INK-USC/RE-Net

RE-Net is a classic research implementation from EMNLP 2020. With 461 stars and nearly 100 forks, it has served as a significant benchmark in the Temporal Knowledge Graph (TKG) research community. However, its defensibility is low (3) because it is a static reference implementation for a specific paper with zero current velocity. In the fast-moving AI landscape, 2020-era RNN-based graph models have largely been superseded by Temporal Graph Neural Networks (T-GNNs) and, more recently, by LLM-based reasoning over graph structures. The 'moat' here is purely academic—being a standard baseline that new papers must cite and compare against. There is no commercial moat or active ecosystem development. Frontier risk is low because this specific niche of structured temporal reasoning is too narrow for generalized labs, but the technology is effectively at the end of its lifecycle in terms of state-of-the-art performance.