Pith. sign in

REVIEW 5 major objections 5 minor 48 references

SCoRE: Streamlined Corpus-based Relation Extraction using Multi-Label Contrastive Learning and Bayesian kNN

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read SCoRE claims that a frozen-embedding, no-fine-tuning pipeline matches or outperforms state-of-the-art relation extraction on five benchmarks while using orders of magnitude less energy.

desk verdict A believable no-finetune RE system whose most interesting result is negative, but the math is sloppy in three places and the headline numbers need error bars before I'd trust them. read the letter →

arxiv 2507.06895 v1 pith:HMOH7ETY submitted 2025-07-09 cs.CL cs.AIcs.IRcs.LG

Luca Mariotti , Veronica Guidetti , Federica Mandreoli This is my paper · ORCID
classification cs.CLcs.AIcs.IRcs.LG
keywords multi-labelrelationextractiondistantsupervisionknowledgegraphenrichmentBayesiankNNcontrastivelearningfrozenpretrainedlanguagemodelsenergy-efficientNLPWiki20d
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 proposes SCoRE, a sentence-level relation extraction system that deliberately avoids the usual recipe of fine-tuning a large pretrained language model. Instead, it runs a single frozen BERT-base pass to average the embeddings of the head and tail entity mentions, trains a small MLP with a multi-label supervised contrastive loss, and classifies new mentions with a Bayesian k-nearest-neighbor rule on the learned hypersphere. The authors claim that on five benchmarks—NYT10M, NYT10D, DisRex, Wiki20m, and their new Wiki20d—this minimal pipeline matches or beats state-of-the-art methods while consuming as little as 0.002 kWh for training and testing, orders of magnitude below a fine-tuned competitor. They also argue that extra complexity hurts when fine-tuning is removed: adding the CLS token, full-sentence attention, or dynamic encoding degrades performance. If right, the paper shows that the relational signal needed for extraction already lives in frozen PLM embeddings, and that heavy fine-tuning is not the price of accuracy.

What carries the argument

The load-bearing machinery is the metric space built by supervised contrastive learning and then exploited by a non-parametric classifier. During dataset creation the frozen PLM maps each sentence once to token embeddings; the head and tail mentions are averaged and concatenated, $\mathbf{x}_{i,j}=[\mathbf{e}_{i,h_j};\mathbf{e}_{i,t_j}]$. A shallow MLP with normalized output projects these vectors onto the hypersphere under a multi-label supervised contrastive loss in which the pair weight $\beta_{ij}$ is the normalized number of shared relation labels, so the distance $d(\cdot,\cdot)$ in the loss becomes relation-aware. At inference, Bayesian kNN estimates each relation posterior from the $k$ nearest neighbors weighted by $\exp(-d/\tau)$ with a flat prior $P(r_h)=1/2$, and a universal threshold produces the final multi-label prediction. The key design point is that the same distance measure is used in training and inference, so the learned geometry and the classifier are one and the same.

What would settle it

Construct a test set where the same head–tail entity pair appears in sentences that express different relations (for example, 'X was born in Y' versus 'X lives in Y'), and check whether SCoRE can separate them, since its input uses only the averaged mention embeddings and ignores the rest of the sentence. A second decisive check is to replace the frozen BERT-base encoder with randomly initialized embeddings of the same dimension: if micro-F1 on NYT10M stays near the reported 77.6, the relational signal is not coming from pretrained language knowledge; if it collapses, the frozen-encoder assumption is doing the work.

Watch

Extended reading notes

Core claim

The paper's claim is that sentence-level multi-label relation extraction does not need fine-tuned pretrained language models. SCoRE encodes every sentence once with frozen BERT-base, averages the embeddings of the head and tail entity mentions, concatenates the two averages, and trains only a small MLP under a multi-label supervised contrastive loss; inference is a Bayesian kNN over the learned hypersphere. On the five benchmarks the authors report the best micro-F1 on NYT10M (77.6) and NYT10D (89.2), the best macro-F1 on Wiki20D (23.9), competitive micro-F1 on DisRex (75.4 vs 77.1) and Wiki20m (83.5 vs 83.8), and the lower CSD on every dataset when compared with the strongest fine-tuned baseline, PARE. Training plus testing consumes 0.002–0.06 kWh, versus 2.11–6.20 kWh for bag-level PARE and much more for its sentence-level adaptation, so the paper concludes that accuracy and cost are not in tension the way the field's fine-tuning-centric practice assumes.

