Elena Cabrio
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Elena Cabrio has a total of 76 co-authors. Here are the top co-authors by number of shared publications:
- Serena Villata
Université Côte d’Azur, Inria, CNRS, I3S, Sophia-Antipolis, France
28 shared publications - Xiaoou Wang
Université Côte d'Azur CNRS Inria I3S France
3 shared publications - Anna De Liddo
The Open University
2 shared publications - Ekaterina Sviridova
Université Côte d’Azur, Inria, CNRS, I3S, France
2 shared publications - Fabien Gandon
Inria, Université Côte d’Azur, CNRS, I3S
2 shared publications - Francesca Rizzo
Politecnico di Milano
2 shared publications
Latest Publications
- Open Access
- 10 September 2026
- Published by IOS Press
- p. 53-64
How Can Similarity Be Explained? Modeling Multiple Points of View with Inverse simFOL
Similarity is often reduced to a single score, sufficient for ranking but insufficient for explanation, as it does not reveal which aspects drive the comparison. We introduce inverse simFOL, an interpretable framework that infers clause and symbol-level importance weights explaining a target similarity score. Building on a fixed and transparent simFOL pipeline over first-order logic (FOL), the problem is formulated as a discrete, constrained optimization problem and solved via genetic search to recover diverse explanations. Experiments on STS-B pairs and an argument-level case study show that similarity admits multiple valid explanations and is strongly context-dependent.
- 10 September 2026
- Published by IOS Press
- p. 53-64
Open AccessHow Can Similarity Be Explained? Modeling Multiple Points of View with Inverse simFOL
Similarity is often reduced to a single score, sufficient for ranking but insufficient for explanation, as it does not reveal which aspects drive the comparison. We introduce inverse simFOL, an interpretable framework that infers clause and symbol-level importance weights explaining a target similarity score. Building on a fixed and transparent simFOL pipeline over first-order logic (FOL), the problem is formulated as a discrete, constrained optimization problem and solved via genetic search to recover diverse explanations. Experiments on STS-B pairs and an argument-level case study show that similarity admits multiple valid explanations and is strongly context-dependent.
- Open Access
- 10 September 2026
- Published by IOS Press
- p. 77-88
Better than the Gold Standard? Evaluating Human vs. LLM Enthymeme Reconstruction in Natural Language Argumentation
The reconstruction of enthymemes (i.e., incomplete arguments) remains a fundamental challenge for Natural Language Processing, bridging the gap between surface-level text understanding and deep pragmatic reasoning. Despite its importance for downstream tasks, research in this area remains scarce. A lack of high-quality benchmarks and a lack of consensus on task formulation complicate further advancements. In this paper, we present a novel high-quality, human-reconstructed dataset based on real-world arguments. To assess this benchmark for enthymeme reconstruction, we conduct a comprehensive comparative study of various Large Language Models (LLMs), evaluating their capacity for logical inference through automatic metrics, a human evaluation framework, and an LLM-as-a-Judge approach. Our results reveal that human experts often find the reconstructions of language models indistinguishable from human-authored benchmarks. Our annotated resource and evaluation framework provide a roadmap for the future of automated inference in computational argumentation.
- 10 September 2026
- Published by IOS Press
- p. 77-88
Open AccessBetter than the Gold Standard? Evaluating Human vs. LLM Enthymeme Reconstruction in Natural Language Argumentation
The reconstruction of enthymemes (i.e., incomplete arguments) remains a fundamental challenge for Natural Language Processing, bridging the gap between surface-level text understanding and deep pragmatic reasoning. Despite its importance for downstream tasks, research in this area remains scarce. A lack of high-quality benchmarks and a lack of consensus on task formulation complicate further advancements. In this paper, we present a novel high-quality, human-reconstructed dataset based on real-world arguments. To assess this benchmark for enthymeme reconstruction, we conduct a comprehensive comparative study of various Large Language Models (LLMs), evaluating their capacity for logical inference through automatic metrics, a human evaluation framework, and an LLM-as-a-Judge approach. Our results reveal that human experts often find the reconstructions of language models indistinguishable from human-authored benchmarks. Our annotated resource and evaluation framework provide a roadmap for the future of automated inference in computational argumentation.
- Top Cited PapersOpen Access
- 11 February 2026
- Published by Springer Nature
- p. 221-251
Enhancing Democratic Deliberations with AI: Insights from the ORBIS Co-creation Journey
As Artificial Intelligence (AI) continues to transform the public sector, particularly in the domain of public deliberation and democratic governance, it also presents distinct challenges. This chapter explores the insights from the ORBIS project, which is situated at the intersection of AI technology and democratic processes. The chapter will critically present and discuss how the project is contributing to advance AI-enhanced democratic and inclusive deliberations by developing AI tools that are ethically grounded and trustworthy by-design. These tools are designed to enhance the accessibility, understandability, and overall quality of deliberative processes, enabling broader participation and facilitating a deeper comprehension of diverse public opinions across various policy issues. By employing advanced data analytics, NLP models, and algorithms for distilling a large volume of public inputs into actionable insights, ORBIS serves as a privilege testbed for discussing how to engage stakeholders into the design of AI-based tools, ensuring their social relevance. This chapter focuses on how the processes of continuous and iterative stakeholder engagement and co-creation have been adopted to ensure alignment of AI functionalities with user needs and democratic values, thus exploring implications towards enhancing the legitimacy of the policymaking process. The discussion highlights how ORBIS serves as a model for integrating AI into scalable democratic practices for public decision-making, reflecting on both the technological advancements and ethical considerations essential for responsible AI deployment in public decision-making.
