Adaptive learning

This article is currently being developed as part of an Eduzendium student project. The project's homepage is at CZ:CIS 700 Special Topics 2008. One of the goals of the course is to provide students with insider experience in collaborative educational projects, and so you are warmly invited to join in here, or to leave comments on the discussion page. However, please refrain from removing this notice. Besides, many other Eduzendium articles welcome your collaboration!

Adaptive learning is a relatively new educational method which uses computers as interactive teaching devices. Computers adapt the presentation of educational material according to students' weaknesses, as indicated by their responses to questions. The motivation is to allow electronic education to incorporate the value of the interactivity afforded to a student by an actual human teacher or tutor. The technology encompasses aspects derived from various fields of study including computer science, education, and psychology.

Adaptive learning systems' primary application is in education, but another popular application is business training. They have been designed as both desktop computer applications and web applications.

History

Adaptive learning or Intelligent tutoring has it origins in artificial intelligence movement and began gaining popularity in the 1970s. At that time, it was commonly accepted that computers would eventually have the human ability of adaptivity. In Adaptive learning, the basic premise is that the tool or system will able to adjust to the student/user's best learning method, which results in a better and more effective learning experience for the user. Back in the 70s the main barrier was the cost and size of the computers rendering the widespread application impractical. Another hurdle in the adoption of early intelligent systems was that the user interfaces were not conducive to aiding in the learning process.

It was not until AutoTutor which was developed by the Institute of Intelligent System around the turn of the century that adaptive learning systems got a voice. This was a major step in adaptive learning systems because it added another medium in communication with the end user. According to the founder and lead on the AutoTutor project - Graesser - "Spoken computational environments may foster social relationships that may enhance learning." Also, in some applications audio content is a necessity, such as language learning applications. Today, the number of new adaptive learning system companies is growing steadily as more classroom are becoming computerized and other industries are finding use for the applications of adaptive learning - such as professional development.

Technology/Methodology

Adaptive learning systems have traditionally been divided into separate components or 'models'. While different model groups have been presented, most systems include some or all of the following models (occasionally with different names) :

• Expert model - The model with the information which is to be taught
• Student model - The model which tracks and learns about the student
• Instructional model - The model which actually conveys the information
• Instructional environment - The user interface for interacting with the system

Expert Model

The expert model has to hold information about the material which is being taught. This can be as simple as the solutions for the question set but it can also include lessons and tutorials and, in more sophisticated adaptive learning systems, even expert methodologies for approaching the questions.

Adaptive learning systems which do not include an expert model will typically incorporate these functions in the instructional model.

Student Model

Student model algorithms have been a rich area for researches over the past twenty years. The simplest means of determining a student's skill level is the method employed in CAT (Computer Adaptive Testing). In CAT, the subject is presented with questions which are selected based on their level of difficulty in relation to the presumed skill level of the subject. As the test proceeds, the computer adjusts the subject's score based on their answers, continuously fine-tuning the score by selecting questions from a narrower range of difficulty.

An algorithm for a CAT-style assessment is simple to implement. A large pool of questions is amassed and rated according to difficulty, either through expert analysis, experimentation, or a combination of the two. The computer then performs what is essentially a binary search, always giving the subject a question which is half way between what the computer has already determined to be the subject's maximum and minimum possible skill levels. These levels are then adjusted to the level of the difficulty of the question, reassigning the minimum if the subject answered correctly, and the maximum if the subject answered incorrectly. Obviously, a certain margin for error has to be built in to allow for scenarios where the subject's answer is not indicative of their true skill level but simply coincidental. Asking multiple questions from one level of difficulty greatly reduces the probability of a misleading answer, and allowing the range to grow beyond the assumed skill level should compensate for possible misevaluations.

Richer student model algorithms look to determine causality and provide a more extensive diagnosis of student's weaknesses by linking 'concepts' to questions and determining strengths and weaknesses in terms of concepts rather than simple 'levels' of ability. Because multiple concepts can influence a single question, questions have to be linked to all relevant concepts. For example, a matrix can list binary values (or even scores) for the intersection of every concept and every question. Then, conditional probability values have to be calculated to reflect the likelihood that a student who is weak in a particular concept will fail to correctly answer a particular question. After a student has taken a full test, the probabilities of weakness in all concepts conditional on incorrect answers in all questions can be calculated using Bayes' Law (these adaptive learning methods are often called bayesian algorithms).

