Pith. sign in

REVIEW 4 major objections 6 minor 64 references

Clustering-based Unsupervised Generative Relation Extraction

T0 review · 4 major / 6 minor · reviewed 2026-08-27 · deepseek-v4-flash

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

Pith's one-line read This paper claims that an unsupervised relation extractor trained by predicting one sentence's dependency path from other sentences with the same entity pair clusters relations more accurately than existing methods on news and UN texts.

desk verdict New self-supervised shortest-path prediction for unsupervised relation extraction; real clustering gains, but F1 results rest on an underspecified and possibly biased label-matching step plus an untested single-relation-per-pair assumption. read the letter →

arxiv 2009.12681 v1 pith:PNEC7MJI submitted 2020-09-26 cs.CL cs.LGstat.ML

Chenhan Yuan , Ryan Rossi , Andrew Katz , Hoda Eldardiry This is my paper · ORCID
classification cs.CLcs.LGstat.ML
keywords relationextractionunsupervisedlearningself-supervisedgenerativemodelclusteringknowledgegraphconstructionsemanticshortestpathencoder-decoder
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 CURE, an unsupervised relation extraction framework that learns to represent relations from raw text alone. Its training signal is self-supervised: given several sentences containing the same entity pair, an encoder reads the dependency-graph paths between the two entities and a decoder must generate the path from one of those sentences; the paper's claim is that this forces the encoder to capture the relation without labels, seed rules, or pre-specified features. The paper then clusters entity pairs by these learned relation vectors and automatically labels each cluster with relation words chosen by word-vector similarity. On the New York Times and United Nations Parallel Corpus datasets, CURE is reported to beat three baselines with average F1 gains of 10.47% and 6.59%, and to agree better with Freebase and YAGO partitions on a rand-index measure.

What carries the argument

The central object is the semantic shortest path: the shortest route between the two entity nodes in a sentence's dependency tree, stored as the word, part-of-speech, and dependency-tag sequences along that route. The load-bearing mechanism is the encoder-decoder prediction loop built on it. Because the model is given several paths for the same entity pair and must generate one of them from the summed encodings of the others, the encoder can only solve the task by keeping what the paths share—the relation—and discarding per-sentence wording. That learned relation vector is what feeds the clustering stage, and the same paths supply the candidate words for automatic cluster labels.

What would settle it

Take the NYT test entity pairs that have more than one Freebase relation, such as a person both born in and living in a place, and compute CURE's precision and recall separately on that subset; if the model mixes those relations into one merged cluster, the joint-encoding target is not separating distinct relations. A cleaner check is to rerun the whole pipeline after deleting all multi-relation entity pairs from training: if the average F1 gain over baselines largely disappears when the conflicting targets are gone, the gain is coming from the single-relation cases that the premise handles.

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that repeated mentions of the same entity pair are a usable, label-free teacher. For each sentence containing an entity pair, CURE extracts the semantic shortest path between the two entities in the dependency tree and records its words, part-of-speech tags, and dependency tags. A Bi-LSTM encoder turns each path into a vector; the vectors for all but one sentence are summed into a single relation vector; a GRU decoder with attention must regenerate the held-out path. Optimizing this prediction is what makes the summed vector a relation representation. At extraction time, that representation is clustered with hierarchical agglomerative clustering, and each cluster is labeled by weighting candidate word vectors by frequency and by dissimilarity to other candidates. The paper reports that this pipeline outperforms Rel-LDA, VAE, and Open-RE on both NYT and UNPC, and that its labels align with ground-truth relations better than common-word labels.

Load-bearing premise

The load-bearing premise is that all sentences containing the same entity pair express the same relation, so one sentence's path can serve as the correct target for predicting from the others; when an entity pair has multiple relations, the training signal points the encoder in conflicting directions.

Editorial extensions

If this is right

  • Relation extraction systems can be trained on unlabeled corpora as long as the corpus contains entity pairs that recur in more than one sentence.
  • Encoding sentences jointly per entity pair, rather than treating each sentence independently, improves clustering F1 and rand index over per-sentence generative models and feature-clustering methods.
  • Automatic cluster labels derived from frequency-weighted word vectors describe relations more accurately than most-common-word labels, making the extracted clusters interpretable without manual labeling.
  • The pipeline transfers across genres and scales: the same settings that produce gains on news text also produce gains on a much larger U.N. corpus, so the result is not tied to one dataset.

Reading between the lines

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

  • Extension: the same self-prediction trick need not be limited to entity pairs; any recurring pair of arguments with a dependency path between them (events, concepts, attributes) could provide the same label-free training signal.
  • Extension: the paper's own observation that 'placeBirth' and 'placeLived' clusters overlap predicts where the method will fail; a version that lets one entity pair be assigned multiple relation vectors instead of one summed vector would be a natural next test.
  • Extension: since the semantic shortest path comes from a dependency parser, the size of the reported gains likely depends on parse quality; evaluating CURE on languages or domains with noisier parses would separate the contribution of the training signal from the contribution of the parser.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Request a human review

A listed scientist reviews the paper for a fee and the review publishes here regardless of verdict. See the reviewers or get listed.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes CURE, an unsupervised relation extraction framework. In a first stage, the model trains an encoder-decoder on semantic shortest paths (SSPs) between entity pairs in dependency graphs: for each entity pair, the encoder sums path encodings and the decoder predicts one SSP given the others, as a self-supervised objective. In a second stage, the trained encoder produces relation representations for entity pairs, which are clustered with Hierarchical Agglomerative Clustering, and each cluster is labeled automatically using word2vec cosine similarity among candidate words. The authors evaluate on NYT and UNPC, comparing against Rel-LDA, VAE, and Open-RE, and report per-relation precision, recall, F1, as well as rand index for cluster quality. The central claim is that CURE outperforms state-of-the-art unsupervised relation extraction models on both datasets, with average F1 gains of 10.47% and 6.59%.

Significance. If the reported results are valid, the main contributions are a practical unsupervised RE pipeline that exploits sentence correlations for the same entity pair during training, an automatic cluster-labeling method based on word vector similarity, and evaluation on two corpora of different genres and scales. The rand-index evaluation is a useful complement to F1 because it measures cluster quality independently of label assignment. The work also gives explicit credit to the difficulty of distinguishing relations such as placeBirth and placeLived. However, the reliability of the headline F1 gains depends on resolving the evaluation-protocol ambiguity for label matching and on addressing the tension between the single-relation training assumption and the multi-relation evaluation rule.

major comments (4)
  1. [§4.3, Tables 4–5] The relation-label matching procedure used to compute F1 is not fully specified and may favor CURE. The text says a cluster is labeled by the relation in Freebase that is similar to the most frequent trigger word in that cluster, but for CURE cluster labels come from Equation 8, which selects words by word2vec cosine similarity. If the same word2vec similarity is used to map predicted labels to Freebase relations, CURE's label selection is optimized for the evaluation metric while the baselines' most-frequent-trigger-word labels are not. Please specify the exact matching function (exact string match, WordNet synonymy, or cosine between word2vec(label word) and word2vec(relation name)), apply identical matching to all systems, and report how many test entity pairs are matched by each method. Without this, the reported average F1 gains may be evaluation artifacts rather than relation extraction improvements.
  2. [§3.2 and §4.3] The self-supervised training signal assumes each entity pair expresses a single relation, but the evaluation explicitly allows an entity pair to have two or more Freebase relations. Section 3.2 justifies Equation 1 by stating that all semantic shortest paths of one entity pair possibly share similar relation information; Section 4.3 then scores an entity pair as correct if the prediction matches any one of its relations. For multi-relation entity pairs, the conditioning paths in Equation 4 can express a different relation from the predicted path, so the summed encoding blends distinct relations and the decoder is trained to predict a target that is not determined by its input. The observation that placeBirth and placeLived clusters partially overlap is consistent with this failure mode. Please quantify the fraction of multi-relation entity pairs in the training and test sets, and either exclude them from training or modify the objective and evaluation to handle them. This is load-bearing because the headline gains are averages over such cases.
  3. [§3.7, Table 4, Table 5, Figure 5] The HAC cluster count, or the distance threshold used as a stopping criterion, is not reported. F1 and rand index depend strongly on the number of clusters, and Section 4.4 even notes that CURE subdivides isPoliticianOf into finer-grained subsets such as president or ambassador. Without reporting the selected k or threshold for CURE and for the baselines, and without a sensitivity analysis, the comparison may reflect favorable cluster granularity rather than better relation representations. Please report the exact clustering parameters and show F1 and rand index across a range of cluster counts for each system.
  4. [Tables 4–5, Figure 5] All reported results are single-point estimates without error bars, significance tests, or repeated training runs. The average F1 gains of 10.47% and 6.59% are the central quantitative claim, but on several relations the difference between CURE and the strongest baseline is small (for example, capital 59.3 versus 59.0 on NYT, and hasNeighbor 62.0 versus 61.8 on UNPC). Please provide confidence intervals or paired significance tests, such as bootstrap over test entity pairs or multiple training runs with different seeds, to establish that the observed gains are not within the range of random variation.
minor comments (6)
  1. [§4.3] There is a typo: "dding sentence correlation" should be "adding sentence correlation".
  2. [Equation (7)] The loss function is typeset in a garbled way; the softmax expression is not readable. Please rewrite it with standard notation and clarify whether the average is over path length n and batch size m as intended.
  3. [§3.3] The dependency relation referred to as "subjective" is likely "nsubj"; please align the terminology with the dependency parser's tag set.
  4. [§1, Contributions] The bullet list says "standard NTY data set" but the corpus is elsewhere called NYT; please fix the abbreviation.
  5. [Table 6] The CW rows run words together (for example, "capitalCW city states help"); please add formatting so that the top-three labels are clearly separated, and explain how the top-three words are selected from the full candidate list.
  6. [§4.2 and §3.4] Several implementation details needed for reproducibility are not reported, including the Word2Vec model and embedding dimension, LSTM and GRU hidden sizes, and the HAC linkage criterion; please add these to the experimental setup.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the self-supervised target is a held-out sentence path, and evaluation uses external Freebase/YAGO relations.

