Entity Linking & Resolution

Text contains mentions ('Dr. Smith', 'the surgeon', 'Smith et al.'). The world contains entities (the actual surgeon, one person). Your graph contains nodes (your records of entities — hopefully one per entity). Two distinct problems connect them: entity linking maps mention→entity ('which Apple does this sentence mean?'), and entity resolution maps node→node ('these two records are the same entity — merge them').

Conflating the three layers is the source of most 'why is Apple-the-fruit connected to iPhone' bugs: a linking error writes a fruit's relations onto a company's node.