Pith. sign in

REVIEW 4 major objections 4 minor 53 references

PathE: Leveraging Entity-Agnostic Paths for Parameter-Efficient Knowledge Graph Embeddings

T0 review · 4 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read PathE stores no entity embeddings; it computes entity representations from relational contexts and paths, achieving state-of-the-art relation prediction and competitive link prediction in relation-rich KGs with under 25% of the parameters…

desk verdict PathE is a genuinely new parameter-efficient KGE architecture with strong relation prediction numbers, but the path-mining ambiguity threatens the link prediction results and the SOTA claim needs a PathCon comparison. read the letter →

arxiv 2501.19095 v1 pith:ZZHKPPZZ submitted 2025-01-31 cs.AI cs.LG

Ioannis Reklos , Jacopo de Berardinis , Elena Simperl , Albert Meroño-Peñuela This is my paper · ORCID
classification cs.AIcs.LG
keywords knowledgegraphembeddingsparameter-efficiententity-agnosticrepresentationslinkpredictionrelationpath-basedreasoningrelationalcontexttransformers
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

PathE is a knowledge-graph embedding method that keeps no embedding table for entities at all. Instead, it learns to compute an entity's representation on demand from the relations that touch it and from random-walk paths through its neighbourhood, storing only relation embeddings and the weights of a few small networks. The paper claims this is enough to match or beat parameter-efficient baselines on relation prediction and to stay competitive on link prediction in relationally rich graphs, using less than a quarter of the parameters of earlier lightweight models. The stakes are scalability and induction: if entity tables can be replaced by relational context, models can handle millions of entities on consumer hardware and embed unseen entities without retraining.

What carries the argument

The central object is the relational-context projection: an entity is encoded as the multiset of relations it participates in, split into incoming and outgoing counts, and passed through a two-layer MLP called the node projector to give a d-dimensional seed vector. Around this, a path sequence model, implemented as a Transformer encoder, contextualises a batch of unique loopless random-walk paths rooted at the entity, with learned entity-focused positional encodings that measure distance from the head or tail. A path aggregator, which can be averaging, an LSTM, or another Transformer encoder, fuses the multiple path occurrences into a single head or tail representation, and separate prediction heads score triples for link prediction or relation prediction. The load-bearing property of this machinery is that every trainable component operates on relation embeddings plus shared projector weights, so the parameter count grows with the number of relations, not the number of entities.

What would settle it

Take a knowledge graph with many entities but very few relation types, for example a social graph with only 'follows' and 'friend' relations, so that most entities have identical incoming and outgoing relation multisets. If PathE's link-prediction MRR then falls to roughly the frequency baseline while an anchor-based model with stored entity tables keeps most of its score, the claim that relational context alone is sufficient for competitive link prediction is refuted.

Watch

Extended reading notes

Core claim

PathE claims that a fully entity-agnostic knowledge-graph embedding model, one that stores parameters only for relations and never for entities, can reach state-of-the-art relation prediction and competitive link prediction on path-rich graphs. The representation of an entity is produced by a node projector that reads the entity's relational context, meaning the counts of each relation in its incoming and outgoing edges, and maps that context through MLPs into a seed vector. That seed is refined by a Transformer over several unique loopless random-walk paths rooted at the entity, using entity-focused positional encodings, and the path occurrences are aggregated into one head or tail vector. On FB15k-237 and CoDEx-Large the paper reports mean reciprocal rank scores of 0.216 and 0.144 with 0.21M and 0.68M parameters respectively, beating the fully entity-agnostic NodePiece-without-anchors baseline and coming close to anchor-based NodePiece. On relation prediction it reports MRR 0.972 on FB15k-237 and 0.874 on WN18RR, above both RotatE and NodePiece. The paper attributes its weaker link prediction on WN18RR and YAGO3-10 to low relational diversity and to the small fraction of entities that have unique relational contexts.

Load-bearing premise

The whole model rests on the assumption that an entity's identity can be recovered from the relations and paths around it, so if many entities share the same relational context, the model has no way to tell them apart.

