A Decision-making Format for the Semantic Web [Position Paper] Eva Blomqvist Marion Ceruti Don McGarry STLab, ISTC-CNR Jeff Waters MITRE Corporation eva.blomqvist@istc.cnr.it Space and Naval Warfare dmcgarry@mitre.org Systems Center Pacific marion.ceruti@navy.mil; jeff.waters@navy.mil ABSTRACT First responders and emergency managers work under dif- This paper describes the work of the W3C Decisions and ficult conditions using current mechanisms for information Decision-making Incubator1 , with the goal to identify re- sharing; they need improved solutions. For example, paper- quirements for a standard decision format, through a set based Incident Command forms provide an initial standard- of use cases, and to develop a first version of a potential ization of emergency information3 . An Incident Command standard format for representing decisions, fulfilling the re- Structure (ICS) can organize responders into a hierarchical quirements of the use cases and exploiting semantic web structure of sections (e.g. Operations, Planning, Logistics, standards. Ongoing efforts include the identification and Finance) and roles (e.g. Incident Commander, Public In- modelling of ‘decision patterns’ and development of proof-of- formation Officer, Safety Officer)4 . XML-based standards concept applications to validate assumptions and patterns. are being developed to improve sharing of emergency infor- mation. The Organization for the Advancement of Struc- tured Information Systems (OASIS) has a family of stan- Keywords dards known as the Emergency Data Exchange Language Decision Making, Decision Format, Ontology Pattern (EDXL)5 . The Emergency Data Exchange Language Com- mon Alerting Protocol (EDXL-CAP) exemplifies simple, use- 1. INTRODUCTION ful, and understandable information-exchange formats. What EDXL-CAP did for alerts, a Common Decision Exchange The time and effort we spend converting our decisions into Protocol (CDEP) could do for decisions [6]. work products, such as briefs, proposals, and communication An important next step is to utilize the semantic web stan- of decisions in meetings, conversations, and emails, could be dards, including RDF, SPARQL, OWL and GRDDL to in- reduced if we had a standard format for representing and tegrate information for dynamic queries across datasets, and sharing decisions. Our tools could be instrumented to gen- for inferencing using the underlying ontologies (e.g. indicat- erate our decisions in a format that could be shared and also ing that the emergency equipment named X in one jurisdic- track the state of decisions within the decision-making pro- tion is the same as the type named Y in another jurisdic- cess. Instrumentation could support the development of a tion). Initial steps in this direction are already being taken, metric of information flow and help us optimize our decision e.g., through the OASIS Distribution Element (DE) sup- processes across our organization or enterprise [7]. Visibil- porting packaging and addressing of emergency management ity of the decisions in their formation and evolution would information for purposes such as routing. The standard enable proactive management and assistance from others [8]. has links to externally-managed ‘lists’ representing concepts 1.1 Usage Scenarios such as ‘senderRole’, ‘receiverRole’ and ‘keywords’. Ontolo- gies should encapsulate, in a machine-understandable man- Sharing decisions across a broad and diverse set of users ner, such information sharing policies. Implicitly present and systems is an important aspect of situational awareness is the underlying decision-making process, continuing at all in many domains, for instance, in emergency management2 . levels through an emergency. The decision format advocated During an emergency, decisions must be shared among emer- in this paper will support the move toward the use of linked gency managers and first responders from multiple organi- data [1], and the recognition of the significance of informa- zations, jurisdictions, and functional capabilities. For exam- tion sharing policies utilizing semantic standards. ple, decisions to route patients must be shared among first The need for representing, sharing and managing deci- responders in the field who are sending the patients, those sions in a machine-understandable format is not exclusive who are doing the transport, the medical facilities receiving to emergency management. One example of another critical the patients, and the patient’s families and relatives. 