Modeling Technology Emergence with Language Models and Graph Analytics. In Data Science & Engineering Master of Advanced Study (DSE MAS) Capstone Projects

This capstone project explores the use of Large Language Models and graph-based learning to model emerging technologies from large-scale innovation data. Patent awards, published papers, and government grants corpora were integrated into a heterogeneous knowledge graph connecting extracted technical terms, institutions, and research outputs over time. Natural Language Processing techniques extracted and embedded technical terminology, which was fused with metadata and graph structure to produce feature-rich node representations. A Graph Neural Network (GNN) was trained for link prediction to expose potential emerging connections between concepts, and these candidate edges were subsequently labeled via prompting and classification models. The resulting system was made available through APIs for novelty detection, citation analysis, and research-velocity tracking, and an interactive UI interface was developed to explore trends and inspect relationships within the evolving innovation landscape. Special derived data structures were created to enable rapid retrieval of oft-accessed information. The overall pipeline—spanning data ingestion, modeling experiments, and visual exploration tools—supports interpretable technology forecasting. The data sources used for this project are collected over the period 2014-2024 and include publications from OpenAlex, accepted patents from the USPTO, and awarded grants from NIH/NSF/SBIR government agencies.

The authors are members of Cohort 10 (2023-2025) of the MAS DSE program at University of California, San Diego.