Visualization of Web-Based Workspace Structures Robert P. Biuk-Aghai Collaborative Systems Laboratory Faculty of Information Technology University of Technology, Sydney P.O. Box 123, Broadway, NSW 2007, Australia robertb@it.uts.edu.au Abstract Web-based collaboration is becoming increasingly common. Often such collaboration exhibits emergent characteristics, for which systems based on a notion of workspaces provide suitable support. Virtual teams collaborating in this manner create virtual organizational structures, the capture and retention of which into organizational memory promises to be of benefit, as it can facilitate reuse. Visualization of workspace structures plays an important role in the effort to identify, and then capture, these structures. The visualization presented here represents workspaces as nodes in a two-dimensional graph, with edges representing inter-workspace relationships. These graphs are visualized using an animated force-directed algorithm. Colour is employed in various ways to reveal additional dimensions of workspace properties. The visualization of intra-workspace relationships is also supported. An example illustrates how the visualization can aid in the identification of collaborative structures using the presented visualization tools. 1. Introduction In recent years, the World Wide Web become the medium through which a host of activities are being carried out. A profusion of systems have been developed to support various activities, including communication, information distribution, workflow, collaborative software development, and many others. At the same time, organizations have begun to take advantage of the opportunities offered by Web-based systems in enabling collaboration by distributed work groups. Virtual teams, even virtual organizations, are becoming increasingly commonplace, and it been claimed that the organization of the 21st century On leave from the Faculty of Science and Technology, University of Macau, Macau S.A.R., China. will be made up of virtual teams and of networks of such teams [7]. In many cases, the work carried out by such teams is of an emergent nature: structure and goal of the work emerge as it is carried out. As we have reported elsewhere [1], conventional workflow management systems, whether Webbased or not, are not suitable for the support of this type of work, as they require work processes to be predefined and encoded in a workflow engine before they can be enacted. Emergent work, being essentially improvisational in nature [8], requires a much greater degree of flexibility, where planning and implementation of work converge. We believe that this category of work processes is better supported by systems which are based on the notion of workspaces. In this context, a workspace is seen as a virtual place bringing together people, artefacts, actions, and communication channels, and associating them with each other through certain defined relationships. We will elaborate on this notion of workspaces later on. Our research group developed a model of collaboration that is based on this notion of workspaces, and implemented this model in the prototype system LiveNet. This system now been operational for over a year, and experiences of its use are now feeding back into its further development. Teams collaborating over the Web using systems such as LiveNet create virtual organizational structures. As with conventional organizational structures, such as group, department or task force, such structures relate organizational resources with each other in certain ways. Part of this structure is related to the composition of the team itself, but other parts are related to the things the team deals with, and the things the team does. Together they make up the structure of the whole virtual collaboration for a given activity. We believe that it is worthwhile to study these structures, as they constitute a part of the organization s memory. Organizational memory been investigated for a number of years [3, 9]. However, most research been directed to-
ward the content, or the what, of collaboration and only little focused on the structure, or the how, of that collaboration. In the literature, these two types of organizational memory have been referred to as declarative and procedural memory, respectively [8]. To be able to recall how virtual collaboration was carried out in the past can be of great value for an organization, especially when it needs to carry out similar collaboration in the future. For instance, when a certain kind of activity was carried out for the first time, it may have been done in an improvisational, emergent fashion. However, subsequently it may be found that the same or a very similar kind of activity needs to be performed again. In this case, time could be saved by reusing the structure which the first performance of the activity arrived at in the end, rather than having to improvise it every time. The investigation of virtual collaboration structures may also reveal generic patterns which can be captured and assembled into a library for later reuse. These include features of the structure s trajectory, i.e. its past and future evolution. If the same kind of structure is found to evolve in similar ways, the evolution can be aided by suitable automated support, such as an agent-based system, drawing upon knowledge of these trajectories. We see the discovery of such structures and patterns as a three-step process consisting of (1) visualization, (2) analysis, and (3) retention. In the first step, users explore the data to gain an initial understanding of the kinds of relationships and correlations, regularities and anomalies that can be found. This then informs the second step, where automated tools perform searches over the database of workspace structures based on user-supplied inputs. Finally, the third step consists of retaining structures obtained through step two in a