Investigating Visualization Ontologies Gao Shu1 Nick J.Avis2 Omer F. Rana2 1 School of Computer Science, Wuhan University of Technology, Hubei 430063, China 2 School of Computer Science, Cardiff University, U.K Abstract The advent of the Grid Computing paradigm and the link to Web Services provides fresh challenges and opportunities for collaborative visualization. Ontologies are a central building block of the Semantic Web. Ontologies define domain concepts and the relationships between them, and thus provide a domain language that is meaningful to both humans and machines. In this paper, the analysis and design of a prototype ontology for visualization are discussed, whose purpose is to provide the semantics for the discovery of visualization services, support collaborative work, curation, and underpinning visualization research and education. Relevant taxonomies are also reviewed and a formal framework is developed. 1.Introduction systems Visualization is a powerful tool for analyzing developing Grid enabled visualization services data and presenting results across a wide range has focused on resource discovery and remote of disciplines. Grid computing offers many new rendering. The Resource Aware Visualization opportunity for Environment associated with visualization the [19],[20], present experience of (RAVE) allows adaptation and depending on resource availability and load to orchestration of the required physical resources. deliver visualization results to end users in a Perhaps the greatest potential of the Grid with timely and transparent manner [20]. However, respect to visualization is its capability to with this experience, we recognize it is support teams imperative to establish common vocabularies collaborative visualization and capture and organise visualization domain geographically participating in marshalling [20]. Our separated efforts [1],[20]. This involves the use of knowledge resources needed to produce visualizations, both machine-to-machine negotiations necessary for computing (software system, HPC, networking) next generation Grid enabled visualization and humans (visualization specialization and end systems. An ontology defines domain concepts user domain experts and users) [2]. The and the relationships between them, and thus capability to interact with resources that are provides a domain language that is meaningful geographically distributed, and allow the via to both humans and machines. An ontology may also be used to deliver significantly improved services, remote hosting environments, etc), (semantic) search and browsing, integration of allows remote services to be accessed and used heterogeneous on demand. Such a computing model assumes improved analytics and knowledge discovery that a user is interested in making use of services capabilities. In this paper we describe the that are not owned locally and/or cannot be development of an ontology for visualization. provisioned at their own local site. The viability This ontology is designed to of such a model has already been demonstrated common vocabulary for: describing visualization through the use of numerical services [3] and data; processes; products; and support the some description and discovery of Web Services; Grid-enabled enabled inform Grid/eScience infrastructure (such as registry emerging as to visualization information sources and provide a sharing of process models between visualization Ontology” held in April 2004, which identified developers and users; curation and provenance the management of visualization processes and data; visualization, and investigated the structure for and collaboration and interaction between such an ontology, giving an overall description distributed sites offering visualization services of what such an ontology should contain [5]. [2]. At the same time, Semantic Web Service However, there was agreement that these initial technologies, such as the Ontology Web efforts were both tentative and incomplete Language (OWL), are developing the means by because the creation of an ontology is an which services can be given richer semantic iterative activity and the establishment of the specifications. Richer semantics can enable ontology for visualization needs consensus more flexible automation of service provision within the visualization community itself. and use, and support the construction of more Brodlie at the University of Leeds proposed the powerful tools and methodologies. This also notion of a “scientific visualization taxonomy” makes it possible for users to define their using the concept of an “E-notation” [6][7], requirements and subsequently connect to which was first developed at the AGOCG services that may be adopted in their work. visualization meeting in 1992. This is seen as a Ontologies are a central building block of the useful classification and