This website requires JavaScript.

Parameter-free Dynamic Graph Embedding for Link Prediction

Jiahao LiuDongsheng LiHansu GuTun LuPeng ZhangNing Gu
Oct 2022
Dynamic interaction graphs have been widely adopted to model the evolution ofuser-item interactions over time. There are two crucial factors when modellinguser preferences for link prediction in dynamic interaction graphs: 1)collaborative relationship among users and 2) user personalized interactionpatterns. Existing methods often implicitly consider these two factorstogether, which may lead to noisy user modelling when the two factors diverge.In addition, they usually require time-consuming parameter learning withback-propagation, which is prohibitive for real-time user preference modelling.To this end, this paper proposes FreeGEM, a parameter-free dynamic graphembedding method for link prediction. Firstly, to take advantage of thecollaborative relationships, we propose an incremental graph embedding engineto obtain user/item embeddings, which is an Online-Monitor-Offline architectureconsisting of an Online module to approximately embed users/items over time, aMonitor module to estimate the approximation error in real time and an Offlinemodule to calibrate the user/item embeddings when the online approximationerrors exceed a threshold. Meanwhile, we integrate attribute information intothe model, which enables FreeGEM to better model users belonging to some underrepresented groups. Secondly, we design a personalized dynamic interactionpattern modeller, which combines dynamic time decay with attention mechanism tomodel user short-term interests. Experimental results on two link predictiontasks show that FreeGEM can outperform the state-of-the-art methods in accuracywhile achieving over 36X improvement in efficiency. All code and datasets canbe found in