A Survey on Knowledge Graph Embeddings for Link Prediction

May combine two claims.

Rossi et al.'s survey selected few models and cannot cover all types of KGE models.

“However, there is no overall classification and summary of the KGE models proposed in recent years in this paper, and the selected models are few, which cannot cover all types of KGE models.”

• cites [28] Rossi, 2021, Knowledge graph embedding for link prediction: A comparative analysis, ACM Trans. Knowl. Discov. Data TKDD sourceDisputed

  Section 3 (p. 5) explicitly proposes a taxonomy of KG link-prediction models, dividing them into three main families and smaller groups. The Introduction (p. 2) also describes it as an educational taxonomy for knowledge graph embedding techniques, contrary to the claim that the paper offers no overall classification. The paper selects 16 representative models, but does not say they cover every type. (LLM, conf=0.89)

  Checked by a reasoning model — may be wrong.

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

TransW uses word embeddings to compose knowledge graph embeddings and can detect unknown facts.

“TransW [101] uses word embeddings to compose knowledge graph embeddings and learns a function mapping from the word embedding space to the knowledge embedding space.Entities and relations are represented in the form of linear combinations of word embeddings in this model, which can detect unknown facts”

• cites [101] Ma, L., Sun, P., Lin, Z., and Wang, H. (2019). Composing Knowledge Graph Embeddings via Word Embeddings. arXiv. sourceSupported

May combine two claims.

Traditional semantic-based models focus only on the information of the triple itself and do not fuse any additional information.

“The traditional semantic-based models focus only on the information of the triple itself and do not fuse any additional information, as DistMult [20] and Complex [21] do.”

• cites [20] Yang, B., Yih, W., He, X., Gao, J., and Deng, L. (2015, January 7–9). Embedding Entities and Relations for Learning and Inference in Knowledge Bases. Proceedings of the ICLR (Poster), San Diego, CA, USA. sourceSupported
• cites [21] Trouillon, T., Welbl, J., Riedel, S., Gaussier, É., and Bouchard, G. (2016). Complex Embeddings for Simple Link Prediction, ICML. sourceSupported

May combine two claims.

TransH introduces a hyperplane, and TransR uses a relation-specific space to handle different relations, excavating more semantic information.

“TransH introduces a hyperplane, and TransR uses a relation-specific space to handle different relations, excavating more semantic information.”

• cites [15] Wang, Z., Zhang, J., Feng, J., and Chen, Z. (2014). Knowledge Graph Embedding by Translating on Hyperplanes, AAAI Press. sourcePartly supported

  The TransH section (p. 3) confirms that the paper introduces relation-specific hyperplanes and projects entity embeddings onto them to represent different roles across relations. The cited paper does not address TransR or its use of a relation-specific space, so it supports only the TransH part of the sentence. (LLM, conf=0.90)

  Experimental — accuracy not yet measured
• cites [17] Lin, Y., Liu, Z., Sun, M., Liu, Y., and Zhu, X. (2015). Learning Entity and Relation Embeddings for Knowledge Graph Completion, AAAI Press. sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

LiteralE introduces literature information, giving priority to numerical literature.

“LiteralE [115] introduces the literature information, giving priority to numerical literature.”

• cites [115] Kristiadi, A., Khan, M.A., Lukovnikov, D., Lehmann, J., and Fischer, A. (2019, January 26–30). Incorporating Literals into Knowledge Graph Embeddings. Proceedings of the ISWC, Auckland, New Zealand. sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

Ji et al. divided research on KGs into four categories: knowledge representation learning, knowledge acquisition, temporal KGs and knowledge-aware applications.

“Ji et al. [31] divided the research on KGs into four categories: knowledge representation learning, knowledge acquisition, temporal KGs and knowledge-aware applications.”

• cites [31] Ji, S., Pan, S., Cambria, E., Marttinen, P., and Yu, P.S. (2020). A Survey on Knowledge Graphs: Representation, Acquisition and Applications. arXiv. sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

SACN improves the GCN model to obtain a weighted graph convolutional network by introducing weights for different types of relations.

