Adaptive Margin Ranking Loss for Knowledge Graph Embeddings via a Correntropy Objective Function

Hamed Shariat Yazdi; Jens Lehmann; Mojtaba Nayyeri; Sahar Vahdati; Xiaotian Zhou

arxiv: 1907.05336 · v1 · pith:LNELJWTHnew · submitted 2019-07-09 · 💻 cs.CL · cs.AI· cs.LG· stat.ML

Adaptive Margin Ranking Loss for Knowledge Graph Embeddings via a Correntropy Objective Function

Mojtaba Nayyeri , Xiaotian Zhou , Sahar Vahdati , Hamed Shariat Yazdi , Jens Lehmann This is my paper

Pith reviewed 2026-05-25 00:31 UTC · model grok-4.3

classification 💻 cs.CL cs.AIcs.LGstat.ML

keywords Adaptive Margin LossTransEKnowledge Graph EmbeddingsLink PredictionCorrentropyMargin Ranking LossTranslation-based Models

0 comments

The pith

Adaptive Margin Loss lets the margin for TransE training expand automatically until it converges, so only a center value needs to be chosen instead of upper and lower bounds.

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

The paper introduces Adaptive Margin Loss (AML) as a new objective for training translation-based knowledge graph embeddings such as TransE. AML is built on a correntropy objective that drives the margin to widen on its own during optimization. This replaces the need to tune separate upper and lower score bounds that earlier margin-based losses require. A reader would care because the change shrinks the hyperparameter search while still pushing positive triple scores toward the small values demanded by the original TransE translation assumption. Experiments on Freebase and WordNet show the approach works for link prediction.

Core claim

The formulation of the proposed loss function enables an adaptive and automated adjustment of the margin during the learning process. Therefore, instead of obtaining two values (upper bound and lower bound), only the center of a margin needs to be determined. During learning, the margin is expanded automatically until it converges. In experiments on standard benchmark datasets including Freebase and WordNet, the effectiveness of AML is confirmed for training TransE on link prediction tasks.

What carries the argument

Adaptive Margin Loss (AML), a correntropy-based ranking loss whose margin widens automatically from a single center parameter until convergence.

If this is right

Only one scalar (the margin center) must be chosen instead of separate upper and lower bounds.
The margin grows automatically during training until it stabilizes.
Positive triple scores are driven toward the small values required by the TransE translation rule.
Performance on link prediction improves on Freebase and WordNet when TransE is trained with AML.

Where Pith is reading between the lines

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

The single-center design could be tested on other margin-based objectives outside knowledge graphs to see whether automatic expansion generalizes.
If convergence proves robust across random seeds, the method might cut the number of validation runs needed when scaling TransE to larger graphs.
A direct comparison of final margin sizes across datasets would reveal whether the converged value correlates with graph density or relation cardinality.

Load-bearing premise

The correntropy-driven margin will converge to a stable value that keeps positive scores small enough for the TransE translation assumption without instability or extra regularization.

What would settle it

A run on one of the paper's benchmark datasets in which the learned margin either diverges, oscillates, or produces positive scores that remain too large, yielding link-prediction hits@10 no better than a fixed-margin baseline.

Figures

Figures reproduced from arXiv: 1907.05336 by Hamed Shariat Yazdi, Jens Lehmann, Mojtaba Nayyeri, Sahar Vahdati, Xiaotian Zhou.

