DeepaMehta – A Semantic Desktop

                   Jörg Richter2 , Max Völkel1 and Heiko Haller1
                       1
                       AIFB, University of Karlsruhe, Germany
                        {mvo,hha}@aifb.uni-karlsruhe.de,
                     http://www.aifb.uni-karlsruhe.de/WBS
          2
            Co-Founder and Lead Architect of DeepaMehta, Berlin, Germany
                   jri@freenet.de, http://www.deepamehta.de



        Abstract. DeepaMehta is an open source semantic desktop application
        based on the Topic Maps standard. It’s conceptualization and especially
        the UI have been guided by findings of cognitive psychology, in order
        to provide a cognitively adequate working environment for knowledge
        workers of all kind. It uses a graph visualization similar to concept maps.
        DeepaMehta aims to evolve nowadays’ separated desktop applications
        into an integrated workspace, enabling the user to organize, describe,
        relate, edit and use almost any information objects.


Introduction In this paper we present the Topic-Map-centric semantic desktop
environment “DeepaMehta”. First we state some psychological requirements for
personal knowledge management (PKM). Then we describe the UI concepts
and their realisation via the Topic Map metaphor. We conclude with a brief
evaluation based on psychological criteria.


Psychological Requirements It should be the main goal of any knowledge
management software, to facilitate creation, externalisation, and (re)construction
of knowledge. Since there is evidence, that conceptual human knowledge is actu-
ally stored in an associative way, comparable to semantic networks [1], it appears
sensible to provide the knowledge worker with a UI, where contents are displayed,
managed, created, and refined in such an associative manner (i. e. items together
with their relations to other items), that enables the construction of semantic
networks—like concept maps [2, 3] do, as well as their more formal derivatives
knowledge maps[4].
    Mapping Techniques 3 allow the knowledge worker to use his natural sense of
spatial orientation wich easily distinguishes spatial positions and layouts (also
in a plane) to gain orientation in his knowledge space. Research in cognitive and
instructional psychology has shown, that the use of concept-map-like techniques
can have various positive effects on learning and problem solving—i. e. knowledge
generation and -use [5, 4, 6].
3
    In this article the term “mapping” is used as coined in the domain of instructional
    psychology, i. e. in the sense of creating and using visual knowledge representations
    called “maps” like mind-maps, concept maps etc.
    A major goal of user interaction design—especially in hypermedia—is to keep
cognitive overhead as low as possible. This is “the additional effort and concen-
tration necessary to maintain several tasks or trails at one time” [7]. Because
human working memory and thus capacity for conscious processing are quite
limited[8], we should avoid wasting it to secondary tasks like worrying about
saving files, dealing with layout and formatting or regaining orientation in the
information environment while writing the actual content.




             Fig. 1. A Topic Map (typical DeepaMehta working screen)


Design The design of DeepaMehta is centered around the model of Topic Maps.
Topic maps are a human-oriented approach to encode knowledge and meta
knowledge (knowledge about knowledge). Topic maps consist of topics, asso-
ciations and occurrences. In the semantic web, this relates to resources, relations
and instances. Topic maps form a self-describing type system, much like RDF
schema. We assume some familiarity with Topic Maps and refer the reader to
ISO standard 13250 [9] and RDF [10].
     DeepaMehta is an application framework with a Topic-Map-based UI (see
fig. 1), the design of which has been guided by findings in cognitive psychology.
Information of any kind as well as the relations between information items can
be displayed and edited in the same space. The user is no longer confronted with
files and programs. There are no overlapping windows, no menu bars and no
dialog boxes. Topic Maps are individual views on interconnected contents. An
application in this context is a collection of topic types, for which specialised and
generic commands are executable by the user (e. g. what’s related, hide, retype,
delete).


