Pith. sign in

REVIEW 4 major objections 5 minor 27 references

Multi-modal Knowledge Graph Generation with Semantics-enriched Prompts

T0 review · 4 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read Selecting an entity's neighbors by how visualizable their relation is and how structurally similar the neighbor is to the entity produces images that are more relevant to both the entity and its knowledge graph than images from…

desk verdict A useful pipeline for generating images from KG neighbors, but the headline 'higher-quality' claim rests on a mis-applied FID and is contradicted by the paper's own human ratings; the relevance and downstream gains are the real contribution. read the letter →

arxiv 2504.13631 v1 pith:EPETYXBF submitted 2025-04-18 cs.AI

Yajing Xu , Zhiqiang Liu , Jiaoyan Chen , Mingchen Tu , Zhuo Chen , Jeff Z. Pan , Yichi Zhang , Yushan Zhu
show 2 more authors
Wen Zhang Huajun Chen
This is my paper · ORCID
classification cs.AI
keywords multi-modalknowledgegraphconstructionneighborselectionpromptengineeringtext-to-imagegenerationimagerewardmodelcompletionembeddings
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

This paper argues that a knowledge graph can be turned into a multi-modal one automatically by generating an image for each entity from a prompt, but the quality of that image hinges on which neighboring triples are allowed into the prompt. It proposes a two-stage neighbor filter: first discard relations whose textual descriptions do not score well on an image-reward model, then keep only one-hop neighbors whose graph embeddings are most similar to the entity's embedding. A language model turns the surviving triples into a short descriptive prompt, and a diffusion model renders the image. On two multi-modal knowledge graph datasets the method beats generating from the entity name alone and from the longest-token neighbor baseline on FID, CLIPscore, and human ratings of entity and graph relevance, and the generated images support link prediction roughly as well as real images.

What carries the argument

The load-bearing object is the two-stage selection score inside VSNS. Stage one, VNS, defines a relation-level visualizability score $r_{\mathrm{vis}}$ as the fraction of sampled triples of that relation whose text-to-image pair is scored above zero by ImageReward, a model trained on human preferences that scores how well an image matches a text prompt, and keeps only relations with $r_{\mathrm{vis}}$ above a threshold $\mu$. Stage two, SNS, embeds the graph with CompGCN, a graph convolutional network that learns entity and relation embeddings, composes a neighbor embedding $e_{(r,t)}=\phi_{\mathrm{CompGCN}}(e_r,e_t)$, and keeps neighbor $t$ of head $h$ when the cosine similarity $\mathrm{sim}(e_h,e_{(r,t)})$ is at least the average similarity over all one-hop neighbors of $h$ through that relation. This score is what turns raw graph structure into a prompt-worthy subset, and it is the component whose removal in the ablation degrades FID and CLIPscore.

What would settle it

Take a set of entities with several one-hop neighbors and compare images generated using SNS-selected neighbors against images generated using the same number of neighbors chosen at random from the same relations; if FID, CLIPscore, and human entity-relevance ratings do not favor the SNS selection, the structural similarity criterion is not doing the work. A second check: find a relation where the highest-similarity neighbor is known to be visually irrelevant (for example, an abstract relation such as 'influences') and show that forcing that neighbor into the prompt lowers human entity-relevance scores.

Watch

Extended reading notes

Core claim

The paper's central claim is that neighbor selection, not prompt length or entity-name-only prompting, is what determines whether a generated image is faithful to an entity and its knowledge graph context. The proposed VSNS method first computes a visualizability score per relation by sampling triples, generating an image from each with a diffusion model, and using an image-reward model to decide whether that relation's semantics can be rendered; relations below a threshold are dropped. For surviving relations, CompGCN embeddings of the head entity and the composed neighbor representation are compared by cosine similarity, and only neighbors at or above the average similarity for that relation are kept. The selected triples are fed to a language model with a fixed instruction to produce a one-sentence semantics-enriched prompt, and Stable Diffusion generates the final image. Quantitative results on two datasets show lower FID and higher CLIPscore than both baselines, human evaluators rank the VSNS images above both baselines on entity relevance and knowledge-graph relevance, and on the DB15K link-prediction test the generated images yield higher MRR and Hits@1 than real images when added to structural features.

