USING STUDENT MENTAL STATE AND LEARNING SENSORY MODALITIES TO IMPROVE ADAPTIVITY IN E-LEARNING

In this paper, we present an innovative solution to improve adaptivity in an e-learning system using Brain Computer Interface (BCI) measures (Attention/Meditation) in order to detect changes in students’ preferred perceptual modes for learning information (VARK model). Our solution is also able to report course units and learning resources that could be difficult for the students


INTRODUCTION
Adaptativelearning systems which focus on users' Learning Styles (LS) have shown to be able to improve the students learning achievements and even their satisfaction during the learning process [1][2] [3][4] [5].
Consequently, many proposals about LS identification have been developed in recent years, which arebased on a) questionnaires answered by the students before whatever interaction, b) the server log file in order to monitor resources that have been accessed by the students, their frequency of use and time spent on them.The last ones are classified in the state of art as automatic ways to identify LS applying, inter alia, Bayesian Networks, Fuzzy Logic, Genetic Algorithms, Neural Networks, Hidden Markov Chains, Decision Trees and statistical tendency measurements to adapt the learning system [6] [7][8] [4][9] [10].
Nevertheless, disadvantages still exist : Many of the preceding works depend of large questionnaires (40 items at least) as a prior step in order to establish the student LS and as is mentioned in [11] [12]; They use very technical language that can cause misinterpretation or misunderstanding because of their size and in some cases students answer them lightly.On the other hand, automatic detection techniques, reported on the literature, are founded on biased data sincethey correspond to activities planned by the teacher for the course; This means that students access an activity or resource just because the teacher planned the use of that resource, so variables as frequency of use and time spenton activities present some uncertainty for identifying student styles.In addition, only afew contributionshave taken into account potential LS changes over the time.It is worth noticethat it has not been proposed any technique that use student internal state (what a student feels at the moment of the interaction).
Based on the foregoing, we propose a technique that takes advantage of users' brain-activity data (attention/meditation) while they interact with e-learning systems.In this manner, we can predict the grade of user engagement (focus/ relaxation) with the interface and detect and monitor the student preferred perceptual mode for learning information.
The article is organized as follows: section 2 presents an introduction to BCI; section 3, describes VARK perception model and the relationship between student engagement and its perceptual mode tendency; section 4 explains the proposed technique to improve adaptivity in an e-learning system and finally, some conclusions and future work are presented.

BRAIN COMPUTER INTERFACES -BCI
BCIs are computational systems that permit interaction between users and the environment by means of their brain activity.This is a new way of communication in which users intentions are sending to external devices such as computers, mobiles, prostheses and wheelchairs, etc. [13] [14] [15] [16].
BCI is an AI system that can acquire brain signals, pre-processes them in order to extract information, identify and gather discriminative information, classify and organize those data and translate them into commands understandable by the connected device [17]

MODAL PREFERENCES IN LEARNING STYLES
A LS is the pattern of behaviour exhibited by the learner in his learning process, which means LS shows the preferred way apprentices used to approach and appropriate knowledge.LS are important because could improve the teaching process doing learning easier for each student.Accordingly, it is extremely important to identify apprentice LS and monitor its changes over the time [11][5].
Considering the state of art in LS, we find the VARK model very interesting in order to use it in e-learning environments, because of common computerlearning resources can be easily transform to Learning Objects compliant with VARK and the patterns of behaviour can be without difficulty translate into guidelines for potential interactions.Thus, we can take advantage of that for improving presentation adaptivity in e-learning environments.

VARK LS Model
The VARK model suggests that learners have a preferred perceptual mode for information inputs and outputs.Accordingly, Visual, Aural, Read/ Write and Kinaesthetic are the possible modal preferences.Table 3 summarizes the features for each type of modal [35].
This model has been widely used because of many learning resources are available in formats that easily could be trace to each type of modality, as well as, the test used to identify the learning profile is very simple, understandable and ease of use [36], [37].