Load-bearing premise

The load-bearing premise is that averaging the frozen BERT-base embeddings of just the head and tail entity mentions preserves enough relational information to separate all relation types; if those averages are relationally uninformative, no amount of contrastive learning or Bayesian kNN can recover the signal.

Editorial extensions

If this is right

  • Because the PLM is used only in a single forward pass during dataset creation, swapping in a newer or larger PLM costs an embedding re-run rather than a full re-training, making the system track the fast-moving PLM ecosystem cheaply.
  • The anti-ablation results imply that without fine-tuning, additional input tokens (CLS), full-sentence attention, and dynamic encoding actively reduce micro- and macro-F1 while multiplying energy use, so SCoRE's minimal design is not a compromise but a better operating point for the no-fine-tuning regime.
  • The CSD analysis shows that a fine-tuned model can earn competitive F1 while conflating inverse relations (e.g., 'followed' vs 'followed-by'), so SCoRE's sentence-level, non-parametric inference is claimed to preserve the directed structure of the KG better.
  • The sensitivity analysis identifies flat prior plus universal threshold as the recommended configuration and argues that @M ranking metrics are unreliable in multi-label RE, with P@R better reflecting real-world recommender use.

Reading between the lines

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

  • A natural extension the paper does not run: replace BERT-base with a larger frozen PLM and check whether the same no-fine-tuning pipeline improves long-tail relations, since modularity claims would predict monotonic gains.
  • The CSD metric could be turned into a training or model-selection signal: if SCoRE's lower CSD is what protects KG structure, one could validate threshold and neighbor choices by CSD rather than F1 alone.
  • In a human-in-the-loop KG-enrichment workflow, SCoRE's stable P@R across thresholds suggests it is best deployed as a ranker that suggests candidate relations for expert confirmation, rather than as a hard classifier.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper presents SCoRE, a sentence-level multi-label relation extraction system. It encodes each sentence once with a frozen BERT-base model, averages the head and tail entity token embeddings, trains a small MLP with a supervised multi-label contrastive loss, and predicts labels at inference using a Bayesian kNN over the training embeddings with a shared probability threshold. The authors evaluate on five benchmarks (NYT10M, NYT10D, DisRex, Wiki20m, Wiki20D), compare with five SOTA systems, report micro/macro-F1 and energy consumption, and propose two new metrics (P@R and CSD) plus a new fully distantly supervised benchmark, Wiki20d. The central empirical claim is that SCoRE matches or surpasses SOTA while consuming far less energy.

Significance. If the claims held, this would be a useful contribution: it would show that a frozen PLM plus contrastive training and non-parametric inference is competitive with heavily fine-tuned systems, with real advantages in modularity and reproducibility. The release of Wiki20d, the explicit energy measurements, and the anti-ablation experiments are concrete strengths. However, the manuscript currently contains a sign inconsistency in the contrastive loss, an over-stated Bayesian interpretation of the kNN posterior, an unsupported claim of outperformance on Wiki20D, and an incorrectly specified new metric. These issues are load-bearing for the headline results and must be resolved before the contribution can be properly assessed.

