The article continues to explore the problem of classifying the content of an e-learning system. The previously developed technology for classifying mathematical content according to the topics of the studied discipline was applied to two new indicators of content quality: the difficulty level and the set of competencies that the task should develop in students. It required a revision of the processing pipeline. Several measures have been taken to improve tokenization rules, address dataset imbalance, implement feature selection, and increase context length. Multi-label classification methods were used to predict the set of competencies developed by a task. A significantly expanded data set, which contains the tasks in the discipline of the theory of functions of a complex variable, was used to test the performance of the proposed approaches, train classifiers, and analyze the classification quality. The proposed combination of methods for preparing and classifying data can form the basis for an automated assessment of the quality of educational content, which could become a promising direction for the development of e-learning systems.