separate database for later reuse. This is similar in kind to the knowledge discovery process in large databases, where visualization plays an important part throughout [2]. In the present paper we shall focus exclusively on the first step of this process, i.e. the visualization of the workspace data. The individual s view in collaboration is always only a partial view of the whole collaborative setting, focusing on the particular part of work in which that individual is involved. The use of information visualization is therefore essential in order to perceive the collaboration structures in their entirety and to thereby enable the discovery of collaboration patterns. We have developed tools for the visualization of the data contained in the LiveNet workspace database which facilitate this discovery. Using these tools, characteristics of different collaboration setups can be identified, and common features of collaboration structures can be detected. The following section will provide an overview of the collaborative model used in LiveNet, which underlies our visualization. Section 3 will then briefly introduce the LiveNet system. In Section 4, the visualization of LiveNet workspace structures will be discussed, and Section 5 will present our workspace visualizer tool. Finally, Section 6 will offer some conclusions. 2. Workspace Model Workspaces provide the locus of collaboration. Whereas physical collaboration takes place in physical space, virtual workspaces provide the substitute for virtual collaboration. They unite objects involved in a collaborative activity. These include people, artefacts, actions, and communication channels. These workspace objects are related with one another according to an underlying model of collaboration. In this model, collaboration is seen as being carried out by actors (or participants) assuming organizational roles. Roles operate within a given organizational context that allows them the performance of certain actions and that affords them the opportunity to communicate with other roles operating in the same or other organizational contexts. Actions commonly involve artefacts, which are created, viewed, or modified by roles. Relationships of roles with other objects are subject to governance rules, which define the level, or scope, of the relationship. For instance, a governance rule may state that a given role may assign new actors to join into a collaboration, or that a role may view but not modify an artefact. Actions in a workspace are of two types: solo actions which are performed by a single action performer (a single role and participant); and interactions which involve multiple roles and/or participants. Interactions may be actionoriented such as joint editing tasks, or communicationoriented such as discussions. Artefacts involved in actions can be any objects and are usually documents of various types such as text, drawings, graphic images, etc. Here we distinguish between two types of artefacts: those which are the direct inputs or outcomes of an activity, and those which are used as background material. Company policies, norms, standards and the like are all examples of background material. Their content is assumed knowledge for those participating in the activity, required for effective participation in the work carried out. Because of its improvisational nature, emergent group work tends to be more communication-intensive than production processes. Therefore, in addition to the above workspace features, communication facilities are essential in workspaces. Informal communication can be handled using actions, as described above, providing for example electronic mail and discussion fora. The latter are supported in workspaces in a dual role: on the one hand they constitute actions, on the other hand they are also artefacts inasmuch as they contain a record of the group s discussions. For more formal communication requirements of the
type-of Message MANAGER ADVISERS POTENTIAL AUTHORS REVIEWS Message type involves Message rule Participant involves involves occupies sent-by Workspace Role performs Solo Action Artefact Action uses, creates Interaction Background Document Discussion Management PROPOSALS CHAPTERS Manuscript planning TENTATIVE PLAN MANUSCRIPT EDITOR Chapter acquisition Reviewing MANUSCRIPT REVIEWERS CHAPTER AUTHOR Figure 1. Meta-model of workspaces and their objects Figure 2. Rich picture of a manuscript preparation process more structured and well-defined aspects of the work activity, special support is required. This is particularly of use for those types of processes which contain some more defined portions within an overall emergent process. Therefore, our model provides facilities for notification using semi-structured messages. These are used to convey asynchronously from one role to another (possibly across workspace boundaries) that a given kind of event occurred. Messages have a given message type. Each workspace can possess a number of message types, each representing a different kind of notification event. Messages are sent according to message rules which represent defined communication channels within the organization and associate pairs of message types and their respective workspaces as well as the receiving role with each other. A meta-model of the workspace objects discussed above and their relationships is shown in Figure 1. Note that for reasons of simplicity, issues of governance have been deliberately omitted from the graph. Workspaces are intended to be highly dynamic structures, continuously evolving along with the work carried out in them. Thus the structure of a workspace at a given point in time represents a snapshot of that instance, and a later snapshot may well differ from a previous one. All workspaces are related with one another in a treelike fashion, with parent and child workspaces. Child workspaces are subsumed to their parent and their use is typically for a