Different concepts of context

1
Different concepts of context

• Research Seminar - 8990
• March 2nd, 2004
• presented by Maciej Janik

2
Content

• What is context ?
• Context in NL (natural languages)
• Context in KR (knowledge represetation)
• Context-aware applications
• Semantic web and context - some proposals

3
Context by Oxford English Disctionary

• Context
• joined word - con (by, near) text (meaning)
• Context (two primary meanings)
• the words around a word, phrase, statement, etc.
  often used to help explain (fix) the meaning.
• the general conditions (circumstances) in which
  an event, action etc. takes place.

4
Context by The Dictionary of Philosophy

• Context (L. contexere, to weave together,
  from con with, and texere to weave). Total
  sum of meanings (associations, ideas,
  assumptions, preconceptions, etc.) that are
• - intimately related to a thing,
• - provide the origins for,
• - influence our attitudes, perspectives,
  judgements, and knowledge of that thing.

5
Context 'is' as context 'does'

• Context in natural language - definition by
  indirect description ...
• Stressor - anything that cauese stress (a
  deviation or distortion of a system from its
  normal state)
• if something IS a stressor depends on person
  experiencing it
• e.g. 'traffic jam', number 666
• Stressor is solely defined by its effects, not by
  attributes or properties.

6
Context 'is' as context 'does'

• Context in natural language is similar to
  stressor.
• Context
• is abstract and can be almost anything,
• can be rich object or very narrow,
• is a source of information that can be used to
  reduce ambiguity, vaguesness or
  underspecification in interpretation,
• is build by speakers during conversation for
  mutual understanding.

7
Context in natural language

• Because it is constructed, in part, by the
  speaker and the interpreter it is not the same as
  context in knowledge representation.
• It provides an additional meaning and clarifies
  given facts.
• It may completely change the literal meaning of
  conversation.

8
Context in natural language

• Example - meaning of 'many' in context
• fact 10 cars
• it is many if one person owns 10 cars
• it is not many if factory produces 10 cars
• maybe treat many as ratio ?
• half students in class got influenze - can
  consider as many
• half students in class want to cancel midterem -
  cannot consider as many

9
Context in knowledge representation

• Differs much from context in natural language.
• Tried to be constructed as mathematic entity and
  formalized.
• Is such a rich object that cannot be completely
  described or captured by logic.
• It is not possible to define a super-context, but
  it is possible to use efficiently small contexts
  (micro-theories proposed by Guha).

10
Formalizing context (KR)

• Basic relations
• ist(c, p) - proposition 'p' is true in context
  'c'
• value(c, e) - design the value of term 'e' in
  context 'c'
• Lifting formulas
• relate the proposition and terms in subcontexts
  to possibly more general propositions and terms
  in the outer context
• formula F1 in C1 "states exactely the same" as
  formula F2 in C2

11
Lifting rules example

• Context
• A walks behind B.
• A says "I like that tree on your left".
• B did not hear, turns around and ask A to repeat
  what he said.
• Context has changed - what should be said now?

A
B
A says "I like that tree on your right".
12
Formalizing context (KR)

• Other relations in and between contexts
• specialize-time(t, c) - specify time 't' in
  context 'c'
• at-time(t, p) - assertion that the proposition
  'p' holds at time 't'
• ist(specialize-time(t, c), p) ist(c,
  at-time(t, p))
• specializes(c1, c2) - context 'c2' involves no
  more assumptions than context 'c1'
• assuming(c, p) - creates new context from context
  'c' where proposition 'p' is assumed to be true

13
Proof theory in contexts

• A proof - a finite sequence of statements that
  lead to proved formula P
• each line has a context (list of contexts)
  associated with it
• each line is a formula or enter/exit context
• each formula in proof must satisfy one of
  following
• formula ist(Cn, ... ist(C1 F)) must be axiom
• formula F is obtained from previous ones by
  inference rules (to formulas or contexts), or by
  enter/exit context

14
Intercontext example

• Difference in definition of the same price()
  function in different contexts
• GE - price is for each part
• NAVY - price means price for part, spare parts
  and warranty
• CGE price(FX22-engine) 3600K CGE
  price(FX22-spare-parts) 5K CGE
  price(FX22-warranty) 6K
• CNAVY price(FX22-engine) 3611K

15
Simplified proof

• CNAVY price(FX22-engine) engine spare-parts
  warranty
• CGE price(FX22-engine) 3600K
• CGE price(spare-parts) 5K
• CGE price(warranty) 6K
• Problem solving context
• Cps spares(CNAVY, FX22-engine)
  spare-parts(CGE)
• Cps warranty(CNAVY, FX22-engine) warranty(CGE)

• Cps GE-price(spares(CNAVY, FX22-engine)) 5K
• Cps GE-price(warranty(CNAVY, FX22-engine))
  6K
• Cps GE-price(FX22-engine) 3600K
• Cps GE-price(spares(CNAVY, FX22-engine))
  GE-price(warranty(CNAVY, FX22-engine))
  GE-price(FX22-engine)
• Cps value(CNAVY, price(FX22-engine)) 3611K

16
Problem solving with contexts

• Lift and Solve
• lift assertions from other context into Problem
  Solving Context and solve within it using
  conventional problem solver
• Switch and Solve
• switch to already exsting context, solve the
  problem and lift answer back to the original
  context

17
Enter and exit the context

• Entering and exiting the context has following
  purposes
• provide focus in the problem solving behavior,
• provide a context for the interaction with the
  system.
• It enables to solve problems from different
  context domains by creating another context (this
  may require some lifting rules).

