dialogues for finding correspondences between partially ... · dialogues for finding...
TRANSCRIPT
Dialogues for finding Correspondences between partially disclosed Ontologies
Terry Payne & Valentina Tamma(in collaboration with Ernesto Jimenez-Ruez & Alessandro Solimando)
Gabrielle Santos, Valentina Tamma, Terry Payne & Floriana Grasso
Dialogue-Based Meaning Negotiation
UK Ontology Network, Newcastle, 14th April 2016Terry PayneUniversity of Liverpool
Using Dialogues to find Alignments
• Different Systems (sensors, devices, services) can assume different ontological models
• Many approaches exist, producing different alignments • But what if fragments of the ontological space are confidential, or
commercially sensitive?
• Agents can exchange knowledge about mappings to find a mutually acceptable alignment.
• Agents disclose preferences on correspondences, terms, axioms • Allows agents to reason with different types of information
2
UK Ontology Network, Newcastle, 14th April 2016Terry PayneUniversity of Liverpool
Approaches to negotiating mappings
3
Aggregating different mappings• Agents selectively identify
what mappings should be disclosed • Strategic choice to avoid undesired
exposure of private knowledge
• Adapts mappings in an evolving environment
Discovering novel mappings• Offer limited ontological
knowledge of seed entities • Agents selectively share
conceptual knowledge to identify localised structural similarity
• Bootstraps the process of aligning different data systems
I know it all! I know nothing!
UK Ontology Network, Newcastle, 14th April 2016Terry PayneUniversity of Liverpool 4
Jimenez-Ruiz E., Payne T.R., Solimando A. and Tamma, V. (2016) Limiting Logical Violations in Ontology Alignment Through Negotiation. In: 15th International Conference on Principles of Knowledge Representation and Reasoning. (KR’16), Cape Town.
Payne T.R., and Tamma, V. (2014) A Dialectical Approach to Selectively Reusing Ontological Correspondences. In: 19th International Conference on Knowledge Engineering and Knowledge Management (EKAW2014), Linköping, Sweden
I know stuff…
Terry Payne & Valentina Tamma(in collaboration with Ernesto Jimenez-Ruez & Alessandro Solimando)
Dialogues for finding Correspondences between partially disclosed Ontologies
But your stuff breaks my stuff!
UK Ontology Network, Newcastle, 14th April 2016Terry PayneUniversity of Liverpool
• Formal Inquiry Dialogue that…• Allows two agents to exchange knowledge
about known mappings
• Aligns only those entities in each agents’ working fragments, without disclosing the ontologies, or all known mappings
• Allows agents to suggest repairs if a mapping introduces a conservatively violation
Extending the Correspondence Inclusion Dialogue (CID)
5
assertrejectCac
ceptC
repair
acceptR
Alice5A
Bob8B
Alice3A
Bob4B
Alice7A
Bob6B
rejectR
assert
rejectCac
ceptC
repair
acceptR
rejectR
join
matched-close
join
Bob2B
Alice1A
matched-close
close
close
D
e
fi
n
i
t
i
o
n
:
T
h
e
m
o
v
e
m
s
= assert for senderx
h
a
s
t
h
e
s
y
n
t
a
x
hx, assert,�,nili, wheret
h
e
f
o
l
l
o
w
i
n
g
c
o
n
d
i
t
i
o
n
s
h
o
l
d
:
• P
r
e
-
c
o
n
d
i
t
i
o
n
s
1
.
x
6= ⇢
2
.
m
o
v
e
t
y
p
e
(ms
�1) 2 {join, endassert, close}
3
.
9� 2 �x
,
s
.
t
.
(
a
)
�
/2 J
B
x
(
b
)
8�0 2 �x
,
i
f
�
0/2 J
B
x
,
d
e
g
r
e
e
(�0) d
e
g
r
e
e
(�)
(
c
)
g
r
o
u
n
d
e
d
(�,Wx)
(
d
)
j
o
i
n
t
e
s
t
(corr(�)) � ✏
• P
o
s
t
-
c
o
n
d
i
t
i
o
n
s
1
.
J
B
x
0= {�} [ J
B
x; J
B
x̂
0= {�} [ J
B
x̂
2
.
x̂
u
p
p
e
r
= d
e
g
r
e
e
(�)
3
.
A
g
0 = A
g
[ {hcorr(�), jointest(�)i}
4
.
⇢
= x
UK Ontology Network, Newcastle, 14th April 2016Terry PayneUniversity of Liverpool
Selecting a subset of possible mappings
• Alignments typically consist of one-to-one mappings• Combining mappings from different alignment fragments can result
in one-to-many mappings; i.e. ambiguity
• Which of these should be selected?• Could it be resolved though objections within the dialogue? • What if the inclusion of a candidate causes a violation?