“In the encoder part, by introducing weights for different types of relations, SACN improves the GCN model to obtain a weighted graph convolutional network (WGCN).”

• cites [26] Shang, C., Tang, Y., Huang, J., Bi, J., He, X., and Zhou, B. (2019). End-to-End Structure-Aware Convolutional Networks for Knowledge Base Completion, AAAI Press. sourceSupported

May combine two claims.

HypER implements nonlinear quadratic combinations of entity and relation embeddings rather than the linear combinations in ConvE, giving HypER richer expressive capacity and fewer parameters.

“HypER implements nonlinear (quadratic) combinations of entity and relation embeddings rather than the linear combinations (weighted sums) in ConvE, which gives HypER the advantages of a much richer expressive capacity and fewer parameters.”

• cites [24] Balazevic, I., Allen, C., and Hospedales, T.M. (2019, January 17–19). Hypernetwork Knowledge Graph Embeddings. Proceedings of the ICANN (Workshop), Munich, Germany. sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

Rossi et al. classified models into three categories: tensor decomposition models, geometric models and deep learning models.

“Rossi et al. [28] classified models into three categories: tensor decomposition models, geometric models and deep learning models.”

• cites [28] Rossi, 2021, Knowledge graph embedding for link prediction: A comparative analysis, ACM Trans. Knowl. Discov. Data TKDD sourceSupported

May combine two claims.

The second group of multiplicative models includes DistMult and Complex, which can outperform additive models by capturing more semantic information.

“The second group of multiplicative models includes DistMult [20] and Complex [21], which can outperform the additive models by capturing more semantic information [19].”

• cites [19] Liu, Z., Sun, M., Lin, Y., and Xie, R. (2016). Knowledge Representation Learning: A Review. J. Comp. Res. Develop., 247–261. sourceNot checked — no full text of the source
• cites [20] Yang, B., Yih, W., He, X., Gao, J., and Deng, L. (2015, January 7–9). Embedding Entities and Relations for Learning and Inference in Knowledge Bases. Proceedings of the ICLR (Poster), San Diego, CA, USA. sourcePartly supported

  The Results (p. 5) report that DistMult outperformed the additive DistAdd model on the evaluated datasets. The Introduction (p. 1) says bilinear embeddings capture the compositional semantics of relations, but the paper does not evaluate Complex or establish that both named models outperform additive models by capturing more semantic information. (LLM, conf=0.86)

  Experimental — accuracy not yet measured
• cites [21] Trouillon, T., Welbl, J., Riedel, S., Gaussier, É., and Bouchard, G. (2016). Complex Embeddings for Simple Link Prediction, ICML. sourceSupported

May combine two claims.

ProjE complements missing information in a knowledge graph by learning joint embeddings of entities and edges and modifying the loss function, improving the KGE model through trivial changes to the network architecture and eliminating complex feature engineering.

“ProjE [176] complements the missing information in a KG by learning the joint embedding of the entities and edges and making modifications to the loss function, thus improving the KGE model through trivial changes to the network architecture and eliminating complex feature engineering.”

• cites [176] Shi, B., and Weningr, T. (2017). ProjE: Embedding Projection for Knowledge Graph Completion, AAAI Press. sourceSupported

May combine two claims.

ConvE obtains local relationships in different dimensions between entities through a convolution layer and a fully connected layer, but ignores the global relationships between triple embeddings of the same dimension.

“ConvE obtains local relationships in different dimensions between entities through a convolution layer and a fully connected layer, which ignores the global relationships between triple embeddings of the same dimension.”

