week 8

84 September/October 2012 Published by the IEEE Computer Society 0272-1716/12/$31.00 © 2012 IEEE

Graphically Speaking Editor: Miguel Encarnação

Toward Visualization in Policy Modeling Jörn Kohlhammer, Kawa Nazemi, Tobias Ruppert, and Dirk Burkhardt Fraunhofer IGD

Information visualization and visual analytics (VA) have become widely recognized research fields applied to a variety of domains and data-related challenges. This development’s main driver has been the rapidly increasing amount of data that must be dealt with daily. Virtually every industry or business, or any political or personal activity, generates vast amounts of data. At the same time, citizens, shareholders, and customers expect highly efficient, informed decision-making based on increasingly complex, dynamic, and in- terdependent data and information.

All this applies in many ways to public-policy modeling. As the recent financial crisis has shown, policy making and regulation are highly challeng- ing tasks. The outcomes of policy choices and in- dividual behavior aren’t easily predictable in our complex society. Ubiquitous computing, crowd- sourcing, and open data, to name just a few exam- ples, are creating masses of data that governments struggle to make sense of for policy modeling.

Increasingly, policy makers are perceiving vi- sualization and data analysis as critical to this sense-making process. Current practice uses visu- alization mainly during postprocessing. Although this is an important step in the right direction, a more promising trend is the integration of vi- sualization tools with simulation and automated analysis. This is clearly in line with the general approach employed by various domains that apply VA techniques for interactive analysis.

Here, we examine the current and future roles of information visualization, semantics visualiza- tion, and VA in policy modeling. Many experts believe that you can’t overestimate visualization’s role in this respect. For example, the recent EU roadmap for this area goes as far as saying that “ensuring appropriate visualisations can … be con- sidered a key component of a mature democracy.”1

The Policy-Making Process Policies are usually defined as principles, rules, and

statements that assist in decision-making and that guide the definition and adoption of procedures and processes. Typically, government entities or their representatives create public policies, which help guide governmental decision-making, legisla- tive acts, and judicial decisions.

Some policy-modeling research emphasizes theo- retical formal modeling techniques for decision- making, whereas some applied research focuses on process-driven approaches. These approaches de- termine effective workflows through clearly defined processes whose performance is then monitored (for example, as in business process modeling). This applied-research approach is widely seen as one way to effectively create, monitor, and opti- mize policies. One aspect of process-driven policy making is the clear definition of the sequence of steps in the process. This ensures the consider- ation of the most relevant issues that might affect a policy’s quality, which is directly linked to its effectiveness.

Ann Macintosh published one of the most widely used policy-making life cycles; it comprises these steps:2

1. Agenda setting defines the need for a policy or a change to an existing policy and clarifies the problem that triggered the policy need or change.

2. Analysis clarifies the challenges and opportuni- ties in relation to the agenda. This step’s goals are examining the evidence, gathering knowl- edge, and a draft policy document.

3. Policy creation aims to create a good workable policy document, taking into consideration a variety of mechanisms such as risk analysis or pilot studies.

4. Policy implementation can involve the develop- ment of legislation, regulation, and so on.

5. Policy monitoring might involve evaluation and review of the policy in action.

The entire life cycle is a loop.

IEEE Computer Graphics and Applications 85

Other, more specific process definitions are avail- able, but these often vary significantly. Depending on the domain—for example, in the social sciences versus in computer science—the definition has dif- ferent scopes and thus very different process steps.

For adopting visualization in policy making, we simplified the general model and introduced three iterative stages (see Figure 1):

1. Information foraging supports policy definition. So, this stage requires visualization techniques that obtain relations between aspects and circumstances, statistical information, and policy-related issues. Such visualized informa- tion enables optimal analysis of the need for a policy.

2. Policy design must visualize the correlating top- ics and policy requirements to ensure a new or revised policy’s functional interoperability.

3. Impact analysis evaluates the designed policy’s potential or actual impact and performance, which must be adequately visualized to support the policy’s further improvement.

All phases involve heterogeneous data sources to allow the analysis of various viewpoints, opinions, and possibilities. Without visualization and inter- active interfaces, handling of and access to such data is usually complex and overwhelming because too much data is available. The key is to provide information in a topic-related, problem-specific way that lets policy makers better understand the problem and alternative solutions.

Today, many data sources support policy model- ing. For example, linked open government data ex- plicitly connects various policy-related data sources (for instance, see http://data.gov.uk/linked-data). Linked data provides type-specific linking of infor- mation, which facilitates information exploration and guided search to get an overview and—later on—a deeper understanding of a specific topic. Massive, multidimensional databases for statisti- cal data also exist—for example, the EuroStat data- base (http://epp.eurostat.ec.europa.eu).

