Development and evaluation of an algorithm for personalizing educational trajectories using artificial intelligence methods

Abstract

This article presents the development of a comprehensive system for personalizing educational trajectories based on artificial intelligence and deep learning methods. A four-level microservice architecture of the system with a dedicated Python service for ML/AI components is proposed. A five-phase personalization algorithm has been developed with three specialized machine learning modules: Deep Knowledge Tracing (DKT) with Long Short-Term Memory (LSTM) networks and an attention mechanism for tracking knowledge evolution, a hybrid architecture combining DKT and Bayesian Knowledge Tracing (BKT) for interpretable predictions, a multi-task performance predictor with an early warning system, and a hybrid recommendation system combining collaborative and content filtering with context-dependent learning. Modern neural network architectures are applied: LSTM with two layers and dropout 0.3, multi-head attention with four heads, multi-task prediction heads for regression and classification. Specialized loss functions are implemented: Bayesian Personalized Ranking (BPR) for implicit feedback and a multi-task function with weighted components. The practical significance lies in the ready-to-implement open-source architecture of ML components, which provides real-time trajectory adaptation with a prediction accuracy of over 77% and a 30% reduction in the risk of dropout.