PR-OWL 2 Case Study: A Maritime Domain Probabilistic Ontology Kathryn Blackmond Laskey, Richard Haberlin, Rommel Novaes Carvalho Paulo Costa Brazilian Office of the Comptroller General Volgenau School of Engineering Brasília, Brazil George Mason University rommel.carvalho@gmail.com Fairfax, VA USA [klaskey, rhaberli, pcosta]@gmu.edu Abstract—Probabilistic ontologies incorporate uncertain and connection between a statement in PR-OWL and a incomplete information into domain ontologies, allowing statement in OWL is not formalized, it is unclear uncertainty in attributes of and relationships among domain how to perform the integration of ontologies that entities to be represented in a consistent and coherent manner. contain statements of both formalisms. The probabilistic ontology language PR-OWL provides OWL constructs for representing multi-entity Bayesian network Carvalho [3] proposed a new syntax and semantics, defined (MEBN) theories. Although compatibility with OWL was a as PR-OWL 2, which improves compatibility between PR- major design goal of PR-OWL, the initial version fell short in OWL and OWL in two important respects. First, PR-OWL 2 several important respects. These shortcomings are addressed by follows the approach suggested by Poole et al. to formalizing the latest version, PR-OWL 2. This paper provides an overview the association between random variables from probabilistic of the new features of PR-OWL 2 and presents a case study of a theories with the individuals, classes and properties from probabilistic ontology in the maritime domain. The case study ontological languages such as OWL. Second, PR-OWL 2 describes the process of constructing a PR-OWL 2 ontology using allows values of random variables to range over OWL an existing OWL ontology as a starting point. datatypes. Keywords- Probabilistic ontology, Multi-Entity Bayesian This paper presents an overview of PR-OWL 2, describes networks, PR-OWL, OWL, Maritime domain ontology, Uncertainty the key features that improve compatibility with OWL, Modeling Process for Semantic Technologies discusses an open-source tool for building PR-OWL 2 probabilistic ontologies, and describes a use case of a PR-OWL I. INTRODUCTION 2 ontology for maritime domain awareness. The emphasis on net-centric operations and the shift to asymmetric warfare have created new challenges for automated II. A PROBABILISTIC ONTOLOGY IN PR-OWL information integration. To meet these challenges, developers A. PR-OWL 1: An Upper Ontology for MEBN Theories are recognizing the need to combine explicit representation of domain semantics with the ability to represent and reason with PR-OWL provides constructs to define probabilistic uncertainty. Probabilistic ontologies allow the representation of ontologies in the OWL ontology language. The initial version, uncertainty about attributes of and relationships among domain PR-OWL 1, is an OWL upper ontology for representing entities. Probabilistic OWL (PR-OWL) [1] is an OWL upper MEBN theories [4]. MEBN is a first-order probabilistic ontology for representing probabilistic ontologies. language (FOPL) [5] that allows probabilities to be assigned in Compatibility with OWL was a major design goal for PR- a consistent way to logical statements. MEBN represents the OWL. However, the initial release of PR-OWL falls short of world as entities that have attributes and are related to other complete compatibility in several important respects. First, entities. Knowledge about the attributes of entities and their there is no mapping in PR-OWL to properties of OWL. relationships to each other is represented as a collection of Second, although PR-OWL has the concept of meta-entities, MEBN fragments (MFrags) organized into MEBN Theories which allows the definition of complex types, it lacks (MTheories). An MFrag represents a conditional probability compatibility with existing types already present in OWL. distribution for instances of its resident random variables given These problems have been noted in the literature [2]: their parents in the fragment graph and the context nodes. An MTheory is a set of MFrags that collectively satisfies PR-OWL does not provide a proper integration of consistency constraints ensuring the existence of a unique joint the formalism of MEBN and the logical basis of probability distribution over instances of the random variables OWL on the meta level. More specifically, as the represented in each of the MFrags within the set. A PR-OWL ontology encodes domain knowledge as a set of MFrags. A PR- This research was partially supported by the Office of Naval Research OWL reasoner uses the probability information encoded in the (ONR), under Contract N00173-09-C-4008. Rommel Carvalho would MFrags to compute responses