• cites [22] Dettmers, T., Minervini, P., Stenetorp, P., and Riedel, S. (2017). Convolutional 2D Knowledge Graph Embeddings, AAAI Press. sourcePartly supported

  The model description (p. 3) says ConvE uses convolutional and fully connected layers to model interactions between entity and relation embeddings. The discussion of 2D convolutions (p. 2) describes interactions between embeddings, but does not say ConvE ignores global relationships between triple embeddings of the same dimension. The Conclusion (p. 7) proposes increasing large-scale structure as future work, which does not establish that specific limitation. (LLM, conf=0.87)

  Experimental — accuracy not yet measured

May combine two claims.

SACN's Conv-TransE decoder removes the reshaping operation and lets convolutional filters operate directly on input entities and relations in the same dimension, so translation properties of TransE remain while the same prediction performance as ConvE is maintained.

“The decoder, called Conv-TransE, removes the reshaping operation from the input entity and relation embeddings and lets the convolutional filters operate directly on input entities and relations in the same dimension.Thus, the translation properties of TransE remain, while the same prediction performance as ConvE is maintained.”

• cites [26] Shang, C., Tang, Y., Huang, J., Bi, J., He, X., and Zhou, B. (2019). End-to-End Structure-Aware Convolutional Networks for Knowledge Base Completion, AAAI Press. sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

SACN is an end-to-end structure-aware convolutional network that takes into account node connectivity, node attributes and relation types simultaneously.

“Weighted GCN, known as SACN [26], is an end-to-end structure-aware convolutional network that takes into account node connectivity, node attributes and relation types simultaneously.”

• cites [26] Shang, C., Tang, Y., Huang, J., Bi, J., He, X., and Zhou, B. (2019). End-to-End Structure-Aware Convolutional Networks for Knowledge Base Completion, AAAI Press. sourcePartly supported

  The Introduction and Conclusion describe SACN as an end-to-end network whose weighted GCN encoder uses graph connectivity, node attributes, and relation types. But the citing sentence conflates the encoder with the full model: WGCN is a component of SACN, which also has a Conv-TransE decoder. (LLM, conf=0.88)

  Experimental — accuracy not yet measured

May combine two claims.

ConvKB uses 1D convolution to maintain the translation characteristics of TransE, which is sufficient for capturing global relationships and transitional characteristics between entities.

“Removing the reshaping operation from ConvE, ConvKB [23] uses 1D convolution to maintain the translation characteristics of TransE, which is sufficient for capturing the global relationships and transitional characteristics between entities.”

• cites [23] Nguyen, D.Q., Nguyen, T.D., Nguyen, D.Q., and Phung, D.Q. (2018, January 1–6). A Novel Embedding Model for Knowledge Base Completion Based on Convolutional Neural Network. Proceedings of the NAACL-HLT, New Orleans, LA, USA. sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

ConvE introduces a 1-N scoring program that takes a head-entity-relation pair and matches all tail entities simultaneously, saving much evaluation time.

“Moreover, ConvE introduces a 1-N scoring program that takes a head-entity-relation pair and matches all the tail entities simultaneously, which is different from other models that use the 1-1 scoring program; thus, much evaluation time is saved.”

• cites [22] Dettmers, T., Minervini, P., Stenetorp, P., and Riedel, S. (2017). Convolutional 2D Knowledge Graph Embeddings, AAAI Press. sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

For link-prediction tasks, R-GCN can be regarded as an autoencoder consisting of an encoder and a decoder.

“For link-prediction tasks, a relational graph convolutional network (R-GCN) can be regarded as an autoencoder consisting of an encoder and a decoder.”

• cites [158] Schlichtkrull, M.S., Kipf, T.N., Bloem, P., Berg, R.v.d., Titov, I., and Welling, M. (2018, January 3–7). Modeling Relational Data with Graph Convolutional Networks. Proceedings of the ESWC, Crete, Greece. sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Rossi et al.'s experiment was not reproduced in a unified environment configuration but used different coding frameworks, such as Python and C++.

“In addition, the experiment of this survey was not reproduced in a unified environment configuration but used different coding frameworks, such as Python and C++.”