18
Weather and the car

• Enter WinterMT
• weather(NorthEast(USA) Snowy)weahter(NorthWest(U
  SA) Rainy)weather(NorthWest(USA)
  Foggy)weather(South(USA) Sunny)weather(BayArea
  Snowy)weather(BayArea Rainy)
• Exit WinterMT

• Enter CarFeatureMTfeature(fog-lights
  Foggy)feature(anti-lock-brakes
  Snowy)feature(anti-lock-brakes
  Rainy)feature(air-condition Sunny)
• Exit CarFeatureMT

Choice of car desired features depends on climate
of the ares where one lives. In South air
condition will be necessary, while in Bay Area
one may mostly need fog lights.
19
Formal context in KR

• Mainly used by AI or expert systems.
• Permit axiomatization in limited contexts to be
  expanded and transcended to other contexts.
• Used for automatic problem solving or in advising
  systems.
• Economy of representation and efficiency in
  reasoning
• Allowing inconsistent knowledge bases
• Resolving lexical ambiguities

20
Context-aware computing

• A system is context-aware if it uses context to
  provide relevant information and/or services to
  the user, where relevancy depends on the users
  task A.K.Dey.
• Features of context-aware application
• presentation of information and services
• automatic execution of service
• tagging of context to information for future use

21
Semantice e-Wallet

• Example Semantic eWallet application for mobile
  devices from Carnegie Mellon University
  Pittsburg, PA.
• Application used for sharing personal information
  (with privacy rules), access services in campus,
  act as 'intelligent mobile device' (e.g.
  scheduler).
• Context here captures environment settings and
  user preferences.

22
Semantic e-Wallet diagram
23
Semantic e-Wallet

• Categories of knowlege and preferences
• static knowledge - context-independent
  information (e.g. name) and preferences (e.g.
  like italian cuisine),
• dynamic knowledge - context-sensitive knowledge,
  mainly derived from preferences (e.g. no instant
  messages while driving),
• service invocation rules - information that helps
  leverage usage of external resources based on
  current contextual attributes,
• privacy preferences - access control rules,
  obfuscation rules.

24
Semantic e-Wallet

• Contextual information consit of
• location,
• current activity,
• calendar activities,
• social and organizational relationships,
• access rights for different users groups,
• rules about used services,
• information obfuscation rules

25
Word about RDF

• RDF - Resource Definition Framework
• Consists of triples Object A, Relation, Object
  B
• Defines sematic relations between objects in
  semantice web
• Set of such triples is represented as a graph

26
Semantic web

• Semantic web - a network of meanings
• Object (entities) and different connections
  (relations) between them
• Schema to structure object and relation types
  (ontology, taxonomy)

27
Why context in semantic web

• Semantic web - the web of next genreration
• Seach not by keywords, but by meaning and
  meningful relations between objects
• Context here should capture user interests or
  preferences
• System (search) answers may be different for
  different users

28
Semantic Context view

• Context represented as set of specific objects
• search relations that pass the selected objects
  are rated as more important
• e.g. relation A-H
• ABEH
• ABDFH
• ABDGH

29
Semantic Context view

• Context represented as a set of specific relation
  types
• e.g. more important relations 'boss', 'employee'
  than 'friend'
• some search paths become more important in
  returned results
• e.g. G - F relation

30
Semantic Context view

• Context represented as a subregion (subgraph) in
  knowledge graph
• all objects and relations within this region are
  rated higher or these are only important to user

31
Semantic Context view

• Context as additional knowledge that introduce
  new objects and relations
• New relations between current objects

32
Semantic Context view

• Context treated as non-primaty (longer path)
  relations between objects
• e.g. consider rel. B-E

33
Semantic Association Ranking

• Example
• PISTA prototype system implemented in LSDIS lab.
• query about connections between Nasir Ali and
  AlQeada

34
Context Summary

• Natural language
• background conversation knowledge and meanings
• Knowledge representation
• mathematical construct for reasoning
• Context-aware computing
• environment, preferences, rules
• Semantic web
• represented mainly by ontology focus on parts of
  meaning and represented knowledge

35
Questions ?
Thank you
36
References

• R.V. Guha, Contexts A Formalization and Some
  Applications, PhD Thesis, Stanford University,
  1991
• B.Aleman-Meza, Ch.Halaschek, I.B.Arpinar,
  A.Sheth, Context-Aware Semantic Association
  Ranking, LSDIS Lab, University of Georgia, 2003
• G.Hirst, Context as a Spurious Concept,
  Department of Computer Science, University of
  Toronto, 1997
• V.Akman, M.Surav, Steps Toward Formaliziing
  Context, Department of Computer Engineering and
  Information Science, Bilkent University, 1996
• A.K.Dey, "Understanding and using context",
  Georgia Institute of Technology, 2001
• F.L.Gandon, N.M.Sadeh, "Semantic Web Technologies
  to Reconcile Privacy and Context Awareness",
  Carnegie Mellon University Pittsburgh, 2003