pith. sign in

arxiv: 1906.11180 · v1 · pith:H6INWO73new · submitted 2019-06-26 · 💻 cs.AI · cs.CL

Canonicalizing Knowledge Base Literals

Jiaoyan Chen , Ernesto Jimenez-Ruiz , Ian Horrocks This is my paper

Pith reviewed 2026-05-25 15:22 UTC · model grok-4.3

classification 💻 cs.AI cs.CL
keywords knowledge baseliteral canonicalizationsemantic typingentity matchingontology reasoningmachine learningdata quality
0
0 comments X

The pith

A framework combining reasoning and machine learning can replace string literals in ontology-based knowledge bases with semantically typed entities.

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

The paper focuses on the problem of string literals appearing in knowledge bases where typed entities would be more appropriate. It introduces a framework that uses logical reasoning over the ontology together with machine learning to decide on suitable entities or new typed entities for each literal. The approach is evaluated on semantic typing and entity matching tasks against existing baselines. A sympathetic reader would care because successful canonicalization would make knowledge bases more consistent and easier to query or integrate. The work treats the combination of the two techniques as the key to handling cases where either method alone falls short.

Core claim

The paper claims that a framework integrating reasoning and machine learning outperforms state-of-the-art baselines when predicting relevant entities and types for the task of canonicalizing literals, that is, replacing string literals with existing KB entities or new entities typed by KB classes.

What carries the argument

The hybrid framework that applies reasoning over ontological structure alongside machine learning models to predict entities and types.

If this is right

  • String literals in the knowledge base can be systematically replaced by either existing entities or newly created typed entities.
  • Semantic typing accuracy improves when reasoning constraints guide the machine learning predictions.
  • Entity matching benefits from the same joint use of logical and statistical signals.
  • Overall knowledge base consistency rises without manual intervention on literal data.

Where Pith is reading between the lines

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

  • The same hybrid pattern could be tested on other data-quality tasks such as relation completion or class assertion checking.
  • Performance gains might diminish on knowledge bases that lack rich class hierarchies or property constraints.
  • Extending the framework with additional reasoning services like consistency checking could further reduce errors in type assignment.

Load-bearing premise

That the knowledge base supplies enough ontological structure for reasoning to combine usefully with machine learning predictions.

What would settle it

An experiment on the same evaluation datasets for semantic typing and entity matching in which the hybrid framework shows no improvement over the baselines would falsify the central claim.

Figures

Figures reproduced from arXiv: 1906.11180 by Ernesto Jimenez-Ruiz, Ian Horrocks, Jiaoyan Chen.

Figure 1
Figure 1. Figure 1: The technical framework for KB literal canonicalization. Candidate class extraction Popular KBs like DBpedia often contain a large number of classes. For efficiency reasons, and to reduce noise in the learning process, we first identify a subset of candidate classes. This selection should be rather inclusive so as to maximize potential recall. In order to achieve this we pool the candidate classes for all … view at source ↗
Figure 2
Figure 2. Figure 2: The architecture of the neural network. The semantics of forward and backward surrounding words is effective in predicting a word’s semantics. For example, “Port” and “Meadow” are more likely to indicate a place as they appear after “Area” and before “Oxford”. To embed such contextual semantics into a feature vector, we stack a layer composed of bidirectional Recurrent Neural Networks (BiRNNs) with Gated R… view at source ↗
Figure 3
Figure 3. Figure 3: (P)recision, (R)ecall and (F1) Score of independent (I) and hierarchical (H) typing for S-Lite, with the scores predicted by the fine tuned AttBiRNN. Sample Refinement [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: [Left] Performance improvement (%) by sample refinement; [Right] Ratio (%) of added (deleted) positive (negative) particular sample per classifier during sample refinement. Entity-Lookup retrieves one or several entities using the whole phrase of the literal, and uses their classes and super classes as the types. Gunaratna [14] matches the literal’s focus term (head word) to an exact class, then an exact e… view at source ↗
read the original abstract

Ontology-based knowledge bases (KBs) like DBpedia are very valuable resources, but their usefulness and usability is limited by various quality issues. One such issue is the use of string literals instead of semantically typed entities. In this paper we study the automated canonicalization of such literals, i.e., replacing the literal with an existing entity from the KB or with a new entity that is typed using classes from the KB. We propose a framework that combines both reasoning and machine learning in order to predict the relevant entities and types, and we evaluate this framework against state-of-the-art baselines for both semantic typing and entity matching.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper studies the problem of canonicalizing string literals in ontology-based KBs (e.g., DBpedia) by replacing them with existing entities or new entities typed from the KB's classes. It proposes a hybrid framework that combines reasoning (leveraging class hierarchies, property domains, etc.) with machine learning to predict relevant entities and types, and evaluates the framework against state-of-the-art baselines on semantic typing and entity matching tasks.

