Data Mining with Linked Open Data (Mine@LOD)

数据挖掘专注于在大数据集中发现模式和规律性。常见的方法假设所有相关数据都存储在一个数据库或数据仓库中，然后使用机器学习或数据挖掘方法扫描该数据库或数据仓库中的模式。随着语义网的不断发展和公共链接数据源的全球可获得性，如何利用这些数据源进行数据挖掘是一个开放的研究问题。在许多现实世界的用例中，为了找到相关和有趣的模式，需要额外的背景知识。例如，当考虑到图书的详细信息(如图书体裁)和书店的社区(如人口结构)时，可以更好地解释不同书店的图书销售数字。由于为所有可能用例提供背景信息几乎是不可能的，因此，MINE@LOD项目旨在开发一种方法，以一种完全自动化的方式，用链接的开放数据中的背景知识丰富数据集。我们预计，有了这些方法，现有的数据挖掘方法可以得到显著改进。为此，必须确定合适的数据源，找到这些数据源中的相关信息，并改进现有的学习和挖掘算法，使它们能够在存在大量弱相关数据源的情况下提供合理的结果。

Data Mining is focused on finding patterns and regularities in large data sets. Common approaches assume all of the relevant data is stored in one database or data warehouse, which is then scanned for patterns using machine learning or data mining methods. With a growing development of the Semantic Web and the global availability of public linked data sources, it is an open research question how to leverage those data sources for data mining processes.In many real-world use cases, additional background knowledge is required in order to find relevant and interesting patterns. For example, better explanations for sales figures of books in different book stores can be found when taking into account detail information both on the books (such as their genre) and the neighborhoods of the book stores (such as the population structure). Since it is hardly feasible to provide that background information for all possible use cases, the project Mine@LOD aims at developing approaches for enriching data sets with background knowledge from Linked Open Data in a fully automated way. We expect that with those approaches, existing data mining methods can be significantly improved. To that end, suitable data sources have to be identified, the relevant information within those data sources needs to be located, and existing learning and mining algorithms have to be improved in a way that they can deliver reasonable results in the presence of a large number of weakly relevant data sources.