**Figure 1.** Figure 1: Illustration of Loss Functions. γ . erefore, positive triples are separated from negative samples. However, using MRL includes the existence of cases where the score of a correct triple (h,r,t) is not suciently small for h + r ' t to hold. A combination of limit-based scoring loss functions for a set of translation-based embedding models [29] have been proposed in order to avoid such cases. By adding a l… view at source ↗

**Figure 2.** Figure 2: Illustration of Adaptive Margin Loss. and [−fr (h 0 ,t 0 ) + γ + ξ]+) and considering Equation 7 and Equation 9, instead of solving Equation 10, the following loss function is minimized: L = λξ 2 + λ+ [fr (h,t) −γ + ξ]+ + λ− [−fr (h 0 ,t 0 ) +γ + ξ]+ (11) e algorithm starts with a value for ξ, i.e. M. Because the loss Equation 11 is minimized, ξ 2 −→ m where m < M. erefore, the margin shrinks from 2M to… view at source ↗

read the original abstract

Translation-based embedding models have gained significant attention in link prediction tasks for knowledge graphs. TransE is the primary model among translation-based embeddings and is well-known for its low complexity and high efficiency. Therefore, most of the earlier works have modified the score function of the TransE approach in order to improve the performance of link prediction tasks. Nevertheless, proven theoretically and experimentally, the performance of TransE strongly depends on the loss function. Margin Ranking Loss (MRL) has been one of the earlier loss functions which is widely used for training TransE. However, the scores of positive triples are not necessarily enforced to be sufficiently small to fulfill the translation from head to tail by using relation vector (original assumption of TransE). To tackle this problem, several loss functions have been proposed recently by adding upper bounds and lower bounds to the scores of positive and negative samples. Although highly effective, previously developed models suffer from an expansion in search space for a selection of the hyperparameters (in particular the upper and lower bounds of scores) on which the performance of the translation-based models is highly dependent. In this paper, we propose a new loss function dubbed Adaptive Margin Loss (AML) for training translation-based embedding models. The formulation of the proposed loss function enables an adaptive and automated adjustment of the margin during the learning process. Therefore, instead of obtaining two values (upper bound and lower bound), only the center of a margin needs to be determined. During learning, the margin is expanded automatically until it converges. In our experiments on a set of standard benchmark datasets including Freebase and WordNet, the effectiveness of AML is confirmed for training TransE on link prediction tasks.

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.

Desk Editor's Note private letter to a colleague

AML gives a practical reduction in hyperparameters for TransE training by letting the margin adapt via correntropy, but the convergence claim has no supporting analysis.

read the letter

The paper's main contribution is a new loss called Adaptive Margin Loss that uses correntropy to adjust the margin automatically during training of translation-based models like TransE. Instead of picking both an upper and lower bound on scores, you only set a center value and the margin expands on its own until it settles. This directly targets the weakness in standard margin ranking loss, where positive triples are not forced close enough to satisfy the head + relation ≈ tail assumption. The abstract reports that this works on Freebase and WordNet for link prediction, which is the expected test bed for this line of work. If the adaptation is stable in practice, it does cut the hyperparameter search space in a useful way for people who train these models regularly. The experiments are presented as confirmation, though the abstract gives no numbers, ablations, or comparison tables, so the strength of the empirical case is hard to judge from what is shown. The soft spot is the missing justification for the central claim. The description says the margin expands automatically until convergence, but there is no derivation of the update rule, no fixed-point argument, and no discussion of conditions on the kernel bandwidth or learning rate that would prevent oscillation or divergence. Without that, the reduction to a single center parameter is not yet shown to be reliable rather than heuristic. The work stays inside the TransE paradigm and does not claim broader applicability. It is the sort of incremental loss tweak that people in knowledge graph embedding research might try if they are already tuning TransE heavily. The idea is clear enough and the motivation is sound, so it is worth sending to referees who can check the full experiments and any additional analysis that may be in the body. I would not cite it yet without seeing those details.

Referee Report

2 major / 2 minor

Summary. The paper proposes Adaptive Margin Loss (AML), a new loss function for training translation-based KG embedding models such as TransE. AML is derived from a correntropy objective and is claimed to enable automatic, adaptive expansion of the margin during training until convergence, thereby requiring only a single hyperparameter (the margin center) instead of separate upper and lower bounds on positive and negative scores. Experiments on standard link-prediction benchmarks (Freebase, WordNet) are reported to confirm improved performance.

