Abstract: 【Purpose/significance】In the Big data environment, scientific literature covers a large number of information in different fields, themes and modes, such as text, images, audio, etc.【Method/process】Therefore, how to effectively filter multimodal redundant data is one of the difficulties at this stage. Therefore, this study proposes a multi-dimensional semantic cross modal retrieval algorithm for scientific literature in the Big data environment. The biased Kalman filter is used to filter the redundant data of scientific literature database. Based on this, the TF-IDF algorithm is used to extract feature words that represent the text. The semantic similarity between feature words is calculated by the value of Document retrieval elements, and the retrieval matrix is generated by combining the rel⁃ evance of the retrieval elements to complete the multi-dimensional semantic cross modal retrieval of scientific documents in the Big data environment.【Result/conclusion】The experimental results show that the proposed algorithm has high retrieval accuracy, high NDCG values, and shorter retrieval time.【Innovation/limitation】The research on this algorithm aims to address the limitations of tradi⁃ tional keyword retrieval methods by integrating multimodal data, utilizing rich semantic information, and addressing semantic gaps. It improves the effectiveness and accuracy of scientific literature retrieval, providing researchers and scholars with more convenient and comprehensive information retrieval services.