Paper—Analysing the Outcome of a Learning Process Conducted Within the System ALS_CORR[LP] Analysing the Outcome of a Learning Process Conducted Within the System ALS_CORR[LP]

This paper presents the results of an experiment, conducted on a sample of computer science students, using the adaptive learning system called ALS_CORR[LP]. Indeed, unlike the traditional LMS, adaptive learning systems provide a personalized learning experience based on specific objectives, prerequisites and learning styles, generating thereby a specific learning path. However, the main issue resides in the fact that they assume that the generated learning path is supposedly the leading one, which is far from being true, as we can always detect some failure cases during the evaluation phase. In this paper we will conduct a learning experiment using the system ALS_CORR[LP], which has the ability to correct the generated learning path by recommending the most relevant learning objects, and update the learner model based on a calculation of similarity in behavior between the struggling learner and the succeeding ones. We will later analyze the results of behavior tracking within the system. Keywords—Learning style, Learning path, Experiment, Adaptive learning system.


Introduction
Adaptive learning systems make content dynamic and interactive, placing the learner at the centre of the learning experience. The techniques used to provide adaptability have been summarized by [1] into several techniques and methods. The authors of [2] made it clear that those systems operate according to two strategies: We will use some explicit methods, [3], [4], [5] and [6], to gather information about the learner's preferences, learning-wise, using mainly e-questionnaires, where they can express their tendencies. We will use implicit methods, [7], [8] and [9], to monitor how the student interacts with the learning system, grouping data on the interaction of the learner with the learning objects. Despite the adaptation provided in both cases, failure is an undeniable reality in the learning process. In the case of adaptive learning systems, failure amounts to a detection of error in the learner's preferences, or sometimes in the relevance of the learning objects.
In this paper, we will present the results of an experiment conducted on bachelor of computer science students, using the current MERISE method as a subject of learning on the adaptive learning platform called ALS_CORR[LP] [10]. The rest of this paper is organized as follows: In Section 2, we will begin with a presentation of the characteristics of the ALS_CORR [LP] system, exposing its learning scenario and architecture. Later, in Section 3, we will focus on the implementation of the experiment. In Section 4, we will analyze the outcome of the experiment and provide some comparisons, using graphical representations of the results. Then, in Section 5, we will provide a graphical representation of the learning style combination of the related students. In Section 6, we will display some statistics about the versions of the learning objects. Finally, some conclusions will be drawn in the last section.

The Style Adaptive Learning System
With the runaway success of e-learning systems in their abilities to meet specific needs, a new kind of system is booming, namely the adaptive learning system. We are currently witnessing a transition from the one-size-fits-all systems [11] to new ones, which are more interested in a personalization of the learning process. In fact, adaptive learning systems are an important class of the e-learning systems; they usually customize the learning process according to the needs, prerequisites, objectives, etc... of each learner, creating thereby a specific learning path. The problem with these systems is that they assume the generated learning path as systematically the leading one, which is not necessarily the case, as we can still detect several negative results during the evaluation phase.
The ALS_CORR [LP] system presented in this paper has the ability to correct the learning path, using a recommendation system operating in two modes: to recommend the most relevant versions of each learning object by calculating the similarities in behavior between the struggling learner and those who have achieved positive results, provided they have the same initial profile, and recommending later their learning objects. The second mode is to update the initial profile in case the behavior observed in the system does not correspond to the original profile generated initially.

System architecture
The ALS_CORR[LP] architecture [12] operates using the components displayed in the following figure: • The Learner model: it is based on Felder-Silverman's test to detect the learning style of each learner and on the prerequisite test. • The Domain model: it is composed of the learning objects, designed according to the SCORM Standards [13], multiple versions of the same learning object and finally the Content Metadata as defined by IEEE Learning object Metadata. • The Instruction model: it is the pedagogical model, responsible for designing the learning object. • The Adaptive model: it enables the system to assign the learning objects according to the characteristics of the learner's profile. • The Evaluation: it represents a critical part [14] in this adaptive learning system, as it remains the only way to correct the learning path in case it appears not to be the leading one.

Learning scenario
For a first-timer, the learner must respond to a prerequisite test and fill in the test of Felder-Silverman's learning style model known as FSLSM [15], in order to determine the initial profile. The system then assigns an appropriate version [16] of the specific course, based on the preferences expressed in the early profile. In case where the generated learning path is not the leading one, which is translated, assessment-wise, by a negative result during the evaluation phase, the system recommends the most relevant versions of the same course based on the calculation of similarity in behavior between the struggling learner and those who have succeeded during the evaluation phase.

