Mis on entity linking ja millised mudelid seda lahendavad?

Entitiy linking is the task of connecting mentions of entities in text to specific records in a knowledge base or ontology (for example, in Wikipedia or a knowledge graph). The goal is to determine which exact entity the mention refers to, considering the context.

Various models are used to solve entity linking:

• Rules and heuristics — early approaches based on string matching and contextual clues.
• Models based on vector representations (embeddings) — use contextual embeddings (e.g., BERT) to match the mention with entities.
• Graph models — consider relationships between entities in a knowledge graph to improve accuracy.
• Neural models with attention — allow taking into account the context and relationships between mentions.

For example: a model might first identify the mention "Apple" in the text, then, using the context (e.g., the words "iPhone", "company"), associate it with the entity "Apple Inc." in the knowledge base, not with the fruit.