full rationale

The claimed derivation chain is self-contained. The relation extractor is trained with the objective in Eq. 1: given semantic shortest paths P_a^1 ... P_a^{u-1} of an entity pair, predict P_a^u. The target path is held out from the encoder input, so this is a legitimate masked/self-supervised task rather than prediction of a fitted value; the training signal comes from co-occurrence of paths for the same entity pair, not from the Freebase/YAGO relations used only at evaluation time. Clustering (HAC on encoder outputs) and label selection (Eq. 8) do not use ground-truth relation labels. The F1 comparison maps predicted cluster-label words to Freebase relations via a similarity criterion applied to all methods, and the rand-index evaluation in Section 4.5 is label-independent; while the exact 'similar' mapping is underspecified and the single-relation-per-entity-pair assumption is a genuine correctness risk (the paper itself notes placeBirth/placeLived cluster overlap), neither reduces the reported gains to the method's own inputs by construction. There are no load-bearing self-citations or imported uniqueness claims, so the circularity score is 0.

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

The method relies on standard ML architecture components and a few domain assumptions. The most consequential free parameter is the HAC cluster count, which is unreported and could be tuned to the gold relation count. All core assumptions are stated or cited, but none are independently verified in this paper.

free parameters (4)
  • HAC cluster count / distance threshold = not reported
    The number of clusters k for hierarchical agglomerative clustering must be specified. The paper does not say how k is chosen; if k equals the number of gold relations in the test set, it is tuned to the evaluation.
  • Encoder and decoder hidden sizes = not reported
    Dimensions nh and nh' of the Bi-LSTM and GRU are not given; a reimplementation must guess them.
  • Embedding dimensions for words, POS, dependency tags = not reported
    The paper says word embeddings have more dimensions than tag embeddings but does not specify values.
  • Word2Vec model and dimension = not specified
    Label selection in Equation 8 depends on a pre-trained Word2Vec model; the corpus and dimension are not disclosed.
assumptions (4)
  • domain assumption All sentences containing the same entity pair share the same relation.
    Section 3.2 uses this to justify predicting one path from others; this is often false for polysemous entity pairs.
  • domain assumption The semantic shortest path between two entities in a dependency tree is sufficient to represent the relation.
    Section 3.3 cites Bunescu and Mooney 2005; the framework encodes only these paths.
  • domain assumption The dependency parser produces correct dependency relations and representative entity nodes.
    All paths are derived from parser output; errors propagate into the encoding and clustering.
  • domain assumption Word2Vec cosine similarity is a meaningful measure for selecting relation words.
    Equation 8 uses Word2Vec to score words; this is an external embedding assumption not verified in the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Clustering-based Unsupervised Generative Relation Extraction." pith.science (2026). https://pith.science/paper/PNEC7MJI