Editorial extensions

If this is right

  • Training and inference parameter counts scale with the relation vocabulary, so graphs with millions of entities but a small relation set become feasible on a single consumer GPU.
  • Unseen entities can be embedded at inference time without retraining, because representations are computed from relational context rather than looked up from a stored table.
  • On relation prediction, the paper reports that PathE beats both a standard embedding model and anchor-based parameter-efficient models on FB15k-237 and WN18RR while using under a million parameters.
  • On path-rich benchmarks, PathE's link prediction is competitive with anchor-based methods at a fraction of the parameter budget, giving it the highest efficiency score among the compared models.
  • The method's scope is explicit: it works best where relational diversity and unique relational contexts are high, and it lags on sparse, relation-poor graphs such as WN18RR and YAGO3-10.

Reading between the lines

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

  • A natural extension the paper leaves implicit is a hybrid model that feeds PathE's contextual embeddings into an anchor-based method, which could repair the sparse-relational failure mode while keeping most of the parameter savings.
  • The entity-focused positional encoding idea is not specific to KG embedding and could transfer to any sequence model that must localise a distinguished token, such as path-based reasoning over temporal or hypergraph data.
  • A cheap pre-test for whether PathE is a good fit is to compute the fraction of entities with unique incoming and outgoing relation multisets; the paper's appendix data suggest that this fraction predicts relative performance on the four benchmarks.
  • The most direct road to web-scale application is an evaluation on Wikidata itself, whose relational profile the paper's appendix notes resembles the path-rich benchmarks, but the paper leaves that experiment to future work.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. PathE proposes a knowledge-graph embedding method that stores no entity-specific embeddings. Entity representations are computed on the fly from (i) relation-context counts via a node projector and (ii) entity-relation paths processed by a Transformer encoder, with an aggregator and task-specific prediction heads for link and relation prediction. The paper evaluates PathE on FB15k-237, WN18RR, YAGO3-10, and CoDEx-Large, reports transductive and inductive results, and presents ablations over aggregation, path count, and positional encodings. Its central claims are that PathE uses less than 25% of the parameters of prior parameter-efficient methods, achieves state-of-the-art relation prediction, and remains competitive in link prediction on path-rich KGs.

Significance. The core idea is genuinely interesting: removing the entity embedding table entirely and deriving entity representations from relational paths is a clean way to obtain parameter efficiency and inductive capability. The paper is also transparent in reporting limitations on relation-sparse datasets, and Appendix E's analysis of relational diversity and unique contexts is a useful contribution. The ablations are informative, and the code is promised to be available. If the path-mining ambiguity is resolved and the empirical claims are matched to the reported numbers, the contribution would be a solid parameter-efficient KG embedding method with a clear applicability profile.

major comments (4)
  1. [3.1] Section 3.1 states that 'Training paths are created by mining random walks from each entity in the KG,' but it never specifies whether the edge set used for path mining is restricted to the training split. Because PathE stores no entity embeddings, the only way an entity can influence its representation is through the relational contexts and paths fed into the node projector and Transformer. If validation or test edges are included in the path-mining graph, then the representation of h and t already encodes information about held-out triples, invalidating the filtered MRR and Hits@K results in Table 1. The inductive section does not resolve this, as it only clarifies that the inference graph is disjoint in that setting. This ambiguity must be resolved by inspecting the released code or by rerunning the experiments with paths mined only from training edges.
  2. [4.3, Table 2] The claim of state-of-the-art relation prediction is not supported by the comparison presented. Section 2.3 cites PathCon [Wang et al., 2021] as achieving state-of-the-art performance on relation prediction, but Table 2 does not include PathCon or any other relation-prediction-specific method, only RotatE and NodePiece. Since PathCon is the stated SOTA baseline, omitting it from the comparison makes the 'state-of-the-art' claim in the abstract and conclusion unsubstantiated. The authors should include PathCon in the relation-prediction experiments or explicitly narrow the claim to parameter-efficient methods.
  3. [Abstract; Section 4.2, Table 1] The claim that PathE 'requires less than 25% of the parameters of previous parameter efficient methods' is not true on CoDEx-Large. From Table 1, PathE has 0.68M parameters while EARL has 2.1M, which is 32.4% of EARL's parameter count. The claim holds for FB15k-237 and WN18RR but is contradicted by the CoDEx-Large row. The paper should either qualify the claim as holding on most benchmarks, compare against a consistent set of parameter-efficient baselines on all benchmarks, or avoid stating a single threshold across all datasets.
  4. [Section 4.2, Table 1] The statement that PathE is 'competitive in link prediction on path-rich KGs' is overstated for CoDEx-Large. PathE's MRR is 0.144, compared with 0.190 for NodePiece with anchors and 0.238 for EARL. These are relative deficits of 24% and 39%, respectively, and PathE also lags behind NodePiece on FB15k-237 by 0.040 MRR. Given the paper's own framing of CoDEx-Large as a path-rich benchmark, the evidence in Table 1 supports 'competitive with NodePiece without anchors' or 'competitive on FB15k-237', but not a general claim of competitiveness on path-rich KGs. The wording should be aligned with the reported numbers.
minor comments (4)
  1. [4.1; Appendix B] The code URLs are inconsistent: Section 4.1 points to https://github.com/IReklos/PathE, while Appendix B gives https://anonymous.4open.science/r/kg embeddings/README.md. Please unify them.
  2. [3.6] In the relation prediction head description, 'This yields a probability distribution over all the possible relations in |E|' should read 'in |R|' since the prediction is over the relation vocabulary, not the entity set.
  3. [4.5] The interpretation of the PCA visualization in Figure 4 is unclear; the sentence 'in CoDEx-Large the separation happens at position 2, hence the model focuses on paths of length 7' does not follow straightforwardly from the figure. Please clarify the relation between embedding separation and path length.
  4. [Appendix C] In the inductive results discussion, RuleN is described as a 'path-based method' in one sentence, but it is a rule-mining method. Please use consistent terminology when categorizing baselines.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: PathE's reported predictions are produced by a trained encoder and evaluated on standard external benchmarks, not by construction from its inputs.

