Large Language Models in Intelligent Education Systems: New Educational Perspectives—A Systematic Review

Large language models (LLMs) are an emerging artificial intelligence-driven technology, based on transformer architecture. LLMs are widely used in modern education, both by learners and tutors, as standalone tools or integrated into e-learning systems, where they can support personalization, adaptive learning, automated assessment and feedback, content generation, and intelligent tutoring. LLMs offer many benefits for learners, but they also have significant limitations. One approach to address the limitations of LLMs is to combine them with other intelligent technologies. The primary goal of this systematic survey is to identify appropriate supporting technologies, mechanisms of use, and methodological approaches able to help overcome the limitations of LLMs and support their responsible and effective use in education. For this reason, analysis and discussion of recent scientific research (published over the last four years) accessible through Google Scholar, ACM, IEEE Xplore, or indexed in Scopus or Web of Science (WoS) is performed. A bibliometric analysis of results from the initial general query strings is used to refine and formulate more specific search queries during the literature retrieval process in the selected databases. Full-text exploration of relevant search results serves as a source for critical analysis and deductions leading to the following conclusion: LLMs should be integrated into e-learning systems, combined with knowledge graphs, ontologies, learning analytics, and multimodal reasoning to enhance reliability, improve pedagogical effectiveness, and enable true personalization. New pedagogical approaches are also needed to ensure the effective use of LLMs in both tutoring and assessment contexts. Therefore, the authors propose methodological guidelines for integrating LLMs in complex modular educational systems.