6
publication article author
submittedPaper reviewedPaper paper editor
UK Ontology Network, Newcastle, 14th April 2016Terry PayneUniversity of Liverpool
• Through the dialogue each agent extends their ontology by including correspondences• however the integrated ontology should not
introduce any change at least in the hierarchy of • Incomplete knowledge about the other agent means the agents
only assess the changes introduced by
• Modify the repair mechanism by Solimando et al. to incrementally check for violations as new correspondences are proposed
Repairs with incomplete knowledge
7
Ox � A � Ox̂
Ox
Ox � A
UK Ontology Network, Newcastle, 14th April 2016Terry PayneUniversity of Liverpool
Example Dialogue
8
Public Knowledge
Private Knowledge
Bob discovers a conservatively violation!!!
b v a
c v a
Ontologyx ⌘ y
z v y
w
Ontologyha,w,⌘i
Alicec = 0.25
ha, x,⌘i
Alicec = 0.9
hb, x,⌘i
Alicec = 0.55
hb, y,⌘i
Alicec = 0.4
hb, z,⌘i
Alicec = 0.6
Correspondence Store �
Correspondence Store �
ha,w,⌘i
Bob
c
= 0.3
ha, x,⌘i
Bob
c
= 0.85
hb, x,⌘i
Bob
c
= 0.5
hb, y,⌘i
Bob
c
= 0.55
hb, z,⌘i
Bob
c
= 0.575
Commitment Store CS
Alice
ha,x,⌘i = 0.9 joint
ha,x,⌘i = 0.85
Bob
hb,y,⌘i = 0.7 joint
hb,y,⌘i = 0.55
Public Knowledge
Private Knowledge
Commitment Store CS
Alice
ha,x,⌘i = 0.9 joint
ha,x,⌘i = 0.85
Bob
hb,y,⌘i = 0.7 joint
hb,y,⌘i = 0.55
a
b c
ax
w
bc y
z
x y
zw
≣≣⊑⊑ ⊑
≣hAlice, assert, hb, z,⌘i, 0.6,?i
As Bob has the axiom z v y, the
inclusion of b ⌘ y and b ⌘ z would
infer: y v z (similarly for x v z)
ax
w
bc y
z
≣≣
UK Ontology Network, Newcastle, 14th April 2016Terry PayneUniversity of Liverpool
Example Dialogue
9
Public Knowledge
Private Knowledge
Bob suggests a repair by weakening the alignment between b and y
b v a
c v a
Ontologyx ⌘ y
z v y
w
Ontologyha,w,⌘i
Alicec = 0.25
ha, x,⌘i
Alicec = 0.9
hb, x,⌘i
Alicec = 0.55
hb, y,⌘i
Alicec = 0.4
hb, z,⌘i
Alicec = 0.6
Correspondence Store �
Correspondence Store �
ha,w,⌘i
Bob
c
= 0.3
ha, x,⌘i
Bob
c
= 0.85
hb, x,⌘i
Bob
c
= 0.5
hb, y,⌘i
Bob
c
= 0.55
hb, z,⌘i
Bob
c
= 0.575
Commitment Store CS
Alice
ha,x,⌘i = 0.9 joint
ha,x,⌘i = 0.85
Bob
hb,y,⌘i = 0.7 joint
hb,y,⌘i = 0.55
Public Knowledge
Private Knowledge
Commitment Store CS
Alice
ha,x,⌘i = 0.9 joint
ha,x,⌘i = 0.85
Bob
hb,y,⌘i = 0.7 joint
hb,y,⌘i = 0.55
a
b c
ax
w
bc y
z
x y
zw
≣≣⊑⊑ ⊑
≣hBob, repair , hb, z,⌘i, 0.575, {hb, y,wi}i
ax
w
bc y
z
≣≣
Either b ⌘ y or b ⌘ z should be weakened!
As joint
hb,y,⌘i < joint
hb,z,⌘i, Bob suggests a repair
that weakens b ⌘ y by removing b w y,leaving the correspondence b v y
UK Ontology Network, Newcastle, 14th April 2016Terry PayneUniversity of Liverpool 10
I don’t know any mappings
Well ask me a question…
Gabrielle Santos, Valentina Tamma, Terry Payne & Floriana Grasso
Dialogue-Based Meaning Negotiation
Santos G., Tamma, V., Payne T.R., and Grasso, F. (2016) A Dialogue Protocol to Support Meaning Negotiation. In: 15th International Conference on Autonomous Agents and MultiAgent Systems. (AAMAS’16), Singapore.