to probabilistic queries. like to thank the Brazilian Office of the Comptroller General (CGU) for their active support since 2008 and for providing the human resources necessary to conduct this research. Figure 1. Probabilistic Ontology for Identifying Ship-of-Interest B. A PR-OWL Ontology for the Maritime Domain ShipCharacteristics MFrags. Knowledge about unusual routes As an example of a PR-OWL ontology, Figure 1 shows a is represented in the UnusualRoute and Meeting MFrags. simple probabilistic ontology developed as part of the Finally, knowledge about hypotheses related to evasive PROGNOS (Probabilistic OntoloGies for Net-centric behavior is represented in the EvasiveBehavior, Eletronics- Operation Systems) project [6]. The ontology is designed for Status, and Radar MFrags. the problem of identifying whether a vessel is a ship of interest. A detailed explanation of this model can be found in [6]. The model is designed to answer the following queries using The model was expanded and extended iteratively as described the following evidence: in [7] to address additional queries and evidence. Overall Goal: Identify whether a ship is a ship of interest, C. An Open Source Tool for Probabilistic Ontologies i.e. if the ship seems to be suspicious in any way. The MFrags shown in Figure 1 are screenshots from the 1. Query: Does the ship have a terrorist crewmember? UnBBayes-MEBN [8], an open source, plug-in-based Java application for building and reasoning with probabilistic a. Evidence: Verify whether a crewmember is ontologies based on the PR-OWL/MEBN framework. 1 It related to any terrorist; features a graphical user interface (GUI), an application b. Evidence: Verify whether a crewmember is programming interface (API) for saving and loading PR-OWL associated with any terrorist organization. ontologies, reasoning algorithms for processing queries, and plugin support for extensions. 2. Query: Is the ship using an unusual route? D. Queries a. Evidence: Verify whether there is a direct report that the ship is using an unusual route; Queries are processed in UnBBayes-MEBN using an implementation of the situation-specific Bayesian network b. Evidence: Verify whether there is a report (SSBN) construction algorithm described in [4]. Figure 2 that the ship is meeting some other ship for no shows an SSBN built using the implemented algorithm. We apparent reason. applied an exact inference algorithm on small-scale problems to test the model and identify logical inconsistencies, 3. Query: Does the ship seem to exhibit evasive differences in query results from those expected by subject- behavior? matter experts, and other flaws in the model. For larger scale a. Evidence: Verify whether an electronic problems, approximate inference algorithms are employed to countermeasure (ECM) was identified by a mitigate scalability issues. We also implemented hypothesis navy ship; management methods [9] to control the complexity of the constructed networks while maintaining acceptable accuracy in b. Evidence: Verify whether the ship has a results. responsive radar and automatic identification system (AIS). III. PR-OWL 2: IMPROVING COMPATIBILITY WITH OWL Each of the nine MFrags of Figure 1 addresses a modular Ideally, it should be possible to use PR-OWL to reason component of the knowledge needed to address the above probabilistically about uncertain aspects of an ontology based queries. Specifically, probabilistic knowledge about hypotheses on the information already available. That is, we would like to related to the identification of a terrorist crewmember is represented in the HasTerroristCrew, TerroristPerson, and 1 UnBBayes is available from http://unbbayes.sourceforge.net/ Figure 2. Situation-Specific Bayesian Network for Identifying Ship-of-Interest be able to begin with an OWL ontology containing information Poole et al. state, “We can reconcile these views by having about a domain, use PR-OWL to define uncertainty about properties of individuals correspond to random variables.” This attributes of and relationships among the entities, and apply a is the approach taken in PR-OWL 2. probabilistic reasoner to reason with available evidence. For The key to building the bridge that connects the example, we might begin with an OWL ontology containing deterministic ontology defined in OWL and its probabilistic classes for ships, routes, persons, and other entities mentioned in the MFrags of Figure1. We would then wish to use PR-OWL extension defined in PR-OWL is to understand how to translate one to the other. On the one hand, given a concept defined in to define the probability distributions represented in the OWL, how should its uncertainty be defined in PR-OWL in a MFrags. way that maintains its semantics defined