Experiences Using Deliberation RuleML 1.01 as Rule Interchange Language

Abstract

After four decades of development, it is now well accepted that rules are one of the most suitable decisioning mechanism for changeintensive applications. Moreover, as web applications are expanding everywhere, rule languages allow to publish, trasmit, deploy, and execute rules on the Web. This paper describes some principles and practices of developing rulebases from computationally independent business models using Deliberation RuleML version 1.01. We present a complete rulebase, based on the UServ Product Derby Case Study, ready to be deployed into rule systems allowing RuleML serialization. We also raise two issues that the next version of Deliberation RuleML may address: i) reference to external domain vocabularies used by rules and ii) distinction between binary object properties, i.e. predicates/relations (whose second argument is a RuleML Ind element), and datatype properties, i.e. predicates/relations (whose second argument is a RuleML Data element), as in OWL.

-

Rule based applications have a long history starting with the early attempts to make logic an executable language [ 13 ], the age of expert system applications [ 9 ], the introduction of object orientation [ 12 ], and the accommodation of rule-based reasoning into mainstream programming languages [ 18 ].

Experiments with rule interchange were pioneered back in 2006-2007 by [ 10, 14, 16 ] within a rule interchange framework R2ML, [ 21, 22 ]. Inspired by RuleML [ 5, 8, 20 ], R2ML provided an approach to integrate the Object Constraint Language (OCL), the Semantic Web Rule Language (SWRL) and the Rule Markup Language (RuleML). Development of the R2ML language ended in 2008 and after that some progress was made by the W3C RIF activity to achieve a standard for rule interoperability. RuleML continued the development [ 7 ] so that by now various families of rule languages are available, including Deliberation RuleML [ 6 ], Reaction RuleML [ 17 ] and Legal RuleML [ 1, 2 ].

Towards restarting the debate about rule interoperability, this paper describes a RuleML 1.01 rulebase accommodating the UServ Product Derby Case Study. The UServ study describes rules using natural language, and provides a scenario simulating a vehicle insurance service. The main purpose is to compute an annual premium for a vehicle insurance policy, belonging to an eligible client. Using a business-speci c language, UServ divides its business rules into sets, each of them addressing di erent goals and contexts i.e. calculates the eligibility scores in order to set the eligibility status for a client/driver/car, as well as pricing and cancelation policies at client portfolio level. The goal of this work is to provide a tutorial for practitioners developing business rule applications from natural language descriptions based on core ontological concepts like classes and variables, to target rule systems. UServ was also implemented using R2ML. 2

Mapping the Data Model into Rules Rules should be based on vocabularies, i.e. the de ned universe of discourse. The UServ case identi es classes such as cars, clients, drivers, their subclasses such as ConvertibleCar, YoungDriver and their properties e.g. theftRating, price, age, and so on. We use F-logic [ 11, 12 ] as the executable language for UServ rules.

We map all universe-of-discourse artefacts according to the mapping rules introduced by R2ML. In addition, all vocabulary items (predicates, functions, constants) are identi ed by URIs, e.g. http://userv.org/ontology/price identies the binary predicate price. Throughout this paper we use the RuleML 1.01 NafNegHornlogEq sub-language, where equality is used just because, in general, Deliberation RuleML 1.01 does not allow the direct de nition of global constants (more in Section 2.4). 2.1

Mapping Classes

UServ classes are mapped to RuleML types or unary predicates. These are the two signi cant options to encode such classes { either by using the @type attribute or by using an explicit class de nition by means of an atom (we assume a declaration of the namespace pre x us: as the iri http://userv.org/ontology/): <!-- 1. type encoding into the variable occurences --> <Var type="us:ConvertibleCar">C</Var> <!-- 2. type explicitly defined as an atom --> <Atom> <Rel iri="us:ConvertibleCar">car</Rel> <Var>C</Var> </Atom> 2.2

Mapping subClassOf Relation

The subClassOf relation is not included by RuleML as a language syntactical construct. However, there are solutions to realize it, either by de ning the subClassOf relation axiomatically or by binding the rulebase to an external vocabulary de ning this relation.

