Entity Extraction in Passage Extractor

Passage Extractor uses entity extraction to provide more concise, specific answers. It attempts to find the shortest possible section of a document that answers the original question. Depending on the question, the answer might be a single word or name, or a few sentences of description.

There are two types of entity extraction that you can use:

  • Eduction. IDOL Eduction provides a set of grammars, which define the entities that you want to find. This method is very powerful for pattern matching, and finding entities that match a particular structure, such as phone numbers. Answer Server includes an embedded IDOL Eduction module for entity extraction.
  • Agent matching. The Passage Extractor Agentstore component stores agents that define entities in the Agentstore component. Passage Extractor sends any candidate answers in an agent query, which returns the matching entity agents. This method is very powerful for entities that have a clearly defined value, such as names.

To get the most out of Passage Extractor, you must configure at least one of an Eduction grammar or Agentstore component for entity extraction. You can use both to make the most out of the different methods, and to get the best answers.