Universities strive for academic excellence by employing various tools to assess study plans and professor performance. Student surveys are crucial in evaluating and enhancing professors’ performance, offering valuable data to improve teaching practices. Despite their importance, processing open-ended survey questions is challenging, especially on a large scale. Therefore, an automated methodology is needed. This research introduces a Natural Language Processing (NLP) approach using Latent Dirichlet Allocation (LDA), sentiment analysis, and transformer models to evaluate students’ perceptions of their teachers. These models generated embeddings that captured semantic relationships, identifying key topics and themes within a dataset of 7,647 surveys from the first semester of 2022. Characteristics like semester, school, career, subject, and comment length were analyzed. The results highlighted frequent and representative topics, providing valuable feedback to teachers and supporting their professional development. These topics were grouped into key concepts transcending individual teacher traits and organized into dimensions related to quantitative survey questions. This organization aided in adjusting weightings for four evaluated dimensions: virtual learning environment, planning, motivation, and communication. Implementing this model for survey analysis offered insights to enhance teacher evaluation processes and provide targeted professional growth support. The findings facilitated evidence-based decision-making and improved teaching practices in higher education institutions.