AUTHORING ADAPTIVE COURSES ALE APPROACH

AUTHORING ADAPTIVE COURSES ALE APPROACH Milos Kravcik, Marcus Specht Fraunhofer FIT Sankt Augustin Germany Abstract Currently there are just a few adaptive hypermedia authoring tools oriented to education. We develop one of them - the Adaptive Learning Environment 1 (ALE) that has been already used by several universities in the area of design and architecture. In this paper we present an approach how to support the authoring process in the case of adaptive (educational) hypermedia. We describe the representation of content materials, of the knowledge about the domain as well as about the student, interconnection of the knowledge with the materials, we mention the support of template based authoring and finally explain what adaptation strategies we implement. Key Words Adaptive hypermedia, authoring, adaptation strategies 1. Introduction Authoring adaptive educational hypermedia is a complex process. It involves writing linear texts, multimedia development, integration of the texts and multimedia into non-linear hypermedia, definition of metadata, creation of a domain model (index in our case) and its interconnection with the content materials. Apparently this work should be reasonably split among several persons forming a team. Various professions should be involved in such a team pedagogy, multimedia development, computing, design, etc. Nowadays there are not many authoring tools for adaptive educational hypermedia [1]. Most of them have been created by researchers for experimental purposes and therefore these tools are not directly usable by authors who want to create adaptive courses. A major objective is to have such authoring tools that a single teacher could use to prepare an adaptive course. In other words the team members do not have to collaborate directly, but they should be able to share the results of their work among each other. 1 The authoring tool integrated in ALE was developed in cooperation with the bureau42 GmbH as author42 The trend is to identify relevant layers in the authoring model [2] to be able to develop different parts of adaptive hypermedia independently and reuse them. This would mean that not only content assets and learning objects are reusable, but the same is true for descriptive knowledge (including metadata) and procedural knowledge (dynamics) of adaptive courses. According to the ALE approach [3-7] authors create learning objects, structure them (hierarchically and by referential hyperlinks), assign them attributes (pedagogic roles, metadata), and specify concepts as an alternative structure. By means of automatic indexing the learning objects are interconnected with concepts, i.e. the system can find for each concept all its occurrences (including synonyms) in learning objects. Then, a student viewing a learning object or a concept is provided with access to all the related information. In what follows we explain this approach in more detail and also describe how adaptation strategies can be specified to provide adaptive behavior of the system. First we describe our understanding of the authoring process and how the hyperspace structure looks like. Then different aspects of knowledge representation are considered and template based authoring in ALE is illustrated. Finally we explain our adaptation strategies and how do we implement them. 2. Authoring Process Learning object development process [8] supports four learning architectures (receptive, directive, guided discovery, exploratory) and allows for both behavioral and constructivist solutions. The authoring part of the process aimed to develop learning experiences consists of three stages: 1. Granular Analysis: identification of the desired performance outcomes 2. Design and Mine: structuring the solution, capturing the learning objectives, looking for existing solutions 3. Reuse and Develop: developing the content The other two stages of this development process are Delivery and Maintenance. Though they are not the topic 416-811 396

of this paper they should be taken into account in the authoring process as well. Especially, the way how the course is to be delivered is determined by adaptation strategies. There can be alternative strategies specified for dynamic delivery, static delivery, mobile learning, blended learning, etc. In the first stage authors define the course prerequisites and the performance goals that should be met. It can be done informally in a special paragraph however it is recommended to do it also in the Description attribute (which is a part of the General category according to the LOM metadata standard [9]), as this information can be viewed by students before they enroll the course. After that ALE authors can search for existing solutions (according to metadata and other attributes), structure them and develop the content. The process of solution structuring and content development is explained in the next section. 3. Course Structure Generally, authors in ALE can create several types of learning objects, specify relationships between them as well as their attributes (metadata). One of the primary objectives supported by the system is the possibility of content exchange and reusability of learning objects in different contexts. To ensure interoperability with other learning management systems export and import facilities for SCORM and XML formats are implemented. The learning objects in ALE form a hierarchical structure, but they can be interconnected by hypertext reference links as well. Similarly to [10] the following basic types of learning objects are defined in ALE: Course Unit: a top-level learning object with specified outcomes (objectives) what the audience will be able to perform once the course delivery is complete Learning Unit: a node in the course hierarchy, a standalone and reusable unit that can be moved from one course to another, it has an objective that is measurable by a practice or by an assessment, it can also have associated learner skills and required knowledge (prerequisites) Learning Element: a self contained chunk of information with a single learning objective learning elements can be content elements or assessment elements A learning element consists of content assets - content blocks. These are considered as atomic units (text, graphics, audio, video, etc) when composing learning objects. The system should enable creation of equivalent content blocks that share learning objectives but differ in media type, media quality, learning modes, or language. This kind of redundancy is essential if we want to support individual pedagogical approaches in various contexts. 