major comments (5)
  1. [3.2, Eq. (4)] Equation (4) defines the contrastive loss with exp(d(z_i,z_j)/tau), where d is a distance (Euclidean or cosine), but Equations (6) and (7) and the surrounding text use exp(-d/tau). With a literal distance in Eq. (4), maximizing the numerator would pull positive pairs apart, the opposite of the claimed clustering behavior. If the implementation uses -d, the printed equation is wrong; if it uses +d, the training objective is inconsistent with the metric used in the kNN likelihood of Eq. (6). Because the paper claims that the kNN exploits the metric induced by CL, this sign issue is central and must be corrected and verified in the released code.
  2. [3.3, Eq. (7)] With the stated flat prior P(r_h)=1/2, the prior cancels algebraically in Eq. (7), leaving the posterior as a distance-weighted average of neighbor labels, sum_i y_i^h exp(-d/tau) / sum_i exp(-d/tau). This is standard distance-weighted kNN, with no class-frequency correction and no informative prior. The claims that the flat prior 'focuses on optimizing recall' or 'adopts a dynamic, local neighborhood-dependent class weight' are not derived from Eq. (7). The authors should either provide a concrete Bayesian contribution that remains after the algebra or stop describing the default configuration as class-prior-aware.
  3. [7.1, Table 2] The text states that SCoRE 'outperforms competitors on NYT10D, NYT10M, and Wiki20d', but Table 2 lists PARE at 67.2 and HiCLRE at 68.3 micro-F1 on Wiki20D, both above SCoRE's 66.9. Furthermore, Table 2 reports the best of five runs with no variance, while Table 6 gives the UU configuration on Wiki20D as 65.1 ± 2.5 over 10 runs; the reported 66.9 is therefore within one standard deviation of the mean, and the SOTA gap is not statistically established. Please report means and standard deviations for the main comparison and qualify the 'matches or surpasses' claim accordingly.
  4. [4.5, Eq. (13)] The definition of n_10 in Eq. (13) uses \hat y_h^j (1 - y_p^j), mixing predicted and true labels inside a single phi coefficient, while n_01 uses (1 - y_h^j) y_p^j. As written, phi is not a well-defined correlation for either the true or the predicted label matrix. If the intended computation applies Eq. (13) separately to the true matrix and to the predicted matrix, the hat in n_10 is a typo and must be removed; otherwise the CSD values in Tables 3 and 6 are not reproducible from the formula given.
  5. [7.1, Table 4] The energy comparison is not on equal footing. The PARE SL and PARE BAG entries appear to include the entire process in one measurement, while SCoRE's 'Train & Test' column excludes the dataset-creation phase reported in a separate column (e.g., 0.63 kWh for NYT10M and 5.75 kWh for Wiki20D). For a fair comparison, the total SCoRE cost (dataset creation plus training and testing) should be compared against the PARE totals; otherwise the headline claim of reducing energy consumption by 'up to three orders of magnitude' is overstated for the smaller datasets.
minor comments (5)
  1. [3.3] There is a duplicated word in 'we estimate testsetrelationtypeprobabilitiesprobabilities right after CL training'; please fix the typo.
  2. [7.1 vs 7.2] The baseline SCoRE numbers in Table 5 differ slightly from Table 2 (e.g., NYT10M 77.4 vs 77.6, Wiki20D 66.8 vs 66.9) with no explanation; clarify the relationship between the two tables.
  3. [3.1] The text calls y_{i,j} a 'one-hot encoding', but for multi-label samples it is a binary label vector with potentially several ones; please use 'binary label vector' to avoid confusion.
  4. [4.5] The margin notation n_{1·} and n_{·1} in Eq. (13) is ambiguous because the text later says phi is computed on the true and predicted matrices separately; please add explicit superscripts to distinguish the two computations.
  5. [7.3] The paragraph states that 'no single setup consistently outperforms the others' but then says UU 'confirms to be the best' with 16 podium positions; make the selection criterion explicit, since podium counts and mean metrics can disagree.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: SCoRE's F1 and energy claims are evaluated against external benchmarks and held-out test metrics, and the flat-prior kNN reduction is a simplification rather than a circular derivation.

full rationale

SCoRE's central claim, that a frozen-PLM embedding plus multi-label contrastive MLP and kNN inference matches or surpasses state-of-the-art relation extraction at lower energy, is evaluated on externally annotated benchmarks (NYT10M, NYT10D, DisRex, Wiki20m, Wiki20D). The training objective in Eq. 4 is a supervised contrastive loss defined over training labels, and the inference rules in Eqs. 5-8 estimate test-set posteriors from stored training labels and distances; neither contains the test labels nor the reported micro-F1/macro-F1 values. The per-dataset choices of c and k are disclosed as validation- or training-set selections, not as fits to the test metrics, so the 'fitted input called prediction' pattern does not apply. The proposed CSD and P@R metrics are computed on held-out test predictions and compared against ground truth, making them post-hoc evaluation measures rather than targets encoded into training. Although the flat prior P(r_h)=1/2 in Eq. 7 cancels algebraically and reduces the 'Bayesian' posterior to a distance-weighted average of neighbor labels, this is an internal simplification of the inference rule, not a circularity: the weights remain functions of learned embeddings and stored training labels, and the reported scores are never fed back into the objective. No load-bearing premise is justified by the present authors' own prior results; the cited probabilistic kNN, supervised contrastive learning, and multi-label CL adaptation results come from external works. Apparent issues such as the sign convention in Eq. 4 and the sensitivity of Table 6 are correctness or reproducibility concerns, not evidence that the derivation reduces to its inputs. Therefore no circular step is exhibited and the appropriate circularity score is 0.

