Pith. sign in

REVIEW 5 major objections 4 minor 61 references

This paper claims that a large language model can fuse a cognitive-neuroscience graph, the Gene Ontology, and the Disease Ontology into one coherent knowledge graph and predict new cross-domain links.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 11:05 UTC pith:JABAEEU2

load-bearing objection The MultiCNKG resource is genuinely new, but the reported F1 and coverage are arithmetically impossible, so the robustness claim does not survive the paper's own formulas. the 5 major comments →

arxiv 2510.06742 v1 pith:JABAEEU2 submitted 2025-10-08 cs.AI cs.LG

MultiCNKG: Integrating Cognitive Neuroscience, Gene, and Disease Knowledge Graphs Using Large Language Models

Ali Sarabadani , Kheirolah Rahsepar Fard This is my paper
classification cs.AI cs.LG
keywords knowledge graph integrationlarge language modelscognitive neuroscienceGene OntologyDisease Ontologyentity alignmentlink predictionGPT-4
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper attempts to show that a single large-language-model pipeline can take three existing knowledge sources — a cognitive-neuroscience graph, the Gene Ontology, and the Disease Ontology — and merge them into one graph, MultiCNKG, that keeps the important relations from each and adds new cross-domain edges the sources do not contain. The intended payoff is a resource that can trace a path from a gene to a disease to a cognitive process, something none of the three input graphs offers by itself. The authors report high alignment quality (precision 85.20%, recall 87.30%, coverage 92.18%, graph consistency 82.50%, expert validation 89.50%) and link-prediction results they interpret as competitive with standard benchmark graphs. Taken at face value, the contribution is a method and a concrete resource for generating gene-disease-cognition hypotheses.

Core claim

The central claim is that LLM-driven entity alignment, semantic relation unification, and iterative graph expansion can reconcile three heterogeneous biomedical/cognitive ontologies into a single directed multigraph with five node types (genes, diseases, cognitive processes, biological pathways, therapeutic targets) and seven relation types (causes, associated with, regulates, involved in, treated by, influences, linked to). The authors argue that the merged graph is semantically coherent and that its predictive signal is usable: across TransE, RotatE, DistMult, ComplEx, ConvE and HolmE, best mean rank is 212 and MRR reaches 0.418, which they compare favourably to FB15k-237, WN18RR, and YAGO

What carries the argument

The mechanism that carries the argument is the LLM as a universal aligner and relation proposer. It encodes entity names and descriptions into embeddings, merges nodes from different ontologies whose cosine similarity exceeds a threshold, maps synonymous relation names (e.g., causes vs. induces) onto a single edge type, and proposes new triples with a confidence score. These new triples enter the graph through the iterative update G(t+1)=G(t)+ΔG(t), and can be removed if validation lowers their confidence. That LLM-generated ΔG is what produces the cross-domain edges MultiCNKG is meant to provide.

Load-bearing premise

The evaluation assumes the LLM's own judgments about alignments and new edges are accurate enough that precision, recall, and coverage can be computed without an independent ground-truth set; if that assumption fails, the reported percentages do not establish coherence.

What would settle it

Recompute the coverage from the published counts using the paper's formula: (6,900 + 11,300) / (57,100 + 88,100) gives roughly 12.5%, not 92.18%; likewise precision 85.20 with recall 87.30 implies F1 ≈ 86.2, not the reported 80.15. A reader who reproduces these discrepancies has direct evidence that at least one evaluation number is inconsistent with the stated definitions, which would need to be resolved before the coherence and robustness conclusion can be trusted.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • MultiCNKG lets a researcher trace a single path from a genetic variant to a neurological disease and then to a cognitive function such as memory or attention.
  • LLM-proposed edges can be filtered by confidence and expert review, so the pipeline is a viable semi-automated hypothesis generator rather than a purely manual curation tool.
  • Link prediction on the merged graph (MRR up to 0.418, P@10 up to 0.571 depending on model) suggests the integrated edge set carries structure that standard embedding models can exploit.
  • The same integration recipe should transfer to other ontology pairs, e.g., drug-disease, since the LLM unifies relation vocabularies rather than relying on one fixed schema.
  • If the authors extend MultiCNKG with drug databases as they propose, the graph would support drug-repurposing queries connecting targets and diseases.

