The use of artificial intelligence in university information systems to personalize learning and support teachers

Abstract

The article considers the integration of artificial intelligence technologies into university information systems as a factor in the transition from standardized learning to adaptive personalized trajectories and as a tool for cognitive support of teachers in the context of increasing heterogeneity of the student audience and the bureaucratization of academic work. It is shown that the digital footprints generated in the LMS and related services can be transformed into recommendation mechanisms and predictive models during analytical processing, which dynamically correlate the complexity of the content with the current level of development, reducing cognitive overload and supporting learning motivation. The analytical part is based on depersonalized data from three federal universities for 2021-2023 and combines a longitudinal comparison of educational results, timing of pedagogical operations, questionnaires, correlation and regression procedures and natural language processing methods for interpreting unstructured texts. The adaptive model recorded an increase in the average score at the end of the semester from 74,32 to 81,95, an increase in the share of excellent results, an increase in the index of residual knowledge from 62.18 to 71.04, while reducing the average time for mastering modules from 42,50 to 36,15 hours, as well as a reduction in dropout rates from 12,84% to 7,42; a decrease in academic performance is interpreted as an alignment of results due to timely correction of trajectories. A redistribution of the time budget has been revealed for teachers: work verification is reduced from 14,55 to 6,20 hours per week, administrative workload – from 6,40 to 3,15, while individual counseling increases from 4,15 to 8,80, and research activity – from 5,20 to 8,35. The accuracy of identifying «at-risk» students was analyzed separately: the transition to multimodal data increases the quality of forecasts to an Accuracy of 94,15% and an AUC-ROC of 0.968, with a high specificity of 95,85%. Student satisfaction and loyalty rates increase with the active use of AI services, and anxiety before exams decreases, which is associated with regular feedback and perceived fairness of assessment. The need for explainability of algorithms and ethically sound work with data is emphasized, since the growth of predictive power directly depends on the expansion of the student's digital profile.