Cognitive Profiling

A key challenge in delivering individualized and personalized learning experiences to every learner is to establish detailed individual profiles for each learner. These profiles should reflect the richness of the cognitive and meta-cognitive processes at play in learning, but also encompass factors known to influence the learning experience, such as Neuro-diversity, gender and socio-economic background. 

Cognitive profiling on an online learning platform involves assessing individual learners' Cognitive abilities, learning styles and preferences to deliver a tailored educational experience.

By analyzing data such as quiz results, interaction patterns, and engagement levels, the platform can create a unique cognitive profile for each user. This profile helps in customizing content delivery, recommending suitable resources, and adapting instructional methods to optimize learning outcomes. Personalized learning pathways can enhance motivation, improve retention, and facilitate a deeper understanding of the material, ensuring that each learner receives the support they need to succeed.

A primary source of information for profiling our students is a detailed 75-question survey administered at the beginning of their journey with our Learning Management System (LMS).

To extract profiles from the survey results, we employ a two-step method, combining Exploratory Factor Analysis (EFA) with Latent Profile Analysis (LPA). EFA is a technique that identifies the hidden structure of observed data. It identifies clusters of  highly inter-correlated questions, termed factors.  We can then infer the corresponding cognitive construct targeted by each factor. The number of factors was selected to achieve a balance between maximizing the variance explained in our data and preventing overfitting. An LPA model could then be fitted to the participants' answers grouped by factor.

LPA is a mixture-model-based latent variable approach (specifically, Gaussian mixtures in our study) that seeks to group individuals into unobserved "profiles" based on continuous variables. As most of our questions utilize a 5-point Likert-style scale, LPA is an appropriate technique for unraveling hidden learner profiles. The number of profiles is chosen to minimize the Bayesian Information Criterion (BIC), which provides an objective decision threshold.