Load-bearing premise

The assumption that carries the method is that a neighbor's structural similarity to an entity, measured by CompGCN embedding cosine similarity, tracks how much that neighbor will help a text-to-image model draw the entity correctly; if structural similarity and visual usefulness diverge, the selection step can discard informative neighbors and keep visually misleading ones.

Editorial extensions

If this is right

  • Knowledge graphs with no existing images can be enriched automatically: entity name, graph neighbors, a language model, and a diffusion model are all that is needed, so the pipeline scales to the size of the graph.
  • Because the generated images improve FID and CLIPscore over name-only and longest-token prompts, the neighbor filter is the effective ingredient, not the prompt length.
  • Generated images can stand in for real images in multi-modal knowledge graph completion: on the DB15K link-prediction experiment, structural features plus generated images beat structural features plus real images on MRR and Hits@1.
  • Filtering hard-to-visualize relations first (VNS) mainly helps entities with few relation types, while pruning neighbors by structure (SNS) mainly helps entities with many neighbors, so the two modules are complementary.
  • The framework targets attribute-style multi-modal knowledge graphs, meaning each entity receives an image attribute; entity-style multimodal graphs would need different construction logic.

Reading between the lines

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

  • A direct test the paper does not run: compare SNS against random neighbors sampled from the same relations at the same count. If random selection matches SNS on FID and human relevance, the structural ranking is not the cause of the gain; this would also isolate how much of the improvement comes from VNS's relation filtering alone.
  • The relation-level visualizability threshold is global ($\mu=0.5$); per-domain thresholds might preserve relations that are easy to visualize for some entity types but not others.
  • The same selection pipeline could be reused for other modalities, such as selecting textual or audio neighbors for entity video generation, wherever a structural-similarity score can be computed.
  • A boundary the authors name is abstract entities: applying the pipeline to emotions or events likely requires additional cues, such as temporal or sentiment information, before the structural-similarity criterion becomes meaningful.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes an automated pipeline for constructing multi-modal knowledge graphs from textual KGs by generating an image for each entity. The core contribution is VSNS, a two-stage neighbor-selection method: VNS filters relations whose triples receive positive ImageReward scores, and SNS ranks the remaining one-hop neighbors by cosine similarity of CompGCN embeddings, keeping neighbors above the mean. The selected neighbors are fed to an LLM to produce semantics-enriched prompts, which are then used by Stable Diffusion to generate entity images. The method is evaluated on MKG-Y and DB15K against two baselines: images from entity names alone (Is) and images from the longest-token neighbor (Im), using FID, CLIPscore, human ratings (IQ, CIE, CIKG), and downstream multi-modal knowledge graph completion with NATIVE.

Significance. If the claims were fully supported, the framework would provide a scalable way to generate contextually relevant images for KG entities without manual prompt engineering, and the downstream MMKGC results suggest that synthetic images can be useful for link prediction. The paper has strengths: it builds on external scoring models (ImageReward, CompGCN, CLIP) rather than tuning on the evaluation metrics; the human-evaluation protocol is described in detail; and the downstream reasoning task gives an application-oriented check. However, the central quality claim is not currently supported: the automatic metric labeled FID is not the standard Fréchet distance, and the human IQ results on MKG-Y contradict the 'higher-quality' part of the abstract. The relevance and KG-consistency improvements are more credible but would benefit from statistical testing. Methodologically, the SNS selection rule's core assumption that structural similarity implies visual prompt usefulness is not tested in isolation.

