Presentation on theme: "Using Ontologies to make Smart Cities Smarter"— Presentation transcript:
1 Using Ontologies to make Smart Cities Smarter Rosario Uceda-Sosa, Biplav Srivastava and Bob SchlossIBM Research@.ibm.comJune 2012
2 Application Developer/ A Semantic Data Model for Smart CitiesA semantic data model (an ontology) of a city, if it is complete and authoritative, (1) simplifies the development of applications that require integrated access to city data sources and (2) enables solution reuse as we move from one city to the next.Independently of using ETL for data consolidation, a semantic data model (3) can extend the metadata with new categories (SanitationServices, CrimesAgainstProperty) without modifying the application or the data sources.SemanticData Model2. Reuse3. Metadata ExtensionsApplication Developer/Consultant[ETL]Data sourcesData ModelAn ontology can make a city interconnected and smart, but it needs to assume thatCities have their own data sources, not necessarily connected, and may not want to consolidate them.Cities have non-standard organizations, departments and competencies.
3 … but, what is an ontology, anyway? What do you think?
4 … but, what is an ontology, anyway? In Computer Science, “An ontology is a formal explicit description of concepts in a domain of discourse (classes (sometimes called concepts)), properties of each concept describing various features and attributes of the concept (slots (sometimes called roles or properties)), and restrictions on slots (facets (sometimes called role restrictions)). An ontology together with a set of individual instances of classes constitutes a knowledge base. In reality, there is a fine line where the ontology ends and the knowledge base begins.” [Noy, 2000]Not to be confused with ontologies (and/or taxonomies) in Philosophy or Life SciencesIn a Smart City domain, we’re concerned with modeling the city data (city activity data, city departments, assets, KPIs), not the city itself (the full set of spatial and temporal relations between people and objects in the city) Ontologies help us to structure and reason about city events, entities and services.Ontology = Class + Relations + ConstraintsKnowledge Base = Ontology + instances + (Standard) Inference and rules
5 Not all ontologies are created equal In practice, ontologies are used -together with inferencing engines and rules-, for a variety of purposes. If we think of them as schemas, there are different waysPurposeInstancesInferencingExamplesAs a deductive systemDeductive System (axioms + deductive rules)Part of the knowledge baseDefined by rules.Expert systems, Planning, Optimization.As a data blueprintConstrain a domainMust conform to the normative schema determined by the ontologySubsumption, class inferencingBiomedical and life sciences (FMA, Radlex)As a data classifierClassify open dataUnknown formatsTag ontologies (MOAT, Echarte, SCOT, NAO, etc.)As a data integratorIntegrating pre-defined model to existing data sourcesInstances are mapped, no constraint enforcement.Subsumption, class, entity inferencingSCRIBEAs data mapping vocabularyMapping to/from existing data sourcesMined instances determine the ontology/schema.D2RQ (a tool)NormativeschemaIntegrativeSchema,dependon instancesSCRIBE belongs to the fourth category: It has no constraints and was designed to support the programming of tools that allow domain experts to deal with entities natural to them (even if the recorded data is actually distributed).
6 What makes a good ontology for data integration? Human Usability A good ontology is a useful ontology, an ontology that both humans and systems can process.Human UsabilityCommunicable. Naming, natural language support, etc.Concise. A simple way to describe the key entities of the model and yet able to infer many factsConsistent. Naming conventions and modeling patternsAuthoritative to domain expertsDocumented, not just descriptions, but also provenanceManaged and maintained by people throughout the model lifecycle.Reusable in similar domains, for similar instances.System UsabilityScalable so large amounts of data can be parsed, stored and retrieved.Efficient query and inferencingProgrammable solutions, both in open and closed data paradigms.Open infrastructure and tools
8 Scribe design decisions Human Usability A good ontology is a useful ontology, an ontology that both humans and systems can process.Human UsabilityCommunicable. Naming, natural language support, etc.Natural language naming, user readable labelsConcise. A simple way to describe the key entities of the model and yet able to infer many factsAnchor classes: events, services, assets, KPIs. Simple and expressive OWL sublanguage, relation taxonomiesConsistent. Naming conventions and modeling patternsClear boundaries between classes and instances.Authoritative to domain expertsAlignment with standardsDocumented, not just descriptions, but also provenanceWealth of annotationsManaged and maintained by people throughout the model lifecycle.Class stewards, involvement of domain experts and end usersReusable in similar domains, for similar instances.Mechanisms for modularization of extensions and customizationsSystem UsabilityScalable so large amounts of data can be parsed, stored and retrieved.Caching mechanisms for DB data (?)Efficient query and inferencingOntology-based inferencing (?)Programmable solutions, both in open and closed data paradigms.Data adapters and schema exploring (?)Open infrastructure and toolsJena, DB2DRQ, Ruby on Rails, etc.
