1. Semantic Precision for Web-based Interoperation
RIACS / NASA AMES
Semantic Precision for Web-based Interoperation
Gio Wiederhold
Stanford University
July 2001
www-db.stanford.edu/people/gio.html
Thanks to Jan Jannink, Shrish Agarwal, Prasenjit Mitra, & Stefan Decker.
11/28/2018
Gio ICEIS 1

2. Outline Setting VG 3 - VG 5 Precision VG 6 - VG 8
Lack of precision VG 9 - VG 11
SKC solution VG 12, VG 20- VG 29
Ontologies VG 13 - VG 19
Early results VG 30
Interoperation VG 31 - VG 32
Tool & examples VG 33 - VG 41
Composition and execution VG 42 - VG 43
Evolution and maintenance VG44 - VG46
Summary – SKC to general science VG 47 - VG 49
11/28/2018
Gio ICEIS 2

3. T r e n d s : 1999
Users of the Internet %  52% of U.S. population
Growth of Net Sites (now 2.2M public sites with 288M pages)
Expected growth in E-commerce by Internet users [BW, 6 Sep.1999]
segment
books %  16.0%
music & video 6.3%  16.4%
toys 3.1%  10.3%
travel %  4.0%
tickets %  4.2%
Overall %  33.0% = $9.5Billion
An unstainable trend cannot be sustained [Herbert Stein]
 new services **
Year / % Ü
% Ü
E-penetration
Toys
Centroid, in 1999
~1% of total market
11/28/2018
Gio ICEIS 3

4. Growth and Perception E-commerce
Gartner: prediction for 2004: 7.3 T$
Revision:2001 prediction for 2004: 5.9 T$ drastic loss?
50 companies, each after
20% of the market
Examples
Artificial Intelligence
Databases
Neural networks
E-commerce
Extrapolated
growth
Disap-
pointment
Combi-
natorial
growth
Realistic growth
Failures
Perceived
growth
Perception level
Perceived initial growth
Invisible growth
11/28/2018
Gio ICEIS 4

5. Our* Information Environment .
B2B, B2C, G2G, G2C, . . .
In the past: Scarcity
Customers needed more information to make better decisions
Today: Excess
The web provides more information than customers can digest
Effect: confusion in decision-making
Must I look at all possibly relevant information?
What is the penalty for missing something ?
What is the cost of looking at everything ?
I am confused, best defer making any decision
11/28/2018
Gio ICEIS 5

6. Need for precision Precision: Few wrong or irrelevant results
More precision is needed as data volume increases
--- a small error rate still leads to too many errors
Information Wall
human limit
hard to move
human with tools?
acceptable limit
data error rate
information quantity
adapted from Warren Powell, Princeton Un.
11/28/2018
Gio ICEIS 6

7. Relationships among parameters
volume retrieved
volume available
100%
50% 0%
space of methods
%tage actually relevant
perfect recall
v.relevant p=
v.retrieved
r =
v.available
perfect precision
Type 1
errors
recall
Type 2
errors
precision
11/28/2018
Gio ICEIS 7

8. Cost of Error types differs2 1
Missed Valid Information
(False Negatives )
causes lost opportunities
cheapest shovel, . . .
suboptimal decision-making by x
valid suppliers
Excess Information
(False Positives )
has to be investigated
attractive-looking supplier - makes toys
Cost-
benefit
Space of results, ordered
Having many cases of excess information
costs more than some missing information
11/28/2018
Gio ICEIS 8

9. A Major Cause of Errors Searches extend over many domains
Domains have their own terminologies
Need autonomy to deal with knowledge growth
The usage of terms in a domain is efficient
Appropriate granularity
Mechanic working on a truck vs. logistics manager
Shorthand notations
PSU vs. PSU
Functions differ in scope
Payroll versus Personnel
getting paid vs. available (includes contract staff)
11/28/2018
Gio ICEIS 9

10. Semantic Mismatches
Information comes from many autonomous sources
Differing viewpoints ( by source )
differing terms for similar items { lorry, truck }
same terms for dissimilar items trunk( luggage, car)
differing coverage vehicles ( DMV, police, AIA )
differing granularity trucks ( shipper, manuf. )
different scope student ( museum fee, Stanford )
different hierarchical structures supplier vs. usage 
Hinders use of information from disjoint sources
missed linkages loss of information, opportunities
irrelevant linkages overload on user or application program
Poor precision when merged
Still ok for web browsing , poor for business & science
11/28/2018
Gio ICEIS 10