Significance. If the claimed convergence behavior holds, AML would meaningfully reduce the hyperparameter burden for margin-based losses while still enforcing the core TransE translation assumption that positive scores remain small. The experimental validation on two widely used datasets constitutes a concrete strength. However, the absence of any derivation, fixed-point analysis, or stability argument for the margin dynamics under correntropy substantially limits the result's theoretical significance.

major comments (2)

[AML formulation (abstract and §3)] The central claim that the correntropy-driven margin expands automatically and converges to a stable value (thereby reducing the search space to a single center parameter) is load-bearing, yet the manuscript supplies neither a derivation of the margin-update dynamics nor any convergence or Lyapunov-style argument. This directly affects the weakest assumption identified in the stress-test note.
[Experiments (§4)] No ablation isolating the adaptive-margin component, no plots or statistics tracking margin evolution across epochs, and no discussion of conditions (kernel bandwidth, learning-rate regimes) under which convergence is guaranteed appear in the experimental section. Without these, the reported gains on Freebase and WordNet cannot be attributed to the claimed adaptive mechanism.

minor comments (2)

[§2] Notation for the correntropy kernel bandwidth and the precise definition of the margin center should be introduced earlier and used consistently.
[Abstract and §4] The abstract states that 'the effectiveness of AML is confirmed' but does not report standard deviation across runs or statistical significance tests; these should be added for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript to strengthen both the theoretical derivation and the experimental validation of the adaptive margin mechanism.

read point-by-point responses

Referee: [AML formulation (abstract and §3)] The central claim that the correntropy-driven margin expands automatically and converges to a stable value (thereby reducing the search space to a single center parameter) is load-bearing, yet the manuscript supplies neither a derivation of the margin-update dynamics nor any convergence or Lyapunov-style argument. This directly affects the weakest assumption identified in the stress-test note.

Authors: We agree that the manuscript would benefit from an explicit derivation of the margin-update dynamics and a convergence argument. While Section 3 motivates the adaptive behavior via the correntropy objective, it does not contain a fixed-point analysis or Lyapunov-style argument. In the revised version we will add a dedicated subsection deriving the margin dynamics from the correntropy formulation and discussing conditions for convergence. revision: yes
Referee: [Experiments (§4)] No ablation isolating the adaptive-margin component, no plots or statistics tracking margin evolution across epochs, and no discussion of conditions (kernel bandwidth, learning-rate regimes) under which convergence is guaranteed appear in the experimental section. Without these, the reported gains on Freebase and WordNet cannot be attributed to the claimed adaptive mechanism.

Authors: We acknowledge the absence of these elements. The revised manuscript will add (i) an ablation isolating the adaptive-margin component, (ii) plots and statistics showing margin evolution across epochs on the benchmark datasets, and (iii) a discussion of the kernel-bandwidth and learning-rate regimes under which stable convergence is observed. revision: yes

Circularity Check

0 steps flagged

No circularity: AML is an independent loss formulation with no self-referential reduction

full rationale

The paper defines AML directly from a correntropy objective and presents the adaptive margin expansion as an emergent property of that definition rather than a fitted parameter or self-cited uniqueness theorem. No equations equate a derived quantity to its own inputs by construction, no margin update is obtained by fitting to a subset of the target data, and no load-bearing premise rests on prior self-citation. The derivation chain is self-contained against external benchmarks and does not rename known results or smuggle ansatzes via citation.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the unproven assumption that the correntropy-driven adaptive margin will converge usefully; one free parameter (margin center) is introduced; no new entities are postulated.

free parameters (1)

margin center
The single value that must still be chosen; all other margin behavior is claimed to emerge automatically during training.

axioms (1)

domain assumption The margin can be expanded automatically during learning until it converges to an effective value.
Invoked when describing how AML replaces manual upper/lower bound selection.

pith-pipeline@v0.9.0 · 5858 in / 1326 out tokens · 22780 ms · 2026-05-25T00:31:05.368793+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