Where Pith is reading between the lines

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

  • Using the paper's own node/edge counts and coverage formula, a direct recomputation gives about 12.5% coverage rather than the reported 92.18%, so the reported evaluation numbers should be treated as unaudited until a reproducible protocol is supplied.
  • A fair test of the LLM's alignments would be to hold out a set of known cross-ontology equivalences (e.g., a disease term that appears in both DO and CNKG) and measure whether the LLM rediscovers them; the authors do not report such a held-out evaluation.
  • The same method could be applied to align other biomedical ontologies (e.g., drug-disease, protein-phenotype), where the LLM's ability to unify heterogeneous relation vocabularies is directly testable against curated mappings.
  • A concrete next experiment is to feed the newly discovered edges to a literature-mining tool and check whether they correspond to recent publications; this would calibrate the novelty score (40.28%) against external evidence.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 4 minor

Summary. The paper describes MultiCNKG, a knowledge graph built by integrating the Cognitive Neuroscience Knowledge Graph (CNKG), Gene Ontology (GO), and Disease Ontology (DO) using GPT-4-based entity alignment, semantic similarity, and graph augmentation. The resulting graph is claimed to contain 6.9K nodes and 11.3K edges. The authors report high precision (85.20%), recall (87.30%), coverage (92.18%), graph consistency (82.50%), and expert validation (89.50%), and present link-prediction results (e.g., TransE MR 391, MRR 0.411; RotatE MR 263, MRR 0.395) as 'competitive' with FB15k-237 and WN18RR.

Significance. If the integration method and evaluation were sound, MultiCNKG would be a potentially useful resource for connecting molecular, disease, and cognitive-neuroscience knowledge, and the use of LLMs for alignment and augmentation is a timely topic. However, the paper's central empirical claims are not supported by the presented evidence. The reported F1 and coverage values contradict the paper's own definitions and numbers, and the link-prediction comparison lacks a defined protocol. The manuscript does not provide code, data, or a reproducible evaluation pipeline, so the quantitative claims cannot be independently checked. The positive aspects are the breadth of the proposed pipeline and the clear research question, but these do not compensate for the load-bearing evaluation defects.

major comments (5)
  1. [§5.1, Table 5] The reported F1-score for MultiCNKG is arithmetically inconsistent with the reported precision and recall. Using the paper's own formula, F1 = 2·P·R/(P+R) = 2·85.20·87.30/(85.20+87.30) ≈ 86.23, not 80.15. An F1 lower than both precision and recall is impossible. This directly undermines the 'robustness' claim built on Table 5.
  2. [§3.3, coverage formula and Table 1] The coverage formula Coverage=(N_merged+E_merged)/Σ(N_i+E_i) gives, with the paper's own numbers, (6.9K+11.3K)/[(2.9K+4.3K)+(11.2K+8.8K)+(43K+75K)] = 18.2K/145.2K ≈ 12.5%, not 92.18%. The reported coverage is off by a factor of about seven. Since coverage is a headline metric in the abstract and conclusion, this is a load-bearing error, not a minor typo.
  3. [§5.2, novelty detection definition] Novelty Score is defined as (LLM-discovered edges)/(total merged edges). By construction, this is the fraction of edges added by the LLM, not a measure of scientific novelty or correctness. Worse, MultiCNKG's novelty score (40.28%) is lower than that of every source graph in Table 5 (GO 44.61, DO 41.05, CNKG 46.38), directly contradicting the paper's narrative that MultiCNKG 'uncovers novel connections.' Without an external gold standard or human adjudication of a sample, this metric does not support the claimed novelty advantage.
  4. [§5.1, precision/recall definitions] Precision and recall are defined over true/false positives and false negatives, but the paper never specifies a ground truth for node/edge alignment. There is no labeled set of correct alignments among CNKG, GO, and DO, nor a description of how TP/FP/FN were counted. Consequently, even the internally consistent numbers would be uninterpretable. This is a fundamental evaluation gap.
  5. [§5.3, Table 6] The link-prediction comparison to FB15k-237, WN18RR, and YAGO3-10 is not meaningful without a protocol. The paper does not state the train/test split, the negative-sampling scheme, whether metrics are filtered or raw, embedding dimensions, training epochs, or hyperparameters. Published benchmark numbers are not comparable to numbers obtained on an arbitrary graph with an unspecified procedure. Moreover, the table's MultiCNKG results are generally worse than the benchmark numbers (e.g., ComplEx MR 212 vs. 144, MRR 0.301 vs. 0.367), so the conclusion that MultiCNKG 'outperforms or matches' these benchmarks is unsupported.
minor comments (4)
  1. [Tables 2 and 3] Table 2 is titled 'Five Different Node Types' and lists five types, but Table 3 is titled 'Five Different Edge Types' while listing seven edge types. The count should be corrected.
  2. [§2, references] Reference [47] (DrugRep-KG) is duplicated as [51], and the two entries are inconsistent in formatting. Also, reference [43] is cited for DrKG, but the cited paper appears to be about COVID-19 drug repurposing; the attribution should be checked.
  3. [§5.2, expert validation] Expert validation is reported as a single percentage with no detail on how many experts, how many edges were reviewed, how agreement was measured, or how sampling was performed. Without this information, 89.50% is just an unsupported point estimate.
  4. [§5.1, graph consistency] Graph consistency is said to use 'semantic and structural checks (ontology-based similarity and OWL-based consistency),' but no operational details or implementation of these checks are given. It is not clear what 82.50% measures.

