Ai-enhanced Measurement of Technological Convergence Uses Heterogeneous Graphs to Define a Two-Dimensional Index

Technological convergence, the increasingly blurred boundaries between disciplines, drives modern innovation, yet accurately measuring this complex phenomenon presents a significant challenge. Siming Deng, Runsong Jia, and Chunjuan Luan, from Dalian University of Technology, along with Mengjia Wu and Yi Zhang, address this problem by developing a new approach to quantify convergence through both its depth and breadth. Their research introduces a Technological Convergence Index (TCI) that utilises advanced artificial intelligence techniques, specifically heterogeneous graph transformers and semantic learning, to analyse patent data and reveal how knowledge integrates across different fields. This multidimensional method not only offers a more comprehensive understanding of convergence than existing measures, but also demonstrates strong reliability through rigorous testing against indicators of patent quality, providing valuable insights for innovation policy and strategic decision-making in cross-domain technologies.

Technology Convergence and Innovation Dynamics

Research consistently demonstrates that technology convergence is a dominant theme in modern innovation, driving both incremental improvements and radical breakthroughs. Studies focus on identifying, analyzing, and predicting how different technologies merge and interact, with increasing emphasis on digital transformation and the simultaneous shift towards sustainability, often termed the “twin transition,” in industries and organizations. Understanding how knowledge flows, is combined, and utilized is central to this research area. Studies explore how firms can effectively manage knowledge to drive innovation through convergence, particularly in the context of sustainability and green technologies, and examine how technology convergence affects industries, competitive landscapes, and firm performance.

Common methodologies include bibliometrics, scientometrics, and network analysis, used to identify trends and knowledge flows, while machine learning and natural language processing are increasingly employed to extract information from text data and understand semantic relationships. This research spans a wide range of applications, from broad cross-industry trends to specific sectors like healthcare, manufacturing, and the automotive industry. Researchers engineered a multidimensional approach that assesses both the depth and breadth of knowledge integration within patent data spanning 2003 to 2024. To calculate depth, the team utilized textual descriptions from the International Patent Classification (IPC) system, constructing a complex network modeled using advanced artificial intelligence techniques, specifically Heterogeneous Graph Transformers and Sentence-BERT, enabling a precise representation of semantic knowledge integration. Complementing this depth analysis, the breadth dimension quantifies technological diversity using the Shannon Diversity Index, measuring the variety of technological combinations present within patents.

The team then integrated these depth and breadth dimensions using the Entropy Weight Method, objectively assigning weights based on their information entropy, ensuring a balanced and representative overall convergence score. To validate the TCI, scientists compared its performance against established convergence measures and conducted a novel robustness test, regressing the TCI against indicators of patent quality, confirming that higher levels of technological convergence are associated with higher-quality innovations. The TCI uniquely assesses convergence along two fundamental dimensions, depth and breadth, providing a more nuanced understanding than previous approaches. Depth calculations leverage textual descriptions from the International Patent Classification (IPC) system, enhanced by incorporating patent metadata into a complex network structure modeled using advanced artificial intelligence techniques, specifically Heterogeneous Graph Transformers and Sentence-BERT. The team developed a method to quantify the semantic strength of connections spanning different fields, revealing how deeply an invention extends beyond its core domain.

Breadth is measured using the Shannon Diversity Index, which captures the variety of IPC-based knowledge combinations within patents. The final TCI is constructed using the Entropy Weight Method, objectively assigning weights to both depth and breadth based on their information entropy. To validate the TCI, researchers compared its performance against established convergence measures, demonstrating its advantages in capturing nuanced patterns of knowledge integration. The TCI uniquely assesses convergence through two key dimensions, depth and breadth, providing a more nuanced understanding than previous approaches. Depth is calculated by analyzing the semantic connections within patent data using advanced artificial intelligence techniques, specifically Heterogeneous Graph Transformers and Sentence-BERT, while breadth quantifies the diversity of technological fields involved in an invention using the Shannon Diversity Index. By integrating these dimensions with an Entropy Weight Method, the TCI objectively evaluates the extent of cross-domain knowledge integration.

The researchers demonstrate the practical relevance of the TCI through rigorous validation, including comparisons with existing convergence measures and regression analysis against indicators of patent quality, strengthening its value for both academic research and practical application. Existing convergence measures often treat depth and breadth as separate concepts or lack robust testing against real-world metrics, limiting their usefulness for guiding innovation policy and strategy. This work overcomes these limitations by offering a unified framework and demonstrating a clear link between the TCI and indicators of successful innovation.