Semantic Web

download report

Transcript Semantic Web

The Semantic Web
Week 20: Agents that
can plan and learn..
Module Website:
http://scom.hud.ac.uk/scomtlm/chs2533
Practical this week:
Today
1.
2.
Relationship between Generative Planning and
OWL-S
More on reasoning with actions
Rest of Course:
Weeks 21 - 23 SW Applications and current activity:
European Projects + Information Management
Recall OWL-S – upper level ontology
OWL-S is a
language for
describing web
services. It is
built from
DAML-S and is
written in OWL.
There are 3
parts to a web
service
specification:
Service
presents
ServiceProfile
supports
describedby
ServiceModel
ServiceGrounding
OWL-S

The Service Model contains a model of the
PROCESS of the Service – and a concrete
definition of Inputs, Outputs, Preconditions, Effects
op( getMoney(Bank,20),
[ ],
[ssc(Bank, [ have_password(Bank,P), balance(Bank,A), ge(A,20), is(AZ,A-20) ],
[have_password(Bank,P), balance(Bank,AZ) ] ),
ssc(M, [haveresource(M,Y), is(YZ,Y+20) ],[haveresource(M,YZ)] ) ], [ ]).
Input Bank = ebank
Preconditions = have_password(ebank,P), balance(ebank,A), haveresource(M,Y),
ge(A,20), is(AZ,A-20) is(YZ,Y+20)
Effects/Outputs = have_password(Bank,P), balance(Bank,AZ), haveresource(M,YZ)
¬ balance(ebank,A), ¬haveresource(M,Y),
Planning program example
task2 :- startOCL([
se(ticket,[have_ticket(ticket,apollo,cats)]),
GOALS
se(dvd2, [have_dvd(dvd2,amazon,terminator)]),
se(dvd1, [have_dvd(dvd1,amazon,star_wars)]) ],
[ ss(ebank, [have_password(ebank,abcd),balance(ebank,100)]),
ss(money, [haveresource(money,10)]),
ss(service,[logged_off]),
ss(apollo, [service(apollo),freeseat(apollo,cats),seat_price(apollo,25)]),
ss(amazon, [
service(amazon),
dvd_price(amazon,terminator,10),
“STATIC”
instock(amazon, terminator),
KNOWLEDGE
instock(amazon, star_wars),
dvd_price(amazon,star_wars,15)]),
ss(dvd1, [have_no_dvd(amazon,star_wars)] ),
ss(dvd2, [have_no_dvd(amazon,terminator)] ),
ss(ticket, [have_no_ticket(apollo,cats)])
] ).
Operator Schema
op( buy_dvd(DVD,Seller,Prod,Price),
[se(Seller, [instock(Seller,Prod),dvd_price(Seller,Prod,Price)]), se(S,[logged_on(S,Seller)] ],
[ssc(M, [haveresource(M,Y), ge(Y, Price), is(YX,Y-Price)],[haveresource(M,YX)]),
ssc(DVD, [have_no_dvd(Seller,Prod)],[have_dvd(DVD,Seller,Prod)]) ], [ ]).
op( book_theatre_seat(Ticket,Seller,Prod),
[se(Seller,[freeseat(Seller,Prod),seat_price(Seller,X)]), se(S,[logged_on(S,Seller)]) ],
[ssc(Money, [haveresource(Money,Y), ge(Y, X), is(YX,Y-X)],[haveresource(Money,YX)]),
ssc(Ticket, [have_no_ticket(Seller,Prod)],[have_ticket(Ticket,Seller,Prod)]) ], [ ]).
op( getMoney(Bank,20),
[ ],
[ssc(Bank, [ have_password(Bank,P), balance(Bank,A), ge(A,20), is(AZ,A-20) ],
[have_password(Bank,P), balance(Bank,AZ) ] ),
ssc(M, [haveresource(M,Y), is(YZ,Y+20) ],[haveresource(M,YZ)] ) ], [ ]).
op( logon(Seller),
[se(money,[haveresource(money,R),gt(R,0)]), se(Seller,[service(Seller)]) ],
[ssc(service, [logged_off],[logged_on(service,Seller)]) ], [ ]).
op(logoff(Seller), [ ],
[ssc(service, [logged_on(service,Seller)],[ logged_off ]) ],
[ ]).
Dynamic
Knowledge
Execution Example
| ?- task2.
..................
goal [se(ticket,[have_ticket(ticket,apollo,cats)]),