1 3 For more information, or to participate in the Decisions Incubator, For examples of incident command forms, see http://training. please review the charter at http://www.w3.org/2005/Incubator/de fema.gov/EMIWeb/IS/ICSResource/ICSResCntr Forms.htm. 4 cision/charter and visit the wiki at http://www.w3.org /2005/Incu- For more information on ICS, see the online training provided bator/decision/wiki/Main Page. by the U.S. Federal Emergency Management Agency, Lesson 3, at 2 http://emilms.fema.gov/IS100A/indexMenu.htm. For more information on emergency and incident management, see 5 for example the National Incident Management System, December For a good overview of EDXL, see http://en.wikipedia.org 2008, published by the U.S. Department of Homeland Security at /wiki/EDXL. The EDXL family of standards is available at the OA- http://www.fema.gov/pdf/emergency/nims/NIMS core.pdf. SIS website: http://www.oasis-open.org/home/index.php. domain of interest is organizational innovation. Each per- son is a ‘decision-maker’ at some level in the organization. The decisions a person makes are critical to the success of an organization, so aspects of decision-making and objective measures of the decision-making process become significant. Decisions involve weighing reasonable options based on met- rics in order to take an action. If we granulize the decision- making process by considering each member of our organi- zation as a decision-maker, then we can support the repre- sentation and sharing of individual innovative actions. Most Figure 1: The AgentRole Content ODP’s graphical organizations attempt to solve this problem through direct representation in UML. or indirect person-to-person communication (e.g., meetings, telecons) or unstructured collaborative tools (email, chat, wiki). XML formats can support notice-type publishing of 2.1 Ontology Design Patterns activities, e.g, RSS or ATOM feeds; however, there remains Under the assumption that classes of problems in ontology an opportunity to showcase semantic standards to capture design can be solved by applying common solutions (as expe- decision-making to improve the querying, inferencing, and rienced in software engineering), ODPs can support design integration with underlying ontology support. reusability. ODPs can be of several types [3], e.g, focusing The focus of this paper is on the information sharing as- on logical language constructs, architectural issues, naming, pects of a decision, which is fostered by a format which is or on the efficient provision of reasoning services. In this concise, generic, i.e., domain independent, and tiered. The paper we focus on Content ODPs (CPs), which are small or more concise the format, the more quickly it can be under- cleverly modularized ontologies with explicit documentation stood and accepted by developers and users alike. of design rationales. CPs can be used as building blocks in ontology design [2]. As an example we describe a CP called 1.2 Project Goals AgentRole. It represents the relation between agents, e.g., The work performed by this incubator activity is designed people, and the roles they play, e.g., manager and meet- to help organizations improve integration of human decisions ing chair. Figure 1 shows the UML diagram7 of the OWL8 into computer systems, to track and manage digitally the building block representing this CP. decision-making process, to enable improved information- CPs are collected in different catalogues, such as the ODP flow metrics, to maintain an archive of the decisions and portal 9 . In addition to their diagrammatic representation, the decision-making process, to enable semi-automation of CPs are described using catalogue-entry fields (c.f. software certain decision-making processes, to improve information pattern templates), such as name, intent, covered require- sharing, and ultimately, to support better, rapid, and agile ments, consequences, and building block, linking to an OWL decision making [7]. The potential standard format should realization of the pattern. The requirements an ODP covers provide concise, generic, structured assessments and deci- are expressed using Competency Questions (CQs) [4], i.e., sions that allow ‘drill down’, support pedigree and confi- typical natural-language queries. dence, enable approvals and vetting, define options consid- ered, including decision criteria with weighting, links to pre- 2.2 eXtreme Design vious decisions and sub-decisions, and support flexible struc- With the name ‘eXtreme Design’ (XD) we identify an ag- turing of complex decisions [7]. However, to reach its full ile approach to ontology engineering [5]. In this paper we potential, the proposed decision format must be compatible focus on XD for CP reuse in ontology design. In XD a devel- with semantic web tools and standards, to provide semantic opment project is characterized by two sets: (i) the problem interoperability and to provide a basis for reasoning that can space, composed of the actual