This paper introduces a comprehensive mining pipeline designed to analyze course feedback comments efficiently. The proposed methodology begins with generating embeddings for student comments, applying Association Rule Mining (ARM) to identify key features, subsequently applying clustering to ascertain optimal k, culminating in the most interpretable topics using Latent Dirichlet Allocation (LDA) and Non-negative Matrix Factorization (NMF) techniques. Our findings reveal that NMF outperforms LDA in terms of topic interpretability and coherence, achieving a notable coherence score of 0.42. This performance enhancement is corroborated through expert manual evaluation, confirming that NMF, combined with precise feature selection and optimal topic quantity, provides a more effective framework for analyzing course feedback.