CAREER: Novel Summarization Techniques for Semi-Structured Data

该项目的总体目标是探索半结构化数据的新型摘要模型，以便在以数据为中心和以文档为中心的数据库中实现近似查询应答。研究了以下关键问题：（a）以数据为中心的XML数据库的近似查询应答。该项目探讨了总结模型，捕捉半结构化数据集的结构和基于值的特征。所开发的模型考虑到了价值内容的异质性，并且当基础数据被修改时可以动态地维护。(b)以文档为中心的XML摘要。摘要模型的探索，使近似计算的距离，根据各种指标，摘要文档和输入查询之间。该项目研究了该模型在文档检索查询的近似查询应答中的应用，以及其在查询处理引擎中的集成，以快速计算候选匹配。(c)连接查询的关系概要。研究了一类新的关系概要，它使用半结构化数据的摘要模型来近似关系数据库的统计特征。这项研究的结果适用于涉及探索大型半结构化数据集的领域，例如，科学数据库或数字图书馆。为了便利对这些目标领域的拟议技术进行评价，将开发系统原型并向公众提供。该项目将对研究生和本科生进行数据简化和近似技术方面的培训，而由此产生的出版物、收集的数据集和开发的软件将在更广泛的网站上提供。dissemination.http://www.cs.ucsc.edu/~alkis/APX/

The overall goal of this project is to explore novel summarization models for semi-structured data, in order to enable approximate query answering within data-centric and document-centric databases. The following key problems are investigated:(a) Approximate query answering for data-centric XML databases. The project explores summarization models that capture both the structural and the value-based characteristics of a semi-structured data set. The developed models take into account the heterogeneity of the value content and can be dynamically maintained when the base data is modified.(b) Document-centric XML summarization. A summarization model is explored that enables the approximate computation of distance, under various metrics, between a summarized document and an input query. The project investigates the application of this model in approximate query answering for document retrieval queries, as well as its integration within the query processing engine for the fast computation of candidate matches.(c) Relational synopses for join queries. A novel class of relational synopses is explored that uses summarization models for semi-structured data in order to approximate the statistical characteristics of a relational database.The results of this research are applicable in areas that involve the exploration of large semi-structured data sets, e.g., scientific databases or digital libraries. To facilitate the evaluation of the proposed techniques in these target areas, system prototypes will be developed and be made publicly available. The project will train graduate and undergraduate students on data reduction and approximation techniques, while the resulting publications, collected data sets, and developed software will be made available on the web for broader dissemination.http://www.cs.ucsc.edu/~alkis/APX/