An Imprecise Probability Approach for Abstract Argumentation based on Credal Sets
Mariela Morveli-Espinoza, Juan Carlos Nieves, Cesar Augusto Tacla
AA N I MPRECISE P ROBABILITY A PPROACH FOR A BSTRACT A RGUMENTATION BASED ON C REDAL S ETS
Mariela Morveli-Espinoza
Graduate Program in Electrical and Computer Engineering (CPGEI),Federal University of Technology - Paraná (UTFPR), Curitiba - Brazil [email protected]
Juan Carlos Nieves
Department of Computing Science of Umeå University,Umeå - Sweden [email protected]
Cesar Augusto Tacla
Graduate Program in Electrical and Computer Engineering (CPGEI),Federal University of Technology - Paraná (UTFPR), Curitiba - Brazil [email protected] A BSTRACT
Some abstract argumentation approaches consider that arguments have a degree of uncertainty,which impacts on the degree of uncertainty of the extensions obtained from a abstract argumentationframework (AAF) under a semantics. In these approaches, both the uncertainty of the arguments andof the extensions are modeled by means of precise probability values. However, in many real lifesituations the exact probabilities values are unknown and sometimes there is a need for aggregatingthe probability values of different sources. In this paper, we tackle the problem of calculating thedegree of uncertainty of the extensions considering that the probability values of the arguments areimprecise. We use credal sets to model the uncertainty values of arguments and from these credalsets, we calculate the lower and upper bounds of the extensions. We study some properties of thesuggested approach and illustrate it with an scenario of decision making. K eywords abstract argumentation · imprecise probability · uncertainty · credal sets The AAF that was introduced in the seminal paper of Dung [1] is one of the most significant developments in thecomputational modelling of argumentation in recent years. The AAF is composed of a set of arguments and a binaryrelation encoding attacks between arguments. Some recent approaches on abstract argumentation assign uncertaintyto the elements of the AAF to represent the degree of believe on arguments or attacks. Some of these works assignuncertainty to the arguments (e.g., [2][3][4][5][6][7][8][9]), others to the attacks (e.g., [6]), and others to both argumentsand attacks (e.g., [10]). These works use precise probability approaches to model the uncertainty values. However,precise probability approaches have some limitations to quantify epistemic uncertainty, for example, to represent groupdisagreeing opinions. These can be better represented by means of imprecise probabilities, which use lower and upperbounds instead of exact values to model the uncertainty values.For a better illustration of the problem, consider a discussion between a group of medicine students (agents). Thediscussion is about the diagnose of a patient. In this context, arguments represent the student’s opinions and the attacksrepresent the disagreements between such opinions. Figure 1 shows the argumentation graph where nodes representarguments and edges the attacks between arguments. In the graph, two arguments represent two possible diagnoses a r X i v : . [ c s . A I] S e p amely measles and chickenpox, there is an argument against measles and two arguments against chickenpox, and thereare three arguments that have no attack relations with the rest of arguments. A = The patient has measlesC = She has blistersE = The patientG = The patient has feverH = The patient’s temperature is 39° B= The patient has chickenpoxD = She only has small red spotspatient has brown eyesF = The patient wasvaccinated for chickenpox
Figure 1: Argumentation graph for the discussion about the diagnose of a patient.Suppose that each opinion – i.e., argument – has a probability value between 0 and 1 that represents the degree ofbelieve of each student. Since there is more than one opinion, this means that each argument has associated a set ofprobability values. Thus, we cannot model these degrees of believe by means of an unique probability value (preciseprobability value), what we need is to represent a range of the possible degrees of believe.To the best of our knowledge, there is no work that models the uncertainty values of arguments by using an impreciseprobability approach. Therefore, we aim to propose an approach for abstract argumentation in which the uncertainty ofthe arguments is modeled by an imprecise probability value. Thus, the research questions that are addressed in thispaper are:1. How to model the imprecise uncertainty values of arguments?2. In abstract argumentation, several semantics have been proposed, which return sets of arguments – calledextensions – whose basic characteristic is that these arguments do not attack each other, i.e. they are consistent.The fact that the arguments that belong to an extension are uncertain, causes that such extension also has adegree of uncertainty. How to calculate the lower and upper bounds of extensions?In addressing the first question, we use credal sets to model the uncertainty values of arguments. Regarding the secondquestion, we base on the credal sets of the arguments to calculate the uncertainty values of extensions obtained undera given semantics. These values are represented by lower and upper bounds. The way to aggregate the credal setsdepends on a causal relation between the arguments.Next section gives a brief overview on credal sets and abstract argumentation. In Section 3, we present the AAF basedon credal sets and the causality graph concept, which are the base for the calculation of the upper and lower bounds ofextension. This calculation is tackled in Section 4. We study the main properties of our approach in Section 5. Relatedwork is presented in Section 6. Finally, Section 7 is devoted to conclusions and future work.
