Coping with an increasing number of les is one of the challenges of current desktops. Adding semantic capabilities is one possible solution. Aligned with this proposal, this work introduces the notion of knowledge folder as a coarse set of documents bound together by a common ontology. The ontology plays the role of a clipboard which can be transparently accessed by the le editors to either export (i.e. annotation) or import (i.e. authoring) metadata within the knowledge folder. Traditional desktop operations are now re-interpreted and framed by this ontology: copy&paste becomes annotation&authoring, and folder digging becomes property traversal. However, a desktop setting requires seamless tooling for these ideas to get through. To this end, this work proposes the use of the mouse as the semantic device. Through the mouse, the user can classify, annotate, author, and locate a le as a resource of the underlying ontology. Moreover, being editor-independent, the mouse accounts for portability and maintainability to face the myriad of formats and editors which characterizes current desktops. The semantic mouse is implemented as a plug-in for Windows.
Current desktops should be enhanced with mechanisms that permit users to abstract away from les.This work builds on the notion of knowledge folder, i.e a coarse set of information elements bound together by a common ontology. The folder contains an ontology, the instantiations, and the resources being annotated. A desktop can hold distinct knowledge folders, and a given le can belong to several knowledge folders. In contrast with current folders, this mechanism attempts to abstract away from how les are physically organised, by providing an ontology-based organisation.
As an example of a knowledge folder, consider all the documentation that goes with a research project. This includes the project proposal (e.g. a Word le), bills being payed by the project funds (e.g. Excel les), etc. These les can be scattered around distinct (physical) folders. Even though, they can belong to the same knowledge folder as some of the following clues indicate, event correlation. Creation/removal of the documents are not totally independent (e.g. a bill does not exist without a project proposal), etc.
Current desktops ignore this situation and treat les as isolated units. Today, we copy and paste the text values from one le to another, and the ontology is kept (and managed) in the users mind. And too often, le location turns into digging through a hierarchical folder tree.
This paper presents how this situation can be improved by the introduction of knowledge folders. Speci cally, copying&pasting becomes annotating&authoring, and folder digging becomes ontology traversal. By tapping current le structures into an ontology, authors can both populate the ontology (i.e. annotation), and reuse the instances of the ontology while authoring a document. The ontology instances play the role of a clipboard which can be transparently accessed by the le editors to either export (i.e. annotation) or import (i.e. authoring) metadata within the knowledge folder. As for le location, les are now resources of an ontology. This permits to enhance and contextualize desktop search based on the ontology properties, and navigate along the ontology associations.
Being in a desktop setting, we can not ignore usability. Handling of knowledge folders should be as seamless as possible. Rather than providing separate tools for exporting/importing (i.e. annotation/authoring), we strive to accommodate to the current tools for traditional copy&paste operations: the mouse. This will certainly facilitate user adoption.
To this end, the semantic mouse (seMouse) is introduced. By clicking on its middle
button, seMouse exports/imports properties from the ontology, regardless of the editor
you are working with. It does not matter whether you are working with Word,
PowerPoint, Netscape, etc, the semantic button is available for annotation/authoring. In this
way, the user does not have to move to a new editor when annotating (like in SMORE
[
Both, the support of knowledge folders as the underlying infrastructure, and the use of the mouse as the device to interact with this infrastructure, are the main contributions of this work towards making desktops semantic.
Next section introduces seMouse through ve scenarios, namely, le classi cation, annotation, authoring, semantic navigation and ontology editing. 2
seMouse at work seMouse is an annotation/authoring device that achieves editor-independence by working at the operating-system level: the mouse. This section introduces seMouse with the help of an example.
As a knowledge folder, consider the cluster of heterogeneous documents that goes with a research project. This includes the project proposal (e.g. one Word le), bills payed with the project funding (e.g. twenty Excel les), papers as deliverables of the project (e.g. twenty les in both .pdf and .doc formats), participants (whose desktop counterpart can be either the homepage, an .html resource, or a .pdf resource) and comments (being realized as either emails or .doc resources).