Currently, policy-modeling approaches don’t use visualization intensively either for the general pro- cess or in any of the three stages. The first research prototypes are close to traditional information visualization techniques, and no visualization ap- proach addresses all the required policy-modeling aspects. As we mentioned before, the goal of intro- ducing more visualization and VA techniques goes one level further. The objective is to ensure more ef- ficient and effective policy modeling by integrating visual methods and automatic analysis methods.

Visual Support for Policy Modeling The simplified policy-modeling process in Figure 1 enables an abstract view of policy evolution. This model can provide only an overview of the stake- holders, political processes, and activators for new or changed policies. Nevertheless, it gives a good foundation for identifying the application of in- formation and communications technology (ICT) to policy creation. In particular, graphical systems that give insight into the heterogeneous, complex, and huge amount of data and information can be adapted to the model’s stages.

Here, we look at various visualization disciplines in this context and classify visualization method- ologies on the basis of the human’s and computer’s roles in the transformation from data to insight. In this process, visualization supports users in var- ious ways. One example is integrated intelligence that recognizes patterns or clusters in data or that considers human abilities or interaction patterns when providing adaptive visualization techniques. In this context, we categorize visualization for policy modeling into information design, infor- mation visualization, semantics visualization, VA, and knowledge discovery in databases (KDD).

Figure 2 illustrates this separation of roles, in which the computer’s role increases and the hu- man’s role decreases from left to right. For example,

Information foraging

Policy design Impact analysis

Figure 1. A simplified policy-modeling process. All three stages involve heterogeneous data sources to allow the analysis of various viewpoints, opinions, and possibilities.

Information design

Information visualization

Human

Computer

Visual analytics

Knowledge discovery in databases

Semantics visualization

Figure 2. The computer’s and human’s roles in visualization disciplines in policy modeling.3 From left to right, the computer’s role increases and the human’s role decreases.

86 September/October 2012

Graphically Speaking

in information design, the computer plays no role per se, whereas in KDD, automatic methods process the underlying data typically without user interac- tion. We consider semantics visualization very rel- evant for policy modeling, as we explain later.

Information Design Political decisions often progress through stages involving stakeholders with heterogeneous skills, knowledge and preferences, and positions in the political chain. The actual decision-making often occurs higher on that chain, with lower-level dis- cussions or workshops taking analysts’ input into account to help determine policy changes. At this lower level, the data and information should be presented adequately. The main target is to con- vey a matter of interest and present the essential issues.

Information design investigates the rules for ad- equate information presentation and the commu- nication of knowledge. Here, humans with their communication skills are the major actors and use outcomes from perception science. Information design focuses on visualization to support human communication during policy making.

Information Visualization To achieve a summary of the most important data and information relevant for decision-making, political analysts must aggregate vast amounts of data, analyze it, and form an overview of it. For these tasks, they can apply information visualiza- tion techniques. Often, automatic analysis prepro- cesses the data, and the analysts usually visualize only these techniques’ outcomes. The analysts can visually and interactively browse through the re- sults and detect the most relevant information to condense for decision-makers.

Every stage of policy modeling can involve in- formation visualization. For example, information foraging could use Gapminder (www.gapminder. org) to explore the evolution of a nation’s wealth to analyze the need for new policies.

Semantics Visualization The increasing amount of semantically annotated open data, especially in the area of linked govern- ment data, justifies investigating semantic tech- nologies for visualization related to policy making. Semantic-technology research focuses on data’s machine readability, whereas semantics visualiza- tion focuses on human-centered approaches for conveying information. Semantics visualization goes beyond ontology visualization, which focuses on visualizing a formal knowledge description in

a certain domain. It provides a comprehensible, interactive view of semantics.4

In visualization, semantics is the meaning- ful relation between two or more data entities. These relations can be described explicitly by for- mal semantic languages (for example, OWL—Web Ontology Language) or gathered implicitly with semantic-mining methods. With the ability to correlate linked government data to data from do- mains unrelated to politics, analysts can find new relations and visualize them for decision-makers.

Semantics visualization is applicable to all three stages of our simplified policy-making model. In- formation foraging can employ the search, explora- tion, and decision-support methods we’ve described. Policy design can employ semantically structured policy formalisms and visual authoring environ- ments. Impact analysis can employ logical infer- ences, predicate logic, and fuzzy cognitive maps to provide a comprehensible comparison of scenarios.

Visual Analytics For complex analysis tasks—for example, during impact analysis—political analysts might have to incorporate complex algorithms and deal with vast amounts of data. Political analysts can’t be ex- perts in every computational discipline that might contribute to the analysis. So, interactive visual displays could help them access the complex com- putational models.

