Highly Liquid Temporal Interaction Graph Embeddings

Chen, Huidi; Xiong, Yun; Zhu, Yangyong; Yu, Philip S.

doi:10.1145/3442381.3449921

Cited by 3 publications

(4 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recurrent models such as Time-LSTM [59], Time-Aware LSTM [3] and RRN [53] capture dynamics of users and items by endowing them with a long short-term memory. HILI [10] is the successor of Jodie [28] and both of them are promising methods for real-world temporal interaction graphs. CTDNE [33] adopts random walk on temporal networks and CAW [51] makes it causal and anonymous to inductively represent sequential networks.…”

Section: Datasetmentioning

confidence: 99%

“…The challenge here is how to estimate the global curvatures accurately for the next period from local structure information. Besides, most of the existing methods [3,10,28,59] also assume that an interaction would only influence corresponding nodes while ignoring the information propagation among the same type of nodes implicitly. New methods are also needed to incorporate the information flow between the same type of nodes.…”

Section: Datasetmentioning

confidence: 99%

“…• Random walk based models: CAW [51], CTDNE [33] are two temporal network models adopting causal and anonymous random walks. • Sequence network embedding models: in this category, JODIE [28], HILI [10] and DeePRed [27] are three state-ofthe-art methods in generating embeddings from sequential networks employing recursive networks. • Hyperbolic models: we compare SINCERE with HGCF [42] which is a hyperbolic method on collaborative filtering.…”

Section: Experiments 41 Datasets and Compared Algorithmsmentioning

confidence: 99%

“…CTDNE [33] and CAW [51] adopt random walk to temporal graph modeling. Co-evolution models consider the influence between users and items and achieve great success [10,12,27,28]. For systematic and comprehensive reviews, readers may refer to [1,26].…”

Section: Related Workmentioning

confidence: 99%

See 3 more Smart Citations

SINCERE: Sequential Interaction Networks representation learning on Co-Evolving RiEmannian manifolds

Zhang

et al. 2023

Proceedings of the ACM Web Conference 2023

View full text Add to dashboard Cite

Sequential interaction networks (SIN) have been commonly adopted in many applications such as recommendation systems, search engines and social networks to describe the mutual influence between users and items/products. Efforts on representing SIN are mainly focused on capturing the dynamics of networks in Euclidean space, and recently plenty of work has extended to hyperbolic geometry for implicit hierarchical learning. Previous approaches which learn the embedding trajectories of users and items achieve promising results. However, there are still a range of fundamental issues remaining open. For example, is it appropriate to place user and item nodes in one identical space regardless of their inherent discrepancy? Instead of residing in a single fixed curvature space, how will the representation spaces evolve when new interaction occurs?To explore these implication for sequential interaction networks, we propose SINCERE, a novel method representing Sequential Interaction Networks on Co-Evolving RiEmannian manifolds. SIN-CERE not only takes the user and item embedding trajectories in respective spaces into account, but also emphasizes on the space evolvement that how curvature changes over time. Specifically, we introduce a fresh cross-geometry aggregation which allows us to propagate information across different Riemannian manifolds without breaking conformal invariance, and a curvature estimator which is delicately designed to predict global curvatures effectively according to current local Ricci curvatures. Extensive experiments on several real-world datasets demonstrate the promising performance of SINCERE over the state-of-the-art sequential interaction prediction methods.

show abstract