In this section, we revise concepts of credal sets and abstract argumentation.
Assume that we have a finite set of events E = { E , ..., E n } and a probability distribution p on this set, where p is amapping p : E → [0 , . According to Levi [11], a closed convex set of probability distributions p is called a credalset. Given an event E , a credal set for E – denoted K ( E ) – is a set of probability distributions about this event and K = { K ( E ) , ..., K ( E n ) } denotes a set of all credal sets. Every credal set has the same number of elements. In thiswork, we assume that the cardinality of the credal sets of K is the same (let us denote it by m ); moreover, we assumethat p i ( E ) denotes the suggested probability of the agent i w.r.t the event E such that ≤ i ≤ m and E ∈ E . Given acredal set K ( E ) , the lower and upper bounds for event E are determined as follows: Lower probability: P ( E ) = inf { p ( E ) : p ( E ) ∈ K ( E ) } (1) Upper probability: P ( E ) = sup { p ( E ) : p ( E ) ∈ K ( E ) } l events { E , ..., E l } ⊆ E and their respective credal sets K ( E ) = { p ( E ) , ..., p m ( E ) } , ..., K ( E l ) = { p ( E l ) , ..., p m ( E l ) } . If { E , ..., E l } are independent events, the lower and upper probabilities are defined as follows: P ( { E , ..., E l } ) = min ≤ j ≤ m { (cid:81) i ≤ li =1 p j ( E i ) } where p j ∈ K ( E i ) (2) P ( { E , ..., E l } ) = max ≤ j ≤ m { (cid:81) i ≤ li =1 p j ( E i ) } On the other hand, when the independence relation is not assumed, the first step is to calculate a credal set for { E , ..., E l } as follows: K ( { E , ..., E l } ) = { p E | p E = min ≤ j ≤ m { p j ( E ) , ...p j ( E l ) }} where p j ( E i ) ∈ K ( E i ) (3)Based on K ( { E , ..., E l } ) , we obtain the lower and upper probabilities: P ( { E , ..., E l } ) = min ( K ( { E , ..., E l } )) (4) P ( { E , ..., E l } ) = max ( K ( { E , ..., E l } )) Example 1
Let { E , E , E } be three events and K ( E ) = { p ( E ) , p ( E ) , p ( E ) } , K ( E ) = { p ( E ) , p ( E ) ,p ( E ) } , and K ( E ) = { p ( E ) , p ( E ) , p ( E ) } their respective credal sets. Next table shows the values of theprobability distributions for each event. E E E p p p E , E , and E are independent, the lower and upper probabilities of ( E , E , E ) are calculated asfollows: P ( E , E , E ) = min { . × . × . , . × . × . , . × . × . } = min { . , . , . } ;hence P ( E , E , E ) = 0 . and P ( E , E , E ) = max { . , . , . } = 0 . .On the other hand, if we assume that E , E , and E are not independent, then the lower and upper probabili-ties are calculated as follows: K ( E , E , E ) = { min { . , . , . } , min { . , . , . } , min { . , . , . }} = { . , . , . } . Thus, P ( E , E , E ) = 0 . and P ( E , E , E ) = 0 . . In this subsection, we will recall basic concepts related to the AAF defined by Dung [1], including the notion ofacceptability and the main semantics.
