A Smart Big Data Analytics and Knowledge Management Framework

根据IBM的一项分析，我们每天创建2.5万亿字节的数据。在分析高速生成的大量不同类型的数据以及管理从数据分析中提取的知识方面，我们面临着巨大的挑战。目前，我们需要专家来识别数据，从而选择正确的分析工具、工作流程和存储系统。基于认知理论，人类在出生后创建并链接数据的配置文件，从而构建知识层，并使他们能够学习和查询知识以进行决策。我们需要一个数据分析框架，它可以像人类专家一样识别数据，通过使用正确的分析工具和工作流程创建数据配置文件，并存储和链接这些知识，以便能够有效地检索它，以进行基于目标的决策。例如，在决定病人的治疗过程时，医生需要查阅病人的医疗记录，其中包含结构化数据（如身高和体重）、非结构化文本注释、相关mri图像，并参考疾病数据库来调查具有相似症状的疾病。我们建议创建一个智能大数据分析和知识管理框架，该框架将应用机器学习模型和各种数据分析工具，从结构和语义内容中识别数据类型，从而应用正确的分析工作流来提取知识，以创建和链接数据配置文件。框架的知识管理部分将提取的知识存储到分布式和混合式知识库中，并创建多级图索引结构，根据目标、知识关联、上下文和情景高效检索正确的知识片段，为决策提供支持。由于大数据需要无所不在的访问和弹性资源，我们将在云资源上实施框架，使其作为云上的数据分析和知识服务可访问。该框架的原型将在三个跨学科用例场景中进行验证：1)使用新闻和金融数据分析预测我们的行业合作伙伴客户的市场状况；2)在初级保健电子病历（EMR）数据中定义和诊断PTSD（创伤后应激障碍），以及3)从金斯顿综合医院重症监护病房的流监测数据中分析心率变异性。拟议的研究项目将培养3名博士、3名硕士和2名本科生使用尖端的大数据分析系统，并培养当今全球市场上备受追捧的数据科学技能。从用例中可以看出，该研究将直接影响技术产业和医疗保健，为未来的智能决策支持系统建立支柱，并保持加拿大在建设智能数字世界方面的领导地位。

According to an IBM analysis, we create 2.5 quintillion bytes of data every day. We face an enormous challenge in analyzing this large volume of many different types of data being generated at a high speed, and managing the knowledge extracted from data analytics. Currently we need experts for recognizing the data, and thereby, selecting the right analytic tools, workflows, and storage systems. Based on cognitive theories, humans create and link profiles of data after birth, which builds layers of knowledge and enables them to learn and query the knowledge for decision making. We need a data analytic framework that can recognize the data like a human expert, create data profiles by using the right analytic tools and workflows, and store and link this knowledge to be able to retrieve it efficiently for goal based decision making. For example, when deciding on a patient's treatment course, a physician is required to consult the patient's medical records containing structured data such as height and weight, unstructured text notes, relevant MRI-image, and refer to a disease database to investigate diseases that have similar symptoms. We propose creating a smart big data analytics and knowledge management framework that will apply machine learning models and a variety of data analytics tools to recognize the data type from structure and semantic content, and thereby, apply the right analytic workflow to extract knowledge to create and link data profiles. The knowledge management part of the framework will store the extracted knowledge into distributed and hybrid knowledge repositories and create a multi-level graph indexing structure to efficiently retrieve the right knowledge pieces for decision support based on goal, knowledge association, context and situation. Due to the necessity of ubiquitous access and elastic resources for big data, we will implement the framework on cloud resources and make it accessible as a data analytic and knowledge service on the cloud. The prototype of the framework will be validated for three interdisciplinary use case scenarios: 1) predicting market status of the clients of our industry partner using news and financial data analytics; 2) defining and diagnosing PTSD (Post Traumatic Stress Disorder) in primary care electronic medical record (EMR) data, and 3) analyzing heart rate variability from streaming monitoring data from the intensive care unit at the Kingston General Hospital. The proposed research program will train 3 PhD, 3 MSc and 2 undergraduate students in using cutting edge big data analytics systems and developing data science skills that are highly sought after in the global market today. As evident from the use cases, the research will directly impact the technology industry and health care, build the backbone for the future intelligent decision support systems, and sustain Canada's leadership in building a smart digital world.