Basically the algorithm considers frequency, position, length and distribution of terms over the documents.
See also: Entity Extractor Architecture
An important component is the usage of Corpus Management to fine-tune the relevancy of a term. See also: Corpus Management - Overview