Advanced Indexing and Search for Efficient Information Discovery

ABSTRACT: In this project, we propose to develop a semantic search system called ActiveSearch, which builds upon the extraction capabilities of NLP engines, and also allows additional ontological information to be used for search. ActiveSearch focuses on two key requirements. The first is that an analyst must be able to quickly and easily convey the types of information that should be targeted. To achieve this capability, we build on recent advances in automatic query expansion and active learning. Our approach allows large expanded queries to be automatically formulated, and then reformulated based on user feedback, so that the system can accurately recognize the targeted content. The second requirement is that the system be fast enough to respond to a user in real-time. To achieve this capability, we propose an approach that can rapidly execute large queries by taking advantage of reverse indices coupled with parallel execution BENEFIT: Today"s analysts are faced an enormous number of documents and yet it can be difficult for them to find the information they need. Search engines can be ineffective for researching topics that cannot be easily be described in a few keywords. On the other end of the spectrum, text extraction engines can identify targeted facts quite precisely, but they need to be trained and laboriously tuned for each task. ActiveSearch represents a middle ground, allowing analysts to pose hard informational queries that can be rapidly tuned for a given task. ActiveSearch is a"research engine"that will be able to save analysts significant time when sifting through large document corpuses. The approach is intended to support military and intelligence analysts, as well as commercial industries that aggregate and compile data, such as the background investigation industry.