Using Preferences in Negotiations over Ontological Correspondences

18
Using preferences in Negotiations over Ontological Correspondences Terry Payne & Valentina Tamma University of Liverpool [email protected] [email protected]

Transcript of Using Preferences in Negotiations over Ontological Correspondences

Using preferences in Negotiations over Ontological

Correspondences

Terry Payne & Valentina TammaUniversity of Liverpool

[email protected]@liverpool.ac.uk

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool

Open Systems, Ontologies and Alignment

• Agents (applications, devices, services) can assume different ontological models

• Modelled implicitly, or explicitly by defining entities (classes, roles etc), typically using some logical theory, i.e. an Ontology

• Alignment Systems align similar ontologies

• If we assume that different alignments exist, how do agents choose which to use?

2

Alignment

Correspondence

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool

Align Everything?

• What does the agent know?• Pre-computed alignments exist, and can be shared

• Different agents may possess different alignment fragments from different sources.

3

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool

Align Everything?

• Do we need everything to be aligned?• An agent may aggregate several ontologies for a variety of domains

• A task may be relevant to only a single module within an ontology

• Fragments of the ontological space may be confidential, or commercially sensitive.

4

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool

Correspondence Inclusion Dialogue (CID)

• Formal Inquiry Dialogue that…• Allows two agents to exchange knowledge about

correspondences to agree upon a mutually acceptable final alignment AL.

• Aligns only those entities in each agents’ working ontologies, without disclosing the ontologies, or all of the known correspondences.

5

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool

Correspondence Inclusion Dialogue (CID)

• Assumptions1. Agents typically possess some knowledge about different

correspondences from different sources

2. This knowledge is partial, asymmetric, and possibly ambiguous; i.e. more than one correspondence exists for a given entity

3. Agents each associate a weight (Degree of Belief) κc to each unique correspondence

4. Joint weights are computed when a correspondence is disclosed

5. Correspondences with a joint weight below the admissibility threshold ϵ should be rejected

6

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool

Dialogue Moves• Dialogue consists of a sequence of moves

• Agents take turns to select and propose a belief they know of, that has not yet been asserted, based on its weight κc

• A shared, or asserted correspondence is:

• accepted based on their combined κc (i.e. joint(c))

• rejected if joint(c) < ϵ, the admissibility threshold • objected to if an agent believes a better correspondence exists for

one of the entities in the correspondence

• The dialogue is presented formally in Payne & Tamma, AAMAS14

7

Payne T.R., and Tamma, V. (2014) Negotiating over Ontological Correspondences with Asymmetric and Incomplete Knowledge. In: 13th International Conference on Autonomous Agents and MultiAgent Systems. (AAMAS’14), Paris.

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool

Dialogue Moves

8

Payne T.R., and Tamma, V. (2014) Negotiating over Ontological Correspondences with Asymmetric and Incomplete Knowledge. In: 13th International Conference on Autonomous Agents and MultiAgent Systems. (AAMAS’14), Paris.

join

matched-close

join

matched-close

object

object

accept

rejectaccept

endassert

endassert

reject

endassert

endassert

assert

assert

join

join

object

object

Alice&Bob

3ABBob2B

Alice2A

Alice1A

Bob1B

Alice5A

Alice4A

Bob6B

Alice6A

Bob4B

object

object

Bob5B

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool

Ambiguity and Objections

9

publication article author

submittedPaper reviewedPaper paper editor

• Alignments typically consist of one-to-one mappings• Combining correspondences from different alignment fragments can

result in one-to-many correspondences; i.e. ambiguity

• Which of these should be selected?• Should selection be deferred until all candidates are found? • Could it be resolved though objections within the dialogue?

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool

Finding an injective alignment?

10

article

publication

draft

paper

0.5

0.6

0.7

• Option 1:• Reduce problem to that of finding a

Matching in a bipartite graph with weighted edges

• The problem becomes one of finding the right objective function • Use Stable Marriage algorithm to prefer stable

(highly weighted) edges • Use Hungarian algorithm to maximise the total

weight of edges

• However: • All candidate correspondences need to be found. • Solution based purely on joint weights

Stable Solution:⟨article,paper,≣⟩

Maximal Solution:⟨article,draft,≣⟩

⟨publication,paper,≣⟩

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).

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool

⟨article, reviewedPaper,≣⟩ 0.45

⟨article, paper,≣⟩ 0.65

11

• Option 2:• Build a simple attack graph during the dialogue

• Alternate correspondences are counter-proposed (through an objection) to form an attack graph

• Attacks can be directed by the difference in the weight of each correspondence • Bi-directional attacks are resolved by random selection of one of the alternatives

• Can then use grounded semantics to determine the extension

Finding an injective alignment?