model of the underlying Semantic Web: they provide formal models of sampled domain knowledge that can be exploited by subsequently visualizing this data, however it intelligent agents. A visualization ontology failed to capture details about how such samples would provide a domain language that is were distributed, or how the visualization was meaningful to humans and machines, describes carried out – both fundamental issues in the configuration of a visualization system and visualization. supports the discovery of available Web Melanie Tory and Torsten Möller in Simon Services. Fraser We show how a prototype ontology for significant work on “visualization taxonomies” visualization may be developed using Protégé. It [8]. Recently, they presented a novel high-level is extendable, and as such, our initial efforts can visualization taxonomy. The taxonomy classifies be viewed as a starting point and catalyst for visualization experts in visualization to create and agree a Algorithms are categorized based on the standard ontology. In Section 2, a summary of assumptions they make about the data being related work is provided. Section 3 introduces visualized. As the taxonomy is based on design the Ontology Web Language and Section 4 models, it is more flexible and considers the describes a popular ontology editing tool -- user’s conceptual model, emphasizing the human Protégé. The development of ontology for aspect of visualization [9]. visualization using Protégé will be discussed in Ed H. Chi at Xerox Parc put forward a new way the Section 5. Conclusions and future work are to presented in Section 6. techniques by using the Data State Model need to establish data which University an may have algorithms taxonomize ontology be also rather information for used for undertaken than data. visualization approach [10]. This research shows that the Data 2.Related Work State Two workshops have recently been held at the understand the design space of visualization UK’s National e-Science Centre (NeSC), one on algorithms, but “Visualization for eScience”, held in January understand how 2003 [4] and the other on “Visualization techniques can be applied more broadly [11]. Model not only also helps researchers helps implementers information visualization Additionally, in an article describing the design inconsistencies in an ontology that conforms to space of information visualization techniques, OWL-DL. Card and Mackinlay constructed a data-oriented in [13]. This taxonomy divides the field of 4.Building OWL Ontology with Protégé visualization subcategories: An OWL ontology can be regarded as a network Scientific Visualization, GIS, Multi-dimensional of classes, properties, and individuals. Classes Plots, Multi-dimensional Tables, Information define names of the relevant domain concepts Landscapes and Spaces, Node and Link, Trees, and their logical characteristics. Properties and Text Transforms. OLIVE is a taxonomy (sometimes also called slots, attributes or roles) assembled Shneiderman’s define the relationships between classes, and information visualization class [14], and divides allow the assignment of primitive values to information visualization techniques into eight instances. Individuals are instances of the classes visual data types: temporal, 1D, 2D, 3D, multi-D, with specific values for the properties. Our Tree, Network, and Workspace. visualization ontology (called here “VO” for taxonomy [12], which is subsequently expanded into by several students in short) is developed using Protégé_3.1_beta with 3.Ontology Web Language (OWL) OWL-plugin. Protégé is an open platform for The OWL language is designed for use by ontology modeling and knowledge acquisition. applications that need to process the content of The OWL Plugin [18] can be used with Protégé, information, presenting and enables a user to load and save OWL files in information to humans. OWL facilitates greater various formats, to edit OWL ontologies with machine interpretation of Web content than that custom-tailored supported by XML, RDF, and RDF Schema provide access to reasoning based on description (RDF-S) by providing additional vocabulary logic. The OWL Plugin user interface provides along with a formal semantics. OWL can be various default tabs for editing OWL classes, used to explicitly represent the meaning of terms properties, forms, individuals, and ontology in vocabularies and define the relationships metadata. between those terms. OWL is part of the As an extension of Protégé, the OWL Plugin has growing stack of W3C recommendations related a large and active user community, a library of to the Semantic Web. Compared with XML, reusable components, and a flexible architecture. XML-Schema, RDF, and RDF-Schema, OWL The OWL Plugin therefore has the potential to adds describing become a standard infrastructure for building properties