pith. sign in

arxiv: 2606.18001 · v1 · pith:KUEVF4OSnew · submitted 2026-06-16 · 💻 cs.LG

Half a Link can Be Enough to Predict a Whole Link: Understanding Generalization in Knowledge Graph Foundation Models

Cosimo Gregucci , Obaidah Theeb , Daniel Hernandez , Antonio Vergari , Steffen Staab This is my paper

Pith reviewed 2026-06-27 01:48 UTC · model grok-4.3

classification 💻 cs.LG
keywords knowledge graph foundation modelszero-shot link predictionhalf-linksgeneralization analysisdiagnostic taxonomyunseen graphs
0
0 comments X

The pith

Knowledge graph foundation models often rely on seeing just one half of a link to predict the full triple on new graphs.

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

This paper investigates how knowledge graph foundation models generalize to predict links on entirely new graphs after a single training run. It shows that performance improves substantially when a test triple shares either its head-relation pair or its relation-tail pair with the training data, even if the full triple is unseen. The authors divide all test cases into four scenarios according to which of these two half-links have been observed and demonstrate that current models exploit the seen halves while struggling when neither half has appeared. Their taxonomy therefore functions as a diagnostic tool that separates easy cases from the harder ones that require genuine zero-shot reasoning.

Core claim

To predict a test triple (h,r,t) it might suffice in practice to have observed the half-link (h,r) or (r,t) in the inference graph. This observation yields a taxonomy of four scenarios based on combinations of seen and unseen half-links. State-of-the-art knowledge graph foundation models achieve higher accuracy precisely in the scenarios where at least one half-link was observed during training, while the fully unseen-half-link scenario remains distinctly harder. The taxonomy therefore serves as a diagnostic protocol that reveals where current models fall short of robust generalization.

What carries the argument

The taxonomy of four half-link observation scenarios for each test triple.

If this is right

  • Models will continue to show higher accuracy on triples that share at least one half-link with the training graph.
  • The no-half-link scenario isolates the portion of generalization that cannot be explained by partial overlap.
  • The four-scenario breakdown can be used to benchmark future models for balanced performance.
  • Improvements targeted at the unseen-half-link cases will be required before KGFMs can be considered fully robust zero-shot learners.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same half-link analysis could be applied to measure how much other graph models depend on partial pattern matching.
  • Training objectives that explicitly penalize reliance on single-half matches might force models to learn more abstract rules.
  • Dataset construction pipelines could deliberately balance the four scenarios to create harder generalization tests.

Load-bearing premise

The performance gaps across the four scenarios are caused by the models' use of seen half-links rather than by other dataset or training factors.

What would settle it

A model that achieves statistically indistinguishable accuracy in all four scenarios across multiple held-out graphs would falsify the claim that seen half-links drive the observed differences.

Figures

Figures reproduced from arXiv: 2606.18001 by Antonio Vergari, Cosimo Gregucci, Daniel Hernandez, Obaidah Theeb, Steffen Staab.

Figure 1
Figure 1. Figure 1: The inference graph supports test triples to fundamentally different scenarios, from both halves seen (SQSA), to one (SQUA, UQSA), to neither (UQUA). On a shared graph, the inference graph Gi (solid) and the test graph Gt (dotted), each scenario refers to a single test triple, the thick, dotted teal edge; the scenario it falls into depends on which of its halves are seen, that is, matched by a thick orange… view at source ↗
Figure 3
Figure 3. Figure 3: UQUA test triples can map to potentially missing relation-graph links. We draw the co-occurrence motifs each test triple would induce, solid if in G r i and dotted if in G r t ; those induced by the specific test triple (teal, left graph of each pair) are thick purple. SQSA (left): the induced t2t between comprises and hasLandmark is present (purple solid), since by construction the answer entity has an in… view at source ↗
read the original abstract