Circularity Check

2 steps flagged

Novelty detection and link-prediction results reduce to the LLM-generated edge set used to construct MultiCNKG, so two headline validation claims are partially circular by construction.

specific steps
  1. self definitional [Section 5.1 (Traditional Evaluation Methods), Novelty Detection; cf. Section 3.4 (Graph Expansion with LLMs)]
    "Novelty Detection: evaluates the ability of the integrated KG to discover new and previously unknown relationships. Novelty Score = NewlyDiscovered Edges (E)_LLM / Total Edges (E)_merged"

    The numerator is precisely the set of edges added by the LLM in Section 3.4: 'A relation is added if P(r_new | h,t) ≥ τ' and 'G(t+1) = G(t) + ΔG(t)' with ΔG(t) described as 'new entities and edges discovered at iteration t.' Therefore the novelty score is definitionally the fraction of LLM-generated edges in the final graph. It restates the construction step as an evaluation result and cannot independently confirm that the edges are new or valid; the metric equals its own input by definition.

  2. fitted input called prediction [Section 5.3 (Link Prediction), Table 6; cf. Section 3.4 (Graph Expansion with LLMs)]
    "Link prediction methods such as TransE, RotatE, DistMult, ComplEx, ConvE, and HolmE were applied to assess predictive strength. ... Experimental results showed that MultiCNKG achieves comparable or superior results to individual graphs (GO, DO, CNKG), particularly in terms of novelty detection and expert validation."

    The MultiCNKG graphs used for Table 6 are the same graphs expanded by the LLM's relation-prediction step in Section 3.4. No train/test split, filtered setting, or external held-out triple set is described for MultiCNKG. The reported MR/MRR/P@K values are therefore evaluated against the edge set the LLM generated during construction; the 'predictive strength' is partly an in-sample measure of how well embedding models recover the LLM's construction-time outputs rather than an independent prediction of unseen relations.

full rationale

The clearest circularity is the Novelty Detection metric: its formula is exactly the fraction of LLM-discovered edges, so the reported 40.28% novelty is a restatement of the graph-expansion process rather than independent evidence. The link-prediction evaluation is also in-sample by construction as reported: no protocol or external triple set is given, so the Table 6 numbers cannot be read as an independent validation of predictive power. These steps are load-bearing because the abstract and conclusion cite novelty detection and link-prediction results to affirm the KG's coherence and predictive value. By contrast, expert validation (89.50%) is an externally grounded check and is not circular in itself. The F1/coverage arithmetic inconsistencies noted in the prompt are correctness risks, not circularity reductions, and are not scored as circular steps. No load-bearing self-citation chain was found; the duplicated references [47]/[51] are to other authors. Because the graph-integration pipeline and the external expert checks retain independent content, the paper is partially circular rather than fully reducible to its inputs.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

No new natural-kind entities are postulated; the novelties are graph nodes, edges, and categories. The central claims rest on several unspecified free parameters and on strong domain assumptions about embedding similarity and expert validation, none of which are independently tested.