VA exactly addresses this problem by combining computers’ data-processing capabilities with the strength of humans’ visual perception. On the one hand, computers process vast amounts of data for aggregation, structuring, or summarization. This provides users with intuitive visual access to the data. On the other hand, users can visually detect interesting patterns in visualizations and can con- trol computers to get a more precise analysis. So, VA can help political analysts incorporate complex ICT into policy making.

KDD VA strongly involves KDD techniques because it employs interactive visual displays to control and use automatic data analysis tools. KDD uses vi- sualization techniques only in restricted ways—for example, in GUIs to set automatic analysis tech- niques’ parameters.

Researchers are already applying KDD tech- niques to policy making, especially during infor- mation foraging and impact analysis. For example, a large KDD community focuses on automatically extracting public opinions from the Web. A VA ap- proach would couple visualization techniques with

IEEE Computer Graphics and Applications 87

such KDD techniques to directly involve policy makers in opinion analysis.

Two Use Cases for Policy Modeling Two European Commission-funded projects that focus on integrating visualization into policy mod- eling are ePolicy and Fupol.

ePolicy The ePolicy (Engineering the Policy-Making Life Cycle; www.epolicy-project.eu) project aims to provide a decision-support system for policy mak- ers. To do this, the project is engineering a policy- making life cycle for regional energy planning. The life cycle will bring to policy makers’ attention both global concerns (for example, regional en- ergy incentives’ impacts, budget constraints, and objectives) and individual concerns (for example, opinions and reactions), guiding them toward bet- ter policy implementation.

Technically, ePolicy integrates these perspectives through global-level optimization, individual-level social simulation, game theory for managing con- flicts and regulating the interaction between these two levels, and opinion mining (see Figure 3). The project’s goals are to uniquely combine these re- search areas and provide intuitive visual-interactive access to the underlying techniques.

The project integrates information visualiza-

tion and VA into policy making, especially during policy design and impact analysis. It uses infor- mation visualization mainly to visualize the vast amounts of data generated by automatic analysis tools such as opinion mining or social simulation. It closely connects VA tools to these analysis tools to set parameters and interactively control them for advanced impact analysis.

This project uses visualization techniques partic- ularly to make complex analysis processes accessible for political analysts supporting decision-makers who are setting political agendas.

Fupol Fupol (Future Policy Modeling; www.fupol.eu), a four-year project that started in late 2011, will integrate multichannel social computing, crowd- sourcing, and semantics visualization into political decision-making. It will consolidate heterogeneous technologies into a core system, which will auto- matically collect, analyze, interpret, and visualize opinions expressed on the Internet. This will en- able governments to gain a better understanding of citizens’ needs. A new governance model will support policy design and implementation. The approach is based on complexity science; it aims to reduce complexity through a spiral life cycle for policy design that’s appropriate for complex soci- etal problems.

Visualization

Global-level optimization

Policy-making life cycle

Policy scenarios

E-participation

Postevent opinion mining

Pre-event opinion mining

Game theory interaction

Equilibrium point

Implementation strategies

Individual-level simulation

Figure 3. The policy-making life cycle in the ePolicy project, which aims to provide a decision support system for policy makers.

88 September/October 2012

Graphically Speaking

The policy design cycle will follow the three stages in Figure 2. Information foraging will apply seman- tics and information visualization to support new policy requirements and changes. Furthermore, Fu- pol will use opinion mining and analysis to gather information on information transparency and vi- sualize it for politicians and citizens. This stage will consider using data from the linked-open- data initiative and other social and semantics data sources for visualization. Usage behavior analysis will be important in this stage. The visualization methods (see Figure 4) will be able to automati- cally adapt the visual structure, visual complexity, and data to be visualized to stakeholders’ abilities and interests.5

Policy design will employ process-driven visualiza- tions. Users’ observed usage patterns will be analyzed to dynamically adapt functionalities and user guid- ance in a policy creation workflow that’s strongly supported by visualizations and simulations.

In impact analysis, semantics visualizations will employ predicate and fuzzy logic to visualize a policy’s impact.

The project is at an early stage that focuses on requirements analysis with the participating cities. Its outcomes will include the governance model,

a policy knowledge database, an ICT framework based on cloud computing, and pilot applications for several European cities and one Chinese city.

The European Commission is investing heav-ily in policy-modeling research through the ICT for Governance and Policy Modeling initia- tive. The Crossroad (http://crossroad.epu.ntua.gr) project has set the tone for various projects in this area, such as ePolicy and Fupol. For the visualiza- tion and VA aspects, Crossroad has relied on the European VA research roadmap that the VisMaster coordination project presented in 2010.6 To create this roadmap, VisMaster pulled together European VA experts; the project also had links to a roadmap written in the US and Canada.

