Information Extraction

Abstract

Information Extraction (IE) techniques aim to extract the names of entities and objects from text and to identify the roles that they play in event descriptions. IE systems generally focus on a specific domain or topic, searching only for information that is relevant to a user's interests. In this chapter, we first give historical background on information extraction and discuss several kinds of information extraction tasks that have emerged in recent years. Next, we outline the series of steps that are involved in creating a typical information extraction system, which can be encoded as a cascaded finite-state transducer. Along the way, we present examples to illustrate what each step does. Finally, we present an overview of different learning-based methods for information extraction, including supervised learning approaches, weakly supervised and bootstrapping techniques, and discourse-oriented approaches. Information extraction (IE) is the process of scanning text for information relevant to some interest, including extracting entities, relations, and, most challenging, events–or who did what to whom when and where. It requires deeper analysis than key word searches, but its aims fall short of the very hard and long-term problem of text understanding, where we seek to capture all the information in a text, along with the speaker's or writer's intention.