@misc{pith2026200912681,
  author       = {Pith},
  title        = {Pith review of: Clustering-based Unsupervised Generative Relation Extraction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PNEC7MJI}},
  note         = {Machine review of arXiv:2009.12681}
}
read the original abstract

This paper focuses on the problem of unsupervised relation extraction. Existing probabilistic generative model-based relation extraction methods work by extracting sentence features and using these features as inputs to train a generative model. This model is then used to cluster similar relations. However, these methods do not consider correlations between sentences with the same entity pair during training, which can negatively impact model performance. To address this issue, we propose a Clustering-based Unsupervised generative Relation Extraction (CURE) framework that leverages an "Encoder-Decoder" architecture to perform self-supervised learning so the encoder can extract relation information. Given multiple sentences with the same entity pair as inputs, self-supervised learning is deployed by predicting the shortest path between entity pairs on the dependency graph of one of the sentences. After that, we extract the relation information using the well-trained encoder. Then, entity pairs that share the same relation are clustered based on their corresponding relation information. Each cluster is labeled with a few words based on the words in the shortest paths corresponding to the entity pairs in each cluster. These cluster labels also describe the meaning of these relation clusters. We compare the triplets extracted by our proposed framework (CURE) and baseline methods with a ground-truth Knowledge Base. Experimental results show that our model performs better than state-of-the-art models on both New York Times (NYT) and United Nations Parallel Corpus (UNPC) standard datasets.

Figures

Figures reproduced from arXiv: 2009.12681 by the authors.

Figure 1
Figure 1. The architecture of relation extractor training stage of CURE [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. The triplets clustering stage of CURE attention weights to incorporate encoding information. hi = дru(hi−1,qi−1 ) qi−1 = attnβ  attnα (hi−1) ⊗ EI ⊕ qi−2  = Wβ  (Wα ⊗ hi−1 + bα  ⊗ EI) ⊕ qi−2  (6) where hi is the output of the i-th GRU unit, which is the predicted probability distribution of the word at that position. qi−1 is the input of the GRU and the weighted information of the previous state and the encod… view at source ↗
Figure 3
Figure 3. % F-1 gain of CURE over baselines on NYT [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: % F-1 gain of CURE over baselines on UNPC [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Rand Index score of CURE and baselines [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

64 extracted references · 51 canonical work pages

  1. [1]

    Gabor Angeli, Melvin Jose Johnson Premkumar, and Christopher D Manning

  2. [2]

    Sören Auer, Christian Bizer, Georgi Kobilarov, Jens Lehmann, Richard Cyganiak, and Zachary Ives. 2007. Dbpedia: A nucleus for a web of open data. In The semantic web. Springer, 722–735

    work page 2007

  3. [3]

    Michele Banko and Oren Etzioni. 2008. The tradeoffs between open and tradi- tional relation extraction. In Proceedings of ACL-08: HLT . 28–36

    work page 2008

  4. [4]

    David M Blei, Andrew Y Ng, and Michael I Jordan. 2003. Latent dirichlet allocation. Journal of machine Learning research 3, Jan (2003), 993–1022

    2003

  5. [5]

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

  6. [6]

    Denny Britz, Anna Goldie, Minh-Thang Luong, and Quoc Le. 2017. Massive Exploration of Neural Machine Translation Architectures. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing. 1442–1451

    work page 2017

  7. [7]

    Razvan Bunescu and Raymond Mooney. 2007. Learning to extract relations from the web using minimal supervision. In Proceedings of the 45th Annual Meeting of the Association of Computational Linguistics . 576–583

    work page 2007

  8. [8]

    Razvan C Bunescu and Raymond J Mooney. 2005. A shortest path dependency kernel for relation extraction. In Proceedings of the conference on human language technology and empirical methods in natural language processing . Association for Computational Linguistics, 724–731

    work page 2005

Show all 64 references
  1. [9]

    Yee Seng Chan and Dan Roth. 2011. Exploiting syntactico-semantic structures for relation extraction. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies-Volume 1. Association for Computational Linguistics, 551–560