Definition 1 (Abstract AF)
An abstract argumentation framework AF is a tuple AF = (cid:104) ARG , R (cid:105) where ARG is afinite set of arguments and R is a binary relation R ⊆ ARG × ARG that represents the attack between two arguments of
ARG , so that ( A, B ) ∈ R denotes that the argument A attacks the argument B . Next, we introduce the concepts of conflict-freeness, defense, admissibility and the four semantics proposed by Dung[1].
Definition 2 (Argumentation Semantics)
Given an argumentation framework AF = (cid:104) ARG , R (cid:105) and a set E ⊆ ARG : • E is conflict-free if ∀ A, B ∈ E , ( A, B ) (cid:54)∈ R . • E defends an argument A iff for each argument B ∈ ARG , if ( B, A ) ∈ R , then there exist an argument C ∈ E such that ( C, B ) ∈ R . • E is admissible iff it is conflict-free and defends all its elements. • A conflict-free E is a complete extension iff we have E = { A | E defends A } . • E is a preferred extension iff it is a maximal (w.r.t set inclusion) complete extension. • E is a grounded extension iff it is the smallest (w.r.t set inclusion) complete extension. • E is a stable extension iff E is conflict-free and ∀ A ∈ ARG and A (cid:54)∈ E , ∃ B ∈ E such that ( B, A ) ∈ R . In this article, there is a set of agents that give their opinions (degrees of belief) regarding each argument in
ARG bymeans of probability distributions. The set of arguments can be compared with the events of set E ; hence, we can say3hat E = ARG . The number of agents that give their opinions determines the cardinality of credal sets. Thus, given m agents and an argument A ∈ ARG , the credal set for A is represented by K ( A ) = { p ( A ) , ..., p m ( A ) } . Finally, K denotes all the credal sets of the arguments in ARG . In this section, we present the definitions of AAF based on credal sets and causality graph. These concepts are importantfor the calculation of the lower and upper bounds of extensions.We use credal sets to model the opinions (degrees of belief) of a set of agents about a set of arguments. Thus, eachargument in an AAF has associated a credal set, which contains probability distributions that represent the opinions ofthe agents about it.
Definition 3 (Credal Abstract Argumentation Framework)
An AAF based on credal sets is a tuple AF CS = (cid:104) ARG , R , K , f CS (cid:105) where (i) ARG is a set of arguments, (ii) R is the attack relation presented in Definition 1, (iii) K is a set of credal sets, and (iv) f CS : ARG → K maps a credal set for each argument in ARG . Recall that the cardinality of every credal set depends on the number of agents. Since all the agents give their opinionsabout all the arguments, all the credal sets have the same number of elements.
Definition 4 (Agent’s opinions)
Let AF CS = (cid:104) ARG , R , K , f CS (cid:105) be a Credal AAF and AGT = { ag , ..., ag m } a set ofagents. The opinion p i of an agent ag i (for ≤ i ≤ m ) is ruled as follows:1. If A ∈ ARG , there is p i ( A ) ∈ K ( A ) where K ( A ) ∈ K .2. ∀ A ∈ ARG , ≤ p i ( A ) ≤ . Regarding the probability values given to the arguments, it is important to consider the notion of rational probabilitydistribution given in [5]. According to Hunter [5], if the degree of belief in an argument is high, then the degree ofbelief in the arguments it attacks is low. Thus, a probability function p is rational for an AF CS iff for each ( A, B ) ∈ R ,if p ( A ) > . then p ( B ) ≤ . where p ( A ) ∈ K ( A ) and p ( B ) ∈ K ( B ) . Example 2
Consider that
