Multi-label text data classification (MLTC) datasets are often characterized by high-dimensionality and complex structures, making them difficult to analyze. An effective feature selection (FS) method is essential for enhancing model performance and interpretability in the context of MLTC. This paper proposes a new FS technique called TOPSIS-ONMF, which combines the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) as filter approach with the optimization capabilities of Non-Negative Matrix Factorization (NMF). By leveraging TOPSIS and optimized NMF (ONMF), TOPSIS-ONMF generates optimal feature seubset. To demonstrate its effectiveness, the paper evaluates TOPSIS-ONMF against six baseline methods, highlighting the significance of this new approach in navigating the challenges associated with MLTC feature selection. Overall, the results demonstrate the superiority of TOPSIS-ONMF in obtaining optimal feature subset for MLTC.