11. Structural Heterogeneity
Same concept?
If incorporated in differing structures?
11/28/2018
Gio ICEIS 11

12. Approach (SKC project)
Scalable Knowledge Composition – Stanford Univ. DB group
Define Terminology in a domain precisely
Schemas, XML DTDs  Ontologies
Develop methods to permit interoperation among differing domains (not integration)
Articulation
Ontology Algebra
Develop tools to support the methods
Ontology matching
11/28/2018
Gio ICEIS 12

13. Functions of Ontologies .
Enable Precision in Understanding
People = designers, implementors, users, maintainers
Systems = sources, mediators, applications
Share the Cost of Knowledge Acquisition & Maintenance
reuse encoded knowledge,
remain up-to-date as domains change
Enable Information Interoperation *
Define the terms that link domains
11/28/2018
Gio ICEIS 13

14. Ancestors of Ontologies
Lexicons: collect terms used in information systems
Taxonomies: categorize, abstract, classify terms
Schemas of databases: attributes, ranges, constraints
Data dictionaries: systems with multiple files, owners
Object libraries: grouped attributes, inherit., methods
Symbol tables: terms bound to implemented programs
Domain object models: (XML DTD): interchange terms
More Knowledge formalized
11/28/2018
Gio ICEIS 14

15. bound to an application
Data and Knowledge
Much Data
Diverse Knowledge
Knowledge Loop
Information is
created at the
confluence of
data -- the state
&
knowledge --
the ability to
select and
project the
state into
the future
Data Loop
Storage
Education
Selection
Recording
Integration
Abstraction
Experience
State changes
Decision-making
Action
bound to an application
11/28/2018
Gio ICEIS 15

16. Two Mismatch Solutions
A Single, Globally consistent Ontology ( Your Hope )
wonderful for users and their programs
too many interacting sources
long time to achieve, 2 sources ( UAL, LH ), 3 (+ trucks), 4, … all ?
costly maintenance, since all sources evolve
no world-wide authority to dictate conformance
Domain-specific ontologies ( XML DTD assumption )
Small, focused, cooperating groups
high quality, some examples - arthritis, Shakespeare plays
allows sharable, formal tools
ongoing, local maintenance affecting users - annual updates
poor interoperation, users still face inter-domain mismatches
11/28/2018
Gio ICEIS 16

17. Global consistency: Hope, but . .
Common assumptions in assembling and integrating distributed information resources
The language used by the resources is the same
Sub languages used by the resources are subsets of a globally consistent language
These assumptions are provably false
Working towards the goal of globally consistency is
1. naïve -- the goal cannot be achieved
inefficient -- languages are efficient in local contexts
unmaintainable – terminology evolves with progress
11/28/2018
Gio ICEIS 17

18. Domain-specific Expertise .
Knowledge needed is huge
Partition into natural domains
Determine domain responsibility and authority
Empower domain owners
Provide tools
Consider interaction
Society of
specialists
11/28/2018
Gio ICEIS 18

19. Domains and Consistency .
a domain will contain many objects
the object configuration is consistent
within a domain all terms are consistent &
relationships among objects are consistent
context is implicit
No committee is needed to forge compromises * within a domain
Domain Ontology
Compromises hide valuable details
11/28/2018
Gio ICEIS 19

20. SKC grounded definition .
Ontology:
a set of terms and their relationships
Term:
a reference to real-world and abstract objects
Relationship:
a named and typed set of links between objects
Reference:
a label that names objects
Abstract object:
a concept which refers to other objects
Real-world object:
an entity instance with a physical manifestation
(or its representation in a factual database)
11/28/2018
Gio ICEIS 20

21. Grounding enables implementation
We use many abstract terms in our work
Needed because we are dealing with many objects
Human thinking is limited to short-term memory
Someone must be able to translate them into code reliably
Each abstract term must have a path to reality
You must provide that path for students and coders
Without a clear path that is not possible
Not automatically at all – machines need specs
Not reliably by human programmers – failures occur
Without implementation there is no benefit
Advice
11/28/2018
Gio ICEIS 21

22. An Ontology Algebra A knowledge-based algebra for ontologies
The Articulation Ontology (AO) consists of matching rules that link domain ontologies
Intersection create a subset ontology
keep sharable entries
Union create a joint ontology
merge entries
Difference create a distinct ontology
remove shared entries
11/28/2018
Gio ICEIS 22

23. Sample Operation: INTERSECTION
Result contains
shared terms
Terms useful
for purchasing
Source Domain 1:
Owned and maintained
by Store
Source Domain 2:
Owned and maintained
by Factory
11/28/2018
Gio ICEIS 23