AGT = { ag , ag , ag , ag } . The Credal AAF for the example given in Introduction is AF CS = (cid:104) ARG , R , K , f CS (cid:105) where:- ARG = { A, B, C, D, E, F, G.H } - R = { ( A, B ) , ( B, A ) , ( F, B ) , ( D, B ) , ( C, A ) } - K = { K ( A ) , K ( B ) , K ( C ) , K ( D ) , K ( E ) , K ( F ) , K ( G ) , K ( H ) } . The table below shows the credal set of eachargument- f CS ( A ) = K ( A ) , f CS ( B ) = K ( B ) , ..., f CS ( H ) = K ( H ) K ( A ) K ( B ) K ( C ) K ( D ) K ( E ) K ( F ) K ( G ) K ( H ) p p p p In a Credal AAF, besides the attack relation between the arguments, there may be a causality relation between them. Tomake this discussion more concrete, consider the following conflict-free sets: • {
G, E } : Having fever does not have to do with the eyes’ color of the patient and vice-verse, so there is norelation between these arguments. This means that they are independent from each other. • { A, G } and { A, F } : In both cases the arguments are related in some way. In the first case, having fever ( G ) isa symptom of (causes) measles ( A ) and in the second case, the fact that the patient is vaccinated for chickenpox( F ) causes that he may have measles and not chickenpox ( A ). Definition 5 (Causality Graph)
Let AF CS = (cid:104) ARG , R , K , f CS (cid:105) be a Credal AAF, a causality graph C is a tuple C = (cid:104) ARG , R CAU (cid:105) such that:i
ARG = ARG ← ∪ ARG → ∪ ARG ◦ is a set of arguments, i R CAU ⊆ ARG × ARG represents a causal relation between two arguments of
ARG (the existence of this relationdepends on the domain knowledge), such that ( A, B ) ∈ R CAU denotes that argument A causes argument B . Itholds that if ( A, B ) ∈ R , then ( A, B ) / ∈ R CAU and ( B, A ) / ∈ R CAU ,iii
ARG ← = { B | ( A, B ) ∈ R CAU } , ARG → = { A | ( A, B ) ∈ R CAU } , and ARG ◦ = { C | C ∈ ARG − ( ARG ← ∪ ARG → ) } ,iv ARG ← and ARG → are not necessarily pairwise disjoint; however, ( ARG ← ∪ ARG → ) ∩ ARG ◦ = ∅ . Example 3
A causality graph for the Credal AAF of Example 2 is C = (cid:104){ A, B, C, D, E, F, G, H } , { ( D, A ) , ( F, A ) , ( H, A ) , ( G, A ) , ( H, G ) , ( G, B ) , ( C, B ) }(cid:105) (see Figure 2), where ARG ← = { A, B, G } , ARG → = { D, F, H, G, C } , and ARG ◦ = { E } . A GFD H B CG E
Figure 2: Causality graph for Example 3. Traced edges represent the causality relation.
Section 2 presented the definition of conflict-free ( cf ) and admissible ( ad ) sets and complete ( co ), preferred ( pr ),grounded ( gr ), and stable ( st ) semantics. Considering the causality graph, the arguments of an extension E x (for x ∈ { cf , ad , co , pr , gr , st } ) may belong to ARG → , ARG ← , or ARG ◦ . Depending on it, the calculation of the probabilisticlower and upper bounds of each extension is different. Thus, we can distinguish the following cases: (i) the extension isempty, (ii) the extension has only one argument, and (iii) the extension includes more than one argument. Definition 6 (Upper and Lower Bounds of Extensions)
Let AF CS = (cid:104) ARG , R , K , f CS (cid:105) be a Credal AAF, C = (cid:104) ARG , R CAU (cid:105) a causality graph, and E x ⊆ ARG (for x ∈ { cf , ad , co , pr , gr , st } ) an extension under semantics x . Thelower and uppers bounds of E x are obtained as follows:1. If E x = {} , then P ( E x ) = 0 and P ( E x ) = 1 , which denotes ignorance.2. If | E x | = 1 , then P ( E x ) = P ( A ) and P ( E x ) = P ( A ) s.t. A ∈ E x , where P ( A ) and P ( A ) are obtained by applying Equation (1).3. If | E x | > , then ( P ( E x ) , P ( E x )) = UL _ BOUNDS ( E x ) (see Algorithm 1).Consider the following functions:- f CAU ( A ) = { B | ( B, A ) ∈ R CAU ∪ f CAU ( B ) } - TOP _ CAU ( E x ) = { A | A ∈ ARG ← ∩ E x and ∀ B s.t. A ∈ f CAU ( B ) , B / ∈ E x } - FREE _ CAU ( E x ) = { A | A ∈ ARG → ∩ E x and ∀ B ∈ f CAU ( A ) , B / ∈ E x } TOP _ CAU and
FREE _ CAU consider only the arguments of E x and their causal relations restricted to E x . The formerreturns the arguments that are caused by any of the other argument in E x but do not cause other argument(s) in E x . Ifthere is an argument that belongs to ARG ← and ARG → in C but the argument(s) caused by it are not in E x , then it isreturned by TOP _ CAU . The latter returns the arguments that belong to
