Personalization in the EPOS project : Personalization in the EPOS project Leo Sauermann, Andreas Dengel, Ludger van Elst, Andreas Lauer, Heiko Maus, Sven Schwarz
DFKI GmbH
12.06.2006 persona created using http://www.sp-studio.de/
Slide2 : Leo Sauermann
1: no personalization without the person : 1: no personalization without the person
2: no person without the subjective data of the person (from the computer’s perspective) : 2: no person without the subjective data of the person (from the computer’s perspective)
Slide5 : Rome Peter Paul’s
mental model Paul’s concepts pimo-things Paul’s files and
digital resources paul:Rome http://www.comune.roma.it/ relations to occurrences
Slide6 : Results of the EPOS project
Szenario of EPOS : Szenario of EPOS Knowledge Worker „Paul“ using Desktop PC
Problem: Data about single ideas are stored in several applications and without context
Files ↔ Emails
Semantic Desktop
dc:language = AEHS : dc:language = AEHS Adaptive Educational Hypermedia System AEHS
Document Space DOCS
User Model UM
Observations OBS
Adaptation Component AC
Three contributions : Three contributions A representation of the user’s personal information items, including e-mails, files, and other data sources using RDF „native resources“ = DOCS
A representation of the user’s mental model in a formal representation, using several layered ontologies. „PIMO“
A desktop service to capture the current actions of the user, representing the actions using RDF and then calculating the current context of the user. „Context Server“ = OBS
UM = DOCS + PIMO + OBS
Adaptive Applications
PIMO = Personal Information Model : PIMO = Personal Information Model
From native structures to PIMO : From native structures to PIMO Native data is expressed in RDF
DOCS
RDF/S vocabularies like foaf, vCard, Dublin Core
data + structures
Personal Information Model PIMO
Personal Concepts
Topics
Places
People
Types
Workflow
with relations to files and folders
PIMO is The “personal ontology” : PIMO is The “personal ontology” “shared” across applications
filled from DOCS, Company ontologies and domain ontologies
used by the – user
creates instances
creates classes & properties (on the fly)
annotates
PIMO is filled : PIMO is filled automatically from data
Data
Paul‘s files & e-mails aperture.sf.net
(check it out!)
PIMO used and extended : PIMO used and extended
PIMO ontology languages : SemDesk Upper Level Person Role Document Organization Time SemDesk Mid-Level Manager Project Contract Company Offer ontology imports Message dfki.de/ont/pim/pimo PIMO ontology languages PIMO-Basic defines the basic language constructs.
PIMO-Upper A domain-independent ontology defining abstract sub-classes of Thing.
PIMO-Mid: More concrete sub-classes of upper-classes. The EPOS mid-level ontology serves to integrate various domain ontologies and provides classes for Person, Project, Company, etc.
PIMO ontology languages : Domain Model: Bibtech A Heiko Car-Ent Report56 Report EPOS dfki.de/ont/pim/pimo PIMO ontology languages Domain ontologies A set of domain ontologies where each describes a concrete domain of interest of the user. The user’s company and its organizational structure may be such a domain, or a shared public ontology.
all PIMO ontology layers : Rep Lang SystemItems SemDesk Upper Level Person Role Document Organization Time SemDesk Mid-Level Manager Project Contract Company Offer Native Data Vocabularies vCard vEvent dublin core foaf
image Person Image Thing sub-classes Domain Model: Bibtech A Heiko Car-Ent Report56 Report EPOS ontology imports Message aperture.semanticdesktop.org/data dfki.de/ont/pim/pimo all PIMO ontology layers
Paul‘s PIMO - Personal Information Model : Paul‘s PIMO - Personal Information Model PIMO of Person:Paul Paul Project Z Report41 File X e-mail2 vCard H Rep Lang SemDesk Upper Level SemDesk Mid-Level Native Data Vocabularies Domain: Bibtech A Domain: Paul’s company
User Model : User Model UM = DOCS + PIMO + OBS
To capture a user model, we need to know
PIMO the categories/model of the user
DOCS the documents/e-mails attached to the categories
OBS the current context of the user
This holistic user model can now be used for several personalized applications
Context Service : Context Service Plugins gather user actions
Elicitation of task concepts
Notification of GUI
Context Representation : Context Representation Context in EPOS
context of a knowledge worker
context shall support (personal) knowledge management
Contextual elements (CEs)
relevant documents, topics, places, actions, tasks, organizational entities, …
from the user's DOCS and PIMO
not alien data, but known, familiar entities and structures
Service Oriented Architecture
ContextService
gathers events using RDF messages from Plugins
represents context as RDF model, using the PIMO S. Schwarz. A context model for personal knowledge management. In Proceedings of the IJCAII WS. on Modeling and Retrieval of Context, Edinburgh, 2005.
