Different individuals exhibit preferences for specific learning styles, which can evolve with changes in learning contexts or through training. The project addresses the challenge of accurately characterizing an individual's learning style, traditionally done via questionnaires, which can be tedious, intrusive, and rely on self-awareness. This project explores automating the process of learning style detection enabling monitoring the web browsing activities of the subjects.

Why Bother to Explore This Topic?

• Knowing learning preferences is valuable because it allows for more effective learning and engagement. By understanding how someone learns best, educators can tailor instruction, improve retention, catering to students specific needs.

• Traditional methods of identifying learning styles, such as questionnaires, are time-consuming and often inaccurate due to the respondents limited self-awareness.

•  Automating this process offers a way to identify changes in learning styles in real-time without relying on surveys. 

• This approach ensures errors caused by respondents ignorance of their learning preferences are avoided.

Implementation

• Baseline Profiling - To set the baseline the Felder Silverman questionnaire was circulated to undergraduate students in Plaksha university, which helped us with the initial profiling of the learning style of the students, making an attempt to understand the correlation between different dimenions of the ILS questionnaire by Felder-Silverman.

• Data Collection Tool - A Chrome web extension was developed to capture detailed browsing data, including:

  • Domain and sub-domain of visited websites

  • Time spent on each page

  • Content modality classification (visual-heavy vs. verbal-heavy)

  • Scroll rate as an engagement indicator

• Data Analysis: Behavioral data from approximately 120 students were collected and analyzed. A Bayesian probabilistic model was employed to infer learning styles, with priors derived from the ILS questionnaire.

• Ontology Integration: An ontology mapping specific browsing behaviors to learning style dimensions provided the semantic framework for the Bayesian system, linking observed behaviors to corresponding learning styles.

• Bayesian Update: The model continuously updated the probability of each learning style given observed behaviors, producing a nuanced, adaptive profile.

P(Learning-Style∣Observed-Behaviors) =P(Observed-Behaviors∣Learning-Style)⋅P(Learning-Style)

prior: P(Learning-Style) derived from ILS questionnaire

likelihood: P(Observed-Behaviors∣Learning-Style) probability of observing the specific behaviors assuming the student has that learning style.

posterior: P(Learning-Style∣Observed-Behaviors) updated probability of a student having a particular learning style given the observed web-browsing behaviors

Future Scope

• As the part of the application we would like  to capture the behavior using  the principles of saliency, dynamically blur non-relevant content on a web page, allowing users to selectively deblur areas of interest. This would enable us to know which modality within a web-page is of interest to the student.

• Using computer vision based affective computing techniques to detect users engagement levels while dealing with any learning material.

For more details refer to the slides as well as the detailed-report for more detail.

This study motivated me to research further on this topic leading to an article in the Morung Express - (Newspaper prevalently read in North Eastern India).