Title:Determining Relevance of Accesses at Runtime (Extended Version)

Abstract:Consider the situation where a query is to be answered using Web sources that restrict the accesses that can be made on backend relational data by requiring some attributes to be given as input of the service. The accesses provide lookups on the collection of attributes values that match the binding. They can differ in whether or not they require arguments to be generated from prior accesses. Prior work has focused on the question of whether a query can be answered using a set of data sources, and in developing static access plans (e.g., Datalog programs) that implement query answering. We are interested in dynamic aspects of the query answering problem: given partial information about the data, which accesses could provide relevant data for answering a given query? We consider immediate and long-term notions of "relevant accesses", and ascertain the complexity of query relevance, for both conjunctive queries and arbitrary positive queries. In the process, we relate dynamic relevance of an access to query containment under access limitations and characterize the complexity of this problem; we produce several complexity results about containment that are of interest by themselves.