major comments (4)
  1. [V-C, Automatic Evaluation; Tables II, IV, V] The reported 'FID' is not the Fréchet Inception Distance defined by Heusel et al. [23]. Section V-C states: 'For FID score calculation, we compare the generated image with each of the three real images separately and select the smallest FID value.' With one generated and one real image, covariance matrices cannot be estimated, so this quantity reduces to a per-image feature distance, not a distribution-level divergence. Taking the minimum over three references rewards matching any one real image and makes the numbers systematically lower. Since the abstract's 'higher-quality images' claim is supported mainly by the FID improvements in Tables II, IV, and V, this claim is not quantitatively established. Please either recompute a standard FID over the full generated and real image sets or relabel the metric and explicitly withdraw the distribution-level quality inference.
  2. [Table III, Human Evaluation] The human evaluation results do not support the abstract's blanket claim of 'higher-quality images that are more relevant' to the KG. On MKG-Y, IQ(Isvns)=2.68 is lower than IQ(Is)=2.72; on DB15K, CIE(Isvns)=2.27 is lower than CIE(Is)=2.31. The human data support only selective improvements, such as CIKG on both datasets and IQ on DB15K. The explanation of lower IQ for neighbor-based images (incomplete body parts, repetitive generation) is descriptive and does not address the comparison against Is. Please either restrict the quality claim to the settings where it is actually observed or provide evidence that the small observed differences are not within annotation noise.
  3. [IV-A2, Eqs. (3)-(5)] The SNS module assumes that the cosine similarity between CompGCN entity embeddings and composed neighbor embeddings (Eq. 4) predicts how useful a neighbor is for visual prompt generation. This assumption is plausible but is not tested in isolation; Table V evaluates the full VSNS pipeline, so the observed improvement could stem from VNS filtering, the LLM prompt template, or the prompt length rather than from the structural ranking. Please provide a controlled comparison that holds relations and prompt template fixed and compares SNS-selected neighbors against random or longest-token neighbors of equal size, and report whether Snei correlates with human CIE/CIKG scores.
  4. [V-C, Human Evaluation] The human evaluation uses small samples (50 entities on MKG-Y, 100 on DB15K) with three annotators, but no inter-annotator agreement, variance, or significance testing is reported. Several margins are tiny (e.g., MKG-Y CIE 2.27 vs 2.06; DB15K IQ 2.26 vs 2.06). Without confidence intervals or a paired test, the claimed advantages may be within annotation noise. Please report per-criterion score distributions, agreement statistics, and appropriate significance tests.
minor comments (5)
  1. [V-A, V-D] The acronym is inconsistent: the paper defines VSNS but Section V-A and V-D repeatedly use 'SVNS'; please standardize the terminology.
  2. [Table II] The column header '#C Number' is unclear; define 'C' (presumably the number of compared entities) and clarify why the two baseline comparisons use different entity subsets, which prevents direct cross-row comparison of FID/CLIPscore values.
  3. [V-C, Implementation Details] The VNS threshold µ=0.5 and the number of sampled triples per relation (ten) are free parameters; please include a sensitivity analysis or justify these choices.
  4. [V-C, Automatic Evaluation] The automatic metrics are computed only on MKG-Y because many DB15K image URLs are unavailable; this limitation should be stated explicitly in the main experimental section rather than only in the implementation details.
  5. [Table VI] The row label 'S+Isvns NATIVE' is redundant and confusing; also, the MMKGC gains over S+I are small (e.g., MRR 0.383 to 0.387 on MKG-Y) and significance is not reported, so the claim that generated images are 'comparable or even superior' to real images should be phrased more cautiously.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the neighbor-selection method, prompt generation, and evaluation rest on external components and independent human/automatic metrics.

full rationale

The derivation chain is self-contained against external benchmarks. The VNS module scores relation visualizability with ImageReward (Eqs. 1-2), the SNS module selects neighbors using CompGCN embeddings and cosine similarity (Eqs. 3-5), and prompts are generated by ChatGPT; none of these inputs are defined in terms of the reported evaluation metrics (FID, CLIPscore, human ratings). The evaluation metrics are external comparisons, and the human evaluation is conducted by independent annotators. The only notable self-citation is NATIVE [26], which shares authors with this paper and is used as a fixed downstream model for knowledge graph completion; it is not used as evidence for the neighbor-selection claim, so it is not load-bearing. The FID computation described in Section V-C ('compare the generated image with each of the three real images separately and select the smallest FID value') deviates from the standard distribution-level FID, but that is a correctness/validity concern, not circularity: the metric is still an external comparison and is not used to construct the method. No equation defines the selected neighbors in terms of the reported outcomes, and no fitted parameter is renamed as a prediction. Therefore no circular step is identified.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The method rests on several domain assumptions: graph embeddings predict visual relevance, ImageReward scores generalize to a relation-level threshold, LLM prompts are faithful, and synthetic images are valid substitutes for real images in downstream tasks. There are two hand-set implementation constants (µ=0.5, 10 triples). No new theoretical entities are introduced.