Knowledge graph (KG) foundation models (KGFMs) are zero-shot generalizers: trained once, they can predict links on unseen graphs without retraining. However, understanding when and how they can robustly generalize across KGs is still an open question. In this paper, we shed some light on their generalization mechanisms highlighting how their performance on unseen KGs is not uniform when it comes to partially seen links, which we call half-links. In fact, we show that to predict a test triple $(h,r,t)$ it might suffice in practice to have observed the half-link $(h,r)$ or $(r,t)$ in the inference graph. This yields a taxonomy of four scenarios when combinations of these half-links are observed or not. In a rigorous stratified analysis over these scenarios, we reveal that SoTA KGFMs use seen half links for predictions, while unseen half-links pose different challenges. As such, our finer-grained taxonomy can be a diagnostic protocol for robust KGFM generalization and highlights where novel KGFMs can improve.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The paper claims that knowledge graph foundation models (KGFMs) generalize on unseen graphs by exploiting 'half-links'—observed (h,r) or (r,t) pairs—for predicting test triples (h,r,t). It introduces a four-scenario taxonomy based on combinations of seen and unseen half-links, and through stratified empirical analysis shows that SoTA KGFMs achieve higher performance when half-links are seen while unseen half-links pose greater challenges; the taxonomy is positioned as a diagnostic protocol for robust generalization.

Significance. If validated with appropriate controls, the half-link taxonomy offers a practical diagnostic for pinpointing generalization failures in KGFMs and could guide development of models robust to fully unseen links. The work identifies a potentially actionable mechanism (partial link observation) that distinguishes current model behavior, which is a useful empirical contribution if the causal interpretation is substantiated.

major comments (3)
  1. [Abstract and §4] Abstract and §4 (stratified analysis): the central claim that performance differences demonstrate models 'use seen half links for predictions' is not supported without controls or matching for confounders such as entity degree, relation frequency, or local graph density; the observed ordering is compatible with purely statistical explanations unrelated to any half-link mechanism inside the model.
  2. [§3] §3 (taxonomy definition): the four-scenario stratification assumes the performance gap is caused by exploitation of half-link patterns, but no ablation, regression, or counterfactual analysis is reported to rule out correlated dataset properties as the driver.
  3. [Abstract] Abstract: no datasets, model implementations, statistical tests, or variance estimates are described, so the support for the empirical finding cannot be evaluated from the provided information.
minor comments (1)
  1. [§3] Ensure the inference graph definition is stated explicitly and consistently when defining the four scenarios.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below and indicate where revisions will strengthen the causal interpretation of our results.

read point-by-point responses

  1. Referee: [Abstract and §4] Abstract and §4 (stratified analysis): the central claim that performance differences demonstrate models 'use seen half links for predictions' is not supported without controls or matching for confounders such as entity degree, relation frequency, or local graph density; the observed ordering is compatible with purely statistical explanations unrelated to any half-link mechanism inside the model.

    Authors: We agree that the current stratification demonstrates a correlation between half-link visibility and performance but does not yet isolate the mechanism from potential confounders. In the revision we will add matched-subset and regression analyses controlling for entity degree, relation frequency, and local density to test whether the performance gap persists after accounting for these factors. revision: partial

  2. Referee: [§3] §3 (taxonomy definition): the four-scenario stratification assumes the performance gap is caused by exploitation of half-link patterns, but no ablation, regression, or counterfactual analysis is reported to rule out correlated dataset properties as the driver.

    Authors: The taxonomy is presented as an observational diagnostic based on the empirical ordering across scenarios. We acknowledge the absence of explicit ablations or counterfactuals. The revised manuscript will include regression models that treat half-link visibility as a predictor while controlling for dataset-level statistics, together with a limitations discussion of alternative explanations. revision: yes

  3. Referee: [Abstract] Abstract: no datasets, model implementations, statistical tests, or variance estimates are described, so the support for the empirical finding cannot be evaluated from the provided information.

    Authors: The abstract is intentionally concise; full details appear in Sections 4–5. To improve evaluability we will expand the abstract with a brief statement naming the primary datasets and models and noting that all reported results include standard deviations across multiple random seeds and statistical significance tests. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical stratification is self-contained

full rationale

The paper performs a post-hoc stratification of test triples into four half-link scenarios and reports observed performance differences. This is a direct empirical measurement on held-out data with no fitted parameters renamed as predictions, no self-definitional equations, and no load-bearing self-citations that reduce the central claim to prior author work. The taxonomy is defined from observable graph structure (presence/absence of (h,r) or (r,t) in the inference graph) and the reported result is the measured accuracy ordering across those strata; nothing in the provided text shows the ordering being forced by construction or by an ansatz smuggled through citation. The analysis therefore stands as an independent diagnostic protocol rather than a tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the domain assumption that half-links are the appropriate units for dissecting generalization behavior, plus the invented concept of the four-scenario taxonomy.

