Abstract: A good interaction pattern between science and technology is the key to generating major innovations, and exploring the identification method of science and technology interaction community fusing text and citation characteristics for scientific and technological innovations represented by papers and patents will help researchers and innovation managers to understand the interaction pattern of science and technology, optimize the transformation of scientific and technological innovations, and discover the path of scientific and technological cross-innovation. Based on the algorithms of text representation learning, graph autoencoder(GAE) and similarity network fusion(SNF), this study proposes a method to identify science and technology interaction communities by fusing textual and citation characteristics of papers and patents, and comprehensively analyzes the science and technology interactions in a specific field from the dimensions of content and intensity of interaction communities. In this study, the field of genetically engineered vaccines is selected for empirical analysis, and the effectiveness of the method is verified through comparative experiments. The results show that the science and technology interaction communities identified in this study can effectively describe the science and technology interaction situation in the field, demonstrate the hotspots of scientific and technological cross-innovation in the field as well as the evolution of interaction, restore the development of the science and technology interaction communities, and provide brand new knowledge units and application scenarios for the study of science and technology interaction.

Key words: Science and technology interaction, Community identification, Graph auto-encoder, Text representation learning, Network fusion