In natural language processing (NLP), topic modeling has come to be an effective technique for identifying and extracting hidden topics from big textual datasets. However, the quality and consistency of the generated topics are significantly luenced by the preprocessing techniques performed on the text data prior to model training. This study compares Latent Dirichlet Allocation (LDA) with Non-Negative Matrix infFactorization (NMF) to see how preprocessing techniques affect topic modeling performance. We show that preprocessing has a significant impact on topic coherence using an evaluation metric. Our findings indicate that NMF, when combined with TF-IDF transformation, achieves a higher coherence score (0.5750) than LDA (0.4345), underscoring the significance of preprocessing in enhancing topic quality and interpretability