axioms (1)
  • domain assumption Half-links (h,r) and (r,t) capture the relevant mechanism for link prediction generalization in KGFMs
    Invoked to justify the taxonomy and stratified analysis.
invented entities (1)
  • half-link no independent evidence
    purpose: Decompose full triples to diagnose generalization
    New analytical unit introduced for this study

pith-pipeline@v0.9.1-grok · 5731 in / 1147 out tokens · 48589 ms · 2026-06-27T01:48:36.039236+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

39 extracted references · 21 canonical work pages · 1 internal anchor

  1. [1]

    Arvindh Arun, Sumit Kumar, Mojtaba Nayyeri, Bo Xiong, Ponnurangam Kumaraguru, Antonio Vergari, and Steffen Staab. 2025. https://doi.org/10.18653/V1/2025.EMNLP-MAIN.1621 SEMMA: A semantic aware knowledge graph foundation model . In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, EMNLP 2025, Suzhou, China, November 4-...

    work page doi:10.18653/v1/2025.emnlp-main.1621 2025

  2. [2]

    Antoine Bordes, Nicolas Usunier, Alberto Garc \' a - Dur \' a n, Jason Weston, and Oksana Yakhnenko. 2013. https://proceedings.neurips.cc/paper/2013/hash/1cecc7a77928ca8133fa24680a88d2f9-Abstract.html Translating embeddings for modeling multi-relational data . In Advances in Neural Information Processing Systems 26: 27th Annual Conference on Neural Inform...

    2013

  3. [3]

    Thu Bui, Carola - Bibiane Sch \" o nlieb, Bruno Ribeiro, Beatrice Bevilacqua, and Moshe Eliasof. 2025. https://doi.org/10.48550/ARXIV.2502.00190 On the effectiveness of random weights in graph neural networks . CoRR, abs/2502.00190

    work page doi:10.48550/arxiv.2502.00190 2025

  4. [4]

    Yihong Chen, Pasquale Minervini, Sebastian Riedel, and Pontus Stenetorp. 2021. https://doi.org/10.24432/C54K5W Relation prediction as an auxiliary training objective for improving multi-relational graph representations . In 3rd Conference on Automated Knowledge Base Construction, AKBC 2021, Virtual, October 4-8, 2021

    work page doi:10.24432/c54k5w 2021

  5. [5]

    Vic Degraeve, Gilles Vandewiele, Femke Ongenae, and Sofie Van Hoecke. 2022. https://doi.org/10.48550/ARXIV.2203.02424 R-GCN: the R could stand for random . CoRR, abs/2203.02424

    work page doi:10.48550/arxiv.2203.02424 2022

  6. [6]

    Tim Dettmers, Pasquale Minervini, Pontus Stenetorp, and Sebastian Riedel. 2018. https://doi.org/10.1609/AAAI.V32I1.11573 Convolutional 2d knowledge graph embeddings . In Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence, (AAAI-18), the 30th innovative Applications of Artificial Intelligence (IAAI-18), and the 8th AAAI Symposium o...

    work page doi:10.1609/aaai.v32i1.11573 2018

  7. [7]

    Boyang Ding, Quan Wang, Bin Wang, and Li Guo. 2018. https://doi.org/10.18653/V1/P18-1011 Improving knowledge graph embedding using simple constraints . In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, ACL 2018, Melbourne, Australia, July 15-20, 2018, Volume 1: Long Papers , pages 110--121. Association for Computa...

    work page doi:10.18653/v1/p18-1011 2018

  8. [8]

    Xin Luna Dong. 2018. https://doi.org/10.1145/3219819.3219938 Challenges and innovations in building a product knowledge graph . In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD 2018, London, UK, August 19-23, 2018 , page 2869. ACM

    work page doi:10.1145/3219819.3219938 2018

  9. [9]

    Mikhail Galkin, Max Berrendorf, and Charles Tapley Hoyt. 2022. https://doi.org/10.48550/ARXIV.2203.01520 An open challenge for inductive link prediction on knowledge graphs . CoRR, abs/2203.01520

    work page doi:10.48550/arxiv.2203.01520 2022

  10. [10]

    Mikhail Galkin, Xinyu Yuan, Hesham Mostafa, Jian Tang, and Zhaocheng Zhu. 2024. https://openreview.net/forum?id=jVEoydFOl9 Towards foundation models for knowledge graph reasoning . In The Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024 . OpenReview.net

    2024

  11. [11]