Conducting the experiment
The aim is to use the system ALS_CORR [LP] to conduct an experiment on a sample of bachelor of computer science students, in order to study their behavior inside the adaptive learning system, and more particularly to check the relevance adaptation performed in the system, by comparing the learning model and the observed behavior.
As part of the Design Method module, where the MERISE method of learning takes place, an experiment was conducted on a sample of 53 students of computer engineering degree at the Faculty of Science and Technology of Tangier. To ensure students' engagement with regard to the process, the final score for each learner was calculated, including the results of the evaluation conducted within the system. In the next section, we will present the results relating to the calculation of Felder-Silverman learning styles within the system.
A first analysis of the results registered in the system reveals the following distribution of profiles according to the Felder-Silverman test: Let us start with an analysis of each dimension of Felder-Silverman's profile used in the system. The following graph shows a comparison of the perception dimension.

Fig. 2. "Perception" dimension analysis
The following graph relates to the analysis of the "input" dimension of Felder-Silverman's profile.

Fig. 3. "Input" dimension analysis
According to the two preceding paragraphs, "Sensing" and "Visual" styles are the most dominant among students' learning preferences. Thus, we can point out that there are four possible combinations, depending on Felder-Silverman's profiles: Here are the results regarding the styles {Sensing, Visual}; in other words, these students prefer to work with examples and facts of visual kind (videos, diagrams, images, illustrations). Here is the second combination of possible learning styles, in which there are learners who tend to favor learning with examples and facts, but of textual nature. This category of learners prefers learning through definitions and algorithms, presented as videos, images, illustrations, etc. The latest combination of learning styles focuses on learners who enjoy learning with definitions and algorithms of a textual nature.

Graphical representation of the learning style combination
It is obvious that the combination of learning styles {Sensing, Visual} is the dominant combination in the system. In fact, students in this class promote learning with visual resources (videos, images, illustrations...) while learning objects are in the form of examples and explanations. Below, we will present an overview of the statistics obtained by the system regarding the learning of the MERISE design method. The graph above shows that the video version of the course introduction is the most viewed of all sections. It is also noteworthy that the video versions of the other sections are the most visited too.
Here is an example of statistics on the most visited versions in seconds.

Evaluation Results
The main objective of the evaluation is to make sure that learners did achieve the educational goals set at the start, and to get a summary of, or judgment on, the teaching strategies adopted. This phase is crucial in the system because it provides it with the necessary information, on which it can rely to reassess the relevance of the calculation made in the Domain model.
Here is the result of the assessment regarding the learning of MERISE method by the computer science students. The figure above shows that 75% were able to overcome this phase successfully, against 25% who experienced difficulties during this phase. The former high percentage could certainly be explained by the adaptation that took place within the system. Nevertheless, we can still see cases of failure during the evaluation phase. Therefore, we will put the focus on how the system has performed during this phase.

Learning Path Correction
We will look at the case of students with negative results during the evaluation phase. We take as an example the case of a student "C-L". Here is the result of the similarity analysis by Bravais Pearson Formula, based on the behavior of other students who have passed this stage successfully. The maximum similarity in behavior is up to 0.896257856337. The system recommends the learner to follow the same course, but with versions of the learner with whom this similarity has proved to be at maximum. He can have access to these new versions as shown in the following figure: Access to the course is possible once again, but this time while offering the versions that match the behavior observed in the learner "C-L" in the system. Here is an example of the displayed versions:

Conclusion and Perspectives
The current paper presents the results of a learning process that took place within the system ALS_CORR [LP]. 53 students took a course in the MERISE method. The results obtained during the evaluation phase were very encouraging, to the extent that they showed that 75% were able to validate the learning process, while the remaining 25% have been experiencing some difficulties during the evaluation phase. Students who had difficulties during the learning phase could follow the same course, but with new versions whose relevance has been deduced through the calculation of similarity in behavior between learners in difficulties and other learners with the same original learning style, and who managed to overcome successfully the evaluation phase. The next step is to adapt our adaptive learning system AIS_CORR[LP] to mobile learning, by studying the constraints of the latter which has experienced a very important development.