free parameters (4)
  • alignment threshold τ
    Used in §3.3 for node alignment (sim(v_i,v_j) ≥ τ) and in §3.4 for relation prediction (P(r_new) ≥ τ). Value is never specified, yet it controls the size and correctness of the merged graph.
  • similarity spread σ
    Appears in the probabilistic similarity model P(r_new)=1-exp(-||v_h-v_t||^2/(2σ^2)) in §3.4. No value is given; this is a free shape parameter for LLM-discovered relations.
  • preprocessing duplicate threshold
    In §3.1, noise removal uses 'threshold-based filtering on similarity scores'. The threshold is unspecified and affects the node/edge counts.
  • confidence decrement δP
    In §3.4, P(r_new) is decreased by δP when expert/automatic feedback lowers confidence. The value of δP is never given.
axioms (4)
  • domain assumption Cosine similarity of LLM embeddings is a valid proxy for entity equivalence across CNKG, GO, and DO.
    The entire node alignment stage (§3.3) depends on this. No labeled alignment set is used to validate it.
  • domain assumption The source KGs (CNKG, GO, DO) are complete and correct enough to serve as the basis for the integrated graph.
    No analysis of source errors, coverage gaps, or inconsistencies is provided; the final graph is then used as ground truth for link prediction.
  • domain assumption Expert validation of LLM-proposed relations is an unbiased ground truth.
    The paper only states that 'most suggested relations were meaningful'. No protocol, sample size, inter-rater agreement, or blinding is described.
  • ad hoc to paper Link prediction results on MultiCNKG can be compared to published benchmark numbers on FB15k-237 and WN18RR.
    The graphs differ in size, relation schema, density, and presumably in split and negative-sampling protocol. The paper provides no justification for this cross-dataset comparison, yet uses it as evidence of competitiveness.

pith-pipeline@v1.3.0-alltime-deepseek · 11537 in / 11551 out tokens · 93349 ms · 2026-08-04T11:05:20.232583+00:00 · methodology

0 comments
read the original abstract

The advent of large language models (LLMs) has revolutionized the integration of knowledge graphs (KGs) in biomedical and cognitive sciences, overcoming limitations in traditional machine learning methods for capturing intricate semantic links among genes, diseases, and cognitive processes. We introduce MultiCNKG, an innovative framework that merges three key knowledge sources: the Cognitive Neuroscience Knowledge Graph (CNKG) with 2.9K nodes and 4.3K edges across 9 node types and 20 edge types; Gene Ontology (GO) featuring 43K nodes and 75K edges in 3 node types and 4 edge types; and Disease Ontology (DO) comprising 11.2K nodes and 8.8K edges with 1 node type and 2 edge types. Leveraging LLMs like GPT-4, we conduct entity alignment, semantic similarity computation, and graph augmentation to create a cohesive KG that interconnects genetic mechanisms, neurological disorders, and cognitive functions. The resulting MultiCNKG encompasses 6.9K nodes across 5 types (e.g., Genes, Diseases, Cognitive Processes) and 11.3K edges spanning 7 types (e.g., Causes, Associated with, Regulates), facilitating a multi-layered view from molecular to behavioral domains. Assessments using metrics such as precision (85.20%), recall (87.30%), coverage (92.18%), graph consistency (82.50%), novelty detection (40.28%), and expert validation (89.50%) affirm its robustness and coherence. Link prediction evaluations with models like TransE (MR: 391, MRR: 0.411) and RotatE (MR: 263, MRR: 0.395) show competitive performance against benchmarks like FB15k-237 and WN18RR. This KG advances applications in personalized medicine, cognitive disorder diagnostics, and hypothesis formulation in cognitive neuroscience.

Figures

Figures reproduced from arXiv: 2510.06742 by Ali Sarabadani, Kheirolah Rahsepar Fard.

Figure 2
Figure 2. Figure 2 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

61 extracted references · 1 canonical work pages

  1. [1]

    Xia, F., Sun , K., Yu, S., Aziz, A., Wan, L., Pan, S., & Liu, H. (2021). Graph learning: A survey. IEEE Transactions on Artificial Intelligence, 2(2), 109–127

    2021

  2. [2]

    Hu, W., Fey, M., Zitnik, M., Dong, Y., Ren, H., Liu, B., Catasta, M., & Leskovec, J. (2020). Open graph b enchmark: Datasets for machine learning on graphs. Advances in Neural Information Processing Systems, 33, 22118–22133

    2020

  3. [3]

    L., Liu, X., Si, S., Li, Y., Bengio, S., & Hsieh, C

    Chiang, W. L., Liu, X., Si, S., Li, Y., Bengio, S., & Hsieh, C. J. (2019). Cluster -GCN: An efficient algorithm for training deep and large graph convolutional networks. Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 257–266

    2019

  4. [4]

    Li, Q., Li, X., Chen, L., & Wu, D. (2022). Distilling knowledge on text graph for social media attribute inference. Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, 2024–2028

