Semantic-Based Link Prediction in Dynamic Technology Innovation Networks

Abstract Technological innovation networks are essential for advancing technologies and fostering scientific breakthroughs. However, a significant challenge remains in their limited capacity to accurately capture semantic relationships and predict future technological developments. To address this, a dynamic innovation network for the technologies is constructed by integrating the LDA topic model and the word2vec algorithm, allowing for the exploration of its evolving patterns. Then, by optimizing the link prediction algorithm, potential relationships within the network are identified, revealing under-explored technological development directions. The results, focusing on hydrogen energy technology, indicate that technologies focused on enhancing the efficiency and performance of power systems, improving the reliability of hydrogen energy systems, and advancing high-efficiency electrolysis-based hydrogen production and supply systems exhibit sustained research momentum. Notably, high-efficiency electrolysis-based hydrogen production systems have experienced significant growth in research attention in recent years. It is anticipated that future advancements will involve greater integration and relevance of technologies aimed at improving power system efficiency, potentially catalyzing a leap development of hydrogen energy technologies.