A Cognitively Adequate User Interface One of the most obvious problems
in current desktop user interfaces is that of context switching. Users currently
have to switch applications for every sub-task. Each switch presents a completely
new interface to the user, even if the underlying concepts are the same: An
address used within a text document is conceptually not much different from the
same address used in an email. Today’s desktop UIs are application-oriented,
not data- or task-oriented. DeepaMehta gives the user stable views, letting
him focus on the task itself, without leaving the work-context: In one and the
same view the user can read an e-mail, link it to an existing topic, attach a note
to it, search for related media, save the search results, make semantic statements
and spatially arrange all these items on the screen. And he will later allways find
his workspace exactly as he left it.
    Browsing the web is easy. Figuring out later where one has been is not that
easy. A browser history is merely time-based (1) and offers no means to attach
any kind of additional information (2). Even worse, after a fixed time interval,
the history is usually erased automatically (3). If not, it becomes so large that it
is virtually impossible to handle (4). Additionally, the browser history contains
no information about other resources accessed than web pages (5). Bookmarks
sometimes do offer an annotation feature, but problems (4) and (5) still remain.
Maintaining bookmarks in a way they remain usable requires quite some addi-
tional effort compared to mere web browsing. DeepaMehta offers constructive
browsing as a solution: Each resource visited (be it a web page or something
else) is represented as a topic in the current workspace. Each new topic is placed
right next to the preceding one. The user can conveniently move these newly
created topics around to other places in his workspace, which is always visible.
Now, surfing the web or accessing other resources automatically creates a map
of viewed objects. Even searches and search results are represented in the same
consistant fashion. This spatially arranged map visualises a work process better
then a list of named URLs and it is persitant, automatically saved and fully
navigable.
    Most traditional applications have a fixed set of objects they can deal with.
The DeepaMehta type system is extensible. The user can construct new topic
and relation types on the fly in the same UI, using a set of very few built-in topic
types like topic type, assoc. type, property, prop. value, data source, search, map,
and workspace. New topic types can be used instantly and serve as the basis for
the UI. Even user management is done using the topic map concept (users are
represented as topics, with relations to groups and shared workspaces).
    DeepaMehta is realised with a service oriented architecture (SOA) offer-
ing many communication and integration options. It features a thin client, a web
interface and a set of desktop adapters. Data can be stored in several back-ends,
exported to XTM or PDF or even shared via SOAP web services. Each topic
type can be provided with it’s own java class, to give it unique functionality.
    For collaboration, users can share their workspace (topics, topic types and
maps) with others, so they can use it e. g. in the web interface relying on the
integrated access control system and central synchronisation server.

Evaluation The DeepaMehta architecture defines a new application model
and gives developers a framework to design DeepaMehta-applications. Such ap-
plications are easy to maintain as the business logic resides on the server. Also a
range of interaction front-ends is offered: a thin client, web front-end and even a
PDA interface4 . The thin client framework provides a solid framework for many
kinds of interaction clients. A more up-to-date user interface is planned.
    This flexibility enabled the successful deployment of several commercial sites
in a variety of domains5 , among which are two eLearning projects and the award-
winning geographic information system “Kiezatlas”. For consultants, a compe-
tence analysis tool was implemented. A third project acts as an information
management system for modern and contemporary artwork.
    In order to evaluate the innovative user interface, we checked6 it against
a set of criteria set up to evaluate visual mapping tools for personal knowledge
management from a cognitive psychological point of view [6].

 – Free Placing an item on the canvas is possible.
 – Free Relations: Stating relations between items in DeepaMehta is possible
   in all degrees of formality (unlinked nodes, unlabeled links, labelled links,
   typed links)
 – Every item can be given an Annotation in natural language.
 – The most useful way to deal with complexity and clarify the macro struc-
   ture of a domain, is to use clustering [11, 8]. DeepaMehta offers only visual
   grouping within a single map but also the ability to create sub-maps.
 – For brainstorming, too many mouse-clicks are currently required and cre-
   ating new items without leaving the keyboard is not possible at all. The
   cognitive overhead should be reduced for standard tasks.
 – The detail and context problem [7, 12] is tackled in DeepaMehta by splitting
   the screen and showing the user always both the actual content (property
   pane, right hand side) and the context information (topic pane, left hand
   side). The interlinked topic-view always displays all related items, while the
   property pane can be navigated classically by traversing links or editing
   property values.

