Big Data Curation

大量结构化数据的广泛可用性使得采用算法和统计方法以新的方式解决数据集成问题成为可能。2009年的一份报告指出，Web上已经有超过1亿个结构化数据表。谷歌的Fusion Tables使发布和共享结构化和半结构化数据变得更加容易。然而，我们还远远不能自动集成和对齐这些数据，以便我们可以在查询回答、决策制定和预测中轻松地使用这些丰富的数据源。网络规模的语义集成是一个“巨大的挑战”问题。越来越多的证据表明，企业规模集成解决方案无法解决的一些细微的语义差异，现在可以使用来自web数据源组合的知识来发现和解决，其中一些是由重视可靠数据创建的激励用户社区策划的。这种数据的深度异构性和动态性不仅需要新的方法来校准，还需要新的方法来跟踪校准决策的来源，以及维护和发布一致的、正确对齐的数据。在这项工作中，我们将开发大数据管理的新方法。数据管理包括为重用或发布数据做准备，对上下文和语义进行适当建模，以及更一般地对数据进行维护，使其对用户具有最大价值。对于大数据，我们必须尽可能地自动化管理任务，并认识到重用用户社区（如开放本体）管理的数据的价值，以增强和指导发布新数据源时的自动化。

The wide-spread availability of vast amounts of structured data has made it possible to adopt algorithmic and statistical methods to solve data integration problems in new ways. A 2009 report indicated there were already over 100 million structured data tables on the Web. Efforts like Google's Fusion Tables make it easy to publish and share structured and semi-structured data. However, we are far from being able to automatically integrate and align this data so that we can easily use these rich data sources collectively in query answering, decision making, and prediction. Semantic integration at web-scale is a "grand challenge" problem. There is emerging evidence that some of the nuanced semantic differences that eluded enterprise-scale integration solutions, can now be discovered and resolved using knowledge from combinations of web data sources, some which have been curated by incentivized user communities who value the creation of reliable data. The deep heterogeneity and dynamic nature of this data demands new methods for not only alignment, but also for tracing the provenance of alignment decisions, and for maintaining and publishing consistent, properly aligned data. In this work, we will develop new methods for big data curation. Data curation involves the preparation of data for re-use or publishing, the proper modeling of context and semantics, and more generally the care of data to make it of maximal value to users. For big data, we must automate curation tasks to the extent possible and recognize the value of reusing data curated by user communities (such as open ontologies) to enhance and guide automation when publishing new data sources.