• cites [28] Rossi, 2021, Knowledge graph embedding for link prediction: A comparative analysis, ACM Trans. Knowl. Discov. Data TKDD sourcePartly supported

  Section 5.1 (pp. 18–19) says Rossi et al. trained and evaluated the models from scratch using implementations from several repositories, which supports the use of different implementations. But it says all experiments ran on the same server environment; it does not establish that the environment configuration was not unified or that the implementations used Python and C++. (LLM, conf=0.83)

  Experimental — accuracy not yet measured

May combine two claims.

SimplEx has low performance for FB15k-237 and high performance for FB15k because FB15k contains inverse relations and SimplEx can model inverse relations appropriately.

“SimplEx [152] has low performance for FB15k-237 and high performance for FB15k because FB15k contains inverse relations and SimplEx can model the inverse relations appropriately.”

• cites [152] Kazemi, S.M., and Poole, D. (2018). SimplE Embedding for Link Prediction in Knowledge Graphs. NeurIPS. sourcePartly supported

  Section 6.1 (p. 8) reports strong SimplE performance on FB15k, and Section 4 (p. 3) says the model uses inverse relations. The paper reports experiments on WN18 and FB15k, not FB15k-237 (Section 6, p. 7); it does not establish low performance on FB15k-237 or attribute its FB15k results to inverse relations in that dataset. (LLM, conf=0.88)

  Experimental — accuracy not yet measured

• AssumptionRests on an assumption: FB15k and FB15k-237 are standard datasets commonly used in this field and are suitable for comprehensively evaluating model performance.

May combine two claims.

SACN also performs well among the implemented models, using relation types and entity attributes in the GCN model structure.

“SACN [26] also performs well among these models; it uses the relation types and entity attributes in the GCN model structure.”

• cites [26] Shang, C., Tang, Y., Huang, J., Bi, J., He, X., and Zhou, B. (2019). End-to-End Structure-Aware Convolutional Networks for Knowledge Base Completion, AAAI Press. sourceSupported

• AssumptionRests on an assumption: FB15k and FB15k-237 are standard datasets commonly used in this field and are suitable for comprehensively evaluating model performance.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

The bilinear+TR model has the shortest training time, with a type regularizer incorporated into the loss function.

“The bilinear+TR model has the shortest training time, with a type regularizer incorporated into the loss function, which fully considers the type information of entities.”

• AssumptionRests on an assumption: Training time results on FB15k are representative for general conclusions about model time efficiency.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Semantic matching models such as DistMult and ComplEx all suffer from longer training times.

“Semantic matching models such as DistMult and ComplEx all suffer from longer training times.”

• AssumptionRests on an assumption: Training time results on FB15k are representative for general conclusions about model time efficiency.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Translational distance models such as HAKE and RotatE all have shorter training times because the translational-distance model has a relatively simple model structure and scoring function without too many parameters.

“Translational distance models such as HAKE and RotatE all have shorter training times because the translational-distance model has a relatively simple model structure and scoring function without too many parameters.”

• AssumptionRests on an assumption: Training time results on FB15k are representative for general conclusions about model time efficiency.

May combine two claims.

LiteralE introduces overhead in the number of parameters compared to the base method, leading to a long training time.

“LiteralE introduces some overhead in terms of the number of parameters compared to the base method, leading to a long training time.”

• AssumptionRests on an assumption: Training time results on FB15k are representative for general conclusions about model time efficiency.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The performance of conventional models such as translation models (TransE) and semantic models (DistMult and ComplEx) is not good.

“The performance of conventional models such as translation models (TransE) and semantic models (DistMult and ComplEx) is not good.”

• AssumptionRests on an assumption: FB15k and FB15k-237 are standard datasets commonly used in this field and are suitable for comprehensively evaluating model performance.

May combine two claims.

HypER has many fewer parameters than ConvE, so it saves much training time.

“While HypER utilizes a 1D relation-specific filter and a nonlinear (quadratic) combination of entity and relation embeddings via hypernetworks to perform weight sharing, it has many fewer parameters than ConvE, so it saves much training time.”

