Entity Salience

Entity salience is a measure of how central a given entity is to a piece of content — how prominently and substantively the content is actually about that entity, as opposed to merely mentioning it in passing. When a system analyzes content, it does not treat every entity present as equally important. It assesses which entities the content is genuinely centered on and which are incidental, and entity salience is that assessment of centrality. Salience influences retrieval and citation. A generative system answering a question about a particular entity searches for content genuinely and substantively about that entity. Content where the entity has high salience — where it is a central subject, discussed in depth, surrounded by related entities and substantive detail — is a strong match and is more likely to be retrieved and cited when that entity is queried. Content where the entity has low salience — mentioned briefly but not the actual subject — is a weak match, and a passing mention does not earn a citation. For an organization, this has a clear and sometimes counterintuitive implication. It is not sufficient for the organization or its key topics to merely appear in content. To earn retrieval and citation, the organization needs content where it, and the topics it intends to be known for, hold genuine salience — content substantively and centrally about those things. This works against a common instinct to seed an organization's name lightly across a large volume of loosely related content; that distributes the name but builds little salience anywhere. The stronger approach is depth: content genuinely and centrally about the organization and its core topics, where salience is real. Entity salience rewards substance over distribution.