De ning subClassOf Axiomatically <Forall> <Var>X</Var> <Implies> <Atom> <Rel iri="us:ConvertibleCar">convertibleCar</Rel> <Var>X</Var> </Atom> <Atom> <Rel iri="us:Car">car</Rel> <Var>X</Var> </Atom> </Implies> </Forall> It is obvious that such an axiom must be added for any pair of classes in the application domain that are subject of in a subClassOf relation. Moreover, assuming that a translation to an executable language will preserve this axiom in this form, inference will be quite expensive as, usually, the subClassOf veri cation is much more of a constraint veri cation than an inference.

Binding a Rulebase to an External Vocabulary

An elegant solution, also discussed in [ 3 ] and [ 10 ], and implemented in OO jDREW, is to de ne this relation at the domain vocabulary level using an appropriate knowledge representation language, for example, RDF Schema: <rdfs:Class rdf:ID="http://userv.org/ontology/ConvertibleCar"> .....

<rdfs:subClassOf rdf:resource="http://userv.org/ontology/Car"/> </rdfs:Class> Of course, the rulebase must provide a reference to this vocabulary, e.g by using type="us:ConvertibleCar", as discussed earlier.

High Order Atom Construct

The Atom content model of the Deliberation RuleML 1.01 could be extended to allow high order constructs1, such as: <Atom> <Rel iri="rdfs:subClassOf">subClassOf</Rel> <arg index="1"> <Rel iri="http://userv.org/ontology/ConvertibleCar" type="http://www.w3.org/2000/01/rdf-schema#Class">convertibleCar</Rel> 1 The older RuleML 0.91 was supporting high order constructs via "Hterm" element. </arg> <arg index="2"> <Rel iri="http://userv.org/ontology/Car" type="http://www.w3.org/2000/01/rdf-schema#Class">car</Rel> </arg> </Atom>

Declarative De nition of Instances

Finally, because the application domain has a limited number of (car) instances, only few of them may be convertible. In this case, from an inference engine perspective, it is less expensive to explicitly de ne cars and their convertibility as facts: <Assert> <And> <Atom> <Rel iri="http://userv.org/ontology/Car">car</Rel> <Ind iri="http://userv.org/ontology/i/a9fccd3a-a851-4e80">car1</Ind> </Atom> <Atom> <Rel iri="http://userv.org/ontology/ConvertibleCar">car</Rel> <Ind iri="http://userv.org/ontology/i/a9fccd3a-a851-4e80">car1</Ind> </Atom> </And> </Assert> Notice that the above code describes the same individual as being an instance of two types but does not encode any subClassOf relationship between them. 2.3

Mapping Properties

UServ de nes object properties and datatype properties. For example, price is a datatype property with a unique value. We map such properties into the traditional rst order logic atom: <Atom> <Rel iri="http://userv.org/ontology/price">price</Rel> <!-- the subject of the property --> <Var>C</Var> <!-- the value of the property --> <Var type="xs:positiveInteger">P</Var> </Atom> Also, to express something like "the car's potential theft rating is high", we encode: <Atom> <Rel iri="http://userv.org/ontology/theftRating">theftRating</Rel> <Var>C</Var> <Data> <!-- @type="us:Rating" --> <us:ratingValue xsi:type="xs:positiveInteger">4</us:ratingValue> <us:ratingVerbalization xml:lang="en">high</us:ratingVerbalization> </Data> </Atom> The structured datatype including an integer (i.e. 4) is more convenient than the plain symbol (i.e., high) as it allows a computer program to better deal with values and their verbalizations. It is worthwhile to mention the Schema.org2 e ort to standardize a large common sense vocabulary for web semantics. Unfortunately, the current version of RuleML does not allow such user-de ned datatypes but only pre-de ned XML datatypes, e.g. <Data xsi:type="xs:string">high</Data> Allowing external datatypes would be an improvement, as common types become more and more adopted by the Web community. However, the current content model of the RuleML <Data> element is xs:any; therefore, above, we were still able to encode the necessary structured data.