free parameters (2)
  • VNS threshold µ = 0.5
    Relation is kept if its mean ImageReward score over 10 sampled triples exceeds 0.5. No sensitivity analysis is reported; the value is stated only in Section V-C.
  • Triples sampled per relation for VNS = 10
    Ten triples per relation are sampled to estimate r_vis (Section V-C-a). The sample size is arbitrary; no variance or confidence interval is reported.
assumptions (4)
  • domain assumption CompGCN embeddings capture structural information that predicts visual relevance of neighbors
    Equations 3-5 assume cosine similarity between entity embedding and composed neighbor embedding ranks visually useful neighbors. The paper does not validate this assumption independently of the full pipeline.
  • domain assumption ImageReward scores, thresholded at 0 and averaged over 10 triples, give a stable relation-level visualizability measure
    Equation 1-2 aggregates binary ImageReward judgments per relation. The threshold of 0 is inherited from ImageReward, and the 10-triple sample is arbitrary; no stability check is provided.
  • domain assumption LLM-generated prompts from triple content are faithful and adequate for text-to-image generation
    Section IV-B relies on ChatGPT to distill and verify neighbor information. The examples are anecdotal; no systematic factuality check is reported.
  • ad hoc to paper Synthetic images can be evaluated against real images using per-entity min/max aggregation and filtering of non-discriminative entities
    Section V-C-b computes FID/CLIPscore only on entities where methods select different neighbors, and aggregates the best matching real image per entity. This is an evaluation-specific design choice, not a standard benchmark protocol.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multi-modal Knowledge Graph Generation with Semantics-enriched Prompts." pith.science (2026). https://pith.science/paper/EPETYXBF

@misc{pith2026250413631,
  author       = {Pith},
  title        = {Pith review of: Multi-modal Knowledge Graph Generation with Semantics-enriched Prompts},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EPETYXBF}},
  note         = {Machine review of arXiv:2504.13631}
}
read the original abstract

Multi-modal Knowledge Graphs (MMKGs) have been widely applied across various domains for knowledge representation. However, the existing MMKGs are significantly fewer than required, and their construction faces numerous challenges, particularly in ensuring the selection of high-quality, contextually relevant images for knowledge graph enrichment. To address these challenges, we present a framework for constructing MMKGs from conventional KGs. Furthermore, to generate higher-quality images that are more relevant to the context in the given knowledge graph, we designed a neighbor selection method called Visualizable Structural Neighbor Selection (VSNS). This method consists of two modules: Visualizable Neighbor Selection (VNS) and Structural Neighbor Selection (SNS). The VNS module filters relations that are difficult to visualize, while the SNS module selects neighbors that most effectively capture the structural characteristics of the entity. To evaluate the quality of the generated images, we performed qualitative and quantitative evaluations on two datasets, MKG-Y and DB15K. The experimental results indicate that using the VSNS method to select neighbors results in higher-quality images that are more relevant to the knowledge graph.

Figures

Figures reproduced from arXiv: 2504.13631 by the authors.

Figure 1
Figure 1. Illustration of multi-modal knowledge graphs [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The framework of translating KGs into MMKGs. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Examples of distilling knowledge from LLM [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Example of Im with lower CIE. D. Main Results a) Automatic Evaluation: To ensure a fair comparison between the VSNS method and the two baselines, we filtered out entities where identical neighbors were selected by both methods. The final number of entities compared, al…
Figure 5
Figure 5. Figure 5: (a): example of Is with lower CIE. (b): examples of generated images of landscapes or places. (c): example of Isvns or Im with lower IQ. TABLE IV THE IMPACT ANALYSIS OF THE VNS MODULE. Methods FID (↓) CLIPscore (↑) #C Number Iwo vns 359.0 0.554 3613 Iw vns 349.3 0.559 …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 21 canonical work pages

  1. [5]

    Draw me like my triples: Leverag- ing generative AI for wikidata image completion,

    R. A. Ahmad, M. Critelli, S. Efeoglu, E. Mancini, C. Ringwald, X. Zhang, and A. Mero ˜no-Pe˜nuela, “Draw me like my triples: Leverag- ing generative AI for wikidata image completion,” in Wikidata@ISWC, ser. CEUR Workshop Proceedings, vol. 3640. CEUR-WS.org, 2023

    work page 2023

  2. [23]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium,

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter, “Gans trained by a two time-scale update rule converge to a local nash equilibrium,” in NIPS, 2017, pp. 6626–6637

    work page 2017

  3. [1]