    In2023 IEEE 39th International Conference on Data Engineering (ICDE)

    Yuxia Geng, Jiaoyan Chen, Jeff Z. Pan, Mingyang Chen, Song Jiang, Wen Zhang, and Huajun Chen. 2023. https://doi.org/10.1109/ICDE55515.2023.00098 Relational message passing for fully inductive knowledge graph completion . In 39th IEEE International Conference on Data Engineering, ICDE 2023, Anaheim, CA, USA, April 3-7, 2023 , pages 1221--1233. IEEE

    work page doi:10.1109/icde55515.2023.00098 2023

  12. [12]

    Cosimo Gregucci, Mojtaba Nayyeri, Daniel Hern \' a ndez, and Steffen Staab. 2023. https://doi.org/10.1145/3543507.3583358 Link prediction with attention applied on multiple knowledge graph embedding models . In Proceedings of the ACM Web Conference 2023, WWW 2023, Austin, TX, USA, 30 April 2023 - 4 May 2023 , pages 2600--2610. ACM

    work page doi:10.1145/3543507.3583358 2023

  13. [13]

    Cosimo Gregucci, Bo Xiong, Daniel Hern \' a ndez, Lorenzo Loconte, Pasquale Minervini, Steffen Staab, and Antonio Vergari. 2025. https://proceedings.mlr.press/v267/gregucci25a.html Is complex query answering really complex? In Forty-second International Conference on Machine Learning, ICML 2025, Vancouver, BC, Canada, July 13-19, 2025 , Proceedings of Mac...

    2025

  14. [14]

    Takuo Hamaguchi, Hidekazu Oiwa, Masashi Shimbo, and Yuji Matsumoto. 2017. https://doi.org/10.24963/IJCAI.2017/250 Knowledge transfer for out-of-knowledge-base entities : A graph neural network approach . In Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence, IJCAI 2017, Melbourne, Australia, August 19-25, 2017 , page...

    work page doi:10.24963/ijcai.2017/250 2017

  15. [15]

    Daniel Scott Himmelstein, Antoine Lizee, Christine Hessler, Leo Brueggeman, Sabrina L Chen, Dexter Hadley, Ari Green, Pouya Khankhanian, and Sergio E Baranzini. 2017. https://doi.org/10.7554/eLife.26726 Systematic integration of biomedical knowledge prioritizes drugs for repurposing . elife, 6:e26726

    work page doi:10.7554/elife.26726 2017

  16. [16]

    Knowledgegraphs.ACM Computing Surveys, 54(4):1–37, 2021

    Aidan Hogan, Eva Blomqvist, Michael Cochez, Claudia d'Amato, Gerard de Melo, Claudio Gutierrez, Sabrina Kirrane, Jos \' e Emilio Labra Gayo, Roberto Navigli, Sebastian Neumaier, Axel - Cyrille Ngonga Ngomo, Axel Polleres, Sabbir M. Rashid, Anisa Rula, Lukas Schmelzeisen, Juan F. Sequeda, Steffen Staab, and Antoine Zimmermann. 2022. https://doi.org/10.1145...

    work page doi:10.1145/3447772 2022

  17. [17]

    Bronstein, \.I smail \.I lkan Ceylan, Mikhail Galkin, Juan L

    Xingyue Huang, Pablo Barcel \' o , Michael M. Bronstein, \.I smail \.I lkan Ceylan, Mikhail Galkin, Juan L. Reutter, and Miguel A. Romero Orth. 2025. https://proceedings.mlr.press/v267/huang25a.html How expressive are knowledge graph foundation models? In Forty-second International Conference on Machine Learning, ICML 2025, Vancouver, BC, Canada, July 13-...

    2025

  18. [18]

    In Proceedings of the 2022 International Conference on Management of Data (Philadelphia, PA, USA) (SIGMOD ’22)

    Ihab F. Ilyas, Theodoros Rekatsinas, Vishnu Konda, Jeffrey Pound, Xiaoguang Qi, and Mohamed A. Soliman. 2022. https://doi.org/10.1145/3514221.3526049 Saga: A platform for continuous construction and serving of knowledge at scale . In SIGMOD '22: International Conference on Management of Data, Philadelphia, PA, USA, June 12 - 17, 2022 , pages 2259--2272. ACM

    work page doi:10.1145/3514221.3526049 2022

  19. [19]