3.1 Attributes Authors can assign certain attributes both to learning objects and content blocks. For each learning object in ALE metadata can be specified according to different standards. Basic metadata schemas that are included in the predefined system are SCORM, LOM, and CANCORE. The metadata help authors during the mining phase of the development process. The attributes in general support the authoring process and influence the adaptation process as explained later on. One of the attributes mentioned above can describe the type (pedagogic role) of a learning object or a content block. Note that the concrete types of learning elements and content blocks can be context dependent and are thus a matter of the system configuration for particular settings. The following is a reasonable proposal of how these types can be chosen in educational applications. According to [10] any learning objective can be classified into one of the five types: Concept a group of objects defined by a single term Fact a statement or data Procedure a sequence of steps Process a flow of events Principle directions for tasks Each of these types has a recommended structure content block types that either must or can be included in a particular learning element type. When the author chooses the learning objective the system can provide a template showing the recommended structure of the corresponding learning element type. To enable practice and assessment special types of learning objects can be provided, like Exercise and Quiz. In [10] the following types of content blocks are mentioned: Definition Example Non-example (not an example) Demonstration Procedure Guideline Introduction Fact Analogy The content block types play a significant role in the adaptation process. The system can adjust the structure of 397

a learning element and its presentation according to these attributes for a certain user model in a particular context. Additionally another record is created that reflects the status of the learning object for the user, for instance with the following meanings: 4. Knowledge Representation learning object O has been requested by the user U test T has been mastered by the user U The knowledge driving the adaptation process is represented in adaptive hypermedia systems both as a domain model and a student model. This is used by the system to provide adaptive link annotations for the student and her tutor. 4.1 Domain Model To model the domain in ALE the network model has been chosen. This is an advanced form of domain model forming a semantic network. Each ALE concept (index term) consists of its description, synonyms, relations with other concepts (the relation types are specified by the author). The system can generate occurrences of the concept in the course in learning elements and in related external documents as well. An index containing all concepts of a particular course can be exported and imported into another course. Because of interoperability with external tools the index can be exported to XML. For instance a CAD system can import the XML version of an index and produce files that can be uploaded into ALE so that ALE can process the occurrences of concepts in these (XML based) files. 4.2 Student Model For student modeling the historic model is employed. It stores all the events related to the student and the status of each learning object regarding the student. According to this status the hyperlinks to learning objects are annotated for the particular student. 4.3 Connecting Concepts with Educational Materials Educational materials are interconnected with the domain model by multi concept indexing. Each learning object can relate to many concepts. Automatic indexing of learning objects simplifies this process. When the author launches re-indexing the system finds for each concept (all the synonyms) its occurrences in all the learning objects of the course. This is full text searching that does not include the attributes (metadata). In this way the system does not distinguish whether a concept is an input (prerequisite) for a learning object or its output (the learning object contributes to the explanation of the concept). However this issue can be easily solved by the means of attributes. Special attributes (e.g. title, keywords) can specify the concepts that are explained in the particular learning object. To comprehend a concept students need first the learning objects that clarify it and then they can discover a broader context of the concept and its relations with other concepts. An event is a record in the database with the information that a particular user performed a certain action with a specific learning object, together with a timestamp when it happened. These kinds of events can be recorded: the user U logged in on 2003-10-22 at 10:33:20 the user U enrolled the course C on 2003-10-22 at 10:43:20 the user U requested the learning object O on 2003-10-22 at 10:53:20 the user U mastered the test T on 2003-10-22 at 10:57:23 Such information can be relevant both for the student wanting to recall the history of her behavior and for the tutor controlling the student s progress. The threshold between the student s privacy and the tutor s control is configurable according to the needs of the users. Figure 1: Enhanced concept based navigation in ALE 398