• AssumptionRests on an assumption: Training time results on FB15k are representative for general conclusions about model time efficiency.

May combine two claims.

For FB15k, only ComplEx obtains good results upon adding literal information, while for FB15-237, only DistMult improves slightly after adding literal information.

“For FB15k, only ComplEx obtains good results upon adding literal information, while for FB15-237, only DistMult improves slightly after adding literal information.”

• AssumptionRests on an assumption: FB15k and FB15k-237 are standard datasets commonly used in this field and are suitable for comprehensively evaluating model performance.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Semantic-matching-based models have been continuously mining deeper semantic information, including path information, entity types and contextual information, to obtain more accurate representations for KGE.

“Since the traditional model was put forward, semantic-matching-based models have been continuously mining deeper semantic information, including path information, entity types, contextual information and so on, to obtain more accurate representations for KGE.”

Probably not a substantial claim: may not pass S3 (load-bearing).

HAKE considers the semantic hierarchy of entities by using polar coordinates, in which concentric circles can naturally reflect the hierarchy.

“HAKE [60] considers the semantic hierarchy of entities by using polar coordinates, in which concentric circles can naturally reflect the hierarchy.”

• cites [60] Zhang, Z., Cai, J., Zhang, Y., and Wang, J. (2020, January 7–12). Learning Hierarchy-Aware Knowledge Graph Embeddings for Link Prediction. Proceedings of the AAAI 2020, New York, NY, USA.Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

Recently proposed semantic models fuse various additional information to obtain better performance to mine deeper semantic information at the bottoms of graphs.

“The former recently proposed models fuse various additional information to obtain better performance to mine deeper semantic information at the bottoms of graphs.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Translation-distance-based models are additive models that usually use distance-based functions to define the scoring function for link prediction.

“These models are additive models, which usually use distance-based functions to define the scoring function for link prediction.”

May combine two claims.

RW-LMLM consists of a random walk algorithm for KGs and a language model-based link-prediction model, and its random walk can capture semantic and syntactic information in KGs by considering entities, relations and order information of paths.

“RW-LMLM is a novel approach for link prediction that consists of a random walk algorithm for KGs (RW) and a language model-based link-prediction model (LMLM).”

• Also stated in: § 3.2. Semantic Information-Based Models · plain

  “RW can capture the semantic and syntactic information in KGs by considering the entities, relations and order information of the paths, in contrast to DeepWalk (which considers only the entities and relations).”

May combine two claims.

Gaussian embedding models take the uncertainties of entities and relations into account but result in a complex model.

“Gaussian embedding models take the uncertainties of the entities and relations into account, but this results in a complex model.”

May combine two claims.

Real-world data are often dynamic and evolving, which makes constructing correct and complete knowledge graphs difficult.

“Second, real-world data are often dynamic and evolving, which leads to difficulty in constructing correct and complete KGs [7].”

• cites [7] Cai, 2017, A Comprehensive Survey of Graph Embedding: Problems, Techniques, and Applications, IEEE Trans. Knowl. Data Eng. sourceNot checked — no full text of the source

May combine two claims.

TransMS considers the semantics between the head entity and the relation as well as between the relation and the tail entity, which previous models do not do.

“In this way, it considers the semantics between the head entity and the relation as well as the semantics between the relation and the tail entity, which is not done in previous models.”

• cites [52] Yang, S., Tian, J., Zhang, H., Yan, J., He, H., and Jin, Y. (2019, January 10–16). TransMS: Knowledge Graph Embedding for Complex Relations by Multidirectional Semantics. Proceedings of the IJCAI, Macao, China. sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Semantic information-based models usually use similarity-based functions to define scoring functions for traditional semantic-matching models or introduce additional information to mine more knowledge for recently proposed models.

“Semantic information-based models usually use similarity-based functions to define scoring functions for traditional semantic-matching models or introduce additional information to mine more knowledge for recently proposed models.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

TransE is one of the earliest models for link prediction and uses the relation to translate the head entity to a tail entity.

