Entity Linking

Entity linking is the process of mapping entity mentions in text to their corresponding canonical entries in a knowledge base or knowledge graph. For example, the mentions "Alice," "Alice Smith," and "A. Smith" in different documents should all be linked to the same canonical entity node representing that person.

Entity linking involves two sub-tasks: mention detection (identifying text spans that refer to entities) and disambiguation (determining which canonical entity a mention refers to when multiple candidates exist). For instance, "Apple" could refer to Apple Inc. or the fruit, and the surrounding context determines the correct linking.

In knowledge graph construction pipelines, entity linking is essential for maintaining data quality. Without it, the same entity gets multiple nodes, relationships are fragmented, and graph queries return incomplete results. Implementation approaches range from simple alias dictionaries and fuzzy string matching to LLM-based linking that uses context to resolve ambiguity. Entity linking quality directly impacts the completeness and usefulness of the resulting knowledge graph.