    2022

  5. [5]

    Zhu, J., Cui, Y., Liu, Y., Sun, H., Li, X., Pelger, M., Yang, T., Zhang, L., Zhang, R., & Zhao, H. (2021). TextGNN: Impro ving text encoder via graph neural network in sponsored search. Proceedings of the Web Conference 2021, 2848–2857

    2021

  6. [6]

    Ma, Y., & Tang, J. (2021). Deep learning on graphs. Cambridge University Press

    2021

  7. [7]

    Shao, Y., Taylor, S., Marshall, N., Morioka, C., & Ze ng- Treitler, Q. (2018). Clinical text classification with word embedding features vs. bag -of-words features. 2018 IEEE International Conference on Big Data (Big Data), 2874–2878. IEEE

    2018

  8. [8]

    Mikolov, T. (2013). Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781

    Pith/arXiv arXiv 2013

  9. [9]

    Qiu, X., Sun, T., Xu, Y., Shao, Y., Dai, N., & Huang, X. (2020). Pre-trained models for natural language processing: A survey. Science China Technological Sciences, 63 (10), 1872 – 1897

    2020

  10. [10]

    Miaschi, A., & D ell'Orletta, F. (2020). Contextual and non-contextual word embeddings: An in -depth linguistic investigation. Proceedings of the 5th Workshop on Representation Learning for NLP, 110–119

    2020

  11. [11]

    X., Zhou, K., Li, J., Tang, T., Wang, X., Hou, Y., Min, Y., Zhang, B., Zhang, J., Dong, Z., et al

    Zhao, W. X., Zhou, K., Li, J., Tang, T., Wang, X., Hou, Y., Min, Y., Zhang, B., Zhang, J., Dong, Z., et al. (2023). A survey of large language models. arXiv preprint arXiv:2303.18223

    Pith/arXiv arXiv 2023

  12. [12]

    L., et al

    Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F. L., et al. (2023). GPT -4 technical report. arXiv preprint arXiv:2303.08774

    Pith/arXiv arXiv 2023

  13. [13]

    Kenton, J. D. M. W. C., & Toutanova, L. K. (2019). BERT: Pre-training of deep bidirectional transformers for language understanding. Proceedings of NAACL-HLT, 1, 2. Minneapolis, Minnesota

    2019

  14. [14]

    Radford, A., Wu, J., Child, R., Luan, D., A modei, D., & Sutskever, I., et al. (2019). Language models are unsupervised multitask learners. OpenAI Blog, 1(8), 9

    2019

  15. [15]

    Raffel, C., et al. (2020). Exploring the limits of transfer learning with a unified text -to-text transformer. Journal of Machine Learning Research, 21, 1–67

    2020

  16. [16]

    B., Yu, J., Soricut, R., Schalkwyk, J., Dai, A

    Team G., Anil, R., Borgeaud, S., Wu, Y., Alayrac, J. B., Yu, J., Soricut, R., Schalkwyk, J., Dai, A. M., Hauth, A., et al. (2023). Gemini: A family of highly capable multimodal models. arXiv preprint arXiv:2312.11805

    Pith/arXiv arXiv 2023

  17. [17]

    W., Sutton, C., Gehrmann, S., et al

    Chowdhery, A., Narang, S., Devlin, J., Bosma, M., Mishra, G., Roberts, A., Barham, P., Chung, H. W., Sutton, C., Gehrmann, S., et al. (2023). PaLM: Scaling language modeling with pathways. Journal of Machine Learning Research, 24(240), 1–113

    2023

  18. [18]

    A., Awan, A

    Abdin, M., Jacobs, S. A., Awan, A. A., Aneja, J., Awadallah, A., Awadalla, H., et al. (2024). Phi -3 technical report: A highly capable language model locally on your phone. arXiv preprint arXiv:2404.14219

    Pith/arXiv arXiv 2024

  19. [19]

    A., Lacroix, T., Rozière, B., Goyal, N., Hambro, E., Azhar, F., et al

    Touvron, H., Lavril, T., Izacard, G., Martinet, X., L achaux, M. A., Lacroix, T., Rozière, B., Goyal, N., Hambro, E., Azhar, F., et al. (2023). LLaMA: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971

    Pith/arXiv arXiv 2023

  20. [20]

    J., & Lu, H

    Webb, T., Holyoak, K. J., & Lu, H. (2023). Emergent analogical reasoning i n large language models. Nature Human Behaviour, 7(9), 1526–1541

