Розробка QA-системи підтримки дистанційного навчання в умовах нестабільної інфраструктури

Abstract

The approach utilizes TF-IDF vectorization and cosine similarity to automate responses to student queries, ensuring reliability without generative hallucinations. A specialized Ukrainian language corpus of 800 entries across six categories was created to handle linguistic variability. The system demonstrates high performance on low-end hardware, maintaining educational continuity during power outages. Testing via Accuracy and F1-score metrics confirms the model's effectiveness as a primary support line. This tool reduces teacher workload while preserving their central role in the pedagogical process.