Towards Improving Long-Tail Entity Predictions in Temporal Knowledge Graphs through Global Similarity and Weighted Sampling

Kanak Mahadik; Mehrnoosh Mirtaheri; Mohammad Rostami; Ryan A. Rossi; Sungchul Kim; Tong Yu; Xiang Chen

REVIEW 4 major objections 5 minor 41 references

Towards Improving Long-Tail Entity Predictions in Temporal Knowledge Graphs through Global Similarity and Weighted Sampling

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Incremental TKG completion gains 10–15% MRR from global similarity and tail sampling.

desk verdict The incremental TKG setup is worth discussing, but the reported 10%/15% MRR gains are uninterpretable because the Section 5.2 protocol trains on test labels before evaluating on them. read the letter →

arxiv 2507.18977 v1 pith:HBZPFYFJ submitted 2025-07-25 cs.AI cs.LG

Mehrnoosh Mirtaheri , Ryan A. Rossi , Sungchul Kim , Kanak Mahadik , Tong Yu , Xiang Chen , Mohammad Rostami This is my paper · ORCID

classification cs.AIcs.LG

keywords temporalknowledgegraphsgraphcompletionincrementallearningcontinuallong-tailentitiesinductivelinkpredictionneuralnetworks

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that temporal knowledge graph completion models can be made to handle an evolving graph—where new facts arrive and new, sparsely connected entities appear—without retraining from scratch. It proposes a model-agnostic enhancement layer that enriches an entity's representation using other entities that have participated in the same relation at earlier times, plus a sampling scheme that up-weights quadruples involving rare entities. On two ICEWS benchmarks, the combined additions improve overall and inductive link prediction over fine-tuning, experience replay, and elastic weight consolidation, with 10% and 15% relative MRR gains on ICEWS14 and ICEWS18. The authors' central point is that long-tail and unseen entities are both a representation problem and a data-balancing problem, and both can be addressed on top of an existing GNN-based TKG model.

What carries the argument

The load-bearing pieces are (1) a relation-based global similarity set $S_t(r)=\{s_i : (s_i,r,o_i,t_i)\in G,\ t_i<t\}$, which defines similar entities by shared relation type rather than graph proximity, and (2) inverse-frequency sampling with a two-phase schedule of weighted and uniform draws. The enhancement function $g(s,r,t)$ is a recency-weighted mean of the embeddings of $S_t(r)$, blended into the base embedding via $\lambda$ and scaled by $\phi(d_s)$; this is the mechanism that lets an infrequent or new entity borrow representation mass from better-connected entities that play the same relational role. The sampling weight is a function $\psi(1/\mathrm{freq}(s), 1/\mathrm{freq}(o))$, instantiated as a minimum, maximum, or mean, and entity frequencies are recomputed incrementally from what has been seen up to the current training step.

What would settle it

Re-run the incremental protocol with each checkpoint trained only on the current snapshot's training split, with no post-training on that snapshot's validation or test sets, and recompute the Average MRR over all prior test sets; if the 10–15% advantage over FT, EWC, and ER shrinks to noise, the reported gains depend on the post-training step.

Watch

Extended reading notes

Core claim

The paper's central claim is that a global, relation-based similarity signal and a frequency-weighted sampling schedule jointly raise the ceiling of incremental TKG completion. For a query $(s,r,?,t)$, the enhancement layer computes $\tilde{e}_s = \lambda f(s) + \phi(d_s)(1-\lambda) g(s,r,t)$, where $g$ aggregates the embeddings of entities $s_i$ that have appeared as subjects of relation $r$ before time $t$, with recency weights $1/(1+\exp(\mu(t-t_i)))$; $\phi(d_s)$ decreases with degree, so sparse entities receive more enhancement. The sampling stage draws a fraction $\alpha$ of each batch with probability inversely related to the frequency of its subject and object, and the remaining $1-\alpha$ uniformly. Integrated into the TiTer model and evaluated on incremental snapshots of ICEWS14 and ICEWS18, this combination reports higher MRR and Hit@k than fine-tuning, EWC, and experience replay, both on the most recent task and averaged over all prior tasks; it also reports gains on queries whose entities were unseen and on frequency-binned long-tail groups.