se(dvd2,[have_dvd(dvd2,amazon,terminator)]),
se(dvd1,[have_dvd(dvd1,amazon,star_wars)])]
achieved by sequence
[getMoney(ebank,20),getMoney(ebank,20),
logon(apollo),book_theatre_seat(ticket,apollo,cats),logoff(apollo),
logon(amazon),buy_dvd(dvd2,amazon,terminator,10),buy_dvd(dvd1,amazon,star_wars,15)]
699 nodes generated
4.51 seconds of cpu
154.9889135254989 nodes search per second
new state is
[ss(dvd1,[have_dvd(dvd1,amazon,star_wars)]),ss(money,[haveresource(money,0)]),ss(dvd2,[
have_dvd(dvd2,amazon,terminator)]),ss(service,[logged_on(service,amazon)]),ss(ticket
,[have_ticket(ticket,apollo,cats)]),ss(ebank,[have_password(ebank,abcd),balance(eban
k,60)]),ss(apollo,[service(apollo),freeseat(apollo,cats),seat_price(apollo,25)]),ss(amazo
n,[service(amazon),dvd_price(amazon,terminator,10),instock(amazon,terminator),instoc
k(amazon,star_wars),dvd_price(amazon,star_wars,15)])]
Planning Algorithm
Nodes = (State, Sequence of actions that got to State)
1. Store the first node (initial state + empty solution)
Repeat
2. pick a node (State, Sequence)
3. pick an action and parameter grounding - ‘A' - that can
be applied to State
4. apply A to State to get State'
5. store new node (State', Sequence + A)
6. if possible, backtrack to 3. and make a different choice.
Until a node (State, Sequence) has been asserted such
that State contains the goal literals
Applying Planning Actions..
4. apply A to State to get State‘ ….
A = getMoney(ebank,20)
op( getMoney(Bank,20), [ ],
[ssc(Bank, [ have_password(Bank,P), balance(Bank,A), ge(A,20), is(AZ,A-20) ],
[have_password(Bank,P), balance(Bank,AZ) ] ),
ssc(M, [haveresource(M,Y), is(YZ,Y+20) ],[haveresource(M,YZ)] ) ], [ ]).
Parameter “Bank” => constant “ebank”
getMoney(ebank,20) =
Precondition = have_password(ebank,P), balance(ebank,A), ge(A,20), is(AZ,A-20) , haveresource(M,Y),
is(YZ,Y+20)
Post condition = have_password(ebank,P), balance(ebank,AZ), haveresource(M,YZ)
Planning Algorithm: applying actions
4. apply A to State to get State‘ …. EXAMPLE
State = [ ss(ebank, [have_password(ebank,abcd), balance(ebank,100)]),
ss(money, [haveresource(money,10)]),
ss(service,[logged_off]), …. ETC] ).
getMoney(ebank,20) =
Precondition = have_password(ebank,P), balance(ebank,A), ge(A,20), is(AZ,A-20) , haveresource(M,Y),
is(YZ,Y+20)
Post condition = have_password(ebank,P), balance(ebank,AZ), haveresource(M,YZ)
Apply A:
1. Are the precondition’s achieved?
Yes, with P =“abcd”, A=100, AZ = 80, M = money, Y = 10, YZ = 30
2. Apply Post Conditions
State’ = [ ss(ebank, [have_password(ebank,abcd), balance(ebank,80)]),
ss(money, [haveresource(money,30)]),
ss(service,[logged_off]), …. ETC] ).
Example Agent Architecture
WEB ENVIRONMENT
Failure
Plan
Execution
Agent
Generative
PLANNER
High
Level
Requests
Response
USER
Feedback
Static
Knowledge
Base
Dynamic
Knowledge
Base
Learning
Component
Summary



Agents have to have a degree of autonomy to
make them useful. They need to be able to reason
with actions, time, events, resources.
Web Agents will be able to effect things by using
Web Services (collect information, supply
information, buy goods, organise transactions..)
Planning algorithms are useful to allow agents to
make plans that achieve goals