Use Case: Composition

1
Use Case Composition

• OWL-S applied to the
• Berardi-Hull
• problem

2
Component Web Services

• op1 book the plane ticket
• op2 register for event
• op3 book a hotel room
• Problem Compose in ways corresponding to needs
  of tourist, researcher, manager
• Variations One-shot vs. reusable composition

3
Representing service
define atomic process take_it_and_leave Inputs
Outputs okay - Boolean, carrier_there,
carrier_back - Airline, fnum_there -
Flight_num, occ_going - Flight_occ,
fnum_back - Flight_num, occ_returning -
Flight_occ Results -- continued below --
4
ResultsBundling Conditions, Outputs, Effects

• results
• when (fl_there fl_back - Flight)
• (tightly_scheduled(fl_there, fl_back))
• available(fl_there, 1)
• available(fl_back, 1))
• -gt (exists (fl_there fl_back - Flight)
• tightly_scheduled(fl_there,
  fl_back)
• air_reserved(cust, fl_there)
• air_reserved(cust, fl_back))
• output okay true,
• carrier_there
  fl_there.carrier,
• carrier_back fl_back.carrier,
• fnum_there fl_there.num,
• fnum_back fl_back.num,
• occ_going fl_there.occ_desig,
• occ_back fl_back.occ_desig

Condition
Effect
Output
5
Failure Case

• when ()
• (not (exists (fl_there fl_back - Flight)
• tightly_scheduled(fl_there,
  fl_back))))
• output okay false

6
Auxiliary Definition

• tightly_scheduled(fl_there, fl_back)
• iff (flight_departs_from(fl_there,
  departure_place)
• flight_lands_at(fl_there,
  arrive_place)
• flight_departs_from(fl_there,
  arrive_place)
• flight_lands_at(fl_there,
  departure_place)
• flight_depart_time(fl_there)
• max_over((? (fl1)
• flight_departs_from(
  fl1, departure_place)
• 
  flight_lands_at(fl1, arrive_place)
• 
  flight_depart_time(fl1) lt date_leave),
• flight_depart_time)
• flight_depart_time(fl_back)
• min_over((? (fl1)
• flight_departs_from(
  fl1, departure_place)
• 
  flight_lands_at(fl1, arrive_place)
• 
  flight_arrive_time(fl1) gt date_back),
• flight_depart_time))

7
Hotel

• define atomic process book_hotel
• Inputs name_hotel String,
• place_hotel Location,
• date_leave, date_back Date
• Outputs okay - Boolean
• Results
• when (h - Hotel)
• name(h) name_hotel
• located(h, place_hotel)
• (forall (d Date)
• on_or_before(date_leave, d)
• before(d, date_back)
• -gt hotel_available(h, d))
• -gt hotel_reserved(cust, h,
• convex_hull(date_leave,
• 
  day_before(date_back)))
• outputs okay true
• failure case okay false

8
Goals Tourist

• reserved(emh, Pamplona_Run_04)
• exists (date - Time_interval)
• beg(date) day(start(Pamplona_Run_04))
• - 1 day
• end(date) day(start(Pamplona_Run_04))
• 1 day
• exists(h - Hotel)
• dist(h, Pamplona) lt 5 km
• cost_per_night(h) lt 100 Euro
• reserved(emh, Pamplona_Run_04)
• hotel_reserved(emh, h, date)
• exists(af - Air_flight)
• ...

9
Goals Manager

• let e pi_meeting(miracle_intelligence,
  qtr(2004,1))
• registered(rr, e)
• let ps set_of_all ((\\ (p) presentation(p)
• at_conf(p,
  e)
• 
  given_by(rr)))
• ch set_of_all
• ((\\ (m) small_group_discussion(m)
  
• participant(m, rr)
• participant(m,
• Program-mgr
• 
  (miracle_intelligence))))
• --CONTINUED--

10
Goals Manager (cont.)

• exists (intvl - Time_interval)
• beg(intvl) min(min_over
• (ps, (?(p)
  beg(scheduled_time(p)))),
• min_over
• (ch, (? (m)
  beg(scheduled_time(m))))
• end(intvl)
• max(max_over
• (ps, (\\ (p)
  beg(scheduled_time(p)))),
• max_over
• (ch, (\\ (m)
  beg(scheduled_time(m))))

11
Goals Manager (cont.)

• exists (af_there af_back - Air_flight)
• ...
• af_there
• time (arg_max
• (? (f)
• before(arrival_time(f)
  ,
• beg(intvl)))
• (? (f) beg(intvl)
• -
  arrival_time(f)))
• af_back ...
• exists (h - Hotel)
• for_all (beg_date end_date - Day)
• beg_date
• end_date
• -gt hotel_reserved
• (rr, h,
• convex_hull(beg_date,
  end_date)))

12
Okay, this looks hopelessly hairybut most of
it is pure computation(maxes, mins, etc.)
13
Planning Paradigm

• Take goal of user agent and match it against
  effects of services.
• Preconditions become new goals the planner must
  find services and inputs that achieve all the
  outstanding goals.
• Composition is automatic

14
Tourist Case
Matches effect of book_hotel

• Goal is
• is Hotel ?h
• dist(?h, Pamplona) lt 5 km
• cost_per_night(?h) lt 100 Euro
• reserved(emh, Pamplona_Run_04)
• hotel_reserved(emh, ?h, date)
• is Air_flight ?af
• beg(?date) day(start(Pamplona_Run_04))
• - 1 day
• end(?date) day(start(Pamplona_Run_04))
• 1 day
• ...

Added to Conditions of book_hotel
15
Where Did the Composition Happen?

• Multiple actions get instantiated, involving
  multiple web services.
• The order may be unimportant, or the Berardi-Hull
  argument may play a role (not clear).

16
Why Book Flights Ahead of Time?
Because you can? Better question Why sometimes
delay booking (e.g, use PriceLine?) First, we
have to make explicit the time-dependence of
prices. Second, we have to allow for
optimization of objective functions.
17
Planning wrt Existing Composite Process

• Suppose an omnibus travel-agent process
  advertises its process model
• define composite process S4tP()
• participants customer // Plus the service
  itself, of course
• inputs
• outputs
• accept(open_session)
• send to customer (confirm_session)
• run_all
• parallel
• ---ticket process---
• iterate
• ---event-booking process

18

• Approach Decompose composite into individual
  control flows.
• Introduce state where needed. Example
  open_session puts us in session open state.
• Each thread through the process becomes a series
  of actions to plan with.