The EU projects on governance and policy mod- eling have just started; it will be interesting to see how far they can introduce visual elements to policy modeling. For example, the European FET (Future & Emerging Technologies) Flagship Initia- tives (http://cordis.europa.eu/fp7/ict/programme/ fet/flagship/home_en.html) include the FuturICT pilot project, which brings together ICT research, complex-systems research, and the social sciences.

Figure 4. Semantics visualization for political information analysis. The figure illustrates the SemaVis visualization framework4 in the information-foraging phase of policy modeling. Using semantics enables the visual abstraction of information in categories, the visualization of relations and dependencies of information entities, and geographic and time-related correlations. (Source: Fraunhofer IGD; used with permission.)

IEEE Computer Graphics and Applications 89

One goal is to give governments and citizens much more capable tools for policy modeling and policy analysis, including a strong visual component.

Similar projects are under way elsewhere, es- pecially in Canada and the US. One example is Vaccine (Visual Analytics for Command, Control, and Interoperability Environments; www.purdue. edu/discoverypark/vaccine), a US Department of Homeland Security Center of Excellence that’s using VA to enhance policy modeling for public safety and public health.

All these initiatives point in one common direc- tion: visualization and VA are vital for informed decision-making and policy modeling in a highly complex information environment overloaded with data and information. We expect that policy mod- eling will be a common application domain in upcoming US and European visualization confer- ences. In the end, it might determine not only our policy makers’ efficiency and proficiency but also citizens’ involvement and confidence in future policy modeling.

References 1. Crossroad—a Participative Roadmap for ICT Research

in Electronic Governance and Policy Modeling, tech. re- port, 2010; http://crossroad.epu.ntua.gr/files/2010/ 02/CROSSROAD_D4.3_Final_Roadmap_Report-v1. 00.pdf.

2. A. Macintosh, “Characterizing E-participation in Policy-Making,” Proc. 37th Ann. Hawaii Int’l Conf.

System Sciences (HICSS 04), IEEE CS, 2004. 3. D.A. Keim et al., “Event Summary of the Workshop

on Visual Analytics,” Computers and Graphics, vol. 30, no. 2, 2006, pp. 284–286.

4. K. Nazemi, C. Stab, and A. Kuijper, “A Reference Model for Adaptive Visualization Systems,” Human- Computer Interaction: Design and Development Ap- proaches, vol. 1, J.A. Jacko, ed., LNCS 6761, Springer, 2011, pp. 480–489.

5. P. Sonntagbauer, “Fupol at a Glance,” Dec. 2011; www.fupol.de/?q=node/6.

6. D.A. Keim et al., Mastering the Information Age: Solving Problems with Visual Analytics, Eurographics Assoc., 2010.

Jörn Kohlhammer is the head of the information visualiza- tion and visual analytics department at Fraunhofer IGD. Contact him at [email protected].

Kawa Nazemi leads the Semantics Visualization group at Fraunhofer IGD. Contact him at kawa.nazemi@igd. fraunhofer.de.

Tobias Ruppert is a researcher in the information visual- ization and visual analytics department at Fraunhofer IGD. Contact him at [email protected].

Dirk Burkhardt is a researcher at Fraunhofer IGD. Con- tact him at [email protected].

Contact department editor Miguel Encarnação at lme@ computer.org.

ADVERTISER PAgE Apple 49 Visweek 2012 Cover 4 Advertising Personnel Marian Anderson: Sr. Advertising Coordinator Email: [email protected] Phone: +1 714 816 2139 | Fax: +1 714 821 4010

Sandy Brown: Sr. Business Development Mgr. Email: [email protected] Phone: +1 714 816 2144 | Fax: +1 714 821 4010

Advertising Sales Representatives (display) Central, Northwest, Far East: Eric Kincaid Email: [email protected] Phone: +1 214 673 3742; Fax: +1 888 886 8599

Northeast, Midwest, Europe, Middle East: Ann & David Schissler Email: [email protected], [email protected] Phone: +1 508 394 4026; Fax: +1 508 394 1707

Southwest, California: Mike Hughes Email: [email protected] Phone: +1 805 529 6790

Southeast: Heather Buonadies Email: [email protected] Phone: +1 973 585 7070; Fax: +1 973 585 7071

Advertising Sales Representative (Classified Line, Jobs Board) Heather Buonadies Email: [email protected] Phone: +1 973 585 7070; Fax: +1 973 585 7071

AdvertiSer/product index • September/october 2012