How Does Research Evolve? Tracing Cross-Domain Trajectories in NLP, ML, and CV with Claim-Grounded Typed Citations

The article introduces the SciTraj corpus, a novel claim-grounded typed citation graph comprising 32,559 papers from NLP, ML, and Vision, connected by 573,126 directed edges across six relation types. This corpus enables more nuanced analysis of research evolution by linking citations to specific claims and includes 287 million typed trajectories for benchmarking link prediction. The findings reveal disciplinary siloing and emergent topics, particularly in Vision and LLM-related research, providing practitioners with insights into citation dynamics and potential areas for innovation.