Significance. If the hybrid method demonstrably outperforms pure-ML baselines and the reasoning component is shown to contribute, the work would provide a concrete, reproducible method for improving KB quality on a common literal-typing issue. The combination of symbolic reasoning with ML is a timely direction, but its value hinges on evidence that the ontological structure in the chosen KBs is rich enough for reasoning to add signal.

major comments (2)
  1. [Abstract (and evaluation sections)] The central claim that the hybrid reasoning+ML framework outperforms baselines rests on the untested assumption that the chosen KBs supply sufficient ontological structure (class hierarchies, domains/ranges) for the reasoning component to improve predictions. No ablation isolating the reasoning module, no quantification of ontology richness, and no analysis of cases where reasoning fails to apply are provided, so performance gains cannot be attributed to the hybrid design rather than the ML component alone.
  2. [Abstract] The abstract states that the framework is evaluated against SOTA baselines for semantic typing and entity matching, yet supplies no metrics, datasets, or results. Without these details it is impossible to verify whether the hybrid approach yields statistically significant gains or whether the evaluation design controls for the contribution of reasoning.
minor comments (1)
  1. [Abstract] Notation for the integration of reasoning and ML (e.g., how logical inferences are encoded as features or constraints) is not introduced in the abstract and should be clarified early.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful comments. We address the major points below and will revise the manuscript accordingly to strengthen the attribution of gains to the hybrid design.

read point-by-point responses

  1. Referee: [Abstract (and evaluation sections)] The central claim that the hybrid reasoning+ML framework outperforms baselines rests on the untested assumption that the chosen KBs supply sufficient ontological structure (class hierarchies, domains/ranges) for the reasoning component to improve predictions. No ablation isolating the reasoning module, no quantification of ontology richness, and no analysis of cases where reasoning fails to apply are provided, so performance gains cannot be attributed to the hybrid design rather than the ML component alone.

    Authors: We agree that an explicit ablation isolating the reasoning module, quantification of ontology richness (e.g., number of classes with hierarchies and properties with domains/ranges), and analysis of reasoning failure cases would strengthen the paper. In the revision we will add these elements to the evaluation sections so that performance gains can be more clearly attributed to the hybrid approach. revision: yes

  2. Referee: [Abstract] The abstract states that the framework is evaluated against SOTA baselines for semantic typing and entity matching, yet supplies no metrics, datasets, or results. Without these details it is impossible to verify whether the hybrid approach yields statistically significant gains or whether the evaluation design controls for the contribution of reasoning.

    Authors: Abstracts are space-constrained summaries; specific metrics and statistical details are presented in the evaluation sections. To improve clarity we will revise the abstract to name the datasets and report the main performance deltas while preserving brevity. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical hybrid framework evaluated against external baselines

full rationale

The paper proposes a hybrid reasoning+ML framework for literal canonicalization and reports performance against state-of-the-art baselines on semantic typing and entity matching. No equations, fitted parameters renamed as predictions, self-definitional steps, or load-bearing self-citations appear in the provided text. The central claim is an empirical superiority result that remains falsifiable by external benchmarks and does not reduce to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based on the abstract alone, no free parameters, axioms, or invented entities are specified.

pith-pipeline@v0.9.0 · 5621 in / 994 out tokens · 30012 ms · 2026-05-25T15:22:25.411151+00:00 · methodology

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

31 extracted references · 31 canonical work pages · 1 internal anchor

  1. [1]

    In: Extended semantic web conference

    Abedjan, Z., Naumann, F.: Synonym analysis for predicate expansion. In: Extended semantic web conference. pp. 140–154. Springer (2013)

    work page 2013

  2. [2]

    In: The semantic web, pp

    Auer, S., Bizer, C., Kobilarov, G., Lehmann, J., Cyganiak, R., Ives, Z.: Dbpedia: A nucleus for a web of open data. In: The semantic web, pp. 722–735. Springer (2007)

    work page 2007

  3. [3]

