Abstract
Belief merging has been an active research field with many important applications. The major approaches for the belief merging problems, considered as arbitration processes, are based on the construction of the total pre-orders of alternatives using distance functions and aggregation functions. However, these approaches require that all belief bases are provided explicitly and the role of agents, who provide the belief bases, are not adequately considered. Therefore, the results are merely ideal and difficult to apply in the multi-agent systems. In this paper, we approach the merging problems from other point of view. Namely, we treat a belief merging problem as a game, in which rational agents participate in a negotiation process to find out a jointly consistent consensus trying to preserve as many important original beliefs as possible. To this end, a model of negotiation for belief merging is presented, a set of rational and intuitive postulates to characterize the belief merging operators are proposed, and a representation theorem is presented.
Similar content being viewed by others
Notes
In related works, the given belief bases may be assumed to be consistent or not. In this work, we do not require that the belief bases be consistent.
More clearly, belief merging processes also consider the number of identical belief bases as well as the equivalent beliefs in the beliefs bases.
These situations are known as drowning effect, a common effect in the merging prioritized belief bases.
ℝ* is the set of non-negative real numbers.
The order of propositions represented in a possible world is peace, resignation, protest, army, publicity, and help, respectively.
Because agents are assumed to be intelligent, they know that the earlier their beliefs are submitted, the higher chance these beliefs are accepted as common beliefs. They also know that submitting a belief which is the consequence of common belief set is not necessary and a belief that causes conflict with the common belief set will never be accepted.
References
Baeza-Yates RA, Ribeiro-Neto BA (1999) Modern information retrieval. ACM Press Addison-Wesley
Baral C, Kraus S, Minker J (1991) Combining multiple knowledge bases. IEEE Trans Knowl Data Eng 3:208–220
Baral C, Kraus S, Minker J, Subrahmanian VS (1991) Combining knowledge bases consisting of first order theories. In: Proceedings of the 6th international symposium on methodologies for intelligent systems, ISMIS ’91. Springer-Verlag, London, UK, pp 92–101
Benferhat S, Dubois D, Kaci S, Prade H (2002) Possibilistic merging and distance-based fusion of propositional information. Ann Math Artif Intell 34:217–252
Bloch I, Lang J (2002) Towards mathematical morpho-logics. Physica-Verlag GmbH, Heidelberg, Germany, pp 367–380
Booth R (2001) A negotiation-style framework for non-prioritised revision. In: Proceedings of the 8th conference on theoretical aspects of rationality and knowledge, TARK ’01. Morgan Kaufmann Publishers Inc, San Francisco, CA, USA, pp 137–150.
Booth R (2006) Social contraction and belief negotiation. Inf Fusion 7:19–34
de Amo S, Carnielli WA, Marcos J (2002) A logical framework for integrating inconsistent information in multiple databases. In: Eiter T, Schewe K-D (eds) Foundations of information and knowledge systems, second international symposium, FoIKS 2002. Lecture notes in computer science, vol 2284. Springer, Salzau Castle, Germany, pp 67–84
Delgrande JP, Jin Y (2012) Parallel belief revision: revising by sets of formulas. Artif Intell 176(1):2223–2245
Everaere P, Konieczny S, Marquis P (2008) Conflict-based merging operators. In: Principles of knowledge representation and reasoning: proceedings of the eleventh international conference, KR 2008. AAAI Press, Sydney, Australia, pp 348–357
Konieczny S (2000) On the difference between merging knowledge bases and combining them. In: Anthony BS, Cohn G, Giunchiglia F (eds) Principles of knowledge representation and reasoning proceedings of the seventh international conference, KR 2000. Morgan Kaufmann Publishers, Breckenridge, Colorado, USA, pp 135–144
Konieczny S (2004) Belief base merging as a game. J Appl Non-Class Log 14(3):275–294