32 extracted references · 32 canonical work pages · 2 internal anchors

[1]

Sam Adams. 2019. Sur/f_ing the Hype Cycle to In/f_inity and Beyond.Research- Technology Management 62, 3 (2019), 45–51

work page 2019
[2]

Anonymous. 2018. Relation Pa/t_tern Encoded Knowledge Graph Embedding by Translating in Complex Space. (2018). anonymous preprint under review

work page 2018
[3]

Stefano A Bini. 2018. Arti/f_icial intelligence, machine learning, deep learning, and cognitive computing: what do these terms mean and how will they impact health care? /T_he Journal of arthroplasty33, 8 (2018), 2358–2361

work page 2018
[4]

Kurt Bollacker, Colin Evans, Praveen Paritosh, Tim Sturge, and Jamie Taylor

work page
[5]

In Proceedings of the 2008 ACM SIGMOD international conference on Management of data

Freebase: a collaboratively created graph database for structuring human knowledge. In Proceedings of the 2008 ACM SIGMOD international conference on Management of data. AcM, 1247–1250

work page 2008
[6]

Antoine Bordes, Sumit Chopra, and Jason Weston. 2014. /Q_uestion answering with subgraph embeddings. arXiv preprint arXiv:1406.3676 (2014)

work page internal anchor Pith review Pith/arXiv arXiv 2014
[7]

Antoine Bordes, Xavier Glorot, Jason Weston, and Yoshua Bengio. 2012. Joint learning of words and meaning representations for open-text semantic parsing. In Arti/f_icial Intelligence and Statistics. 127–135

work page 2012
[8]

Antoine Bordes, Nicolas Usunier, Alberto Garcia-Duran, Jason Weston, and Oksana Yakhnenko. 2013. Translating embeddings for modeling multi-relational data. In Advances in neural information processing systems . 2787–2795

work page 2013
[9]

Antoine Bordes, Jason Weston, Ronan Collobert, and Yoshua Bengio. 2011. Learn- ing structured embeddings of knowledge bases. In Twenty-Fi/f_th AAAI Conference on Arti/f_icial Intelligence

work page 2011
[10]

Stephen Boyd and Lieven Vandenberghe. 2004. Convex optimization. Cambridge university press

work page 2004
[11]

Andrew Carlson, Justin Be/t_teridge, Bryan Kisiel, Burr Se/t_tles, Estevam R Hr- uschka, and Tom M Mitchell. 2010. Toward an architecture for never-ending language learning. In Twenty-Fourth AAAI Conference on Arti/f_icial Intelligence

work page 2010
[12]

Huimin Chen, Maosong Sun, Cunchao Tu, Yankai Lin, and Zhiyuan Liu. 2016. Neural sentiment classi/f_ication with user and product a/t_tention. InProceedings of the 2016 conference on empirical methods in natural language processing . 1650– 1659

work page 2016
[13]

Shizhu He, Kang Liu, Yuanzhe Zhang, Liheng Xu, and Jun Zhao. 2014. /Q_uestion answering over linked data using /f_irst-order logic. InProceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) . 1092– 1103

work page 2014
[14]

Konrad H¨oﬀner, Sebastian Walter, Edgard Marx, Ricardo Usbeck, Jens Lehmann, and Axel-Cyrille Ngonga Ngomo. 2016. Survey on Challenges of /Q_uestion Answering in the Semantic Web. Semantic Web Journal (2016)

work page 2016
[15]

Rodolphe Jena/t_ton, Nicolas L Roux, Antoine Bordes, and Guillaume R Obozinski

work page
[16]

InAdvances in Neural Information Processing Systems

A latent factor model for highly multi-relational data. InAdvances in Neural Information Processing Systems. 3167–3175

work page
[17]

Guoliang Ji, Shizhu He, Liheng Xu, Kang Liu, and Jun Zhao. 2015. Knowledge graph embedding via dynamic mapping matrix. In Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers) , Vol. 1. 687–696

work page 2015
[18]

Jens Lehmann, Robert Isele, Max Jakob, Anja Jentzsch, Dimitris Kontokostas, Pablo Mendes, Sebastian Hellmann, Mohamed Morsey, Patrick van Kleef, S¨oren Auer, and Chris Bizer. 2015. DBpedia - A Large-scale, Multilingual Knowledge Base Extracted from Wikipedia. Semantic Web Journal 6, 2 (2015), 167–195. Outstanding Paper Award (Best 2014 SWJ Paper)

work page 2015
[19]

Yankai Lin, Zhiyuan Liu, Maosong Sun, Yang Liu, and Xuan Zhu. 2015. Learning entity and relation embeddings for knowledge graph completion. InTwenty-ninth AAAI conference on arti/f_icial intelligence

work page 2015
[20]

Yankai Lin, Shiqi Shen, Zhiyuan Liu, Huanbo Luan, and Maosong Sun. 2016. Neural relation extraction with selective a/t_tention over instances. InProceed- ings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , Vol. 1. 2124–2133

work page 2016
[21]

Weifeng Liu, Puskal P Pokharel, and Jos´e C Pr´ıncipe. 2007. Correntropy: Proper- ties and applications in non-Gaussian signal processing. IEEE Transactions on Signal Processing 55, 11 (2007), 5286–5298

work page 2007
[22]

George A Miller. 1995. WordNet: a lexical database for English. Commun. ACM 38, 11 (1995), 39–41

work page 1995
[23]

Mojtaba Nayyeri, Sahar Vahdati, Jens Lehmann, and Hamed Shariat Yazdi. 2019. So/f_t Marginal TransE for Scholarly Knowledge Graph Completion.arXiv preprint arXiv:1904.12211 (2019)

work page internal anchor Pith review Pith/arXiv arXiv 2019
[24]

Maximilian Nickel, Volker Tresp, and Hans-Peter Kriegel. 2012. Factorizing yago: scalable machine learning for linked data. In Proceedings of the 21st international conference on World Wide Web. ACM, 271–280

work page 2012
[25]

Kasey Pane/t_ta. 5. trends emerge in the gartner hype cycle for emerging tech- nologies, 2018. Retrieved November 4 (5), 2018

work page 2018
[26]

Wei Shen, Jianyong Wang, Ping Luo, and Min Wang. 2013. Linking named entities in tweets with knowledge base via user interest modeling. InProceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 68–76

work page 2013
[27]

Sean Szumlanski and Fernando Gomez. 2010. Automatically acquiring a seman- tic network of related concepts. In Proceedings of the 19th ACM international conference on Information and knowledge management . ACM, 19–28

work page 2010
[28]

/T_h´eo Trouillon, Johannes Welbl, Sebastian Riedel, ´Eric Gaussier, and Guillaume Bouchard. 2016. Complex embeddings for simple link prediction. In International Conference on Machine Learning . 2071–2080

work page 2016
[29]

Zhen Wang, Jianwen Zhang, Jianlin Feng, and Zheng Chen. 2014. Knowledge graph embedding by translating on hyperplanes. In Twenty-Eighth AAAI confer- ence on arti/f_icial intelligence

work page 2014
[30]

Fuzheng Zhang, Nicholas Jing Yuan, Defu Lian, Xing Xie, and Wei-Ying Ma

work page
[31]

In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining

Collaborative knowledge base embedding for recommender systems. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining . ACM, 353–362

work page
[32]

Xiaofei Zhou, Qiannan Zhu, Ping Liu, and Li Guo. 2017. Learning knowledge embeddings by combining limit-based scoring loss. In Proceedings of the 2017 6 ACM on Conference on Information and Knowledge Management . ACM, 1009– 1018. 7

work page 2017

[1] [1]