Assumptions & free parameters 7 free parameters · 6 assumptions · 0 invented entities

The central claim rests on a small set of model choices: a frozen PLM encoder, a contrastive projection MLP, a Bayesian kNN classifier, and per-dataset hyperparameters. No new theoretical entities are introduced. The most fragile assumptions are that frozen BERT embeddings contain the relational signal and that the CL loss as written (with the distance sign convention) actually learns the intended metric. The many grid-searched hyperparameters are disclosed but their final values are only partially reported.

free parameters (7)
  • MLP architecture (depth l, width m, output dim m_h) = final values not visible in manuscript; selected via grid search
    Grid search over l in [3,4,5], l/m in [0.01,0.05,0.1], m_h in [5,10,15]; chosen on validation performance.
  • temperature tau = from [0.01, 0.05, 0.1, 0.2]
    Controls sharpness of contrastive loss; selected per dataset.
  • learning rate = from [1e-4, 5e-3, 1e-3]
    Grid-searched as part of the common configuration.
  • batch size = from [64, 128, 256]
    Grid-searched as part of the common configuration.
  • kNN k per dataset = NYT10M:50, NYT10D:100, DisRex:50, Wiki20M:100, Wiki20D:150
    Selected on validation or training set; controls the local density estimate in the Bayesian kNN.
  • probability threshold c per dataset = NYT10M:0.6, NYT10D:0.7, DisRex:0.5, Wiki20M:0.5, Wiki20D:0.7
    Converts posterior probabilities to binary labels; tuned per dataset.
  • distance measure d = euclidean or cosine, selected per dataset
    Used in both contrastive loss and kNN; the sign convention in Eq. 4 vs Eq. 6 is inconsistent, affecting the definition of the objective.
assumptions (6)
  • domain assumption Frozen BERT-base embeddings contain sufficient relational information.
    The entire dataset creation relies on a single forward pass of an off-the-shelf PLM; no task adaptation is performed (Section 3.1).
  • ad hoc to paper Supervised contrastive loss from [38] is valid for multi-label RE.
    Eq. 4 adapts [19]/[38]; the paper does not prove convergence or that the learned metric is appropriate for kNN classification.
  • domain assumption Bayesian kNN with flat prior is appropriate for DS noise.
    Section 3.3 assumes P(r_h)=1/2 and that this prioritizes recall; this is a design choice with no validation across all datasets.
  • domain assumption Distant supervision labels (KG-derived) are treated as ground truth.
    Dataset creation assumes KG triples are correct; noise is acknowledged but not modeled (Section 5, Wiki20d description).
  • standard math Pearson phi and Frobenius norm are suitable for measuring label-correlation distance.
    Used in CSD; a simple linear correlation measure with a standard matrix norm.
  • domain assumption Grid search on validation (or training for NYT10D) selects optimal hyperparameters without overfitting.
    NYT10D has no validation set, so c and k are selected on training data, risking optimism in reported results (Section 6.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of SCoRE: Streamlined Corpus-based Relation Extraction using Multi-Label Contrastive Learning and Bayesian kNN." pith.science (2026). https://pith.science/paper/HMOH7ETY

@misc{pith2026250706895,
  author       = {Pith},
  title        = {Pith review of: SCoRE: Streamlined Corpus-based Relation Extraction using Multi-Label Contrastive Learning and Bayesian kNN},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HMOH7ETY}},
  note         = {Machine review of arXiv:2507.06895}
}
read the original abstract

The growing demand for efficient knowledge graph (KG) enrichment leveraging external corpora has intensified interest in relation extraction (RE), particularly under low-supervision settings. To address the need for adaptable and noise-resilient RE solutions that integrate seamlessly with pre-trained large language models (PLMs), we introduce SCoRE, a modular and cost-effective sentence-level RE system. SCoRE enables easy PLM switching, requires no finetuning, and adapts smoothly to diverse corpora and KGs. By combining supervised contrastive learning with a Bayesian k-Nearest Neighbors (kNN) classifier for multi-label classification, it delivers robust performance despite the noisy annotations of distantly supervised corpora. To improve RE evaluation, we propose two novel metrics: Correlation Structure Distance (CSD), measuring the alignment between learned relational patterns and KG structures, and Precision at R (P@R), assessing utility as a recommender system. We also release Wiki20d, a benchmark dataset replicating real-world RE conditions where only KG-derived annotations are available. Experiments on five benchmarks show that SCoRE matches or surpasses state-of-the-art methods while significantly reducing energy consumption. Further analyses reveal that increasing model complexity, as seen in prior work, degrades performance, highlighting the advantages of SCoRE's minimal design. Combining efficiency, modularity, and scalability, SCoRE stands as an optimal choice for real-world RE applications.