    2011

  2. [10]

    Jinxiu Chen, Donghong Ji, Chew Lim Tan, and Zheng-Yu Niu. 2005. Unsupervised feature selection for relation extraction. In Companion Volume to the Proceedings of Conference including Posters/Demos and tutorial abstracts

    2005

  3. [11]

    Mark Craven, Johan Kumlien, et al. 1999. Constructing biological knowledge bases by extracting information from text sources.. In ISMB, Vol. 1999. 77–86. 9

    1999

  4. [12]

    Oier Lopez De Lacalle and Mirella Lapata. 2013. Unsupervised relation extraction with general domain knowledge. InProceedings of the 2013 Conference on Empirical Methods in Natural Language Processing . 415–425

    2013

  5. [13]

    Hady Elsahar, Elena Demidova, Simon Gottschalk, Christophe Gravier, and Fred- erique Laforest. 2017. Unsupervised open relation extraction. In European Se- mantic Web Conference. Springer, 12–16

    2017

  6. [14]

    Oren Etzioni, Michele Banko, Stephen Soderland, and Daniel S Weld. 2008. Open information extraction from the web. Commun. ACM 51, 12 (2008), 68–74

    2008

  7. [15]

    Oren Etzioni, Michael Cafarella, Doug Downey, Ana-Maria Popescu, Tal Shaked, Stephen Soderland, Daniel S Weld, and Alexander Yates. 2005. Unsupervised named-entity extraction from the web: An experimental study. Artificial intelli- gence 165, 1 (2005), 91–134

    2005

  8. [16]

    Anthony Fader, Stephen Soderland, and Oren Etzioni. 2011. Identifying relations for open information extraction. In Proceedings of the conference on empirical methods in natural language processing . Association for Computational Linguis- tics, 1535–1545

    2011

  9. [17]

    Zhou GuoDong, Su Jian, Zhang Jie, and Zhang Min. 2005. Exploring various knowledge in relation extraction. In Proceedings of the 43rd annual meeting on as- sociation for computational linguistics . Association for Computational Linguistics, 427–434

    2005

  10. [18]

    Takaaki Hasegawa, Satoshi Sekine, and Ralph Grishman. 2004. Discovering relations among named entities from large corpora. In Proceedings of the 42nd annual meeting on association for computational linguistics . Association for Com- putational Linguistics, 415

    2004

  11. [19]

    Philipp Heim, Sebastian Hellmann, Jens Lehmann, Steffen Lohmann, and Timo Stegemann. 2009. RelFinder: Revealing relationships in RDF knowledge bases. In International Conference on Semantic and Digital Media Technologies . Springer, 182–187

    2009

  12. [20]

    Sepp Hochreiter. 1998. The vanishing gradient problem during learning recurrent neural nets and problem solutions. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems 6, 02 (1998), 107–116

    1998

  13. [21]

    Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory.Neural computation 9, 8 (1997), 1735–1780

    1997

  14. [22]

    Raphael Hoffmann, Congle Zhang, and Daniel S Weld. 2010. Learning 5000 relational extractors. In Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics. 286–295

    2010

  15. [23]

    Scott B Huffman. 1995. Learning information extraction patterns from examples. In International Joint Conference on Artificial Intelligence . Springer, 246–260

    1995

  16. [24]

    Jing Jiang and ChengXiang Zhai. 2007. A systematic exploration of the fea- ture space for relation extraction. In Human Language Technologies 2007: The Conference of the North American Chapter of the Association for Computational Linguistics; Proceedings of the Main Conference...

    2007

  17. [25]

    Natthawut Kertkeidkachorn and Ryutaro Ichise. 2017. T2KG: An end-to-end system for creating knowledge graph from unstructured text. In Workshops at the Thirty-First AAAI Conference on Artificial Intelligence

    2017

  18. [26]

    Moldovan

    Jun-Tae Kim and Dan I. Moldovan. 1995. Acquisition of linguistic patterns for knowledge-based information extraction. IEEE transactions on knowledge and data engineering 7, 5 (1995), 713–724

    1995

  19. [27]

    Natalia Konstantinova. 2014. Review of Relation Extraction Methods: What Is New Out There?. In Analysis of Images, Social Networks and Texts , Dmitry I. Ignatov, Mikhail Yu. Khachay, Alexander Panchenko, Natalia Konstantinova, and Rostislav E. Yavorsky (Eds.). Springer Interna...

