In higher education, open-text comments from Student Evaluations of Teaching (SET) provide valuable insights into instructional strategies. However, processing these comments can be challenging, leading to limited feedback for instructors. This research aims to develop Natural Language Processing (NLP) strategies to transform student comments into actionable feedback. Two research questions guide this study: 1) How can NLP methods diagnose the effectiveness or mismatch of instruction in higher education? and 2) How can these diagnoses inform personalized recommendations for contextually relevant teaching practices? Using cosine similarity between vector representations of student comments and literature-based statements it is diagnosed the presence of effective teaching practices. This diagnosis will inform personalized feedback recommendations. Preliminary work has used Exploratory Factor Analysis was used to analyze latent dimensions in the comment-statement similarity matrix and results suggest that correlations are linked to pedagogically relevant latent variables. This methodology seems to be a valid strategy for diagnosing the effectiveness or mismatch of teaching practices in higher education. Future research directions include exploring text data representations from different theoretical perspectives on education and investigating the impact and implementation of teaching practices suggested by language models compared to those recommended by human agents.