“TransE [13] is one of the earliest models for link prediction, which uses the relation for translating the head entity to a tail entity.”

• cites [13] Bordes, A., Usunier, N., Garcia-Duran, A., Weston, J., and Yakhnenko, O. (2013, January 5–8). Translating Embeddings for Modeling Multi-Relational Data. Proceedings of the NIPS, Lake Tahoe, NV, USA.Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Traditional semantic-based models suffer from higher computational complexity.

“These models suffer from higher computational complexity.”

May combine two claims.

TransC combines structural information with entity concepts to improve KGE models and introduces a novel type of semantic similarity to measure the distinctness of entity semantics using concept information.

“The TransC [149] model combines structural information with entity concepts (referring to the categories of entities) to improve KGE models and introduces a novel type of semantic similarity to measure the distinctness of entity semantics by using the concept information.”

• cites [149] Wang, Y., Liu, Y., Zhang, H., and Xie, H. (2019). Leveraging Lexical Semantic Information for Learning Concept-Based Multiple Embedding Representations for Knowledge Graph Completion. Asia-Pacific Web (APWeb) and Web-Age Information Management (WAIM) Joint International Conference on Web and Big Data, Springer. sourceNot checked — no full text of the source

May combine two claims.

Translation-based models have insufficient expressiveness due to the direct addition of h + r and the lack of different semantic information.

“Although the embedding operation of this type of model is relatively simple, the expressiveness is not sufficient due to the direct addition of h + r and the lack of different semantic information.”

May combine two claims.

TransE has difficulty dealing with multirelational graphs and is limited by its simple translation operation and lack of a discrimination policy for all kinds of relations.

“TransE is the earliest translation-based embedding model, and it has difficulty dealing with multirelational graphs; it is limited by its simple translation operation as well as its lack of a discrimination policy for all kinds of relations.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The three categories of KGE models reflect the mainstream models of KGE development in three main stages.

“These three categories reflect the mainstream models of KGE development in three main stages.”

• AssumptionRests on an assumption: The three categories (translation-distance-based, semantic-matching-based and neural-network-based) exhaustively cover mainstream KGE models for link prediction.

May combine two claims.

Neural-network-based models consider the type of entity or relation, temporal information, path information and substructure information.

“These models consider the type of entity or relation, temporal information, path information and substructure information.”

Probably not a substantial claim: may not pass S2 (checkable).

FB15k and FB15k-237 could be used to evaluate model performance more comprehensively.

“Therefore, these datasets could be used to evaluate the performance of the model more comprehensively.”

• AssumptionRests on an assumption: FB15k and FB15k-237 are standard datasets commonly used in this field and are suitable for comprehensively evaluating model performance.

Probably not a substantial claim: may not pass S3 (load-bearing).

Additional information used by KGE models includes path information, order information, concepts, entity attributes and entity types.

“The additional information includes path information, order information, concepts, entity attributes, entity types and so on.”

• Also stated in: § 3.3. Neural Network-Based Models · plain

  “The additional information also includes path information, order information, concepts and so on.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The implementation used the code provided in the original paper with the same settings to obtain the best results.

“We also used the code provided in the original paper with the same settings to obtain the best results.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

There are few studies on dynamic graphs, which are better able to reflect the real world.

“There are few studies on dynamic graphs, which are better able to reflect the real world.”

May combine two claims.

Dai et al. considered only two aspects and did not provide an overall table.

“Regarding additional information, they considered only these two aspects and also did not provide an overall table.”

• cites [19] Liu, Z., Sun, M., Lin, Y., and Xie, R. (2016). Knowledge Representation Learning: A Review. J. Comp. Res. Develop., 247–261. sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

The GCN can be scaled linearly on the edge of the graph to learn hidden layer representations for encoding local graphical structures and features of nodes.

“It can be scaled linearly on the edge of the graph to learn hidden layer representations for encoding the local graphical structures and features of nodes.”