Figures

Figures reproduced from arXiv: 2507.06895 by the authors.

Figure 1
Figure 1. Visual representation of SCoRE, showcasing the dataset creation, training, and testing stages. 𝑡𝑟𝑎𝑖𝑛 = {(𝐱𝑖 , 𝐲𝑖 )}𝑖=1,…,𝑁 𝑡𝑒𝑠𝑡 = {(𝐱̃ 𝑖 , 𝐲̃ 𝑖 )}𝑖=1,…,𝑁̃ where 𝑁 = ∑ 𝑖∶𝑠𝑖∈′ 𝑚𝑖 and 𝑁̃ = ∑ 𝑖∶𝑠𝑖∈′′ 𝑚𝑖 are the total number of training and test data samples respectively, with the double indexing dropped for clarity. Note that the LLM encoder is employed solely during dataset creation, serving as an informed prior. T… view at source ↗
Figure 2
Figure 2. Wiki20D relation class distribution [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. PARE and SCoRE worst correlation distance entries [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 43 canonical work pages

  1. [1]

    Recon: Relation extraction using knowledge graph contextinagraphneuralnetwork.ProceedingsoftheWebConference 2021

    Bastos, A., Nadgeri, A., Singh, K., Mulang’, I.O., Shekarpour, S., Hoffart, J., 2020. Recon: Relation extraction using knowledge graph contextinagraphneuralnetwork.ProceedingsoftheWebConference 2021

    work page 2020

  2. [2]

    DiS-ReX: A multilingual dataset for distantly supervised relation extraction, in: Muresan, S., Nakov, P., Villavicencio, A

    Bhartiya, A., Badola, K., Mausam, 2022. DiS-ReX: A multilingual dataset for distantly supervised relation extraction, in: Muresan, S., Nakov, P., Villavicencio, A. (Eds.), Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), Association for Computational Linguistics, Dublin, Ireland. pp. 849–863

    work page 2022

  3. [3]

    Cleangraph: Human-in-the- loop knowledge graph refinement and completion

    Bikaun, T., Stewart, M., Liu, W., 2024. Cleangraph: Human-in-the- loop knowledge graph refinement and completion. arXiv preprint arXiv:2405.03932

    arXiv 2024

  4. [4]

    Learningtoextractrelationsfromthe web using minimal supervision, in: Proceedings of the 45th annual meetingoftheassociationofcomputationallinguistics,pp.576–583

    Bunescu,R.,Mooney,R.,2007. Learningtoextractrelationsfromthe web using minimal supervision, in: Proceedings of the 45th annual meetingoftheassociationofcomputationallinguistics,pp.576–583

    work page 2007

  5. [5]

    Empowerdistantlysupervisedrelationextractionwithcollab- orative adversarial training, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp

    Chen, T., Shi, H., Liu, L., Tang, S., Shao, J., Chen, Z., Zhuang, Y., 2021a. Empowerdistantlysupervisedrelationextractionwithcollab- orative adversarial training, in: Proceedings of the AAAI Conference on Artificial Intelligence, pp. 12675–12682

    work page

  6. [6]

    Chen, T., Shi, H., Tang, S., Chen, Z., Wu, F., Zhuang, Y., 2021b. Cil:Contrastiveinstancelearningframeworkfordistantlysupervised relationextraction,in:Proceedingsofthe59thAnnualMeetingofthe AssociationforComputationalLinguisticsandthe11thInternational Joint Conference on Natural Language Processing (Volume 1: Long Papers), pp. 6191–6200

    work page

  7. [7]

    Improving distantly-supervised relationextractionthroughbert-basedlabelandinstanceembeddings

    Christou, D., Tsoumakas, G., 2021. Improving distantly-supervised relationextractionthroughbert-basedlabelandinstanceembeddings. IEEE Access 9, 62574–62582

    work page 2021

  8. [8]

    mlco2/codecarbon: v2.4.1

    Courty, B., Schmidt, V., Luccioni, S., Goyal-Kamal, et al., 2024. mlco2/codecarbon: v2.4.1

    work page 2024

Show all 48 references
  1. [9]

