Optimizing dyslexia intervention through an adaptive sequential recommender system

Abstract

Children with dyslexia face significant learning difficulties that require personalized and intensive interventions. Although computer-based support programs exist, they often fail to adapt to the unique needs of each child, representing a major challenge in the field of educational intervention. This article presents a new adaptive sequential guidance system for personalized dyslexia intervention that addresses these limitations. The proposed methodology incorporates several key innovations: (1) a dynamic word generator that creates phonetically modified words and pseudowords from seed words, (2) a three-dimensional matrix structure (E, W, and F) to effectively manage word difficulty and user performance, and (3) a recommendation algorithm based on matrix factorization. To mitigate cold-start problems, the system implements a heuristic initiation process and uses an extension technique to detect difficulties in specific derived words. Additionally, the concept of “virtual children” generated from real data and based on Bayesian Knowledge Tracking is introduced, allowing thorough testing and optimization of the system prior to its actual implementation. The evaluation of the system demonstrates three main results: (1) the use of heat maps and 3D visualization of the E matrix allows identifying specific areas of difficulty for each user, facilitating more targeted interventions; (2) extensive testing confirms the robustness of the system to reduce error rates in multiple trials; and (3) a parametric study evidences the ability of the system to adapt through adjustable parameters, keeping each child in his or her optimal learning zone