DeepaMehta’s UI and interaction paradigm takes a consequent approach of min-
imalist design, where only relevant controls are shown. As this differs from com-
mon interfaces, it requires some initial time to get acquainted.
    Providing zooming capability and a grouping feature would surely increase
its utility, especially for the use of larger and more complex maps.
4
  Download at http://www.deepamehta.de/docs/deepamobil.html
5
  http://www.kiezatlas.de, http://artfacts.net
6
  Due to space limitations, the more detailed and elaborate evaluation can only be
  found in the long version of this paper
Conclusion Wether DeepaMehta will succeed in replacing nowadays standard
applications or not, in any case it introduces and combines several quite promis-
ing innovative approaches to personal knowledge management and user interface
design. Without a bloated interface it offers many useful features well-thought-
through.
    As is common for prototypes, usability is still improvable, however it becomes
clear that DeepaMehta beares a high potential, combining the advantages of
visual mapping techniques and semantically specified topic maps. Furthermore
it offers a solid and web service enabled back-end for collaborative creation and
use of knowledge bases ranging from informal collections of notes to fully fledged
ontologies. And it has allready proven it’s utility in several production-status
commercial projects.
    It is a networked environment for personal ontology-based management of
unified knowledge.

Acknowledgments: Research reported in this paper has been partially financed
by the EU in the IST-2003-507482 project Knowledge Web7 and is supported by the
German Federal Ministry of Education and Research (BMBF) under the SmartWeb
project. We would like to thank our colleagues for fruitful discussions.


References
 1. Quilian, M.R.: Semantic Memory. In: M. Minski (ed.). Semantic Information Pro-
    cessing. MIT Press, Cambridge, MA (1968)
 2. Novak, J.D., Gowin, D.B.: Learning how to learn. Cambridge University Press,
    New York (1984)
 3. Novak, J.D.: (The theory underlying concept maps and how to construct them)
    Available through Institute for Human and Machine Cognition, The University of
    West Florida http://cmap.coginst.uwf.edu/info/ (July 2002).
 4. O’Donnell, A.M., Dansereau, D.F., Hall, R.: Knowledge Maps as Scaffolds for
    Cognitive Processing. Educational Psychology Review 14 (2002)
 5. Jonassen, D.H., Beissner, K., Yacci, M.: Structural Knowledge: Techniques for
    Representing, Conveying and Acquiring Structural Knowledge. Lawrence Erlbaum
    Associates, Inc (1993)
 6. Haller, H.: Mappingverfahren zur Wissensorganisation (2003) availlable online at
    http://heikohaller.de/literatur/diplomarbeit/.
 7. Conklin, J.: Hypertext: an introduction and survey. Computer 20 (1987) 17–41
 8. Miller, G.A.: The magical number seven, plus or minus two: Some limits on our
    capacity for processing information. Psychological Review 63 (1956) 81–97
 9. Michel Biezunski, Martin Bryan, S.R.N.: ISO/IEC 13250:2000 Topic Maps. Tech-
    nical report (1999)
10. Pepper, S.: Ten theses on Topic Maps and RDF. Technical report (2002)
11. Anderson, J.R.: Cognitive Psychology and Its Implications. 6th edn. Worth Pub-
    lishers (2005)
12. Furnas, G.W.: Generalized Fisheye Views. In: proceedings of Human Factors in
    Computing Systems CHI ’86. (1986) available online: http://www.si.umich.edu/
    ~furnas/Papers/FisheyeCHI86.pdf.

7
    see http://knowledgeweb.semanticweb.org