    Constructing biological knowl- edgebasesbyextractinginformationfromtextsources.,in:ISMB,pp

    Craven, M., Kumlien, J., et al., 1999. Constructing biological knowl- edgebasesbyextractinginformationfromtextsources.,in:ISMB,pp. 77–86

    1999

  2. [10]

    A survey on relation extraction

    Detroja, K., Bhensdadia, C., Bhatt, B.S., 2023. A survey on relation extraction. Intelligent Systems with Applications 19, 200244

    2023

  3. [11]

    Exploitingglobalcontextandexternal knowledge for distantly supervised relation extraction

    Gao,J.,Wan,H.,Lin,Y.,2023. Exploitingglobalcontextandexternal knowledge for distantly supervised relation extraction. Knowledge- Based Systems 261, 110195. URL: https://www.sciencedirect. com/science/article/pii/S0950705122012916,doi: https://doi.org/10. 1016/j.knosys.2022.110195

    2023

  4. [12]

    Manual evaluation matters: Reviewing test protocols of distantly supervised relation extraction, in: Zong, C., Xia, F., Li, W., Navigli, R

    Gao, T., Han, X., Bai, Y., Qiu, K., Xie, Z., Lin, Y., Liu, Z., Li, P., Sun, M., Zhou, J., 2021. Manual evaluation matters: Reviewing test protocols of distantly supervised relation extraction, in: Zong, C., Xia, F., Li, W., Navigli, R. (Eds.), Findings of the Association for C...

    2021

  5. [13]

    Gøttcke, J.M.N., Zimek, A., 2021. Handling class imbalance in k- nearest neighbor classification by balancing prior probabilities, in: Similarity Search and Applications: 14th International Conference, SISAP 2021, Dortmund, Germany, September 29–October 1, 2021, Proceedings 14...

    2021

  6. [14]

    Knowledgegraphs

    Hogan, A., Blomqvist, E., Cochez, M., D’amato, C., Melo, G.D., Gutierrez, C., Kirrane, S., Gayo, J.E.L., Navigli, R., Neumaier, S., Ngomo, A.C.N., Polleres, A., Rashid, S.M., Rula, A., Schmelzeisen, L.,Sequeda,J.,Staab,S.,Zimmermann,A.,2021. Knowledgegraphs. ACM Comput. Surv. 54

    2021

  7. [15]

    An overview of distant supervision for relation extractionwithafocusondenoisingandpre-trainingmethods

    Hogan, W., 2022. An overview of distant supervision for relation extractionwithafocusondenoisingandpre-trainingmethods. arXiv preprint arXiv:2207.08286

    2022 arXiv

  8. [16]

    Hu, Z., Cao, Y., Huang, L., Chua, T.S., 2021. How knowledge graph and attention help? a qualitative analysis into bag-level rela- tion extraction, in: Proceedings of the 59th Annual Meeting of the AssociationforComputationalLinguisticsandthe11thInternational Joint Conference o...

    2021

  9. [17]

    Jia,W.,Dai,D.,Xiao,X.,Wu,H.,2019. Arnor:Attentionregulariza- tion based noise reduction for distant supervision relation classifica- tion,in:Proceedingsofthe57thannualmeetingoftheassociationfor computational linguistics, pp. 1399–1408

    2019

  10. [18]

    Jiang, X., Wang, Q., Li, P., Wang, B., 2016. Relation extraction with multi-instance multi-label convolutional neural networks, in: Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pp. 1471–1480

    2016

  11. [19]

    Supervised contrastive learning

    Khosla, P., Teterwak, P., Wang, C., Sarna, A., Tian, Y., Isola, P., Maschinot, A., Liu, C., Krishnan, D., 2020. Supervised contrastive learning. Advances in neural information processing systems 33, 18661–18673

    2020

  12. [20]

    Hiclre: A hierarchical contrastive learning framework for distantly supervised relationextraction,in:FindingsoftheAssociationforComputational Linguistics: ACL 2022, pp

    Li, D., Zhang, T., Hu, N., Wang, C., He, X., 2022. Hiclre: A hierarchical contrastive learning framework for distantly supervised relationextraction,in:FindingsoftheAssociationforComputational Linguistics: ACL 2022, pp. 2567–2578

    2022

  13. [21]

    Self-distilled transitive instance weighting for denoised distantly supervised relation extraction, in: Findings of the Association for Computational Linguistics: EMNLP 2023, pp

    Lin, X., Jia, W., Gong, Z., 2023. Self-distilled transitive instance weighting for denoised distantly supervised relation extraction, in: Findings of the Association for Computational Linguistics: EMNLP 2023, pp. 168–180

    2023

  14. [22]

    Knowledge graph attention mechanism for distant supervision neural relation extraction

    Liu, M., Zhou, F., He, J., Yan, X., 2022. Knowledge graph attention mechanism for distant supervision neural relation extraction. Knowledge-Based Systems 256, 109800. URL: https: //www.sciencedirect.com/science/article/pii/S0950705122009145, doi:https://doi.org/10.1016/j.knosy...

