Abstract: The timely and accurate identification of breakthrough scientific literature is of crucial strategic significance for the efficient allocation of scientific research resources, seizing the initiative in technological development, and enhancing national core competitiveness. However, existing identification methods have limitations such as single-dimensional indicators and insufficient recognition efficiency. From the contextual perspective of knowledge innovation, this study constructs a characteristic system that distinguishes breakthrough papers from ordinary papers from three aspects: knowledge foundation, research team, and academic attention, and proposes a machine learning-based early identification method for breakthrough papers. Experiments in the field of biomedicine show that the F1-score of the model in this study reaches 0.838, which verifies the effectiveness of the method. Among the factors, early-stage impact, number of references, and Price Index have the most significant impact on the model's recognition results. This study enriches and expands the theoretical framework and methodological system for breakthrough paper identification from the novel perspective of innovation context.

Key words: Breakthrough papers；Machine learning models；Early recognition；Biomedical field, Knowledge innovation