9 Organization/Operation profile SCRIBE data modelSCRIBE is a non-normative, authoritative, modular, extensible semantic model for Smarter Cities.It consists of a Core Model that includes common classes (events and messages, stakeholders, departments, services, city landmarks and resources, KPIs, etc.), extensions by domain and customizations by city.Simple languageClasses + Inheritance + Relations + InferencingBased on standards (OWL-QL, SPARQL)Mappable to UMLMetadata annotations and TaggingSimple languageClasses + Inheritance + Relations + InferencingBased on standards (OWL-QL, SPARQL)Mappable to UMLMetadata annotations and TaggingSimple languageClasses + Inheritance + Relations + InferencingBased on standards (OWL-QL, SPARQL)Mappable to UMLMetadata annotations and TaggingCommonbuilding blocksSCRIBE Core ModelCity CustomizationExtensionOrganization/Operation profileAuthoritativeAligned with standards (CAP, NIEM, MISA/MRM, UCore)Validate with customer scenariosValidated with open city dataAuthoritativeAligned with standards (CAP, NIEM, MISA/MRM, UCore)Validate with customer scenariosValidated with open city dataAuthoritativeAligned with standards (CAP, NIEM, MISA/MRM, UCore)Validate with customer scenariosValidated with open city dataWeatherWaterTransportationBuildingAndParcelAssetManagementFeatures
10 The key concepts of the SCRIBE Ontology 1. Describes messages, events and services as they flow through the systemBefore/aftertriggersBefore/aftertriggersBefore/aftertriggersMessage(Advisory)Event(Storm, RoadWork)WorkItem(RoadWorkWI)Protocol(InfrastructureWorkP)Asset(pipe, valve)2. Represents types of city services (not the city organization itself) so the administrative structure of a city can be assembled from SCRIBE building blocksCityServiceAreaCityServiceAreaOwnsAgency(WhitePlainsTraffic)For example, RegressionModel can be used for classification –as well as prediction- because multiple regression equations can be combined to predict categorical values.
11 CAP UCore NIEM MISA/MRM City and Government Standards and SCRIBE While most of the standards relevant to Smarter Cities are message exchange models (CAP, UCore, NIEM) or business planning (MISA/MRM) , SCRIBE integrates the (1) message-based models with (2) asset management and (3) services and their KPIs in an extensible model.CAPUCoreNIEMMISA/MRMCore entitiesAlert, message certainty, security, urgencyIncidentPeople, Places, Events and ThingsProgram, service, outcome, target group, outcome.AdvantagesSimple to implement and read. Established standardExtension mechanisms defined. Supported by DoD, DHS, DoJ.Tools for search and subset extraction (SSTG) Established standard. Well defined extension process (IEPD)International, municipality basedIssuesSubject and related resources are underdefinedNot mature enough, incomplete.Large (4000 concepts) and cumbersome (even with support tools) Not deep in any domainRepresents administration, business planning of a city, not its operation. Cumbersome to extend.Representational LanguageXMLXML with schema substitution for inheritanceXML (rdfs?)For example, RegressionModel can be used for classification –as well as prediction- because multiple regression equations can be combined to predict categorical values.
12 (Person, Organization) Smarter City Standards and SCRIBE(1) A message is an event (with publisher/subscribers or requestors/responders) AND it has as a subject an (external/processing) event. In principle, a message could refer to another message.Entity(Person, Organization,- item)Role(Person, Organization)Organization(CityOrganization)ServiceArea(Public Safety, etc.)causesisStakeholder ->hasRole ->StakeholderEventIs-a(1)AssetExternalEventWorkItemMessagesubjectFor example, RegressionModel can be used for classification –as well as prediction- because multiple regression equations can be combined to predict categorical values.Maximo-BasedOverlap, superset, etc.Tom TravisPlannerTransportationDeptCAP-BasedOverlap, superset, etc.Stakeholder1RoadRepairWorkOrderIntersection: MainAnd HamiltonNIEM-BasedOverlap, superset, etc.
13 The SCRIBE MetadataFor example, RegressionModel can be used for classification –as well as prediction- because multiple regression equations can be combined to predict categorical values.