and classes: among others, relations ontology-based Semantic Web applications [16]. instead additional of vocabulary just for graphical widgets, and to between classes (e.g. disjointedness), cardinality (e.g. “exactly one”), equality, richer typing of 5.Ontology for visualization properties, characteristics of properties (e.g. symmetry), and enumerated classes. OWL has 5.1 The overview of VO three sub-languages: A taxonomy is a good mental starting point for OWL Lite, OWL DL, and OWL Full [15]. Our building an ontology. Unfortunately, there has ontology for visualization is represented in OWL not been a universally, well-accepted taxonomy DL-- because OWL DL is much more expressive for visualization developed as yet. In our opinion, than OWL-Lite and is based on Description the ontology for visualization must be a Logics, it is therefore possible to automatically compromise between function, understandability compute a classification hierarchy and check for and uniformity. As a starting point therefore, we increasingly-expressive synthesize some existing taxonomies, mainly concepts and their relations as possible to based on Ken Brodlie [6] and Melanie Tory, provide machine-readable formal specifications Torsten Möller’s [8] taxonomies, and present for the discovery of visualization services. these in an organized structure which highlights Moreover, some of decisions made in the Data_Model chosen for /by has-component transform from/by/into Primitive_Set has-component has-component Data_ Representation Visualization_ Techniques used for/by Figure 1 Relationship between the four main classes in the VO the connection visualization between techniques data and models, process of building VO were not rigorously data justified but rather based on the authors’ the representation. Taking the data model for intuition example, firstly, it is categorized into two types: limitations of the tools currently available. discrete model and continuous model, and then The VO consists of classes, properties and the continuous model is classified according to individuals. Classes are interpreted as sets that the type of each variable: scalar, vector, tensor, contain individuals. They are described using point or multivariate, and each of which is formal descriptions that state precisely the broken the requirements for membership of the class. dimensionality and number of variables, and is Properties illustrate relationships between two denoted by the E-notation [6]. The factor of time individuals. And individuals represent objects in is neglected here, and the relationship between the domain that we are interested in. the corresponding The VO has four abstract classes representing visualization techniques should be considered the main concepts in the visualization domain: mainly in the process of category. The discrete Data_Model, model is classified into either a connected or Data_Representation, and Primitive_Set. Their unconnected model, and the unconnected model relationship is shown in figure 1. is to Data_Model describes the user’s data model, dimensionality of the data involved, each of Visualization_Techniques defines a variety of which is also denoted by the E-notation. A more techniques and algorithms used to transform or detail description is presented in Section 5.3. visualize the data model, Data_Representation Our VO is a web-accessible and was envisioned contains multimodal attributes, which enables a to be a prototype ontology used to enable user to chose an appropriate representation, automatic whereas Primitive_Set includes the elements down data broken further model down and according the further composition to according of Web-based visualization services. So we do not attempt to capture all knowledge about the visualization domain, rather our main goal is to cover as many and sometimes directed by the Visualization_Techniques, used as the building block of above classes. 5.2 Primitive Set class 5.3 Data_Model class The subclasses of Primitive_Set class are the The hierarchies within the Data_Model class, basic concepts which might be linked through which are shown in figure 3, are basically the has- property with corresponding classes. In a same as our taxonomy for visualization. sense they serve as a set of basic concepts upon However, we can go further with an ontology. which the ontology is built. Figure 2 shows the portion of Primitive_Set as corresponding to blocks for the Data_Model. Figure 2. The contents of the class Primitive_Set To create any ontology, it is necessary to provide concise definitions of the base concepts involved. Figure 3. The content of Data_Model It is often the case, however, that some of these Where: EnSm denotes a (n) scalar entity on a concepts come from different fields and it is m-dimensional domain. better to have them formalized by experts in