    2023

  21. [21]

    A., Tourigny, N., Rigault, P., & Morissette, J

    Belleau, F., Nolin, M. A., Tourigny, N., Rigault, P., & Morissette, J. (2008). Bio2RDF: Towards a mashup to build bioinformatics knowledge systems. Journal of Biomedical Informatics, 41(5), 706–716

    2008

  22. [22]

    S., Lizee, A., Hessler, C., Brueggeman, L., Chen, S

    Himmelstein, D. S., Lizee, A., Hessler, C., Brueggeman, L., Chen, S. L., Hadley, D., Green, A., Khankhanian, P., & Baranzini, S. E. (2017). Systematic integration of biomedical knowledge prioritizes drugs for repurposing. eLife, 6 , e26726

    2017

  23. [23]

    Luo, R., Sun, L., Xia, Y., Qin, T., Zhang, S., Poon, H., & Liu, T. Y. (2022). BioGPT: Generative pre -trained transformer for biomedical text generation and mining. Briefings in Bioinformatics, 23(6), bbac409

    2022

  24. [24]

    S., Chen, F., Gong, C., Bracken-Clarke, D., Xue, E., Yang, Y., Sun, J., & Lu, Z

    Jin, Q., Wang, Z., Floudas, C. S., Chen, F., Gong, C., Bracken-Clarke, D., Xue, E., Yang, Y., Sun, J., & Lu, Z. (2023). Matching patients to clinical trials with large language models. arXiv

    2023

  25. [25]

    Yuan, H., Yuan, Z., Gan, R., Zhang, J., Xie, Y., & Yu, S. (2022). BioBART: Pretraining and ev aluation of a biomedical generative language model. arXiv preprint arXiv:2204.03905

    Pith/arXiv arXiv 2022

  26. [26]

    A., Rouvier, M., & Dufour, R

    Labrak, Y., Bazoge, A., Morin, E., Gourraud, P. A., Rouvier, M., & Dufour, R. (2024). BioMistral: A collection of open-source pretrained large language models for med ical domains. arXiv preprint arXiv:2402.10373

    Pith/arXiv arXiv 2024

  27. [27]

    J., & Dumontier, M

    Alharbi, R., Ahmed, U., Dobriy, D., Lajewska, W., Menotti, L., Saeedizade, M. J., & Dumontier, M. (2023). Exploring the role of generative AI in constructing knowledge graphs for drug indications with medical context. Proceedings http://ceur-ws.org

    2023

  28. [28]

    A., Rahman, F., & Grimm, C

    Wawrzik, F., Rafique, K. A., Rahman, F., & Grimm, C. (2023). Ontology learning applications of knowledge base construction for microelectronic systems info rmation. Information, 14(3), 176

    2023

  29. [29]

    R., & Wu, Y

    Peng, C., Yang, X., Yu, Z., Bian, J., Hogan, W. R., & Wu, Y. (2023). Clinical concept and relation extraction using prompt-based machine reading comprehension. Journal of the American Medical Informatics Association, 30(9), 1486 – 1493

    2023

  30. [30]

    Pan, S., Luo, L., Wang, Y., Chen, C., Wang, J., & Wu, X. (2024). Unifying large language models and knowledge graphs: A roadmap. IEEE Transactions on Knowledge and Data Engineering

    2024

  31. [31]

    Khorashadizadeh, H., Mihindukulasooriya, N., T iwari, S., Groppe, J., & Groppe, S. (2023). Exploring in -context learning capabilities of foundation models for generating knowledge graphs from text. arXiv preprint arXiv:2305.08804

    Pith/arXiv arXiv 2023

  32. [32]

    G., Ghaffari, P., & Hokamp, C

    Boylan, J., Mangla, S., Thorn, D., Ghalandari, D. G., Ghaffari, P., & Hokamp, C. (2024). KGValidator: A Framework for Automatic Validation of Knowledge Graph Construction. arXiv preprint arXiv:2404.15923

    Pith/arXiv arXiv 2024

  33. [33]

    P., & Groth, P

    Allen, B. P., & Groth, P. T. (2024). Evaluating Class Membership Relations in Knowledge Graphs using Large Language Models. arXiv preprint arXiv:2404.17000

    Pith/arXiv arXiv 2024

  34. [34]

