Section: New Results

Reasoning with conflicts and decision support

Participants : Pierre Bisquert, Patrice Buche, Michel Chein, Madalina Croitoru, Jérôme Fortin, Alain Gutierrez, Abdelraouf Hecham, Martin Jedwabny, Michel Leclère, Rallou Thomopoulos, Bruno Yun.

The work carried out during this year can be structured into two main research directions: structured logic-based argumentation and collective decision making.

Structured argumentation

To solve real-world problems we sometimes need to consider features that cannot be expressed purely (or naturally) in classical logic. Indeed, real world information is often “imperfect”: it can be partially contradictory, vague or uncertain, etc. During the evaluation period, we mostly considered reasoning in presence of conflicts. To handle this issue, as a reasoning method robust to contradiction, we have used structured argumentation, where arguments have an internal logical structure representing an inference step (i.e. some premises inducing a conclusion). In this context, arguments and their interaction are typically generated from an inconsistent knowledge base. Such arguments are in contrast to those employed in abstract argumentation where they are considered a black box (usually provided as input to a problem and not computed).

More precisely, this year, we mainly worked on two issues: the first one concerns the question of scrutinizing a structured argument, i.e. checking both the validity ("is the conclusion induced by the premisses?") and its soundness ("is the argument valid and are its premisses true?"). This is interesting in the context of collective decision making, where participants utter arguments that can be assessed. The second one relates to the computational complexity of generating arguments from a knowledge base. Indeed, it can potentially produce a huge number of arguments, which impedes the usability of argumentation for big knowledge bases.

Formalizing argument schemes and fallacies

More precisely, we have presented a logical framework allowing us to express assessment of facts and arguments together with a proof system to answer these questions. Our motivation was to clarify the notion of validity in the context of logic-based arguments along different aspects (such as the formulas used and the inference scheme). Originality lies in the possibility for the user to design their own argument schemes, i.e. specific inference patterns (e.g. expert argument, analogy argument). We showed that classical inference obtains when arguments are based on classical schemes (e.g. Hilbert axioms). We went beyond classical logic by distinguishing “proven” formulas from “uncontroversial” ones (whose negation is not proven) and provided a definition of a fallacious argument in this context.

• LPNMR 2019 [20]. In collaboration with Florence Dupin de Saint-Cyr and Philippe Besnard (IRIT).

Optimising argumentation frameworks

Another problem addressed was the large number of logical arguments that can be potentially constructed from a knowledge base. To address this problem we have proposed a compact representation of the structured argumentation system under the form of hypergraphs and implemented it in the NAKED prototype. The tool allows to import a knowledge base (expressed in the existential rule framework), generate, visualise and export the corresponding argumentation hypergraph. These functions, paired with the aim of improving the extension computation efficiency, make this software an interesting tool for non-computer science experts, such as people working in the agronomy domain.

• AAMAS 2019 [33]. In collaboration with Srdjan Vesic (CRIL).

Collective decision making

In this setting we have focused towards the deliberation and voting techniques. We have investigated how deliberation can help generate or impact the structure of preferences underlying the voting process. We have implemented the PAPOW prototype [27] that allows for filtering of voters depending on their individual characteristics.

Argumentation as a tool to generate new preferences

We have investigated how argumentation can solve the Condorcet paradox by using the notion of extension (maxi-consistent sets of arguments) in order to compute new preferences. Our research hypothesis is that a decision made by a group of participants understanding the qualitative rationale (expressed by arguments) behind each other’s preferences has better chances to be accepted and used in practice. Accordingly, we proposed a novel qualitative procedure which combines argumentation with computational social choice for modeling the collective decision-making problem. We showed that this qualitative approach produces structured preferences that can overcome major deficiencies that were exhibited in the social choice literature and affect most of the major voting rules. More precisely, we have dealt with the Condorcet Paradox and the properties of monotonicity and homogeneity, which are unsatisfiable by many voting rules.

• PRAI 2019 [14]. In collaboration with Christos Kaklamanis and Nikos Karanikolas (CTI, Greece).

Argumentation as a tool to modify individual preferences

The previous approach implies that voters are replaced by the extensions which, while it allows to circumvent the Condorcet Paradox, might prove difficult to implement as it disregards the notion of (voters') majority. Hence, we proposed a decision-making procedure based on argumentation and preference aggregation which permits us to explore the effect of reasoning and deliberation along with voting for the decision process. We represented the deliberation phase by defining a new voting argumentation framework, that uses vote and generic arguments, and its acceptability semantics based on the notion of pairwise comparisons between alternatives. We proved for these semantics some theoretical results regarding well-known properties from argumentation and social choice theory.

Moreover, we also studied the notion of unshared features (i.e., alternatives' criteria that constitute justifications of preferences for some agents but not for others) and showed under which conditions it is possible to reach a Condorcet consensus. We provided a deliberation protocol that ensures that, after its completion, the number of unshared features of the decision problem can only be reduced, which would tend to show that deliberation allows to lower the risk of Condorcet Paradox.

• ICAART 2019 [28]. In collaboration with Christos Kaklamanis and Nikos Karanikolas (CTI, Greece). PRIMA 2019 [21].

Discovering and qualifying authority links

We finalized this year the description of the engine SudoQual, devoted to the evaluation of link quality in document bases, developed in collaboration with ABES, the French National Agency for Academic Libraries (http://www.abes.fr), in the context of ANR Qualinca research project (2012-2016) (https://www.lirmm.fr/qualinca/). We presented the methodology and general algorithms used to discover and qualify so-called authority links (which are coreference links between entities mentioned in descriptions of documents and entities described in referential bases). Moreover, ABES has put in production this year a professional tool for documentalists, called Paprika (https://paprika.idref.fr/), whose kernel is the SudoQual engine.

• KCAP 2019 [25].