Load-bearing premise

The performance claims rest on the assumption that an evaluation procedure which runs extra training on each snapshot's validation and test sets before advancing to the next snapshot does not leak the answers those checkpoints are later scored on; if that assumption fails, the reported gains could come from memorization rather than incremental generalization.

Editorial extensions

If this is right

Existing GNN-based TKG completion models can adopt the enhancement layer and weighted sampling without architectural retraining, so tail-focused prediction becomes an add-on rather than a redesign.
Because the final checkpoint is evaluated on every prior snapshot's test set, the reported Average numbers describe retention of old knowledge when new snapshots arrive.
Inductive queries, meaning links involving entities absent from all earlier snapshots, improve, indicating that role-based similarity can bootstrap representations for genuinely new entities.
On the two ICEWS benchmarks, the full method reports the best MRR and Hit@k in both current-task and averaged-over-time evaluations, making fine-tune-plus-enhancement a plausible default for streaming event graphs.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The relation-sharing similarity rule is defined without timestamps in the entity-selection step, so the same enhancement should transfer to static knowledge graphs; a direct test would separate the temporal weighting from the global-role signal.
Extending inverse-frequency weighting to relation types as well as entities is a natural follow-up, since sparse relations create the same long-tail failure mode the paper documents for entities.
Because the evaluation protocol post-trains on each snapshot's validation and test sets before the next task, a strict no-leak re-run would show how much of the 10–15% MRR gap comes from genuine continual-learning updates rather than warm starts on answer-bearing labels.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The thing you should know before reading: the headline numbers are not trustworthy. Section 5.2 describes a two-checkpoint protocol where, before moving to the next temporal snapshot, the model is post-trained on both the validation and test sets for a few epochs. That means every later model is initialized from weights that have been supervised on earlier test labels, and the reported "Average" metrics are computed over exactly those test sets. The claimed 10% and 15% MRR improvements over FT, EWC, and ER could easily be memorization rather than incremental generalization. The enhancement layer and weighted sampling may absorb leaked patterns more effectively than the baselines, so the comparison is stacked in the authors' favor.

What is actually new: the paper's own contribution is the incremental training regime with catastrophic-forgetting baselines. The enhancement layer and weighted sampling come from the authors' WWW 2024 companion paper. The problem setup—multiple temporal extrapolation tasks with new entities appearing over time—is useful and worth having as a benchmark. The ablation study is sensible, and the complexity analysis is fine. Credit where due: the authors are explicit about the protocol, so the leak is not hidden; that makes it easier to fix.

Soft spots, in proportion: the leak is load-bearing. Aside from that, only Titer is used as the base model, so "model-agnostic" is an overstatement. There are no error bars or significance tests, and with two datasets and modest gains that matters. The relation-based similarity is a simple grouping rule rather than a learned measure, which is fine but less novel than the presentation implies. A minor issue: hyperparameters are selected on validation performance, and no sensitivity analysis is reported.

Who this is for: researchers working on continual learning for temporal KGs will find the problem formulation and baseline comparisons a useful starting point, but they should not rely on the reported numbers. If the authors rerun without the leak, the core idea—global similarity plus weighted sampling for long-tail entities—might hold up, but that is currently an open question.

Recommendation: reject the current version. If I were handling it, I would send it back with a clear request to remove the post-training on test sets, report results from the clean checkpoints, and include error bars over multiple runs. That revision, if it comes back clean, would deserve another look. For peer review: this is not a desk reject—the flaw is identifiable and fixable, and the problem is relevant enough to merit serious referee time.

Referee Report

4 major / 5 minor

Summary. This paper proposes an incremental training framework for temporal knowledge graph completion, consisting of (i) a model-agnostic enhancement layer that augments local GNN-based entity representations with a global relation-based similarity aggregation, and (ii) a weighted frequency-based sampling strategy that up-weights quadruples involving rare entities. The framework is instantiated with TiTer as the base model and evaluated on ICEWS14 and ICEWS18 under fine-tuning, EWC, and experience-replay baselines. The authors claim improved total link prediction, inductive link prediction, and long-tail entity performance, including 10% and 15% MRR gains, and they provide ablations, a forgetting analysis, and a complexity/runtime analysis.