    Flock: A Knowledge Graph Foundation Model via Learning on Random Walks

    Jinwoo Kim, Xingyue Huang, Krzysztof Olejniczak, Kyungbin Min, Michael M. Bronstein, Seunghoon Hong, and \.I smail \.I lkan Ceylan. 2025. https://doi.org/10.48550/ARXIV.2510.01510 Flock: A knowledge graph foundation model via learning on random walks . CoRR, abs/2510.01510

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2510.01510 2025

  20. [20]

    Jaejun Lee, Chanyoung Chung, and Joyce Jiyoung Whang. 2023. https://proceedings.mlr.press/v202/lee23c.html Ingram: Inductive knowledge graph embedding via relation graphs . In International Conference on Machine Learning, ICML 2023, 23-29 July 2023, Honolulu, Hawaii, USA , Proceedings of Machine Learning Research, pages 18796--18809. PMLR

    2023

  21. [21]

    Kostylev

    Shuwen Liu, Bernardo Cuenca Grau, Ian Horrocks, and Egor V. Kostylev. 2021. https://proceedings.neurips.cc/paper/2021/hash/0fd600c953cde8121262e322ef09f70e-Abstract.html INDIGO: gnn-based inductive knowledge graph completion using pair-wise encoding . In Advances in Neural Information Processing Systems 34: Annual Conference on Neural Information Processi...

    2021

  22. [22]

    Lorenzo Loconte, Nicola Di Mauro, Robert Peharz, and Antonio Vergari. 2023. http://papers.nips.cc/paper\_files/paper/2023/hash/f4b768188be63b8d2680a46934fd295a-Abstract-Conference.html How to turn your knowledge graph embeddings into generative models . In Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Proces...

    2023

  23. [23]

    Xin Lv, Xu Han, Lei Hou, Juanzi Li, Zhiyuan Liu, Wei Zhang, Yichi Zhang, Hao Kong, and Suhui Wu. 2020. https://doi.org/10.18653/V1/2020.EMNLP-MAIN.459 Dynamic anticipation and completion for multi-hop reasoning over sparse knowledge graph . In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, EMNLP 2020, Online, Novem...

    work page doi:10.18653/v1/2020.emnlp-main.459 2020

  24. [24]

    Suchanek

    Farzaneh Mahdisoltani, Joanna Biega, and Fabian M. Suchanek. 2015. http://cidrdb.org/cidr2015/Papers/CIDR15\_Paper1.pdf YAGO3: A knowledge base from multilingual wikipedias . In Seventh Biennial Conference on Innovative Data Systems Research, CIDR 2015, Asilomar, CA, USA, January 4-7, 2015, Online Proceedings . www.cidrdb.org

    2015

  25. [25]

    Chaitanya Malaviya, Chandra Bhagavatula, Antoine Bosselut, and Yejin Choi. 2020. https://doi.org/10.1609/AAAI.V34I03.5684 Commonsense knowledge base completion with structural and semantic context . In The Thirty-Fourth AAAI Conference on Artificial Intelligence, AAAI 2020, The Thirty-Second Innovative Applications of Artificial Intelligence Conference, I...

    work page doi:10.1609/aaai.v34i03.5684 2020

  26. [26]

    Aisha Mohamed, Shameem Puthiya Parambath, Zoi Kaoudi, and Ashraf Aboulnaga. 2020. http://proceedings.mlr.press/v124/mohamed20a.html Popularity agnostic evaluation of knowledge graph embeddings . In Proceedings of the Thirty-Sixth Conference on Uncertainty in Artificial Intelligence, UAI 2020, virtual online, August 3-6, 2020 , Proceedings of Machine Learn...

    2020

  27. [27]

    Maximilian Nickel, Xueyan Jiang, and Volker Tresp. 2014. https://proceedings.neurips.cc/paper_files/paper/2014/file/e6c6717ff809b4dba22d1cbfdc1c6696-Paper.pdf Reducing the rank in relational factorization models by including observable patterns . In Advances in Neural Information Processing Systems 27: Annual Conference on Neural Information Processing Sy...

