Section: New Results

Ontology-Based Query Answering with Existential Rules

Participants : Jean-François Baget, Mélanie König, Michel Leclère, Marie-Laure Mugnier, Michaël Thomazo.

Note that for this section, as well as all sections in New Results, participants are given in alphabetical order.

In collaboration with: Sebastian Rudolph (Karlsruhe Institute of Technology)

We have pursued the work on the existential rule framework in the context of Ontology-Based Query Answering. See the 2011 activity report for details on this framework also known as Datalog+/-. The ontology-based query answering problem consists of querying data while taking into account inferences enabled by an ontology (described by existential rules in our case).

From 2009 to 2011, we mainly investigated decidability and complexity issues. In 2012, while still interested in deepening decidability and complexity results, we tackled the next step: algorithms. Our aim is to develop algorithms with good theoretical properties (at least they should run in “the good worst-case complexity class”) and with good performance in practice. There are two main ways of processing rules, namely forward chaining and backward chaining. In forward chaining, rules are applied to enrich the initial facts and query answering is solved by evaluating the query against the “saturated” facts (as in a classical database system). When it is finite, the backward chaining process can be divided into two steps: first, the query is rewritten into a first-order query (typically a union of conjunctive queries) using the rules; then the rewritten query is evaluated against the initial facts (again, as in a classical database system).

Forward Chaining Algorithms

Considering the expressive class of greedy bounded-treewidth set of rules (in short gbts), which we defined in 2011, we have designed a query answering algorithm which has several advantages over 2011 algorithm, while staying optimal with respect to worst-case combined and data complexities.

  1. It is much more implementable (previous algorithm was using an oracle).

  2. It is generic in the sense that it works for any class of rules that fulfills the gbts property, but it can also be easily specialized for specific gbts subclasses whith lower complexities, such as frontier-guarded or guarded rules, in such a way that it runs in the good complexity class.

  3. It allows for separation between offline and online processing steps: the knowledge base can be compiled independently from queries, which are evaluated against the compiled form.

One of the lightweight description logics used for ontology-based query answering is ℰℒ. We designed a subclass of existential rules that covers ℰℒ with the same complexity of reasoning, while allowing for any predicate arity and some cycles on variables. We also added complex role inclusions like transitivity and right/left identity rules to enhance expressivity, while staying polynomial in data complexity and generalizing existing results.

  • Results published in [36] , [37] and [32] (invited conference). See also our research report [49] for a longer version.

  • A journal version extending the papers at IJCAI 2011 and KR 2012 is in preparation, to be submitted to a major artificial intelligence journal.

Backward Chaining Algorithms

We consider query rewriting techniques that output a union of conjunctive queries, which we see as a set of conjunctive queries. More specifically, only the most general elements of this set need to be kept in the output. We first proved that all sound and complete query rewriting algorithms necessarily produce the same result (up to redundancy) when restricted to their most general elements. It follows that comparing existing algorithms with respect to the size of the produced query is pointless.

Existing query rewriting algorithms accept only specific classes of existential rules (mainly corresponding to the translation of some lightweight description logics). We designed an algorithm that accept as input any set of existential rules and stops if this set of rules fufills so-called fus property (meaning that the set of most general rewritings of any initial conjunctive query is finite). This algorithm has been implemented and first experimentations have been led on rule bases obtained by translating description logic bases.

  • Results published in [31] (best paper price)

Querying Optimization (Work in Progress)

Our current work aims at improving previous algorithms, in particular: the online querying step in the gbts algorithm; the query rewriting algorithm, by avoiding generating several times equivalent rewritings; for specific subclasses, query rewriting into a set of so-called semi-conjunctive queries instead of conjunctive queries, which reduces the size of the output query.