# A Matrix Approach for Weighted Argumentation Frameworks: a Preliminary Report

The assignment of weights to attacks in a classical Argumentation Framework allows to compute semantics by taking into account the different importance of each argument. We represent a Weighted Argumentation Framework by a non-binary matrix, and we characterize the basic extensions (such as w-admissible, w- stable, w-complete) by analysing sub-blocks of this matrix. Also, we show how to reduce the matrix into another one of smaller size, that is equivalent to the original one for the determination of extensions. Furthermore, we provide two algorithms that allow to build incrementally w-grounded and w-preferred extensions starting from a w-admissible extension.

• 7 publications
• 18 publications
• 4 publications
09/10/2012

### A matrix approach for computing extensions of argumentation frameworks

The matrices and their sub-blocks are introduced into the study of deter...
02/11/2022

### On the preferred extensions of argumentation frameworks: bijections with naive extensions

This paper deals with the problem of finding the preferred extensions of...
11/11/2014

### Exploiting Parallelism for Hard Problems in Abstract Argumentation: Technical Report

Abstract argumentation framework () is a unifying framework able to enco...
07/08/2002

### On Concise Encodings of Preferred Extensions

Much work on argument systems has focussed on preferred extensions which...
12/12/2012

### ConArg: a Tool to Solve (Weighted) Abstract Argumentation Frameworks with (Soft) Constraints

ConArg is a Constraint Programming-based tool that can be used to model ...
12/17/2018

### Trichotomic Argumentation Representation

The Aristotelian trichotomy distinguishes three aspects of argumentation...
03/12/2020

### Querying and Repairing Inconsistent Prioritized Knowledge Bases: Complexity Analysis and Links with Abstract Argumentation

In this paper, we explore the issue of inconsistency handling over prior...

## 1 Introduction

An Abstract Argumentation Framework (AF[dung] is represented by a pair consisting of a set of arguments and a binary relation of attack defined between some of them. Given a framework, it is possible to examine the question on which set(s) of arguments can be accepted, hence collectively surviving the conflict defined by . Answering this question corresponds to define an argumentation semantics. The key idea behind extension-based semantics is to identify some sets of arguments (called extensions) that survive the conflict “together”. A very simple example of AF is , where two arguments and attack each other. In this case, each of the two positions represented by either or can be intuitively valid, since no additional information is provided on which of the two attacks prevails. However, having weights on attacks results in such additional information, which can be fruitfully exploited in this direction. For instance, in case the attack is stronger than (or preferred to) , taking the position defined by may result in a better choice for an intelligent agent, since it can be regarded as more reliable or relevant on the framework.

In a recent work, Xu and Cayrol represent an AF by a binary matrix and they give a characterization for stable, admissible and complete extensions by analysing sub-blocks of this matrix [xu15]. Also, they present the reduced matrix w.r.t. conflict-free subsets, by which the determination of extensions becomes more efficient, and that allows to determine -grounded and -preferred extensions.

Our aim is to extend the above mentioned results to Weighted Argumentation Frameworks (WAFs) by adopting the paradigm introduced in [DBLP:conf/cilc/BistarelliPS10, DBLP:conf/flairs/BistarelliRS16] for the semiring-based version of classical semantics. In particular, (i) we characterize -conflict-free, -admissible, -stable and -complete extensions by analysing sub-blocks of a non-binary matrix representing a given WAF, (ii) we show how to reduce this matrix to another one of smaller size that allows to more efficiently determine extensions, and (iii) we provide two algorithms that allow to build incrementally grounded and preferred extensions.

This paper is organized as follows: we first recall the basic definitions on AFs and on WAFs, then we give characterizations for weighted extensions by analysing the matrix associated with the given WAF. Finally, we present the matrix reductions of WAFs based on contraction and division of WAFs, and we provide methods for incrementally building -grounded and -preferred extensions.

