Section: New Results

Distributed data management

Participants : Serge Abiteboul, Émilien Antoine, Cristina Sirangelo, Nadime Francis, Luc Segoufin.

Distributed knowledge base.

We are developing the system Webdamlog [16] , [13] , [14] to address the challenges faced by everyday Web users, who interact with inherently heterogeneous and distributed information. Managing such data is currently beyond the skills of casual users. In Webdamlog, we see the Web as a knowledge base consisting of distributed logical facts and rules. The objective is to enable automated reasoning over this knowledge base, ultimately improving the quality of service and of data. The system supports the Webdamlog language, a Datalog style language with rule delegation.

Deduction in uncertain worlds.

Motivated by reasoning in distributed environments in which disagreements arise between different actors, we study in [17] deduction (captured by datalog programs) in the presence of inconsistencies (induced by functional dependency (FD) violations). We adopt an operational semantics for datalog with FDs based on inferring facts one at a time, while never violating the FDs. This yields a set of possible worlds that we capture by c-tables of possibly exponential size. We propose to use probabilities to measure this nondeterminism and define a probabilistic semantics that can be captured by probabilistic conditional tables. Not surprisingly, we show that computing the probability of a query answer in our setting is expensive, which leads us to introduce a sampling algorithm to estimate answer probabilities. We then turn our attention to the problem of explaining why a particular answer holds. This leads us to consider two novel notions: the most influential extensional facts, and the most likely proofs for an answer. We study algorithms for ranking facts and proofs based on their contribution to the derivation of an answer. Finally, we consider how our framework can be adapted to a distributed setting, and in particular, how sampling can be performed in a distributed manner.

Access rights in a distributed setting.

We started considering access right issues in Webdamlog. This is related to specifying access right on views in standard databases. There is also the issues of controlling rules that are run locally but were specified by other peers.

Incomplete information in Web data.

Incomplete information often arises from the integration of different Web data sources, as well as from the exchange of data between communicating Web applications. The semantics of incompleteness (i.e. which possible complete databases are represented by an incomplete one) depends on the context and the particular scenario where incompleteness raises from. We have studied how to deal with the presence of incomplete information under different possible semantics. We have in particular studied in which condition it is possible to query incomplete data “naively", i.e. as if it were complete. We have exhibited “natural" fragments of first order logic for which naive evaluation is possible, under different semantics.

Graph data management.

Graph structured data can be found in new emerging applications such as RDF and linked data, or social networks. The peculiarity of queries over graphs is that they are interested in both data carried by the graph and in the graph topology; they are often based on reachability patterns. In a distributed setting it is very common to be able to query only a partial description or a "view" of the graph. We studied the problem of answering queries using only the information provided by the views. The presence of a form of recursion in views and queries presents new challenges. We found restricted classes of graph views and queries that allow efficient query answering over views.