Significance. If the empirical results were valid, the paper would contribute a practical, modular recipe for continual and inductive TKG completion, and the global-similarity enhancement is an interesting departure from purely local GNN aggregation. The two released benchmark datasets and the complexity analysis are also useful. However, the main evaluation protocol is contaminated by supervised use of validation and test labels during incremental training, so the central empirical claims cannot be assessed from the reported experiments. The model-agnostic claim is also supported only by a single base model, and no uncertainty quantification is provided.

major comments (4)

[Section 5.2] The incremental training procedure is leaky: the text states that at each step the authors keep one checkpoint trained on D_train_t for evaluation and another obtained by 'post-training on both the validation and test sets for a few epochs before proceeding' to D_train_{t+1}. Consequently, every checkpoint after t=1 is initialized from weights that have been supervised on D_test_1 through D_test_{t-1} and on the corresponding validation sets, while the Average columns in Tables 2 and 3 are computed over exactly those earlier test sets. This is future-label leakage: the comparisons against FT, ER, and EWC in Tables 2-4 and Figure 4 measure partly memorization or leaked adaptation rather than incremental generalization. The claims in Section 6.1 of 10% (ICEWS14) and 15% (ICEWS18) MRR improvements, and the forgetting analysis in Figure 4a, are unsubstantiated without a rerun that uses only D_train_t for weight updates and initialization.
[Section 6.2 and Table 3] The inductive link prediction evaluation is invalidated by the same leak. Entities are classified as unseen if they do not appear in previous training quadruples, but because earlier D_test sets were post-trained on before advancing to the next task, many of those 'unseen' entities have in fact appeared in the model's training history. The Average results over the union of inductive test sets are therefore not a measure of inductive generalization. In addition, Table 3 appears malformed: the rows for +ER and +EWC contain fewer entries than the columns require, and the claimed First-snapshot equality of FT, ER, and EWC is not visible from the table. The inductive claim cannot be evaluated from the current tables.
[Abstract and Section 1] The model-agnostic claim is not supported by the experiments. The abstract and introduction state that the enhancement layer can be added to 'any existing TKG completion method' or 'most GNN-based TKG completion models,' but all experiments use a single base model, TiTer. Even setting aside the leak, a one-base-model study cannot establish model-agnosticism; at minimum a second base model with a different inductive mechanism is needed.
[Tables 2-4 and Section 5.2] The experiments report point estimates without seeds, error bars, or significance tests. Several headline differences are small (e.g., Table 2, ICEWS18 Average MRR: Ours 0.323 vs EWC 0.319; ICEWS14 Average MRR: Ours 0.421 vs FT 0.413), so it is not possible to tell whether the proposed components, rather than noise or the leaked data, drive the reported improvements. The hyperparameter search is performed on a validation set, which is itself later used for post-training, compounding the selection bias.

minor comments (5)

[Section 2] In the Related Work section, 'corrsponding' should be 'corresponding'.
[Equations (2)-(3)] The similarity set S_t(r) is defined without excluding the query entity s itself; if s has past events with relation r, the enhancement g(s,r,t) may include s's own embedding. The authors should clarify whether this is intentional and, if so, how self-reinforcement is handled.
[Section 5.1 and Table 1] The notation '28k/3.7k/4k' and the header 'G1 Avg|Q| Gi>1' are not defined; the reader should be told whether these are train/validation/test counts and what the average is taken over.
[Section 6.1] The text says '10% relative improvement' and '15% improvement' in MRR, but from Table 2 these percentages are relative to Titer trained only on the first snapshot, not relative to the best incremental baseline. The reference point should be stated explicitly in both the abstract and Section 6.1.
[Table 3] The table formatting appears corrupted: the +ER and +EWC rows have fewer entries than the columns require, and the mapping between values and the 'First'/'Average' column groups for ICEWS18 and ICEWS14 is hard to infer. The table should be reconstructed carefully.