## 2 Weighted Argumentation Frameworks

In this section, we recollect the main definitions at the basis of AFs [dung], and introduce c-semirings for dealing with attack-weights. We then rephrase some of the classical definitions, with the purpose to parametrise them with the notion of weighted attack and c-semiring. Last, we give definitions about the matrix representation for AFs.

### 2.1 Abstract Argumentation Frameworks

In his pioneering work [dung], Dung proposed Abstract Frameworks for Argumentation, where (as shown in Figure 1) an argument is an abstract entity whose role is solely determined by its relations to other arguments:

###### Definition 1.

An Abstract Argumentation Framework (AF) is a pair of a set of arguments and a binary relation on , called attack relation. , (or ) means that attacks ( is asymmetric).

Let be an AF and . denotes the set of arguments attacked by (a set attacks a set if exist and with ). denotes the set of arguments attacking . denotes the set of arguments which are not attacked (also called initial arguments of ).

An argumentation semantics is the formal definition of a method ruling the argument evaluation process. In the extension-based approach, a semantics definition specifies how to derive from an AF a set of extensions, where an extension of an AF is simply a subset of . In Definition 2 we define conflict-free sets:

###### Definition 2 (Conflict-free).

A set is conflict-free iff no two arguments and in exist such that attacks .

All the following semantics rely (explicitly or implicitly) upon the concept of defence:

###### Definition 3 (Defence [dung]).

An argument is defended by a set (or defends ) iff for any argument , if then s.t., .

###### Definition 4 (Extension-based semantics).
• A conflict-free set is admissible iff each argument in is defended by .

• An admissible extension is a complete extension iff each argument that is defended by is in .

• A preferred extension is a maximal (w.r.t. set inclusion) admissible subset of .

• A grounded extension is a minimal (w.r.t. set inclusion) complete subset of .

• A conflict-free set is a stable extension iff for each argument which is not in , there exists an argument in that attacks it.

### 2.2 C-semirings

C-semirings are commutative ( is commutative) and idempotent semirings (i.e., is idempotent), where defines a partial order . The obtained structure can be shown to be a complete lattice.

###### Definition 5 (c-semirings).

A commutative semiring is a tuple such that is a set, , and are binary operators making the triples and commutative monoids (semi-groups with identity), satisfying i) (distributivity), and ii) (annihilator). If , the semiring is said to be absorptive.

Well-known instances of c-semirings are:

• 111Boolean c-semirings can be used to model crisp problems and classical Argumentation [dung].,

• ,

• ,

• ,

• .

C-semirings provide a structure that reveals to be suitable for Weighted Argumentation Frameworks. In fact, values in can be used as weights for relations, while the operators and allow to define an ordering among weights.

### 2.3 Weighted AFs

The following definition reshapes the notion of Weighted Argumentation Framework into semiring-based WAF, called :

###### Definition 6 (Semiring-based WAF).

A semiring-based WAF () is a quadruple , where is a c-semiring , is a set of arguments, the attack binary-relation on , and is a binary function. Given and , then means that attacks with a weight . Moreover, we require that iff .

In Figure 2, we provide an example of a WAF describing the defined by , , with , and (i.e., the weighted semiring).

Therefore, each attack is associated with a semiring value that represents the “strength” of an attack between two arguments. We can consider the weights in Figure 2 as supports to the associated attack, as similarly suggested in [DBLP:journals/ai/DunneHMPW11]. A semiring value equal to the top element of the c-semiring (e.g., for the weighted semiring) represents a no-attack relation between two arguments. On the other side, the bottom element, i.e., (e.g., for the weighted semiring), represents the strongest attack possible. In the following, we will use to indicate the operator of the c-semiring on a set of values:

###### Definition 7 (Attacks to/from sets of arguments).

Let be a . A set of arguments attacks a set of arguments and the weight of such attack is , if

 W(B,D)=⨂b∈B,d∈DW(b,d)=k.

For example, looking at Figure 2, we have that , , and .