    W., Morris, J

    Soman, K., Rose, P. W., Morris, J. H., Akbas, R. E., Smith, B., Peetoom, B., Villouta -Reyes, C., Cerono, G., Shi, Y., Rizk - Jackson, A., et al. (2023). Biomedical knowledge graph - enhanced prompt generation fo r large language models. arXiv preprint arXiv:2311.17330

    Pith/arXiv arXiv 2023

  35. [35]

    Cm, S., Prakash, J., & Singh, P. K. (2023). Question answering over knowledge graphs using BERT based relation mapping. Expert Systems, 40(10), e13456

    2023

  36. [36]

    Guo, Q., Cao, S., & Yi, Z. (2022). A medical question answering system using large language models and knowledge graphs. International Journal of Intelligent Systems, 37(11), 8548–8564

    2022

  37. [37]

    Wu, Y., Hu, N., Bi, S., Qi, G., Ren, J., Xie, A., & Song, W. (2023). Retrieve -rewrite-answer: A KG -to-text enhanced LLMs framework for knowledge graph question answering. arXiv preprint arXiv:2309.11206

    Pith/arXiv arXiv 2023

  38. [38]

    Choudhary, N., & Reddy, C. K. (2023). Complex logical reasoning over knowledge graphs using large language models. arXiv preprint arXiv:2305.01157

    Pith/arXiv arXiv 2023

  39. [39]

    K., & Ekbal, A

    Varshney, D., Zafar, A., Behera, N. K., & Ekbal, A. (2023). Knowledge grounded medical dialogue generation using augmented graphs. Scientific Reports, 13(1), 3310

    2023

  40. [40]

    Jiang, P., Xiao, C., Cross, A., & Sun, J. (2023). Graphcare: Enhancing healthcare predictions with personalized knowledge graphs. arXiv preprint arXiv:2305.12788

    Pith/arXiv arXiv 2023

  41. [41]

    D., Ho, J

    Xu, R., Shi, W., Yu, Y., Zhuang, Y., Jin, B., Wang, M. D., Ho, J. C., & Yang, C. (2024). RAM -EHR: Retrieval augmentation meets clinical predictions on electronic health r ecords. arXiv preprint arXiv:2403.00815

    Pith/arXiv arXiv 2024

  42. [42]

    W., Churpek, M

    Gao, Y., Li, R., Croxford, E., Tesch, S., To, D., Caskey, J., Patterson, B. W., Churpek, M. M., Miller, T., Dligach, D., et al. (2023). Large language models and medical knowledge grounding for diagnosis prediction. medRxiv, 2023–11

    2023

  43. [43]

    Few -shot Link Prediction via Graph Neural Networks for COVID -19 Drug -repurposing,

    V. N. Ioannidis et al., "Few -shot Link Prediction via Graph Neural Networks for COVID -19 Drug -repurposing,"

  44. [45]

    Gene Ontology: Tool for the Unification of Biology,

    M. Ashburner et al., "Gene Ontology: Tool for the Unification of Biology," Nature Genetics, vol. 25, no. 1, 2000, pp. 25–29. DOI:10.1038/75556

    doi:10.1038/75556 2000

  45. [46]

    KG -Predict: A Knowledge Graph Computational Framework for Drug Repurposing,

    Z. Gao, P. Ding, and R. Xu, "KG -Predict: A Knowledge Graph Computational Framework for Drug Repurposing," Journal of Biomedical Informatics , vol. 132, 2022, 104133. DOI:10.1016/j.jbi.2022.104133

    arXiv 2022

  46. [47]

    DrugRep -KG: Toward L earning a Unified Latent Space for Drug Repurposing Using Knowledge Graphs,

    Z. Ghorbanali et al., "DrugRep -KG: Toward L earning a Unified Latent Space for Drug Repurposing Using Knowledge Graphs," Journal of Chemical Information and Modeling, vol. 63, no. 8, 2023, pp. 2532–2545

    2023

  47. [48]

    Disease Ontology: A Backbone for Disease Semantic Integration,

    L. Schriml et al., "Disease Ontology: A Backbone for Disease Semantic Integration," Nucleic Acids Research, vol. 40, no. D1, 2011, pp. D940–D946. DOI:10.1093/nar/gkr972

    work page doi:10.1093/nar/gkr972 2011

  48. [49]

    DrugBank 5.0: A Major Update to the DrugBank Database for 2018,

    D. S. Wishart et al., "DrugBank 5.0: A Major Update to the DrugBank Database for 2018," Nucleic Acids Research , vol. 46, no. D1, 2018, pp. D1074 –D1082. DOI:10.1093/nar/gkx1037

    doi:10.1093/nar/gkx1037 2018

  49. [50]