Circularity Check

2 steps flagged · score 6.0 of 10

The evaluation protocol post-trains on validation and test labels before every later snapshot, so the reported 'Average' MRR gains over FT/ER/EWC partly measure test-set memorization; the enhancement layer itself is inherited from the authors' own prior paper.

fitted input called prediction [Section 5.2 (Evaluation Setup), Section 6.1 (Overall Performance / Table 2)]
"Thus, at training step t, we maintain two checkpoints: one post-training on D_train_t for evaluation and another post-training on both the validation and test sets for a few epochs before proceeding to D_train_{t+1}."

Every checkpoint that initializes the next snapshot has been supervised on validation and test labels from all earlier snapshots. The reported 'Average' metrics average performance on those same earlier test sets, so the final model's average scores are not predictions for unseen future facts but retrievals of labels already used as training signal. The central claim of a 10%/15% MRR improvement over FT/EWC/ER therefore reduces, at least in part, to a comparison of how well each method memorizes the post-training test data. A leak-free rerun is needed; the paper does not provide one.
self citation load bearing [Section 4.1, Eq. (3) and footnote 1; Section 1 contributions]
"The enhanced representation of a subject entity s is computed within the context of the query [18]. ... g(s,r,t) = (sum_{s_i in S_t(r)} w_i e_{s_i}) / (sum w_i), w_i = 1/(1+exp(mu(t-t_i))). ... Preliminary results of this work are presented in the 2024 ACM on Web Conference [18]."

The paper presents the enhancement layer as a novel contribution ('We propose... novel architectural components'), but the contextual similarity function is taken from the authors' own WWW'24 paper [18], which the footnote identifies as 'Preliminary results of this work.' The core mechanism is therefore justified by self-citation rather than derived or independently re-validated in this manuscript; combined with the test-leak evaluation, the claimed advantage of this layer is not independently established here.

full rationale

This is an empirical paper rather than a formal derivation, so most equations are self-contained and not circular in the usual sense. The serious circularity is in the evaluation protocol: Section 5.2 explicitly post-trains the model on both the validation and test sets before proceeding to the next snapshot, and Section 6.1 reports 'Average' performance over all prior test sets. As a result, the final model's average MRR/Hits on earlier snapshots reflects labels it has already been trained on, making the headline 10%/15% improvements over fine-tuning, EWC, and ER at least partially forced. The inductive-link results in Section 6.2 suffer from the same leak: entities 'unseen' at step t were seen in the post-training on earlier test sets. Additionally, the enhancement layer (Eq. 3) is attributed to the authors' own prior work [18], so the novelty claim is partly self-citational, though the equations are fully specified here. No evidence of equation-level tautology was found; the leak is the load-bearing issue. Because the central empirical claim reduces by construction to training-on-test, the score is 6 rather than 0-2.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central empirical claims rest on tuned hyperparameters and on unvalidated domain assumptions about relation-based similarity and inverse-frequency sampling. The evaluation also depends on a non-standard protocol that trains on validation and test sets between tasks; this is the largest burden on the reported numbers.

free parameters (6)

lambda = grid {0.3, 0.5, 0.7}, selected on validation
Balance between base embedding f(s) and enhanced embedding g(s) in Eq (1).
mu = grid {0.1, 0.3, 0.5}, selected on validation
Recency decay in Eq (3): w_i = 1 / (1 + exp(mu * (t - t_i))).
number of similar entities = grid {10, 15, 20, 25}, selected on validation
Subset size for the enhancement layer's similar-entity set S_t(r).
alpha = grid {0, 0.1, 0.2, 0.5, 0.8, 1}, selected on validation
Fraction of training data drawn by inverse-frequency weighted sampling in Section 4.2.
psi aggregation function = not reported
Section 4.2 says psi can be min, max, or mean of inverse frequencies, but no selected function or grid is reported; a hidden degree of freedom.
epoch count for validation/test post-training = not reported ('a few epochs')
Section 5.2 post-trains checkpoints on validation and test sets for an unspecified number of epochs, affecting all subsequent checkpoints.