###### Definition 8 (w-defence [DBLP:conf/flairs/BistarelliRS16]).

Given a , , -defends iff such that , we have that .

A set -defends an argument from , if the of all attack weights from to is worse222When considering the partial order of a generic semiring, we use “worse” or “better” because “greater” or “lesser” would be misleading: in the weighted semiring, , i.e., lesser means better. (w.r.t. ) than the of the attacks from to . For example, the set in Figure 2 defends from because , i.e., ().

###### Definition 9 (w-conflict-free).

Given a , a subset of arguments is -conflict-free if .

Given a , a -conflict-free set is -admissible iff the arguments in are -defended by from the arguments in .

###### Definition 11 (w-complete).

A -admissible extension is also a -complete extension iff each argument such that is -admissible belongs to , i.e., .

###### Definition 12 (w-preferred and w-grounded).

A -preferred extension is a maximal (w.r.t. set inclusion) -admissible subset of . The least (w.r.t. set inclusion) -complete extension is the -grounded extension.

###### Definition 13 (w-stable).

Given , a -admissible set is also a -stable extension iff such that .

## 3 The Matrix Representation for WAFs

Given an AF , we can obtain a matrix representing by using Definition 4 in [xu15]. We extend this definition to represent WAFs through matrices.

###### Definition 14.

Let be a WAF with . The matrix of corresponding to the permutation of , denoted by , is a matrix of order , its elements being determined by the following rules: iff and ; iff .

###### Example 1.

Given as in Figure 3. The matrices of corresponding to the permutations and are

## 4 Characterizing extensions of a WAF

In this section, we mainly focus on the characterization of various extensions in the matrix representing a WAF.

### 4.1 Characterizing the w-conflict-free subsets

The basic requirement for extensions is conflict-freeness. So, we will discuss the matrix condition which insures that a subset of a WAF is conflict-free.

###### Definition 15.

Let be a WAF with and . The sub-block

 Mi,j=⎛⎜ ⎜ ⎜ ⎜ ⎜⎝ai1,i1ai1,i2…ai1,ikai2,i1ai2,i2…ai2,ik⋮⋮⋱⋮aik,i1aik,i2…aik,ik⎞⎟ ⎟ ⎟ ⎟ ⎟⎠

of is called the cf-sub-block of , and denoted by for short. We use this sub-block to find conflict-free subsets of arguments.

###### Claim 1.

Given with , is -conflict-free iff all the elements in the cf-sub-block are .

###### Example 2.

Consider the WAF of Figure 3. We have that , and . By Theorem 1, is -conflict-free, while and are not.

### 4.2 Characterizing the w-admissible subsets

From Definition 10, we know that arguments belonging to a -admissible subset are -defended from the arguments in .

###### Definition 16.

Let be a WAF with , and . The sub-block

 Mi1,i2,…,ikj1,j2,…,jh=⎛⎜ ⎜ ⎜ ⎜ ⎜⎝ai1,j1ai1,j2…ai1,jhai2,j1ai2,j2…ai2,jh⋮⋮⋱⋮aik,j1aik,j2…aik,jh⎞⎟ ⎟ ⎟ ⎟ ⎟⎠

of is called the s-sub-block of , and denoted by for short. The sub-block of

 Mj1,j2,…,jhi1,i2,…,ik=⎛⎜ ⎜ ⎜ ⎜ ⎜⎝aj1,i1aj1,i2…aj1,ikaj2,i1aj2,i2…aj2,ik⋮⋮⋱⋮ajh,i1ajh,i2…ajh,ik⎞⎟ ⎟ ⎟ ⎟ ⎟⎠

is called the -sub-block333In [xu15], is denoted as and it is called the a-sub-block. of , and denoted by .

###### Theorem 1.

Given with , a -conflict-free subset is -admissible iff , , where

refers to the column vector

of the s-sub-block and refers to the column vector of the -sub-block .