full rationale

The paper's claimed derivation chain is empirical and self-contained rather than definitional. PathE stores only relation embeddings; entity representations are computed from relational contexts and mined paths through the node projector (Eq. 1) and path/aggregator Transformers (Sections 3.3-3.5), then scored by separate link-prediction and relation-prediction heads trained with standard losses on benchmark train splits (Section 3.6, Section 4.1). Evaluations use the original train/validation/test splits and filtered ranking over full entity corruptions, so the reported MRR and Hits@K are genuine held-out predictions, not rearrangements of fitted quantities. The relation-prediction head is trained with cross-entropy on (head, tail, relation) triples and evaluated on held-out relations; this is ordinary supervised evaluation, not a fitted parameter renamed as a prediction. The Effi metric is explicitly defined as MRR divided by parameter count and is a reporting choice, not an input to training. No load-bearing self-citation chain or imported uniqueness theorem appears: prior-work citations are for baseline results, benchmark conventions, and generic components such as negative-sampling loss, none of which define PathE's outputs. The paper's own Appendix E is an honest limitation analysis showing that WN18RR has only 8% unique relational contexts, which explains rather than conceals the model's weak performance there. The skeptic's concern about possible path-mining leakage from the full KG is a potential data-handling validity issue, not circularity: even if paths were mined from the full graph, that would be test-set leakage affecting the validity of the benchmark numbers, not an equation that reduces the derivation to its inputs. Accordingly, there is no circular step to quote and the circularity score is 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The ledger is dominated by hyperparameters tuned on validation and by the domain assumption that relational context alone is sufficient for entity discrimination. No new physical or conceptual entities are introduced.

free parameters (5)
  • Embedding dimension d = 64 (FB15k-237, YAGO), 128 (WN18RR, CoDEx)
    Chosen via random search on validation; affects parameter count and capacity.
  • Paths per entity ppe = 4 (FB15k-237, CoDEx-L), 2 (WN18RR, YAGO)
    Tuned via random search; ablation shows diminishing returns.
  • Path transformer config (layers, heads, feedforward dim) = varies per dataset, see Table 7
    Selected by random search; one of the main capacity knobs.
  • Entity aggregation strategy = Transformer encoder (all datasets in best config)
    Ablation shows small gains over averaging; chosen via search.
  • Loss function and label smoothing = CE, smoothing 0.01-0.2
    Tuned via random search.
assumptions (4)
  • domain assumption Entities in path-rich KGs can be discriminated by their relational contexts (the multiset of relation types they appear with).
    The node projector (Section 3.3) builds entity representations only from relation counts; the paper's own analysis (Appendix E) shows this fails when unique relational contexts are rare (WN18RR, YAGO3-10).
  • domain assumption Random walks with no loops, of length up to 20, sampled with equal probability as incoming or outgoing, provide sufficient contextual information for entity representation.
    Path mining described in Section 3.1; no theoretical guarantee that these paths capture all relevant structure.
  • domain assumption Transformer encoders can learn to aggregate path information into discriminative entity representations.
    Architectural assumption; supported empirically only on two datasets.
  • standard math Standard filtered evaluation with all entities as corruptions is an unbiased measure of link prediction quality.
    Standard in KGE literature; used consistently.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PathE: Leveraging Entity-Agnostic Paths for Parameter-Efficient Knowledge Graph Embeddings." pith.science (2026). https://pith.science/paper/ZZHKPPZZ