    In: International Semantic Web Conference

    Auer, S., Lehmann, J., Hellmann, S.: Linkedgeodata: Adding a spatial dimension to the web of data. In: International Semantic Web Conference. pp. 731–746. Springer (2009)

    work page 2009

  4. [4]

    In: AAAI (2019)

    Chen, J., Jimenez-Ruiz, E., Horrocks, I., Sutton, C.: Colnet: Embedding the semantics of web tables for column type prediction. In: AAAI (2019)

    work page 2019

  5. [5]

    In: Proceedings of the 2014 Conference on Empirical Methods in Natural Lan- guage Processing

    Cho, K., van Merrienboer, B., Gulcehre, C., Bahdanau, D., Bougares, F., Schwenk, H., Ben- gio, Y .: Learning phrase representations using rnn encoder–decoder for statistical machine translation. In: Proceedings of the 2014 Conference on Empirical Methods in Natural Lan- guage Processing. pp. 1724–1734 (2014)

    work page 2014

  6. [6]

    In: European semantic web conference

    Debattista, J., Londo ˜no, S., Lange, C., Auer, S.: Quality assessment of linked datasets using probabilistic approximation. In: European semantic web conference. pp. 221–236. Springer (2015)

    work page 2015

  7. [7]

    In: International Semantic Web Conference

    Dimou, A., Kontokostas, D., Freudenberg, M., Verborgh, R., Lehmann, J., Mannens, E., Hellmann, S., Van de Walle, R.: Assessing and refining mappingsto rdf to improve dataset quality. In: International Semantic Web Conference. pp. 133–149. Springer (2015)

    work page 2015

  8. [8]

    In: International Conference on Web Information Systems Engineering

    Dongo, I., Cardinale, Y ., Al-Khalil, F., Chbeir, R.: Semantic web datatype inference: Towards better rdf matching. In: International Conference on Web Information Systems Engineering. pp. 57–74 (2017)

    work page 2017

  9. [9]

    In: International Semantic Web Conference

    Efthymiou, V ., Hassanzadeh, O., Rodriguez-Muro, M., Christophides, V .: Matching web ta- bles with knowledge base entities: from entity lookups to entity embeddings. In: International Semantic Web Conference. pp. 260–277. Springer (2017)

    work page 2017

  10. [10]

    Semantic Web 9(1), 77–129 (2018)

    F ¨arber, M., Bartscherer, F., Menne, C., Rettinger, A.: Linked data quality of dbpedia, free- base, opencyc, wikidata, and yago. Semantic Web 9(1), 77–129 (2018)

    work page 2018

  11. [11]

    In: International Semantic Web Confer- ence

    Fleischhacker, D., Paulheim, H., Bryl, V ., V¨olker, J., Bizer, C.: Detecting errors in numerical linked data using cross-checked outlier detection. In: International Semantic Web Confer- ence. pp. 357–372 (2014)

    work page 2014

  12. [12]

    In: Proceedings of the 23rd ACM International Conference on Conference on Information and Knowledge Management

    Gal ´arraga, L., Heitz, G., Murphy, K., Suchanek, F.M.: Canonicalizing open knowledge bases. In: Proceedings of the 23rd ACM International Conference on Conference on Information and Knowledge Management. pp. 1679–1688 (2014)

    work page 2014

  13. [13]

    In: International Semantic Web Conference

    Gangemi, A., Nuzzolese, A.G., Presutti, V ., Draicchio, F., Musetti, A., Ciancarini, P.: Au- tomatic typing of dbpedia entities. In: International Semantic Web Conference. pp. 65–81. Springer (2012) Canonicalizing Knowledge Base Literals 17

    work page 2012

  14. [14]

    In: European Semantic Web Conference

    Gunaratna, K., Thirunarayan, K., Sheth, A., Cheng, G.: Gleaning types for literals in rdf triples with application to entity summarization. In: European Semantic Web Conference. pp. 85–100 (2016)

    work page 2016

  15. [15]

    In: Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing

    Kartsaklis, D., Pilehvar, M.T., Collier, N.: Mapping text to knowledge graph entities using multi-sense lstms. In: Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. pp. 1959–1970 (2018)

    work page 2018

  16. [16]