To support context exploration enhanced concept based navigation (Figure 1) is provided by ALE. Together with the currently displayed learning object all related concepts can be listed, and for each such concept all its occurrences (in learning elements) as well. Alternatively one can observe relationships between concepts and learning elements or between related concepts also on an interactive concept map. These facilities can help the student to comprehend the context relationships and to access the relevant concepts or learning elements in an easy way reducing the cognitive overhead of learners and supporting exploratory learning. <table border="0" cellspacing="0" cellpadding="10" class="definition"> <tr> <td valign="top" align="left"> [AuthorTag ContentBlock] allowtext = "true" [/AuthorTag] </td> </tr> </table> As we have already mentioned there can be two basic alternative structures specified for a course one is the default hierarchical course structure (an analogy of the table of contents in a book) and the second is formed by the concepts (like the index in a book). Students can view the both structures in the course player, together with the current learning object. Should it cause cognitive overhead the student can close any of the additional modules. 5. Template Based Authoring ALE has been considered as currently the most advanced form-based interface among adaptive educational hypermedia systems [1]. This kind of interface is more intuitive for authors than that one provided by markup based authoring tools. In the meantime the ALE authoring interface has been further developed to support template based authoring. This should make the authoring process more intuitive. The system enables reusability of learning objects and content blocks as well as their representation in various multimedia formats. Additionally it allows separation of the content and the layout for both learning elements and content blocks by means of predefined design templates (in HTML and CSS). Figure 2: Creating a new learning object in ALE To create a new learning object the author first chooses its type from a predefined list of templates (Figure 2). A template defines a specific type and structure to keep a certain consistency across the particular installation of the system. Inside a learning element the author creates individual content blocks (text, multimedia, or URL), defines their attributes (options) and specifies the templates (Figure 3). In Options the author can specify how individual content blocks are to be integrated into a coherent learning object. Before a new course is started to be developed the content analysis is usually performed. Then in the template configuration module system administrators can define learning element (paragraph) types and their templates to adapt it to different application domains. Authors can use predefined templates, but these can also be customized and put into the repository so that others can share them. ALE provides an embedded HTML editor operating in the WYSIWYG mode. There is a special (HTML like) language used for tagging of templates. The template system enables definition of the layout as well as restriction of media elements that can be inserted in certain parts of the paragraph (e.g. an image with a specified resolution, a QuickTime movie). The following example shows a simple template for the content block Definition: Figure 3: Learning element authoring in ALE 399

6. Adaptation Strategies Separation of the content and its presentation in electronic documents is not new. The main idea was to let the authors focus on the document semantics (structure and contents) and leave the layout issues for designers. Then the same content can be reused and presented with different styles in different environments. Developing our adaptation strategies we have realized that the content of educational materials on one side and the encoded knowledge driving the adaptation process on the other can in general be created by different persons. Similarly, the descriptive knowledge encoded as certain predefined attributes (e.g. pedagogic roles) and metadata can be generated independently of the procedural knowledge that can be for instance specified in the form of adaptation rules. Definition of the procedural knowledge usually requires algorithmic thinking and perhaps also some programming experience. This skill can not always be expected from authors coming for instance from the area of art and humanities. On the other hand, they should be able to specify the descriptive knowledge in the form mentioned in the Section 3. However these authors might use texts, pictures and other media types created independently by completely other persons. Some of these assets can be just referenced as remote objects without the right to edit them. Therefore we consider it wise to separate the content, the descriptive and the procedural knowledge in adaptive courses. From our point of view, adaptation strategies correspond with the procedural knowledge encoded in adaptive hypermedia systems. Generally speaking an adaptation strategy should specify how individual objects (learning objects and content blocks) will be presented by the system based on their attributes and the current parameters of the student model (e.g. preferences). One way how to do it is assignment of various weights to the objects according to their attributes and the current user model (context), e.g. with respect to the pedagogic role or the media type of the object. Then these weights can be used by the presentation module to select the appropriate objects, to sort them (according to the pedagogic weight) and present them in a suitable way (based on the media weight). Of course, the procedural knowledge does not have to be specified in this way and alternative representations can be employed to express the adaptive behavior of the system. The definition of the procedural knowledge has been elaborated in [11]. Three layers of the procedural knowledge are distinguished the highest layer is formed by the adaptation strategies, which can be specified by the means of the adaptation language, consisting of simple adaptation techniques. In the following section we describe what an adaptation language should be able to perform to support our approach. Note that this is quite a simple language comparing with the language proposed in [11]. 