pith. sign in

arxiv: 1907.03007 · v1 · pith:6JMEL44Bnew · submitted 2019-07-05 · 💻 cs.IR · cs.AI· cs.CL

NeuType: A Simple and Effective Neural Network Approach for Predicting Missing Entity Type Information in Knowledge Bases

Jon Arne B{\o} Hovda , Dar\'io Garigliotti , Krisztian Balog This is my paper

Pith reviewed 2026-05-25 01:47 UTC · model grok-4.3

classification 💻 cs.IR cs.AIcs.CL
keywords entity typingknowledge basesneural networksDBpediamissing informationsemantic typesinformation retrieval
0
0 comments X

The pith

Simple neural networks significantly improve prediction of missing entity types in knowledge bases

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

The paper introduces two neural network architectures designed to assign semantic types to entities in knowledge bases when that information is missing. These models take as input short textual descriptions of entities and, optionally, details about related entities. Evaluated on the DBpedia knowledge base, the architectures achieve significant improvements over existing methods. A sympathetic reader would care because incomplete type information limits the utility of knowledge bases in search and other tasks, and this approach offers an effective way to complete them automatically.

Core claim

The central claim is that neural networks processing short entity descriptions can accurately predict entity types from a taxonomy, outperforming the current state of the art on DBpedia.

What carries the argument

Two neural network architectures that process short entity descriptions and optionally related-entity information to predict types.

Load-bearing premise

Short textual descriptions of entities contain sufficient information to determine their semantic types accurately.

What would settle it

A test on DBpedia where the neural models fail to outperform the previous state-of-the-art methods on type prediction accuracy.

Figures

Figures reproduced from arXiv: 1907.03007 by Dar\'io Garigliotti, Jon Arne B{\o} Hovda, Krisztian Balog.

Figure 1
Figure 1. Figure 1: Neural architectures. Arrows indicate fully [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
read the original abstract

Knowledge bases store information about the semantic types of entities, which can be utilized in a range of information access tasks. This information, however, is often incomplete, due to new entities emerging on a daily basis. We address the task of automatically assigning types to entities in a knowledge base from a type taxonomy. Specifically, we present two neural network architectures, which take short entity descriptions and, optionally, information about related entities as input. Using the DBpedia knowledge base for experimental evaluation, we demonstrate that these simple architectures yield significant improvements over the current state of the art.

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

0 major / 3 minor

Summary. The paper introduces two simple neural network architectures (NeuType) for the task of predicting missing semantic types for entities in a knowledge base taxonomy. The models take short textual entity descriptions as primary input, with an optional extension incorporating information about related entities. Experiments are conducted on the DBpedia knowledge base, with the central claim being that these architectures produce significant improvements over prior state-of-the-art methods for entity type prediction.

Significance. If the reported gains hold under rigorous evaluation, the work is significant for demonstrating that straightforward neural models can outperform more elaborate prior approaches on a practically important KB completion task. The emphasis on simplicity is a strength, as it lowers barriers to adoption and reproduction for information access applications that rely on complete type information. The modeling assumption that short descriptions suffice is standard but here shown to be effective at scale.

minor comments (3)
  1. The abstract asserts 'significant improvements' without any quantitative details, baselines, or statistical tests; move a concise summary of the key results (e.g., F1 gains and significance tests) into the abstract for clarity.
  2. Section 4 (experimental setup) should explicitly state the train/validation/test splits, the exact DBpedia version used, and the full list of baselines with citations to ensure reproducibility.
  3. Figure 1 and the architecture diagrams would benefit from clearer labeling of input dimensions and the optional related-entity branch to avoid ambiguity in the optional extension.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment of our work and the recommendation for minor revision. We appreciate the recognition that the emphasis on simple neural architectures is a strength for practical adoption in KB completion tasks.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper proposes two neural network architectures for entity type prediction from short descriptions (optionally with related entities) and evaluates them empirically on DBpedia. No equations, derivations, or mathematical claims appear in the provided text. The central result is an empirical performance improvement over prior SOTA, which rests on standard supervised learning rather than any self-referential construction, fitted-input prediction, or load-bearing self-citation chain. The modeling assumption that descriptions contain sufficient signal is a conventional empirical premise, not a definitional loop. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Review performed on abstract only; no explicit free parameters, axioms, or invented entities are stated.