The values of multi-valued properties such as airbag are mapped into RuleML lists (using <Plex>): <Atom> <Rel iri="http://userv.org/ontology/airbag">airbags</Rel> <Var>C</Var> <Plex> <Data xsi:type="xs:string">driverAirbag</Data> <Data xsi:type="xs:string">frontPassengerAirbag</Data> </Plex> </Atom> Object-valued properties are mapped following the same principles. While the "positional" RuleML used above adopted a logic programming syntax, the above method for de ning properties (binary predicates) may not be ideal. However, "slotted" RuleML, inspired by F-logic, as in Object-Oriented RuleML [ 3 ] and PSOA RuleML [ 4 ], could be used instead. Because UServ is built on an object-oriented approach (as are many other real-life applications), properties are binary relations. Therefore, it is obvious that knowledge representation languages that de ne vocabularies are more suited for de ning such properties. Here is an example using OWL: <owl:DatatypeProperty rdf:ID="price"> <rdf:type rdf:resource="owl:FunctionalProperty"/> <rdfs:domain rdf:resource="http://userv.org/ontology/Car"/> <rdfs:range rdf:resource="xs:positiveInteger"/> </owl:DatatypeProperty> 2 See http://schema.org and http://lists.w3.org/Archives/Public/public-vocabs/ Such a declaration would allow a rulebase validator to check the consistency of rules, e.g. fail when the below "multi-priced" atom is encountered: <Atom> <Rel iri="http://userv.org/ontology/price">p</Rel> <Var>C</Var> <Plex> <Data xsi:type="xs:positiveInteger">23000</Data> <Data xsi:type="xs:positiveInteger">23500</Data> </Plex> </Atom> 2.4

Mapping Built-in Predicates and Global Constants

UServ deals with all kinds of conditions such as numerical constraints, e.g. "price is greater than $45,000", or list membership conditions "the car model is on the list of High Theft Probability Auto". The mapping uses external built-in predicates, speci cally the ones introduced by the W3C RIF Datatypes and Built-Ins 1.0. <!-- price is greater than $45,000 --> <Atom> <Rel iri="pred:numeric-less-than-or-equal">gt</Rel> <Var type="xs:positiveInteger">P</Var> <Data xsi:type="xs:integer">45000</Data> </Atom> <!-- the car model is on the list of High Theft Probability Auto --> <Atom> <Rel iri="prolog:member">member</Rel> <Var>M</Var> <Ind type="rdf:Bag">HighTheftProbabilityAutoList</Ind> </Atom> Unlike logic programming, object oriented applications deal with global constants that are available to all objects. Similarly such a construct will greatly improve rulebase readability.

Deliberation RuleML cannot directly represent global constants but uses equality to explicitly de ne a collection as an individual such as the above-used HighTheftProbabilityAutoList, e.g. <Equal oriented="yes"> <Ind>HighTheftProbabilityAutoList</Ind> <Plex> <Ind iri="http://userv.org/ontology/i/a9fccd3a-a851-4e80">car1</Ind> <Ind iri="http://userv.org/ontology/i/5bff7cc6-72f7-4be3">car3</Ind> <Ind iri="http://userv.org/ontology/i/5eb91389-f1f3-473c">car5</Ind> <Ind iri="http://userv.org/ontology/i/3c308aa3-3507-40fa">car6</Ind> </Plex> </Equal> However, such an approach requires the use of a hornlogeg RuleML language ("eq" referring to equality), while there may be applications which do not really require full equality but only the fragment needed for de ning global constants. However, from the perspective of interoperability, it may be the responsibility of a translator to identify such constructs and translate them into global constants, while still keeping the rest of the rule translation in the realm of a hornlog RuleMl language. 2.5

Expressing Facts

UServ facts are descriptions of cars such as: Car 4, 2005 Honda Odyssey, Full Airbags, Alarm System, Price $39,000, Type Luxury Car,

Not on "High Theft Probability List" it is obvious to any object oriented modeler that car1 is the id of an object of the type LuxuryCar together with a set of properties (e.g. carModel:'Honda Odyssey', modelYear:2005). This can be mapped to Deliberation RuleML 1.01 using the traditional logic programming approach: We shall observe that the representation of "Not on High Theft Probability List" corresponds to negation as failure (no positive fact). In addition, the global constant HighTheftProbabilityAutoList does not contain this car among its values. This solution has the advantage of preserving the logic programming approach, allowing a translator to still use the NafNegHornlog sub-language. There are object-oriented rule languages, notably F-logic, Drools [ 19 ], and PSOA RuleML allowing declarative descriptions of objects e.g., car4:LuxuryCar [ price -> 39000; modelYear -> 2005; hasAlarm-> true; airbags -> ['driverAirbag','frontPassengerAirbag', 'sideAirbag'] ].

