cs.LGNov 18, 2026

Interpreting Timestamp Permutations in Dynamic Link Prediction

Authors: Ilia Karpov, Kirill Koloskov, Valeriy Kuleshov, Dmitry Ilvovsky

Organizations: International Laboratory for Applied Network Research, HSE University, Moscow, Russia · Independent Researcher, Russia · HSE University, Moscow, Russia

Abstract

Timestamp permutations are used to test whether dynamic link predictors exploit temporal information, but a performance gap can reflect changes in training, evaluation history, or both. We separate these effects with a crossed diagnostic and measure input exposure: whether altered histories actually change the predictor's inputs. On MOOC, TCL trained with binary cross-entropy gains approximately eight pairwise-accuracy points from original-chronology training under either evaluation history, while changing history at fixed weights gives only 0.03 points. The small history contrast reflects unchanged inputs for 68.37% of queries, stable rankings for most others, and cancellation of gains and losses. Under original evaluation history, suppressing explicit time encoding reduces the training contrast from 8.03 to 3.64 points, a paired reduction of 4.39 points (95% interval [3.68, 5.09]) across six seeds. DyGFormer shows corresponding training and history contrasts of 22.34 and 0.38 points across two seeds, despite greater exposure. Partial history permutations further separate nominal intervention strength, input changes, and aggregate accuracy. These comparisons turn a single permutation gap into interpretable statements about training conditions and model responses, while leaving the internal mechanisms of temporal learning open.

Figures & tables

Explore similar work

CardsList
  1. Revisiting Temporal Link Prediction through Structural Proximity

    Oct 28, 2026Yuzhang Fei, Yanping Zheng, Lu Yi +1Temporal Link Prediction

  2. HAWKEYE: Seeing One Layer Deeper -- A Cohesion-Aware Structural Channel for Temporal Link Prediction

    Aug 24, 2026Jiacheng Ding, Xiaofei ZhangTemporal Link PredictionTemporal GNNs

  3. StreamTab: Training-Free Temporal Link Prediction with Tabular Foundation Models

    Nov 6, 2026Mingqi Yang, Renqi WuTemporal Link PredictionDynamic Graph Learning