NLP Wiki

User Tools

Site Tools

Trace: • diffusion_models • natural_language_interfaces_to_databases • multi-task_learning • text_analytics • graphical_models • visualizing_neural_networks • hugging_face • language_identification • contrastive_learning • model_editing_and_unlearning
nlp:entity_linking

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision		Previous revision
nlp:entity_linking [2021/07/02 21:56] – [Papers] jmflanignlp:entity_linking [2025/06/06 23:29] (current) – jmflanig
Line 9: Line 9:
 ===== Papers ===== ===== Papers =====
 See [[https://www.aclweb.org/anthology/search/?q=entity+linking|ACL Anthology - Entity linking]] See [[https://www.aclweb.org/anthology/search/?q=entity+linking|ACL Anthology - Entity linking]]
 +  * [[https://aclanthology.org/W12-3016.pdf|Lin et al 2012 - Entity Linking at Web Scale]] 
   * [[http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.297.6050&rep=rep1&type=pdf|Rao et al 2014 - Entity Linking: Finding Extracted Entities in a Knowledge Base]]  Explains why it's good to use a ranking approach to training an entity linking system in section 5.  "One natural approach to learning would be classification, in which each possible y ∈ Y is classified as being either correct or incorrect. However, such an approach enforces strong constraints: we not only require the correct KB entry to be classified positively, but all other answers to be classified negatively. Additionally, we can expect very unbalanced training, in which the vast majority of possible answers are incorrect... Instead, we select a single correct candidate for a query using a supervised machine learning ranker."   * [[http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.297.6050&rep=rep1&type=pdf|Rao et al 2014 - Entity Linking: Finding Extracted Entities in a Knowledge Base]]  Explains why it's good to use a ranking approach to training an entity linking system in section 5.  "One natural approach to learning would be classification, in which each possible y ∈ Y is classified as being either correct or incorrect. However, such an approach enforces strong constraints: we not only require the correct KB entry to be classified positively, but all other answers to be classified negatively. Additionally, we can expect very unbalanced training, in which the vast majority of possible answers are incorrect... Instead, we select a single correct candidate for a query using a supervised machine learning ranker."
   * [[http://blender.cs.illinois.edu/paper/edl2014overview.pdf|Ji et al 2014 - Overview of TAC-KBP2014 Entity Discovery and Linking Tasks]]   * [[http://blender.cs.illinois.edu/paper/edl2014overview.pdf|Ji et al 2014 - Overview of TAC-KBP2014 Entity Discovery and Linking Tasks]]
Line 23: Line 24:
   * [[https://www.aclweb.org/anthology/2020.emnlp-main.630.pdf|Botha et al 2020 - Entity Linking in 100 Languages]]   * [[https://www.aclweb.org/anthology/2020.emnlp-main.630.pdf|Botha et al 2020 - Entity Linking in 100 Languages]]
   * [[https://arxiv.org/pdf/2101.09969.pdf|Ravi et al 2021 - CHOLAN: A Modular Approach for Neural Entity Linking on Wikipedia and Wikidata]]   * [[https://arxiv.org/pdf/2101.09969.pdf|Ravi et al 2021 - CHOLAN: A Modular Approach for Neural Entity Linking on Wikipedia and Wikidata]]
 +
 +==== Joint NER + EL ====
 +  * [[https://aclanthology.org/P19-2026.pdf|Martins et al 2019 - Joint Learning of Named Entity Recognition and Entity Linking]]
  
 ==== Evaluation ==== ==== Evaluation ====
Line 29: Line 33:
 ==== EL in Dialog ==== ==== EL in Dialog ====
   * [[https://www.aclweb.org/anthology/2021.naacl-industry.4.pdf|Shang et al 2021 - Entity Resolution in Open-domain Conversations]]   * [[https://www.aclweb.org/anthology/2021.naacl-industry.4.pdf|Shang et al 2021 - Entity Resolution in Open-domain Conversations]]
 +
 +===== Candidate Entities =====
 +Most EL systems use a "name dictionary" or "alias table," which is a list from strings to candidate entities they might refer to.  These are usually created using rule-based heuristics, such as running a string edit distance between entities in the knowledge graph and the train, dev and test set text beforehand (this is too slow for an online system).  The papers in this section describe methods of finding candidate entities (and perhaps also linking them at the same time).
 +
 +  * [[https://arxiv.org/pdf/1912.01070.pdf|Bansal et al 2019 - Simultaneously Linking Entities and Extracting Relations from Biomedical Text Without Mention-level Supervision]] "Candidate generation" section talks about using string-edit distance for candidate entities.
 +  * [[https://arxiv.org/pdf/1909.10506.pdf|Gillick et al 2019 - Learning Dense Representations for Entity Retrieval]] Uses dense representations to retrieve candidate entities
  
 ===== Datasets ===== ===== Datasets =====
nlp/entity_linking.1625262997.txt.gz · Last modified: 2023/06/15 07:36 (external edit)
Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki