Extracting partition statistics from semistructured data

The effective grouping, or partitioning, of semistructured data is of fundamental importance when providing support for queries. Partitions allow items within the data set that share common structural properties to be identified efficiently. This allows queries that make use of these properties, such as branching path expressions, to be accelerated. Here, we evaluate the effectiveness of several partitioning techniques by establishing the number of partitions that each scheme can identify over a given data set. In particular, we explore the use of parameterised indexes, based upon the notion of forward and backward bisimilarity, as a means of partitioning semistructured data; demonstrating that even restricted instances of such indexes can be used to identify the majority of relevant partitions in the data.