<p>Automated collection and analysis of student feedback within Intelligent Tutoring Systems are vital for the continuous refinement of both educational content and software performance, ensuring that learning environments remain responsive to student needs. This study presents the creation and annotation of Serbian student feedback corpora within an Intelligent Tutoring System, intending to enhance both software functionality and educational experiences. The research addresses gaps in existing studies by implementing a transparent and standardized data annotation process, with Inter-Annotator Agreement scores confirming the reliability of the annotation process. The resulting datasets were then processed using fine-tuned multilingual transformer models, with data augmentation techniques enhancing the analysis. Additionally, a few-shot prompting of a large language model was explored to further improve classification accuracy. The experimental results show that fine-tuned transformer models, combined with data augmentation, significantly enhance the accuracy of feedback analysis, achieving performance levels comparable to human annotators and surpassing baseline models. This automated approach to analyzing student feedback provides substantial time and resource savings for educators and software developers, enabling more efficient and timely improvements to both the software and the educational strategies. This work not only contributes to the development of Serbian language resources but also establishes a foundation for future research in Crowd-based Requirements Engineering and Text-based Emotion Detection within educational contexts.</p>