Search papers, labs, and topics across Lattice.
This paper introduces HAWKEYE, a cohesion-aware structural channel designed to enhance temporal link prediction by focusing on 2-hop cohesive bridges rather than the traditional 1-hop common-neighbour signal, which is shown to be ineffective in sparse temporal graphs. By incrementally maintaining a family of cohesiveness indicators, HAWKEYE significantly improves the performance of existing models like DyGFormer, achieving test average precision and mean reciprocal rank increases of up to 10.8 points across multiple datasets. The method demonstrates robust scalability and effectiveness, particularly in scenarios where the graph's 2-hop discriminative AUC is high, while also providing insights into its limitations in degenerate or saturated graphs.
HAWKEYE nearly doubles the baseline performance in temporal link prediction by leveraging deeper structural insights that traditional methods overlook.
State-of-the-art temporal-link-prediction (TLP) models are, in essence, multi-channel information aggregators: they combine an interaction-history channel, a time-encoding channel, and a structure channel. The first two have been refined relentlessly; the structure channel remains a crude afterthought -- DyGFormer encodes it as a 1--2-bit neighbour-cooccurrence count. We begin with a measurement: on sparse temporal graphs the classical 1-hop common-neighbour signal is near-random (discriminative AUC $\approx 0.50$), because two nodes almost never share a direct neighbour; the genuinely discriminative signal lies one hop deeper -- the 2-hop cohesive bridge, whose discAUC reaches 0.73--0.98, on both bipartite and non-bipartite graphs. Motivated by this, we propose HAWKEYE, a cohesion-aware structural channel that incrementally maintains the classical k-family of cohesiveness indicators (degree $\to$ k-core $\to$ k-truss) and forms 2-hop cohesive-bridge features. HAWKEYE is a drop-in replacement for a temporal-graph model's native structure channel, with no change to the backbone. Swapping HAWKEYE into DyGFormer improves test AP/MRR over the cooccurrence channel by +0.6 to +10.8 points across six multi-seed-validated datasets (uci, enron, USLegis, CanParl, reddit, mooc). On the bipartite recommendation benchmark tgbl-subreddit, a 3-seed single-pass struct-only ablation shows HAWKEYE nearly doubling the baseline test MRR (0.103$\pm$0.003 $\to$ 0.204$\pm$0.005, +10.1 points across all three seeds); the streaming pipeline scales to the 67M-edge tgbl-flight in five minutes per pass. We further characterise when it helps: the gain tracks a graph's training-free 2-hop discAUC and vanishes on degenerate or saturated graphs -- a predictable boundary. All code, data, and figure-generation scripts are released.
