A Deterministic Model for Semistructured and Structured Data

随着互联网的发展，数据库正在以前所未有的规模被复制、转换和修改。随之而来的是对数据进行注释的需要：为了纠正数据、评论数据或记录数据的来源。然而，传统的数据库具有规则的结构，并且没有为注释提供“空间”，这些注释通常是不规则的，并且在所有粒度级别上都需要。注释需要某种形式的半结构化表示。本文研究了一种新的结构化和半结构化数据模型，该模型旨在促进数据标注过程。通常的半结构化数据模型是基于边缘标记的图，并且是“不确定的”，因为给定的顶点可能有几个具有相同标签的外边缘。这里提出的模型更具限制性，因为这些标签都必须不同。它的限制性较小，因为标签本身可能携带数据。事实上，标签本身可能是一小块半结构化数据。这种模式有许多优点。与XML一样，数据表示完全符合语法。所有的操作都是语法性的，并且数据的序列化是即时的。特别是对象标识符是语法结构，它们具有结构，并且可以通过对该结构的操作来实现某些数据库转换。调查的目标是开发模型和相关接口，以便它可以用作半结构化和结构化数据的公共表示。第二个目标是开发数据注释工具。http://db.cis.upenn.edu/Research/

With the development of the Internet, databases are being copied, transformed, and corrected on an unprecedented scale. With this comes the need to annotate data: for correcting it, commenting on it, or for recording its provenance. However, conventional databases have a regular structure and provide no "room" for annotations which are typically irregular and required at all levels of granularity. Annotations require some form of semistructured representation. This investigation centers on a new model for structured and semistructured data which is intended to facilitate the process of data annotation. The usual models of semistructured data are based on an edge-labeled graph and are "nondeterministic" in that a given vertex may have several out-edges with the same label. The model proposed here is more restrictive in that these labels must all be different. It is less restrictive because the labels themselves may themselves carry data. In fact a label may itself be a small piece of semistructured data. A number of advantages are claimed for this model. Like XML, the data representation is entirely syntactic. All operations are syntactic,, and serialization of data is immediate. Object identifiers in particular are syntactic constructs, they have structure, and certain database transformations may be achieved by manipulation of this structure. The goal of the investigation is to develop the model and associated interfaces so that it can be used as a common representation for both semistructured and structured data. A second goal is to develop tools for data annotation. http://db.cis.upenn.edu/Research/