Konieczny S, Pérez RP (1999) Merging with integrity constraints. In: Symbolic and quantitative approaches to reasoning and uncertainty, European conference, ECSQARU’99. Lecture notes in computer science. Springer, London, UK, pp 233–244
Konieczny S, Pérez RP (2002) Merging information under constraints: a logical framework. J Log Comput 12(5):773–808
Konieczny S, Lang J, Marquis P (2004) Da2 merging operators. Artif Intell 157:49–79
Konieczny S, Lang J, Marquis P (2005) Reasoning under inconsistency: the forgotten connective. In: Proceedings of the nineteenth international joint conference on artificial intelligence, IJCAI-05. Professional Book Center, Edinburgh, Scotland, UK, pp 484–489
Levi I (1977) Subjunctives, dispositions and chances. Synthese 34:423–455
Lew MS, Sebe N, Djeraba C, Jain R (2006) Content-based multimedia information retrieval: state of the art and challenges. ACM Trans Multimedia Comput Commun Appl 2(1):1–19
Liberatore P, Schaerf M (1998) Arbitration (or how to merge knowledge bases). IEEE Trans Knowl Data Eng 10:76–90
Lin J (1996) Integration of weighted knowledge bases. Artif Intell 83:363–378
Olfati-Saber R, Fax JA, Murray RM (2007) Consensus and cooperation in networked multi-agent systems. Proc IEEE 95(1):215–233
Pattichis CS, Pattichis MS, Micheli-Tzanakou E (2001) Medical imaging fusion applications: an overview. In: The thirty-fifth asilomar conference on signals, systems & computers, vol 2, pp 1263–1267
Qi G, Liu W, Bell DA (2006) Merging stratified knowledge bases under constraints. In: Proceedings of the 21st national conference on artificial intelligence, vol 1. AAAI Press, pp 281–286
Reddy MP, Prasad BE, Reddy PG, Gupta A (1994) A methodology for integration of heterogeneous databases. IEEE Trans Knowl Data Eng 6(6):920–933
Ren W, Beard R, Atkins E (2005) A survey of consensus problems in multi-agent coordination. In: Proceedings of American control conference, vol 3, pp 1859–1864
Revesz PZ (1995) On the semantics of arbitration. Int J Algebra Comput 7:133–160
Serra J (1983) Image analysis and mathematical morphology. Academic Press Inc, Orlando, FL, USA
Sliwko L, Nguyen NT (2007) Using multi-agent systems and consensus methods for information retrieval in internet. Int J Intell Inf Database Syst 1(2):181–198
Tran TH, Vo QB, Kowalczyk R (2011) Merging belief bases by negotiation. In: König A, Dengel A, Hinkelmann K, Kise K, Howlett RJ, Jain LC (eds) Knowledge-based and intelligent information and engineering systems - 15th international conference, KES 2011. Lecture notes in computer science, vol 6881. Springer, Kaiserslautern, Germany, pp 200–209
Zhang D (2010) A logic-based axiomatic model of bargaining. Artif Intell 174:1307–1322
Acknowledgements
This research was partially supported by Polish Ministry of Science and Higher Education under grant no. N N519 407437 (2009-2012).
Author information
Authors and Affiliations
Corresponding author
Appendix—Proofs
Appendix—Proofs
Proposition 1
ΔN satisfies for IC0, IC1, IC2, IC3, IC7, IC8; it does not satisfy properties IC4, IC5, and IC6.
Proof
For the sake of proof, we recall about the axiomatic approach of Social Contraction function as follows:
Given a belief vector \(\overrightarrow{S} \in \mathcal{B}^n\) and a function \(f: \mathcal{B}^n \rightarrow \mathcal{B}^n\). We have:
Theorem 2
[7] Function f is a Social Contraction function, i.e. f = f N , if and only if f satisfies the following postulates:
-
SC1. \(\overrightarrow{S} \subseteq f(\overrightarrow{S})\) ;
-
SC2. \(f(\overrightarrow{S})\) is consistent;
-
SC3. if \(\overrightarrow{S}\) is consistent, then \(\overrightarrow{S} = f(\overrightarrow{S})\) ;
-
SC4. \(f_n(\overrightarrow{S}) = S_n\) .
Now, we refer to these postulates to prove Proposition 1 as follows:
-
IC0 is satisfied, it means that \(\Delta^N(\overrightarrow{S}) \subseteq S_n\). Indeed, by SC4 we have \(f_n(\overrightarrow{S}) = S_n\). Therefore, by (1) we have:
$$ \Delta^N(\overrightarrow{S}) =\bigcap\nolimits_{i=1}^{n}{f_i(\overrightarrow{S})} = \bigcap\nolimits_{i=1}^{n-1}{f_i(\overrightarrow{S})} \bigcap f_n(\overrightarrow{S}) = \bigcap\nolimits_{i=1}^{n-1}{f_i(\overrightarrow{S})} \bigcap S_n. $$It implies that \(\Delta^N(\overrightarrow{S}) \subseteq S_n\).
