X-STREAMS: Cross- Stream Textual Realtime Multi-document Summarizer

Period of Performance: 08/07/2012 - 02/06/2013


Phase 1 SBIR

Recipient Firm

Manifest Labs, Inc.
2900 W. Anderson Lane C-200-301
Austin, TX 78757
Principal Investigator


In response to the Navy s N121-078 solicitation, Manifest Labs, Inc., proposes X-STREAMS, a real-time summarization system that improves upon the current state-of-the-art results on the novel information reporting of entities and events found in textual data sources. Using a novel combination of mature techniques, and a new semantic layering methodology, X-STREAMS will increase the value of streaming document summarization capabilities, by merging information across streams and improving the timeliness and accuracy of automated knowledge discovery. The ultimate goal of the X-STREAMS research is to automate much of the summarization of documents and other forms of communication which may be represented as text, such as IM chat, voice and image transcriptions. The strength of X-STREAMS is that it uses a data-driven, unsupervised learning approach to train adaptable summarization models. These models can be trained in any language, and do not require special rules or linguists to develop or maintain the system. To minimize redundant information in reports, X-STREAMS employs a parallelized implementation of the leading methodology for determining the maximum marginal relevance in automated document summarization.