14 Inferencing and object properties There are three types of ‘horizontal’ relations:HasAttribute (inv. attributeOf) for properties and attributes (name, identifier, etc.)HasAggregateMember (inv aggregateMemberOf) for parts or members (hasChild, a process has process steps as members)AssociatedTo (its own inverse) for everything else We can do inferencing on extensions to SCRIBEFor example, RegressionModel can be used for classification –as well as prediction- because multiple regression equations can be combined to predict categorical values.
15 Application Developer/ SCRIBE toolingSCRIBE is written using standard RDF/OWL editors and software (Jena)Application Developer/ConsultantEndUserModel ToolingEdit, extend modelCustomize ModelIntegrate with DataQuery/Navigate Model and DataStandard OWL/XML (TopBraid, Protégé, Pellet, SPARQL, etc. )MIDO, DB2RQL, R2DQ, etc.Form-based queries? Record-based navigation?ImplementationContentSCRIBE Core ModelSemantic model of events, city assets, geography and resources, city organization and services, KPIs, processes,Simple subset of OWL, directly mappable to UMLCity CustomizationCity Data CatalogSCRIBE is alsoa. A modeling processB. Tools to make the model usable. The first tool we’ve worked on, MIDO (Mapping Instance Data to Ontologies), allows the mapping of existing data to the SCRIBE model and is part of the process of customizing SCRIBE to a new city.MIDODatabase Schema
16 Customizing Scribe in different cities Scribe is NOT closed. We know that cities have different organizations, different service levels and different KPIs. The Scribe model is designed to provide the building blocks (service types, city departments, KPI taxonomies, CAP messages) that can be customized to define the overall operations of a cityStandards (CAP, NIEM, MISA/MRM, etc.)Scenarios/Data (cities open data)Scribe COREMIDOMaps city data toScribe.Populates modelwith instancedataWashington D.C.ChicagoDublinServicesDepartmentsAssetsKPIsServicesDepartmentsAssetsKPIsServicesDepartmentsAssetsKPIs
18 311 events in Washington D.C. Suppose a Smarter City application that manages city operations wants to display citizen complaints (311 calls) on a map, filtered by a few user-defined constraints (times, locations, type of call, etc.)A fraction of the 311 incident table (from DC Open Data) is below. Among the data we have:IdentifierType of service (code + description)Time (ServiceOrderDate, ServiceResolution date, etc.)Place (Lon/Lat, Ward, PSA, District, etc.)The agency that should handle the requestVarious qualifiers (enum types): priority, resolution, etc.)311 Requests (2010)
19 How to map 311 events to an existing model The application may access directly the 311 table by querying incidents according to given criteria:“SELECT SERVICEREQUESTID SERVICETYPECODE LATITUDE LONGITUDE WARD DISTRICT PSA DATEREPORTED FROM DC911 WHERE SomeConstraintHere”ORThe application may define an intermediate (data model) layer that:AEventBDefines a ServiceRequest object that knows how to retrieve all the data from one or more tables.Defines two objects, ServiceRequest, where all the common data to all service requests is, and DC311SvceReq, which captures the info specific to DC.IDServiceRequestTypeCDateOrderedLon/LatIS-ANotice that in (C), inheritance can be applied to locations (wards, districts, addresses, Lon/Lat points are ways to describe a location)Also, we could push the model further and have all kinds of abstractions, say, an event class that captures ID, Time, Location and Type.…DC311SvceReqWard311 Requests (2010)
20 Mapping 911 (crime) incidents Now suppose that the application wants to add the visualization of crime incidents. The corresponding open data table is shown below. Notice that it looks similar to DC311… but not quite:ID’s have different formatTime is ReportDateTime, and has a time of day, not just a dateOffenses do not have codesThere’s no referring agencyFrom the point of view of the application:We can create another query for the DC911 table and consolidate the information at the application level (requires recompiling)We can add types and data to the object model, but this bloats the objects.We can use the inheritance hierarchy to refactor the information in the model. IF the model is well thought out, the changes are minimal… But we’ll need inferencing, infrastructure to keep the graphs… We’ll be replicating RDF/OWLABCCrime Incidents (2010)
21 The right data integration point. A semantic model approach … And there are net benefits to a model-driven, semantic approach:Applications can be coded ‘in the abstract’. E.g., Display all current events independently of whether they are 311 or 911.Applications can refine the metadata without having to touch the code or the underlying data. E.g. Display all sanitation requestsApplications can be shielded from the details of the databases, like in the case of implicit joins. E.g. Display the names of the dispatchers associated with active requests.The SCRIBE model captures enough information about events to allow a small customization to work.