24. color =table(colcode)
Sample Intersections
Articulation
ontology
matching rules :
size = size
color =table(colcode)
style = style
Ana-
tomy
Material inventory {...}
Employees { }
Machinery { }
Processes { }
Shoes { }
Shoe Factory
Shoe Store
Shoes { }
Customers { }
Employees { }
{. . . }
Hard-
ware
foot = foot
Employees Employees
Nail (toe, foot) Nail (fastener)
Department
Store
11/28/2018
Gio ICEIS 24

25. Other Basic Operations
UNION: merging
entire ontologies
DIFFERENCE: material
fully under local control
Arti-
culation
ontology
typically prior
intersections
11/28/2018
Gio ICEIS 25

26. Features of an algebra The record of past operations can be
Operations can be composed
Operations can be rearranged
Alternate arrangements can be evaluated
Optimization is enabled
The record of past operations can be
kept and reused
(experience: 3 months  1 week for Webster's annual update,
 2 weeks for OED (6 x size [Jannink:01] )
11/28/2018
Gio ICEIS 26

27. INTERSECTION support Articulation ontology Store Ontology Matching
rules that use
terms from the
2 source domains
Factory
Terms useful
for purchasing
11/28/2018
Gio ICEIS 27

28. Primitive Operations Model and Instance ... Constructors create object
create set
Connectors
match object
match set
Editors
insert value
edit value
move value
delete value
Converters
object - value
object indirection
reference indirection
Unary
Summarize -- structure up
Glossarize - list terms
Filter - reduce instances
Extract - circumscription
Binary
Match - data corrobaration
Difference - distance measure
Intersect - schem discovery
Blend - schema extension
...
11/28/2018
Gio ICEIS 28

29. Matching rules (in process)
A equals U
A equals {u, … }
A equals Union (U, V)
A equals Union (U, {v, ...})
A equals Intersection (U, W)
A equals Intersection (U, {w, ...})
A equals Difference (U, X)
A equals Difference (U, {x, ...})
must
obey algebraic
properties
11/28/2018
Gio ICEIS 29

30. Sample Processing in HPKB
What is the most recent year an OPEC member nation was on the UN security council (SC)?
(An DARPA HPKB Challenge Problem)
SKC resolves 3 Sources
CIA Factbook ‘96 (nations)
OPEC (members, dates)
UN (SC members, years)
SKC obtains the Correct Answer
1996 (Indonesia)
Other groups obtained more,
but factually wrong answers;
they relied on one global source, the CIA factbook.
Problems resolved by SKC
Factbook – a secondary source -- has out of date OPEC & UN SC lists
Indonesia not listed
Gabon (left OPEC 1994)
different country names
Gambia => The Gambia
historical country names
Yugoslavia
UN lists future security council members
Gabon 1999
needed ancillary data
11/28/2018
Gio ICEIS 30

31. Interoperation via Articulation
At application definition time
Match relevant ontologies where needed
Establish articulation rules among them.
Record the process
At execution time
Perform query rewriting to get to sources
Optimize based on the ontology algebra.
For maintenance
Regenerate rules using the stored process formulation
11/28/2018
Gio ICEIS 31

32. Generation of the rules
Provide library of automatic match heuristics
Lexical Methods -- spelling
Structural Methods -- relative graph position
Reasoning-based Methods
Nexus 
Hybrid Methods
Iteratively, with an expert in control
GUI tool to
- display matches and
- verify generated matches using the human expert
- expert can also supply matching rules
11/28/2018
Gio ICEIS 32

33. Articulation Generator
Being built by Prasenjit Mitra
Thesaurus
OntA
Context-based
Word Relator
Phrase Relator
Driver
Semantic
Network
(Nexus)
Structural
Matcher
Ont1
Ont2
Human Expert
11/28/2018
Gio ICEIS 33

34. Lexical Methods Preprocessing rules. - Expert-generated seed rules.
e.g., (Match O1.President O2.PrimeMinister)
- Context-based preprocessing directives.
Thesaurus - synonyms, generalizations
yellow  ochre, canary
Nexus – term relationship graph
Owner = buyer
( Distance of words as measure of relatedness )
11/28/2018
Gio ICEIS 34

35. Tools to create articulations
Graph matcher
for
Articulation-
creating
Expert
Vehicle
ontology
Transport
ontology
Suggestions
for articulations
11/28/2018
Gio ICEIS 35