ARG → but whose caused arguments do not belongto extension E x . Example 4 (Cont. Example 2 considering the causality graph of Example 3). After applying the semantics presentedin Definition 2, we obtain that E CO = E PR = E GR = E ST = { C, E, F, D, H, G } . Since this extension has more than oneelement, the Algorithm 1 has to be applied: • We first evaluate the number of the caused arguments: E y ∩ ARG ← = { G } (for y ∈ { CO , PR , GR , ST } ), then we obtain TOP _ CAU ( E y ) = { G } and f CAU ( G ) = { H } ; hence, E G = { G, H } .At last, we calculate the credal set for E G by applying Equation (3): K ( E G ) = { . , . , , . } . • Next, we obtain those arguments that belong to the extension and that neither cause any other argument norare caused by any other argument:
ARG (cid:48)◦ = { E } . lgorithm 1 Function UL _ BOUNDS
Input:
An extension E x and a causality graph C = (cid:104) ARG , R CAU (cid:105)
Output: ( P ( E x ) , P ( E x )) if ( E x ∩ ARG ← ) (cid:54) = ∅ then ARG (cid:48)(cid:48)← = TOP _ CAU ( E x ) for i = 1 to | ARG (cid:48)(cid:48)← | do E iA = A ∪ ( f CAU ( A ) ∩ E x ) Calculate K ( E iA ) //Calculate the credal set for E iA by applying Equation (3) end for end if ARG (cid:48)◦ = E x ∩ ARG ◦ if ( E x ∩ ARG → ) (cid:54) = ∅ then ARG (cid:48)(cid:48)→ = FREE _ CAU ( E x ) end if //* — E x contains only one set of related arguments — *// if | ARG (cid:48)(cid:48)← | == 1 && ARG (cid:48)◦ == ∅ && ARG (cid:48)(cid:48)→ == ∅ then // Apply Equation (4) for obtaining the lower and upper bounds of E x P ( E x ) = P ( ARG (cid:48)(cid:48)← ) , P ( E x ) = P ( ARG (cid:48)(cid:48)← ) else //Apply Equation (2) for obtaining the lower and upper bounds of E x P ( E x ) = P ( (cid:83) i ≤| ARG (cid:48)(cid:48)← | i =1 E iA ∪ ARG (cid:48)◦ ∪ ARG (cid:48)(cid:48)→ ) , P ( E x ) = P ( (cid:83) i ≤| ARG (cid:48)(cid:48)← | i =1 E iA ∪ ARG (cid:48)◦ ∪ ARG (cid:48)(cid:48)→ ) end if return ( P ( E x , P ( E x ) • Then, we evaluate the number of causing arguments: E y ∩ ARG ← = { C, D, F } and we obtain FREE _ CAU ( E y ) = { C, D, F } . • Since E y do not contains only related arguments, we apply Equation (2) considering K ( E G ) , K ( E ) , K ( C ) , K ( D ) , and K ( F ) . • Finally, we obtain: ( P ( E y ) , P ( E y ) = [0 . , . .Let us also take some conflict-free sets: E CF = { A, F, H, D, E, G } , E CF = { A, F, H, D, G } , E CF = { B, C, G, H } , and E CF = { A } . The lower and upper bounds for these extensions are: ( P ( E CF ) , P ( E CF )) = [0 . , . , ( P ( E CF ) , P ( E CF ))= [0 . , . , ( P ( E CF ) , P ( E CF )) = [0 . , . , and ( P ( E CF ) , P ( E CF )) = [0 . , . . So far, we have calculated the lower and upper bounds of extensions obtained under a given semantics. The nextstep is to compare these bounds in order to determine an ordering over the extensions, which can be used to choosean extension that resolves the problem. In this case, the problem was making a decision about a possible diagnosisbetween two alternatives: measles or chickenpox. We are not going to tackle the problem of comparing and orderingthe extensions because it is out of the scope of this article; however, we can do a brief analysis taking into account theresult of the previous example. Arguments A and B represent each of the alternatives. The unique extension underany semantics y does not include any of the alternatives. On other hand, free-conflict sets E CF , E CF and E CF includeargument A and conflict free set E CF includes argument B . We can notice that there is a notorious difference betweenthe lower and upper bounds of E y and the lower and upper bounds of any of the other conflict-free sets. In fact, thelower and upper bounds of the conflict-free sets