AI-enhanced measurements of technical convergence define two-dimensional indexes using heterogeneous graphs

The convergence of technology, an increasingly blurred boundary between disciplines, encourages modern innovation, but accurate measurement of this complex phenomenon poses an important challenge. Dang, Runsong Jia, Chunjuan Luan of Dalian Institute of Technology, and Shimming of Mengjia Wu and Yi Zhang address this issue by developing a new approach to quantifying convergence through both its depth and width. Their research uses advanced artificial intelligence technologies, particularly heterogeneous graph transformers and semantic learning, to analyze patent data and introduce technical convergence index (TCI), which illustrates how knowledge is integrated in different fields. This multidimensional method not only provides a more comprehensive understanding of convergence than existing measurements, but also demonstrates strong reliability through rigorous testing of patent quality indicators, providing valuable insights into innovation policies and strategic decision-making in cross-domain technologies.

Technology convergence and innovation dynamics

Research consistently demonstrates that technology convergence is the dominant theme in modern innovation, facilitating both progressive improvement and fundamental breakthroughs. Research focuses on identifying, analyzing and predicting how different technologies blend and interact, focusing on simultaneous shifts towards digital transformation and sustainability, often referred to as industry or organizational “twin transitions.” Understanding how knowledge flows, combines and utilizes is central to this field of research. The research explores how companies can effectively manage their knowledge to promote innovation through convergence, particularly in the context of sustainability and green technology, and examines how technology convergence affects industry, competitive landscapes, and corporate performance.

While general methodologies include bibliographic, scientific and network analysis, they are used to identify trends and knowledge flows, machine learning and natural language processing are increasingly being adopted to extract information from textual data and understand semantic relationships. This study spans a wide range of applications ranging from broad industrial trends to specific sectors such as healthcare, manufacturing and the automotive industry. The researchers designed a multidimensional approach that assesses both the depth and width of knowledge integration within patent data from 2003 to 2024. To calculate depth, the team utilized textual descriptions from the International Patent Classification (IPC) system to build complex networks modeled using advanced artificial intelligence technologies. Complementing this depth analysis, the width dimension quantifies the diversity of technology using the Shannon diversity index and measures the diversity of technology combinations present within the patent.

The team then integrated these depth and width dimensions using the entropy gravity method, and objectively allocated weights based on information entropy to ensure a balanced, representative overall convergence score. To verify the TCI, scientists compared its performance with established convergence measurements, conducted new robustness tests, regressing the TCI from patent quality indicators, and verifying that higher levels of technical convergence are associated with high-quality innovations. TCI uniquely evaluates convergence along two fundamental dimensions, depth and width, providing a more nuanced understanding than previous approaches. Depth calculation utilizes textual explanations from the International Patent Classification (IPC) system. This was enhanced by incorporating patent metadata into complex network structures modeled using advanced artificial intelligence technologies, particularly heterogeneous graph trances and statements. The team has developed a method to quantify the semantic strength of connections across different fields, revealing that inventions are beyond the core domain.

Width is measured using the Shannon Diversity Index to capture various combinations of IPC-based knowledge within the patent. The final TCI is constructed using the entropy gravity method and objectively assigns weights to both depth and width based on information entropy. To validate TCI, researchers compared its performance with established convergence measures and demonstrated its advantages in capturing subtle patterns of knowledge integration. TCI uniquely evaluates convergence through two important dimensions, depth and width, providing a more nuanced understanding than previous approaches. Depth is calculated by analyzing semantic connections within patent data using advanced artificial intelligence technologies, particularly non-uniform graph transformers and statements, while width quantifies the diversity of the technical fields involved in inventions using Shannon Diversity Index. By integrating these dimensions with the entropy gravimetry, TCI objectively assesses the extent of cross-domain knowledge integration.

Researchers demonstrate the practical relevance of TCI through rigorous verification, including comparing with existing convergence measures, regression analysis of patent quality indicators, and strengthening the value of both academic research and practical applications. Existing convergence measures either treat depth and width as separate concepts or lack robust testing of actual metrics, limiting their usefulness in guiding innovation policies and strategies. This work overcomes these limitations by providing a unified framework and demonstrating a clear link between TCI and indicators of successful innovation.