What is your next invention? — A framework of mining technological development rules and assisting in designing new technologies based on BERT as well as patent citations

In the past 20 years, with the rapid development of technology, the number of granted patents has also been increasing over years. How to utilize these data effectively is very important for making R&D policies and assisting in designing new technologies. In this paper, we map patents into a low-dimensional vector space, which is constructed by International Patent Classification (IPC) codes, through a deep learning model, i.e., Bidirectional Encoder Representations from Transformers (BERT). Then, this research makes the following contributions: first, we find that the generated vectors can describe the new technologies’ invention perspectives of patents accurately according to their texts; second, these vectors are combined with the physical meaning of patent citations (technological application) for the first time to solve some issues in designing new technologies from a different view; third, the citation relations and vectors of patents are adopted to explore the development rules of technology in terms of new technology’s invention perspective; fourth, an approach is raised to assist inventors in designing new technologies through the forward citations of patents, whose vectors have great similarities to the initial ones’; finally, the patents granted by USPTO in the past 20 years are used to verify the effectiveness of our framework.