sub-activity of the parent workspace, or an activity related to that of the parent. Workspaces are also intended to support a particular organizational goal, with activity in the workspace corresponding to goal-directed activity in the real world. A number of workspaces may possess a common goal, and thus be related with one another. To illustrate the application of this workspace model to a real-life example, consider the case of the preparation of a manuscript with chapters contributed by different authors. While the preparation of chapters is the authors responsibility, the collection of all chapters and the production of the finished manuscript is the responsibility of a publication team. Looked at from a high level, this involves a number of activities, roles, and documents, and can be represented as in Figure 2. The notation used is an adapted form of rich pictures where clouds represent activities (more precisely, interactions), boxes represent artefacts, and the stand-alone text strings represent roles. Arrows linking artefacts and activities signify involvement of the artefacts in the activities: production or modification (if pointing to the artefact), and consumption (if pointing from the artefact). Lines linking roles and activities signify involvement of the roles in the activities. In this example, a set of workspaces would be set up, one for each activity, and would be populated with the roles and artefacts identified. Other workspace objects could also be added, if known at this stage, or may be added once the work is underway. The following section will show how our prototype system LiveNet can be applied to this example. 3. LiveNet Prototype We have developed the prototype system LiveNet which is based on the workspace model introduced above. LiveNet is implemented as a Web-based three-tier system: users interact with the LiveNet client, which communicates through the network with a LiveNet server, which in turn interacts with a relational database management system that maintains the workspace data. Two different clients exist: an HTML-only client implemented using servlets and aimed at end-users; and a Java applet client providing more functionality and aimed at workspace developers. As both LiveNet clients are web-based, distributed collaboration across the Internet is made possible.
4.1. Structural elements Figure 3. Manuscript planning workspace, as seen through the LiveNet Java client An example of the LiveNet client user interface (for the Java applet client) is displayed in Figure 3. It shows a view into the publication management workspace (corresponding to the Management activity in Figure 2), with five windows open showing documents, background material, roles, participants, and discussions. Other workspace objects may exist, however the user chosen only to view information on these types of objects. Thus the view of a workspace is typically only a partial view. Also, because of the way that governance is defined for a given workspace, one role s view of a workspace is usually different from the view of other roles, as each role is typically assigned a different set of workspace objects. Moreover, relationships existing between objects within and also between workspaces can not be easily perceived through the individual user s interface. Thus, in order to effectively visualize the, possibly complex, relationships among workspace objects in a single workspace, as well as the relationships across workspaces, use of the existing LiveNet client is not sufficient. We have therefore developed a dedicated workspace visualizer tool to accomplish this. Before presenting this tool, however, we first introduce elements and requirements of visualization. 4. Workspace Visualization Visualization of workspace structure, both within and between workspaces, aims to aid the discovery of collaborative patterns in these structures. To this end, it is necessary to identify the structural elements to be visualized. These can be grouped into three categories: workspace objects, intra-workspace relationships, and inter-workspace relationships. Workspace objects were already introduced in Section 2 and were shown in a meta-model in Figure 1. They include: roles, participants, documents, backgrounds, actions, discussions, message types, message rules, and messages. Intra-workspace relationships relate workspace objects with one another, according to the governance rules defined within the workspace. The types of relationships that can exist are: Role Participant: Assignment of a participant to a role. Role Document: Assignment of a document to a role. Role Background: Assignment of a background to a role. Role Action: Assignment of an action to a role. Role Discussion: Assignment of a discussion to a role. Role Message: Receipt of a message by a role. Role Message Rule: Definition of a role as message recipient in a message rule. Message Message Rule: Specification of the rule governing a message. Message Message Type: Definition of the type of a message. Inter-workspace relationships relate workspaces with one another, according to their hierarchy, the workspace objects they contain, or their goal. Following relationships can exist between a pair of workspaces: Parent-Child: One workspace is the parent workspace, the other is its child workspace. The kinds of visualizations which can be performed on workspaces can be divided into two main categories: firstly, visualizations of the structure of collaboration and its evolution over time; secondly, visualizations of collaborative behaviour within this structure. The initial focus of our visualization work is concerned with the former, and only to a limited extent with the latter. Shared Participants: The workspaces have one or more participants in common. Shared Documents: The workspaces have one or more documents in common. Shared Backgrounds: The workspaces have one or more backgrounds in common.
