entity disambiguation
**Entity disambiguation** resolves **which specific entity a mention refers to** — determining whether "Jordan" means the country, Michael Jordan, or Jordan River, using context clues to select the correct entity from multiple candidates.
**What Is Entity Disambiguation?**
- **Definition**: Resolve ambiguous entity mentions to specific entities.
- **Problem**: Same name can refer to multiple entities.
- **Goal**: Select correct entity based on context.
**Ambiguity Types**
**Name Ambiguity**: "Washington" (person, city, state, president).
**Metonymy**: "White House" (building or administration).
**Abbreviations**: "MIT" (university, other organizations).
**Common Names**: "John Smith" (thousands of people).
**Cross-Lingual**: Same entity, different names in different languages.
**Disambiguation Signals**
**Context**: Surrounding words provide clues.
**Co-Occurring Entities**: Other entities mentioned nearby.
**Document Topic**: Overall document subject.
**Entity Popularity**: More famous entities more likely.
**Entity Types**: Expected type from context (person, place, organization).
**Temporal**: Time period of document.
**Geographic**: Location context.
**AI Techniques**
**Feature-Based**: Context features, entity features, compatibility scores.
**Embedding-Based**: Entity and context embeddings, similarity matching.
**Graph-Based**: Entity coherence in knowledge graph.
**Neural Models**: BERT-based disambiguation, entity-aware transformers.
**Collective Disambiguation**: Resolve all mentions jointly for coherence.
**Evaluation**: Accuracy on benchmark datasets (AIDA CoNLL, MSNBC, ACE).
**Applications**: Knowledge base population, question answering, information extraction, semantic search.
**Tools**: DBpedia Spotlight, TagMe, BLINK, spaCy entity linker, Wikifier.