Sam Adams. 2019. Sur/f_ing the Hype Cycle to In/f_inity and Beyond.Research- Technology Management 62, 3 (2019), 45–51

work page 2019

[2] [2]

Anonymous. 2018. Relation Pa/t_tern Encoded Knowledge Graph Embedding by Translating in Complex Space. (2018). anonymous preprint under review

work page 2018

[3] [3]

Stefano A Bini. 2018. Arti/f_icial intelligence, machine learning, deep learning, and cognitive computing: what do these terms mean and how will they impact health care? /T_he Journal of arthroplasty33, 8 (2018), 2358–2361

work page 2018

[4] [4]

Kurt Bollacker, Colin Evans, Praveen Paritosh, Tim Sturge, and Jamie Taylor

work page

[5] [5]

In Proceedings of the 2008 ACM SIGMOD international conference on Management of data

Freebase: a collaboratively created graph database for structuring human knowledge. In Proceedings of the 2008 ACM SIGMOD international conference on Management of data. AcM, 1247–1250

work page 2008

[6] [6]

Antoine Bordes, Sumit Chopra, and Jason Weston. 2014. /Q_uestion answering with subgraph embeddings. arXiv preprint arXiv:1406.3676 (2014)

work page internal anchor Pith review Pith/arXiv arXiv 2014

[7] [7]

Antoine Bordes, Xavier Glorot, Jason Weston, and Yoshua Bengio. 2012. Joint learning of words and meaning representations for open-text semantic parsing. In Arti/f_icial Intelligence and Statistics. 127–135

work page 2012

[8] [8]

Antoine Bordes, Nicolas Usunier, Alberto Garcia-Duran, Jason Weston, and Oksana Yakhnenko. 2013. Translating embeddings for modeling multi-relational data. In Advances in neural information processing systems . 2787–2795

work page 2013

[9] [9]

Antoine Bordes, Jason Weston, Ronan Collobert, and Yoshua Bengio. 2011. Learn- ing structured embeddings of knowledge bases. In Twenty-Fi/f_th AAAI Conference on Arti/f_icial Intelligence

work page 2011

[10] [10]

Stephen Boyd and Lieven Vandenberghe. 2004. Convex optimization. Cambridge university press

work page 2004

[11] [11]

Andrew Carlson, Justin Be/t_teridge, Bryan Kisiel, Burr Se/t_tles, Estevam R Hr- uschka, and Tom M Mitchell. 2010. Toward an architecture for never-ending language learning. In Twenty-Fourth AAAI Conference on Arti/f_icial Intelligence

work page 2010

[12] [12]

Huimin Chen, Maosong Sun, Cunchao Tu, Yankai Lin, and Zhiyuan Liu. 2016. Neural sentiment classi/f_ication with user and product a/t_tention. InProceedings of the 2016 conference on empirical methods in natural language processing . 1650– 1659

work page 2016

[13] [13]

Shizhu He, Kang Liu, Yuanzhe Zhang, Liheng Xu, and Jun Zhao. 2014. /Q_uestion answering over linked data using /f_irst-order logic. InProceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) . 1092– 1103

work page 2014

[14] [14]

Konrad H¨oﬀner, Sebastian Walter, Edgard Marx, Ricardo Usbeck, Jens Lehmann, and Axel-Cyrille Ngonga Ngomo. 2016. Survey on Challenges of /Q_uestion Answering in the Semantic Web. Semantic Web Journal (2016)

work page 2016

[15] [15]

Rodolphe Jena/t_ton, Nicolas L Roux, Antoine Bordes, and Guillaume R Obozinski

work page

[16] [16]

InAdvances in Neural Information Processing Systems

A latent factor model for highly multi-relational data. InAdvances in Neural Information Processing Systems. 3167–3175

work page

[17] [17]

Guoliang Ji, Shizhu He, Liheng Xu, Kang Liu, and Jun Zhao. 2015. Knowledge graph embedding via dynamic mapping matrix. In Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers) , Vol. 1. 687–696

work page 2015

[18] [18]