-
By SC2 we have \(f(\overrightarrow{S})\) is consistent, it means that \(\bigcap_{i=1}^{n}{f_i(\overrightarrow{S})} \neq \emptyset\). Thus, by (1) we have \(\Delta^N(\overrightarrow{S}) \neq \emptyset\), it implies that \(\Delta^N(\overrightarrow{S})\) is consistent, so IC1 is satisfied.
-
By SC3, if \(\overrightarrow{S}\) is consistent then \(f(\overrightarrow{S}) = \overrightarrow{S}\), therefore by (1), we have \(\Delta^N(\overrightarrow{S}) =\bigcap_{i=1}^{n}{f_i(\overrightarrow{S})} = \bigcap_{i=1}^{n}{S_i}\). Thus, IC2 is satisfied.
-
IC3 is obvious.
-
IC7 and IC8 are satisfied. Truly, we consider \(S_{\mu}\in \mathcal{B}\) and \(\overrightarrow{S'}\) from \(\overrightarrow{S}\) by replacing S N by S′ N = S N ∩ S μ . IC7 and IC8 are understood equivalently as follows:
-
IC7′. \(\Delta^N(\overrightarrow{S}) \bigcap S_{\mu} \subseteq \Delta^N(\overrightarrow{S'})\).
-
IC8′. if \(\Delta^N(\overrightarrow{S}) \bigcap S_{\mu} \neq \emptyset\), then \( \Delta^N(\overrightarrow{S'}) \subseteq \Delta^N(\overrightarrow{S}) \bigcap S_{\mu} \).
Firstly, if \(\Delta^N(\overrightarrow{S}) \bigcap S_{\mu} = \emptyset\) then IC7′ is trivially satisfied. Then, we consider the case \(\Delta^N(\overrightarrow{S}) \bigcap S_{\mu} \neq \emptyset\), we need to prove that both IC7′ and IC8′ are satisfied, it is equivalent to \(\Delta^N(\overrightarrow{S'}) = \Delta^N(\overrightarrow{S}) \bigcap S_{\mu}\). Indeed, by (1), we have:
$$ \begin{array}{llll} \Delta^N(\overrightarrow{S'}) & = & \bigcap_{i=1}^{n}{f_i(\overrightarrow{S'})} &\\ & = & \bigcap_{i=1}^{n-1}{f_i(\overrightarrow{S'})} \bigcap S'_{N} &\textrm{(by $SC4$, we have } f_n(\overrightarrow{S'})= S'_{N})\\ &=& \bigcap_{i=1}^{n-1}{f_i(\overrightarrow{S'})} \bigcap (S_{N} \bigcap S_{\mu})& \textrm{(because of } S'_{N} = S_{N} \bigcap S_{\mu}) \\ &=& \bigcap_{i=1}^{n-1}{f_i(\overrightarrow{S})} \bigcap (S_{N} \bigcap S_{\mu})&(because~S_i= S'_i, \forall i \in [1..n-1]) \\ &=& \bigcap_{i=1}^{n}{f_i(\overrightarrow{S})} \bigcap S_{\mu} &\textrm{(by $SC4$, we have } S_{N} = f_n(\overrightarrow{S}))\\ &=& \Delta^N(\overrightarrow{S}) \bigcap S_{\mu}& \mbox{(by (\ref{DN:eq}))} \end{array} $$ -
□
Proposition 2
If F is a Simultaneous Concession Solution, then Δ G satisfies properties IC1, IC2, IC5, IC7, and IC8; it does not satisfy properties IC0, IC3, IC4, and IC6.
