Section: New Results
Managing Dynamic Linked Data
Complex Event Processing
Complex event processing requires to answer queries on streams of complex events, i.e., nested words or equivalently linearizations of data trees, but also to produce dynamically evolving data structures as output.
The topic of the PhD project of M. Sakho supervised by Niehren and Boneva is to generalize algorithms for querying streams to hyperstreams. These are collections of linked streams as naturally produced as intermediate results of complex events processing. Hyperstreams are incomplete descriptions of relational structures, so they can queried similarly to incomplete databases, for which the notion of a certain query answer is most appropriate.
In a paper published at RP , they studied certain query answering for hyperstreams with simple events. Such hyperstreams can be identified with compressed string patterns. They proved that the certain query answering for regular queries on compressed string patterns is PSPACE-complete, independently of whether the finite automata defining the regular queries are assumed deterministic or not, and independently of whether compression is permitted or not. They also showed that the problem is in PTIME when restricted to linear string patterns (possibly with compression) and to deterministic finite automata.
In a paper published at LATA , the studied certain query answering on hyperstreams of complex events. Such hyperstreams can be modeled by compressed tree pattern with context variables. They showed that certain query answering for regular queries on compressed tree pattern with context variables is EXP -complete, independently of whether the tree automata defining the regular queries are assumed deterministic or not, and independently of whether compression is permitted or not. They also showed that the problem is in PTIME when restricted to linear tree patterns (possibly with compression) and to deterministic tree automata.
In his PhD project – belonging to the ANR Colis– Gallot with his supervisors Salvati and Lemay presented higher order tree transducers which extend macro tree transducers. Moreover they obtained nice properties such as the closure of the transducers under composition. Algorithms to compute such compositions are proposed. Those algorithms perform partial evaluation and are guided by semantic interpretations over finite domains.
Another virtue of higher-order transducers is that their linear syntactic restriction make them equivalent to logically defined MSO transductions. One of the composition algorithm proposed preserves the linearity. Furthermore, we have also showed that we can decrease the order of linear transducer (i.e. the complexity of the functions it handles) when this one is larger than 4.
These results are unpublished paper for now.