Ranked search over structured and semi-structured data

Abstract

Traditionally, relational database systems have been designed for precise queries over structured data, while information retrieval systems have been designed for more flexible ranked keyword search queries over unstructured (text) data. However, many new and emerging applications require data management capabilities that combine the benefits of database and information retrieval systems, e.g., e-commerce and content management applications. In this dissertation, we have proposed and initiated steps toward the larger goal of integrating relational database and information retrieval systems.

First, we consider the problem of ranked text search in relational databases, where the traditional ranking paradigms and techniques developed for stand-alone unstructured documents are not directly applicable. We thus propose a new ranking paradigm that uses structured data values to score the results of text search queries. Our experimental results on real and synthetic data sets show that we can support the new ranking paradigm efficiently in relational databases.

Second, we explore a novel problem of discovering rich relationships in databases based on user queries with text search and structural query conditions. Toward this end, we have introduced the notion of data topology and developed efficient algorithms for computing ranked topologies based on user queries. We have evaluated our algorithms using a real biological database, the Biozon database (http://www.biozon.org).

Third, we consider the problem of efficiently producing results for keyword search queries over semi-structured XML documents. We present the XRANK system that is built to address the novel challenges for effective XML keyword search as apposed to HTML keyword search. Our experimental results show that XRANK offers both space and performance benefits when compared with existing approaches.