A further extension of identifying weaknesses in terms of concepts is to program the student model to analyze incorrect answers. This is especially applicable for multiple choice questions. Consider the following example:

Simplify: 2x² + x³

a) Can't be simplified b) 3x⁵ c) ... d) ...

Clearly, a student who answers (b) is adding the exponents and failing to grasp the concept of like terms. In this case, the incorrect answer provides additional insight beyond the simple fact that it is incorrect.

Instructional Model

The instructional model generally looks to incorporate the best educational tools that technology has to offer (such as multimedia presentations) with expert teacher advice for presentation methods. The level of sophistication of the instructional model depends greatly on the level of sophistication of the student model. In a CAT-style student model, the instructional model will simply rank lessons in correspondence with the ranks for the question pool. When the student's level has been satisfactorily determined, the instructional model provides the appropriate lesson. The more advanced student models which assess based on concepts need an instructional model which organizes its lessons by concept as well. The instructional model can be designed to analyze the collection of weaknesses and tailor a lesson plan accordingly.

When the incorrect answers are being evaluated by the student model, some systems look to provide feedback to the actual questions in the form of 'hints'. As the student makes mistakes, useful suggestions pop up such as "look carefully at the sign of the number". This can again fall in the domain of the instructional model, with generic concept-based hints being offered based on concept weaknesses, or the hints can be question-specific in which case there is really an overlap between the student, instructional, and expert models.

Implementations

Classroom Implementation

Adaptive learning which is implemented in the classroom environment using information technology is ofter referred to as an Intelligent Tutoring System or an Adaptive Learning System. Intelligent Tutoring Systems operate upon three basic principals :

• Systems need to be able to dynamically adapt to the skills and abilities of a student.
  • Environments utilize cognitive modeling to provide feedback to the student while assessing student abilities and adapting the curriculum based upon past student performance.
  • Inductive logic programming (ILP) is a way to bring together inductive learning and logic programming to an Adaptive Learning System. Systems using ILP are able to create hypothesis from examples demonstrated to it by the programmer or educator and then use those experiences to develop new knowledge to guide the student down paths to correct answers.
• Systems must have the ability to be flexible and allow for easy addition of new content.
  • Cost of developing new Adaptive Learning Systems is often prohibitive to educational institutions so re-usability is essential.
  • School districts have specific curriculum which the system needs to be to utilize to be effective for the district. Algorithms and cognitive models should be broad enough to teach mathematics, science, and language.
• Systems need to also adapt to the skill level of the educators.
  • Many educators and domain experts are not skilled in programming or simply do not have enough time to demonstrate complex examples to the system so it should adapt to the abilities of educators.

Distance Learning Implementation

Adaptive Learning systems can be implemented on the Internet for use in distance learning applications, group collaboration, and for web personalization.

Distance learning is a relatively new area of research with regards to adaptive learning. Initial systems without adaptive learning were able to provide automated feedback to students who are presented questions from a preselected question bank. That approach however lacks the guidance which teachers in the classroom can provide. Current trends in distant learning call for the use of adaptive learning to implement an environment with intelligent dynamic behavior in the learning environment.

During the time a student spends learning a new concept they are tested on their abilities and databases track their progress using one of the models. The latest generation of distance learning systems take into account the students' answers and adapt themselves to the student's cognitive abilities using a concept called 'cognitive scaffolding'. Cognitive scaffolding is the ability of an automated learning system to create a cognitive path of assessment from lowest to highest based on the demonstrated cognitive abilities. A current successful implementation of adaptive learning in web-based distance learning is the Maple engine of WebLearn by RMIT university. WebLearn is advanced enough that it can provide assessment of questions posed to students even if those questions have no unique answer like those in the Mathematics field.

Group collaboration is also a hot field in the adaptive learning research area. Group collaboration is a key field in Web 2.0 which extends the functionality of distance learning. Adaptive learning can be incorporated into group collaboration in the following ways: