the establishment of a new set of relationships, which codify a learner’s “status” with respect to a learning entity or concept. Such relationships may refer to directly observable learner behavior (e.g., whether a learner has read, or has not read a node in the learning material), or to inferred status drawn from multiple sources, including results of exercises, etc. (e.g., knows, does not know, or is ready for).
Arguably, the only standard available today that has extended provisions for modelling fine-grained user activities is PAPI. The PAPI standard reflects ideas from intelligent tutoring systems where the performance information is considered as the most important information about a learner, and also stresses the importance of inter-personal relationships (Vassileva et al., 2003). The strengths of PAPI in relation to ALEs stem from its support for representing learner activities in quite structured manner and in as great detail as necessary. Further to the above, PAPI provides a variety of bindings (multiple codings, APIs and protocols), which facilitate its employment in different scenarios within ALEs.
Although PAPI might be more appropriate for modelling users in the context of adaptive ALEs (as compared, for example, to IMS LIP), it is far from being adequate in all its dimensions. Dolog and Nejdl (2003), for example, report on recent work carried out in the context of the EU/IST Elena project, towards the development of the first version of a learner profile to support simple personalization techniques. To cater for omissions or weaknesses in each individual standard (as identified through scenario development and analysis), the RDF-based learner profile they propose is based on subsets of both IMS LIP and PAPI.
Whichever standard (or combination of standards) one might use as a basis for standardisation in ALEs, there exists an additional issue that needs to be addressed. Specifically, it would be necessary to agree upon ways of deriving portions of the learner model from the domain / course model (at least for as long as the learner is “taking” a course), as well as upon when and how such detailed information gets “summarised” into the more coarse-grained models that exist today. This is of particular importance in the case of ALEs that employ what are known as “overlay” models, to relate the learner’s current progress in a course, with the course model itself.
A newly created course is characterized by its authors as “fast” and “introductory”. Nevertheless, in practice, students need to spend three times the anticipated time and effort before they can get an acceptable level of familiarity with the material; additionally, upon completion, students are capable of solving problems from an associated repository at all levels of difficulty. It should be clear that selecting this course purely on the basis of its associated metadata might lead to serious mistakes (e.g., in the process of content filtering). Adding information from its actual use provides a more “informed” view of the course and has the potential to lead to better personalization as a direct consequence.
Maintaining detailed information about a user’s activities within an ALE also gives rise to a new opportunity in terms of group identification and modelling. Specifically, if one can refer to learner activities in a standardised way, then one can also identify dimensions of activities that should be used as predictors or measures for determining group membership. For example, one could identify that learners are to be grouped along the dimension “willingness to interact with peers”, which is to be inferred from (among other things) the user’s active participation in on-line discussion fora.
Unlike the case of learner modelling, group modelling as discussed in this paper is only cursorily covered by existing standards. In fact, PAPI seems to be the only specification that provides sufficient support for describing the characteristics of groups. However, the very features of PAPI that constitute its strengths in the case of learner modelling, turn into potential stumbling blocks in the case of group modelling: PAPI is mainly oriented towards the activities / performance of individuals or groups; however, it makes no explicit provisions for describing the characteristics / attributes that are shared between the group participants. As a result, semantic information over what actually qualifies a person as a member of a group can only be indirectly modeled.
No comments:
Post a Comment