Figure 1. Course concept abstraction levels and an example of few user group
domains.
Ossi Nykänen
Tampere university of technology (TUT), Department of mathematics, PO Box
692, FIN-33101 Tampere, Finland.
Email: ossi.nykanen@.cc.tut.fi
Martti Ala-Rantala
Tampere university of technology (TUT), Department of mathematics, PO Box
692, FIN-33101 Tampere, Finland.
Email: martti.ala-rantala@iki.fi
Abstract. Our paper presents a design for a Hypermedia-Based Learning Environment, HBLE for short. It will be an environment for teaching and learning mathematical sciences with distributed interactive hypermedia systems. HBLE will offer tools and methods for course development, teaching, maintenance and different learner-centered studying strategies. The system will also have information acquisition functionality for research purposes.
HBLE may be used either as learning material to supplement a regular university course or by an independent student. In either case, graded examinations will still retain their traditional form. The system will be integrated with our university's student registration and course enrollment databases.
The learning material is presented as a directed acyclic graph called knowledge graph. Information is divided into cells and arcs combining them. The content of one cell teaches one simple issue whereas arcs define the prerequisite relations between cells; to fully understand a cell's content one must understand all prerequisite cells. Learning material may be therefore conveniently presented in a form of a generalized hypertext document with link semantics of its own. Although knowledge graph resembles the idea of a concept map, these two learning aids need necessary not to be the same. Knowledge graph and concept map usually share their cells, but may have different arc topology.
Cells in HBLE comprise five types of material: theory, exercises, theoretical and practical examples as well as tests. These are divided into difficulty levels by the author. Material is composed of (basically) arbitrary combination of text, graphics, simulations, audio, video, interfaces to mathematical software and other media types. In addition, cells store measurement data about the student's progress in cell's subject. The system uses this data in conjunction with the difficulty levels to adapt to an individual student's learning success.
Another form of adaptation is learning strategy planning. This is achieved with special measures of student's contextual learning. Student is offered a selection of strategies, e.g., basic skills improvement and target-oriented studying.
The underlying technology is WWW. On the client side, any Java-enabled Web browser with necessary audio and video plug-ins can be used. The system is intended for use through Web, but it can also be used, with limited functionality, from a CD-ROM. Java applets will be used extensively, for example, for user interaction and illustrative simulations. Learning material database will reside on the server side. HTML pages are created on the fly on the server side and partially by Java applets on the client side. In addition to the learning material, the server stores various usage statistics plus the students' learning progress data. The server side functionality will be implemented using Java and C++.
The high-level architectural design of HBLE is a distributed and modular framework facilitating future extensions like additional media types, enhancements to the graph theory and teaching deductive sciences other than mathematics.
key words: WWW, adaptive learning systems, educational hypermedia
There is an increasing interest in creating WWW-based learning tools and learning environments. Such tools use the general methodology and techniques that have resulted from the work of computer-based education and artificial intelligent pioneers in the overlapping fields of the computer aided instruction (CAI), intelligent tutoring (IT) and adaptive hypermedia (AH). WWW has made the production and deployment of distributed learning material easy and ought to let the designers of such material concentrate on the content and structure of the material.
Learning environments should provide students with tools and methods for motivating and effective self-studying. Basically it means applying student adaptive techniques and learning by doing ideology when creating learning environments. Learning environment should also strive to provide the feedback and feeling of success required for motivated learning just like a human tutor would.
This is where educational software designers come in. Information presentation, user adaptation and collaboration are the fundamental sectors of learning environments and an ideal system should combine all three. The task of creating educational hypermedia is far from trivial since it involves coupling number of different types of theories and techniques such as goal-oriented teaching, hypermedia management and student modeling. One apparent problem has been that sometimes new technologies have been adapted for the sake of themselves, with little or no concern for the actual study process. This has lead to building prototypes with little long-term use.
We are using and revising ideas resulting from previous work at TUT in the field of hypermedia-based educational systems. This work includes Matriisilaskenta I, a Web course on elementary matrix algebra, and Johdatus Korkeakoulumatematiikkaan, an introduction to university-level mathematics.
