Currently, artificial intelligence-driven scientific research (AI for Science, AI4S) has become the high ground of international scientific and technological (Sci&Tech) competition. The essence of artificial intelligence lies in knowledge acquisition and application. In the process of rapid development of Sci&Tech large models and the advancement of AI4S research paradigms, a critical issue faced by many scientific research teams is the lack of reliable high-quality data and specialized domain knowledge. As the main carrier of human knowledge, Sci&Tech literature is rich in rigorous, credible, systematic, and cutting-edge knowledge, which has important value for supporting the training of Sci&Tech large models and driving the AI4S scientific research paradigm transformation in China. In this paper, the concept of Sci&Tech literature knowledge base to support AI4S is proposed. This paper reviews the requirements for constructing knowledge base of Sci&Tech literature supporting AI4S. The overall idea of constructing AI4S Sci&Tech literature knowledge base through Sci&Tech literature data mining to support Sci&Tech large models training and scientific research paradigm transformation is elaborated. The National Science Library, Chinese Academy of Sciences has been actively exploring the scientific knowledge and high-quality data embedded in the domain Sci&Tech literature and making efforts to construct the domain knowledge base and Sci&Tech literature large model to support AI4S. This paper concludes by proposing the future development direction and strategy.