PROOF-IT Deep Semantic Learning System

PROOF-IT深度语义学习系统

• 批准号: 971576
• 金额: $ 6.18万
• 依托单位: PRECISIONLIFE LIMITED
• 依托单位国家: 英国
• 项目类别: Small Business Research Initiative
• 财政年份: 2017
• 资助国家: 英国
• 起止时间: 2017 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=971576
• 关键词: PROOF; Deep; Semantic; Learning; System

Title: PROOF-IT Abstract: The PROOF-IT Phase 1 project will apply deep semantic learning and indexing technology on the provided Innovate UK Funded Projects dataset and existing publicly available Innovate UK guidance and advice documents. It will use the same underlying semantic learning engine to provide proofs-of-concept for several use cases, including: 1. automated tools for more efficient and accurate assessor allocation, 2. rapid, automated detection of undeclared resubmissions, 3. a partnering portal, to rapidly identify potential collaborative R&D project partners, and 4. more accurate answering of users’ natural language questions in FAQ/support/enquiry forum settings There will be 3 proof-of-concept interfaces demonstrated in Phase 1 - an integrated grant application centric system combining use cases 1 & 2, and two search interfaces for use cases 3 and 4. The PROOF-IT system will be built on RowAnalytics’ existing spot.my deep semantic learning system, developed in part with Innovate UK support, which has been used to efficiently semantically analyse, index and search large text collections such as scientific, patent & legal literature, e-retail listings to identify brand protection violations, image collections for predicting the effectiveness of digital advertising and even complex representations such as chemical structure data. The spot.my deep semantic learning engine requires no training and so is much more efficient, flexible, cheaper and easier to operate & use than other Machine Learning and AI approaches. While learning the deep semantic relationships between concepts, crucially it does not require re-training or curation of new controlled vocabularies when new or previously unencountered technical domains and keywords are required to be matched. It can semantically index huge document sets in minutes using standard GPU computing, and the system support tens of millions of keywords and billions of documents. Document sets can be continually updated using incremental indexing to make information available within minutes of it becoming available. Phase 2 of the project will build on these applications, scaling out the document sets to covered to the full text of all applications received and automating full semantic equivalence checks for undeclared resubmissions (even when wording has been changed to obfuscate similarities). An assessor allocation tool will also be built based on semantic matching of the set of available assessors’ resumes to the received applications for a call and knowledge of other attributes for the assessors that Innovate UK thinks relevant, e.g. number/type of existing projects, conflicts of interest, location etc. The user facing information search tools could also be extended to include context sensitive and semantically powered chatbots for dealing with routine user enquiries, support requests and general Q&A. The semantic engine would also provide a substrate for detailed analysis and correlation of projects’ thematic and/or application features with outcome metrics such as project forecasting accuracy and monitoring reporting scores.

标题：PROOF-IT 摘要：PROOF-IT 第一阶段项目将在提供的 Innovate UK 资助项目数据集和现有的公开可用的 Innovate UK 指导和建议文件上应用深度语义学习和索引技术。它将使用相同的底层语义学习引擎为多个用例提供概念验证，包括：1.用于更高效、更准确地分配评估者的自动化工具，2.快速自动检测未声明的重新提交，3.一个合作门户，用于快速识别潜在的协作研发项目合作伙伴，以及4.在常见问题解答/支持/查询论坛设置中更准确地回答用户的自然语言问题 第一阶段展示了 3 个概念验证界面 - 一个以综合拨款申请为中心的系统，结合了用例 1 和 2，以及用例 3 和 4 的两个搜索界面。PROOF-IT 系统将构建在 RowAnalytics 现有的 Spot.my 深度语义学习系统上，该系统部分是在 Innovate UK 的支持下开发的，该系统已用于高效地对大型文本集合进行语义分析、索引和搜索，例如科学、专利和法律文献、电子零售 用于识别品牌保护违规行为的列表、用于预测数字广告有效性的图像集合，甚至是化学结构数据等复杂的表示形式。 Spot.my 深度语义学习引擎无需培训，因此比其他机器学习和人工智能方法更高效、灵活、更便宜且更易于操作和使用。在学习概念之间的深层语义关系时，至关重要的是，当需要匹配新的或以前未遇到的技术领域和关键字时，不需要重新训练或管理新的受控词汇。它可以使用标准 GPU 计算在几分钟内对庞大的文档集进行语义索引，并且系统支持数千万个关键字和数十亿个文档。可以使用增量索引不断更新文档集，以便在信息可用后几分钟内即可使用信息。该项目的第二阶段将建立在这些应用程序的基础上，扩展文档集以覆盖所有收到的应用程序的全文，并对未声明的重新提交自动进行完整的语义等效检查（即使措辞已更改以混淆相似性）。还将根据一组可用评估员简历与收到的电话申请的语义匹配以及创新英国认为相关的评估员其他属性的知识来构建评估员分配工具，例如现有项目的数量/类型、利益冲突、位置等。面向用户的信息搜索工具还可以扩展到包括上下文敏感和语义驱动的聊天机器人，用于处理日常用户查询、支持请求和一般问答。语义引擎还将为项目的主题和/或应用程序功能与项目预测准确性和监控报告分数等结果指标的详细分析和关联提供基础。