• cites Kipf et al.Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S4 (based).

KGE models embed entities and relations into a low-dimensional vector space while preserving the structure of the KG and its underlying semantic information.

“KGE models embed entities and relations into a low-dimensional vector space while preserving the structure of the KG and its underlying semantic information.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Dai et al. described models with two categories: triplet fact-based representation learning models and description-based representation learning models.

“Dai et al. [19] described these models with two categories: triplet fact-based representation learning models and description-based representation learning models.”

• cites [19] Liu, Z., Sun, M., Lin, Y., and Xie, R. (2016). Knowledge Representation Learning: A Review. J. Comp. Res. Develop., 247–261. sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

ProjR can model N-1, N-N, 1-N-1 and one-relation circle structures with the advantage of a combination operator design.

“It can model N-1, N-N, 1-N-1 and one-relation circle structures with the advantage of a combination operator design.”

• cites [178] Zhang, W., Li, J., and Chen, H. (2018, January 26–30). ProjR: Embedding Structure Diversity for Knowledge Graph Completion. Proceedings of the NLPCC, Hohhot, China. sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

Traditional distance-based and semantic-matching-based models cannot meet the requirements of KGE.

“Traditional distance-based and semantic-matching-based models cannot meet the requirements of KGE.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The use of convolutional neural networks or attention mechanisms helps to generate better embeddings.

“The use of convolutional neural networks or attention mechanisms also helps to generate better embeddings.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Link prediction is one of the most common tasks for evaluating the performance of KGE, and it has no formal definition.

“Link prediction is one of the most common tasks for evaluating the performance of KGE, which has no formal definition.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

It is difficult to model the heterogeneity of graphs.

“It is difficult to model the heterogeneity of graphs.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based). May combine two claims.

Neural networks have the abilities of self-study, associative storage and high-speed optimization.

“Neural networks have been a promising solution in many fields and have the abilities of self-study, associative storage and high-speed optimization.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The paper's experiments on representative models were conducted in a unified environment, including the server type and programming language (pytorch).

“In addition, our experiments on some representative models were conducted in a unified environment, including the server type and programming language (pytorch).”

• AssumptionRests on an assumption: Conducting experiments in a unified environment, including server type and programming language (pytorch), ensures fair comparison across models.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

A neural-network model was introduced into KGE to propagate neighborhood information.

“Recently, to obtain better and more effective entity and relation embeddings, a neural-network model was also introduced into KGE to propagate neighborhood information.”

Probably not a substantial claim: may not pass S3 (load-bearing).

All experiments targeting the link-prediction task were conducted in a unified environment with a T640 Dell server, Ubuntu 16.04, and an NVIDIA-SMI 418.67 GPU.

“We conducted all the experiments targeting the link-prediction task in a unified environment with a T640 Dell server, the Ubuntu 16.04 system, and an NVIDIA-SMI 418.67 GPU.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

There are many models for KGE, but there are still many challenges in obtaining better entity and relation embeddings for link prediction.

“As described above, there are many models for KGE, but there are still many challenges in obtaining better entity and relation embeddings for link prediction.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Kipf et al. proposed a simple but effective hierarchical-propagation rule running directly on a graph with a neural-network model, a GCN.

“Kipf et al. proposed a simple but effective hierarchical-propagation rule running directly on a graph with a neural-network model, a GCN.”

• cites Kipf et al.Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

The graph convolution neural network has a strong ability to mine the underlying semantics of knowledge graphs.

“Especially, the graph convolution neural network has a strong ability to mine the underlying semantics of knowledge graphs.”

• AssumptionRests on an assumption: FB15k and FB15k-237 are standard datasets commonly used in this field and are suitable for comprehensively evaluating model performance.

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

If the node information of a multi-hop domain can be aggregated, the accuracy of the model in specific tasks can be greatly improved.

“In addition, if the node information of a multi-hop domain can be aggregated, the accuracy of the model in specific tasks can be greatly improved.”

