1. Guided Interactive Discovery ofGuided Interactive Discovery of
e-Government Servicese-Government Services
Giovanni Maria Sacco
Dipartimento di Informatica, Università di Torino
Corso Svizzera 185, 10149 Torino, Italy
sacco@di.unito.it
Where is the knowledge we have lost in information?
T.S. Eliot, The Rock
2. e-Government Services for citizens represent one of the most
frequent and critical points of contact between citizens and
public administrations.
THE PUBLIC FACE OF GOVERNMENT
e-services represent the only practical way of providing
incentives and support to specific classes of citizens.
THE FRIENDLIERFACE OF GOVERNMENT
3. DISCOVERY of e-services
rather than plain RETRIEVAL
is a critical functionality in e-government systems
But it is managed by search rather than explorative
technology
5. Since the vast majority of information is essentially textual and
unstructured in nature
information retrieval techniques are extensively used both in
pull and push strategies
BUT…
6. 1. almost 80% of relevant documents are not retrieved
2. extremely wide semantic gap between the user model
(concepts) and the system model (words)
3. users have no or very little assistance in formulating queries
4. results are presented as a flat list with no systematic
organization: browsing is difficult or impossible.
7. RICHSEMANTIC SCHEMATA (ONTOLOGIES)
• End-users do not understand them
• Agent mediators required: costly to implement, not
transparent, hard to understand what they do
• Schemata hard to design and maintain
8. Traditional research has focussed on
RETRIEVAL OF INFORMATION
BUT
The most common task is BROWSING:
FIND RELATIONSHIPS
THIN ALTERNATIVES OUT
9. Finding opportunities/services
Finding a job
Finding the laws and regulations that apply
BUT ALSO
Buying a digital camera
Finding a restaurant for tonight
Finding the cause of a malfunction
Selecting a photo
Finding a suspect/missing person from a photobank
….
11. Dynamic Taxonomies:
the first model to fully exploit multidimensional and
faceted classifications
Sacco, G.M., “Dynamic taxonomies: a model for large
information bases”, IEEE Trans. o n Data and Kno wle dg e
Eng ine e ring , May/June 20 0 0
US Patent n. 6,763,349 (EU pending)
DYNAMIC TAXONOMIES
12. Representation
Intension: The infobase is described by a taxonomy designed
by an expert (the schema)
Extension: Documents can be classified at any level of
abstraction and each document is classified under n concepts
(n>1)
No relationships otherthan subsumptions (IS-A, PART-OF)
need to be represented in the schema.
DYNAMIC TAXONOMIES
13. What is a concept?
A concept is a label which identifies a set of documents
(classified under that concept)
A nominalistic approach: concepts are described by instances
ratherthan by properties
Subsumptions require that an inclusion constraint is maintained:
If D(C) denotes the set of documents classified under C and C’ is a descendant
of C in the hierarchy, D(C’)⊆D(C)
DYNAMIC TAXONOMIES
14. How do concepts relate?
By subsumptions (IS-A, PART-OF)
By the Extensional Inference Rule:
Two concepts C and C’ are related if there is at least a
document Dwhich is classified both underC and C’ orone of
theirdescendants
Because of the inclusion constraint, IS-A, PART-OF relationships are a special
case of the Extensional Inference Rule.
DYNAMIC TAXONOMIES
17. Important consequence:
Relationships among concepts need not be anticipated but can
be inferred from the actual classification
Advantages:
a simpler schema
adapts to new relationships (dynamic)
finds unexpected relationships (discovery)
DYNAMIC TAXONOMIES
19. 1. Initial step: Tree picture of the entire infobase
AN EXAMPLE
The infobase schema is used for
browsing
The initial focus is the entire infobase
20. 2. Zoom on a concept and see related concepts
AN EXAMPLE
This is the central operation:
1. The new focus is ANDed with the previous focus
2. The entire infobase is reduced to the documents in the
current focus
3. The taxonomy is reduced in order to show all and only those
concepts which are extensionally related to the selected
focus (filtering)
21. 3. Iterate until the number of documents is
sufficiently small
AN EXAMPLE
3 zoom operations are sufficient to select an3 zoom operations are sufficient to select an
average 10 documents frominfobases withaverage 10 documents frominfobases with
1,000,000 documents, described by a compact1,000,000 documents, described by a compact
taxonomy with 1,000 concepts.taxonomy with 1,000 concepts.
22. • Simple and familiar interface (the only new operation is
the Zoom, which is easily understood)
• The user is effectively guided to reach his goal: at each
stage he has a complete list of all related concepts (i.e.
a complete taxonomic summary of his current focus)
• Completely symmetric interaction: if A and B are
related, the user will find B if he zooms on A, and A if
he zooms on B (most systems are asymmetric)
• Discovery of unexpected relationships
BENEFITS
23. • TRANSPARENCY: the user is in charge and knows
exactly what’s happening
• EXCELLENT CONVERGENCE very few iterations
needed
BENEFITS
24. • Easy multilingual support (just translate concept labels)
• Easy to unobtrusively gather user interests
• Easy to accommodate reviews, popularity, etc.
• Effective push strategies
dbworldx.di.unito.it
BENEFITS
25. • Simple integration with other retrieval techniques (IR,
DB):
dynamic taxonomies as a prefilter:
they establish the context for the query
dynamic taxonomies as a conceptual summary:
they summarize long result lists
BENEFITS
26. CONCLUSIONS
Dynamic taxonomies provide a single and
simple access model that solves the vast
majority of the information dissemination needs
of public administrations
In fact they are so versatile that can be used for:
laws and regulations, e-commerce, medical
guidelines, human resource management,
multimedia information bases…
27. Universal Knowledge Processor
High-performance dynamic taxonomy engine
• Microsoft Windows environment
• A set of high performance multithreaded COM objects
• Intension and extension in RAM even for large
databases (20Mb for 1M documents)
• Extremely fast operation: 327 reduced taxonomies per
second on a 800K item infobase
CONCLUSIONS
28. THE SYSTEM IS AVAILABLE AT
www.knowledgeprocessors.com
Thankyou!
CONCLUSIONS