⟨article, submittedPaper,≣⟩ 0.5

⟨publication, paper,≣⟩ 0.7

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool 12

• Option 2:• Build a simple attack graph during the dialogue

• Alternate correspondences are counter-proposed (through an objection) to form an attack graph

• Attacks can be directed by the difference in the weight of each correspondence • Bi-directional attacks are resolved by random selection of one of the alternatives

• Can then use grounded semantics to determine the extension

Finding an injective alignment?

⟨article, submittedPaper,≣⟩ 0.5

⟨publication, paper,≣⟩ 0.7 publication

article author

submittedPaper reviewedPaper paper editor

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool 13

• Option 3:• Build a preference-based argumentation graph during the

dialogue• Based on the work of Parsons, Wooldridge, & Amgoud (2003)

• Arguments are modelled as a tuple S = (H, c) where

• c = the claim or conclusion of the argument - i.e. the correspondence

• H = the support, such that the claim holds

• Attacks occur as:

• rebuts - if the claim of one argument negates another (e.g. c1 ≣ ¬c2)

• undercut - if the claim of one argument negates an element of the support S

Finding an injective alignment?

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool

• Agents represent the correspondences symbolically within a (potentially inconsistent) stratified knowledge base 𝛴.• A propositional symbol is used to represent the correspondences

• The knowledge base is stratified into disjoint sets corresponding to the weights of the correspondences

• Correspondences accepted by both agents are added to the sets, based on their weights

Stratifying Arguments

14

⌃0.7 = {i}⌃0.65 = {j}⌃0.5 = {k}⌃0.45 = {m}

p c c

i hpublication, paper,⌘i 0.70

j harticle, paper,⌘i 0.65

k harticle, submittedPaper,⌘i 0.5

l harticle, reviewedPaper,⌘i 0.45

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool

• Objections within the dialogue are modelled within the knowledge bases.• If a correspondence c was raised as an objection to another

c’ (due to an ambiguity), then they must be disjoint:

Modelling Objections

15

p c c

i hpublication, paper,⌘i 0.70

j harticle, paper,⌘i 0.65

k harticle, submittedPaper,⌘i 0.5

l harticle, reviewedPaper,⌘i 0.45

⌃0.7 = {i, i $ ¬j}⌃0.65 = {j, j $ ¬i, j $ ¬k, j $ ¬l}⌃0.5 = {k, k $ ¬j, k $ ¬l}⌃0.45 = {l, l $ ¬j, l $ ¬k}

hAlice, object, hpublication, paper, 0.7i, harticle, paper, 0.65iihBob, object, harticle, paper, 0.65i, harticle, submittedPaper, 0.5iihBob, object, harticle, paper, 0.65i, harticle, reviewedPaper, 0.45iihAlice, object, harticle, submittedPaper, 0.5i, harticle, reviewedPaper, 0.45ii

if ambiguous(c, c0), then (c =) ¬c0) ^ (c0 =) ¬c)

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool

Modelling Objections

16

p c c

i hpublication, paper,⌘i 0.70

j harticle, paper,⌘i 0.65

k harticle, submittedPaper,⌘i 0.5

l harticle, reviewedPaper,⌘i 0.45

1. a1 undercuts a2 2. As a consequence

2.1. a1 undercuts a3 2.2. a1 undercuts a5

a1 = ({i, i $ ¬j},¬j) a5 = ({j, j $ ¬l},¬l)a2 = ({j, j $ ¬i},¬i) a6 = ({l, l $ ¬j},¬j)a3 = ({j, j $ ¬k},¬k) a7 = ({k, k $ ¬l},¬l)a4 = ({k, k $ ¬j},¬j) a8 = ({l, l $ ¬k},¬k)

3. As a7 is not attacked by a3: 3.1. a7 undercuts a8 3.2. a7 undercuts a6

Solution: publication article author

submittedPaper reviewedPaper paper editor

a1, a4, a7 = {i, k, i $ ¬j, k $ ¬j, i $ ¬l}AL = {hpublication, paper,=i,

harticle, submittedPaper,=i}.

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool

Conclusions• Developed a formal Inquiry Dialogue that supports the sharing of

ontological correspondences between agents• Only those correspondences relating to the agents working ontology are

aligned, thus avoiding unnecessary alignment

• Implemented a full version of the dialogue for evaluation• The resulting alignment performs significantly better in most cases than the

average performance of other approaches, when tested with a reference alignment (AAMAS ‘2014)

• Preference based argumentation can be used as a modelling paradigm for objections• Currently investigating how to encode other ontological constraints

propositionally within this framework

17

Using preferences in Negotiations over Ontological CorrespondencesTerry Payne University of Liverpool

Questions

18

For other papers on this and our other related work:

http://www.csc.liv.ac.uk/~trp/Knowledge-Based-Agents.html