EnVm denotes an entity has (n-)vector data on a these fields. VO may need some fundamental m-dimensional domain. mathematical concepts to be included. One way En_nTm denotes an entity has a (n;n) tensor to do this is to link the ontology to another data on a m-dimensional domain(where n=3). already developed ontology. For example, the EPm denotes a m-dimensional domain of Mathematics on the Net [MONET 2004][17] point-data (where n=1,2). project is developing OWL-based mathematical EMulti_var ontologies using the OpenMath and MathML multi-dependent variables. initiatives. However, at the moment, since they Em are still being refined, we hope to investigate the unconnected data on m-dimensional domain. denotes denotes an entity an entity has has discrete and option of using them at a later time, so VO defines some simple mathematical concepts on We can put more complex and meaningful its own. But the approach seems to be most machine understandable knowledge into the promising and is likely to be used in the future VO by using stronger semantic connections in when ontologies from base knowledge domains the OWL-based ontology: OWL is used to relate are finalized. classes through relationships and simple superclass/subclass properties, and also distinguishes between necessary (inherited) visualized by algorithms included in classes conditions (superclasses) and necessary and AnVm and A3V3, which relates this data model sufficient to the corresponding algorithm. Meanwhile, it conditions (equivalent classes). ∀ has-DatatypeSet Furthermore, an OWL-based description allows inherits restriction condition users to restrict classes or related classes with Vector and ∀ has-StatespaceSet Continuous from its superclass EV, which is shown in figure description logic ∀, ∃, ¬,∩,∪, where: symbols, such as 4. Through these restrictions on properties and ∀ means allValuesFrom, logic statements, our VO can provide Semantic ∃ someValuesFrom, Web agents with background knowledge about ¬ complementOf, domain concepts and their relationships. This ∩ intersectionOf, knowledge can be exploited in various ways, for ∪ unionOf. example At the same time, one of the key features of an functions, i.e. agents can use their ontological ontology that are described using OWL-DL is knowledge to match a user data model with the that they can be processed by a reasoner, such as available visualization services offering the RACER in Protégé. This allows checks for corresponding algorithms [21]. Taking EnS3 as inconsistencies, an hidden dependencies, to drive context-sensitive search example, if the user’s data model is EnS3, redundancies, and misclassification. Taking the matchmaking agent can search the services E3V3 as an example, we define its necessary & which offer the algorithms, included in the class sufficient condition as EV, has-Dimension = 3, AnS3 has-Vector = 3, which mean that E3V3 is the Marching_Cube or Contour_Connecting and so subclass of EV, the number of its vector on, to visualize the data model. All of these can argument equals 3 and dimension is 3. We can be done by means of EnS3’s necessary condition also define that its necessary condition is ∀ is-transformed-by AnS3. ∀ is-transformed-by (AnVm 5.4 Visualization_Techniques class The subclasses of Visualization_Techniques ∪ A3V3), where n , m=2,3, which means it can be transformed or shown in figure Figure 4 The definition of the class E3V3 in Protégé OWL plugin 5, such as define a set of techniques and algorithms which 5.5 Data_ Representation class can be used to transform or create visual The reason that Data_Representation class representations of data using a data model. So should be included is that some models of techniques/algorithms are categorized based on representation are specific to classes of data, for the data model rather than on the data itself, example work in flow visualization and graph which visualization facilitates the transformation or have distinct categories of visualization of the data model. The relationship representations. So users should be allowed to between techniques/algorithms and data model is make the appropriate representation choice achieved by three properties: is-transformed-by, based on their information need. Its subclasses transform-from and transform-into-model. The contain multimodal attributes, i.e. visual, haptic, property transform-from is an inverse property sonic, olfactory, physical attribute etc. The visual of is-transformed-by. The former means an attribute is further split into four subclasses algorithm spatialization, timing, color and transparency can be used to visualize the corresponding data model, the latter the inverse. [7]. The range of the property transform-into-model is Visualization_Techniques, which means we 6.Conclusion and future work can use an algorithm to transform a data model In