Temporal link prediction improves by tracking who is active
Your Temporal Link Predictor Is Blind to Who Is Active: A Missing Factor That Transfers Across Models
Machine LearningSocial and Information Networks
Summary
Temporal link prediction tries to guess who will interact with whom in the future based on past interactions. The authors found that current methods ignore an important factor: how active the person initiating the interaction is. They developed a simple model to track source activity, which alone sometimes outperforms more complex methods. Adding this factor to existing models improves their accuracy across many tests and speeds up training significantly.
What this means in practice
- •For social media platform engineers: Improve friend suggestion or content recommendation by integrating source activity modeling to better predict who will initiate future interactions.$Commercial implications: Enables more accurate and faster prediction models for social interactions, enhancing recommender systems as a marketable service.
- •For network security analysts: Enhance detection of unusual activity patterns by factoring in the activity level of suspicious sources within temporal interaction data.
Authors
Ji Zhang, Zixin Liu, Yiran Ding, Jiayi Wang, Yilu Du, Weijia Xuan
Abstract
An interaction has two parts: someone decides to act, and then chooses whom to act on. Temporal link prediction has concentrated on the second, and we show that it is blind to the first by construction: a standard negative keeps the real source and swaps the destination, and we prove that this cancels the source's activity exactly from the optimal score, so no model trained and evaluated this way is ever rewarded for learning it. Under the harder historical and inductive negatives, whose sources differ, the same factor becomes the dominant signal. We model it with Source Node Activity Modeling (SNAM), a self-exciting event intensity fitted by an exact point-process likelihood to decayed interaction counts the history states already contain; it has fewer than 20 parameters. On their own, never looking at the destination, these parameters beat DyGFormer and TPNet on four of five datasets under historical negatives. Added to the frozen scores of TPNet, TGN, DyGFormer and DSRD, four models of different design, without retraining anything, they raise AP on almost every backbone-dataset pair in both settings, by up to 25 points. Our full model ranks first overall against eleven baselines on 13 datasets and three protocols, and on million-event streams trains an epoch 9-100x faster than TPNet and DyGFormer. We conclude that source activity is a blind spot of temporal link prediction, and a cheap, transferable one to close. Code is available at https://github.com/Erutaner/Your-Temporal-Link-Predictor-Is-Blind-to-Who-Is-Active.