Knowledge Graph Mining for the Linked Open Data Cloud

在过去几年中，通过关联数据范例(也称为数据网络)在Web上开发结构化数据受到了相当大的关注，特别是随着Google和Microsoft等公司采用知识图来增强其搜索引擎。链接开放数据云(LOD)是一组跨域互联的数据集，代表了数据网络的最新发展。我们见证了已发布的RDF数据集数量的巨大发展。LOD的主要兴趣之一是它有可能提供语义搜索能力，例如使用直接答案而不是文档集来回答查询，以及a)从各种知识来源找到答案和b)通过推理机制推断答案的能力。 然而，发布的数据集的数量带来了新的挑战，因为没有任何既定的机制来确保它们的质量。特别是，在没有适当的模式和本体模型的情况下泛滥的RDF数据对有效的查询回答几乎没有兴趣。另一个挑战是当前Web内容的表示和覆盖，特别是领域知识。尽管数据网络在增长，但大多数网络内容仍然是以非结构化格式和文本表示的。因此，有必要用高质量的数据和模式扩展当前的LOD；开发从非结构化Web内容中学习本体模式和知识库的方法和工具；测量和评估LOD质量，并为当前Web结构化数据设计纠正和完成策略；以及开发具体显示Web数据对查询回答的兴趣的用例。 在这项发现基金中，我们将专注于网络上的一种特殊资源，即跨域百科全书维基百科。在Web of Data的上下文中，Wikipedia链接到LOD上的主要资源之一DBpedia，DBpedia是基于Wikipedia信息框和类别的Wikipedia的RDF表示。作为LOD上的枢纽，DBpedia已经成为几个语义分析任务的中心资源。特别是，DBpedia是几个新的工业和学术语义标注服务(例如IBM的AlChemy、DBpedia Spotlight)的骨干，这些服务用于标记Web内容并识别文本中的实体和概念。然而，DBpedia面临着之前在覆盖维基百科内容、缺乏适当的本体模式和事实错误方面所描述的相同的质量问题。该计划的目标是开发工具和方法来纠正、公理和扩展DBpedia知识库。我们还将展示学习的知识库对语义Web上下文中更好的语义标注和查询回答的兴趣。

The development of structured data on the Web, through the linked data paradigm (aka the Web of data), has received considerable attention during the last few years, especially with the adoption of knowledge graphs by companies such as Google and Microsoft to enhance their search engines. The Linked Open Data Cloud (LOD), a set of interconnected data sets across domains, represents the latest developments of the Web of data. We have witnessed a huge development in the number of published RDF datasets. One of the main interests of the LOD is its potential to offer semantic search capabilities such as query answering with direct answers instead of sets of documents, and the ability to a) find answers from various knowledge sources and b) infer answers through reasoning mechanisms. However, the quantity of published datasets poses new challenges as there aren't any established mechanisms to ensure their quality. In particular, the deluge of RDF data without proper schemas and ontological models is of little interest for efficient query answering. Another challenge is the representation and coverage of current Web content, especially for domain knowledge. Despite the growth of the Web of data, the majority of Web content is still represented in unstructured formats and texts. Thus there is a necessity to expand the current LOD with data and schemas of good quality; to develop methods and tools that learn ontological schemas and knowledge bases from unstructured Web content; to measure and evaluate LOD quality and design correction and completion strategies for current Web structured data; and to develop use cases that concretely show the interest of the Web of data for query answering. In this discovery grant, we will focus on one particular resource on the Web which is the cross-domain encyclopedia Wikipedia. In the context of the Web of data, Wikipedia is linked to one of the main resources on the LOD, DBpedia, which is the RDF representation of Wikipedia based on Wikipedia infoboxes and categories. Being a hub on the LOD, DBpedia has become a central resource for several semantic analysis tasks. In particular, DBpedia is the backbone of several new industrial and academic semantic annotation services (e.g. IBM's Alchemy, DBpedia Spotlight) which are used to tag Web content and recognize entities and concepts in text. However, DBpedia suffers from the same quality problems previously described in terms of coverage of Wikipedia content, lack of proper ontological schema, and factual errors. The objective of this program is to develop tools and methods to correct, axiomatize and expand the DBpedia knowledge base. We will also demonstrate the interest of the learned knowledge base for better semantic annotation and query answering in the context of the Semantic Web.