this paper, we have described our experience into a simpler form. It can therefore be seen that of building an ontology for visualization, it is possible to match data for specific although it is accepted that this is both applications to the most appropriate visualization incomplete and tentative. Our VO is designed as techniques/algorithms by means of the above an OWL-based prototype ontology whose three properties. A fragment of subclass of purpose is to provide a vocabulary by which Visualization_Techniques is shown in figure 5 users (for example, AS2 denotes that its subclasses communicate, and can be used to transform or visualize the data matchmaking portal model ES2). visualization services [21]. Several specific and visualization it systems is being for used discovery can in of directions for future work include: z To populate the VO in terms of Operators, Techniques, Algorithms and (low-level) Transformation with respect to the Data Model. z To determine the limits of the expressiveness of the VO. z To extend the above VO to include a user model and actions. z To extend the above VO to include temporal aspects/conditions. z To determine if the VO can be used to characterize and describe lower levels of details such as NAG-EXPLORER meshes Figure 5. The content of Visualization_Techniques using the same constructs. z To create a system which uses the VO to automatically build ALL permissible visualizations given knowledge of the data. Techniques using the Data State Reference Model. In Proceedings of the Symposium on Information References Visualization (InfoVis '00), pp. 69--75. IEEE Press, 1. Brodlie, K. W., Duce, D. A., Gallop, J. R., Walton, 2000. Salt Lake City, Utah. J. P. R. B. & Wood, J. D. Distributed and Collaborative Visualization.Computer 12. S. K. Card, J. D. Mackinlay. The Structure of the Information Graphics Design Space. Proceedings of IEEE Symposium on Information Forum 23 (2), 223-251,2004 2. D. J. Duke, Visualization Visualization (InfoVis ’97), Phoenix, Arizona, K.W. Brodlie and D. A. Duce, 92-99 Color Plate 125, 1997. Building an Ontology of Visualization 3. Simone Ludwig, O. F. Rana, J. A. Padget and W. 13. S. K. Card, J. D. Mackinlay and B. Shneiderman. for Information Visualization: Using Vision to Think. Mathematical Web Services, Journal of Grid Morgan-Kaufmann, San Francisco, California, Computing, 2006. 1999. Naylor, Matchmaking framework 4. National e-Science Centre. Visualization ontologies. http://www.nesc.ac.uk/talks/393/vis 14. OLIVE: On-line Library of Visualization ontology report.pdf. Information Environments. http://otal.umd.edu/Olive/, 1999. 5. National e-Science Centre. Visualization for 15. OWL Web Ontology Language Overview : e-science. http://www.w3.org/TR/2004/REC-owl-features-20 http://www.nesc.ac.uk/esi/events/130/workshop 040210/ report.pdf. 16. Holger Knublauch, Ray W. Fergerson, Natalya F. Noy, Mark A. Musen. The Protégé OWL Plugin: 6. K.W. Brodlie, 1992.Visualization Techniques, in and An Open Development Environment for Semantic Applications, edited by K.W. Brodlie, L.A. Web Applications. Third International Semantic Carpenter, Web Conference - ISWC 2004 Scientific Visualization R.A. - Techniques Earnshaw, J.R. Gallop, R.J.Hubbold, A.M. Mumford, C.D. Osland and P. 17. MONET Consortium. MONET Home Page, www. Quarendon, Chapter 3, pp 37-86, Springer-Verlag. 7.K.W. Brodlie, 1993. A Classification Scheme for Scientific Visualization, in Animation Available from http://monet.nag.co.uk. 18 Holger Knublauch, Mark A. Musen, and Alan L. and Rector.Editing description logics ontologies with Scientific Visualization, edited by R. A. Earnshaw the Prot´eg´e OWL plugin. In International and D. Watson, pp 125-140, Academic Press. Workshop on Description Logics, Whistler, BC, Canada, 2004. 8. Melanie Tory and Torsten Moller. A model-based visualization taxonomy. SFU-CMPT-TR2002-06, Technical Computing Report Science Dept., Simon Fraser University, 2002. 19 I J Grimstead, D W Walker and N J Avis. Resource Aware Visualization Environment – RAVE, accepted for publication in Concurrency 9. Melanie Tory and Torsten Möller. Rethinking Visualization: A High-Level Taxonomy, IEEE and Computation: Practice and Experience. 20 Ian J. Grimstead, David W. Walker, Symposium on Information Visualization, pp. Nick J. Avis. Collaborative Visualization: A 151-158, Oct. 2004. Review 10. Ed H. Chi and J. T. Riedl. An Operator Interaction and International Taxonomy. Symposium In on Ninth IEEE Distributed Framework for Visualization Systems. Symposium Simulation and Real-Time Applications, 2005, pp. on 61-69. Information Visualization (InfoVis '98), Research Triangle Park, North Carolina: 63-70, 1998. 11. Ed H. Chi. A Taxonomy of Visualization 21 Gao Shu Omer F. Rana, Nick J Avis and Chen Dingfang. Ontology_based Semantic Matchmaking Approach. Accepted of publication.