Enabling queries on relational data on the Web

虽然万维网为公众提供了对有史以来创建的最大、最异构的数据集合之一的访问，但它的查询功能仍然非常局限于关键字搜索。这降低了搜索的效率和可以探索或检索的信息的范围。更具体地说，当搜索不限于选择并且涉及对文档内容的连接和设置操作时，关键字搜索是不够的。例如，关键字搜索不适合查询关系，而Web包含大量的关系。同样，当搜索结果的粒度小于文档，并且结果是满足某些条件的一组更精细的项或实体时，关键字搜索也不太适用。本提案旨在研究和解决一些阻碍在Web上查询关系数据的问题，特别是：(1)在提取和聚合信息方面，我们将研究不同编码方案之间的关系，以及两种编码可以引用或描述同一事实或表达同一关系的条件。例如，理解这种关系是检测查询是否与数据片段匹配、聚合相关片段和过滤不相关片段的关键。(2)在相关性排序方面，我们将研究能够解释事实或关系更可靠或可信的条件以及可信度与相关性之间关系的模型。(3)在数据清理和过滤方面，我们将探索可用于清理可能在结果中引入噪声的资源的策略和过滤器，以及更多可扩展的数据结构和算法来实现这些过滤器。(4)更好地理解上述问题和我们开发的解决方案将使我们能够寻找更高级的搜索和查询公式。研究这些公式将进一步为我们提供更多机会来探索适用于Web设置的更有效的查询评估策略。

While the World Wide Web provides public access to one of the largest and most heterogeneous data collections ever created, its querying capabilities have remained very much limited to keyword search. This reduces the efficiency of the search and the scope of the information that can be explored or retrieved. More specifically, a keyword search is not sufficient when the search is not bounded to a selection and involves join and set operations over the contents of the documents. For instance, a keyword search is not appropriate for querying relationships, and the Web contains relationships galore. Also, a keyword search is not very applicable when the granularity of the search result is smaller than a document, and the result is a more refined set of items or entities that satisfy some conditions.This proposal is to study and to address some of the issues that hinder querying relational data on the Web, in particular: (1) On extracting and aggregating information, we will study the relationships between different encoding schemes and the conditions under which two encodings may refer to or describe the same fact or express the same relationship. Understanding such relationships is a key in, for example, detecting if a query matches a piece of data, aggregating related pieces and filtering unrelated pieces. (2) On relevance ranking, we will be investigating models that can explain the conditions under which a fact or relationship is more reliable or trustworthy and the relationships between trustworthiness and relevance. (3) On data cleaning and filtering, we will explore strategies and filters that can be used to clean up resources that may introduce noise in the result, and more scalable data structures and algorithms to implement those filters. (4) A better understanding of the aforementioned issues and the solutions that we develop will enable us to look for more advanced searching and querying formulations. Investigating these formulations would further provide us with more opportunities to explore more efficient query evaluation strategies that are applicable in the setting of the Web.