Applications : Applications
Context Assistance : Context Assistance Sidebar
can be switched off
shows current context
documents
people
projects
topics
changes dynamically
use: open related information, pro-active, non-obtrusive assistance system
Application: Drop Box : Application: Drop Box Helps filing information
uses PIMO structures concepts and folders
uses DOCS for text similarity
Knows the users model and is trained by using it
process flow
files are stored into a Drop-Box folder
files are text-analysed and possible target folders are suggested
Drop-Box user interface shows
user selects a folder, classify
files are moved and classified
not used, but obvious: OBS – current context
Semantic Search : Semantic Search search over EPOS data (PIMO)
can be personalized using rules SPARQL queries
example
# found something? -> infer other representations via SPARQL
(?hit retrieve:item ?x) ->
querySparql('CONSTRUCT { ?x pimbasic:hasOtherRepresentation ?y } ')
# found a project? -> also show members
(?hit retrieve:item ?project),
(?project rdf:type org:Project) ->
querySparql('CONSTRUCT {
?project org:containsMember ?m.
}).
Innovation
search result expansion using SPARQL
customized rules for search – only when word “x” is searched, include these results, etc
Semantic Search : Semantic Search
Evaluation : Evaluation
Methodology used for Evaluation in EPOS : Methodology used for Evaluation in EPOS Case Study Method
Case Study with 8 researchers from DFKI
Preperation Phase three months: users learn the system, bugfixing
Evaluation period of 1 week with daily usage
Daily interviews with questionnaire
Usage data collection
Explicit user feedback for proposals from context elicitation – user had to check if results were correct
General observations from questionnaire
Personal Ontology represents the view of the user (90% positive answers)
it is “valuable” for searching and classifying information
Semantic Desktop is perceived as “helpful in their daily work”
Example findings from Case Study : Example findings from Case Study Move & Classify via EPOS DropBox
Filing is faster than before due to proposal of locations?
Yes on 40% of all days
There was still manual filing, but only in 8% of all reported filings
Multiple classification have been used
2,5 categories per file: multicriterial classification
PIMO is populated on-the-fly
Lessons learned: changing the linker : Lessons learned: changing the linker Before:
Evaluation
Manual linking with the Linker was seldom used
Whereas semi-automatic linking was appreciated by means of Move & Classify, Topic-linking, PIM mapping
Redesign
After:
Resources can be “Tagged”, the metaphor is known from Web 2.0 applications. Tags are searched semi-automatically
Benefit is immediately seen
User interface is simpler
… to be evaluated again ….
Semantic Search : Semantic Search evaluated at Siemens SBS by Mark Siebert and Pierre Smits
customization features used
integrated a proposal ontology
conclusion: Through peer search and semantic enhancements, recall was increased. Precision may decrease, depending on the setup and scenario.
Outlook : Outlook
Our goal : Our goal This is my personal
computer
Slide34 : Semantic Applications
Desktop Search Gnowsis Server Aperture Crawlers Outlook e-mail server filesystem Outlook Crawler Ont. Matching Files e-mail Gui invocation Tagging Clustering Domain Ontologies Lucene Index Web 2.0 Interfaces
Open source and reusable : EPOS will be continued
our results are code for you Open source and reusable
gnowsis-beta http://www.gnowsis.org
Nepomuk http://nepomuk.semanticdesktop.org
MyMory http://www.dfki.de/mymory
Slide36 : note:
Demo of Nepomuk today at the EU projects session
I can demo gnowsis beta 0.9
Summary : Summary The PIMO ontology stack and Paul’s PIMO allow us to personalize using precise knowledge about the user
User observation components identify contexts based on PIMO
Applications use PIMO + context in combination
open source, will be continued
Questions : Questions persona created using http://www.sp-studio.de/