    Knowledge graphs meet multi-modal learning: A comprehensive survey,

    Z. Chen, Y . Zhang, Y . Fang, Y . Geng, L. Guo, X. Chen, Q. Li, W. Zhang, J. Chen, Y . Zhu et al. , “Knowledge graphs meet multi-modal learning: A comprehensive survey,” arXiv preprint arXiv:2402.05391 , 2024

    arXiv 2024

  4. [2]

    Large language models and knowledge graphs: Opportunities and challenges,

    J. Z. Pan, S. Razniewski, J. Kalo, S. Singhania, J. Chen, S. Dietze, H. Jabeen, J. Omeliyanenko, W. Zhang, M. Lissandrini, R. Biswas, G. de Melo, A. Bonifati, E. Vakaj, M. Dragoni, and D. Graux, “Large language models and knowledge graphs: Opportunities and challenges,” TGDK, vol. 1, no. 1, pp. 2:1–2:38, 2023

    work page 2023

  5. [3]

    Unsupervised semantic correspondence using stable diffusion,

    E. Hedlin, G. Sharma, S. Mahajan, H. Isack, A. Kar, A. Tagliasacchi, and K. M. Yi, “Unsupervised semantic correspondence using stable diffusion,” in NeurIPS, 2023

    work page 2023

  6. [4]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” in NeurIPS, 2020

    2020

  7. [6]

    Oracle bone inscriptions infor- mation processing based on multi-modal knowledge graph,

    J. Xiong, G. Liu, Y . Liu, and M. Liu, “Oracle bone inscriptions infor- mation processing based on multi-modal knowledge graph,” Comput. Electr . Eng., vol. 92, p. 107173, 2021

    work page 2021

  8. [7]

    Imgpedia: A linked dataset with content-based analysis of wikimedia images,

    S. Ferrada, B. Bustos, and A. Hogan, “Imgpedia: A linked dataset with content-based analysis of wikimedia images,” in ISWC (2) , ser. Lecture Notes in Computer Science, vol. 10588. Springer, 2017, pp. 84–93

    work page 2017

Show all 27 references
  1. [8]

    GAIA: A fine-grained multimedia knowledge extraction system,

    M. Li, A. Zareian, Y . Lin, X. Pan, S. Whitehead, B. Chen, B. Wu, H. Ji, S. Chang, C. R. V oss, D. Napierski, and M. Freedman, “GAIA: A fine-grained multimedia knowledge extraction system,” in ACL (demo). Association for Computational Linguistics, 2020, pp. 77–86

    2020

  2. [9]

    Visualsem: a high-quality knowledge graph for vision and language,

    H. Alberts, T. Huang, Y . Deshpande, Y . Liu, K. Cho, C. Vania, and I. Calixto, “Visualsem: a high-quality knowledge graph for vision and language,” CoRR, vol. abs/2008.09150, 2020

    2008 arXiv

  3. [10]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” in CVPR. IEEE Computer Society, 2009, pp. 248–255

    2009

  4. [11]

    Answering visual-relational queries in web- extracted knowledge graphs,

    D. O ˜noro-Rubio, M. Niepert, A. Garc ´ıa-Dur´an, R. Gonzalez-Sanchez, and R. J. L ´opez-Sastre, “Answering visual-relational queries in web- extracted knowledge graphs,” in AKBC, 2019

    2019

  5. [12]

    MMKG: multi-modal knowledge graphs,

    Y . Liu, H. Li, A. Garc ´ıa-Dur´an, M. Niepert, D. O ˜noro-Rubio, and D. S. Rosenblum, “MMKG: multi-modal knowledge graphs,” in ESWC, ser. Lecture Notes in Computer Science, vol. 11503. Springer, 2019, pp. 459–474

    2019

  6. [13]

    TIV A-KG: A multimodal knowledge graph with text, image, video and audio,

    X. Wang, B. Meng, H. Chen, Y . Meng, K. Lv, and W. Zhu, “TIV A-KG: A multimodal knowledge graph with text, image, video and audio,” in ACM Multimedia. ACM, 2023, pp. 2391–2399