@misc{pith2026250119095,
  author       = {Pith},
  title        = {Pith review of: PathE: Leveraging Entity-Agnostic Paths for Parameter-Efficient Knowledge Graph Embeddings},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZZHKPPZZ}},
  note         = {Machine review of arXiv:2501.19095}
}
read the original abstract

Knowledge Graphs (KGs) store human knowledge in the form of entities (nodes) and relations, and are used extensively in various applications. KG embeddings are an effective approach to addressing tasks like knowledge discovery, link prediction, and reasoning. This is often done by allocating and learning embedding tables for all or a subset of the entities. As this scales linearly with the number of entities, learning embedding models in real-world KGs with millions of nodes can be computationally intractable. To address this scalability problem, our model, PathE, only allocates embedding tables for relations (which are typically orders of magnitude fewer than the entities) and requires less than 25% of the parameters of previous parameter efficient methods. Rather than storing entity embeddings, we learn to compute them by leveraging multiple entity-relation paths to contextualise individual entities within triples. Evaluated on four benchmarks, PathE achieves state-of-the-art performance in relation prediction, and remains competitive in link prediction on path-rich KGs while training on consumer-grade hardware. We perform ablation experiments to test our design choices and analyse the sensitivity of the model to key hyper-parameters. PathE is efficient and cost-effective for relationally diverse and well-connected KGs commonly found in real-world applications.

Figures

Figures reproduced from arXiv: 2501.19095 by the authors.

Figure 1
Figure 1. Schematic overview of a PathE architecture, using an example triple [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Example portion of a training batch highlighting a positive [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Link prediction performance of PathE (MRR and [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Visualisation of the relative positional embeddings for en [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 47 canonical work pages

  1. [1]

    Boxe: A box embedding model for knowledge base completion

    Ralph Abboud, \.I smail \.I lkan Ceylan, Thomas Lukasiewicz, and Tommaso Salvatori. Boxe: A box embedding model for knowledge base completion. In NeurIPS , 2020

    work page 2020

  2. [2]

    PyKEEN 1.0: A Python Library for Training and Evaluating Knowledge Graph Embeddings

    Mehdi Ali, Max Berrendorf, Charles Tapley Hoyt, Laurent Vermue, Sahand Sharifzadeh, Volker Tresp, and Jens Lehmann. PyKEEN 1.0: A Python Library for Training and Evaluating Knowledge Graph Embeddings . JMLR , 22(82):1--6, 2021

    work page 2021

  3. [3]

    PRRL: path rotation based knowledge graph representation learning method

    Changhao Bai and Peng Wu. PRRL: path rotation based knowledge graph representation learning method. In BDCAT , 2021

    work page 2021

  4. [4]

    Freebase: a collaboratively created graph database for structuring human knowledge

    Kurt Bollacker, Colin Evans, Praveen Paritosh, Tim Sturge, and Jamie Taylor. Freebase: a collaboratively created graph database for structuring human knowledge. In SIGMOD , 2008

    work page 2008

  5. [5]

    Translating embeddings for modeling multi-relational data

    Antoine Bordes, Nicolas Usunier, Alberto Garc \' a - Dur \' a n, Jason Weston, and Oksana Yakhnenko. Translating embeddings for modeling multi-relational data. In NeurIPS , 2013

    work page 2013

  6. [6]