6.1 Implementation To illustrate our ideas concerning adaptation strategies let us outline one possible implementation of them. There are various classifications of learning styles. According to [12] four learning styles can be distinguished: Concrete experience (CE) Reflective observation (RO) Abstract conceptualization (AC) Active experimentation (AE) In a two-dimensional space abstractness vs. concreteness x action vs. reflection there exist four quadrants and each of them specifies one learner style with some typical strengths: Converger (AC, AE) practical application of ideas Diverger (CE, RO) imaginative ability Assimilator (AC, RO) ability to create theoretical models Accommodator (CE, AE) doing things An adaptation strategy can be defined assigning weights to the content block types for each learner style. These weights should specify the relevance of the particular content block type according to its pedagogical role. Similarly, weights can be defined for different media types as well. The following example illustrates a possible assignment of pedagogic weights for two learner styles. Accommodator: Definition 4 Example 10 Non-example 10 Demonstration 10 Procedure 6 Guideline 6 Introduction 8 Fact 6 Analogy 4 Assimilator: Definition 10 Example 4 Non-example 4 Demonstration 4 Procedure 6 400

Guideline 6 Introduction 8 Fact 6 Analogy 10 These weights do not have to be fixed and can be optimized according to evaluation results. The system can adjust them also for each particular user based on the observations of her behavior. For instance collapsing of a content block by the learner will decrease the pedagogic weight assigned to the content block, while un-collapsing will increase it. Based on the pedagogic weights the system can choose the most appropriate content blocks for a particular student in a particular context and sort them. If equivalent content blocks are available, the most suitable one (e.g. according to the media weight) is chosen, taking into account the current context (including the user preferences). To summarize this implementation strategy, it is based on assignment of weights for particular content block (or learning object) attributes (e.g. pedagogical roles or media types) and for a particular user preference attribute (e.g. learning style). According to the concrete context the threshold is set up. Then the filtering mechanism selects the content blocks that are to be presented. The selected ones are sorted according to the weights. Some additional rules can be specified, e.g. one saying that if Introduction is selected it should always be presented as first. We think these rules are from the author s point of view rather simpler than those shown in [11]. But to compare the usability, expressive power and performance of both approaches their evaluation is needed. 7. Conclusion In this paper we have presented our approach to support authoring of adaptive educational hypermedia. One of the main objectives is to enable sharing of partial results between different professions involved in the process. We develop and implement our own adaptation strategies that should simplify the authoring process and support reusability of adaptive courses (or their parts). We believe that especially the authors without computing skills can benefit by these strategies. Currently we are doing an empirical evaluation of the adaptation approach described here, but the results were not ready for this paper. However, the approach corresponds with established standards and recommendations. Concerning the adaptation strategies separation of the content, descriptive and procedural knowledge in adaptive courses seems to be a natural approach successfully applied already in related areas. References [1] P. Brusilovsky, Developing Adaptive Educational Hypermedia Systems: from Design Models to Authoring Tools, Authoring Tools for Advanced Technology Learning Environments (Kluwer, 2003). [2] A. Cristea, A. de Mooij, LAOS: Layered WWW AHS Authoring Model and their corresponding Algebraic Operators, Proc. of WWW 03, Budapest, 2003. [3] M. Specht, M. Kravcik, L. Pesin, R. Klemke, Integrated Authoring Environment for Web Based Courses in WINDS, Proc. of the ICL 2001 Workshop, Villach, 2001. [4] M. Specht, M. Kravcik, L. Pesin, R. Klemke, Authoring Adaptive Educational Hypermedia in WINDS, Online-Proc. the ABIS 2001 Workshop (Adaptivität und Benutzermodellierung in interaktiven Softwaresystemen), Dortmund, 2001. [5] M. Specht, M. Kravcik, R. Klemke, L. Pesin, R. Hüttenhain, Adaptive Learning Environment for Teaching and Learning in WINDS, Proc. of the 2nd International Conference on Adaptive Hypermedia and Adaptive Web Based Systems, Malaga, 2002. [6] M. Specht, M. Kravcik, R. Klemke, L. Pesin, R. Hüttenhain, Adaptive Learning Environment in WINDS, Proc. of the ED-MEDIA 2002 Conference, Denver, 2002. [7] M. Specht, M. Kravcik, R. Klemke, L. Pesin, Information Brokering for the Adaptive Learning Environment, Proc. of the e-learn 2002 Conference, Montreal, 2002. [8] Reusable Learning Object Strategy: Designing and Developing Learning Objects for Multiple Learning Approaches (Cisco Systems, 2003). [9] Draft Standard for Learning Object Metadata (IEEE, 2002). [10] Reusable Learning Object Authoring Guidelines: How to Build Modules, Lessons, and Topics (Cisco Systems, 2003). [11] A. Cristea, L. Calvi, The three Layers of Adaptation Granularity, Proc. of UM 03, Pittsburgh, 2003. [12] D. A. Kolb, Experiential learning: experience as the source of learning and development, (Prentice Hall, 1984). 401