SBIR-STTR Award

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project will be improved effectiveness of document management systems in U.S. Businesses. The project integrates novel approaches to unsupervised machine learning, concept identification, and ontology construction to create a sustainable content management system that will allow companies to find and associate information more accurately and efficiently. Corporations run on information, and routine operations depend on finding information efficiently. For example, corporate acquisitions require filing information quickly in the acquiring company's systems; employee turnover necessitates intelligent analysis to enable continuing operations; and regulatory compliance and legal retention requirements demand consistent categorization and correct retention of records. While creating electronic documents is easy, finding and analyzing them remain difficult tasks. The proposed project is intended to provide effective assistance to companies, within everyday business practices, without requiring major investments in change. Distribution of information between corporate data centers and the cloud further necessitates tools to help with classification consistency and searchability. If successful, this project will provide an encompassing framework within which company workflows are integrated and corporate workers can more easily and efficiently extract usable information from corporate IT systems. This Small Business Innovation Research (SBIR) Phase I project provides new software tools for knowledge workers. Industrial information technology requires the integration of proven methods in a robust, sustainable framework. The investigators' prior work in artificial intelligence demonstrated that a well-designed framework, with open source packages and interstitial software, can provide an effective knowledge management system. In this project the company intends to mine and extend research ideas from knowledge management, artificial intelligence, natural language processing, machine learning, information retrieval and human-computer interfaces. Work in artificial intelligence has shown that domain knowledge is necessary for high performance problem solving. The company intends to leverage corporate knowledge to augment keyword search with semantics of the domain. Concept identification methods developed for natural language processing will be used to augment the powerful statistical tools provided by unsupervised machine learning and information retrieval technology.

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase II project will be more effective and efficient use of unstructured data. Limiting analysis to structured data ignores the massive amount of information in reports, memos, articles, and other written documents. Workers require information on past work and ongoing projects, best practices, current events and competitor and customer activities. However, companies with thousands of workers have millions of documents. It is prohibitively expensive to index them manually so that they can be found, analyzed and acted on. Moreover, overloaded workers do not have the time or training required to take on the task. The problem is exacerbated by mergers and acquisitions. In addition, over years, it is common for companies to accumulate large numbers of duplicate and out-of-date documents that workers do not take the time to rationalize and delete. The result is inflated storage costs and reduced productivity as workers struggle to find the relevant, up-to-date information. Inconsistent information governance also puts organizations at risk - litigation (retaining documents without legal or business value), safety (using out-of-date process safety management procedures) and operational (not leveraging best practices and lessons learned across the enterprise and beyond).This Small Business Innovation Research (SBIR) Phase II project addresses the problem that text documents - especially those internal to an organization - are very difficult to locate and analyze unless they are classified and tagged. But manual classification and tagging are too expensive and inconsistent for large collections. Large companies store many millions of documents. And there is even more relevant information on the Web. The objective of the proposed research to is to provide software assistants that classify documents into pre-specified categories, add tags to describe what each document is about, and the entities named in the documents (e.g., oilfields). The assistants identify relevant documents and help people to learn of new developments by sending alerts when new documents of interest appear on the web or in the company's computers. The primary technical result will be a suite of software assistants that companies can adopt singly or as an ensemble to help manage information sustainably. These assistants build upon the proposed research to develop and integrate novel approaches to unsupervised machine learning, concept identification, and ontology construction. They will enable companies to overcome major problems, including overload, finding relevant, up-to-date information, analyzing unstructured information, and identifying unneeded documents for elimination.