###### Example 3.

Let’s consider the -conflict-free subsets and (see Figure 3). We have and , the weight associated to the column vector of is while the weight associated to the row vector of is . Since , is not -admissible in according to Theorem 1.
However, from and , we know that the weight associated to the column vector of is while the weight associated to the row vector of is . Since , we claim that is -admissible in by Theorem 1.

### 4.3 Characterizing the w-stable extensions

We can say whether a -admissible subset is also a -stable extension by checking if all arguments in are attacked by arguments in . On this purpose, we can use the already defined matrix .

###### Theorem 2.

Given with , a -admissible subset is a -stable extension iff each column vector of the s-sub-block of contains only elements different from , where is a permutation of .

###### Example 4.

Let’s consider the -admissible subset (see Figure 3). Since the only column vector of contains some elements different from , we claim that is a -stable extension of , according to Theorem 2.

### 4.4 Characterizing the w-complete extensions

From the definition of -complete extension, it comes that in addition of considering relations between arguments all inside and between arguments in and those outside , we also need to take into account attacks thoroughly outside . We give the following definition and theorem.

###### Definition 17.

Let be a WAF with , and .The sub-block

 Mj1,j2,…,jhj1,j2,…,jh=⎛⎜ ⎜ ⎜ ⎜ ⎜⎝aj1,j1aj1,j2…aj1,jhaj2,j1aj2,j2…aj2,jh⋮⋮⋱⋮ajh,j1ajh,j2…ajh,jh⎞⎟ ⎟ ⎟ ⎟ ⎟⎠

of is called the c-sub-block of , and denoted by for short.

###### Theorem 3.

Given with , a -admissible subset is -complete iff

1. if some column vector of the s-sub-block contains only elements, then its corresponding column vector of the c-sub-block contains some element different from and

2. for each column vector of the c-sub-block appearing in (1), which contains some element different from , there is at least one element of such that , where refers to the column vector of the s-sub-block , where refers to the column vector of the s-sub-block , and .

###### Example 5.

Given as in Figure 4. According to Definition 14, the matrix of is as follows

 M(AF)=⎛⎜ ⎜ ⎜⎝0400008000000300⎞⎟ ⎟ ⎟⎠

By Theorem 1, we have that is -admissible. Note that the matrix has a column vector corresponding in to the column vector . For in , the corresponding column vector in has . Since , according to Theorem 3, we claim that is a -complete extension of .

## 5 Matrix reduction for WAFs

Most of the time, it is convenient to reduce the size of the matrix before performing further operations on it. Below, we provide a method to contract the -conflict-free subset of a matrix into a single entity, without affecting the computation of the extensions. Moreover, we show an iterative procedure for building -grounded and -preferred extensions.

### 5.1 Matrix reduction by contraction

Starting from a conflict-free sub-block, we can characterize -admissible, -stable and -complete extensions of a WAF. Contracting such a sub-block, we obtain a new matrix of smaller size, but with the same semantics status as the original one.

###### Definition 18.

Let be the matrix of a WAF. The combination of two rows and of the matrix consists in “combining” the elements in the same position of the rows. If and are elements in the same position of the rows and respectively, their combination is given by the rule . The combination of two columns of the matrix is similar as the combination of two rows.

For a -conflict-free subset , we can contract the sub-block into a single entry in the matrix. This new entry will have the same status as w.r.t. the extension-based semantics. Thus the matrix can be reduced into another matrix with order by applying the following rules: let , for each such that and ,

1. combine rows to the row ;

2. combine column to the column ;

3. delete row and column .

The matrix obtained in this way is called the reduced matrix w.r.t. the conflict-free subset . Also, the original WAF can be reduced into a new one with arguments by applying the following rules. Let and . For each such that and , and each such that , set and . Then,

1. if , combine to and set ;

2. if , combine to and set ;

3. delete and from .

Let denote the new relation and , then is a new AF called the reduced AF w.r.t. . Obviously, the reduced matrix is exactly the matrix obtained from and .

