claws-lab/jodie

Implements dynamic embedding trajectories using RNN-based architecture to capture evolving node behavior in temporal interaction networks, enabling both temporal link prediction and node state change detection. The framework introduces t-Batch, a specialized batching algorithm that parallelizes independent edges for scalability on large graphs. Targets downstream tasks including recommendation systems, anomaly detection, and fraud detection across social networks, e-commerce, and financial domains.

414 stars. No commits in the last 6 months.