Santos G., Tamma, V., Payne T.R., and Grasso, F. (2015) Dialogue Based Meaning Negotiation. In: 15th Workshop on Computational Models of Natural Argument (CMNA 2015), Bertinoro, Italy
UK Ontology Network, Newcastle, 14th April 2016Terry PayneUniversity of Liverpool
Meaning-based Dialogue
• Explores a cognitive approach to reaching consensus over possible correspondences…• Agents identify possible concepts that may be ontologically
equivalent in their respective ontologies • Each then seeks further evidence over the locality of each concept to
verify if these are structurally similar. • Both agents have the opportunity to ask questions • Correspondences only accepted if both agents accept the same
underlying support
11
UK Ontology Network, Newcastle, 14th April 2016Terry PayneUniversity of Liverpool
Cognitive Approach
12
Open
Propose
Close
Confirm
Agent initiates the dialogue • Identify the name of the entity to be aligned. • Opponent can chose to offer a candidate
corresponding entity, or reject the dialogue
Alice3A
Bob2B
Alice1A
Bob4B
Alice6A
Alice8A
Bob5B
Bob7B
fail
initiate
rejectC
propose
rejectC
justify
testify
justify
justify
assert
rejectS
justify
testify
assertrejectS
accept
accept
Alice3A
Bob2B
Alice1A
Bob4B
Alice6A
Alice8A
Bob5B
Bob7B
fail
initiate
rejectC
propose
rejectC
justify
testify
justify
justify
assert
rejectS
justify
testify
assertrejectS
accept
accept
UK Ontology Network, Newcastle, 14th April 2016Terry PayneUniversity of Liverpool
Propose
13
Open
Propose
Close
Confirm
Allows agent to gather supporting evidence in favour of a correspondence
• Asks for triples representing the entities locality • Attempts to match this to its own triples
If sufficient support found, then it can assert a correspondence
Alice3A
Bob2B
Alice1A
Bob4B
Alice6A
Alice8A
Bob5B
Bob7B
fail
initiate
rejectC
propose
rejectC
justify
testify
justify
justify
assert
rejectS
justify
testify
assertrejectS
accept
accept
Alice3A
Bob2B
Alice1A
Bob4B
Alice6A
Alice8A
Bob5B
Bob7B
fail
initiate
rejectC
propose
rejectC
justify
testify
justify
justify
assert
rejectS
justify
testify
assertrejectS
accept
accept
UK Ontology Network, Newcastle, 14th April 2016Terry PayneUniversity of Liverpool
Confirm
14
Open
Propose
Close
Confirm
Opponent verifies the support for the correspondence
• Can accept the correspondence • May ascertain its own evidence to augment the
support (i.e returning to the propose phase)
Alice3A
Bob2B
Alice1A
Bob4B
Alice6A
Alice8A
Bob5B
Bob7B
fail
initiate
rejectC
propose
rejectC
justify
testify
justify
justify
assert
rejectS
justify
testify
assertrejectS
accept
accept
Alice3A
Bob2B
Alice1A
Bob4B
Alice6A
Alice8A
Bob5B
Bob7B
fail
initiate
rejectC
propose
rejectC
justify
testify
justify
justify
assert
rejectS
justify
testify
assertrejectS
accept
accept
UK Ontology Network, Newcastle, 14th April 2016Terry PayneUniversity of Liverpool
Close
15
Open
Propose
Close
ConfirmAlice3A
Bob2B
Alice1A
Bob4B
Alice6A
Alice8A
Bob5B
Bob7B
fail
initiate
rejectC
propose
rejectC
justify
testify
justify
justify
assert
rejectS
justify
testify
assertrejectS
accept
accept
Alice3A
Bob2B
Alice1A
Bob4B
Alice6A
Alice8A
Bob5B
Bob7B
fail
initiate
rejectC
propose
rejectC
justify
testify
justify
justify
assert
rejectS
justify
testify
assertrejectS
accept
accept
Dialogue fragment terminates if • Both agents accept the correspondence • No support can be found that is acceptable to
both agents
UK Ontology Network, Newcastle, 14th April 2016Terry PayneUniversity of Liverpool
Conclusions• Dialogue mechanisms can support anytime computing of alignments
• Agents can selectively disclose private correspondences given their perceived correctness.
• Extensions allow incremental check/repair for conservativity violations• A modified notion of mutually acceptable repairs mitigate problems due to an agents’
incomplete knowledge about the other agent’s ontology
• A cognitive approach explores possible mappings through asking questions
• mappings are proposed with supporting evidence
• CID Dialogue has been empirically evaluated using OAEI datasets• results have been published at various conferences, and journal versions in progress…
16
http://cgi.csc.liv.ac.uk/~trp/Knowledge-Based-Agents.html