Abstract: Sci-tech novelty retrieval is an important approach to evaluate scientific and technological innovation activities, but the quality of novelty retrieval report is greatly affected by the subjective factors of novelty searchers, and the sci-tech novelty retrieval work also needs to be transformed. This paper proposed that the LDA topic model based on big data can effectively improve the authority of sci-tech novelty searching, and provide strategic reference for the development of sci-tech novelty retrieval. This paper introduced the theoretical basis and application of LDA topic model, formed a sci-tech novelty search method based on LDA topic model, discussed the specific application of this method in different aspects of novelty search, and analyzed the effect of this method through practical cases. The introduction of LDA subject model method in the sci-tech novelty retrieval can effectively help the novelty searchers to understand the novelty retrieval points, assist the auditor to rapidly review reports and inspire the clients to evaluate scientific research from multiple perspectives. It not only improves the readability and objectivity of the novelty search report, but also explores the in-depth intelligence service capability of the novelty search station, laying a solid foundation for the transformation of the sci-tech novelty retrieval. 

Key words: Sci-tech novelty retrieval, LDA, Topic model, Topic evolution