Afrl - Ri - Rs - Tm - 2007 -

As stated at the 2003 Workshop on Satellite Data Applications and Information Extraction, there is a pressing need in today’s world to be able to “facilitate quick war fighting decisions that fully leverage the huge volumes of available information.” However, this can be a daunting task due to the massive amounts of data used to find intelligence about current situations and to predict future events. Information extraction technologies are currently being developed to aid intelligence analysts in the process of finding pieces of key information in text, audio, image, and video documents. This technology will have the ability to automatically pull out relevant pieces of intelligence and structure them in a way that analysts can easily understand, enabling them to utilize more data, make decisions faster, and stay on top of global situations. This paper describes the need for information extraction technologies within the military, some of the current technologies available, and the problems associated with them. It also looks at some of the ongoing research projects in areas of multimedia information extraction. Finally, it looks at the StreamSage audio extraction software and the demonstration of this software, explains how to run the original software and critiques it, and describes the demo platform developed during the author's summer employment at AFRL-Rome. Background Intelligence analysts must sort through large amounts of data to determine which data sources are relevant to their needs and to find pertinent pieces of information within that data. With the heightened security threats in today’s world, there is an increased need to identify possible threats quickly and more accurately. However, the volume of data used to find intelligence is increasing while the number of intelligence analysts is decreasing due to downsizing within the Department of Defense. Therefore, the need for faster methods of searching intelligence data is growing. Information extraction (IE) is the process of extracting data from retrieved documents and saving it in a useful manner, such as a database. In the future, the use of IE technology will reduce the amount of time and labor spent on finding key pieces of information, therefore enabling analysts to develop reports faster and find relevant information in a shorter amount of time. Text information extraction technology is one area of IE technology. It is divided into three levels of complexity: shallow, intermediate, and deep extraction. Current work in text IE has been at the shallow and intermediate levels. Shallow extraction finds basic information such as named entities (the names of people, places, and organizations), numerical information (monetary values, percentages, etc), and simple events denoted by action verbs. Technology in this area is currently available for use in the intelligence industry, thereby giving analysts the ability to sort through large amounts of text documents quickly to find basic information. Intermediate extraction technologies possess the capabilities to extract relationships between entities (e.g. Person B works for Person A) and find more meaningful event information, such as people taking part in an action, as well as the time and place of an action. This area is currently being researched by various organizations, including the United States Air Force Research Laboratory Rome Research Site. Future research in text IE will be at the deep extraction level. Deep extraction will recognize event scenarios and complex relationships, as well as infer information from non-explicit events. Deep extraction technology will also find information