Searching for entities: When retrieval meets extraction

Retrieving entities inside documents instead of documents or web pages themselves has become an active topic in both commercial search systems and academic information retrieval research. Our method of entity retrieval is based on a two-layer retrieval and extraction probability model (TREPM) for integrating document retrieval and entity extraction together. The document retrieval layer finds supporting documents from the corpus, and the entity extraction layer extracts the right entities from those supporting documents. We theoretically demonstrate that the entity extraction problem can be represented as TREPM model. The TREPM model can reduce the overall retrieval complexity while keeping high accuracy of locating target entities. The experiment is based on the document retrieval and entity extraction as well as the overall task. The preliminary results are promising and deserve for further exploration. Keywords: entity retrieval, document retrieval, entity extraction