    Meta relational learning for few-shot link prediction in knowledge graphs

    Mingyang Chen, Wen Zhang, Wei Zhang, Qiang Chen, and Huajun Chen. Meta relational learning for few-shot link prediction in knowledge graphs. In EMNLP , 2019

    work page 2019

  7. [7]

    Pan, and Huajun Chen

    Mingyang Chen, Wen Zhang, Zhen Yao, Yushan Zhu, Yang Gao, Jeff Z. Pan, and Huajun Chen. Entity-agnostic representation learning for parameter-efficient knowledge graph embedding. In AAAI , 2023

    work page 2023

  8. [8]

    Go for a walk and arrive at the answer: Reasoning over paths in knowledge bases using reinforcement learning

    Rajarshi Das, Shehzaad Dhuliawala, Manzil Zaheer, Luke Vilnis, Ishan Durugkar, Akshay Krishnamurthy, Alex Smola, and Andrew McCallum. Go for a walk and arrive at the answer: Reasoning over paths in knowledge bases using reinforcement learning. In ICLR , 2018

    work page 2018

Show all 53 references
  1. [9]

    Convolutional 2d knowledge graph embeddings

    Tim Dettmers, Pasquale Minervini, Pontus Stenetorp, and Sebastian Riedel. Convolutional 2d knowledge graph embeddings. In AAAI , 2018

    2018

  2. [10]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805 , 2018

    2018 arXiv

  3. [11]

    Denis, Jiapeng Wu, and William L

    Mikhail Galkin, Etienne G. Denis, Jiapeng Wu, and William L. Hamilton. Nodepiece: Compositional and parameter-efficient representations of large knowledge graphs. In ICLR , 2022

    2022

  4. [12]

    A survey on knowledge graph-based recommender systems

    Qingyu Guo, Fuzhen Zhuang, Chuan Qin, Hengshu Zhu, Xing Xie, Hui Xiong, and Qing He. A survey on knowledge graph-based recommender systems. IEEE Trans. Knowl. Data Eng. , 34(8):3549--3568, 2022

    2022

  5. [13]

    Long short-term memory

    Sepp Hochreiter and J \" u rgen Schmidhuber. Long short-term memory. Neural Comput. , 9(8):1735--1780, 1997

    1997

  6. [14]

    Shaoxiong Ji, Shirui Pan, Erik Cambria, Pekka Marttinen, and Philip S. Yu. A survey on knowledge graphs: Representation, acquisition, and applications. IEEE Trans. Neural Networks Learn. Syst. , 33(2):494--514, 2022

    2022

  7. [15]

    Path-specific knowledge graph embedding

    Yantao Jia, Yuanzhuo Wang, Xiaolong Jin, and Xueqi Cheng. Path-specific knowledge graph embedding. Knowl. Based Syst. , 151:37--44, 2018

    2018

  8. [16]

    Roboplanner: autonomous robotic action planning via knowledge graph queries

    Ajay Kattepur and Balamuralidhar P. Roboplanner: autonomous robotic action planning via knowledge graph queries. In ACM/SIGAPP SAC , 2019

    2019

  9. [17]

    Suchanek

    Jonathan Lajus, Luis Gal \' a rraga, and Fabian M. Suchanek. Fast and exact rule mining with AMIE 3. In ESWC , 2020

    2020

  10. [18]

    Improving knowledge graph completion via increasing embedding interactions

    Weidong Li, Rong Peng, and Zhi Li. Improving knowledge graph completion via increasing embedding interactions. Appl. Intell. , 52(8):9289--9307, 2022

    2022

  11. [19]

    Modeling relation paths for representation learning of knowledge bases

    Yankai Lin, Zhiyuan Liu, Huan - Bo Luan, Maosong Sun, Siwei Rao, and Song Liu. Modeling relation paths for representation learning of knowledge bases. In EMNLP , 2015

    2015

  12. [20]

    Relation path embedding in knowledge graphs

    Xixun Lin, Yanchun Liang, Fausto Giunchiglia, Xiaoyue Feng, and Renchu Guan. Relation path embedding in knowledge graphs. Neural Comput. Appl. , 31(9):5629--5639, 2019

    2019

  13. [21]

    Adapting meta knowledge graph information for multi-hop reasoning over few-shot relations