Deliberation RuleML could achieve such a representation as in Object-Oriented RuleML or by extending the slotted notation to allow relations as slot names: <Assert> <Equal oriented="yes"> <Ind iri="http://userv.org/ontology/i/a9fccd3a-a851-4e80">car4</Ind> <Plex> <slot> <Rel iri="http://userv.org/ontology/price"></Rel> <Data xsi:type="xs:positiveInteger">39000</Data> </slot> <slot> <Rel iri="http://userv.org/ontology/hasAlarm"></Rel> <Data xsi:type="xs:boolean">true</Data> </slot> ...

</Plex> </Equal> </Assert> 3

Rules and Rulebases

Deliberation RuleML 1.01 de nes rulebases as collections of implications, facts and equations, where the <Rule> element is specialized to an <Implies> <Rulebase xml:id="theftRating"> <!-- Processing car's theftRating --> <meta> <Atom>

<Rel iri="http://userv.org/ontology/theftRating">theftRating</Rel> </Atom> </meta> <!-- If the car is a convertible, then the car's potential theft rating is high. --> <Forall> <Var type="us:ConvertibleCar">C</Var> <Atom> <Rel iri="http://userv.org/ontology/theftRating">theftRating</Rel> <Var type="us:ConvertibleCar">C</Var> <Data> <!-- @type="us:Rating" --> <us:ratingValue xsi:type="xs:positiveInteger">4</us:ratingValue> <us:ratingVerbalization xml:lang="en">high</us:ratingVerbalization> </Data> </Atom> </Forall> <!-

If the car's price is greater than $45,000, then the car's potential theft rating is high.

--> <Forall> <Var type="us:Car">C</Var> <Implies direction="backward"> <And> <Atom> <Rel iri="http://userv.org/ontology/price">price</Rel> <Var>C</Var> <Var type="xs:positiveInteger">P</Var> </Atom> <Atom> <Rel iri="pred:numeric-greater-than">gt</Rel> <Var type="xs:positiveInteger">P</Var> <Data xsi:type="xs:integer">45000</Data> </Atom> </And> <Atom> <Rel iri="http://userv.org/ontology/theftRating">theftRating</Rel> <Var>C</Var> <Data> <!-- @type = "us:Rating"--> <us:ratingValue xsi:type="xs:positiveInteger">4</us:ratingValue> <us:ratingVerbalization xml:lang="en">high</us:ratingVerbalization> </Data> </Atom> </Implies> </Forall> <!-- other rules ... --> </Rulebase>

Conclusion and Future Work This work provides some insight into the Deliberation RuleML 1.01 capabilities to serialize business rule scenarios. We hope that our ndings will contribute to improvements in the next RuleML version. We also raise the issue that the next version of Deliberation RuleML may provide: i) reference to external domain vocabularies used by rules and ii) distinction between binary object properties, i.e. predicates/relations (whose second argument is a RuleML Ind element), and datatype properties, i.e. predicates/relations (whose second argument is a RuleML Data element), as in OWL. Future work will be concerned with a family of XSLT transformations allowing translation from speci c sub-languages of Deliberation RuleML to Prolog and F-logic.

Acknowledgments References

We would like to thank to Tara Athan, Harold Boley, Adrian Giurca and Adrian Paschke for their essential feedback and insights that made this work possible.

