III: Small: Knowledge Graph Query Processing and Benchmarking

今天，如果用户在b谷歌或必应上有问题，她仍然需要浏览多个网页才能找到答案。由于移动设备的兴起，这种模式正在发生变化。在过去的十年里，我们见证了许多系统的目标是直接回答查询，例如，使用从互联网上收集的知识图或通过众包。信息搜索领域真正的巨变即将到来！智能警务、个人协助、个性化医疗保健、法律服务、科学文献搜索以及最近的机器人技术等领域正在出现广泛的新应用。该项目将服务于这些应用程序，并在查询无处不在的异构知识图方面取得根本性进展。它将大大简化这些应用程序中的查询公式，并提高搜索质量/速度。考虑到知识图中数据的高度异构性，对于普通用户来说，编写完全符合数据规范的结构化查询是极其困难的，而关键字查询可能过于模糊，无法反映用户的搜索意图。当同一实体或关系有不同的表示时，情况会变得更糟。期望一个复杂的查询系统能够支持不同的概念表示，而不强迫用户使用非常受控制的词汇表。它应该为用户提供简单的机制，以便他们能够快速地提出一个正确的查询，无论是显式的还是隐式的（例如，通过相关性反馈）。本提案将开发这样的系统，使其具有用户友好性和可扩展性。该研究计划建立一个灵活的查询基准，该基准能够处理异构、大规模的知识图，以及用户指定的配置和性能指标。基准测试对于数据库研究的快速发展是必不可少的。有许多成功的例子表明，强大而有意义的基准可以极大地加快一个研究领域的发展。本项目提出的查询基准是非常必要的。它将(1)提供一种标准化的方法来公平和全面地评估不同的知识图查询算法，(2)提高对现有查询引擎的理解，(3)通过让研究人员参与到构建更好、更快、更智能的方法的同一领域来推进这一领域。欲了解更多信息，请参阅该项目的网站：http://www.cs.ucsb.edu/~xyan/kg.html

Today, if a user has a question, using Google or Bing, she still has to read through multiple web pages to find answers. This paradigm is now changing due to the rise of mobile devices. Over the last decade, it was witnessed that many systems aim to answer queries directly, e.g., using knowledge graphs collected from the Internet or through crowdsourcing. A real sea change in information search is coming! A broad range of new applications are emerging in intelligent policing, personal assistance, individualized healthcare, legal services, scientific literature search, and recently robotics. This project will serve these applications and make fundamental advances in querying heterogeneous knowledge graphs, which are ubiquitous. It is going to significantly ease query formulation and improve search quality/speed in these applications.Given the high data heterogeneity in knowledge graphs, writing structured queries that fully comply with data specification is extremely hard for ordinary users, while keyword queries can be too ambiguous to reflect user search intent. The situation becomes even worse when there are various representations for the same entity or relation. It is expected that a sophisticated query system shall be able to support different concept representations without forcing users to use very controlled vocabulary. It shall provide simple mechanisms to users so that they can quickly come up with a right query either explicitly or implicitly (e.g., via relevance feedback). This proposal is going to develop such system, make it user-friendly and scalable. The proposed research includes a plan to build a flexible query benchmark that is able to cope with heterogeneous, large-scale knowledge graphs, as well as user specified configurations and performance metrics. Benchmarks are indispensable for rapid development of database research. There were many successful examples of how robust and meaningful benchmarks can greatly expedite the development of a research area. The query benchmark proposed in this project is very needed. It is going to (1) provide a standardized way to fairly and comprehensively evaluate different knowledge graph query algorithms, (2) improve the understanding of the existing query engines, and (3) advance the area by getting researchers involved in the same play ground for building better, faster, and more intelligent methods.For further information see the project web site at: http://www.cs.ucsb.edu/~xyan/kg.html