    2014

  28. [28]

    Tara Safavi and Danai Koutra. 2020. https://doi.org/10.18653/V1/2020.EMNLP-MAIN.669 Codex: A comprehensive knowledge graph completion benchmark . In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, EMNLP 2020, Online, November 16-20, 2020 , pages 8328--8350. Association for Computational Linguistics

    work page doi:10.18653/v1/2020.emnlp-main.669 2020

  29. [29]

    Zhiqing Sun, Zhi - Hong Deng, Jian - Yun Nie, and Jian Tang. 2019. https://openreview.net/forum?id=HkgEQnRqYQ Rotate: Knowledge graph embedding by relational rotation in complex space . In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019 . OpenReview.net

    2019

  30. [30]

    Teru, Etienne G

    Komal K. Teru, Etienne G. Denis, and William L. Hamilton. 2020. http://proceedings.mlr.press/v119/teru20a.html Inductive relation prediction by subgraph reasoning . In Proceedings of the 37th International Conference on Machine Learning, ICML 2020, 13-18 July 2020, Virtual Event , Proceedings of Machine Learning Research, pages 9448--9457. PMLR

    2020

  31. [31]

    Kristina Toutanova and Danqi Chen. 2015. https://doi.org/10.18653/V1/W15-4007 Observed versus latent features for knowledge base and text inference . In Proceedings of the 3rd Workshop on Continuous Vector Space Models and their Compositionality, CVSC 2015, Beijing, China, July 26-31, 2015 , pages 57--66. Association for Computational Linguistics

    work page doi:10.18653/v1/w15-4007 2015

  32. [32]

    Th \' e o Trouillon, Johannes Welbl, Sebastian Riedel, \' E ric Gaussier, and Guillaume Bouchard. 2016. http://proceedings.mlr.press/v48/trouillon16.html Complex embeddings for simple link prediction . In Proceedings of the 33nd International Conference on Machine Learning, ICML 2016, New York City, NY, USA, June 19-24, 2016 , JMLR Workshop and Conference...

    2016

  33. [33]

    Yingxin Wu, Xiang Wang, An Zhang, Xiangnan He, and Tat - Seng Chua. 2022. https://openreview.net/forum?id=hGXij5rfiHw Discovering invariant rationales for graph neural networks . In The Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022 . OpenReview.net

    2022

  34. [34]

    Wenhan Xiong, Thien Hoang, and William Yang Wang. 2017. https://doi.org/10.18653/V1/D17-1060 Deeppath: A reinforcement learning method for knowledge graph reasoning . In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, EMNLP 2017, Copenhagen, Denmark, September 9-11, 2017 , pages 564--573. Association for Computation...

    work page doi:10.18653/v1/d17-1060 2017

  35. [35]

    Yongqi Zhang and Quanming Yao. 2022. https://doi.org/10.1145/3485447.3512008 Knowledge graph reasoning with relational digraph . In WWW '22: The ACM Web Conference 2022, Virtual Event, Lyon, France, April 25 - 29, 2022 , pages 912--924. ACM

    work page doi:10.1145/3485447.3512008 2022

  36. [36]

    Yucheng Zhang, Beatrice Bevilacqua, Mikhail Galkin, and Bruno Ribeiro. 2024. https://proceedings.mlr.press/v269/zhang25a.html TRIX: A more expressive model for zero-shot domain transfer in knowledge graphs . In Learning on Graphs Conference, 26-29 November 2024, Virtual, Proceedings of Machine Learning Research, page 12. PMLR

    2024

  37. [37]

    Jincheng Zhou, Beatrice Bevilacqua, and Bruno Ribeiro. 2023. https://openreview.net/forum?id=KIISfvU806 A multi-task perspective for link prediction with new relation types and nodes . In NeurIPS 2023 Workshop: New Frontiers in Graph Learning

    2023

  38. [38]

    Zhaocheng Zhu, Xinyu Yuan, Michael Galkin, Louis - Pascal A. C. Xhonneux, Ming Zhang, Maxime Gazeau, and Jian Tang. 2023. http://papers.nips.cc/paper\_files/paper/2023/hash/b9e98316cb72fee82cc1160da5810abc-Abstract-Conference.html A*net: A scalable path-based reasoning approach for knowledge graphs . In Advances in Neural Information Processing Systems 36...

    2023

  39. [39]

    Zhaocheng Zhu, Zuobai Zhang, Louis - Pascal A. C. Xhonneux, and Jian Tang. 2021. https://proceedings.neurips.cc/paper/2021/hash/f6a673f09493afcd8b129a0bcf1cd5bc-Abstract.html Neural bellman-ford networks: A general graph neural network framework for link prediction . In Advances in Neural Information Processing Systems 34: Annual Conference on Neural Info...

    2021