Combining Structural and Textual Contexts for Compressing Semistructured Databases