Proof
In the original work, the author does not mention about the integrity constraint, but in the spirit of the work, we can understand the integrity constraint as a formula μ added in the highest priority level of the demand set, which has the biggest height of the hierarchy. The merging operators are modified as \(\Delta^G(G) = \bigcup_{a_i\in\mathcal{A}} f_i(G)\bigcup\{\mu\)}. For the convenience of proof, we remind the following result from original work:
Definition 16
[30] Given a belief profile \(G=(\{(X_i, \succcurlyeq_i)|{a_i\in \mathcal{A}}\},\mu) \in g^{\mathcal{A},\mathcal{L}}\), belief profile \(G'=(\{(X'_i, \succcurlyeq'_i)|{a_i\in \mathcal{A}}\},\mu)\in g^{\mathcal{A},\mathcal{L}}\) is a max proper sub-game of G, denoted \(G' \sqsubset_{\max} G\) if for all \(a_i\in \mathcal{A}\), we have:
-
1.
X′ i ⊂ X i and \(\not\exists X\) such that X′ i ⊂ X ⊂ X i ;
-
2.
\(\phi \in X'_i, \psi \in X_i, \psi \succcurlyeq_i \phi\) implies ψ ∈ X′ i ;
-
3.
\(\succcurlyeq'_i = \succcurlyeq_i \bigcap (X'_i \times X'_i)\).
Theorem 3
[30] Given belief profile \(G=(\{(X_i, \succcurlyeq_i)|{a_i\in \mathcal{A}}\},\mu) \in g^{\mathcal{A},\mathcal{L}}\) , function \(f:g^{\mathcal{A},\mathcal{L}}\rightarrow \mathcal{O}(G)\) is a Simultaneous Concession Solution, i.e f = F, if and only if it satisfies the following axioms:
-
Axiom1. \(\bigcup_{a_i\in \mathcal{A}}f_i(G)\) is consistent.
-
Axiom2. \(\forall a_i\in \mathcal{A}, \phi \in f_i(G), \psi \in X_i\) and \(\psi \succcurlyeq_i \phi\) implies ψ ∈ f i (G).
-
Axiom3. if \(\bigcup_{a_i\in \mathcal{A}} X_i\) is consistent, then \(\forall a_i\in \mathcal{A}(f_i(G) = X_i)\) .
-
Axiom4. if \(\exists a_i\in \mathcal{A}(f_i(G) = \emptyset)\) , then \(\forall a_i\in \mathcal{A}(f_i(G) = \emptyset)\) .
-
Axiom5. if \(G' \sqsubset_{\max} G\) and \(\bigcup_{a_i\in \mathcal{A}}X_i\) is inconsistent, then f(G) = f(G′).
Now, we refer to the these axioms to prove Proposition 2 as follows:
-
IC0 is not satisfied. Indeed, suppose that for some \(a_i\in \mathcal{A}\), f i (G) = ∅, then by Axiom4 f i (G) = ∅ for all \(a_i\in \mathcal{A}\), thus by (2) we have ΔG(G) = ∅, it implies that \(\Delta^G(G) \nvdash \mu\).
-
Axiom1 requires that f(G) is consistent, it means that ΔG(G) is also consistent by (2), thus IC1 is satisfied.
-
ByAxiom3, if \(\bigcup_{a_i\in \mathcal{A}} X_i\) is consistent, then \(\forall a_i\in \mathcal{A}(f_i(G) = X_i)\) if G is non-conflictive. Therefore, \(\Delta^G(G) =\bigcup_{i=1}^n f_i(G) = \bigcup_{i=1}^nX_i \bigcup \{\mu\}\), it implies that IC2 is satisfied.
-
IC4 is fail. Truly, let us consider bargaining game
$$ G = (\{(X_1,\succcurlyeq_1),(X_2,\succcurlyeq_2)\},\top), $$in which X 1 = {α, α→β} and \(\alpha\succcurlyeq_1\alpha \rightarrow \beta\), \(X_2 = \{\lnot\beta, \alpha \rightarrow \beta\}\) and \(\alpha \rightarrow \beta \succcurlyeq_2 \lnot\beta\). We have \(\Delta^G(G) = \{\alpha, \alpha \rightarrow \beta\} = X_1\), so \(\Delta^G(G)\vdash X_1\) and \(\Delta^G(G)\nvdash X_2\). It violates IC4.
-
By means of Axiom5, we have
$$\label{DZ:pr} \bigcup f(G_1\bigcup G_2)\subseteq(\bigcup{f(G_1)}) \bigcup (\bigcup{f(G_2)}) $$(4)and f(G 1),f(G 2), and f(G 1 ∪ G 2) are consistent (by Axiom1), thus IC5 is satisfied.