Shared Actions: The workspaces have one or more actions in common. Shared Discussions: The workspaces have one or more discussions in common. Related by Message Rule: One workspace is the source and the other the target of a message rule. Common Goal: The workspaces have the same goal. 4.2. Visualization requirements To support the goal of discovering workspace structures and structural patterns, the following requirements exist: To visualize relationships of a possibly large number of workspaces in a single display. To support visualization of a subset of workspaces and relationship types by selectively hiding or showing workspaces and their relationships. To support the identification of clusters of related workspaces. To provide details of workspace internals where required. To provide additional information to enable comparison of workspaces, based on various workspace measures. Connected with the last requirement, we have developed following workspace measures: Workspace density, which measures the number of objects contained in a workspace. These include documents, backgrounds, discussions, actions, message types, and message rules. However, as the number of workspace objects assigned to a particular role in the workspace can be smaller than that of all objects contained in it, a further distinction is necessary; therefore, the absolute number of objects in a workspace is denoted as absolute workspace density, while the number of objects visible to a particular role is denoted as role workspace density. As this may differ among all the roles in the workspace, we distinguish further between the minimum role workspace density, the maximum role workspace density, and the mean role workspace density, i.e. the minimum, maximum, and mean of all the individual role densities, respectively. Evolution intensity measures the extent to which a workspace undergoes evolution. This is measured as the number of new objects per unit of time. Evolution intensity measures the dynamic aspect of structure and indicates the relative stability of a workspace. Evolution recency measures the extent to which a workspace recently undergone evolution. This is measured as the number of new objects over a period of time up until the time of observation. Evolution recency can identify those workspaces which are currently actively evolving. Visualizations of workspace behaviour are possible too, namely related to messages. Two measures deal with this kind of behaviour: Message intensity, which measures the number of messages sent or received since the creation of a workspace. Message recency measures the number of messages which a workspace sent or received over a period of time up until the time of observation, and identifies workspaces which are currently actively communicating. 5. LiveNet Visualizer The representational form chosen for the LiveNet Visualizer tool is that of a graph, which been widely used for representing relationships between objects [4, 5, 6]. In the LiveNet Visualizer, nodes in the graph represent workspaces, and edges represent relationships. As workspaces have parent-child relationships among each other, the basic structure of the graph is that of a tree, while other types of inter-workspace relationships, which may additionally exist, constitute additional edges among tree nodes. To facilitate the dynamic exploration of the graph, not all of which may be visible at the same time, it is visualized employing a force-directed animated visualization algorithm. This algorithm is similar to that of Huang et al. [6], with some modifications. Use of this algorithm also aids in the discovery of workspace clusters, as will be explained shortly. A typical display of workspaces is shown in Figure 4. Only parent-child relationships are visualized in this display, and workspace names are abbreviated where necessary to limit node size. The display of other relationship types can be enabled from a map control panel, which is shown in Figure 5. In addition, the control panel also allows other control actions to be performed on the graph, such as node colouring to represent certain workspace properties, switching between abbreviated and full workspace names in the display, modification of the edge length, enabling and disabling of the graph s animation, etc. It is often a requirement to be able to identify workspaces which have something in common for example some documents, or participants, or their goal. In order to understand how the animated display can help in this regard, it is necessary to understand some basic features of the visualization algorithm. The force-directed animated visualization algorithm is based on a conceptual model of nodes and edges which have two basic properties: nodes behave toward one