• AssumptionRests on an assumption: FB15k and FB15k-237 are standard datasets commonly used in this field and are suitable for comprehensively evaluating model performance.

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

The selected models are representative for gaining deep insight into existing models, and they are conventional, recently proposed or highly cited.

“Therefore, these selected models are representative for gaining deep insight into the existing models, and they are conventional, recently proposed or highly cited.”

May combine two claims.

The rotation operation used in translational-distance models and the hypernetwork used in CNN-based models play important roles in improving performance.

“which indicates that the rotation operation used in translational-distance models and the hypernetwork used in CNN-based models play important roles in improving performance.”

• AssumptionRests on an assumption: FB15k and FB15k-237 are standard datasets commonly used in this field and are suitable for comprehensively evaluating model performance.

May combine two claims.

RotatE uses complex space and mines different types of relations, so different aspects of semantic information are modeled well by this integration strategy.

“RotatE uses the complex space and mines different types of relations (symmetry/antisymmetry, inversion and composition); thus, different aspects of semantic information are modeled well by this integration strategy.”

• AssumptionRests on an assumption: FB15k and FB15k-237 are standard datasets commonly used in this field and are suitable for comprehensively evaluating model performance.

May combine two claims.

DistMult's scoring function can only capture pairwise interactions of the same dimension between entities, so this simple embedding can only deal with symmetric relations.

“Its scoring function can only capture pairwise interactions of the same dimension between entities.Therefore, this simple embedding can only deal with symmetric relations.”

May combine two claims.

Compared with traditional models, the latest models have achieved better results, benefiting from their unique model structures and sampling technologies or from adding important additional information.

“Compared with traditional models, the latest models have their own advantages.Generally, they have achieved better results, benefiting from their own unique model structures and sampling technologies or adding important additional information.”

• AssumptionRests on an assumption: FB15k and FB15k-237 are standard datasets commonly used in this field and are suitable for comprehensively evaluating model performance.

May combine two claims.

Because ComplEx introduces a complex vector space and can deal with asymmetric relations, it has a good response to literal information for FB15k but not FB15k-237.

“Because ComplEx introduces a complex vector space and can deal with asymmetric relations, it has a good response to literal information for FB15k but not FB15k-237.”

• AssumptionRests on an assumption: FB15k and FB15k-237 are standard datasets commonly used in this field and are suitable for comprehensively evaluating model performance.

Probably not a substantial claim: may not pass S4 (based).

Models with additional information, such as node attributes, node types, relationship types, and prior knowledge, have better performance.

“Models with additional information, such as node attributes, node types, relationship types, prior knowledge and so on, have better performance.”

• AssumptionRests on an assumption: FB15k and FB15k-237 are standard datasets commonly used in this field and are suitable for comprehensively evaluating model performance.

May combine two claims.

RW-LMLM takes into account both order information and the random walk algorithm, and it has the capability of dealing with underlying semantic information.

“RW-LMLM [91] takes into account both the order information and random walk algorithm, and it has the capability of dealing with underlying semantic information.”

• cites [91] Wang, C., Yan, M., Yi, C., and Sha, Y. (2019, January 26–30). Capturing Semantic and Syntactic Information for Link Prediction in Knowledge Graphs. Proceedings of the ISWC, Auckland, New Zealand. sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

For HAKE, polar coordinates may have great benefits because their particular structure enables them to mine considerable hidden semantic structure information.

“For HAKE [60], we believe that the polar coordinates may have great benefits because of their particular structure, which enables them to mine considerable hidden semantic structure information.”

• cites [60] Zhang, Z., Cai, J., Zhang, Y., and Wang, J. (2020, January 7–12). Learning Hierarchy-Aware Knowledge Graph Embeddings for Link Prediction. Proceedings of the AAAI 2020, New York, NY, USA.Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• AssumptionRests on an assumption: FB15k and FB15k-237 are standard datasets commonly used in this field and are suitable for comprehensively evaluating model performance.