Topic modeling (TM) is a powerful technique for uncovering latent thematic structures within extensive document collections. However, traditional TM techniques often prioritize single objectives, such as maximizing document coherence or topic interpretability, which may overlook the diverse needs of various stakeholders. This paper proposes a new Multi-Objective Topic Modeling framework to address the inherent trade-offs among multiple conflicting objectives in document analysis. The research is specifically applied to a collection of conference papers on Neural Information Processing Systems (NIPS) released between 1987 and 2017, providing a concrete context for the proposed framework. Five phases are applied in this research: the pre-processing phase for cleaning the dataset to be in a more suitable form to use in the following phases, applying the new Sentences Reduction based on Length and weight (SR-LW) technique, which reduces the minor significance sentences based on their length and weight within the corpus, embedding the sentences using Sentence-Bidirectional Encoder Representation from Transformer (S-BERT), Reducing the dimensionality of the sentences embedding by utilizing Uniform Manifold Approximation and Projection (UMAP). Finally, it uses different clustering methods to generate the topics. The performance is comprehensively assessed by measuring the three objectives (coherence, diversity, and execution time) of the generated topics, instilling confidence in the robustness of the research. The experiment has reached the higher values of the three objectives with the Mini-batch algorithm, respectively (0.605, 0.972 in 4.762 s), demonstrating the effectiveness of the proposed SR-LW in producing more coherent, diverse topics in less execution time than traditional algorithms.