SBIR Phase I: Authoring Assistance via Contextual Semantic Labeling

这个小型企业创新研究（SBIR）第一阶段项目的更广泛影响是推进自然语言处理（NLP），以提高企业的生产力，合规性和洞察力。文件是业务的基础结构，因为它们包含与客户、供应商、合作伙伴和监管机构之间的详细协议、义务、要求和条款，对业务运营至关重要。然而，文档仍然代表“暗数据”，独立于自动化业务流程，无法访问。商业真实的房地产、保险、专业服务、金融服务、法律的事务所等企业产生和消费包含具有无数变化的类似模式的许多文档。创作和执行这些协议是费力且容易出错的，但是很难自动化这些半结构化信息的使用。该项目开发了一系列复杂的步骤，从叙事文本中识别结构和信息，应用人工智能中几个思想流派的最新技术。该项目将使知识工作者能够获得人工智能的帮助，以更轻松、高效、精确、保密、合规和洞察力的方式撰写和执行商业协议。这个小型企业创新研究（SBIR）第一阶段项目是通过结构化的计算机优化版本来增强非结构化的以人为本的文本，每个文档的“影子”表示，使用XML和数据库技术为用户和组织提供创新的软件帮助。该研究采用多方面的方法，应用计算机视觉，然后使用深度学习，贝叶斯，进化，符号和经典NLP技术创建新算法的管道。 该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact of this Small Business Innovation Research (SBIR) Phase I project is to advance Natural Language Processing (NLP) to improve productivity, compliance and insight for businesses. Documents are the underlying fabric of business as they hold detailed agreements, obligations, requirements and terms central to business operations with customers, suppliers, partners and regulators. However, documents still represent "dark data", separate and inaccessible to automated business processes. Businesses like commercial real estate, insurance, professional services, financial services, legal firms and many others produce and consume many documents containing similar patterns with innumerable variations. Authoring and executing these agreements is laborious and error-prone, but it is difficult to automate the use of this semi-structured information. This project develops a series of sophisticated steps to discern structure and information from narrative text, applying the latest techniques from several schools of thought in artificial intelligence. This project will enable knowledge workers to gain the assistance of artificial intelligence to author and execute commercial agreements with greater ease, efficiency, precision, confidentiality, compliance and insight.This Small Business Innovation Research (SBIR) Phase I project is to enhance unstructured human-centered text with a structured computer-optimized version, a "shadow" representation of each document that uses XML and database technology to enable innovative software assistance for users and organizations. The research takes a multi-faceted approach, applying computer vision and then creating a pipeline of new algorithms using techniques from Deep Learning, Bayesian, Evolutionary, Symbolic and Classic NLP. The process operates on "small" datasets (10-30 documents) with high accuracy as well as large datasets.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.