in OWL? On the other The difficulty with this idea is that PR-OWL 1 has no hand, given a random variable defined in PR-OWL, how mapping between the random variables used in PR-OWL and should it be represented in OWL in a way that respects its the properties used in OWL. For example, suppose we have uncertainty already defined in PR-OWL? defined an OWL class Ship with property isShipOf- PR-OWL 2 formalizes the relationship between OWL Interest, intended to represent whether a ship is a ship-of- properties and PR-OWL random variables using the relation interest. We might want to use the PR-OWL random variable definesUncertaintyOf [3]. In our previous example, we isShipOfInterest(ship) to define the uncertainty would use the relation definesUncertaintyOf [3] to associated with this property. We might use the ShipOfInterest relate the OWL property isShipOfInterest to the PR- MFrag of Figure 1 to specify its probability distribution. However, despite the syntactic similarity between the property OWL 2 random variable isShipOfInterest(ship). An name and the random variable name, PR-OWL 1 has no way to additional complexity arises because MEBN can represent n- specify formally that the random variable isShipOfInter- ary functions and predicates, whereas OWL has only binary properties. We must ensure that not only is the random variable est(ship) defines the uncertainty of the OWL property linked to its associated OWL property by defines- isShipOfInterest. Thus, even if we had information UncertaintyOf, but also its arguments are linked to their about whether a particular ship, say Ship379, is a ship-of- respective OWL properties by either isSubjectIn or interest, we would not be able to instantiate the random isObjectIn, depending on whether they refer to the domain variable isShipOfInterest(ship) for Ship379. or range of the OWL property, respectively. This feature is Poole et al. [10] point out the need to relate the random especially important when dealing with n-ary random variables from probabilistic theories to the individuals, variables, where each argument of the random variable will be properties and classes of ontological languages like OWL. associated with a different OWL property. Figure 3. Mapping of PR-OWL Random Variables and OWL Properties Figure 3 shows a schematic for the mapping between OWL over the type definition and type hierarchy in PR-OWL is properties and PR-OWL random variables. A full discussion delegated to OWL. of the formal mapping between OWL properties and PR-OWL In PR-OWL 2, therefore, the possible values or outcomes of random variables can be found in [3]. The mapping provides the basis for a formal definition of consistency between a PR- a random variable are instances of classes and data types. When specifying that a random variable will have individuals OWL probabilistic ontology and an OWL ontology, in which rules in the OWL ontology correspond to probability one of a class as its possible outcomes, it is reasonable to assume that all known individuals of that class form a set of assertions in the PR-OWL ontology. A formal notion of consistency can lead to development of consistency checking collectively exhaustive outcomes. However, the assumptions about individuals in OWL are not enough to guarantee these algorithms. individuals are mutually exclusive. More specifically, although Another major difference between PR-OWL 1 and PR- OWL provides a way to express unique names, it also allows OWL 2 is that the separate definition of entity in PR-OWL is two different names to point to the same object in the real replaced by OWL’s built-in notion of classes and data types. world. To address this issue, PR-OWL 2 follows the MEBN That is, a PR-OWL entity is now identified with either a class and PR-OWL 1 convention, and assumes that every individual or a data type in OWL. Moreover, since OWL supports has a unique ID associated to it. multiple inheritance, so does PR-OWL 2. Thus, all the control Figure 5. Entity-Relationship Diagram for Maritime Ship Ontology a. Evidence: Identify the size of the ship; We note that there are certain aspects of the full PR-OWL b. Evidence: Confirm the ship is a warship; semantics that are not fully captured in OWL-DL, and c. Evidence: Identify the primary mission of therefore cannot be handled by OWL-DL reasoners, but are expected to be respected by PR-OWL reasoners. In particular, the ship based on its weapons and sensors. to specify the restriction that a random variable defines the 2. Query: What nation has flagged the ship? uncertainty of a property would require OWL Full. For this reason, the restriction is not explicitly represented in PR-OWL, a. Evidence: Identify the nation under which but it is expected to be enforced by a