    In: Proceedings of the 23rd international conference on World Wide Web

    Kontokostas, D., Westphal, P., Auer, S., Hellmann, S., Lehmann, J., Cornelissen, R., Zaveri, A.: Test-driven evaluation of linked data quality. In: Proceedings of the 23rd international conference on World Wide Web. pp. 747–758. ACM (2014)

    work page 2014

  17. [17]

    In: International Semantic Web Conference

    Krompaß, D., Baier, S., Tresp, V .: Type-constrained representation learning in knowledge graphs. In: International Semantic Web Conference. pp. 640–655. Springer (2015)

    work page 2015

  18. [18]

    In: AAAI

    Luo, X., Luo, K., Chen, X., Zhu, K.Q.: Cross-lingual entity linking for web tables. In: AAAI. pp. 362–369 (2018)

    work page 2018

  19. [19]

    In: Proceedings of the 7th international conference on semantic systems

    Mendes, P.N., Jakob, M., Garc´ıa-Silva, A., Bizer, C.: Dbpedia spotlight: shedding light on the web of documents. In: Proceedings of the 7th international conference on semantic systems. pp. 1–8. ACM (2011)

    work page 2011

  20. [20]

    Efficient Estimation of Word Representations in Vector Space

    Mikolov, T., Chen, K., Corrado, G., Dean, J.: Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781 (2013)

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  21. [21]

    me-weaving chinese linking open data

    Niu, X., Sun, X., Wang, H., Rong, S., Qi, G., Yu, Y .: Zhishi. me-weaving chinese linking open data. In: International Semantic Web Conference. pp. 205–220. Springer (2011)

    work page 2011

  22. [22]

    Semantic web 8(3), 489–508 (2017)

    Paulheim, H.: Knowledge graph refinement: A survey of approaches and evaluation methods. Semantic web 8(3), 489–508 (2017)

    work page 2017

  23. [23]

    In: International semantic web conference

    Paulheim, H., Bizer, C.: Type inference on noisy rdf data. In: International semantic web conference. pp. 510–525 (2013)

    work page 2013

  24. [24]

    In: International Semantic Web Conference

    Paulheim, H., Gangemi, A.: Serving dbpedia with dolce–more than just adding a cherry on top. In: International Semantic Web Conference. pp. 180–196. Springer (2015)

    work page 2015

  25. [25]

    In: International Semantic Web Conference

    Pujara, J., Miao, H., Getoor, L., Cohen, W.: Knowledge graph identification. In: International Semantic Web Conference. pp. 542–557. Springer (2013)

    work page 2013

  26. [26]

    In: International Semantic Web Conference

    Raad, J., Beek, W., Van Harmelen, F., Pernelle, N., Sa ¨ıs, F.: Detecting erroneous identity links on the web using network metrics. In: International Semantic Web Conference. pp. 391–407. Springer (2018)

    work page 2018

  27. [27]

    Data Mining and Knowledge Discovery 22(1-2), 31–72 (2011)

    Silla, C.N., Freitas, A.A.: A survey of hierarchical classification across different application domains. Data Mining and Knowledge Discovery 22(1-2), 31–72 (2011)

    work page 2011

  28. [28]

    AI Magazine 36(1), 75–86 (2015)

    Sleeman, J., Finin, T., Joshi, A.: Entity type recognition for heterogeneous semantic graphs. AI Magazine 36(1), 75–86 (2015)

    work page 2015

  29. [29]

    In: Proceedings of the 2018 World Wide Web Conference on World Wide Web

    Vashishth, S., Jain, P., Talukdar, P.: Cesi: Canonicalizing open knowledge bases using em- beddings and side information. In: Proceedings of the 2018 World Wide Web Conference on World Wide Web. pp. 1317–1327 (2018)

    work page 2018

  30. [30]

    In: Proceedings of the 27th ACM International Conference on Information and Knowledge Management

    Wu, T.H., Wu, Z., Kao, B., Yin, P.: Towards practical open knowledge base canonicalization. In: Proceedings of the 27th ACM International Conference on Information and Knowledge Management. pp. 883–892 (2018)

    work page 2018

  31. [31]

    Semantic Web7(1), 63–93 (2016)

    Zaveri, A., Rula, A., Maurino, A., Pietrobon, R., Lehmann, J., Auer, S.: Quality assessment for linked data: A survey. Semantic Web7(1), 63–93 (2016)

    work page 2016