Our paper presents a design for a Hypermedia-Based Learning Environment (HBLE) designed primarily for teaching university level mathematics. The system will offer tools for authoring learning material, studying (apparently!), and performing research on Web-based learning systems. Our work is in the analysis and design phase and the actual system is yet to be implemented. We describe the system as we are designing it, leaving implementation details open.
Designed as hypermedia-based learning environment, HBLE couples different types of educational hypermedia disciplines and techniques. This includes structuring course content, communication and recognition of different types of user groups.
Course material is presented and arranged in HBLE using hypertext approach. The hypertext presentation is not the result of a straightforward conversion from existing linear lecture notes, but a newly constructed simple semantic network containing the course topics is used instead. Therefore, the course material must be developed in this format from the beginning. Currently material development is based heavily on teaching mathematics but in the future scope of our approach may be broadened into other disciplines as well.
Abstract course is the learning material content comprising the topics and relations between them. A concrete course, or course, corresponds to a real university course. It includes the content information (abstract course), student information (participants), authoring information (staff) and dynamic information (logs, statistics and deadlines). Course may be a stand-alone web course or extension of a regular lecture course. A student instance is a personal copy of a concrete course that stores the student's progress measurement data. (See Figure 1).
Figure 1. Course concept abstraction levels and an example of few user group
domains.
From the student's point of view the learning material is presented as a tree. Information is divided into cells and arcs combining them. Each cell introduces a single topic. Each topic consists one simple issue to be studied, e.g., a definition or a theorem. Arcs define the prerequisite relations between cells; to fully understand the content of a cell one must understand all prerequisite cells.
Course material is internally presented in a form of a directed acyclic graph that we call knowledge graph. By construction, knowledge graph is essentially a special kind of a weighted directed graph. Knowledge graph has certain constrains, though; parallel arcs not allowed, extra arcs in paths between cells are reduced, etc. Arcs and cells have also various kinds of attributes, e.g., for student data, identifiers, labels and visual properties. Figure 2 presents an example of a simple knowledge graph.
Figure 2. A sample knowledge graph.
Each topic is presented as a small hypertext document. Arcs define the prerequisite relations between the topics. For example, in Figure 2, the student must understand the concepts set and scalar to understand vector.
An approach using relatively small, linked knowledge elements has many advantages. First, it makes possible to effectively define specific goals for studying. Second, each student's knowledge of each topic can be estimated separately enabling automated navigation guidance and third, the hierarchical structure of information is stated explicitly.
The basic structure of knowledge graph as stated above, however, is not rich enough to meet the needs of usable learning environment. Information is scattered around knowledge graph with no clear educational guidelines to follow. The structure, logical as it is, does not necessarily support motivated study. In order to provide context and order to studying we introduce virtual sections. Virtual sections present no new information in the system but define logical groups of topics instead. They bind the relevant topics together in a meaningful fashion providing overview of the topics discussed with general background, examples, exercises and tests. Virtual sections may be combined into even larger entities.
Topics and virtual sections refer to abstract documents presenting the actual content of the course. Abstract documents consist of different kinds of hierarchical elements. They are stored in a document database in a general presentation form. Document instances (e.g. displayable documents) are created on demand. Some elements, e.g., examples and exercises, consists of adaptive subelements, which is the basis of content adaptation. Individual elements may include different media types, e.g., static images, video, audio and hypertext. Students receive documents adjusted to their skills while teachers and designers receive more information based on their individual classification and needs.
Collaborative activities, that is, working in a group and communicating with the teacher, are essential for the learning process. Sometimes the designers of computer-based learning environments try to reproduce a class room environment or parts thereof in hypermedia format using, for example, real-time discussion channels. We do not feel that this is always very useful because of the limited presentation and communication capabilities of computer as opposed to personal discussion and presence.
There are several forms of collaboration and communication supported by HBLE. The student may ask the teacher questions about the course. Each concrete course has a discussion forum associated with it. The forum is organized like the customary bulletin boards or news systems. There is a FAQ (Frequently Asked Questions) list per abstract course, which has two roles: first, it is a place to store solutions to common problems until the course content is changed so as to remove these problems, and secondly, it stores solutions to common problems and causes of misconception that cannot be easily remedied. Further, students, identified visitors and teachers may send comments about the course material to the author. Real-time discussion is not supported at this stage.