-
However, even if ( ∪ f(G 1)) ∪ ( ∪ f(G 2)) is consistent, but because of (4), IC6 is not satisfied.
-
Now, we prove that ΔG satisfies both IC7 and IC8. First, we consider a bargaining game G and two formulas μ and ψ as the integrity constraints. We have G′ and G′′ by respectively adding μ and \(\mu \land \psi\) into the highest priority level of the demand set, which has the biggest height of the hierarchy of G. We have the counters of IC7 and IC8 as follows:
-
IC7′′. \(\Delta^G(G') \land \psi \vdash \Delta^G(G'')\).
-
IC8′′. if \(\Delta^G(G') \land \psi\) is consistent then \(\Delta^G(G'') \vdash \Delta^G(G') \land \psi\).
If \(\Delta^G(G') \land \psi\) is inconsistent, IC7′′ is trivially satisfied. We consider the case when \(\Delta^G(G') \land \psi\) is consistent, then to prove IC7′′ and IC8′′ we need to prove \(\Delta^G(G') \land \psi = \Delta^G(G'')\). Indeed, we have ΔG(G′) = ( ∪ f(G)) ∪ {μ}, thus ψ is consistent with μ and ψ is consistent with ( ∪ f(G)) and obviously, ( ∪ f(G)) is consistent with μ. Therefore, ∪ f(G) is consistent with \((\mu\land \psi)\), so \(\Delta^G(G') \land \psi = \Delta^G(G) \land (\mu \land \psi) =\Delta^G(G'')\). It is clear that both IC7 and IC8 are satisfied.□
-
Proposition 3
If Δ M(G) is a Negotiation Belief Merging operator, it satisfies for IC0, IC1, IC2, IC7, IC8 and, it does not satisfy properties IC3, IC4, IC5, IC6
Proof
-
Because of (3), it is easy to see that IC0 is satisfied.
-
Suppose that μ is consistent, we also have \(\neg \mu\) consistent. Thus, CO implies that \(\bigcup_{a_i\in\mathcal{A}}f_i(G)\) is consistent and by (3) we also have ΔM(G) consistent. It means that IC1 is satisfied.
-
We assume that \(\bigcup_{a_i\in\mathcal{A}}X_i \bigcup \{\mu\}\) is consistent. We have
$$ MAXCONS(\bigcup\nolimits_{a_i\in\mathcal{A}}X_i,\mu) =\{\bigcup\nolimits_{a_i\in\mathcal{A}}X_i\bigcup \{\mu\}\}. $$Therefore, by CP we have
$$ \bigcup\nolimits_{a_i\in\mathcal{A}}X_i\bigcup \{\mu\} \subseteq \bigcup\nolimits_{a_i\in\mathcal{A}}f_i^M(G)\bigcup \{\mu\}. $$(*)On the other hand, by IR we have
$$ \bigcup\nolimits_{a_i\in\mathcal{A}}f_i^M(G)\bigcup \{\mu\} \subseteq \bigcup\nolimits_{a_i\in\mathcal{A}}X_i\bigcup \{\mu\}. $$(**)Thus, from (*) and (**) we have
$$ \bigcup\nolimits_{a_i\in\mathcal{A}}f_i^M(G)\bigcup \{\mu\} = \bigcup\nolimits_{a_i\in\mathcal{A}}X_i\bigcup \{\mu\} $$(***)Lastly, from (3) and (***), we have
$$ \Delta^M(G) = \bigcup\nolimits_{a_i\in\mathcal{A}}X_i \bigcup \{\mu\}. $$It means that IC2 is satisfied.