    Building a Knowledge Graph to Enable Precision Medicine,

    P. Chandak, K. Huang, and M. Zitnik, "Building a Knowledge Graph to Enable Precision Medicine," Scientific Data, vol. 10, no. 1, 2023, 67. https://doi.org/10.1038/s41597-023-01960-3

    doi:10.1038/s41597-023-01960-3 2023

  50. [51]

    DrugRep -KG: Toward Learning a Unified Latent Space for Drug Repurposing Using Knowledge Graphs,

    Z. Ghorbanali et al., "DrugRep -KG: Toward Learning a Unified Latent Space for Drug Repurposing Using Knowledge Graphs," Journal of Chemical Information and Modeling, vol. 63, no. 8, 2023, pp. 2532–2545

    2023

  51. [52]

    Freebase: A Collaboratively Created Graph Database for Structuring Human Knowledge,

    K. Bollacker, C. Evans, P. Paritosh, T . Sturge, and J. Taylor, "Freebase: A Collaboratively Created Graph Database for Structuring Human Knowledge," in Proceedings of the 2008 ACM SIGMOD International Conference on Management of Data (SIGMOD '08), New York, NY, USA, 2008, pp. 1247 –1250. https://doi.org/10.1145/1376616.1376746

    arXiv 2008

  52. [53]

    WordNet: A Lexical Database for English,

    G. A. Miller, "WordNet: A Lexical Database for English," Communications of the ACM, vol. 38, no. 11, 1995, pp. 39 –41. https://doi.org/10.1145/219717.219748

    arXiv 1995

  53. [54]

    Yago: A Core of Semantic Knowledge,

    F. M. Suchanek, G. Kasneci, and G. Weikum, "Yago: A Core of Semantic Knowledge," in Proceedings of the 16th International Conference on World Wide Web (WWW '07) , New York, NY, US A, 2007, pp. 697 –706. https://doi.org/10.1145/1242572.1242667

    arXiv 2007

  54. [55]

    Translating Embeddings for Modeling Multi - relational Data,

    A. Bordes, N. Usunier, A. Garcia-Durán, J. Weston, and O. Yakhnenko, "Translating Embeddings for Modeling Multi - relational Data," in Proceedings of the 26th International Conference on Neural Information Processing Systems (NIPS '13), Red Hook, NY, USA, 2013, pp. 2787 –2795. https://hal.science/hal-00920777

    2013

  55. [56]

    Rotate: Knowledge Graph Embedding by Relational Rotation in Complex Space,

    Z. Sun et al., "Rotate: Knowledge Graph Embedding by Relational Rotation in Complex Space," arXiv preprint, 2019. https://doi.org/10.48550/arXiv.1902.10197

    Pith/arXiv arXiv doi:10.48550/arxiv.1902.10197 2019

  56. [57]

    Emb edding Entities and Relations for Learning and Inference in Knowledge Bases,

    B. Yang et al., "Emb edding Entities and Relations for Learning and Inference in Knowledge Bases," in International Conference on Learning Representations,

  57. [58]

    Complex Embeddings for Simple Link Prediction,

    T. Trouillon et al., "Complex Embeddings for Simple Link Prediction," in Proceedings of the International Conference on Machine Learning, New York, USA, 2016, pp. 2071–2080

    2016

  58. [59]

    Convolutional 2D Knowledge Graph Embeddings,

    T. Dettmers et al., "Convolutional 2D Knowledge Graph Embeddings," in Proceed ings of the AAAI Conference on Artificial Intelligence, New Orleans, USA, 2018. https://doi.org/10.1609/aaai.v32i1.11573

    doi:10.1609/aaai.v32i1.11573 2018

  59. [60]

    HolmE: Low -dimensional Hyperbolic KG Embedd ing for Better Extrapolation,

    Z. Zheng et al., "HolmE: Low -dimensional Hyperbolic KG Embedd ing for Better Extrapolation," in Proceedings of the ESWC Conference, 2024

    2024

  60. [2014]

    https://doi.org/10.48550/arXiv.1412.6575

    Pith/arXiv arXiv doi:10.48550/arxiv.1412.6575

  61. [2020]

    DOI:10.48550/arXiv.2007.10261

    Pith/arXiv arXiv doi:10.48550/arxiv.2007.10261 2007