Analysis of Dynamic Topic Modeling for Textual Data Using A Novel Approach of Hybrid Deep Learning with NMF and NTD
C. B. Pavithra, J. Savitha · International Journal of Basic and Applied Sciences · 2025
In this research, we conduct an in-depth analysis of dynamic topic modeling techniques for real-time and evolving textual data, leveraging methods such as Non-Negative Matrix Factorization (NMF), Supervised NMF (SNMF), Non-Negative Tucker Decomposition (NTD), and hybrid models, Hybrid HDP-CT-DTM, Hybrid DTM-RNN, and proposed Hybrid Convolutional Neural Networks (CNNs) with NMF and NTD. Using the "Advanced Topic Modeling for Research Articles 2.0" dataset, which comprises 14,000 documents, this research paper evaluates the effectiveness of these methods based on perplexity, coherence, precision, recall, F-score, and accuracy. Our findings indicate that the proposed hybrid CNN with NTD model outperforms other techniques across all evaluation metrics, demonstrating superior ability in capturing complex topic structures and maintaining high accuracy. This performance is attributed to the rich feature extraction capabilities of CNNs and the higher-order interaction modeling provided by NTD. This research work highlights the potential of advanced hybrid models for enhancing the quality and interpretability of topic models in dynamic and large-scale textual datasets.