Boosting Entity Linking Performance by Leveraging Unlabeled Documents

Le, Phong Ba; Titov, Ivan

doi:10.18653/v1/p19-1187

Cited by 42 publications

(53 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our model is competitive with Plato in the semi-supervised setting, which additionally uses 50 million documents as unlabeled data. Le and Titov (2019)'s setting is quite different from ours that their model is a global model (requires document input) and trained on Wikipedia and 30k newswire documents from the Reuters RCV1 corpus (Lewis et al 2004). Their model is potentially trained on domain-specific data since the CoNLL-YAGO dataset is derived from the RCV1 corpus.…”

Section: Resultsmentioning

confidence: 99%

Fine-Grained Entity Typing for Domain Independent Entity Linking

Onoe

Durrett

2020

AAAI

View full text Add to dashboard Cite

Neural entity linking models are very powerful, but run the risk of overfitting to the domain they are trained in. For this problem, a “domain” is characterized not just by genre of text but even by factors as specific as the particular distribution of entities, as neural models tend to overfit by memorizing properties of frequent entities in a dataset. We tackle the problem of building robust entity linking models that generalize effectively and do not rely on labeled entity linking data with a specific entity distribution. Rather than predicting entities directly, our approach models fine-grained entity properties, which can help disambiguate between even closely related entities. We derive a large inventory of types (tens of thousands) from Wikipedia categories, and use hyperlinked mentions in Wikipedia to distantly label data and train an entity typing model. At test time, we classify a mention with this typing model and use soft type predictions to link the mention to the most similar candidate entity. We evaluate our entity linking system on the CoNLL-YAGO dataset (Hoffart et al. 2011) and show that our approach outperforms prior domain-independent entity linking systems. We also test our approach in a harder setting derived from the WikilinksNED dataset (Eshel et al. 2017) where all the mention-entity pairs are unseen during test time. Results indicate that our approach generalizes better than a state-of-the-art neural model on the dataset.

show abstract

Section: Resultsmentioning

confidence: 99%

Fine-Grained Entity Typing for Domain Independent Entity Linking

Onoe

Durrett

2020

AAAI

View full text Add to dashboard Cite

show abstract

“…the traditional methods on standard benchmark (e.g., AIDA-CoNLL). A line of follow-up work (Le and Titov 2018;2019a;2019b) investigate potential improvement solution or other task settings based on that.…”

Section: Introductionmentioning

confidence: 99%

Improving Entity Linking by Modeling Latent Entity Type Information

Chen

Wang

Jiang

et al. 2020

AAAI

View full text Add to dashboard Cite

Existing state of the art neural entity linking models employ attention-based bag-of-words context model and pre-trained entity embeddings bootstrapped from word embeddings to assess topic level context compatibility. However, the latent entity type information in the immediate context of the mention is neglected, which causes the models often link mentions to incorrect entities with incorrect type. To tackle this problem, we propose to inject latent entity type information into the entity embeddings based on pre-trained BERT. In addition, we integrate a BERT-based entity similarity score into the local context model of a state-of-the-art model to better capture latent entity type information. Our model significantly outperforms the state-of-the-art entity linking models on standard benchmark (AIDA-CoNLL). Detailed experiment analysis demonstrates that our model corrects most of the type errors produced by the direct baseline.

show abstract

“…It is clear, that the above-mentioned methods cannot guarantee correct labelling of the samples, however, such imperfect data can still be used in weak supervision. This strategy is used extensively for named entity recognition [20], relation extraction [21], [22], entity linking [23] and text classification [24]. As weak supervision can introduce different types of noise into a model, in our research to infer the sense label of the unannotated sample, we combined the predicted class probabilities of the three weakly supervised models alongside uncertainty estimation.…”

Section: Related Workmentioning

confidence: 99%

Weakly Supervised Word Sense Disambiguation Using Automatically Labelled Collections

Loukachevitch¹

2021

Proceedings of ISP RAS

View full text Add to dashboard Cite

State-of-the-art supervised word sense disambiguation models require large sense-tagged training sets. However, many low-resource languages, including Russian, lack such a large amount of data. To cope with the knowledge acquisition bottleneck in Russian, we first utilized the method based on the concept of monosemous relatives to automatically generate a labelled training collection. We then introduce three weakly supervised models trained on this synthetic data. Our work builds upon the bootstrapping approach: relying on this seed of tagged instances, the ensemble of the classifiers is used to label samples from unannotated corpora. Along with this method, different techniques were exploited to augment the new training examples. We show the simple bootstrapping approach based on the ensemble of weakly supervised models can already produce an improvement over the initial word sense disambiguation models.

show abstract

Boosting Entity Linking Performance by Leveraging Unlabeled Documents

Cited by 42 publications

References 14 publications

Fine-Grained Entity Typing for Domain Independent Entity Linking

Fine-Grained Entity Typing for Domain Independent Entity Linking

Improving Entity Linking by Modeling Latent Entity Type Information

Weakly Supervised Word Sense Disambiguation Using Automatically Labelled Collections

Contact Info

Product

Resources

About