    2014

  20. [28]

    Sebastian Krause, Hong Li, Hans Uszkoreit, and Feiyu Xu. 2012. Large-scale learning of relation-extraction rules with distant supervision from the web. In International Semantic Web Conference. Springer, 263–278

    2012

  21. [29]

    ChunYang Liu, WenBo Sun, WenHan Chao, and Wanxiang Che. 2013. Con- volution neural network for relation extraction. In International Conference on Advanced Data Mining and Applications . Springer, 231–242

    2013

  22. [30]

    Yang Liu, Furu Wei, Sujian Li, Heng Ji, Ming Zhou, and Houfeng Wang. 2015. A dependency-based neural network for relation classification. In53rd Annual Meet- ing of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Pro...

    2015

  23. [31]

    Diego Marcheggiani and Ivan Titov. 2016. Discrete-state variational autoencoders for joint discovery and factorization of relations. Transactions of the Association for Computational Linguistics 4 (2016), 231–244

    2016

  24. [32]

    Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013. Efficient estimation of word representations in vector space.arXiv preprint arXiv:1301.3781 (2013)

    2013 arXiv

  25. [33]

    Mike Mintz, Steven Bills, Rion Snow, and Dan Jurafsky. 2009. Distant supervision for relation extraction without labeled data. In Proceedings of the Joint Conference of the 47th Annual Meeting of the ACL and the 4th International Joint Conference on Natural Language Processing...

    2009

  26. [34]

    Ndapandula Nakashole, Gerhard Weikum, and Fabian Suchanek. 2012. PATTY: a taxonomy of relational patterns with semantic types. In Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Com- putational Natural Language Learning. Associa...

    2012

  27. [35]

    Dat PT Nguyen, Yutaka Matsuo, and Mitsuru Ishizuka. 2007. Exploiting syntac- tic and semantic information for relation extraction from wikipedia. In IJCAI Workshop on Text-Mining & Link-Analysis (TextLink 2007)

    2007

  28. [36]

    Dat PT Nguyen, Yutaka Matsuo, and Mitsuru Ishizuka. 2007. Relation extraction from wikipedia using subtree mining. In Proceedings of the National Conference on Artificial Intelligence, Vol. 22. Menlo Park, CA; Cambridge, MA; London; AAAI Press; MIT Press; 1999, 1414

    2007

  29. [37]

    Truc-Vien T Nguyen and Alessandro Moschitti. 2011. End-to-end relation extrac- tion using distant supervision from external semantic repositories. InProceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies: short pape...

    2011

  30. [38]

    Hoifung Poon and Pedro Domingos. 2009. Unsupervised semantic parsing. In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: Volume 1-Volume 1. Association for Computational Linguistics, 1–10

    2009

  31. [39]

    Sebastian Riedel, Limin Yao, and Andrew McCallum. 2010. Modeling relations and their mentions without labeled text. InJoint European Conference on Machine Learning and Knowledge Discovery in Databases . Springer, 148–163

    2010

  32. [40]

    Lorenza Romano, Milen Kouylekov, Idan Szpektor, Ido Dagan, and Alberto Lavelli

  33. [41]

    Benjamin Rosenfeld and Ronen Feldman. 2006. Ures: an unsupervised web relation extraction system. In Proceedings of the COLING/ACL on Main conference poster sessions. Association for Computational Linguistics, 667–674

    2006

  34. [42]

    Binjamin Rozenfeld and Ronen Feldman. 2006. High-performance unsupervised relation extraction from large corpora. In Sixth International Conference on Data Mining (ICDM’06). IEEE, 1032–1037

    2006

  35. [43]

    Sunita Sarawagi and William W Cohen. 2005. Semi-markov conditional random fields for information extraction. In Advances in neural information processing systems. 1185–1192

    2005

  36. [44]

    Stephen Soderland, David Fisher, Jonathan Aseltine, and Wendy Lehnert. 1995. CRYSTAL: Inducing a conceptual dictionary. arXiv preprint cmp-lg/9505020 (1995)

    1995 arXiv

  37. [45]

    Fabian M Suchanek, Gjergji Kasneci, and Gerhard Weikum. 2007. Yago: a core of semantic knowledge. In Proceedings of the 16th international conference on World Wide Web. 697–706

    2007

  38. [46]

    Shingo Takamatsu, Issei Sato, and Hiroshi Nakagawa. 2012. Reducing wrong labels in distant supervision for relation extraction. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics: Long Papers- Volume 1. Association for Computational Ling...

