Upload
matilda-fitzgerald
View
216
Download
1
Tags:
Embed Size (px)
Citation preview
1
On the role of a Librarian Agent in ontology-
based Knowledge Management Systems
Nenad StojanovicInstitute AIFB
WM 2003
Luzern, 2. – 4. 4. 2003
2
Agenda
• Motivation
• Architecture
• Implementation
• Conclusion
3
The Management of the Searching for Information in Semantic Portals
Challenges for the ontology-based Information Retrieval:
(inherited from the Web IR)
- a user posts short queries
- a user reads only top-ten results
(inherited from the Digital Libraries)
- users’ needs vary
Query Management
Ranking Schema
Collection Management
4
Approach: Simulating bricks-and-mortar environment
I need something about ontologies and
evolution 1. Do I have that in the repository
2.
Please, refine the request evolution 4.
How?5.
A lot of users search for
+Editor 6.
+Editor , but visual
7.
You got 20 results
8.
I have time to see three only. Are these the best
ones 9.To notice: one query more for ontologies, ...
10.
Librarian Agent
1000 results!
This query is so
ambiguous
3.
Collection Management Query Management
Do I have that in the repository
Please, refine the request evolution
To notice: one query more for ontologies, ...
A lot of users search for
+Editor
1000 results!
This query is so
ambiguous
Ranking
You got 20 results
5
Agenda
• Motivation
• Architecture
• Implementation
• Conclusion
6
query
response
query/response/action
Log file
Access mechanism
Information repository
User
Librarian Agent
Domainontology
Query Management
Collection Management
quer
y
resp
onse
quer
y
1.
ambi
guity
2.
refin
emen
ts
3.
feedback
5.changerecommendations
6.
Librarian Agent - Architecture
Ranking
4.
7
1. The Ambiguity of a Query
Definition: A query is ambiguous if the information need behind that query cannot be determined uniquely
Types:
A. semantic ambiguity – regarding the used vocabulary
- incompleteness
- unclearness
- redundancy
- un-satisfiable query
B. content ambiguity – regarding the information repository
- equivalency
- generalisation
- specialisation
8
1.A. Semantic Ambiguity
X:Doctor[worksIn->>KM] and X[inGroup->>KM] and X:Researcher.
Compactness:
X:Doctor or X:Researcher Researcher
Doctor
Professor Dozent
Clarity:
X:Doctor means
X:Professor or X:Dozent
Redundancy:
X[worksIn->>KM] is
redundant informationRule:
FORALL X, X[worksIn->>KM] <-X[inGroup->>KM].
9
1.B. Content Ambiguity
For a Query-Answering pair: (M, O)
Structural equivalence (=):
Structural subsumption (parent-child) (<):
212211 ),(),( OOOMOM
212211 ),(),( OOOMOM
query_termsquery_objects
Parameters:
Largest equivalent query:
Smallest equivalent query
Uniquness
Covering
CoveringTerms
),(max a
ia OMQ
adiri
10
1.B Content Ambiguity – Query Map
Query: “Prof. and Project and KM“ NumberOfTerms:=3NumberOfObjects:=3...MaxEquality:= 2LargestEqu:= “Prof. and Project and KM and Group“MinEquality:= 1SmallestEqu:=(“Prof. and Project“) or (“Prof. and KM“)Uniquness:=P5
Generalisation:“Lecturer and Group and KM“ NumberOfTerms:=3NumberOfObjects:=9...Covering = 90%
Generalisation:“Researcher and Project and KM“ NumberOfTerms:=3NumberOfObjects:=8...Covering = 50%
Sibling:“PhD and Project and KM“ NumberOfTerms:=3NumberOfObjects:=1...
Specialisation:“Prof. and EU-Project and KM“ NumberOfTerms:=5NumberOfObjects:=1...
Relat
ions
gen
erat
ed b
y FC
A
11
2. Query Refinement
X:Doctor[worksIn->>KM] and X[inGroup->>KM] and X:Researcher;
a) the structure of the ontology
X:Researcher -> X:Doctor
+ X[project->>.]
„project“ is an appropriate classifier
b) the capacity of the knowledge repository
P1[inGroup->>KM;worksIn->>KM;course->>X; project->>Onto1]
P2[inGroup->>KM;worksIn->>KM;course->>Y; project->>Onto1]
...
P3[inGroup->>KM;worksIn->>KM;course->>W; project->>Onto2]
P4[inGroup->>KM;worksIn->>KM;course->>Z; project->>Onto2]
c) user’s behaviour (how do users refine their queries)
QueryLog:
Collection
Sessionization Query-patterns
Analyse
12
Agenda
• Motivation
• Architecture
• Implementation
• Conclusion
13
Librarian Agent in the “Usage Mining Loop”
Server
Browser
Ontology Manager
UsageOLAP cube
Content vs UsageOLAP cube
Ontology +Knowledge WH
Server
LogRDBMS
OntologyInformation
Content
Logfile
Ontology information
Ontology EvolutionSemantic crawlerOther Ontology Management tools
HTTP Request
Inference Engine
…
Web page+ Ontologyinformation
semantic Log file
SemiPort
14
Librarian Agent in the KAON framework
Persistence, Transaction, Security
RDF API KAON RDF SERVER
KAON API
Data layer
Middleware
KAON PORTALApplications & Services Query Recommendation
Domainontology
Information repository
Map Visualizer
Vision Portal
Usageontology
Usage Log
AmbiguityMeasurement
QueryRefinement
Query Management
RankingModule
CollectionManagement
MapBuilder
UsageLogging
15
Agenda
• Motivation
• Architecture
• Implementation
• Conclusion
16
Conclusion
• Librarian Agent simulates the role of a shop assistant in searching in the brick and mortar environement
• Key issues:
• measuring ambiguity of a query
• ordering query space using FCA
• Very applicable in the Knowledge Portals
• Implementation in the KAON is on the way
• Planed large-scale evaluation for the MEDLINE
17
Thank you for the attention !