modeling issues (local prob- ease development of advanced applications. lems), e.g., to model steps in a decision making process; (ii) In summary the main goals of the incubator are: the solution space, made up of reusable modeling solutions, e.g., a piece of an ontology that models sequences of events • To discover a set of requirements for a standard deci- (a CP). Each CP, as well as the local problem, is related to sion format, through a set of use cases. ontology requirements expressed as CQs or sentences. If a • To develop a draft of a potential standard format for local problem can be described in terms of the CQs of a CP representing decisions, fulfilling the requirements of then that CP can be reused for building the solution. XD the use case and exploiting semantic web standards. does not prescribe a specific method for matching the local problem to patterns, and at the moment the only tool sup- port available are search functionalities utilizing the textual 2. METHODOLOGY BACKGROUND descriptions of the patterns. Creating a vocabulary for expressing decisions that ex- XD is a test-driven and task-focused approach that re- ploits semantic web standards means, in practice, creating sults in highly modular ontologies. The main principles of a set of ontology modules that can be linked in a network, XD include the intensive use of CPs, and extensive collabo- to be used independently or together in different combina- ration [5]. The iterative workflow of XD contains 12 steps. tions. The main tools we use for this practical task is the The project is initiated in the first four steps, which in- eXtreme Design ontology engineering methodology and the 7 notion of Ontology Design Patterns (ODPs), supported by For notation details, see: http://www.topquadrant.com/products/ TB Composer.html the ontology development environment NeOn Toolkit6 . 8 http://www.w3.org/2004/OWL/ 6 9 http://www.neon-toolkit.org http://www.ontologydesignpatterns.org clude, scoping, and requirements engineering (e.g., deriving base new decisions on the collected knowledge in the the CQs from user stories). In steps five through nine the recorded decisions of the organization. CQs are divided into into small, coherent sets and ontology modules produced realize those sets of CQs. These steps 3.1.1 Measuring Information Flow include unit tests on each module before its release. The Research shows that an analytical solution of information three final steps integrate modules into a coherent solution, velocity is intractable but metrics that support the under- focusing on collaboration and integration. standing of information flow can be useful [8]. An agent- based model for information flow can be used to character- 3. ONGOING WORK ize physical analogs to causal measures [6]. In this use case, interactions and exchanges can be modeled as physical prop- In this section we describe our ongoing efforts and how erties. Information, its suppliers, and consumers are then we apply the XD methodology to support these efforts. We treated as agents. The behavior of the agents and system as proceed in a bottom-up fashion, starting from the use cases a whole can be discussed and infodynamic analogs of ther- and deriving requirements for a representation format that modynamic and other physical quantities associated with can be realized as ontology modules based on ODPs. How- these processes could be explored [8]. The use of concep- ever, we have also encountered a number of cases where this tual analogs from the physical domain implies the viability leads to the development of general ODPs themselves. of future ontologies to characterize information flow. 3.1 Use Cases 3.1.2 Automatic Assessment of Options Use cases are in our context general scenarios, horizon- Design considerations have been described and exempli- tal with respect to application domains (i.e., they are rep- fied for implementing a decision-acquisition system based on resented in multiple domains), where the envisioned deci- a CDEP [7]. CDEP is an XML- and REST-based protocol sion format can give some substantial benefit. So far, five for representing generic human decisions in a simple, inter- use cases have been identified (the list is continuously ex- operable format. The characteristics of decisions can be ex- tended). The use cases are intended to be general and not pressed using CDEP and its proposed XML format [7]. The domain specific, in terms of industry domain. Their detailed CDEP concepts will be considered, and enhanced, within description, including resulting requirements in the form of the currently envisioned decision format, and a conversion CQs can be found in the Incubator wiki10 . Background and XSLT stylesheet will enable interoperability across these for- related work for two of the use cases are described more in mats as needed. The use case ontology would allow for the depth in Sections 3.1.1 and 3.1.2. consideration of multiple data sources, multiple decision op- • Measuring Information Flow - Where a decision tions, and the tracking of decision confidence throughout the process representation can help answering questions decision-making process. such as ‘When did a certain process begin and end?’, ‘How much time was spent on a certain step in the 3.2 Decision Patterns process?’, and ‘What is the average time for making a The decision patterns include concrete decision format certain type of decision?’