Jens Lehmann, Robert Isele, Max Jakob, Anja Jentzsch, Dimitris Kontokostas, Pablo Mendes, Sebastian Hellmann, Mohamed Morsey, Patrick van Kleef, S¨oren Auer, and Chris Bizer. 2015. DBpedia - A Large-scale, Multilingual Knowledge Base Extracted from Wikipedia. Semantic Web Journal 6, 2 (2015), 167–195. Outstanding Paper Award (Best 2014 SWJ Paper)

work page 2015

[19] [19]

Yankai Lin, Zhiyuan Liu, Maosong Sun, Yang Liu, and Xuan Zhu. 2015. Learning entity and relation embeddings for knowledge graph completion. InTwenty-ninth AAAI conference on arti/f_icial intelligence

work page 2015

[20] [20]

Yankai Lin, Shiqi Shen, Zhiyuan Liu, Huanbo Luan, and Maosong Sun. 2016. Neural relation extraction with selective a/t_tention over instances. InProceed- ings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , Vol. 1. 2124–2133

work page 2016

[21] [21]

Weifeng Liu, Puskal P Pokharel, and Jos´e C Pr´ıncipe. 2007. Correntropy: Proper- ties and applications in non-Gaussian signal processing. IEEE Transactions on Signal Processing 55, 11 (2007), 5286–5298

work page 2007

[22] [22]

George A Miller. 1995. WordNet: a lexical database for English. Commun. ACM 38, 11 (1995), 39–41

work page 1995

[23] [23]

Mojtaba Nayyeri, Sahar Vahdati, Jens Lehmann, and Hamed Shariat Yazdi. 2019. So/f_t Marginal TransE for Scholarly Knowledge Graph Completion.arXiv preprint arXiv:1904.12211 (2019)

work page internal anchor Pith review Pith/arXiv arXiv 2019

[24] [24]

Maximilian Nickel, Volker Tresp, and Hans-Peter Kriegel. 2012. Factorizing yago: scalable machine learning for linked data. In Proceedings of the 21st international conference on World Wide Web. ACM, 271–280

work page 2012

[25] [25]

Kasey Pane/t_ta. 5. trends emerge in the gartner hype cycle for emerging tech- nologies, 2018. Retrieved November 4 (5), 2018

work page 2018

[26] [26]

Wei Shen, Jianyong Wang, Ping Luo, and Min Wang. 2013. Linking named entities in tweets with knowledge base via user interest modeling. InProceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 68–76

work page 2013

[27] [27]

Sean Szumlanski and Fernando Gomez. 2010. Automatically acquiring a seman- tic network of related concepts. In Proceedings of the 19th ACM international conference on Information and knowledge management . ACM, 19–28

work page 2010

[28] [28]

/T_h´eo Trouillon, Johannes Welbl, Sebastian Riedel, ´Eric Gaussier, and Guillaume Bouchard. 2016. Complex embeddings for simple link prediction. In International Conference on Machine Learning . 2071–2080

work page 2016

[29] [29]

Zhen Wang, Jianwen Zhang, Jianlin Feng, and Zheng Chen. 2014. Knowledge graph embedding by translating on hyperplanes. In Twenty-Eighth AAAI confer- ence on arti/f_icial intelligence

work page 2014

[30] [30]

Fuzheng Zhang, Nicholas Jing Yuan, Defu Lian, Xing Xie, and Wei-Ying Ma

work page

[31] [31]

In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining

Collaborative knowledge base embedding for recommender systems. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining . ACM, 353–362

work page

[32] [32]

Xiaofei Zhou, Qiannan Zhu, Ping Liu, and Li Guo. 2017. Learning knowledge embeddings by combining limit-based scoring loss. In Proceedings of the 2017 6 ACM on Conference on Information and Knowledge Management . ACM, 1009– 1018. 7

work page 2017