UServProductDerbyCaseStudy.ruleml <!-

If the car model is on the list of High Theft Probability Auto, then the cars potential theft rating is high. --> <Forall> <Var type="http://userv.org/ontology/Car">C</Var> <Implies direction="backward"> <And> <Atom> <Rel iri="http://userv.org/ontology/carModel">carModel</Rel> <Var>C</Var> <!-

If the car has no airbags, then the cars potential occupant injury rating is extremely high. --> <Forall> <!-

If none of the following is true, then the auto eligibility is eligible - Auto eligibility is not eligible - Auto eligibility is provisional <Atom>

<Rel iri="http://userv.org/ontology/SeniorDriver">seniorDriver</Rel>

If the following are not true, then eligible driver

- Young driver

If driver has taken drivers training from a licensed driver training company, then driver has training certification --> <Forall> <Var type="http://userv.org/ontology/Driver">D</Var> <Implies direction="backward"> <Atom> <Rel iri="http://userv.org/ontology/hasTrainingFromLicensedDriverTrainingCompany"> hasTrainingFromLicensedDriverTrainingCompany </Rel> <Var type="http://userv.org/ontology/Driver">D</Var> </Atom> <!-

If the driver has been convicted of a DUI, then the driver qualifies as a High Risk Driver --> <Forall> <Var type="http://userv.org/ontology/Driver">D</Var> <Implies direction="backward"> <And> <Atom> <Rel iri="http://userv.org/ontology/noOfDUI">noOfDUI</Rel> <Var type="http://userv.org/ontology/Driver">D</Var> <Var>DUI</Var> </Atom> <Atom> <Rel iri="http://userv.org/ontology/DriverRiskRating">driverRiskRating</Rel> <Var>D</Var> <Data> <us:value xsi:type="xs:positiveInteger">4</us:value> <us:verbalization xml:lang="en">high</us:verbalization> </Data> </Atom> </Implies> </Forall> <!-

If the number of accidents the applicant has been involved in is greater than 2, then the driver qualifies as a High Risk Driver. --> <Forall> <Var type="http://userv.org/ontology/Driver">D</Var> <Implies direction="backward"> <And> <Atom> <Rel iri="http://userv.org/ontology/noOfAccidentsI">noOfAccidents</Rel> <Var type="http://userv.org/ontology/Driver">D</Var> <Var>accidents</Var> </Atom> <Atom> <Rel iri="pred:numeric-greater-than">gt</Rel> <Var type="xs:positiveInteger">accidents</Var> <Data xsi:type="xs:integer">2</Data> </Atom> </And> <!-

If the driver has had more than 3 moving violations in the last two years, then the driver qualifies as a High Risk Driver.

References

1. Tara Athan , Harold Boley, Guido Governatori, Monica Palmirani, Adrian Paschke, and Adam Wyner . Legalruleml: From metamodel to use cases - (a tutorial) . In Leora Morgenstern , Petros S. Stefaneas, Francois Levy , Adam Wyner, and Adrian Paschke, editors, RuleML , volume 8035 of Lecture Notes in Computer Science, pages 13 { 18 . Springer, 2013 .

2. Tara Athan , Harold Boley, Guido Governatori, Monica Palmirani, Adrian Paschke, and Adam Wyner . Oasis legalruleml . In Enrico Francesconi and Bart Verheij , editors, ICAIL , pages 3 { 12 . ACM, 2013 .

3. Harold Boley . Object-oriented ruleml: User-level roles, uri-grounded clauses, and order-sorted terms . In Michael Schroeder and Gerd Wagner , editors, RuleML , volume 2876 of Lecture Notes in Computer Science, pages 1 { 16. Springer, 2003 .

4. Harold Boley . A rif-style semantics for ruleml-integrated positional-slotted, objectapplicative rules . In Nick Bassiliades , Guido Governatori, and Adrian Paschke, editors, RuleML Europe , volume 6826 of Lecture Notes in Computer Science, pages 194 { 211 . Springer, 2011 .

5. Harold Boley , Mike Dean , Benjamin N. Grosof , Michael Kifer, Said Tabet, and Gerd Wagner . Ruleml position statement . In Rule Languages for Interoperability. W3C , 2005 .

6. Harold Boley , Rolf Grutter, Gen Zou, Tara Athan, and Sophia Etzold . A datalog + plus ruleml 1.01 architecture for rule-based data access in ecosystem research . In Antonis Bikakis, Paul Fodor, and Dumitru Roman, editors, RuleML , volume 8620 of Lecture Notes in Computer Science, pages 112 { 126 . Springer, 2014 .

