Search for answers as an application of the problem of relationship extraction

Abstract

The volume of unstructured information in the form of text documents from various sources is ever increasing. In order to access it and obtain knowledge that can be applied to a variety of tasks, you must first "structure" that information. The extraction of the relational structure between entities, in the form of triplets based on a verb, can be applied to the problem of finding answers. This work has made use of efficient techniques of shallow text analysis. The system built has a precision and recall comparable to other state of the art systems. The extracted triplets form the knowledge base on which queries are made to obtain answers to questions in natural language. The results over a question database to posed to a corpus of 2052 web documents on Salta demonstrate the validity of this approach.