. components, as well as generic patterns, hence, both: • Linked Data Supporting Decisions - Where linked • The ontology modules that we propose as a starting data [1] supports decision making, and a decision rep- point for creating a standard in this field, resentation format could help answer questions such as ‘What data support this decision?’, ‘What were the • and the more general ODPs that we discover and de- options and the criteria used for this decision?’, and velop as a result of this effort. ‘How were the options assessed?’ The first module draft that was produced corresponds to the • Automatic Assessment of Options - Where a de- use case of ‘Measuring Information Flow’ listed above. This cision format is intended to support semi-automatic ontology module is a specialization of the Transition ODP11 . decision making by automatic assessment of options In this case we found an ODP already available that we could through some metric. In this case questions are for specialize and create a specific decision-process pattern. In instance ‘What are the metrics for this decision and to other cases, such as when viewing a decision as a past event, what options do they apply?’, ‘What are the relative no ‘event-pattern’ was available in the ODP portal. There- weights of different metrics?’, and ‘How will the met- fore, we are creating general ODPs to be specialized in the rics be combined to generate an overall assessment?’ decision ontology modules. By treating general (rather than domain-specific) use cases of decision-making, we make sure • Interoperability - For example, a shared decision that the developed modules are actually reusable patterns, representation can support interoperation between dif- rather than a solution tailored to one specific application. ferent command and control units and between deci- All decision patterns will be implemented in RDF/OWL. sion makers and people implementing decisions. Eight patterns are identified so far, but need to be created. Four examples are described below: • Situational Awareness - A representation of de- cisions and the decision-making process can support • A ‘Statement with variable’-pattern, to describe queries, systems and/or organizations to be aware of the de- such as the question underlying a decision. cision status, to identify situations, such as the situ- ation when important information is missing, and to • ‘Filter’ and ‘Aggregation’-patterns, where a filter would 10 represent criteria applicable to some data, e.g., a set http://www.w3.org/2005/Incubator/decision/wiki/Final Report 11 Use Cases http://ontologydesignpatterns.org/wiki/Submissions:Transition of options, and an aggregation would represent a way 3.2). We intend to submit any patterns developed (both to combine data, e.g., grouping options. general and specific to decision-making) to the ODP portal. We expect to present a set of proof-of-concept applications • A ‘Normalization’-pattern that models transformations (see Section 3.3). These applications will show the poten- of values into a common scale, for comparing options. tial of our draft patterns. The applications will be used to • A ‘Weighting’-pattern to express the relative impor- validate our results against current practices in different do- tance of data, e.g., weighting of assessment criteria. mains, e.g., to validate the hypothesis that linked data are suitable to support decision making and that automatic as- 3.3 Proof-of-concept Application sessment of options is possible in certain use cases. During To verify the requirements and the ontology modules, and the project, we will make the problems and possible solu- to demonstrate the usefulness of such a format, a demonstra- tions visible in different communities, e.g., the semantic web tion system is being developed at the Space and Naval War- community, domain specific interest groups, and standards fare Systems Center Pacific. Initially, the system will focus organizations. We envision that at the end of the project on enabling decision making using open linked data sets [1]. we can