Deploying Information Deploying Information Agents on the Web - - PowerPoint PPT Presentation
Deploying Information Deploying Information Agents on the Web Agents on the Web Craig A. Knoblock Knoblock Craig A. University of Southern California University of Southern California and and Fetch Technologies Fetch Technologies Craig
Craig Knoblock University of Southern California 1
Craig A. Craig A. Knoblock Knoblock University of Southern California University of Southern California and and Fetch Technologies Fetch Technologies
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 2 2
Information Agents Research Group Research Group
Steve Minton, Fetch Tech.
Jose Luis Ambite, USC
Greg Barish Barish, Fetch Tech. , Fetch Tech.
Kristina Lerman Lerman, USC , USC
Martin Michalowski Michalowski, USC , USC
Ion Muslea Muslea, SRI , SRI
Maria Muslea Muslea, USC , USC
Sheila Tejada Tejada, UNO , UNO
Snehal Thakkar, USC
Rattapoom Tuchinda Tuchinda, USC , USC
Electric Elves
Hans Chalupsky Chalupsky, USC , USC
Yolanda Gil, USC
Jean Oh, CMU
David V. Pynadath Pynadath, USC , USC
Thomas A. Russ, USC
Milind Tambe Tambe, USC , USC
Funding
DARPA
AFOSR
NSF
Microsoft
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 3 3
Sites provide limited capabilities for personalization personalization
Few sites are designed to be integrated with
Build agents that can perform retrieval, integration, and monitoring tasks on any integration, and monitoring tasks on any
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 4 4
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 5 5
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 6 6
Elves project goal: Apply agent technology to Elves project goal: Apply agent technology to support human organizations support human organizations
Applications: Office Elves and Travel Elves
W W W A g e n t P r o x i e s F o r P e o p l e I n f o r m a t i o n A g e n t s O n t o l o g y - b a s e d M a t c h m a k e r s
GRID
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 7 7
Office Elves created as an application of the Electric Elves Electric Elves
Given travel itinerary, generates set of agents for anticipating travel anticipating travel-
related failures and
Price changes
Schedule changes
Flight delays & cancellations
Earlier and close connections
Finding the closest restaurant given GPS coordinates
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 8 8
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 9 9
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 10 10
Travel Itinerary
W W W A g e n t P r o x i e s F o r P e o p l e I n f o r m a t i o n A g e n t s O n t o l o g y - b a s e d M a t c h m a k e r s
GRID
Flight Prices & Schedules Weather Flight Status Restaurants
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 11 11
Flight-
Status Agent:
Flight delayed message:
Your United Airlines flight 190 has been delayed. Your United Airlines flight 190 has been delayed. It was originally scheduled to depart at 11:45 AM It was originally scheduled to depart at 11:45 AM and is now scheduled to depart at 12:30 PM. and is now scheduled to depart at 12:30 PM. The new arrival time is 7:59 PM. The new arrival time is 7:59 PM.
Flight cancelled message:
Your Delta Air Lines flight 200 has been cancelled. Your Delta Air Lines flight 200 has been cancelled.
Fax to hotel message:
Attention: Registration Desk Attention: Registration Desk I am sending this message on behalf of David I am sending this message on behalf of David Pynadath Pynadath, who has a reservation at your hotel. David , who has a reservation at your hotel. David Pynadath Pynadath is on United Airlines 190, which is now is on United Airlines 190, which is now scheduled to arrive at IAD at 7:59 PM. Since the scheduled to arrive at IAD at 7:59 PM. Since the flight will be arriving late, I would like to request flight will be arriving late, I would like to request that you indicate this in the reservation so that the that you indicate this in the reservation so that the room is not given away. room is not given away.
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 12 12
Airfare Agent: Airfare dropped message
The airfare for your American Airlines itinerary The airfare for your American Airlines itinerary (IAD (IAD -
LAX) dropped to $281.
