TREND: TempoRal Event and Node Dynamics for Graph Representation Learning

Wen, Zhihao

doi:10.1145/3485447.3512164

Cited by 41 publications

(21 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…(2) TREND. In [88], the formation of each edge ((𝑣 𝑖 , 𝑣 𝑗 ), 𝑡) is defined as an event with unique properties specified by the states of end nodes (𝑣 𝑖 , 𝑣 𝑗 ) at time step 𝑡. Namely, events may occur due to different reasons at different time steps.…”

Section: Discussionmentioning

confidence: 99%

“…On the one hand, the encoders and loss functions of some methods cannot capture the variation of weighted topology. For instance, most approaches with the unevenly-spaced edge sequence description rely on some stochastic processes on unweighted graph topology (e.g., temporal random walk [61,87] and temporal point process [52,88,89]), which do not have the hypothesis regarding dynamic edge weights. On the other hand, the decoders of some approaches [48,49,74,81] are only designed for unweighted graphs, treating TLP as a binary edge classification task.…”

Section: Advanced Topics and Future Directionsmentioning

confidence: 99%

See 1 more Smart Citation

Temporal Link Prediction: A Unified Framework, Taxonomy, and Review

Meng¹,

Yeung²

2022

Preprint

View full text Add to dashboard Cite

Dynamic graphs serve as a generic abstraction and description of the evolutionary behaviors of various complex systems (e.g., social networks and communication networks). Temporal link prediction (TLP) is a classic inference task on dynamic graphs, which aims to predict possible future linkage using historical dynamic topology. The predicted future topology can be used to support some advanced applications on real-world systems (e.g., resource pre-allocation) for better system performance. This survey provides a comprehensive review of existing representative TLP methods. Concretely, we first give the formal statements regarding data models, task settings, and learning paradigms that are commonly used in related research. A hierarchical fine-grained taxonomy is further introduced to categorize existing methods in terms of their data models, learning paradigms, and techniques. From a generic perspective, we propose a unified encoder-decoder framework to formulate all the methods reviewed, where different approaches only differ in terms of some components of the unified framework. Moreover, we envision serving the community with an open-source project OpenTLP 1 that refactors or implements some representative TLP methods using the proposed unified framework and summarizes other public sources. To conclude this survey, we also discuss some advanced topics in recent research and highlight possible future directions. CCS Concepts: • General and reference → Surveys and overviews; • Networks → Network dynamics; • Information systems → Temporal data.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Advanced Topics and Future Directionsmentioning

confidence: 99%

Temporal Link Prediction: A Unified Framework, Taxonomy, and Review

Meng¹,

Yeung²

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…To prevent the deficiency, another line of work aims to learn the temporal processing of continuously evolving events. DyRep [25] and its variants [26,27] employ temporal Hawkes process to model the temporal properties in networks. CTDNE [28] extends the standard random walk to time-respect random walk, FiGTNE [29] proposes timereinforced random walk to embed fine-grained networks, and CAW-N [9] leverages causal anonymous walk to capture motifs in temporal graphs.…”

Section: Related Workmentioning

confidence: 99%

TPGNN: Learning High-order Information in Dynamic Graphs via Temporal Propagation

Wang¹,

Li²,

Yu³

2022

Preprint

View full text Add to dashboard Cite

Temporal graph is an abstraction for modeling dynamic systems that consist of evolving interaction elements. In this paper, we aim to solve an important yet neglected problemhow to learn information from high-order neighbors in temporal graphs?-to enhance the informativeness and discriminativeness for the learned node representations. We argue that when learning high-order information from temporal graphs, we encounter two challenges, i.e., computational inefficiency and over-smoothing, that cannot be solved by conventional techniques applied on static graphs. To remedy these deficiencies, we propose a temporal propagationbased graph neural network, namely TPGNN. To be specific, the model consists of two distinct components, i.e., propagator and node-wise encoder. The propagator is leveraged to propagate messages from the anchor node to its temporal neighbors within k-hop, and then simultaneously update the state of neighborhoods, which enables efficient computation, especially for a deep model. In addition, to prevent over-smoothing, the model compels the messages from nhop neighbors to update the n-hop memory vector preserved on the anchor. The node-wise encoder adopts transformer architecture to learn node representations by explicitly learning the importance of memory vectors preserved on the node itself, that is, implicitly modeling the importance of messages from neighbors at different layers, thus mitigating the oversmoothing. Since the encoding process will not query temporal neighbors, we can dramatically save time consumption in inference. Extensive experiments on temporal link prediction and node classification demonstrate the superiority of TPGNN over state-of-the-art baselines in efficiency and robustness.

show abstract

“…In vanilla NP, it infers the distribution of link prediction function 𝑃 (𝑧 𝑇 |G ≤𝑇 ) from historical context data, and fixes it for future link prediction. However, in dynamic graphs, the links arrive in different frequencies [5,13,36]. Therefore, it is inappropriate to use a static distribution over time, although we can incorporate the new-coming links into the context data and update the distribution by using Eq.…”

Section: Sequential Ode Aggregatormentioning

confidence: 99%

“…SNP [32] adopts a RNN aggregator to consider the sequential information, but it still fails to consider the derivative of the underlying distribution and uses a static distribution through time. For instance, as shown in Figure 2b, the links in dynamic graphs arrive irregularly [5,13,36]. Additionally, an important event could lead to a large number of links occurring in a short time, which is shown by the spikes in Figure 2b.…”

Section: Introductionmentioning

confidence: 99%

Graph Sequential Neural ODE Process for Link Prediction on Dynamic and Sparse Graphs

Luo,

Haffari,

Pan

2022

Preprint

View full text Add to dashboard Cite

Link prediction on dynamic graphs is an important task in graph mining. Existing approaches based on dynamic graph neural networks (DGNNs) typically require a significant amount of historical data (interactions over time), which is not always available in practice. The missing links over time, which is a common phenomenon in graph data, further aggravates the issue and thus creates extremely sparse and dynamic graphs. To address this problem, we propose a novel method based on the neural process, called Graph Sequential Neural ODE Process (GSNOP). Specifically, GSNOP combines the advantage of the neural process and neural ordinary differential equation (ODE) that models the link prediction on dynamic graphs as a dynamic-changing stochastic process. By defining a distribution over functions, GSNOP introduces the uncertainty into the predictions, making it generalize to more situations instead of overfitting to the sparse data. GSNOP is also agnostic to model structures that can be integrated with any DGNN to consider the chronological and geometrical information for link prediction. Extensive experiments on three dynamic graph datasets show that GSNOP can significantly improve the performance of existing DGNNs and outperform other neural process variants. CCS CONCEPTS• Networks → Network algorithms; • Computing methodologies → Machine learning.

show abstract

TREND: TempoRal Event and Node Dynamics for Graph Representation Learning

Cited by 41 publications

References 25 publications

Temporal Link Prediction: A Unified Framework, Taxonomy, and Review

Temporal Link Prediction: A Unified Framework, Taxonomy, and Review

TPGNN: Learning High-order Information in Dynamic Graphs via Temporal Propagation

Graph Sequential Neural ODE Process for Link Prediction on Dynamic and Sparse Graphs

Contact Info

Product

Resources

About