OASIS: Ontology Reasoning over Frequently-changing and Streaming Data

依赖于对松散结构和大规模数据集的智能管理的高级应用程序在医疗保健、商业和政府等领域发挥着关键作用。基于本体的技术是许多此类应用的核心。简而言之，基于本体的数据管理系统(ODMS)通过提供用于在本体中表示关于应用的背景知识的手段，并利用自动推理技术来推断数据和本体中隐含的信息，从而实现智能信息处理。然而，现有的ODMS不太适合需要实时分析快速变化的数据的应用。例如，石油和天然气公司持续监测传感器读数，以检测设备故障并预测维护需求；网络提供商分析流量数据，以识别流量异常和拒绝服务攻击；知识图不断更新；物联网(IoT)应用，如智能城市，需要对来自多种类型设备的数据进行实时分析。ODMS经常从数据库文献中借用实现技术，其中对快速变化的数据的实时分析使用两种主要方法。(1)在流处理系统中，输入数据在概念上被视为流经系统的带有时间戳的元组的无限序列；数据只能在一次处理中使用，系统存储的信息本身就不完整。流作业是长期运行的：查询只部署一次，并继续产生结果，直到被删除。最先进的系统，如阿帕奇风暴、阿帕奇火花流、谷歌的Millheel、Linked In的Samza和阿帕奇Flink，通过在集群中分配流工作负载来实现亚秒级延迟，这需要复杂的调度和容错技术。(2)在实时数据库中，数据被视为不断演变的记录的有限集合。这种传统的有限和持久集合的概念在数据库世界中无处不在，非常适合需要一致和完整的数据视图的应用程序。实时数据库与传统数据库的关键区别在于，与流系统类似，它们允许客户端订阅长时间运行的连续查询，这些查询可以即时推送增量更新。然而，在使这些办法适用于药品监督管理系统时，出现了许多理论和实践上的困难。在OASIS项目中，我们将解决这些困难，并为能够实时摄取和处理快速变化的数据的新一代ODMS奠定基础。这类系统将通过支持智能决策的复杂分析管道的快速执行来支持上述应用程序。此外，我们将利用所获得的见解来实现原型并在实际部署中对其进行测试。

Advanced applications relying on intelligent management of loosely structured and large-scale datasets play a key role in domains such as healthcare, business and government. Ontology-based technologies lie at the core of many such applications. In a nutshell, an ontology-based data management system (ODMS) enables intelligent information processing by providing means for representing background knowledge about the application in an ontology, and exploiting automated reasoning techniques to infer information that is implicit in the data and the ontology.State-of-the art ODMSs are, however, not well-suited for applications which require real-time analysis of rapidly changing data. For instance, oil and gas companies continuously monitor sensor readings to detect equipment malfunction and predict maintenance needs; network providers analyse flow data to identify traffic anomalies and Denial of Service attacks; knowledge graphs are continuously updated; and Internet of Things (IoT) applications such as Smart Cities require real-time analysis of data stemming from multiple types of device.ODMSs often borrow implementation techniques from the database literature, where real-time analysis of rapidly changing data has been tackled using two main approaches.(1) In a stream processing system, the input data is conceptually seen as an unbounded sequence of time-stamped tuples that flow through the system; data is only available for processing in a single pass and information stored by the system is inherently incomplete. Streaming jobs are long-running: queries are deployed once and continue to produce results until removed.State-of-the art systems, such as Apache Storm, Apache Spark Streaming, Google's Millwheel, Linked In's Samza, and Apache Flink, achieve sub-second latencies by distributing the streaming workload in a cluster, which requires sophisticated scheduling and fault-tolerance techniques. (2) In a real-time database, the data is seen as a finite collection of records that is continuously evolving. This traditional concept of a finite and persistent collection is ubiquitous in the database world is well-suited for applications requiring a consistent and complete view of the data.The key feature that distinguishes real-time from traditional databases is that, similarly to streaming systems, they allow clients to subscribe to long-running continuous queries that instantaneously push incremental updates. Many theoretical and practical difficulties arise, however, when adapting these approaches to ODMSs. In the OASIS project, we will address these difficulties and lay the foundations for a new generation of ODMSs capable of ingesting and processing rapidly changing data in real time. Such systems will support the aforementioned applications by enabling fast execution of complex analytics pipelines supporting intelligent decisions. Moreover, we will exploit the resulting insights to implement a prototype and test it in real-life deployments.