36. continue from initial point
Also suggest similar terms
for further articulation:
by spelling similarity,
by graph position
by term match repository
Expert response:
1. Okay
2. False
3. Irrelevant
to this articulation
All results are recorded
Okay’s are converted into articulation rules
11/28/2018
Gio ICEIS 36

37. Candidate Match Nexus * free, we also have an OED-based nexus.
Term linkages automatically extracted from 1912 Webster’s dictionary *
* free, we also have an OED-based nexus.
Based on processing headwords ý definitions using algebra primitives
Notice presence of 2 domains: chemistry, transport
11/28/2018
Gio ICEIS 37

38. Using the nexus
11/28/2018
Gio ICEIS 38

39. Navigating the match repository
11/28/2018
Gio ICEIS 39

40. Example: NATO Country Graphs
Austria:
bundestag
....
11/28/2018
Gio ICEIS 40

41. To be matched to 70% of documented matches found automatically
Great Britain
parliament
....
70% of documented matches found automatically
Remainder required human interaction with our tools.
11/28/2018
Gio ICEIS 41

42. Broader Applications Compose
Composed ontology for
applications using A,B,C,E
Articulation
ontology
for U (A B) (B C) (C E)
Articulation
ontology
Legend:
U : union U
: intersection U (C E)
Articulation
ontology
for
ontology
for resource E
Ontology
for C (A B) U U (B C) U (C D)
Ontology
for resource A
Ontology
for resource B
Ontology for
resource D
11/28/2018
Gio ICEIS 42

43. Exploiting the result Result has links to source
Future work
Result has links
to source
Processing & query evaluation is best performed within Source Domains & by their engines
11/28/2018
Gio ICEIS 43

44. Architectural development
time W2 W1 D2 D6 D4 W3 I1 D1 D5 I2 M1 M2 A1 A4 A5 A2 A6 b. A3
datasources
wrappers
mediators
Network
middleware
integrators
applications D3
11/28/2018
Gio ICEIS 44

45. Transform Data to Information
data and simulation resources
value-added services
decision-makers at workstations
Application Layer
Mediation Layer
Foundation Layer
11/28/2018
Gio ICEIS 45

46. Mediation and maintenance
Application
Interface
Changes of
user needs
Domain
ontology
changes
Software & People
Owner / Creator
Maintainer
Lessor - Seller
Advertisor
Models, programs,
rules, caches,
Resource Interfaces
Resource functional &
ontological changes
Tools can help, but changes have to be dealt with rapidly.
Automated learning is typically too slow, requires many instances
11/28/2018
Gio ICEIS 46

47. Domain Specialization
Knowledge Acquisition (20% effort) &
Knowledge Maintenance (80% effort *)
to be performed by
Domain specialists
Professional organizations
Field teams of modest size
autonomously
maintainable
Empowerment
* based on experience with software
11/28/2018
Gio ICEIS 47

48. SKC Synopsis Research Objective:
Precise answers from heterogeneous, imperfect, scalably many data sources
Sources for Ontologies:
General: CIA World Factbook ‘96, UN-www, OPEC-www
Webster’s Dictionary, Thesaurus, Oxford English Dictionary
Topical: NATO, BattleSpace Sensors, Logistics Servers
Theory:
Rule-based algebra over ontologies
Translation & Composition primitives
Sponsor and collaboration
AFOSR; DARPA DAML program; W3C; Stanford KSL and SMI; Univ. of Karlsruhe, Germany; others.
11/28/2018
Gio ICEIS 48

49. Innovation in SKC
No need to harmonize full ontologies
Focus on what is critical for interoperation
Rules specific for articulation
Tools for creation and maintenance
Maintenance is distributed
to n sources
to m articulation agents in mediators
Potentially many sets of articulation rules
is m < n2 , depends on semantic architecture density
a research question: density
11/28/2018
Gio ICEIS 49

50. Conclusion Today solved by human intermediate experts
High precision is important for enterprise applications
cost of overload versus opportunity loss
Semantic differences cause problems
Today solved by human intermediate experts
Will need automation support
Tools so that expert knowledge is captured and maintainable
Scalability requires a thorough foundation
Algebra provides composition, formal basis, delegation
Formal composition supports maintenance
Delegation of responsibility and authority enhances quality
Many research tasks left
11/28/2018
Gio ICEIS 50

51. Long Range Science Vision
Artificial
Intelligence
knowledge mgmt
domain expertise
uncertainty
Databases
access
storage
algebras
Systems
Engineering
analysis
documentation
costing
. . .
. . .
Integration Science
Gio ICEIS 51