    Xin Lv, Yuxian Gu, Xu Han, Lei Hou, Juanzi Li, and Zhiyuan Liu. Adapting meta knowledge graph information for multi-hop reasoning over few-shot relations. In EMNLP , 2019

    2019

  14. [22]

    Fine-grained evaluation of rule- and embedding-based systems for knowledge graph completion

    Christian Meilicke, Manuel Fink, Yanjie Wang, Daniel Ruffinelli, Rainer Gemulla, and Heiner Stuckenschmidt. Fine-grained evaluation of rule- and embedding-based systems for knowledge graph completion. In ISWC , 2018

    2018

  15. [23]

    Anytime bottom-up rule learning for knowledge graph completion

    Christian Meilicke, Melisachew Wudage Chekol, Daniel Ruffinelli, and Heiner Stuckenschmidt. Anytime bottom-up rule learning for knowledge graph completion. In IJCAI , 2019

    2019

  16. [24]

    Wordnet: a lexical database for english

    George A Miller. Wordnet: a lexical database for english. Communications of the ACM , 38(11):39--41, 1995

    1995

  17. [25]

    SKILL: structured knowledge infusion for large language models

    Fedor Moiseev, Zhe Dong, Enrique Alfonseca, and Martin Jaggi. SKILL: structured knowledge infusion for large language models. In ACL , 2022

    2022

  18. [26]

    Compositional vector space models for knowledge base completion

    Arvind Neelakantan, Benjamin Roth, and Andrew McCallum. Compositional vector space models for knowledge base completion. In ACL , 2015

    2015

  19. [27]

    Maximilian Nickel, Lorenzo Rosasco, and Tomaso A. Poggio. Holographic embeddings of knowledge graphs. In AAAI , 2016

    2016

  20. [28]

    Rule-guided compositional representation learning on knowledge graphs

    Guanglin Niu, Yongfei Zhang, Bo Li, Peng Cui, Si Liu, Jingyang Li, and Xiaowei Zhang. Rule-guided compositional representation learning on knowledge graphs. In AAAI , 2020

    2020

  21. [29]

    SAFRAN: an interpretable, rule-based link prediction method outperforming embedding models

    Simon Ott, Christian Meilicke, and Matthias Samwald. SAFRAN: an interpretable, rule-based link prediction method outperforming embedding models. In AKBC , 2021

    2021

  22. [30]

    Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas K \" o pf, Edward Z. Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Stei...

    2019

  23. [31]

    You CAN teach an old dog new tricks! on training knowledge graph embeddings

    Daniel Ruffinelli, Samuel Broscheit, and Rainer Gemulla. You CAN teach an old dog new tricks! on training knowledge graph embeddings. In ICLR , 2020

    2020

  24. [32]

    Codex: A comprehensive knowledge graph completion benchmark

    Tara Safavi and Danai Koutra. Codex: A comprehensive knowledge graph completion benchmark. arXiv preprint arXiv:2009.07810 , 2020

    2009 arXiv

  25. [33]

    Learning multi-granularity and adaptive representation for knowledge graph reasoning

    Ziyu Shang, Peng Wang, Wenjun Ke, Jiajun Liu, Hailang Huang, Guozheng Li, Chenxiao Wu, Jianghan Liu, Xiye Chen, and Yining Li. Learning multi-granularity and adaptive representation for knowledge graph reasoning. In IJCAI , 2024

    2024

  26. [34]

    M-walk: Learning to walk over graphs using monte carlo tree search

    Yelong Shen, Jianshu Chen, Po - Sen Huang, Yuqing Guo, and Jianfeng Gao. M-walk: Learning to walk over graphs using monte carlo tree search. In NeurIPS , 2018

    2018

  27. [35]

    Rotate: Knowledge graph embedding by relational rotation in complex space

    Zhiqing Sun, Zhi - Hong Deng, Jian - Yun Nie, and Jian Tang. Rotate: Knowledge graph embedding by relational rotation in complex space. In ICLR , 2019

    2019

  28. [36]

    Teru, Etienne G

    Komal K. Teru, Etienne G. Denis, and William L. Hamilton. Inductive relation prediction by subgraph reasoning. In ICML , 2020

    2020

  29. [37]