    2023

  7. [14]

    Mmpedia: A large-scale multi-modal knowledge graph,

    Y . Wu, X. Wu, J. Li, Y . Zhang, H. Wang, W. Du, Z. He, J. Liu, and T. Ruan, “Mmpedia: A large-scale multi-modal knowledge graph,” in ISWC. Springer, 2023, pp. 18–37

    2023

  8. [15]

    Richpedia: A large-scale, comprehensive multi-modal knowledge graph,

    M. Wang, H. Wang, G. Qi, and Q. Zheng, “Richpedia: A large-scale, comprehensive multi-modal knowledge graph,” Big Data Res. , vol. 22, p. 100159, 2020

    2020

  9. [16]

    Aspectmmkg: A multi- modal knowledge graph with aspect-aware entities,

    J. Zhang, J. Wang, X. Wang, Z. Li, and Y . Xiao, “Aspectmmkg: A multi- modal knowledge graph with aspect-aware entities,” in CIKM. ACM, 2023, pp. 3361–3370

    2023

  10. [17]

    Hierar- chical text-conditional image generation with CLIP latents,

    A. Ramesh, P. Dhariwal, A. Nichol, C. Chu, and M. Chen, “Hierar- chical text-conditional image generation with CLIP latents,” CoRR, vol. abs/2204.06125, 2022

    2022 arXiv

  11. [18]

    Prompt engineering for healthcare: Methodologies and applications,

    J. Wang, E. Shi, S. Yu, Z. Wu, C. Ma, H. Dai, Q. Yang, Y . Kang, J. Wu, H. Hu, C. Yue, H. Zhang, Y . Liu, X. Li, B. Ge, D. Zhu, Y . Yuan, D. Shen, T. Liu, and S. Zhang, “Prompt engineering for healthcare: Methodologies and applications,” CoRR, vol. abs/2304.14670, 2023

    2023 arXiv

  12. [19]

    Imagereward: Learning and evaluating human preferences for text-to- image generation,

    J. Xu, X. Liu, Y . Wu, Y . Tong, Q. Li, M. Ding, J. Tang, and Y . Dong, “Imagereward: Learning and evaluating human preferences for text-to- image generation,” in NeurIPS, 2023

    2023

  13. [20]

    Composition- based multi-relational graph convolutional networks,

    S. Vashishth, S. Sanyal, V . Nitin, and P. P. Talukdar, “Composition- based multi-relational graph convolutional networks,” in ICLR. Open- Review.net, 2020

    2020

  14. [21]

    Wikidata: a free collaborative knowl- edgebase,

    D. Vrandecic and M. Kr ¨otzsch, “Wikidata: a free collaborative knowl- edgebase,” Commun. ACM, vol. 57, no. 10, pp. 78–85, 2014

    2014

  15. [22]

    Dbpedia - A large-scale, multilingual knowledge base extracted from wikipedia,

    J. Lehmann, R. Isele, M. Jakob, A. Jentzsch, D. Kontokostas, P. N. Mendes, S. Hellmann, M. Morsey, P. van Kleef, S. Auer, and C. Bizer, “Dbpedia - A large-scale, multilingual knowledge base extracted from wikipedia,” Semantic Web, vol. 6, no. 2, pp. 167–195, 2015

    2015

  16. [24]

    Clipscore: A reference-free evaluation metric for image captioning,

    J. Hessel, A. Holtzman, M. Forbes, R. L. Bras, and Y . Choi, “Clipscore: A reference-free evaluation metric for image captioning,” in EMNLP (1). Association for Computational Linguistics, 2021, pp. 7514–7528

    2021

  17. [25]

    Learning transferable visual models from natural language supervi- sion,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, and I. Sutskever, “Learning transferable visual models from natural language supervi- sion,” in ICML, ser. Proceedings of Machine Learning Research, vol

  18. [26]

    Native: Multi-modal knowledge graph completion in the wild,

    Y . Zhang, Z. Chen, L. Guo, Y . Xu, B. Hu, Z. Liu, W. Zhang, and H. Chen, “Native: Multi-modal knowledge graph completion in the wild,” in SIGIR. ACM, 2024, pp. 91–101

    2024

  19. [139]

    8748–8763

    PMLR, 2021, pp. 8748–8763

    2021

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.