###### Theorem 4.

Given with , let be conflict-free and . Then is stable (resp. admissible, complete, preferred) in AF iff is stable (respectively admissible, complete, preferred) in the reduced .

### 5.2 Matrix reduction by division

Let be a WAF. The -grounded extension of can be viewed as the union of two subsets and : consists of the initial arguments of and is the -grounded extension, -defended by , of the remaining sub-AF w.r.t.  (that is , where ). On the other hand, a -preferred extension coincides with an admissible extension , -defended by , from which the associated remaining sub-AF (where ) has no nonempty admissible extension. We have the following theorem.

###### Theorem 5.

Let be a WAF, be a -admissible extension of , and . If is a -admissible (resp. -stable, -complete, -preferred) extension, -defended by , of the remaining sub-AF w.r.t.  (), then is a -admissible (resp. -stable, -complete, -preferred) extension of .

###### Example 6.

Given in Figure 4, consider and , with . is -admissible in and is -admissible in . Then, for Theorem 5, is a -admissible extension of .

### Building w-grounded extensions

A -grounded extension can be built incrementally by starting from a -admissible extension. Let be the set of initial arguments of , then is a -admissible extension. If has no initial arguments, then the -grounded extension of is empty. Otherwise, let be the set of initial arguments of . We proceed to construct by computing the sets as follows:

1. ;

2. and ;

1. compute ;

2. , , , ;

3. (with ), if then and ;

4. repeat (c) until ;

5. , with ;

3. repeat 3. until or .

This process can be done repeatedly until, for some , , where . From Theorem 5, we know that the set union between -admissible extensions is a -admissible extension in turn. At this point, the set is the -grounded extension of . Note that the set coincides with the set of undec arguments in the labelling of where is the set of in arguments.

###### Example 7.

Let be a WAF as in Figure 5. We have , so we look for the sets . , so , and . Consider that implies . is the -grounded extension of .

### Building w-preferred extensions

A -preferred extension can be built incrementally by starting from some -admissible extension. Since the -preferred semantics admits more extensions, different -preferred extensions can be built, depending on both the initial extension and the selection of the nonempty -admissible on each step of the procedure. Let be any -admissible extension of and compute:

1. ;

2. and ;

1. compute ;

2. , , , ;

3. (with ), if then and ;

4. repeat (c) until ;

5. , with ;

3. repeat 3. until or .

This process can be done repeatedly until, for some , , where . At this point, by Theorem 5, the set is the -preferred extension of .

###### Example 8.

Let be a WAF as in Figure 5. Let’s consider the -admissible extension of . Thus , and . Since and , is the -preferred extension of .

Computational Complexity. We analysed the above described algorithms from the computational point of view. The first algorithm, which computes -grounded extensions, has an overall time complexity of . The algorithm for -preferred extensions reveals worse performance than the first one, with a time complexity of . This is due to the fact that an admissible extension has to be found at each execution of step 3. A more extended study of the complexity is left for future work.

## 6 Conclusion and Future Work

In this work, we introduce a matrix approach for studying extensions of semiring-based semantics. A WAF is represented as a matrix in which all elements correspond to weights assigned to relations among arguments. In particular, by extracting sub-blocks from this matrix, it is possible to check if a set of arguments is an extension for some semantics. Also, we describe an incremental procedure for building -grounded and -preferred extensions and we study how to reduce the number of arguments of a WAF in order to obtain a contracted matrix with the same status as the original one (w.r.t. the semantics). A possible application for this approach could be the identification of equational representation of semiring-based extensions, by using the method proposed in [DBLP:conf/ecsqaru/Gabbay11]. We plan to extend our current implementation [DBLP:conf/ictai/BistarelliS11, DBLP:conf/tafa/BistarelliS11] with the proposed approaches, and to test their performance on real applications. Finally, we would like to investigate whether such methodologies can be applied when considering coalitions of arguments [DBLP:journals/fuin/BistarelliS13].