-
We consider a belief profile \(G=(\{(X_i, \succcurlyeq_i)|{a_i\in \mathcal{A}}\},\mu)\) and a formula μ 1. If \(\Delta^M(G)\land \mu_1\) is inconsistent, IC7 is obviously satisfied. We consider the other case when \(\Delta^M(G)\land \mu_1\) is consistent. Let \(G'=(\{(X_i, \succcurlyeq_i)|{a_i\in \mathcal{A}}\},\mu\land\mu_1)\). In order to prove that both IC7 and IC8 are satisfied, we need to show that \(\Delta^M(G)\land \mu_1 \equiv \Delta^M(G')\). Indeed, because \(\Delta^M(G)\land \mu_1\) is consistent, μ and μ 1 are consistent, and f i (G) and μ 1 are also consistent. Therefore, \(f_i(G') \equiv f_i(G)\land \mu_1 \), and together (3) we have:
$$\begin{array}{rll} \bigcup\nolimits_{a_i\in \mathcal{A}} f_i(G')&\equiv & \bigcup\nolimits_{a_i \in \mathcal{A}} (f_i(G)\land \mu_1)\\ &\equiv& \bigcup\nolimits_{a_i\in \mathcal{A}} f_i(G)\land \mu_1. \end{array}$$Consequently, we have:
$$\begin{array}{rll} \Delta^M(G') &=& \bigcup\nolimits_{a_i\in \mathcal{A}} f_i(G') \land (\mu \land \mu_1)\\ &\equiv & \left(\bigcup\nolimits_{a_i \in \mathcal{A}}f_i(G)\land \mu\right) \land \mu_1\\ & = & \Delta^M(G) \land \mu_1. \end{array}$$Thus, we have \(\Delta^M(G)\land \mu_1 \equiv \Delta^M(G')\) if \(\Delta^M(G)\land \mu_1\) is consistent.
Generally, IC7 and IC8 are satisfied.
-
IC4 is fail. Truly, let us consider bargaining game \(G = ((X_1,\succcurlyeq_1),(X_2,\succcurlyeq_2),\top)\), in which X 1 = {α, α→β} and \(\alpha\succcurlyeq_1\alpha \rightarrow \beta\), \(X_2 = \{\lnot\beta, \alpha \rightarrow \beta\}\) and \(\alpha \rightarrow \beta \succcurlyeq_2 \lnot\beta\). We have \(f^M(G) = \{\alpha, \alpha \rightarrow \beta\} = X_1\), so \(f^M(G)\vdash X_1\) and \(f^M(G)\nvdash X_2\). It violates IC4.□
Theorem 1
Negotiation Belief Merging solution f M satisfies properties IR , CO , CP , PO , and SY .
Proof
We assume to have a Negotiation Belief Merging solution f M for any belief merging negotiation model \(\mathcal{M}\). We need to show that f M satisfies properties IR, CO, CP, PO, and SY. With a initial negotiation S = (G,O), where \(G=(\{(X_i, \succcurlyeq_i)|{a_i\in \mathcal{A}}\},\mu)\in g^{\mathcal{A},\mathcal{L}}\) and \(O = \{\emptyset|a_i\in \mathcal{A}\}\), we have \(s^M(S) =\lambda = (S_0,\ldots,S_k)\).
-
For IR, we need to show that if \(S_k = (G, O^*)\) and f(G) = O * then \(f_i(G)\subseteq X_i\). Indeed, because of choice function c, an agent is chosen if she still has some belief to submit and by update function w, she only allows to submit the belief in her belief set. Therefore, an agent obtains the set of beliefs which does not exceed her initial belief set i.e. \(f_i(G)\subseteq X_i\).
-
Because the integrity constraint μ is put into the initial constructing common belief set and in each round of negotiation the chosen beliefs do not allow to jointly cause conflict with the common belief set, thus the solution f(G) is always jointly consistent with μ. Consequently, CO is satisfied.
-
In each round of negotiation, one of maximal consistent subsets of the current common belief set and submitting belief with the current common belief set as integrity constraint is chosen. Thus, both CP and PO are satisfied.
-
Because the possible outcome is gradually constructed by based on the beliefs and the preferences of agents on these beliefs without the identifier of agents, therefore SY is satisfied.□
Proposition 4
if Δ M(G) is a Negotiation Belief Merging operator,
Proof
Because ΔG(G) has to satisfy Axiom1 and Axiom2, it is easy to see that ΔM(G) ⊢ ΔG(G) by CO and CP. □
Rights and permissions
About this article
Cite this article
Tran, T.H., Nguyen, N.T. & Vo, Q.B. Axiomatic characterization of belief merging by negotiation. Multimed Tools Appl 65, 133–159 (2013). https://doi.org/10.1007/s11042-012-1136-7
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11042-012-1136-7