Earlier-
Flight Agent: Earlier flights message
The status of your currently scheduled flight is: The status of your currently scheduled flight is: # 190 LAX (11:45 AM) # 190 LAX (11:45 AM) -
IAD (7:29 PM) 45 minutes Late If you would like to return earlier, the following If you would like to return earlier, the following United Airlines flights will arrive earlier than your United Airlines flights will arrive earlier than your scheduled flights: scheduled flights: # 946 LAX (8:31 AM) # 946 LAX (8:31 AM) -
IAD (3:35 PM) 11 minutes Late
# 388 LAX (9:25 AM) -
DEN (12:25 PM) 10 minutes Late # 1534 DEN (1:20 PM) # 1534 DEN (1:20 PM) -
IAD (6:06 PM) On Time
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 13 13
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 14 14
<YAHOO_WEATHER>
<TEMP>25</TEMP> <OUTLOOK>Sunny</OUTLOOK> <HI>32</HI> <LO>19</LO> <APPARTEMP>25</ APPARTEMP > <HUMIDITY>35%</HUMIDITY> <WIND>E/10 km/h</WIND> <VISIBILITY>20 km</VISIBILITY> <DEWPOINT>9</DEWPOINT> <BAROMETER>959 mb</BAROMETER> </ROW> </YAHOO_WEATHER>
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 15 15
Define both beginning and end of required information on the page information on the page Name: Joel’s <p> Phone: <i> (310) 777-1111 </i><p> Review: … SkipTo(Phone) SkipTo(<i>) SkipTo(</i>)
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 16 16
Decomposes a hard problem into several easier
Extracts items independently of each other
Inductive Learning System
Extraction Rules
EC Tree Labeled Pages
GUI
Inductive Learning System
Extraction Rules
EC Tree EC Tree Labeled Pages
GUI
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 17 17
Search Space:
SkipTo( <b> ( ) ...
Training Examples:
Name: Del Taco <p> Phone (toll free) : <b> ( 800 ) 123-4567 </b><p>Cuisine ... Name: Burger King <p> Phone : ( 310 ) 987-9876 <p> Cuisine: …
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 18 18
System selects most informative examples to label label
Want to achieve 100% accuracy with as few examples as possible examples as possible
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 19 19
Name: Joel’s <p> Phone: (310) 777-1111 <p>Review: The chef… SkipTo( Phone: ) Name: Kim’s <p> Phone: (213) 757-1111 <p>Review: Korean … Name: Chez Jean <p> Phone: (310) 666-1111 <p> Review: … Name: Burger King <p> Phone:(818) 789-1211 <p> Review: ... Name: Café del Rey <p> Phone: (310) 111-1111 <p> Review: ... Name: KFC <p> Phone:<b> (800) 111-7171 </b> <p> Review:...
Unlabeled Examples Training Examples
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 20 20
Two ways to find start of the phone
Name: KFC <p> Phone: (310) 111-1111 <p> Review: Fried chicken …
SkipTo( Phone: ) BackTo( ( Nmb ) )
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 21 21
Unlabeled data
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 22 22
Name: Joel’s <p> Phone: (310) 777-1111 <p>Review: ... SkipTo( Phone: ) Name: Kim’s <p> Phone: (213) 757-1111 <p>Review: ... BackTo( (Nmb) )
Name: Chez Jean <p> Phone: (310) 666-1111 <p> Review: ... Name: KFC <p> Phone:<b> (800) 111-7171 </b> <p> Review:... Name: Burger King <p> Phone: (818) 789-1211 <p> Review: ... Name: Café del Rey <p> Phone: (310) 111-1111 <p> Review: ...