User Engagement and Learning Tendency
There are few studies that have shown the relationship between students' engagement or affective response and their LS tendency owing to the difficulty to estimate that in an e-learning environment.Notwithstanding, technological advances have allowed some interesting studies.The most interesting are: a) [38]it was conducted an experiment in order to investigate the relationship between LS, engagement and Visual Programming Languages.Their conclusion was that Visual style learners exhibited higher engagement labels.For the experiment VARK model LS, Venkatesh's questionnaire for measure the engagement and Scrath worked with log-file server data in order to predict disengagement in an e-learning system using data mining techniques.They found that there is a relation between the potential disengagement situation and student's learning performance (evaluation activities).The disengagement prediction took into account data from learners' interaction such number of accessed pages, time spend per activity, etc.
Considering as well foregoing studies that show the existence of a relation between LS and level engagement, the strength of this proposal is to use Attention/Meditation measurement in order to predict and monitoring learning modal tendency and thus improve adaptivity in an e-learning system.

ADAPTIVITY PROCESS
The adaptivity process comprises three components, one runs on the client side and others on the side server.These components are connected as shown in figure 1.

Modal Preferences
People that easily understand and assimilate information presented in charts, graphs, and other symbolic modes instead of words.People who prefer to use spoken material and talking.Individuals who easy understand and appropriate information from different kinds of texts.People who need to go into direct practice, doing muscular movements or having movement sensations in order to understand.

Main Features
Visual (V)

CONCLUSION
This work describes the use of student's mental data (Attention/Meditation) as way to monitor VARK modal preferences and thus improve adaptivity.For this type of data, the use of measures of central tendency in the Inference-Maker algorithm was appropriate because it allows to identify to what extent the data is grouped or spread around the intervals defined for the target variables.Additionally, BCI data could be used as a metric for the design of learning resources because they allow to establish the extent to which students are engage in learning activities.Multimodal is a combination of VARK, it could be bimodal, or more.In those cases, UI turns into the strongest modal preference for the index page and the others pages levels could take the other modalities.For balance users, it means with low difference between their modalities, the tendency will be selected randomly because that situation point out that user could take advantage of whatever modal option.Link to Cu structure learning resources related to student tendency (as a first priority).Link Cu structure learning resources changes to the next in priority Refresh statistical information on "Difficulty topics" professor panel.Future work will be dedicated to further applying the adaptivity process in a complete courses supported by Moodle platform to identify the relation between high engagement learning resources and effectiveness of students learning process.

CONDITION
1 programming Languagewere used.b)[39]found out that students' emotion (frustration, anxiety, focus, etc.) is affected by the type of materials they use during the learning process, and suggest that students should receive materials compliant with their cognitive learning style in order to stimulate learning interest and performance.This work was tested for Visualizer and Verbalizer styles and used heart rate variability as a measure of emotional state by means of emWave device.c)[40]  studied the relationship between students' LS and EEG data when students perform mental rotation tasks.They could determine by means of different statistical measurements that different brain zones were activated according to the student's LS and its gender.They used EGI 64-channel HydroCel Geodesic Sensor and Fielder-Solomon Inventory for LS.d)[41]

:Tabla 3 .
It is an autonomous component that getsBCI data (Attention/Meditation) form TGC (Neurosky controller), via socket connection.When interaction ends the EEG-Captured create and stores a XML file on the server.It isnoteworthy VARK LEARNING STYLES SUMMARY

Tabla 4 .
ADAPTATION RULESEVENTACTIONThe user interface turns into an iconic map, chart or graphs, using symbols.The user interface turns into a vocal modal.The user interface turns into a mental map, list or table, with strong highlight in text.The user interface needs to have animation and movement, with balance between symbols, text.

Figure 2 .
Figure 2. Multimodality Read/Write -Aural and Recommendation for change the resource.

Table 4
, shows main adaptation rule.Adaptive Maker is able to build the concrete user interface by means of RIA technologies (CCS, JavaScript, and HTML5) and Web Speech API.For best understanding of user interface changes, figure2 and 3shows a prototype of a learning resource for Multimodal Read/Write-Aural tendencie and the recommendation for access another resources according with the model.