assumptions (4)

domain assumption Entities that share a relation type as subjects are semantically similar enough that aggregating their temporal embeddings improves sparse-entity representations.
Eqs (2)-(3) define S_t(r) by shared relation and use it to build g(s,r,t); no validation of this similarity criterion is provided.
domain assumption Frequency-weighted sampling with inverse-frequency weights and a mixing fraction alpha improves long-tail generalization without overfitting.
Section 4.2 assumes rare-entity oversampling is beneficial; the method is an inductive bias, not derived.
ad hoc to paper Training on the validation and test sets before advancing to the next snapshot is a legitimate way to initialize later models.
Section 5.2 explicitly post-trains on validation and test sets; this is introduced by the paper and is the source of potential evaluation leakage.
domain assumption ICEWS14 and ICEWS18 interval splits with a 50% first snapshot and 7-day later snapshots are representative of incremental TKG scenarios.
Section 5.1 constructs the benchmarks by choosing a specific ICEWS interval and snapshot sizes; no justification is given that this captures real streaming dynamics.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Towards Improving Long-Tail Entity Predictions in Temporal Knowledge Graphs through Global Similarity and Weighted Sampling." pith.science (2026). https://pith.science/paper/HBZPFYFJ

@misc{pith2026250718977,
  author       = {Pith},
  title        = {Pith review of: Towards Improving Long-Tail Entity Predictions in Temporal Knowledge Graphs through Global Similarity and Weighted Sampling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HBZPFYFJ}},
  note         = {Machine review of arXiv:2507.18977}
}

read the original abstract

Temporal Knowledge Graph (TKG) completion models traditionally assume access to the entire graph during training. This overlooks challenges stemming from the evolving nature of TKGs, such as: (i) the model's requirement to generalize and assimilate new knowledge, and (ii) the task of managing new or unseen entities that often have sparse connections. In this paper, we present an incremental training framework specifically designed for TKGs, aiming to address entities that are either not observed during training or have sparse connections. Our approach combines a model-agnostic enhancement layer with a weighted sampling strategy, that can be augmented to and improve any existing TKG completion method. The enhancement layer leverages a broader, global definition of entity similarity, which moves beyond mere local neighborhood proximity of GNN-based methods. The weighted sampling strategy employed in training accentuates edges linked to infrequently occurring entities. We evaluate our method on two benchmark datasets, and demonstrate that our framework outperforms existing methods in total link prediction, inductive link prediction, and in addressing long-tail entities. Notably, our method achieves a 10\% improvement and a 15\% boost in MRR for these datasets. The results underscore the potential of our approach in mitigating catastrophic forgetting and enhancing the robustness of TKG completion methods, especially in an incremental training context

Figures

Figures reproduced from arXiv: 2507.18977 by the authors.

**Figure 2.** Continual Extrapolation Setup. The arrow depicts the time. For extrapolation, the edge timestamps in [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** Local vs. Relation-Based Entity Similarity Enhancements.(a) A traditional neighborhood-based ap [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Analysis of model components for catastrophic forgetting and long-tail entities. [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 26 canonical work pages

[1]

Antoine Bordes, Nicolas Usunier, Alberto Garcia-Duran, Jason Weston, and Oksana Yakhnenko. 2013. Translating embeddings for modeling multi-relational data.Advances in neural information processing systems26 (2013)

2013
[2]

E Boschee, J Lautenschlager, S O’Brien, S Shellman, J Starz, and M Ward. 2015. Integrated Crisis Early Warning System (ICEWS) Coded Event Data.URL: https://dataverse. harvard. edu/dataverse/icews(2015)

work page 2015
[3]

Yuanning Cui, Yuxin Wang, Zequn Sun, Wenqiang Liu, Yiqiao Jiang, Kexin Han, and Wei Hu. 2023. Lifelong embedding learning and transfer for growing knowledge graphs. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 37. 4217–4224

work page 2023
[4]

