CoChat
Sign Up
Cited by 146Topic ModelingMultimodal Machine Learning ApplicationsNatural Language Processing Techniques

The DOI matches its Crossref record; 33 of the 33 references it cites were found in a registry or an index.

E-BERT: Efficient-Yet-Effective Entity Embeddings for BERT

Open access
Poerner, Nina, Waltinger, Ulli, Schütze, Hinrich·2020·Findings of the Association for Computational Linguistics: EMNLP 2020
DOI 10.18653/v1/2020.findings-emnlp.71
Open this paper in CoChat — free accountOpen PDF

Abstract

We present a novel way of injecting factual knowledge about entities into the pretrained BERT model ( Devlin et al., 2019): We align Wikipedia2Vec entity vectors (Yamada et al., 2016) with BERT's native wordpiece vector space and use the aligned entity vectors as if they were wordpiece vectors.The resulting entity-enhanced version of BERT (called E-BERT) is similar in spirit to ERNIE (Zhang et al., 2019) and KnowBert (Peters et al., 2019), but it requires no expensive further pretraining of the BERT encoder.We evaluate E-BERT on unsupervised question answering (QA), supervised relation classification (RC) and entity linking (EL).On all three tasks, E-BERT outperforms BERT and other baselines.We also show quantitatively that the original BERT model is overly reliant on the surface form of entity names (e.g., guessing that someone with an Italian-sounding name speaks Italian), and that E-BERT mitigates this problem.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.18653/v1/2020.findings-emnlp.71

Cites

33 of 33 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Schematic depiction of E-BERT-concat.
Figure 1 Schematic depiction of E-BERT-concat.CC-BY · From the PDF
Figure 2 Left y-axis (bars): delta in mean Hits@1 relative to BERT on individual LAMA relations. Right y-axis (crosses): frequency of questions where the answer is a substring of the subject entity name (i.e., questions that would be deleted by the string match filter). Model size: BASE . Due to space constraints, we only show relations with max absolute delta ≥ 0 . 075 .
Figure 2 Left y-axis (bars): delta in mean Hits@1 relative to BERT on individual LAMA relations. Right y-axis (crosses): frequency of questions where the answer is a substring of the subject entity name (i.e., questions that would be deleted by the string match filter). Model size: BASE . Due to space constraints, we only show relations with max absolute delta ≥ 0 . 075 .CC-BY · From the PDF
Figure 3 Mean Hits@k for different k . Model size: BASE . The x-axis is on a logarithmic scale.
Figure 3 Mean Hits@k for different k . Model size: BASE . The x-axis is on a logarithmic scale.CC-BY · From the PDF
Figure 4 Schematic depiction of E-BERT-MLM in inference mode, predicting an entity vector for the name 'Platt' in context. Blue: E BERT wordpiece vectors. Red: E E -BERT entity vectors. The candidates A and their priors p ( a ) are given by the candidate generator. Assume that the entity Tony Adams (footballer) was decoded in a previous iteration (see 'Iterative refinement').
Figure 4 Schematic depiction of E-BERT-MLM in inference mode, predicting an entity vector for the name 'Platt' in context. Blue: E BERT wordpiece vectors. Red: E E -BERT entity vectors. The candidates A and their priors p ( a ) are given by the candidate generator. Assume that the entity Tony Adams (footballer) was decoded in a previous iteration (see 'Iterative refinement').CC-BY · From the PDF
Figure 5 AIDA dev set micro F1 after every epoch.
Figure 5 AIDA dev set micro F1 after every epoch.CC-BY · From the PDF
Figure 6 Relation classification: Expected maximum macro F1 (dev set) as a function of the number of hyperparameter configurations.
Figure 6 Relation classification: Expected maximum macro F1 (dev set) as a function of the number of hyperparameter configurations.CC-BY · From the PDF
Figure 7 Entity linking: Expected maximum micro F1 (dev set) as a function of the number of hyperparameter configurations.
Figure 7 Entity linking: Expected maximum micro F1 (dev set) as a function of the number of hyperparameter configurations.CC-BY · From the PDF

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 33 references

Research works printed but not linked: 2

  1. Hady Elsahar, Pavlos Vougiouklis, Arslen Remaci, Christophe Gravier, Jonathon Hare, Frédérique Laforest, and Elena Simperl. 2018. T-REx: A large scale alignment of natural language with knowledge base triples. In LREC, pages 3448-3452, Miyazaki, Japan.
  2. Diederik P Kingma and Jimmy Ba. 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980. · arXiv

Cited by

Sorted by citations

See all 146

Related

Metadata

Authors
Poerner, N., Waltinger, U., & Schütze, H.
Affiliations
  • Siemens (Germany) (DE)
  • Ludwig-Maximilians-Universität München (DE)
ORCID
Venue
Findings of the Association for Computational Linguistics: EMNLP 2020
Publisher
Association for Computational Linguistics
Published
2020
Pages
803-818
DOI
10.18653/v1/2020.findings-emnlp.71
OpenAlex
W3021191241
Cited by
146
References
33
Citations by year
2026: 5 · 2025: 11 · 2024: 21 · 2023: 24 · 2022: 34 · 2021: 34 · 2020: 15
FWCI
10.21
Type
conference paper
Language
en
Open access
Licence
cc-by
Locations
  • Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP) (publishedVersion, cc-by)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
Funders
Siemens
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Multimodal Machine Learning Applications
  • Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
Keywords
entity embeddings, BERT, knowledge injection, relation classification, entity linking

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review