axioms (1)
  • domain assumption Neural networks can map short textual descriptions to semantic types from a fixed taxonomy
    Core modeling assumption implicit in the proposed architectures

pith-pipeline@v0.9.0 · 5635 in / 1044 out tokens · 22874 ms · 2026-05-25T01:47:44.567069+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

14 extracted references · 14 canonical work pages

  1. [1]

    Krisztian Balog. 2018. Entity-Oriented Search. /T_he Information Retrieval Series, Vol. 39. Springer

    work page 2018

  2. [2]

    Krisztian Balog and Robert Neumayer. 2012. Hierarchical target type identi/f_ica- tion for entity-oriented queries. In Proc. of CIKM. 2391–2394

    work page 2012

  3. [3]

    Aldo Gangemi, Andrea Giovanni Nuzzolese, Valentina Presu/t_ti, Francesco Draic- chio, Alberto Muse/t_ti, and Paolo Ciancarini. 2012. Automatic typing of DBpedia entities. In International Semantic Web Conference. 65–81

    work page 2012

  4. [4]

    Dar´ıo Gariglio/t_ti, Faegheh Hasibi, and Krisztian Balog. 2019. Identifying and exploiting target entity type information for ad hoc entity retrieval. Information Retrieval Journal 22, 3 (2019), 285–323

    work page 2019

  5. [5]

    Mohit Iyyer, Varun Manjunatha, Jordan Boyd-Graber, and Hal Daum´e III. 2015. Deep Unordered Composition Rivals Syntactic Methods for Text Classi/f_ication. In Proc. of ACL-IJCNLP. 1681–1691

    work page 2015

  6. [6]

    Tom´aˇs Kliegr and Ondˇrej Zamazal. 2016. LHD 2.0: A text mining approach to typing entities in knowledge graphs. Web Semantics: Science, Services and Agents on the World Wide Web 39 (2016), 47–61

    work page 2016

  7. [7]

    Le and Tomas Mikolov

    /Q_uoc V. Le and Tomas Mikolov. 2014. Distributed Representations of Sentences and Documents. In Proc. of ICML, Vol. 32. 1188–1196

    work page 2014

  8. [8]

    /T_homas Lin, Mausam, and Oren Etzioni. 2012. No Noun Phrase Le/f_t Behind: Detecting and Typing Unlinkable Entities. In Proc. of EMNLP-CoNLL. 893–903

    work page 2012

  9. [9]

    Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. 2013. Distributed representations of words and phrases and their compositionality. In Proc. of NIPS. 3111–3119

    work page 2013

  10. [10]

    Ndapandula Nakashole, Tomasz Tylenda, and Gerhard Weikum. 2013. Fine- grained semantic typing of emerging entities. In Proc. of ACL. 1488–1497

    work page 2013

  11. [11]

    Heiko Paulheim and Christian Bizer. 2013. Type inference on noisy RDF data. In Proc. of ISWC. 510–525

    work page 2013

  12. [12]

    Heiko Paulheim and Christian Bizer. 2014. Improving the quality of Linked Data using statistical distributions. International Journal on Semantic Web and Information Systems (IJSWIS) 10, 2 (2014), 63–86

    work page 2014

  13. [13]

    Yadollah Yaghoobzadeh, Heike Adel, and Hinrich Sch¨utze. 2017. Noise Mitigation for Neural Entity Typing and Relation Extraction. In Proc. of EACL. 1183–1194

    work page 2017

  14. [14]

    Yadollah Yaghoobzadeh and Hinrich Sch¨utze. 2015. Corpus-level Fine-grained Entity Typing Using Contextual Information. In Proc. of EMNLP. 715–725

    work page 2015