Skip to content

Topic 04: Entity Linking and Entity Disambiguation

Jump to bottom
Sherry Lin edited this page Oct 9, 2020 · 17 revisions

Surveys and Analysis

  1. Entity Linking with a Knowledge Base: Issues, Techniques, and Solutions (TKDE 2014) 🌟
  2. Neural Entity Linking: A Survey of Models based on Deep Learning (2020) [Paper]
  • a survey of state-of-the-art neural entity linking models
  • a survey of entity embedding techniques;
  • a discussion of recent domain-independent (zero-shot) and cross-lingual EL approaches;
  • a survey of EL applications to modeling word representations.
  1. Error analysis of the well known DeepED model [Link]

General Papers

  1. Zero-Shot Entity Linking by Reading Entity Descriptions (ACL 2019) [Paper][Code and Datasets]
  2. Keyphrase Overlap Relatedness for Entity Disambiguation (CIKM 2012), LSH 🌟
  3. Old is Gold: Linguistic Driven Approach for Entity and Relation Linking of Short Text (NAACL 2019), with Relation Linking
  4. Improving Entity Linking by Modeling Latent Relations between Mentions (ACL 2018)
  5. Entity Linking for Tweets (ACL 2013)
  6. Pangloss: Fast Entity Linking in Noisy Text Environments (KDD 2018) [Presentation] 🌟
  7. THINKER - Entity Linking System for Turkish Language (TKDE 2018) 🌟
  8. SHINE+: A General Framework for Domain-Specific Entity Linking with Heterogeneous Information Networks (TKDE 2018) 🌟

Global Coherence

  1. Relational Inference for Wikification (ACL 2013)
  2. Robust Disambiguation of Named Entities in Text (EMNLP 2011)
  3. Liege: Link Entities in Web Lists with Knowledge Base (KDD 2012)🌟
  4. Collective entity linking in web text: A graph-based method (SIGIR 2011)
  5. Collective Annotation of Wikipedia Entities in Web Text (KDD 2009)🌟
  6. Local and Global Algorithms for Disambiguation to Wikipedia (ACL 2011)
  7. Learning entity representation for entity disambiguation (ACL 2013)
  8. Robust named entity disambiguation with random walks (Semantic Web 2018)
  9. Graph ranking for collective named entity disambiguation (ACL 2014)
  10. Personalized page rank for named entity disambiguation (ACL 2015)
  11. Collective entity resolution with multi-focal attention (2016)
  12. To link or not to link? a study on end-to-end tweet entity linking (NAACL 2013)
  13. An entity-topic model for entity linking (EMNLP 2012)
  14. Deep joint entity disambiguation with local neural attention (EMNLP 2017)
  15. Improving entity linking by modeling latent relations between mentions (ACL 2018)
  16. Neural Collective Entity Linking Based on Recurrent Random Walk Network Learning (IJCAI 2019), introduces external knowledge to model the semantic interdependence between different EL decisions
  17. ELDEN: Improved Entity Linking using Densified Knowledge Graphs (NAACL-HLT 2018) [Paper][Code], supervised EL system
  18. KBPearl: A Knowledge Base Population System Supported by Joint Entity and Relation Linking (VLDB 2020) [Paper], with relation linking 🌟
  19. Joint Embedding in Named Entity Linking on Sentence Level [Paper]

Relax the Global Coherence Assumption

  1. Joint entity linking with deep reinforcement learning (WWW 2019) [Paper]

Reinforcement learning, apply LSTM to be able to maintain long term memory for previous decisions.

  1. Learning Dynamic Context Augmentation for Global Entity Linking (DCA, ACL 2019) [Paper]

Reinforcement learning, previous decisions are collected as dynamic context to improve the following predictions.

  1. Global Entity Disambiguation with Pretrained Contextualized Embeddings of Words and Entities (2020) [Paper]
  • BERT+MLM.
  • Note: Paper 1,2,3 address ED as a sequential decision task that disambiguates mentions one by one, and uses words and already disambiguated entities to disambiguate new mentions.
  1. Joint Learning of Local and Global Features for Entity Linking via Neural Networks (COLING 2016), CNN+RNN
  2. Pair-Linking for Collective Entity Disambiguation: Two Could Be Better Than All (TKDE 2018) 🌟
  3. KBPearl: A Knowledge Base Population System Supported by Joint Entity and Relation Linking (VLDB 2020) 🌟

Some related works:

  1. Evaluating the Impact of Knowledge Graph Context on Entity Disambiguation Models (CIKM 2020, short) [Paper]

Employ SPASQL to fetch triples of target entities and incorporate the triples as kg context in pre-trained ED models. DCA is used as baselines on the AIDA-CONLL dataset.