Angel Daruna, Mehul Gupta, Mohan Sridharan, and Sonia Chernova. 2021. Continual learning of knowledge graph embeddings.IEEE Robotics and Automation Letters6 (4 2021), 1128–1135. Issue 2. https://doi.org/10.1109/LRA.2021. 3056071

doi:10.1109/lra.2021 2021
[5]

Shib Sankar Dasgupta, Swayambhu Nath Ray, and Partha Talukdar. 2018. Hyte: Hyperplane-based temporally aware knowledge graph embedding. InProceedings of EMNLP. 2001–2011

work page 2018
[6]

Zhen Han, Peng Chen, Yunpu Ma, and Volker Tresp. 2020. Dyernie: Dynamic evolution of riemannian manifold embeddings for temporal knowledge graph completion.arXiv preprint arXiv:2011.03984(2020)

arXiv 2020
[7]

Zhen Han, Peng Chen, Yunpu Ma, and Volker Tresp. 2020. xERTE: Explainable Reasoning on Temporal Knowledge Graphs for Forecasting Future Links.arXiv preprint arXiv:2012.15537(2020)

work page Pith review arXiv 2020
[8]

Zhen Han, Zifeng Ding, Yunpu Ma, Yujia Gu, and Volker Tresp. 2021. Learning Neural Ordinary Equations for Forecasting Future Links on Temporal Knowledge Graphs. InProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, Online and Punta Cana, Dominican Republic, 8352–8364. https://d...

doi:10.18653/v1/2021.emnlp-main.658 2021

Show all 41 references

[9]

Zijie Huang, Zheng Li, Haoming Jiang, Tianyu Cao, Hanqing Lu, Bing Yin, Karthik Subbian, Yizhou Sun, and Wei Wang. 2022. Multilingual Knowledge Graph Completion with Self-Supervised Adaptive Graph Alignment.arXiv preprint arXiv:2203.14987(2022)

2022 arXiv
[10]

Prachi Jain, Sushant Rathi, Soumen Chakrabarti, et al. 2020. Temporal Knowledge Base Completion: New Algorithms and Evaluation Protocols.arXiv preprint arXiv:2005.05035(2020)

2020 arXiv
[11]

Woojeong Jin, Meng Qu, Xisen Jin, and Xiang Ren. 2019. Recurrent event network: Autoregressive structure inference over temporal knowledge graphs.arXiv preprint arXiv:1904.05530(2019)

2019 arXiv
[12]

Woojeong Jin, Meng Qu, Xisen Jin, and Xiang Ren. 2020. Recurrent Event Network: Autoregressive Structure Inference over Temporal Knowledge Graphs. InProceedings of EMNLP. 6669–6683

2020
[13]

Gjergji Kasneci, Maya Ramanath, Fabian Suchanek, and Gerhard Weikum. 2009. The YAGO-NAGA approach to knowledge discovery.ACM SIGMOD Record37, 4 (2009), 41–47

2009
[14]

James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, et al. 2017. Overcoming catastrophic forgetting in neural networks.Proceedings of the national academy of ...

2017
[15]

Julien Leblay and Melisachew Wudage Chekol. 2018. Deriving validity time in knowledge graph. InCompanion Proc. of the The Web Conference. 1771–1776

2018
[16]

Zhizhong Li and Derek Hoiem. 2018. Learning without forgetting.IEEE Transactions on Pattern Analysis and Machine Intelligence40, 12 (2018), 2935–2947. ACM Trans. Knowl. Discov. Data., Vol. 1, No. 1, Article 1. Publication date: January 2024. Towards Improving Long-Tail Entity ...