    Representing text for joint embedding of text and knowledge bases

    Kristina Toutanova, Danqi Chen, Patrick Pantel, Hoifung Poon, Pallavi Choudhury, and Michael Gamon. Representing text for joint embedding of text and knowledge bases. In EMNLP , 2015

    2015

  30. [38]

    Compositional learning of embeddings for relation paths in knowledge base and text

    Kristina Toutanova, Xi Victoria Lin, Wen - tau Yih, Hoifung Poon, and Chris Quirk. Compositional learning of embeddings for relation paths in knowledge base and text. In ACL , 2016

    2016

  31. [39]

    Complex embeddings for simple link prediction

    Th \' e o Trouillon, Johannes Welbl, Sebastian Riedel, \' E ric Gaussier, and Guillaume Bouchard. Complex embeddings for simple link prediction. In ICML , 2016

    2016

  32. [40]

    Talukdar

    Shikhar Vashishth, Soumya Sanyal, Vikram Nitin, and Partha P. Talukdar. Composition-based multi-relational graph convolutional networks. In ICLR , 2020

    2020

  33. [41]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. In NeurIPS , 2017

    2017

  34. [42]

    Relational message passing for knowledge graph completion

    Hongwei Wang, Hongyu Ren, and Jure Leskovec. Relational message passing for knowledge graph completion. In SIGKDD , 2021

    2021

  35. [43]

    Deeppath: A reinforcement learning method for knowledge graph reasoning

    Wenhan Xiong, Thien Hoang, and William Yang Wang. Deeppath: A reinforcement learning method for knowledge graph reasoning. In EMNLP , 2017

    2017

  36. [44]

    One-shot relational learning for knowledge graphs

    Wenhan Xiong, Mo Yu, Shiyu Chang, Xiaoxiao Guo, and William Yang Wang. One-shot relational learning for knowledge graphs. In EMNLP , 2018

    2018

  37. [45]

    Subgraph neighboring relations infomax for inductive link prediction on knowledge graphs

    Xiaohan Xu, Peng Zhang, Yongquan He, Chengpeng Chao, and Chaoyang Yan. Subgraph neighboring relations infomax for inductive link prediction on knowledge graphs. In IJCAI , 7 2022. Main Track

    2022

  38. [46]

    Modeling complex relationship paths for knowledge graph completion

    Ping Zeng, Qingping Tan, Xiankai Meng, Haoyu Zhang, and Jianjun Xu. Modeling complex relationship paths for knowledge graph completion. IEICE Trans. Inf. Syst. , 101-D(5):1393--1400, 2018

    2018

  39. [47]

    Knowledge graph reasoning with relational digraph

    Yongqi Zhang and Quanming Yao. Knowledge graph reasoning with relational digraph. In WWW , 2022

    2022

  40. [48]

    Discriminative path-based knowledge graph embedding for precise link prediction

    Maoyuan Zhang, Qi Wang, Wukui Xu, Wei Li, and Shuyuan Sun. Discriminative path-based knowledge graph embedding for precise link prediction. In ECIR , 2018

    2018

  41. [49]

    Path-rotate: Knowledge graph embedding by relational rotation of path in complex space

    Xiaohan Zhou, Yunhui Yi, and Geng Jia. Path-rotate: Knowledge graph embedding by relational rotation of path in complex space. In IEEE/CIC ICCC , 2021

    2021

  42. [50]

    Zhaocheng Zhu, Zuobai Zhang, Louis - Pascal A. C. Xhonneux, and Jian Tang. Neural bellman-ford networks: A general graph neural network framework for link prediction. In NeurIPS , 2021

    2021

  43. [51]

    A*net: a scalable path-based reasoning approach for knowledge graphs

    Zhaocheng Zhu, Xinyu Yuan, Mikhail Galkin, Sophie Xhonneux, Ming Zhang, Maxime Gazeau, and Jian Tang. A*net: a scalable path-based reasoning approach for knowledge graphs. In NeurIPS , 2024

    2024

  44. [52]

    A survey on application of knowledge graph

    Xiaohan Zou. A survey on application of knowledge graph. In Journal of Physics: Conference Series , volume 1487, page 012016. IOP Publishing, 2020

    2020

  45. [53]

    write newline

    " write newline "" before.all 'output.state := FUNCTION fin.entry add.period write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence '...

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.