PR-OWL probabilistic the ship is registered. reasoner. This enables consistency checking of the The entity-relationship diagram of Figure 5 presents a deterministic part of a PR-OWL ontology using a DL reasoner. simplified design of the Military Ship Ontology illustrating the primary attributes used to answer these queries. The decision IV. PR-OWL 2 CASE STUDY maker desires to know the class of warship that he faces. A The following case study demonstrates the application of class of ships has a consistent hull design and a standardized probability to an existing ontology to represent uncertainty in suite of weapons and sensors. These weapons and sensors knowledge about instance attributes. In this case, an existing work in concert to provide synergy in executing the primary ontology of Western European warships identifies the major mission of each type of ship. By combining a ship type with characteristics of each combatant class through the attributes of the nation that operates it, a logical prediction of warship class size, sensors, weapons, missions, and nationality. Figure 5 may be obtained. shows an entity-relationship diagram for the ontology. The decision maker is trying to determine the warship class of a International law of the sea requires that each merchant ship contact about which he has limited information. By adding is registered and sails under a single nation for the purpose of probability to the existing ontology, we can identify the most regulation, certification, and pollution control. That process is likely class of ship he is encountering when provided only known as flagging, and an individual ship is flagged by a partial or uncertain information. The model is designed to nation. It is not required that a ship is flagged under the same answer the following query using the following evidence: nation as its owner; a “flag of convenience” allows a ship to be operated under an alternate nation to reduce operating costs and Overall Goal: Given uncertain or absent attribute regulations. However, warships are always flagged under the information about a specific ship, what is the most likely nation of ownership. European warship class that satisfies these attributes? 1. Query: What is the type of warship? Figure 6. Military Ship Probabilistic Ontology The Gross Naval Class is a naval schema that delineates hasWarshipType(ship) and hasFlag(ship). The warships from merchant ships, and is mutually exclusive. WarshipType MFrag may be further decomposed into the Through identification of weapon and sensor attributes, as ShipSize, GrossNavalClass, and PrimaryMission MFrags. well as overall ship size, a Gross Naval Class estimate may The GrossNavalClass MFrag is influenced by both the be made for the unknown ship. While it can be assumed that ShipSize and ShipSensor MFrags through the all ships have a radar sensor, only military ships have sensors hasShipSize(ship) and hasSensor(ship) RVs, associated with weapons systems. The presence of a weapon while the PrimaryMission MFrag is influenced by the system, or a weapon-associated sensor, provides reasonable ShipSensor and ShipWeapon MFrags with evidence that a ship is a warship. hasSensor(ship) and hasWeapon(ship) RVs. Warships are of different types based on their primary With the MTheory complete as shown in Figure 6, the Local mission. Most ships have multiple mission capabilities, but Probability Distribution (LPD) must be populated. for this ontology we assume the following primary mission areas by ship type: Prior probabilities for the hasFlag RV were obtained from an estimate of merchant ship registrations available Anti-Air Warfare (AAW): through open source information. Similarly, − Aircraft Carrier (CV, CVN) hasShipSize represents a finite and exhaustible set of − Cruiser (CG) ship lengths (LengthLess150m, Length150to- − Guided Missile Destroyer (DDG) 100m, LengthGreater200m) into which each ship is − Guided Missile Frigate (FFG) categorized. Prior probability estimates were again obtained Anti-Surface Warfare (ASuW): via open source literature. Priors for hasSensor and − Destroyer (DD) hasWeapon were obtained through subject-matter-expert Anti-Submarine Warfare (ASW): review of open source literature and represent the proportion − Frigate (FF) of warships with each of the types of sensors. LPDs for the GrossNavalClass and PrimaryMission MFrags require By observing the combination of weapons and sensors, it is conditional statements about relationships from the input possible to infer the most likely mission area. This, nodes shown in Figure 6. A detailed description of these combined with an estimate of ship size, provides an relationships is described in a forthcoming paper. indication of the type of warship. Queries to