Abstract: The subject matter of this specification can be embodied in, among other things, a method that includes identifying resources relating to an entity, where each resource includes multiple terms and is included in a corpus of resources relating to multiple entities. Candidate terms from the resources for potentially associating with the entity and a category associated with the entity are identified. A relative frequency of the candidate terms in the identified resources is compared to a frequency of the candidate terms associated with other entities. Each of the candidate terms are weighted, for example, based on a source of the candidate term and the relative frequency of the candidate term. A weighted frequency of each candidate term is calculated based on the weights, and candidate terms are selected as representative terms for the entity based on the weighted frequency.