22 Step 1. Customizing SCRIBE for Washington D.C. SCRIBE captures the basics of events, service types, dates, etc. but we don’t expect the model to be comprehensive. For example, we didn’t model all the types of services that the 311 table had.To customize SCRIBE, we created a new file for DC, importing the core model.We may want to customize SCRIBE for a variety of reasonsSuperCans is a DC-specific program and it will likely remain in the DC specific classes.CollectingIllegalDumping or SeasonalCollection were not contemplated in the core, and they may be marked for promotion at a later date (using the modelPromotion annotation)Adding a new data property to a core class, like a DC-specific identifierNote that constraints and rules in the DC model do not need to be reflected in the mapping to SCRIBE.
23 ServiceTypeDescriptor Step 2. Mapping instance data to the modelNext, we map the data in the columns to either a data property (transferring the data into that data property, like in the case of SIMPLEREQUESTID) OR a class (to match an enumerated type, which in the case of SCRIBE is represented as a taxonomy of classes.)ServiceRequestServiceRequestIDassociatedToServiceTypehasDescriptorServiceTypeDescriptorcodeDataThis mapping is done through a mapping model and tool called MIDO, whose details are not covered here. However, we can assume that the columns in the two tables have been mapped to the SCRIBE model AND the instance data can be accessed through the SCRIBE model.
24 … Step 3. Query through the model. Query abstract classes The data from DC Service Requests and Crime Incidents can now be queried together as events, not just as service requests or criminal incidents.Query: All Events in DC, with type, District and Ward…Notice that some of the data is missing in the original table… That’s still ok
25 Step 3. Query through the model. Annotation Metadata As shown previously. The inferencing in the ontology can be leveraged in a query.Query: Public Sanitation Service Requests
26 Step 3. Query through the model. Implicit join Everything in a semantic model is connected. The service request can be linked to the name of the dispatcher of the department.Query: Select events associated to dept of Public Works and his dispatcher
28 Scribe design decisions Human Usability A good ontology is a useful ontology, an ontology that both humans and systems can process.Human UsabilityCommunicable. Naming, natural language support, etc.Key to management and model validationConcise. A simple way to describe the key entities of the model and yet able to infer many factsBalance between simple language (RDF), conciseness and inferencing power is key to usability. Map to UML.Consistent. Naming conventions and modeling patternsUse of relation taxonomy to infer relations despite extensions.Authoritative to domain expertsMerging standards is not enough. Alignment with standards allows a consistent model.Documented, not just descriptions, but also provenanceLimited benefit to end users unless coupled with sample instances or data entry formsManaged and maintained by people throughout the model lifecycle.People not always available for the full lifecycleReusable in similar domains, for similar instances.Mechanisms for promotion of changes to the core.System UsabilityScalable so large amounts of data can be parsed, stored and retrieved.Not clear whether data should remain in RDBEfficient query and inferencingImpact analysis queries may require a few seconds. This is OK.Programmable solutions, both in open and close data paradigms.dA standard library of data adapters and mappings to SCRIBE are needed.Open infrastructure and toolsWe used Jena, DB2DRQ, Ruby on Rails, etc.
29 For more information http://researcher. ibm. com/view_project. php For more information OR29
30 ReferencesA direct map of relational data to RDF, W3C working draft 14 March, 2011,R2RML: RDB to RDF Mapping Language, W3C Working Draft 24 March 2011,The D2RQ Platform v0.7 - Treating Non-RDF Relational Databases as Virtual RDF Graphs, ,Hannes Bohring and Soren Auer, Mapping XML to Ontologies, citeseerx.ist.psu.edu/viewdoc/download?doi=T. nfRodrigues, P. Rosa, J. Cardoso, Mapping XML to existing OWL ontologies, citeseerx.ist.psu.edu/viewdoc/download?doi=DB2OWL, A tool for automatic Database-To-Ontology mapping, summary?doi=Municipal Information Systems Association/Municipal Reference Model (MISA/MRM),National Information Exchange Model,D. Gonzales, C. Ohlandt, E. Landree, C. Wong, R. Bitar and J. Hollywood. The Universal Core Information Exchange Framework, Assessing its Implications for Acquisition Programs, RAND report, 2011,D. Allemang, J. Hendler, Semantic Web for the Working Ontologist, Effective Modeling in RDF and OWL, Morgan Kaufman, 2008.Noy, McGuinness, Ontology Development 101: A Guide to Creating Your First Ontology.