Transcript
Ontologies in SPECTER DFKI Workshop on Ontologies for Personal Memory 25th May 2005 Alexander Kröner
German Research Center for Artificial Intelligence
Outline • About SPECTER – Overview – Memory model
• Ontologies in SPECTER – Structure – Selected Applications
• Outlook – Future research
1
What is SPECTER? • SPECTER is about… – … context- and affect-aware personal assistance – … in instrumented environments – … using a long-term memory
• Major research issues – – – – –
Extension of perception Learning about behavior and affect Augmentation of decision making and effecting Reflection and introspection Usability engineering
Overview
Perceptions
User Feedback
RFID (smart objects, location)
Memory
Introspection
Sensor Input
GPS (location)
Interpretation
Web Interaction (shopping, weather, …)
Support
Biosensors (user feedback)
2
Sample Interaction With Demonstrator Instrumented shop with tagged items and RFID sensors integrated into shelf and basket U visits a Web store; S logs the visit and acquires additional information
In a shop: U inspects a product
U decides on a product
U: User S: SPECTER PJ: Personal Journal
S uses the PJ to create a comparison with previously seen products
Home again: Accompanied by S, U reviews the course of the day and S’s performance
Memory Model Components and Processes Stream of raw sensor data Filtering
Long-term Memory
Registering
Sensor n Updating
Perceptions Archiving
Context Log
Interpreting
Abstracting
User Support
Short-term Memory
...
Sensor 1
Personal Journal Learning
User Model
Introspection
Sensory Memory
3
Memory Model From Sensor Input To Journal Entry • Sensor Memories – Physical (e.g., RFID for object movement) – Virtual (e.g., current weather) – Provide an interface, which communicates sensed information as perceptions encoded in RDF
• Perceptions usually contain small pieces of information, e.g.: – (1): “12:15: user took the camera X from the shelf” – (2): “12:17: user took the camera Y from the shelf” – A perception’s content is modeled using the OWL “subset” of the IEEE SUMO and MILO
• Journal entries are created from perceptions – The abstraction process combines perceptions to small episodes stored in journal entries, e.g.: – (1) and (2): “12:15-12:17: user compared camera X and camera Y”
Ontologies in SPECTER
Structure and selected applications
4
Ontology Structure
High-level commonsense knowledge, e.g., “human”, “building”
Domain-specific knowledge, e.g., about products
IEEE SUMO
Mid-level commonsense knowledge, e.g., “store”
IEEE MILO
About SPECTER, e.g., about services
SPECTER App.
User Model
App. Domain
Personal Journal
Application of Ontologies in SPECTER (examples) • Representing the Environment – Modeling entities, actions,…
• Memory Structure – Perceptions, journal entries,…
• Memory Navigation – Navigation within memory during introspection
• Visualizing Resources – Assigning context-dependant style information to ontology resources
• User Support – Characterization of situations
5
Memory Structure • Perceptions – Containers for individuals of sumo:Process
• Personal Journal Entries – Make reference to perceptions – Contain a fixed set of annotations • • • • •
Time Location Source Ratings …
Navigation • In the GUI, resources may have attached navigation options, e.g., – related journal entries, – detailed information about the resource
• These options are selected on-the-fly for the resource based on – the particular ontology class – available memory content (e.g., are there other entries referring the resource)
6
Visualization •
Ontology resources may have attached display “hints” – CSS information – XSL with Java extensions
•
This visualization information can be specified for varying application contexts, e.g., – List – Detail
•
Styles may be inherited according to the ontology’s hierarchy, example: – milo:Arriving may inherit its style from sumo:Process
•
Styles defined for resources composed of other resources may optionally make use of these resources’ styles
User Support Binding Services to Situations (1) • Goal: Assist the user in the specification of application rules for services provided by SPECTER
• Binding services to situations – Examples
You looked for a new phone – this shop has a phone on sale: …
If the user has entered a shop ⇒ Look for interesting items on sale If shop==SmartShop and value of basket >100€ ⇒ check bank account
– Approach: interactive definition of classifiers for situations
7
User Support Binding Services to Situations (2) Overview of registered bindings
Quick binding of situation/service
(Opt.) Define a situation’s features
•
Determining a situation’s features – Candidate feature set created by the system based on statistical relevance – User may criticize this set and add semantically related concepts (using the underlying ontology)
Summary • Memory model – Sensory memory, short-term memory, long-term memory – Focus on semantic and episodic information
• Ontology – Uses IEEE SUMO/MILO – Is applied for a variety of processes in SPECTER • Modeling the environment, visualization, user support,…
8
Outlook • Scenario – Given several users with a SPECTER-like system at a time
• Exchange of information – Learning from others by exchanging episodes • Where to get the right information? – Looking up information sources in an ad-hoc scenario
• What happens with my information? – How “published” information can be applied
• What do they know about me? – Introspection of other’s memories
• …
9