Figure 4. Map of all workspaces, showing only parent-child relationships Figure 6. Map of all workspaces, showing parent-child and goal relationships. Four clusters of workspaces with common goals are revealed Figure 5. Workspace map control panel another like the same poles of magnets do: they repel each other; edges behave like mechanical springs: they possess a certain length when at rest, and when compressed or expanded they tend back towards that length. These two properties in combination produce a self-optimizing graph which, to the extent possible within its bounds, will lay itself out reasonably well. However, our visualization algorithm is a modified version of the standard one in which the spring force that applies to edges can be turned on or off for certain types of inter-workspace relationships. When turned off for a certain relationship type, nodes which are connected by edges of that type will float freely to a point where they are far enough away from other nodes so that repulsion between nodes is minimal. However, when the spring force is turned on, nodes will be pulled together until edges are at, or close to, their natural length. This can be useful when trying to discover clusters of related workspaces. For instance, if it is required to discover which sets of workspaces have common goals, the spring force (pull) and visibility for edges of type Goal are enabled in the map control panel. The resulting map, as shown in Figure 6, now reveals four clusters of workspaces with common goals (indicated by differently coloured edges), with 13, 5, 4, and 2 workspaces, respectively. Once a set of workspaces been identified for further investigation, other workspaces in the graph can be hidden by focusing on one node. This will only show the focus node and all those nodes which have at least one visible edge connecting them to the focus node.
Figure 7. Map of workspaces used for manuscript preparation, showing parentchild, document and participant relationships Figure 8. Map of workspaces used for manuscript preparation, with detail map for each workspace. The detail maps show roles, participants, documents, backgrounds, and discussions To return to the example of manuscript preparation introduced earlier, we could find the set of workspaces used for manuscript preparation from a global map, such as the one shown in Figure 4, then focus on the Publication- Management workspace. The resulting map now only contains this workspace as the focus node, together with its parent workspace and its three child workspaces. When the display of all relationship types is turned on, it is revealed that the only types of relationships existing between these workspaces are the parent-child, shared document, and shared participant relationships (see Figure 7). The numbers attached to the participant edges indicate the number of participants in common in the pair of workspaces. In this case, there are six participants in common among three of the workspaces (Manuscript-Planning, Chapter- Acquisition, and Reviewing). This relatively high number of shared participants indicates that the three workspaces are quite likely used as part of the same collaborative activity. Once the exploration of the workspace network reached this stage, where sets of mutually related workspaces have been identified, the analysis of workspace internals can provide additional insight into the collaborative structures. For this purpose, workspace detail maps can be opened for the workspaces under investigation. This is shown in Figure 8. These detail maps reveal relationships among workspace objects, and show the proportions of different types of workspace objects in relation to one another. For instance, an initial analysis of these detail maps reveals that: (1) all of them have one discussion forum which been assigned to all roles; (2) all of them have some documents which have been assigned to all roles; (3) three of them have a role Manuscript-Editor with the same participant (Valerie) assigned to it; (4) the same three workspaces also have a role with many (at least four) participants assigned to it; as well as several other similarities. Together these constitute elements of what may turn out to be a pattern common among other sets of workspaces. The discovery of such pattern elements from a visualization can then prompt focused searches in the database to ascertain whether or not these elements actually constitute patterns. Further information supporting comparison between different workspaces can be shown