have a better location. This may indicate that lower and upper boundsof extensions that include one of the alternatives are better than others of extensions that do not include any of thealternatives. This in turn indicates that using uncertainty in AAF may improve the resolutions of some problems, whichwas demonstrated in [3] for precise uncertainty and it is showed in the example by using imprecise uncertainty. In this section, we study two properties of the proposed approach that guarantee (i) that the approach can be reduced tothe AAF of Dung and (ii) that the values of both the lower and upper bounds of the extensions are between 0 and 1.6iven a Credal AAF AF CS = (cid:104) ARG , R , K , f CS (cid:105) , AF CS is maximal if ∀ A ∈ ARG it holds that p i = 1 ( ≤ i ≤ m )where p i ∈ K ( A ) and K ( A ) = f CS ( A ) and AF CS is uniform if ≤ p i ≤ . Be maximal transforms an AF CS into astandard AAF of Dung, which means that every agent believes that every argument is believed without doubts. Thenext proposition shows that a AF CS can be reduced to an AAF that follows Dung’s definitions. Proposition 1
Given a credal AAF AF CS = (cid:104) ARG , R , K , f CS (cid:105) and a extension E x ( x ∈ { cf , ad , co , pr , gr , st } ) . If AF CS is maximal, then ∀ E x ⊆ ARG , P ( E x ) = P ( E x ) = 1 . Proof 1
Since AF CS is maximal, then ∀ A ∈ ARG , K ( A ) = { , ..., m } . In order to obtain the P ( E x ) and P ( E x ) ,Equations (1), (2), or (4) have to be applied. For Equation (1): the inf { , ..., } = sup { , ..., } = 1 . For Equation(2): ∀ A, (cid:81) { , ..., } = 1 , so the minimum and maximum of a set composed of 1s is always 1. The same happens withEquation (4). Proposition 2
Given a credal AAF AF CS = (cid:104) ARG , R , K , f CS (cid:105) and a extension E x ( x ∈ { cf , ad , co , pr , gr , st } ) . If AF CS is uniform, then ∀ E x ⊆ ARG , ≤ P ( E x ) ≤ and ≤ P ( E x ) ≤ . Proof 2
In order to obtain the P ( E x ) and P ( E x ) , Equations (1), (2), or (4) have to be applied. Since AF CS is uniform,we can say that the minimums (infimums) and maximums (supremums) are always between 0 and 1. Besides, the productof two numbers between 0 and 1 is always between 0 and 1. In this section, we present the most relevant works – to the best of our knowledge – that study probability and abstractargumentation. These works assign probability to the arguments, to the attacks, or to the extensions and all of them useprecise probabilistic approaches. Thus, as far as we know, we are introducing the first abstract argumentation approachthat employs imprecise probabilistic approaches.Dung and Thang [2] propose an AF for jury-based dispute resolution, which is based on probabilistic spaces, fromwhich are assigned probable weights – between zero and one – to arguments. In the same way, Li et al. [10] presentand extension of Dung’s original AF by assigning probabilities to both arguments and defeats. Hunter [3] bases on thetwo articles previously presented and focuses on studying the notion of probability independence in the argumentationcontext. The author also propose a set of postulates for the probability function regarding admissible sets and extensionslike grounded and preferred. Following the idea of using probabilistic graphs, the author assigns a probability value toattacks in [6].Thimm [4] focuses on studying probability and argumentation semantics. Thus, he proposes a probability semanticssuch that instead of extensions or labellings, probability functions are used to assign degrees of belief to arguments. Anextension of this work was published in [9]. Gabbay and Rodrigues [7] also focus on studying the extensions obtainedfrom an argumentation framework. Thus, they introduce a probabilistic semantics based on the equational approach toargumentation networks proposed in [12].