Entity Linking with Type Info

  1. Improving Entity Linking through Semantic Reinforced Entity Embeddings (ACL 2020) [Paper] [Data and Code] [Details]
  • Fine-grained semantic types of entities can let the linking models learn contextual commonality about semantic relatedness.
  • fine-grained semantic words appear frequently as apposition (e.g., Defense contractor Raytheon), coreference (e.g., the company) or anonymous mentions (e.g., American defense firms). These fine-grained types of entities can help capture local contexts and relations of entities.
  1. Improving Entity Linking by Modeling Latent Entity Type Information (AAAI 2020) [Paper]
  • Conduct error analysis of the well known DeepED1 model (Ganea and Hofmann 2017) on the development set of AIDA-CoNLL, and found that more than half of their error cases fall into the category of type errors where the predicted entity’s type is different from the golden entity’s type.
  • Inject latent entity type information into the entity embeddings by modeling the immediate context surrounding the mention.
  • Apply pre-trained BERT to represent the entity context.
  1. A joint model for entity analysis: Coreference, typing, and linking (TACL 2014)
  2. Joint entity recognition and disambiguation (EMNLP 2015)
  3. J-nerd: joint named entity recognition and disambiguation with rich linguistic features (TACL 2016)
  • Paper 3,4,5 integrate type information into the entity linking task by jointly NER+EL, which captures the mutual dependency between them using structured CRF. These methods mainly differ in the design of hand-engineered features.
  1. Joint learning of named entity recognition and entity linking. (ACL: Student Research Workshop, 2019)
  • Multi-task learning using learned features by extending Stack-LSTM.

Joint NER and EL

  1. Joint Learning of Named Entity Recognition and Entity Linking (ACL 2019) [Paper]
  • Multi-task learning of NER and EL based on Stack-LSTM approach.
  • Supervised EL system with learned features.
  • Future extension: training entity contextual embeddings and extend it to be cross-lingual.
  1. Re-ranking for joint named-entity recognition and linking (CIKM 2013) [Paper]
  • The reranking model then chooses among the set of all possible mention and entity link labelings for the whole phrase to determine the best choice. It can use features for known relationships between the television channel ABC and the television program The_View to encourage these as outputs. For efficiency, we use the pipeline models to prune the set of all possible candidate mentions and entity links to a manageable size while maintaining high recall. The reranking model can then use more sophisticated features for collective classification over this pruned set.
  • Rely on existing NER tools. Only NER is beneficial to EL, not vice versa.
  • Hand-engineered features.
  • Uses a large number of heuristically obtained Noun phrase (NP) chunks and word n-grams as additional input to the EL stage.
  1. To link or not to link? a study on end-to-end tweet entity linking (NAACL 2013)
  • Only suitable for short-text such as tweets.
  1. Joint entity recognition and disambiguation (EMNLP 2015) [Paper]
  • NER is beneficial to EL. EL is also beneficial to NER.
  • Supervised EL system.
  • Hand-engineered features.
  1. J-nerd: joint named entity recognition and disambiguation with rich linguistic features (TACL 2016) [Paper] [Code]
  • Supervised, non-linear probabilistic graphical model that captures mention spans, mention types, and the mapping of mentions to entities in a knowledge base.
  • Hand-engineered features.
  • Relies on fully labeled training data where each tagged entity needs to have an NER and EL label.
  1. A Joint Model for Entity Analysis: Coreference, Typing, and Linking (TACL 2014) [Paper] [System] [GitHub]
  • Joint learning of entity typing, EL, and coreference.
  • Hand-engineered features.
  1. Contextualized End-to-End Neural Entity Linking [Paper]
  • An end-to-end differentiable neural EL model that jointly performs MD and ED, based on BERT, while eliminating external knowledge so that we can study the impact of external knowledge to the EL model.
  1. Noise-robust Named Entity Understanding for Virtual Assistants [Paper]
  • Combining NER and EL information in a joint reranking module for noisy spoken language queries in the context of a digital voice assistant, our proposed framework improves accuracy in both tasks.
  1. End-to-End Neural Entity Linking [Paper]
  • The main idea is to consider all possible spans as potential mentions and learn contextual similarity scores over their entity candidates that are useful for both MD and ED decisions.

Entity Embeddings (Pre-trained)

  1. Wembedder: Wikidata entity embedding web service [Intro] [Github] [Web service]
  • Web service: only the "most similar" service.
  1. Pre-trained embeddings for Wikidata [Link]
  2. Pre-trained embeddings [Link]
  • 100-dimention and 50-dimention, parsed by numpy.memmap. However, I can only read one float (instead of a vector) for each entity. No sure whether there is mistake.

Notes for Entity Linking

  1. Candiate Entity Ranking: https://www.jianshu.com/p/90e2c7a5c9f5
  2. http://nlpprogress.com/english/entity_linking.html
  3. Recent Trend for Entity Linking [Notes]
Clone this wiki locally