through node colouring. For example, Figure 9 shows how the display of absolute workspace density node colouring can identify the relatively small set of workspaces in which absolute workspace density is very high (shown in a dark colour). The workspaces involved in manuscript preparation are shown as having an average (low) workspace density, with values ranging from 3 to 5 objects. This can be indicative of a relatively young workspace which as it evolves may grow to have a higher density. Other node colouring is possible too, such as evolution intensity, message intensity, and others (these measures were previously briefly presented in Section 4.2). The use of such measures, expressed graphically, can support the homing in on workspaces which are to be investigated in more detail. It can also be of value
Organizations which need to collaborate across a distance are increasingly making use of Web-based systems to facilitate this collaboration. Systems based on the notion of workspaces are suitable for the support of emergent work processes. These workspaces constitute virtual organizational structures. Therefore, it can be of value for an organization to retain parts of this structure, with the view to reusing them in future collaboration. An important first step towards the capture of such structures is their identification. As these structures can consist of numerous elements, and these elements can be related to each other in numerous ways, such structures can be highly complex. Additionally, each structural element may possess multiple properties, adding to the difficulty of adequately perceiving the structure. Therefore, the visualization of workspace structures is of great importance if useful information is to be identified and captured. This paper presented visualization of workspaces and workspace networks as applied to the LiveNet system. Different features of the LiveNet Visualizer tool were presented and applied to an example of workspaces used in manuscript preparation. We plan to continue our work in this area by developing methods and tools that will allow us to take the next step after visualization, namely the capture of structures identified during visualization, and their storage in appropriate form in libraries of collaboration structures. Furthermore, we intend to mine collaboration patterns from the workspace database, driven by inputs which we will obtain from the visualization of workspaces. The visualization presented is not limited to the LiveNet system. Any system which maintains a semantically rich model similar to the one shown in Figure 1 can be visualized in a similar manner. Acknowledgments The support from the University of Macau and from the University of Technology, Sydney, which made this research possible, is gratefully acknowledged. The author is also thankful to the anonymous reviewer who provided comments on an earlier version of this paper. Figure 9. Map of workspaces with node colouring indicating workspace density when trying to identify anomalies or unusual setups within workspaces. 6. Conclusions References [1] R. P. Biuk-Aghai. Virtual workspaces for web-based emergent processes. In Pacific Asia Conference on Information Systems 2000, Hong Kong, China, June 1 3, 2000. To appear. [2] R. J. Brachman, T. Khabaza, W. Kloesgen, G. Piatetsky- Shapiro, and E. Simoudis. Mining business databases. Communications of the ACM, 39(11):42 48, Nov. 1996. [3] E. J. Conklin. Capturing organizational memory. In R. M. Baecker, editor, Readings in Groupware and Computer- Supported Cooperative Work: Assisting Human-Human Collaboration, pages 561 565. Morgan Kaufmann Publishers, 1993. [4] S. G. Eick and G. J. Wills. Navigating large networks with hierarchies. In S. K. Card, J. D. Mackinlay, and B. Shneiderman, editors, Readings in Information Visualization: Using Vision to Think, pages 207 214. Morgan Kaufmann Publishers, Inc., San Francisco, CA, USA, 1999. [5] T. He. Internet-based front-end to network simulator. In E. Gröller, H. Löffelmann, and W. Ribarsky, editors, Data Visualization 99, pages 247 252, Vienna, Austria, May 1999. Springer-Verlag. [6] M. L. Huang, P. Eades, and J. Wang. On-line animated visualization of huge graphs using a modified spring algorithm. Journal of Visual Languages and Computing, 9(6):623 645, 1998. [7] J. Lipnack and J. Stamps. Virtual teams. Executive Excellence, 16(5):14 15, May 1999. [8] C. Moorman and A. S. Miner. Organizational improvisation and organizational memory. Academy of Management Review, 23(4):698 723, Oct. 1998. [9] J. P. Walsh and G. R. Ungson. Organizational memory. Academy of Management Review, 16(1):57 91, Jan. 1991.