One method of assessing the student's comprehension of a topic is to have the student answer questions in essay format. The answers are graded by a human teacher and used as a basis for progress measurement just like other assessment methods.
HBLE has two main user groups: students and staff (see Figure 3). Students fall into three categories depending on the way they intend to use the system.
An unidentified visitor is just like a student flipping through the pages of a textbook in a bookstore. This category is essentially for demo purposes, and an unidentified visitor does not participate in any concrete course in any way. The usage of the system is limited: the learning results are not stored, and the use of supplementing software, e.g., Matlab and Maple, may be limited. The collaborative functions are disabled.
An identified visitor is similar to a student who buys and reads a textbook and may take an exam if it is allowed for a particular course (some courses require exercises or projects to be completed). An unidentified visitor may become an identified visitor by registering as one. The progress measurement data is collected for this type of users. Identified visitors are not enrolled on university courses by HBLE. They do not have to be registered at the university.
A student is registered at the university. When he/she enrolls on a course through HBLE, a notification is sent to the university's course enrollment system. The student takes the class according to the standard university procedure and has the right to use the aspects of HBLE that require teacher contribution. Basically, students use HBLE as one form of course material.
Staff users are divided into four categories: authors, teachers, researchers, and administrators.
Authors produce the course content and are responsible for the FAQ lists. They create topics, relations between them and group the topics into virtual sections.
Teachers or instructors make use of the system as teaching material. There is one teacher assigned to each course. The teacher follows the progress of students, answers the questions posed by the students and participates the discussion about the course. The teacher also grades the answers to essay problems.
Researchers use the system for information acquisition purposes. The system collects statistics about the paths students take through the material, the usage of various features of the system, the frequencies of visits to individual topics, etc. The system produces reports out of this data.
Administrator manages the HBLE system, for example, adds users and cleans up after a course is through. This is a super-user role that requires strict security measures.
Since our aim is to design and create an adaptive learning environment with elementary tutoring capabilities, the need for a student model is obvious. Our approach suggests the structure of hypertext topics as a natural starting point for student modeling.
Subject matter, the abstract course, is divided into cells, the smallest knowledge items identified by the user model. Cells refer to individual topics or virtual sections. Each cell is assigned a topic knowledge measure, a scalar representing the student's measured knowledge of the topic.
The basic version of knowledge graph consists of only cells and prerequisite relations between them (see Figure 2). The idea of the prerequisite relation is simple. Cell A is said to be prerequisite to cell B if there exists a path from A to B. (This implies that prerequisite graph is actually tree-like, since circles are obviously not allowed. We keep using the term "graph" anyway for the sake of notational symmetry.) In other words, the prerequisite version of the knowledge graph indicates what topics should be known before studying the topic at hand. Arcs of the graph may be weighted to compensate for the differences in the amount information presented in individual topics. Richer semantic relations may be introduced later when experience is gained in using the system.
The prerequisite semantics makes the guidance and analysis of goal-oriented studying possible. Prerequisite graph is equipped with a comprehension measure, a scalar-valued function based on the topic knowledge measures and the structure of the graph leading to selected cells. Comprehension measure allows analyzing user's contextual state of knowledge at large entities. With the comprehension measure, automated guidance system is able to pinpoint student's strong and weak areas of knowledge and thereby help student to focus his or her efforts on related areas.
Since the validity of the student model is based on the accuracy of the topic knowledge measurements, updating them is critical. In general, human-computer interaction the communication channel is very narrow; computer systems are incapable of effectively using natural languages, they lack the ability to understand nonverbal messages, and so on. This indicates that in order to gain relevant information from the user an exact frame of interaction has to be employed. In our case updating student model is based on controlled exercises defined by student model interface.
Student model interface defines exactly how and when student model is updated. When the set of learning tools available for a student evolves, new features to the interaction between a student and the student model may be added.
Figure 4. Event trace diagram of student running an exercise.
The actual measurable interaction is handled with interaction modules working according to the internal frame defined the student model interface (see Figure 4). Interaction modules are, for instance, adaptive applications specialized in running different types of controlled exercises. This modularity ensures that the interaction modules can be used with a variety of content domains without rewriting any of HBLE components.