propose a standardization effort in the context of The user has four modules, or screens. In the Topic screen, W3C. We can pursue several use cases and application ideas the user enters the key question of the decision, keywords, as separate research projects. and where the decision result will appear. The keywords will drive a search for relevant open-linked data sets. Next, 5. ACKNOWLEDGMENTS the user selects a data set from which the entries provide a The authors thank the Office of Naval Research for their named set of options. From the Options screen, the user se- support of this work. This paper is the work of U.S. Govern- lects the properties to use as metrics. On the Metrics screen, ment employees performed in the course of their employment the user selects filtering criteria to reduce the options. The and no copyright subsists therein. user can additionally assign weights to the metrics. When a similar decision is encountered, users can efficiently select a named set of Options or Metrics to aid reuse of decision 6. REFERENCES components. A semi-automatic learning process will be con- [1] C. Bizer, T. Heath, and T. Berners-Lee. Linked Data - sidered for future releases, proposing named sets of options The Story So Far. International Journal on Semantic or metrics found useful to other users, based on similar- Web and Information Systems, 5(3):1–22, 2009. ity of questions and keywords. On the Assessment screen, [2] E. Blomqvist. OntoCase-Automatic Ontology the filtered options appear in an ordered list based on the Enrichment Based on Ontology Design Patterns. In weighted metrics. The user selects one or more options as ISWC 2009, 8th International Semantic Web the answer to the decision question. The user is returned Conference, volume 5823 of LNCS, pages 65–80, to the Topic screen where the answer(s) is/are recorded and Washington, DC, November 2009. Springer. visible. Throughout the process, the time spent in the vari- [3] A. Gangemi and V. Presutti. Ontology Design Patterns. ous stages is tracked to assess information flow. Future ver- In Handbook on Ontologies, 2nd Ed., International sions of this system will support manual entry of decisions, Handbooks on Information Systems. Springer, 2009. a more robust set of filtering criteria, integration of multiple [4] M. Gruninger and M. S. Fox. The role of competency datasets, and mobile applications for efficiency in the field. questions in enterprise engineering. In Proceedings of The decision format discussed here will be used to manage the IFIP WG5.7 Workshop on Benchmarking - Theory the decision as a whole, and its modular components. and Practice, 1994. 3.4 Experiences [5] V. Presutti, E. Daga, A. Gangemi, and E. Blomqvist. eXtreme Design with Content Ontology Design An important outcome, apart from the requirements and a Patterns. In Proc. of WOP 2009, collocated with proposed decision representation, will be experiences related ISWC-2009, volume 516. CEUR Workshop to the XD methodology and ODPs. XD has been used in the Proceedings, November 2009. project both as a framework for the modelling but also as a [6] J. Waters and M. Ceruti. Modeling and simulation of means for teaching ontology engineering to participants less information flow: A study of infodynamic quantities. In familiar with semantic technologies. So far we found that Proc. of the 15th International Command and Control the level of detail of the XD methodology is highly benefi- Research and Technology Symposium (ICCRTS 2010), cial for teaching ontology engineering to novice modelers. It Santa Monica, CA, June 2010. introduces an intuitive way of scoping the problem, through modularization, and it allows the modeler to draw on previ- [7] J. Waters, M. Ceruti, R. Patel, and J. Eitelberg. ous experiences of others through ODPs. We envision that Decision-acquisition system based on a common the project will benefit the further development of XD, and decision-exchange protocol. In Proc. of the 15th XD will be validated through valuable experiences. International Command and Control Research and Technology Symposium (ICCRTS 2010), Santa Monica, CA, June 2010. 4. OUTLOOK [8] J. Waters, R. Patel, J. Eitelberg, and M. Ceruti. In September 2010, the project reached its half-way point Information velocity metric for the flow of information and should be completed by the end of March 2011. By through an organization: Application to decision that time the project will have a set of requirements for a support. In Proc. of the 14th ICCRTS (ICCRTS 2009), potential decision-representation standard, i.e., the use cases Washington, DC, June 2009. (initial set in Section 3.1), and a first draft of such a repre- sentation, i.e., the decision patterns (initial ideas in Section