7. Harold Boley , Adrian Paschke, and Omair Sha q. Ruleml 1.0: The overarching speci cation of web rules . In Mike Dean , John Hall, Antonino Rotolo, and Said Tabet, editors, RuleML , volume 6403 of Lecture Notes in Computer Science, pages 162 { 178 . Springer, 2010 .

8. Harold Boley , Said Tabet, and Gerd Wagner . Design rationale for ruleml: A markup language for semantic web rules . In Isabel F. Cruz, Stefan Decker, Jero^me Euzenat, and Deborah L. McGuinness, editors, SWWS , pages 381 { 401 , 2001 .

9. Randall Davis , Bruce G. Buchanan, and Edward H. Shortli e. Production rules as a representation for a knowledge-based consultation program . Artif. Intell. , 8 ( 1 ): 15 { 45 , 1977 .

10. Adrian Giurca and Gerd Wagner . Rule modeling and interchange . In Negru et al. [ 15 ], pages 485 { 491 .

11. Michael Kifer . Rules and ontologies in f-logic . In Norbert Eisinger and Jan Maluszynski , editors, Reasoning Web , volume 3564 of Lecture Notes in Computer Science, pages 22 { 34 . Springer, 2005 .

12. Michael Kifer and Georg Lausen . F-logic: A higher-order language for reasoning about objects, inheritance, and scheme . In James Cli ord, Bruce G. Lindsay, and David Maier, editors, SIGMOD Conference , pages 134 { 146 . ACM Press, 1989 .

13. Robert A. Kowalski . Predicate logic as programming language . In IFIP Congress , pages 569 { 574 , 1974 .

14. Milan Milanovic , Dragan Gasevic, Adrian Giurca, Gerd Wagner, and Vladan Devedzic . Towards sharing rules between owl/swrl and uml/ocl . ECEASST, 5 , 2006 .

15. Viorel Negru , Tudor Jebelean, Dana Petcu, and Daniela Zaharie, editors. Proceedings of the Ninth International Symposium on Symbolic and Numeric Algorithms for Scienti c Computing , SYNASC 2007 , Timisoara, Romania, September 26-29 , 2007 . IEEE Computer Society, 2007 .

16. Oana Nicolae , Adrian Giurca, and Gerd Wagner . On interchange between jboss rules and jess . In Costin Badica and Marcin Paprzycki , editors, IDC, volume 78 of Studies in Computational Intelligence , pages 135 { 144 . Springer, 2007 .

17. Adrian Paschke , Harold Boley, Zhili Zhao , Kia Teymourian , and Tara Athan . Reaction ruleml 1 . 0: Standardized semantic reaction rules . In Antonis Bikakis and Adrian Giurca , editors, RuleML , volume 7438 of Lecture Notes in Computer Science, pages 100 { 119 . Springer, 2012 .

18. Mark Proctor . Relational declarative programming with jboss drools . In Negru et al. [15], page 5.

19. Mark Proctor . Drools: A rule engine for complex event processing . In Andy Schurr, Daniel Varro, and Gergely Varro, editors, AGTIVE , volume 7233 of Lecture Notes in Computer Science, page 2. Springer, 2011 .

20. Gerd Wagner , Grigoris Antoniou, Said Tabet, and Harold Boley . The abstract syntax of ruleml - towards a general web rule language framework . In Web Intelligence , pages 628 { 631 . IEEE Computer Society, 2004 .

21. Gerd Wagner , Carlos Viegas Damasio, and Grigoris Antoniou . Towards a general web rule language . Int. J. Web Eng. Technol. , 2 ( 2 /3): 181 { 206 , 2005 .

22. Gerd Wagner , Adrian Giurca, and Sergey Lukichev . A general markup framework for integrity and derivation rules . In Francois Bry , Francois Fages, Massimo Marchiori, and Hans Jurgen Ohlbach, editors, Principles and Practices of Semantic Web Reasoning , volume 05371 of Dagstuhl Seminar Proceedings. Internationales Begegnungs- und Forschungszentrum fur Informatik (IBFI) , Schloss Dagstuhl, Germany, 2005 . < Atom> <Rel iri="pred:numeric-greater-than">gt</Rel> <Var type="xs:positiveInteger">DUI</Var> <Data xsi:type="xs:integer">0< /Data> </Atom> </And>