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 23 23
SkipTo(Phone:) BackTo( (Nmb) )
… Phone: (800) 171-1771 <p> Fax: (111) 111-1111 <p> Review: … … Phone:<i> (800) 555-5555 </i><p> Review: A century ago (1891) … … Phone: (800) 171-1771 <p> Fax: (111) 111-1111 <p> Review: …
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 24 24
Learn “content description” for item to be extracted
Too general for extraction
( Nmb Nmb ) ) Nmb Nmb – – Nmb Nmb can’t tell a can’t tell a phone number phone number from a from a fax fax number number
Useful at discriminating discriminating among among query candidates query candidates
Learned content descriptions
Starts with: ( ( Nmb Nmb ) )
Ends with: Nmb Nmb – – Nmb Nmb
Contains: Nmb Nmb Punct Punct
Length: [6,6] [6,6]
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 25 25
Query: randomly chosen contention point
Output: rule with fewest mistakes on queries
Query: contention point that most violates weak view weak view
Output: committee vote (2 rules + weak view)
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 26 26
33 most difficult of the 140 tasks from [Kushmerick Kushmerick ’97] ’97]
5 10 15 20 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20+ Examples to 100% accuracy Random sampling
Extraction Tasks
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 27 27
5 10 15 20 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20+ Examples to 100% accuracy Naïve Co-Testing Random sampling
Extraction Tasks
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 28 28
5 10 15 20 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20+ Examples to 100% accuracy Aggressive Co-Testing Naïve Co-Testing Random sampling
Extraction Tasks
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 29 29
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 30 30
Name Street Phone Art’s Deli 12224 Ventura Boulevard 818-756-4124 Teresa's 80 Montague St. 718-520-2910 Steakhouse The 128 Fremont St. 702-382-1600 Les Celebrites 155 W. 58th St. 212-484-5113 Name Street Phone Art’s Delicatessen 12224 Ventura Blvd. 818/755-4100 Teresa's 103 1st Ave. between 6th and 7th Sts. 212/228-0604 Binion's Coffee Shop 128 Fremont St. 702/382-1600 Les Celebrites 160 Central Park S 212/484-5113
Zagat’s Restaurants
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 31 31
Learn importance of attributes for matching records Zagat’s
Dept of Health
Art’s Deli 12224 Ventura Boulevard 818-756-4124 Art’s Delicatessen 12224 Ventura Blvd. 818/755-4100
Name Street Phone
Mapping rules: Name > .9 & Street > .87 => mapped Name > .95 & Phone > .96 => mapped
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 32 32
Set of Mapped Objects Choose initial examples Generate committee of learners
Learn Rules Classify Examples Votes Votes Votes
Choose Example
USER
Learn Rules Classify Examples Learn Rules Classify Examples
Label Label
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 33 33
Art’s Deli, Art’s Delicatessen CPK, California Pizza Kitchen Ca’Brea, La Brea Bakery Yes Yes Yes Yes No Yes No No No
Examples M1 M2 M3 Committee
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 34 34
Primary data source may be insufficient to determine mappings determine mappings
Secondary sources can help reduce the uncertainty uncertainty
Examples of secondary sources
Geocoder
Maps street addresses into lat/long coordinates
Business directories
Provide company officers and locations
Area code updates
Provide changes in area codes over time
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 35 35
Record Linkage
Matched Records
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 36 36
Record Linkage
Matched Records Secondary Source
26 Beach Cafe 26 Washington St. Venice, CA 26 Beach Cafe 26 Washington Boulevard Marina Del Rey, Calif
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 37 37
Secondary source reduces the depth of
# Labeled Examples Total Correct Matches Precision Recall Average DT Depth Precision Recall Average DT Depth 25 109 51% 33% 5 66% 51% 1 35 109 73% 57% 8 81% 68% 3 50 109 83% 81% 10 85% 85% 3 Without Secondary Source With Secondary Source
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 38 38
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 39 39
Information gathering may involve accessing and integrating data from many sources and integrating data from many sources
Total time to execute these plans may be large
Slow remote sources
Unpredictable network latencies
Binding patterns
Source cannot be queried until a previous query has been answered been answered
Result: execution is often I/O-
bound
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 40 40
Plan language and and execution system execution system for Web for Web-
based information integration
Expressive enough for monitoring a variety of sources
Efficient enough for real-
time monitoring
Theseus
Executor
PLAN myplan { INPUT: x OUTPUT: y BODY { Op (x : y) } } 01010101010110 00011101101011 11010101010101
Plan Input Data
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 41 41
Examples: : Wrapper Wrapper, , Select Select, etc. , etc.