    2012

  39. [47]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in neural information processing systems . 5998–6008

    2017

  40. [48]

    Denny Vrandečić and Markus Krötzsch. 2014. Wikidata: a free collaborative knowledgebase. Commun. ACM 57, 10 (2014), 78–85

    2014

  41. [49]

    Xiang Wang, Dingxian Wang, Canran Xu, Xiangnan He, Yixin Cao, and Tat-Seng Chua. 2019. Explainable reasoning over knowledge graphs for recommendation. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 33. 5329–5336

    2019

  42. [50]

    Daniel S Weld, Raphael Hoffmann, and Fei Wu. 2009. Using wikipedia to bootstrap open information extraction. Acm Sigmod Record 37, 4 (2009), 62–68

    2009

  43. [51]

    Fei Wu and Daniel S Weld. 2010. Open information extraction using Wikipedia. In Proceedings of the 48th annual meeting of the association for computational linguistics. Association for Computational Linguistics, 118–127

    2010

  44. [52]

    Chenyan Xiong, Russell Power, and Jamie Callan. 2017. Explicit semantic rank- ing for academic search via knowledge graph embedding. In Proceedings of the 26th international conference on world wide web . International World Wide Web Conferences Steering Committee, 1271–1279

    2017

  45. [53]

    Yulan Yan, Naoaki Okazaki, Yutaka Matsuo, Zhenglu Yang, and Mitsuru Ishizuka

  46. [54]

    Limin Yao, Aria Haghighi, Sebastian Riedel, and Andrew McCallum. 2011. Struc- tured relation discovery using generative models. InProceedings of the Conference on Empirical Methods in Natural Language Processing . Association for Computa- tional Linguistics, 1456–1466

    2011

  47. [55]

    Alexander Yates, Michele Banko, Matthew Broadhead, Michael J Cafarella, Oren Etzioni, and Stephen Soderland. 2007. Textrunner: open information extraction on the web. In Proceedings of Human Language Technologies: The Annual Conference of the North American Chapter of the Asso...

    2007

  48. [56]

    Daojian Zeng, Kang Liu, Siwei Lai, Guangyou Zhou, Jun Zhao, et al. 2014. Relation classification via convolutional deep neural network. (2014). 10

    2014

  49. [57]

    Dongxu Zhang and Dong Wang. 2015. Relation classification via recurrent neural network. arXiv preprint arXiv:1508.01006 (2015)

    2015 arXiv

  50. [58]

    Yuyu Zhang, Hanjun Dai, Zornitsa Kozareva, Alexander J Smola, and Le Song

  51. [59]

    Michał Ziemski, Marcin Junczys-Dowmunt, and Bruno Pouliquen. 2016. The united nations parallel corpus v1. 0. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC’16) . 3530–3534. 11

    2016

  52. [2006]

    In 11th Conference of the European Chapter of the Association for Computational Linguistics

    Investigating a generic paraphrase-based approach for relation extraction. In 11th Conference of the European Chapter of the Association for Computational Linguistics

  53. [2008]

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

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

    2008

  54. [2009]

    Unsupervised relation extraction by mining wikipedia texts using infor- mation from the web. In Proceedings of the Joint Conference of the 47th Annual Meeting of the ACL and the 4th International Joint Conference on Natural Lan- guage Processing of the AFNLP: Volume 2-Volume 2...

  55. [2015]

    Leveraging linguistic structure for open domain information extraction. In Proceedings of the 53rd Annual Meeting of the Association for Computational Lin- guistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 344–354

  56. [2018]

    In Thirty-Second AAAI Conference on Artificial Intelligence

    Variational reasoning for question answering with knowledge graph. In Thirty-Second AAAI Conference on Artificial Intelligence

Pith tools

Reviewed August 27, 2026 · model on record in the stance chip above.