Certain and possible XPath answers

Formulating an XPath query over an XML document is a difficult chore for a non-expert user. This paper introduces a novel approach to ease the querying process. Instead of specifying a query, the user simply marks positive examples X⁺ of nodes that fit her information need. She may also mark negative examples X^- of undesirable nodes. A deductive method, to suggest additional nodes that will interest the user, is developed in this paper. To be precise, a node y is a certain answer if every query returning all positive examples X⁺, and not returning any negative example from X ^-, must also return y. Similarly, y is a possible answer if there exists a query returning X⁺ and y, while not returning any node in X^-. Thus, y is likely to be of interest to the user if y is a certain answer, and unlikely to be of interest if y is not even a possible answer. The complexity of finding certain and possible answers, with respect to various classes of XPath, is studied. It is shown that for a wide variety of XPath queries (including child and descendant axes, wildcards, branching and attribute constraints), certain and possible answers can be found efficiently, provided that X⁺ and X^- are of bounded size. To prove this result a novel algorithm is developed.