The development of the student model is non-monotonic. Student's knowledge state is changed according to the correctness of the student's exercise answers. Correct answers increase the estimate of student's knowledge state according to the difficulty of the exercises while poorly solved exercises clearly below the student's abilities decrease it.
The governing design goal is to create a layered, loosely coupled object-oriented framework that is easily extensible and modifiable. HBLE is expected to remain in use for several years to come as a means of study and research. During this time, the system will be amended in many ways just like any large software system. A well thought out architecture is vital to this end.
HBLE has two main components: the client running in the Web browser and the server running in a Unix workstation. The client is implemented as Web pages, residing in a number of frames, that communicate with each other via a well-defined interface. Any Java-enabled Web browser that supports frames and has the necessary plug-ins installed can be used. The server is implemented as a set of CGI (Common Gateway Interface) scripts written in C++ and Perl. RCS, a popular Unix package for version control, is used for both learning material and software components. Maple and Matlab engines are used for producing exercise material and presentations as well as directly by students through Web front-ends.
Figure 5. High-Level Architecture of HBLE.
HBLE architecture is akin to the three-tier architecture in the database realm (see Figure 5). The server-side databases are hidden by the persistence module, so named as it makes the objects on the client side persistent by implementing their load and save operations. Note that "database" here does not refer to a RDBMS (Relational Database Management System) but rather to storage in general. While active, the objects live in the client-side process space.
Persistence module is the domain model interface that the student model and staff user tools talk to. It models the concepts that the student model and staff tools are based on, accepts requests from both and translates them into database queries. For example, after the student has logged in and chosen the course for the session, the student model requests from the persistence module the course instance for that particular student. The module loads the necessary data from the databases, creates the student instance and sends it across the network. During the session, the actual hypermedia contents of the topics are loaded by the persistence module on demand.
User database stores the user authentication data and privilege information for all user types. Student database stores the students' personal information, the progress measurement data and the private annotations. The course material database stores the abstract courses, that is, the topics and the graphs presenting the prerequisite relations between the topics. Everything else concerning an abstract course is also stored here, e.g., students' comments on the course material. The course database stores the concrete courses: course dates, announcements, bulletin board messages, questions, etc. A course may contain up to hundreds of components that are edited over the time and a large course might have more than one author. Therefore a version control system is an important part of the functionality of the course material database.
Session management takes care the of login and logoff procedures plus the necessary authentication. Student and staff sessions alike are based on the session manager.
The GUI is based on a Web browser. It presents the actual learning material to and interacts with the student. GUI offers functionality for both inter-topic and intra-topic navigation. The former means essentially moving between topics according to the selected strategy or at student's discretion, and the former stands for accessing different types of material within a topic.
The student model stores the student's instance of the course. Student model acts as a filter producing the presentation adapted to the student's individual learning progress. Furthermore, the student model stores and updates a student's progress measurement data and creates on request the study plan based on this data and the strategy chosen by the student. For unidentified visitors, the student model is disabled and the material is presented with fixed filtering.
Each staff user category has a set of tools for performing their tasks. The teachers have tools, e.g., for following the progress of the students and viewing and answering their questions. Researchers have various reporting tools whereas administrators have user and course management tools. For security reasons, all tools are not available through Web.
Creating educational hypermedia and related software involves coupling of different types of theories and technologies. Information presentation, user adaptation and collaboration are the fundamental sectors of learning environments and an ideal system should combine all of them. Today most systems work effectively in one of these areas. A brief view on the field indicates that WWW has become a rather popular test field for running stand-alone curriculums. Sites fall typically in the following categories; Web-based curriculums and non-interactive static courses, individual tutorials and interactive exercises and educational ftp-like resources. Some sites, such as The World Lecture Hall and Mathematics Archives WWW Server, gather links to various WWW course resources providing a good starting point for the casual browser.
Most sites available in the WWW present the subject matter in a book-like fashion sometimes powered with some tutoring capabilities. Sites such as ELM-ART, Mathmania and Interactive Real Analysis present straightforward learning material in an interesting fashion but offer little or no interaction. In addition to individually crafted courses, there are also few large-scale development environments for creating and running Web-based courses. A good example of such an environment is WebCT, a system for creating, authoring and teaching Web courses. The only significant problems with WebCT are that courses created are practically linear, lack adaptivity, and use rather limited range of exercises.