2018
[17]

Ke Liang, Lingyuan Meng, Meng Liu, Yue Liu, Wenxuan Tu, Siwei Wang, Sihang Zhou, Xinwang Liu, and Fuchun Sun. 2022. Reasoning over different types of knowledge graphs: Static, temporal and multi-modal.arXiv preprint arXiv:2212.05767(2022)

2022 arXiv
[18]

Mehrnoosh Mirtaheri, Ryan A Rossi, Sungchul Kim, Kanak Mahadik, Tong Yu, Xiang Chen, and Mohammad Rostami
[19]

Mehrnoosh Mirtaheri, Mohammad Rostami, and Aram Galstyan. 2023. History Repeats: Overcoming Catastrophic Forgetting For Event-Centric Temporal Knowledge Graph Completion.arXiv preprint arXiv:2305.18675(2023)

2023 arXiv
[20]

Shirui Pan, Linhao Luo, Yufei Wang, Chen Chen, Jiapu Wang, and Xindong Wu. 2023. Unifying Large Language Models and Knowledge Graphs: A Roadmap.arXiv preprint arXiv:2306.08302(2023)

2023 arXiv
[21]

David Rolnick, Arun Ahuja, Jonathan Schwarz, Timothy Lillicrap, and Gregory Wayne. 2019. Experience replay for continual learning.Advances in Neural Information Processing Systems32 (2019)

2019
[22]

Mohammad Rostami. 2021. Lifelong domain adaptation via consolidated internal distribution.Advances in neural information processing systems34 (2021), 11172–11183

2021
[23]

Mohammad Rostami, Soheil Kolouri, and Praveen Pilly. 2019. Complementary Learning for Overcoming Catastrophic Forgetting Using Experience Replay.. InIJCAI

2019
[24]

Tom Schaul, John Quan, Ioannis Antonoglou, and David Silver. 2016. Prioritized experience replay. InIJCLR

2016
[25]

Tong Shen, Fu Zhang, and Jingwei Cheng. 2022. A comprehensive overview of knowledge graph completion.Knowledge- Based Systems(2022), 109597

2022
[26]

Hanul Shin, Jung Kwon Lee, Jaehong Kim, and Jiwon Kim. 2017. Continual learning with deep generative replay. In NeurIPS. 2990–2999

2017
[27]

Hyun Je Song and Seong Bae Park. 2018. Enriching translation-based knowledge graph embeddings through continual learning.IEEE Access6 (2018), 60489–60497. https://doi.org/10.1109/ACCESS.2018.2874656

2018
[28]

Haohai Sun, Jialun Zhong, Yunpu Ma, Zhen Han, and Kun He. 2021. TimeTraveler: Reinforcement learning for temporal knowledge graph forecasting.arXiv preprint arXiv:2109.04101(2021)

2021 arXiv
[29]

Binh Tang and David S Matteson. 2021. Graph-Based Continual Learning. InInternational Conference on Learning Representations

2021
[30]

Rakshit Trivedi, Hanjun Dai, Yichen Wang, and Le Song. 2017. Know-evolve: Deep temporal reasoning for dynamic knowledge graphs. InInternational Conference on Machine Learning. PMLR, 3462–3471

2017
[31]

Junshan Wang, Guojie Song, Yi Wu, and Liang Wang. 2020. Streaming Graph Neural Networks via Continual Learning. InProceedings of the 29th ACM International Conference on Information & Knowledge Management. 1515–1524

2020
[32]

Liang Wang, Wei Zhao, Zhuoyu Wei, and Jingming Liu. 2022. SimKGC: Simple Contrastive Knowledge Graph Completion with Pre-trained Language Models.arXiv preprint arXiv:2203.02167(2022)

2022 arXiv
[33]

Zhihao Wang and Xin Li. 2019. Hybrid-TE: Hybrid Translation-Based Temporal Knowledge Graph Embedding. In IEEE ICTAI. IEEE, 1446–1451

2019
[34]

Jiapeng Wu, Meng Cao, Jackie Chi Kit Cheung, and William L Hamilton. 2020. Temp: Temporal message passing for temporal knowledge graph completion.arXiv preprint arXiv:2010.03526(2020)

2020 arXiv
[35]

Jiapeng Wu, Yishi Xu, Yingxue Zhang, Chen Ma, Mark Coates, and Jackie Chi Kit Cheung. 2021. Tie: A framework for embedding-based incremental temporal knowledge graph completion. InProceedings of the 44th international ACM SIGIR conference on research and development in informa...

