Evolution AI, WMG Data Science Group, Dun & Bradstreet: Artificial Intelligence-Driven Information Extraction from Documents

许多企业以商业文档(如合同、发票、交货单、工资单)和财务文档(如合规声明、监管文件、账户)的形式持有大量的半结构化信息作为扫描图像或PDF。这些复杂的文档不容易被当今可用的技术处理或‘理解’，这些技术包括自然语言处理(NLP)、一种机器学习和光学字符识别(OCR)。在这个项目中，我们建议开发一种产品，从包含表单和表格的半结构化文档中准确地提取信息，而不需要人工键入。这类用例的准确率目前非常低，约87%，远远低于人类的认知能力。合作伙伴提议使用颠覆性的大数据方法来解决这个问题，该方法利用了最近从研究基地涌现的一类新算法。我们的项目是一个高度创新的产业研发合作项目。它由提供自主阅读和理解人类语言产品的英国中小企业Evolution AI领导；由全球市场信息和数据解决方案提供商Evolution的客户(Dun&BradStreet)支持；以及世界领先的大学(华威大学、WMG数据科学集团)。我们将基于Evolution AI在NLP中的能力及其专有的大数据处理架构，以及WMG在深度学习和计算机视觉以及邓白氏数据方面的专业知识，开发基于深度人工神经网络的新算法，最终形成市场就绪的产品。我们的创新提供了戏剧性的机会，以提高后台处理的效率，引入自动化和提高生产率；还提供了大规模获取新的智能知识的机会。我们预计我们的产品将为Evolution AI的客户创造显著的生产力提升，并帮助其加速增长，到2023年成为价值34-45亿美元的出口市场，估计每年增长16%-38%。到2023年，这将创造至少42个新的就业机会，为公共财政带来大约7-12倍的投资回报率。演进人工智能成立于2015年，提供五项核心功能：1\。信息提取；2.语义匹配；3.深度分类；4.自主网络爬行；5.自然语言生成。Evolution AI专有的自然语言处理(NLP)算法利用了使用神经网络的深度学习技术的最先进进展，运行在我们的高性能计算架构上。其客户是银行、保险、专业服务、政府、卫生、零售和媒体等领域的主要蓝筹组织。

Many businesses hold vast volumes of semi-structured information in the form of business documents (e.g. contracts, invoices, delivery notes, payslips) and financial documents (e.g. compliance statements, regulatory filings, accounts) as scanned images or pdfs. These complex documents are not easily processed or 'understood' by technologies available today, including Natural Language Processing (NLP), a type of machine learning, and Optical Character Recognition (OCR). In this project we propose to develop a product to extract information accurately from semi-structured documents containing forms and tables, obviating the need for manual keying by human beings. Accuracy rates for such use-cases are currently very low, ~87%, well below human cognitive capability.The partners propose to solve this problem using a disruptive 'big data' approach exploiting a new class of algorithms which has recently emerged from the research base. Our project is a highly innovative industrial R&D collaboration. It is led by between Evolution AI, a UK SME which provides products that read and understand human language autonomously; backed by Evolution's customer (Dun & Bradstreet), the global market information and data solutions provider; and a world-leading university (University of Warwick, WMG Data Science Group). We will build on Evolution AI's capability in NLP and its proprietary big data processing architecture, along with WMG's expertise in deep learning and computer vision and Dun & Bradstreet's data, to develop new algorithms based on deep artificial neural networks, culminating in a market-ready product.Our innovation provides dramatic opportunities to make efficiencies in back-office processing, introduce automation and improve productivity; and also to capture new, intelligent knowledge at scale. We expect our product to create significant productivity gains for Evolution AI's customers and to help accelerate its growth into an export market worth $3.4 - $4.5Bn by 2023 growing at an estimated 16% - 38% per year. This will create at least 42 new jobs and an ROI to the public purse of approximately 7-12 times investment by 2023\.Evolution AI was incorporated in 2015 and provides five core capabilities: 1\. Information extraction; 2\. Semantic matching; 3\. Deep categorisation; 4\. Autonomous web crawling; 5\. Natural language generation. Evolution AI's proprietary Natural Language Processing (NLP) algorithms exploit state-of-the-art advances in deep learning technology using neural networks, running on our high-performance compute architecture. Its clients are major blue-chip organisations in banking, insurance, professional services, government, health, retail and media.