Regardless of their format and folder location, it is likely that a high degree of content reuse as well as frequent contextual navigations within this le space happens. This is what makes this set of les a knowledge unit. Being in a participant -an html le-, you frequently need to locate her project proposals -Word les-, or being in a project proposal, the associated papers -PDF les- are commonly accessed.
A knowledge folder comprises an ontology ( gure 1 shows the one for the sample problem). Five classes are identi ed. Each class is characterized by a set of value-based properties (e.g. title, keyword, abstract). Associations are de ned between these classes (e.g. a project is supervisedBy a participant) (termed ObjectProperty in OWL). And the expressiveness of OWL can be used to de ne inverse and transitivity properties between the associations.
Although the ontology is at the core of the semantic desktop, this paper focuses on authoring and annotating resources of the ontology. We do not address how the inference power of the ontology can achieve its full potential in a desktop setting.
Once the ontology has been set, the population process begins. The key idea is to use the mouse as the semantic device so that interactions with the underlying ontology are achieved via mouse clicks. Speci cally, pressing the middle button on the mouse causes an interaction with the ontology manager. This interaction is context-sensitive, i.e. the button accomplishes distinct operations depending on the place the pointer sits on. Next paragraphs introduce ve scenarios of the use of the semantic mouse.
Scenario 1: le classi cation. First of all, les need to be identi ed as instances of any of the ontology's classes. This is achieved by opening a le, and pressing the middle button. A menu pops up for the user to indicate to which class this le is a resource.
Scenario 2: annotation (see gure 2 and 3). Annotation&authoring becomes the counterpart of copy&paste in traditional desktops, with the difference that now these operations are conducted along the ontology net. What is being exported(i.e. copy) is no longer a string but a class property of the ontology.
If a le has already been categorised, the annotation process may begin. If some text is selected, the mouse is used to export this text as part of the value of a property as it is shown in gure 2. Of course, the set of properties will depend on the class of Fig. 2. Scenario 2: annotation. Some text is selected. Being a deliverable le, the menu displays properties of this class. The text will become the value of the chosen property. the resource. In the example, title, keyword and abstract correspond to properties of the deliverable class.
On the other hand, if no text is selected, the middle button is used to establish associations with other les. This situation is exempli ed in gure 3. In this case, the CORDIS project template for EEC projects has been used. When the middle button is pressed, a menu pops up for the user to link the current resource with other target resources. The menu is customised for the current resource, that is, the associations are restricted to those available for the current resource, whereas the target les are also limited to those of the appropriated class. In the example, the association can only be established with deliverables les since this is the destination class of the delivers association.
Scenario 3: authoring (see gure 4). Associations being set during annotation can now be exploited. For instance, the project resource can import the title of its associated deliverable resources. In the example, the article Authoring and Annotation of Web Pages in CREAM appears as a deliverable of the current le. By selecting this article, the menu is extended rightwise to show up its properties. The user can select one of these properties, and its value is inserted at the cursor place.
Scenario 4: semantic navigation . File location in current desktops frequently implies folder digging. By contrast, semantic navigation strives to exploit the associative behaviour of the human memory. A resource can be located from the resources it is related to. That is, the ontology provides the context to facilitate resource location.
Once a le has been selected, semantically-related les can be located by pressing
the middle button, regardless of the folders where these les are physically located,
providing a resource-centric navigation. This facilitates location of neighbour resources,
but it may be cumbersome whenever browsing is required. In this case, a graph-based
RDF visualizer can be a better option (see [
Conclusions This work strives to lower the adoption barrier of the semantic desktop by providing seamless tooling. To this end, we support the notion of knowledge folder as the underlying infrastructure, and the semantic mouse as the interactive device. Being editorindependent, the mouse accounts for portability and maintainability to face the myriad of formats and editors which characterizes current desktops. Similar to other areas of computing, a balance is needed between generality (e.g. format-independence, editorindependence, etc), and functionality (i.e. the semantic tooling available). seMouse illustrates a semantic-lite approach where a compact set of functions are available to no matter which editor within Windows.