Most of the individual exercises in the Web are based on the intuitive use of HTML forms or Java applets. The main problem with these is that with few exceptions, most of them are created from scratch with little systematic planning. Lists of exercises can be found, e.g., from the site Mathematics Archive: JAVA and Other Interactive WWW Pages . Another type of educational resources is, for instance, Mathematica or Maple resource packages consisting software-specific features.
All the research related to the area of educational hypermedia is not published as HTML or distributed trough URL addresses. From our point of view the most interesting conventional publications deal with formalizing human-computer interaction and structuring information. Publications are not discussed here in detail, relevant references are listed in the end of this document.
One interesting field study of exploratory learning strategies in an average office environment conducted by Rieman (Rieman 1996), however, is worth mentioning. Rieman studied the learning strategies of 14 informants with different backgrounds using applications like word processors, programming tools and email and database programs. Rieman found evidence that task-driven and "just-in-time" approaches are the main strategies when learning under time pressures. This is what most researchers would have expected. The rather striking overall finding was, however, that on-line help systems were inferior to more conventional manuals. The fact that informants preferred paper manuals implies that on-line help systems need revising. This, on the other hand, suggests how the development of goal-oriented learning environments should evolve. Implications are clear: User-Centered Design (UCD) should play major role in designing educational hypermedia.
In this article we have presented a design for hypermedia-based learning environment. Presently, the system is in the design and analysis phase. Our approach combines WWW, a method for learning material structuring, and adaptivity based on student modeling. As the structure of the material is radically different from the customary, linear lecture note structure, a user-friendly set of tools for authors will have to be created. After the first functional version is up and running, it will be used to gather information to aid in further development of the system.
The research is supported by the Academy of Finland, the Finnish Ministry of Education and Tampere University of Technology. Special thanks to our colleagues for their helpful comments.
Brusilovsky, P.: 1996, 'Methods and Techniques of Adaptive Hypermedia'. User Modeling and User-Adapted Interaction 6, 87-129.
Carbonaro A., Maniezzo V., Roccetti M., and Salomoni P.: 1996, ' Modelling the Student in Pitagora 2.0'. User Modeling and User-Adapted Interaction 4, 233-251.
Cockburn, A, and Jones, S.: 1996, 'Which way now? Analysing and easing inadequacies in WWW navigation'. Human-Computer Studies 45, 105-129.
Dillon A., and Watson C.: 1996, ' User analysis in HCI - the historical lessons from individual differences research'. Human-Computer Studies 45, 619-637.
Hohl, H., Böcker, H-D., and Gunzenhäuser R.: 1996, 'Hypadapter: An daptive Hypertext System for Exploratory Learning and Programming'. User Modeling and User-Adapted Interaction 6, 131-156.
Kaplan, G., Fenwick, J., and Chen., J.: 1993, 'Adaptive Hypertext Navigation Based On User Goals and Context'. User Modeling and User-Adapted Interaction 3, 193-220.
Lee, K., Lee, Y., and Berra., P.: 1997, 'Managment of Multi-strctured Hypermedia Documents: A Data Model, Query Language, and Indexing Scheme'. Multimedia Tools and Applications 4, 199-223.
Lucarella, D., and Zanzi, A.: 1996, 'A Visual Retrieval Environment for Hypermedia Information Systems'. ACM Transactions on Information Systems 14/1, 3-29.
Norman, D., and Spohrer, J.: 1996, 'Learner-Centered Education'. Communications of the ACM 39/4, 24-27.
Pohjolainen, S., Multisilta, J. and Antchev, K.: 1996, 'Matrix algebra with hypermedia'. Education and Information Technologies 1, 123-141.
Powell, D.: 1996, 'Group Communication'. Communications of the ACM 39/4, 50-53.
Ragnemalm, E. L.: 1996, 'Student Diagnosis in Practice; Bridging a Gap'. User Modeling and User-Adapted Interaction 5, 93-116.
Rieman, J.: 1996, 'A Field Study of Exploratory Learning Strategies'. ACM Transactions on Computer-Human Interaction 3/3, 189-218.
Soloway, E., and Pryor, A.: 1996, 'The Next Generation in Human-Computer Interaction'. Communications of the ACM 39/4, 16-18.