Abstarct: The rapid growth of scientific publications and the increasing need for researchers to quickly comprehend the content of multiple related documents have highlighted the importance of developing effective methods for extracting and summarizing integrated and coherent knowledge. The central rationale of this study is baxsed on the notion that, in scientific texts, conceptually related information is often scattered across several shared topics. Leveraging this latent topical structure can enable multi-document summarization to be performed in a more focused, concise, and coherent manner. The primary limitation of baxseline multi-document summarization methods is that, without considering the topical structure across documents, a considerable amount of redundant information is produced, leading to reduced thematic coherence in the generated summaries. To address this issue, the proposed method first applies topic modeling to extract the thematic structure of the documents. Subsequently, through topic segmentation, the abstract texts of the papers are divided into topical segments. Segments sharing the same topic—obtained from both the reference abstracts and the main abstract of each sample in the Multi-XScience dataset—are then merged to form topic-specific documents. Each topic document is further processed through text preprocessing and single-document extractive summarization to emphasize key information and eliminate unnecessary content. Finally, the collection of topic documents is fed into the PRIMERA multi-document summarization model to generate the final abstractive summary while preserving topical coherence. Experimental results on the Multi-XScience dataset demonstrate that the proposed approach achieves significant improvements over the baxseline across evaluation metrics. Qualitative analysis further confirms that the generated summaries exhibit reduced redundancy, broader coverage of scientific concepts, and improved thematic coherence. These findings indicate that incorporating topic modeling into the baxseline frxamework can be an effective step toward enhancing the quality of multi-document summarization for scientific texts.