Operators produce and consume data
Operators “fire” upon any input data
Wrapper Select Join Wrapper
Address 100 Main St., Santa Monica, 90292 520 4th St. Santa Monica, 90292 2 Ocean Blvd, Venice, 90292
City State Max Price Santa Monica CA 200000
Input relation Output relation Plan
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 42 42
MUL MUL ADD a b c d
Operations scheduled by data availability availability
Independent operations execute in parallel
Maximizes horizontal parallelism
Example: computing : computing (a*b) + (c*d)
Operations emit data as soon as possible possible
Independent data processed in parallel
Maximizes vertical parallelism
Producer Consumer MUL MUL ADD a b c d
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 43 43
Prices of used cars
Safety ratings
Reviews
2002 Midsize coupe/hatchback
$4K-
$12K,
No Oldsmobiles Oldsmobiles
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 44 44
meet criteria
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 45 45
meet criteria
Oldsmobiles
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 46 46
meet criteria
Oldsmobiles
reviews for each
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 47 47
meet criteria
Oldsmobiles
reviews for each
reviews of each
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 48 48
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 49 49
WRAPPER
ConsumerGuide Search
(Midsize coupe/hatchback, $4000 to $12000, 2002) ((http://cg.com/summ/20812.htm),
((http://cg.com/full/20812.htm),
search criteria
WRAPPER
ConsumerGuide Summary
WRAPPER
ConsumerGuide Full Review
(car reviews)
WRAPPER
Edmunds Search
((Oldsmobile Alero), (Dodge Stratus), (Pontiac Grand Am), (Mercury Cougar)) JOIN SELECT
maker != "Oldsmobile"
WRAPPER
NHTSA Search
(safety reports)
((Dodge Stratus), (Pontiac Grand Am), (Mercury Cougar))
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 50 50
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 51 51
General problem
Means for issuing and confirming predictions
Two new operators
Speculate: Makes predictions based on "hints" : Makes predictions based on "hints"
Confirm: Prevents errant results from exiting plan : Prevents errant results from exiting plan
Speculate
answers hints confirmations predictions/additions
Confirm
confirmations probable results actual results
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 52 52
J S W W W W W
BEFORE
Same criteria yields same cars
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 53 53
AFTER
Same criteria yields same cars
J S W W Speculate
hints predictions/additions confirmations answers
W Confirm W W
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 54 54
J S W W Speculate W Confirm W W
2002 Midsize coupe $4000-$12000
Time = 0.0 sec
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 55 55
J S W W Speculate W Confirm W W
Oldsmobile Alero T1 Dodge Stratus T2 Pontiac Grand Am T3 Mercury Cougar T4
Time = 0.1 sec
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 56 56
J S W W Speculate W Confirm W W
Dodge Stratus T2 Pontiac Grand Am T3 Mercury Cougar T4
Time = 0.2 sec
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 57 57
J S W W Speculate W Confirm W W
Oldsmobile Alero Dodge Stratus Pontiac Grand Am Mercury Cougar
Time = 1.0 sec
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 58 58
J S W W Speculate W Confirm W W
T1 T2 T3 T4
Time = 1.1 sec Additions (corrections), if any
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 59 59
J S W W Speculate W Confirm W W
Dodge Stratus (safety) (review) T2 Pontiac Grand Am (safety) (review) T3 Mercury Cougar (safety) (review) T4
Time = 3.2 sec
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 60 60
J S W W Speculate W Confirm W W
Dodge Stratus (safety) (review) Pontiac Grand Am (safety) (review) Mercury Cougar (safety) (review)
Time = 3.3 sec
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 61 61
Confirm blocks predictions (and results of) from blocks predictions (and results of) from exiting plan before verification exiting plan before verification
CPU
Speculative operations use "speculative threads"
Lower priority threads
Memory and bandwidth
Speculative operations allocate "speculative resources"
Drawn from "speculative pool" of memory / objects
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 62 62
Use predicted cars to speculate about the ConsumerGuide ConsumerGuide summary and full URLs summary and full URLs
Optimistic performance
Execution time: max
max { {1.2, 1.4, 1.5, 1.6
1.2, 1.4, 1.5, 1.6}
} =
= 1.6 sec 1.6 sec
Speedup over streaming dataflow: (4.2/1.6) (4.2/1.6) = = 2.63 2.63
W J S W W SPEC CONF SPEC W W SPEC
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 63 63
Successive runs of the plan benefit
Even with different input data
Consider optimizing only the most expensive path ( path (MEP MEP) )
Until overhead of further optimization
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 64 64
