SBIR Phase I: Customizable Question Answering System for Homeland Security and Commercial Applications

点击翻译按钮获取中文摘要

This Small Business Innovation Research (SBIR) Phase I project proposes a novel Question Answering (QA) technology. The intellectual merits of the proposed activity are three fold: (1) it provides automatic decomposition of high level questions into logical, informative sets of fact-seeking questions, such that very complex questions can be automatically answered. This novel QA approach advances the state-of-the-art technology, which is currently limited to simple factual questions, enabling the answering of complex questions that model scenarios observed in actual customer environments. (2) It radically improves the accuracy of current state-of-the-art QA by using a logic prover to extract and justify answers. Language Computer Corporation (LLC) plans to develop an inference mechanism for question answering that is capable of extracting answers based on semantic inference chains, rather than on superficial keyword-based metrics. (3) It introduces a novel approach that adapts open-domain QA technology to domain-specific information using automatically acquired ontologies, seamlessly integrated with the system's open-domain knowledge base. This capability provides a rapid and efficient customization method for various domains of interests, such as weapons of mass destruction.The broader impact of the proffered technology is as follows: (1) the proposed approach allows QA technology to expand its capability, now restricted to synthetic evaluations based on simple, factual questions, to actual commercial applications with complex questions and scenarios. This places LLC in a position to target both government and commercial markets, where the accuracy, coverage, reliability and usability of the retrieved information are crucial. Ideal applications for this QA technology include homeland defense, Customer Relationship Management (CRM), education, medical, and legal. (2) The proposed model uses a logic proving mechanism that associates every extracted answer with a logical, easily understandable explanation of the answer correctness. Furthermore, LLC proposes the introduction of an automatic procedure to quickly adapt an open-domain QA system to domain-specific scenarios. This set of features makes this QA system an ideal tool for the intelligence business, where the quality of the information extracted is paramount, and where switching between completely different domains of interest is frequent.