Score!: Scalable and Complete Reasoning with Incomplete Ontology Reasoners

工业、政府和医疗保健的决策越来越依赖于改善数字信息的获取和处理。这导致迫切需要更强大和更灵活的信息系统。新一代信息系统需要有效地处理大数据集，利用机器可读的领域知识，并回答查询，同时考虑知识和数据。基于本体的信息系统（OIS）构成了一个快速成熟的技术，以满足这些要求的潜力。本体提供用户熟悉的术语的词汇表，以及描述这些术语的含义的公理。OIS可以利用本体中丰富的领域知识来提供统一的数据视图，并使用自动推理机使用隐含的信息丰富查询答案。OWL及其修订版OWL 2提供了一种强大而灵活的本体建模语言，它可以捕获诸如类层次结构、不完整信息、否定信息等特征。OWL本体正在越来越多的应用中使用，并正在成为访问、收集和共享知识和数据的核心技术。然而，涉及大量数据的应用，仍然对OIS的适用性构成严重挑战。OIS的适用性问题通常源于相互冲突的应用需求。对复杂的应用领域建模需要丰富的本体语言。细粒度的信息访问需要强大的查询语言。在大型数据集上进行查询需要可扩展的推理机。依赖于信息访问的关键决策要求查询答案是完整的，或者在不完整的地方是很好理解的。由于相关推理问题的最坏情况下的高复杂性，可扩展性通常是在使用强大的本体和查询语言的冲突，许多应用程序放弃完整性，以实现所需的可扩展性。因此，现有的OIS无法满足这些要求中的一个或多个：它们只支持弱本体或查询语言，它们不能扩展到所需的数据量，或者它们不能保证查询答案的完整性。我们在这个项目中的目标是为满足所有上述要求的新一代OIS奠定基础，因此，提供了表达能力，可伸缩性和完整性的理想组合。为了实现这样一个雄心勃勃的目标，我们观察到表达能力，可伸缩性和完整性之间的权衡所施加的限制适用于语言级别：也就是说，它们涉及以给定的本体、查询和数据建模语言表达的每个本体、查询和数据集的最坏情况复杂度界限。然而，与特定应用程序相关的本体、查询和数据集的类别受到更多限制。例如，尽管应用程序数据通常是未知的或频繁变化的，但本体本身在设计时是固定的，或很少变化。因此，对于给定的查询和本体语言，已知一般不完整的推理器可能会产生与手头的应用程序的完整推理器相同的结果。识别这样的情况是具有挑战性的，但它会有巨大的附加值：应用程序可以利用可扩展的不完整的推理，同时仍然享有完整性保证，从而实现“两全其美”。我们相信，我们的主要目标可以通过设计OIS，优化的本体，查询和数据集相关的应用程序在手。这样的OIS将最大限度地提高可扩展性，同时确保查询答案的完整性，即使是对于丰富的本体，大规模的数据集和复杂的用户查询。

Decisions in industry, government and health care increasingly depend on improved access to and processing of digital information. This has led to a pressing demand for more powerful and flexible information systems. New generation information systems will need to efficiently process large data sets, exploit machine-readable domain knowledge, and answer queries by taking into account both knowledge and data.Ontology-based information systems (OISs) constitute a rapidly maturing technology with the potential to meet these requirements. An ontology provides a vocabulary of terms that are familiar to the user, together with axioms describing the meaning of those terms. OISs can exploit the rich domain knowledge in an ontology to provide a unified view of the data and enrich query answers with implicit information using an automated reasoner.Several standards for ontology and query languages have been developed, including RDF, OWL, OWL 2, and SPARQL. OWL and its revision OWL 2 provide a powerful and flexible ontology modelling language that can capture features such as class hierarchies, incomplete information, negative information, and so on. OWL ontologies are being used in an increasing range of applications, and are becoming a core technology for accessing, gathering, and sharing knowledge and data.Applications involving large amounts of data, however, still pose serious challenges to the applicability of OISs. Problems in the applicability of OISs typically originate from conflicting application requirements.- Modelling complex application domains requires rich ontology languages.- Fine-grained access to information requires powerful query languages.- Answering queries over large data sets requires scalable reasoners.- Critical decisions that depend on access to information require query answers that are either complete, or where the incompleteness is well-understood.Due to high worst-case complexity of the relevant reasoning problems, scalability is usually in conflict with the use of powerful ontology and query languages, and many applications give up completeness to achieve the desired scalability. As a result, existing OISs fail to meet one or more of these requirements: they support only weak ontology or query languages, they do not scale to the required volumes of data, or they do not provide guarantees as to the completeness of query answers.Our goal in this project is to lay the foundations for a new generation of OISs that meet all the aforerementioned requirements, thus providing the ideal combination of expressive power, scalability and completeness.To accomplish such an ambitious goal, we observe that the limitations imposed by the trade-offs between expressivity, scalability and completeness apply at the language level: that is, they involve worst-case complexity bounds for every ontology, query, and data set expressed in given ontology, query and data modeling languages. The class of ontologies, queries and data sets that are relevant to a particular application is, however, much more restricted. For example, although application data is often unknown or frequently changing, the ontology itself is fixed at design time, or changes infrequently. As a result, a reasoner known to be incomplete in general for given query and ontology languages might yield the same results as a complete reasoner for the application at hand. Identifying such cases is challenging, but it would have tremendous added value: applications could exploit scalable incomplete reasoners while still enjoying completeness guarantees, thus achieving 'the best of both worlds'.We believe that our main goal can be accomplished by designing OISs that are optimised for the ontologies, queries and data sets relevant to the application at hand. Such OISs would maximise scalability while ensuring completeness of query answers, even for rich ontologies, large-scale data sets, and complex user queries.