Associate a hint with a predicted value
2002 Midsize coupe 4K-
12K
Olds Alero Alero, Dodge Stratus, Pontiac Grand Am, Mercury Cougar , Dodge Stratus, Pontiac Grand Am, Mercury Cougar
Use features of a hint to predict value
EXAMPLE: : Predicting car list from Edmunds Predicting car list from Edmunds type = SUV : (Nissan Pathfinder, Ford Explorer) type = Midsize : :...min <= 10000 : (Olds Alero, Dodge Stratus) min > 10000 : (Honda Accord, Toyota Camry)
Cache Decision list
Year Type Min Max Car list 2002 Midsize 8000 15000 (Oldmobile Alero, Dodge Stratus) 2002 Midsize 7500 14500 (Oldmobile Alero, Dodge Stratus) 2002 SUV 14000 20000 (Nissan Pathfinder, Ford Explorer) 2001 Midsize 11000 18000 (Honda Accord, Toyota Camry) 2002 SUV 18000 22000 (Nissan Pathfinder, Ford Explorer)
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 65 65
"http://cg.com/summary/" : ε : COPY
"." :
ε : COPY
Transducers are FSM that translate hints into predictions http://cg.com/summary/20812.htm http://cg.com/full/20812.htm
To create full review URL:
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 66 66
1000 2000 3000 4000 5000 6000 7000 8000 CarInfo RepInfo TheaterLoc FlightStatus StockInfo Time to last tuple (ms)
No speculation 50% correct 100% correct
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 67 67
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 68 68
Goal: Automatically compose new services from existing web services existing web services
We developed services that can dynamically compose information producing services compose information producing services
Builds on data integration techniques to construct plans
Turns the plans into Theseus plans for efficient execution execution
We are extending this work to more complex services that can change the world (side effects) services that can change the world (side effects)
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 69 69
250 750 1250 1750 2250 12/8/2002 12/13/2002 12/18/2002 12/23/2002 12/28/2002 1/2/2003 1/7/2003
Date Price
American Airlines flights192 & 223, LAX-BOS, departing on Jan. 2 & 9
Agents can go beyond gathering and monitoring
They can help make decisions by exploiting the wealth
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 70 70
Agents can go beyond gathering and monitoring
They can help make decisions by exploiting the wealth
Developed a learning system, Hamlet, to predict whether it is better to wait or buy whether it is better to wait or buy [
[Etzioni Etzioni, Knoblock, , Knoblock, Tuchinda Tuchinda, Yates, KDD’03] , Yates, KDD’03]
Collected data on airline prices over several months months
Learned a model of the pricing
In our simulation on collected data, Hamlet saved $198,074 out of a possible $320,572 saved $198,074 out of a possible $320,572 (61.8% of optimal) (61.8% of optimal)
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 71 71
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 72 72
Wrapper learning
Supervised [Kushmerick Kushmerick ’97, Hsu & Dung ’98] ’97, Hsu & Dung ’98]
Unsupervised [Lerman Lerman et al. ’01, et al. ’01, Crescenzi Crescenzi ’01] ’01]
Record linkage
Learning [Cohen ’00, Sarawagi Sarawagi & & Bhamidipaty Bhamidipaty ’02] ’02]
Statistics [Winkler ’98]
Name matching [Bilenko Bilenko et al. ’03, Cohen et al. ’03] et al. ’03, Cohen et al. ’03]
Efficient plan execution
Network query engines [Ives et al. 1999, Naughton Naughton et et
Hellerstein et al. 2001] et al. 2001]
Agent execution systems [Firby Firby ’94, Myers et al. 1996] ’94, Myers et al. 1996]
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 73 73
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 74 74
Web provides the ideal environment for developing and testing software agents developing and testing software agents
Noted by Etzioni Etzioni, AAAI’96 in his talk on , AAAI’96 in his talk on Softbots Softbots
Yet few have seized this opportunity…why?
Like robotics, wide variety of hard technical problems problems
With Web Services, the Semantic Web, etc. the infrastructure is improving infrastructure is improving
Great opportunity for AI
Ability to demonstrate and test technologies in a real-
world setting
Opportunity to apply technologies to make a difference in people’s lives in people’s lives
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 75 75
Many interesting technical challenges for building software agents: software agents:
Wrapping online sources
Linking records across sites
Efficiently executing agent plans
Extraction from text documents
Aligning ontologies
across sources
Planning to integrate data sources
Learning to improve performance and capabilities
Integrating these capabilities in a robust architecture that can: can:
Respond to failures
Explain its behavior
Communicate appropriately
Craig Knoblock Craig Knoblock University of Southern California University of Southern California 76 76
Craig Knoblock University of Southern California 77