Proposal and Evaluation of a Technique of Discovering Xml Structures for Efficient Retrieval

We propose an adaptable approach to discovery of database schemas for well-formed XML data such as EDI, news, and digital libraries, which we interchange, filter, or download for future retrieval and analysis. The generated schemas usually consist of more than one table. Our approach controls the number of tables to be divided by use of statistics of XML so that the total cost of processing queries is reduced. We generate schemas appropriate for complex data such as text formatting tags and child elements with the small maximum number of occurrences in order to reduce the number of tables. To this end, we introduce three functions NULL expectation, Large Leaf Fields, and Large Child Fields for controlling the number of tables to be divided. We described how to translate queries in XQuery into those in SQL. We also describe the concept of short paths contained by generated database schemas and their effects on the performance of query processing. We discuss when and how it is necessary to change the resultant database schemas in case of updates of the original XML data. We evaluate typical XML queries over the generated schemas and normalized schemas as another approach and measure and compare both of the costs in order to validate our approach.