    2022

  15. [23]

    Distantsupervision for relation extraction without labeled data, in: Proceedings of the Joint Conference of the 47th Annual Meeting of the ACL and the 4th L

    Mintz,M.,Bills,S.,Snow,R.,Jurafsky,D.,2009. Distantsupervision for relation extraction without labeled data, in: Proceedings of the Joint Conference of the 47th Annual Meeting of the ACL and the 4th L. Mariotti et al.:Preprint submitted to Elsevier Page 14 of 15 SCoRE Internat...

    2009

  16. [24]

    Nadgeri,A.,Bastos,A.,Singh,K.,Mulang’,I.O.,Hoffart,J.,Shekar- pour, S., Saraswat, V., 2021. KGPool: Dynamic knowledge graph context selection for relation extraction, in: Zong, C., Xia, F., Li, W.,Navigli,R.(Eds.),FindingsoftheAssociationforComputational Linguistics:ACL-IJCNLP...

    2021

  17. [25]

    Uni- fying Large Language Models and Knowledge Graphs: A Roadmap

    Pan, S., Luo, L., Wang, Y., Chen, C., Wang, J., Wu, X., 2024. Uni- fying Large Language Models and Knowledge Graphs: A Roadmap . IEEE Transactions on Knowledge & Data Engineering 36, 3580– 3599

    2024

  18. [26]

    The ultimate guide to fine-tuning llms from basics to breakthroughs: An exhaustive review of technologies, research, best practices, applied research challenges and opportunities

    Parthasarathy, V.B., Zafar, A., khan, A.I., Shahid, A., 2024. The ultimate guide to fine-tuning llms from basics to breakthroughs: An exhaustive review of technologies, research, best practices, applied research challenges and opportunities. ArXiv abs/2408.13296

    2024 arXiv

  19. [27]

    Knowledge graphs: Opportunities and challenges

    Peng, C., Xia, F., Naseriparsa, M., Osborne, F., 2023. Knowledge graphs: Opportunities and challenges. Artificial Intelligence Review 56, 13071–13102

    2023

  20. [28]

    Peng, H., Gao, T., Han, X., Lin, Y., Li, P., Liu, Z., Sun, M., Zhou, J., 2020. Learning from context or names? an empirical study on neural relationextraction, in: Proceedings of the2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 3661–3672

    2020

  21. [29]

    Rathore, V., Badola, K., Singla, P., Mausam, 2022. PARE: A simpleandstrongbaselineformonolingualandmultilingualdistantly supervised relation extraction, in: Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), Associ...

    2022

  22. [30]

    Ren, F., Zhang, L., Zhao, X., Yin, S., Liu, S., Li, B., 2022. A simple but effective bidirectional framework for relational triple extraction, in: Proceedings of the Fifteenth ACM International Conference on WebSearchandDataMining,AssociationforComputingMachinery, New York, NY...

    2022

  23. [31]

    Riedel, S., Yao, L., McCallum, A., 2010. Modeling relations and theirmentionswithoutlabeledtext,in:Proceedingsofthe2010Euro- pean Conference on Machine Learning and Knowledge Discovery in Databases: Part III, Springer-Verlag, Berlin, Heidelberg. p. 148–163

    2010

  24. [32]

    The principles of deep learning theory

    Roberts, D.A., Yaida, S., Hanin, B., 2022. The principles of deep learning theory. volume 46. Cambridge University Press Cambridge, MA, USA

    2022

  25. [33]

    Match- ing the blanks: Distributional similarity for relation learning, in: Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp

    Soares, L.B., Fitzgerald, N., Ling, J., Kwiatkowski, T., 2019. Match- ing the blanks: Distributional similarity for relation learning, in: Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp. 2895–2905

    2019

  26. [34]

    Sun, X., Liu, Q., Wu, S., Wang, Z., Wang, L., 2023. Noise-robust semi-supervised learning for distantly supervised relation extraction, in:Bouamor,H.,Pino,J.,Bali,K.(Eds.),FindingsoftheAssociation for Computational Linguistics: EMNLP 2023, Association for Com- putational Lingu...