This work presents an approach for abstract argumentation under imprecise probability. We defined a credal AAF, inwhich credal sets are used to model the the uncertainty values of the arguments, which correspond to opinions of a setof agents about their degree of believe about each argument. We have considered that – besides the attack relation –there also exists a causality relation between the arguments of a credal AAF. Based on the credal sets and the causalityrelation, the lower and upper bounds of the extensions – obtained from a semantics – are calculated.We have done a brief analysis about the problem of comparing and ordering the extensions based on their lower andupper bounds; however, a more complete analysis and study are necessary. In this sense, we plan to follow this directionin our future work. We also plan to further study the causality relations, more specifically in the context of credalnetworks [13]. Finally, we want to study the relation of this approach with bipolar argumentation frameworks [14].
Acknowledgment
This work is partially founded by CAPES (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior).7 eferences [1] Phan Minh Dung. On the acceptability of arguments and its fundamental role in nonmonotonic reasoning, logicprogramming and n-person games.
Artificial intelligence , 77(2):321–357, 1995.[2] Phan Minh Dung and Phan Minh Thang. Towards (probabilistic) argumentation for jury-based dispute resolution.
International Conference on Computational Models of Argument , 216:171–182, 2010.[3] Anthony Hunter. Some foundations for probabilistic abstract argumentation.
International Conference onComputational Models of Argument , 245:117–128, 2012.[4] Matthias Thimm. A probabilistic semantics for abstract argumentation. In
European Conference on ArtificialIntelligence , volume 12, pages 750–755, 2012.[5] Anthony Hunter. A probabilistic approach to modelling uncertain logical arguments.
International Journal ofApproximate Reasoning , 54(1):47–81, 2013.[6] Anthony Hunter. Probabilistic qualification of attack in abstract argumentation.
International Journal ofApproximate Reasoning , 55(2):607–638, 2014.[7] Dov M Gabbay and Odinaldo Rodrigues. Probabilistic argumentation: An equational approach.
Logica Universalis ,9(3):345–382, 2015.[8] Régis Riveret, Dimitrios Korkinof, Moez Draief, and Jeremy Pitt. Probabilistic abstract argumentation: aninvestigation with boltzmann machines.
Argument & Computation , 6(2):178–218, 2015.[9] Matthias Thimm, Pietro Baroni, Massimiliano Giacomin, and Paolo Vicig. Probabilities on extensions in abstractargumentation. In
International Workshop on Theorie and Applications of Formal Argumentation , pages 102–119.Springer, 2017.[10] Hengfei Li, Nir Oren, and Timothy J Norman. Probabilistic argumentation frameworks. In
International Workshopon Theorie and Applications of Formal Argumentation , pages 1–16. Springer, 2011.[11] Isaac Levi.
The enterprise of knowledge: An essay on knowledge, credal probability, and chance . MIT press,1983.[12] Dov M Gabbay. Equational approach to argumentation networks.
Argument & Computation , 3(2-3):87–142,2012.[13] Fabio G Cozman. Credal networks.
Artificial intelligence , 120(2):199–233, 2000.[14] Leila Amgoud, Claudette Cayrol, Marie-Christine Lagasquie-Schiex, and Pierre Livet. On bipolarity in argumen-tation frameworks.