the Military Ship Probabilistic Ontology are At this point an MTheory is created to determine processed in UnBBayes-MEBN using an implementation of hasWarshipClass(ship) in the WarshipClass MFrag the situation-specific Bayesian network (SSBN) construction algorithm. Instances of unknown ships and representative for some unknown ship. The eight MFrags associated with evidence are entered via the OWL ontology through the this determination are shown in Figure 6. Inputs to hasWarshipClass RV are the RVs from the UnbBayes GUI to reflect partial or uncertain information WarshipType and Nationality MFrags, representing the concepts introduced above with the RVs Figure 7 Situation-Specific Bayesian Network Military Ship Classification about ship attributes. These are checked against known PR-OWL to define uncertainty about attributes of and characteristics provided by subject-matter experts. relationships among the entities, and apply a probabilistic reasoner to reason with available evidence. Using the For example, suppose the following evidence is obtained formalized construct introduced in PROWL-2, we map each about a ship of interest: of the RVs in the MFrags of the probabilistic ontology to the • UID: Surcouf existing OWL property in the original ontology. This is accomplished through the probabilistic ontology building • hasNavalGun(Surcouf): True sequence executed on the UnbBayes software. For example, • hasFlag(Surcouf): France the WarshipType class in OWL has an object property of hasPrimaryMission. This object property is mapped to • hasShipSize(Surcouf): <150m the hasPrimaryMission(ship) RV of the Executing a query of the isWarshipClass node PrimaryMission MFrag. Mappings produced for each RV produces the SSBN found in Figure 7. In this case, there is a and its associated property in OWL allow us to use PR-OWL 68% chance that Surcouf is a member of the French to reason probabilistically about uncertain aspects of an LaFayette Class of frigates, which is the correct existing ontology based on the information already available. classification. V. CONCLUSION As discussed in Section III, our goal is to begin with an Combining uncertainty reasoning with semantic OWL ontology containing information about a domain, use technology is necessary for robust, interoperable, net-centric fusion and decision support systems. The probabilistic ontology language PR-OWL provides a way to represent and reason with probabilistic ontologies. PR-OWL 2 improves compatibility with OWL in several important respects. Through a case study, this paper describes the construction of a probabilistic ontology obtained by enhancing an existing OWL ontology with probability information. REFERENCES [1] P. C. G. Costa, Bayesian semantics for the semantic web, PhD dissertation, Fairfax, VA, George Mason University (Jul. 2005). [2] L. Predoiu, H. Stuckenschmidt, Probabilistic extensions of semantic web languages - a survey, in: The Semantic Web for Knowledge and Data Management: Technologies and Practices, Idea Group Inc, 2008. [3] R. N. Carvalho, Probabilistic ontology: Representation and modeling methodology, PhD dissertation, Fairfax, VA, George Mason University (Jun. 2011). [4] K. B. Laskey, MEBN: a language for First-Order bayesian knowledge bases, Artificial Intelligence 172 (2-3) (2008) 140–178. [5] B. Milch, S. Russell, First-Order probabilistic languages: Into the unknown, in: Inductive Logic Programming, Vol. 4455 of Lecture Notes in Computer Science, Springer Berlin / Heidelberg, 2007, pp. 10–24. [6] R. N. Carvalho, P. C. G. Costa, K. B. Laskey, and K. Chang, “PROGNOS: predictive situational awareness with probabilistic ontologies,” in Proceedings of the 13th International Conference on Information Fusion, Edinburgh, UK, Jul. 2010. [7] Carvalho, R.N., Haberlin, R., Costa, P., Laskey, K.B. and Chang, K.C., Modeling a Probabilistic Ontology for Maritime Domain Awareness. Proceedings of the Fourteenth International Conference on Information Fusion, July 2011. [8] Matsumoto, S., Carvalho, R., Costa, P., Laskey, K.B., Santos, L.L. and Ladeira, M. There’s No More Need to be a Night OWL: on the PR-OWL for a MEBN Tool Before Nightfall. in Introduction to the Semantic Web: Concepts, Technologies and Applications, G. Fung, Ed. iConcept Press, 2011. [9] R. Haberlin, P. C. G. da Costa, K. B. Laskey, Hypothesis management in support of inferential reasoning, in: Proceedings of the Fifteenth Inter- national Command and Control Research and Technology Symposium, Santa Monica, CA, USA, 2010. [10] D. Poole, C. Smyth, R. Sharma, Semantic science: Ontologies, data and probabilistic theories, in: Uncertainty Reasoning for the Semantic Web I, Vol. 5327 of Lecture Notes in Computer Science, Springer Berlin / Heidelberg, 2008, pp. 26–40.