Rules workshop Day 2 Use cases 1 Christiane Goldbreich Prez Sharing anatomical knowledge Brain ontology developed, published on UniRennes web site Problem is to label brain parts in digitally processed images (MRI) from anatomical ontology and rules Claims that ontology language needs be integrated with rule language because she would miss some solutions if not; as far as CSMA understands, the problem is navigating ontology's associations/constraints Requirements on ontology lang: OWL DL expressiveness + qualified cardinality constraints Requirements on Rule weblanguage See slide N-ary predicates seem to be one of the main issue Discussion Are the rules and ontology available in machine readable form? A: yes, see URL on slides Harold: what language for rules do you need? Do you need functions? Is it Datalog? A: We do not need functions, but this is not a question for me: you candidate technology guys should look at the use case and tell me; Harold: do you need negative? Ian: you need more expressive power, no matter where do you get it from? Rules or Owl or what? Would you be happy with full FOL? A: We need all of OWL DL, expressed in OWL or an equivalent rule language Phil Archer Prez Content labelling; move from PICS to RDF-based labels is under way Two different but related uses for rules (the problem is that when you have millions of pages, you do not want to have to meta-tag them all): associating a description with a URI (I give the URI and you give me the description): ; currently do it in RDF, defining rulesets of the form: default label for domain isXXX; then list of rules in sequential mode of form if condition then override default with yyy; reducing data to a binary (I get a list of labels (no violaence, there are bare breast etc) and I want to answer the question: can children see that (Y/N)?); RDF data + User profile result in Match/no match Discussion Q: how is user profile goes in there BenG: We should basically be able to do it PaulV: very good use case, that is where rules hit the common man use of internet; in your first case, do you hit scalability problems? A: do not know yet, system just rolled out Doug Clarck Prez Most orgs today are just collections of projects Project management institute the standardisation org in the profession Opportunity areas for rules: work breakdown (PMI standard exist), scheduling (standard under development), Perf (standard exists), maturity (standard), status Rule language requirements: unique rule iDs; needs modality (alethic, hard-soft), compound rules; fulle programmatic capabilities Use rules to generate states of knowledge for decision making, reporting etc; priorities, weight, confidence of rules important Discussion MarkL: what do you mean with "target"? A: the data at the end of the URI, the data to which the rule will be applied; unique URI based on data structure etc XXX what are coumpound rules? A: Rulesets Chris Matheus Prez Problem domain: situation awareness = understanding of what is going on in an evolving situation Requires domain knowledge, abstraction, time dependent reasoning, sensor fusion RuleVISOR; builds ontology first, then rules/ground them on top of ontology Domain: supply logistics SIXA: reason about tracks data using pedigree ontology and rules Converts SWRL rules to Jess or BaseVisor using XSLT; establish input stream of event annotated using the ontology; use Jess or RuleVisor to reason on the annotated events Issues with SWRL: restriction to binary predicate is a pain (9 relatively simple rules turned into >1000 lines of SWRL) mismatch between declarative semantics and implementation: SWRL built-ins: need them that specify specify input and output terms; no explicit generation/assertion of new facts => ; issue with vars in head that are unbound in bodyneed assert time issues: need to make decision from partial information => requires NAF (could be within scoped context à la N3); need time-dependent attriobutes; some computed info needed only occasional see slides Wishlist: Rules defined on top of OWL ontologies NAF perhaps with scoped context Procedural attachemen,t Non-binary predicates Explicit generation of new facts (assert, gensym) Functionally defined buid-ins Graphical means of defining/understanding rules Means of generating simple explanation Real-time or near real-time performance Build-in support for reasoning about undertainty Discussion Dave: ontology essentially as categories? A: no we use the other properties of OWL as well, e.g. hierarchy Harold: is your problem with N-ary relations a real issue or would an appropriate presentation layer be enough? A: makes sense Srini Prez Requirement for interoperability industry data standards (MISMO) regulatory compliance requirements Nothing new; current solution are stick with one vendor: see slide design proprietary vendor neutral rule language: pb is that thi sis resource intensive Requirements for an interoperable rule language expressiveness: specify the rules based on an object model; open and extensible constructs deterministic characteristics: rule managemebnt (versioning, permissions, effective/expiration dates etc); describe groups of shareable rules consistent integration: conflict resolution and preconditions; synchro and asynchro execution Discussion Ed: what do you mean with expressional completeness? A: defines the syntactic requirements we have Xxx: is MISMO an object model or an XML schema? A: XSD; Q2: do you really need rules based on an object model or would an XSD be enough? A: don't really know at this point; MISMO is too new to know Finkelstein: are the rules for MISMO actually published? A: MISMO defines a data standard, they do not publish anything beyond a dictionary and a data model Dave Reynolds Prez Warning: does not represent all rule activities at HP; only the semantic web oriented ones; there are BR activities etc as well, will not speak for them Little hindsight about use of rules/rules interchange in Jena; only indirect based on user group traffic, questions; but little evidence of rule interchange between users Based on N3, with negation except for extensional predicates; RETE+tabled LP targeted at RDF processing Use cases: deductive rule usage, see requirements for interoperability and expressiveness on slides integrity rule usage xxx Discussion Csma: integrity usage, policy expression case) no evidence of interoperabiloity requirements in the case of source composition? A: No, intuitively I would have expected that, but we see no evidence of it (but we have no direct evidence anyway) Tim: Stefan: no evidence of backward-chaining rules; you have any explanation for that? A: Eric: would you mind if we ask general questions on the list about users' use of rules or whatever? A: no