Inferring a Relax NG Schema from XML Documents

Abstract

An XML schema specifies the structural properties of XML documents generated from the schema and, thus, is useful to manage XML data efficiently. However, there are often XML documents without a valid schema or with an incorrect schema in practice. This leads us to study the problem of inferring a Relax NG schema from a set of XML documents that are presumably generated from a specific XML schema. Relax NG is an XML schema language developed for the next generation of XML schema languages such as document type definitions (DTDs) and XML Schema Definitions (XSDs). Regular hedge grammars accept regular tree languages and the design of Relax NG is closely related with regular hedge grammars. We develop an XML schema inference system using hedge grammars. We employ a genetic algorithm and state elimination heuristics in the process of retrieving a concise Relax NG schema. We present experimental results using real-world benchmark.