2021
[36]

Jiapeng Wu, Yishi Xu, Yingxue Zhang, Chen Ma, Mark Coates, and Jackie Chi Kit Cheung. 2021. TIE: A Framework for Embedding-based Incremental Temporal Knowledge Graph Completion.SIGIR 2021 - Proceedings of the 44th International ACM SIGIR Conference on Research and Development ...

2021
[37]

Chengjin Xu, Mojtaba Nayyeri, Fouad Alkhoury, Hamed Shariat Yazdi, and Jens Lehmann. 2019. Temporal knowledge graph embedding model based on additive time series decomposition.arXiv preprint arXiv:1911.07893(2019)

2019 arXiv
[38]

Friedemann Zenke, Wulfram Gerstner, and Surya Ganguli. 2017. The temporal paradox of Hebbian learning and homeostatic plasticity.Curr. opinion in neuro.43 (2017), 166–176

2017
[39]

Fan Zhou and Chengtai Cao. 2021. Overcoming Catastrophic Forgetting in Graph Neural Networks with Experience Replay. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 35. 4714–4722

2021
[40]

Cunchao Zhu, Muhao Chen, Changjun Fan, Guangquan Cheng, and Yan Zhan. 2020. Learning from History: Modeling Temporal Knowledge Graphs with Sequential Copy-Generation Networks.arXiv preprint arXiv:2012.08492(2020). ACM Trans. Knowl. Discov. Data., Vol. 1, No. 1, Article 1. Publ...

2020 arXiv
[2024]

InCompanion Proceedings of the ACM on Web Conference 2024

Tackling Long-Tail Entities for Temporal Knowledge Graph Completion. InCompanion Proceedings of the ACM on Web Conference 2024. 497–500

2024

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

Antoine Bordes, Nicolas Usunier, Alberto Garcia-Duran, Jason Weston, and Oksana Yakhnenko. 2013. Translating embeddings for modeling multi-relational data.Advances in neural information processing systems26 (2013)

2013

[2] [2]

E Boschee, J Lautenschlager, S O’Brien, S Shellman, J Starz, and M Ward. 2015. Integrated Crisis Early Warning System (ICEWS) Coded Event Data.URL: https://dataverse. harvard. edu/dataverse/icews(2015)

work page 2015

[3] [3]

Yuanning Cui, Yuxin Wang, Zequn Sun, Wenqiang Liu, Yiqiao Jiang, Kexin Han, and Wei Hu. 2023. Lifelong embedding learning and transfer for growing knowledge graphs. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 37. 4217–4224

work page 2023

[4] [4]

Angel Daruna, Mehul Gupta, Mohan Sridharan, and Sonia Chernova. 2021. Continual learning of knowledge graph embeddings.IEEE Robotics and Automation Letters6 (4 2021), 1128–1135. Issue 2. https://doi.org/10.1109/LRA.2021. 3056071

doi:10.1109/lra.2021 2021

[5] [5]

Shib Sankar Dasgupta, Swayambhu Nath Ray, and Partha Talukdar. 2018. Hyte: Hyperplane-based temporally aware knowledge graph embedding. InProceedings of EMNLP. 2001–2011

work page 2018

[6] [6]

Zhen Han, Peng Chen, Yunpu Ma, and Volker Tresp. 2020. Dyernie: Dynamic evolution of riemannian manifold embeddings for temporal knowledge graph completion.arXiv preprint arXiv:2011.03984(2020)

arXiv 2020

[7] [7]

Zhen Han, Peng Chen, Yunpu Ma, and Volker Tresp. 2020. xERTE: Explainable Reasoning on Temporal Knowledge Graphs for Forecasting Future Links.arXiv preprint arXiv:2012.15537(2020)

work page Pith review arXiv 2020

[8] [8]

Zhen Han, Zifeng Ding, Yunpu Ma, Yujia Gu, and Volker Tresp. 2021. Learning Neural Ordinary Equations for Forecasting Future Links on Temporal Knowledge Graphs. InProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, Online and Punta Cana, Dominican Republic, 8352–8364. https://d...

doi:10.18653/v1/2021.emnlp-main.658 2021