    2023

  27. [35]

    Szép, M., Rueckert, D., von Eisenhart-Rothe, R., Hinterwimmer, F.,

  28. [36]

    Wan, Z., Cheng, F., Liu, Q., Mao, Z., Song, H., Kurohashi, S.,

  29. [37]

    Rescueimplicitandlong-tailcases:Nearestneighborrelationextrac- tion, in: Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pp

    Wan, Z., Liu, Q., Mao, Z., Cheng, F., Kurohashi, S., Li, J., 2022. Rescueimplicitandlong-tailcases:Nearestneighborrelationextrac- tion, in: Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pp. 1731–1738

    2022

  30. [38]

    Wang,R.,Dai,X.,etal.,2022. Contrastivelearning-enhancednearest neighbor mechanism for multi-label text classification, in: Proceed- ingsofthe60thAnnualMeetingoftheAssociationforComputational Linguistics (Volume 2: Short Papers), pp. 672–679

    2022

  31. [39]

    A survey of human-in-the-loop for machine learning

    Wu, X., Xiao, L., Sun, Y., Zhang, J., Ma, T., He, L., 2022. A survey of human-in-the-loop for machine learning. Future Gener. Comput. Syst. 135, 364–381

    2022

  32. [40]

    Investigating why contrastive learning benefits robustness against label noise, in: International Conference on Machine Learning, PMLR

    Xue, Y., Whitecross, K., Mirzasoleiman, B., 2022. Investigating why contrastive learning benefits robustness against label noise, in: International Conference on Machine Learning, PMLR. pp. 24851– 24871

    2022

  33. [41]

    Distantly supervised relation extrac- tion via contextual information interaction and relation embeddings

    Yin, H., Liu, S., Jian, Z., 2023. Distantly supervised relation extrac- tion via contextual information interaction and relation embeddings. Symmetry 15, 1788

    2023

  34. [42]

    Distant supervision for relation extraction with hierarchical attention-based networks

    Zhang, J., Cao, M., 2023. Distant supervision for relation extraction with hierarchical attention-based networks. Expert Syst. Appl. 220

    2023

  35. [43]

    A Bidirectional Extraction-Then-Evaluation Framework for Complex Relation Extraction

    Zhang, W., Wang, J., Chen, C., Lu, W., Du, W., Wang, H., Liu, J., Ruan, T., 2024. A Bidirectional Extraction-Then-Evaluation Framework for Complex Relation Extraction . IEEE Transactions on Knowledge & Data Engineering 36, 7442–7454

    2024

  36. [44]

    A comprehensive survey on relation extraction: Recent advances and new frontiers

    Zhao, X., Deng, Y., Yang, M., Wang, L., Zhang, R., Cheng, H., Lam, W., Shen, Y., Xu, R., 2024. A comprehensive survey on relation extraction: Recent advances and new frontiers. ACM Comput. Surv. 56

    2024

  37. [45]

    ExploringPrivileged Features for Relation Extraction With Contrastive Student-Teacher Learning

    Zhao,X.,Yang,M.,Qu,Q.,Xu,R.,Li,J.,2023. ExploringPrivileged Features for Relation Extraction With Contrastive Student-Teacher Learning . IEEE Transactions on Knowledge & Data Engineering 35, 7953–7965

    2023

  38. [46]

    Distantly supervised relation extraction with kb-enhanced reconstructed latent iterative graph networks

    Zhou, Q., Zhang, Y., Ji, D., 2023. Distantly supervised relation extraction with kb-enhanced reconstructed latent iterative graph networks. Knowledge-Based Systems 260, 110108. URL: https: //www.sciencedirect.com/science/article/pii/S0950705122012047, doi:https://doi.org/10.10...

    2023

  39. [2023]

    2580–2585

    Relation extraction with weighted contrastive pre-training on distantsupervision,in:FindingsoftheAssociationforComputational Linguistics: EACL 2023, pp. 2580–2585

    2023

  40. [2024]

    arXiv preprint arXiv:2411.09539

    A practical guide to fine-tuning language models with limited data. arXiv preprint arXiv:2411.09539

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.