- 11 February 2026
- Published by Springer Nature
- p. 221-251
Top Cited PapersOpen AccessEnhancing Democratic Deliberations with AI: Insights from the ORBIS Co-creation Journey
As Artificial Intelligence (AI) continues to transform the public sector, particularly in the domain of public deliberation and democratic governance, it also presents distinct challenges. This chapter explores the insights from the ORBIS project, which is situated at the intersection of AI technology and democratic processes. The chapter will critically present and discuss how the project is contributing to advance AI-enhanced democratic and inclusive deliberations by developing AI tools that are ethically grounded and trustworthy by-design. These tools are designed to enhance the accessibility, understandability, and overall quality of deliberative processes, enabling broader participation and facilitating a deeper comprehension of diverse public opinions across various policy issues. By employing advanced data analytics, NLP models, and algorithms for distilling a large volume of public inputs into actionable insights, ORBIS serves as a privilege testbed for discussing how to engage stakeholders into the design of AI-based tools, ensuring their social relevance. This chapter focuses on how the processes of continuous and iterative stakeholder engagement and co-creation have been adopted to ensure alignment of AI functionalities with user needs and democratic values, thus exploring implications towards enhancing the legitimacy of the policymaking process. The discussion highlights how ORBIS serves as a model for integrating AI into scalable democratic practices for public decision-making, reflecting on both the technological advancements and ethical considerations essential for responsible AI deployment in public decision-making.
- Open Access
- 11 February 2026
- Published by Springer Nature
- p. 89-118
ORBIS: An AI-Enabled Architecture for Scaling Online Deliberation
The chapter presents the process behind the development of the ORBIS toolkit as a set of software components and services de-signed to facilitate large-scale online deliberation through the combination of human (collective intelligence) and artificial intelligence technologies. The platform is built on a modular and flexible architecture that combines different AI heterogeneous components. The main layers of the toolkit are: (i) Enabling layer: Provides core services such as authentication, authorisation, and resources management. (ii) Data Layer: Manages deliberation content, participant profiles, AI models, and connects to external knowledge bases, (iii) Service Layer: Contains the key AI-powered deliberation functionalities such as Argument Mining, Explanations Generator, Feedback Aggregation, Conflict Detection and Policy Recommendation, and (iv) User Interface Layer: Composed by existing well-adopted deliberation services designed to consume in a seamless to the end-user and orchestrated way, the external deliberation augmentation services. The chapter points out the key innovations of ORBIS toolkit, which include: (i) argument visualisation and clustering to organise lengthy discussions, (ii) AI-assisted explanation and interpretation of complex issues, (iii) summarisation of key ideas and points in long deliberations, and (iv) reporting mechanisms to assist policy-makers in their decision making procedures. By combining collective and artificial intelligence, the chapter ultimately describes how ORBIS aims to achieve scaling of online deliberations while maintaining discussion quality. Ultimately, it is discussed how the further promising research that this open architecture invites, and how it can inform future development and advances in the emerging field of AI-augmented collective intelligence.
- 11 February 2026
- Published by Springer Nature
- p. 89-118
Open AccessORBIS: An AI-Enabled Architecture for Scaling Online Deliberation
The chapter presents the process behind the development of the ORBIS toolkit as a set of software components and services de-signed to facilitate large-scale online deliberation through the combination of human (collective intelligence) and artificial intelligence technologies. The platform is built on a modular and flexible architecture that combines different AI heterogeneous components. The main layers of the toolkit are: (i) Enabling layer: Provides core services such as authentication, authorisation, and resources management. (ii) Data Layer: Manages deliberation content, participant profiles, AI models, and connects to external knowledge bases, (iii) Service Layer: Contains the key AI-powered deliberation functionalities such as Argument Mining, Explanations Generator, Feedback Aggregation, Conflict Detection and Policy Recommendation, and (iv) User Interface Layer: Composed by existing well-adopted deliberation services designed to consume in a seamless to the end-user and orchestrated way, the external deliberation augmentation services. The chapter points out the key innovations of ORBIS toolkit, which include: (i) argument visualisation and clustering to organise lengthy discussions, (ii) AI-assisted explanation and interpretation of complex issues, (iii) summarisation of key ideas and points in long deliberations, and (iv) reporting mechanisms to assist policy-makers in their decision making procedures. By combining collective and artificial intelligence, the chapter ultimately describes how ORBIS aims to achieve scaling of online deliberations while maintaining discussion quality. Ultimately, it is discussed how the further promising research that this open architecture invites, and how it can inform future development and advances in the emerging field of AI-augmented collective intelligence.
- Open Access
Stakeholder Suite: A Unified AI Framework for Mapping Actors, Topics and Arguments in Public Debates
Public debates surrounding infrastructure and energy projects involve complex networks of stakeholders, arguments, and evolving narratives. Understanding these dynamics is crucial for anticipating controversies and informing engagement strategies, yet existing tools in media intelligence largely rely on descriptive analytics with limited transparency. This paper presents Stakeholder Suite, a framework deployed in operational contexts for mapping actors, topics, and arguments within public debates. The system combines actor detection, topic modeling, argument extraction and stance classification in a unified pipeline. Tested on multiple energy infrastructure projects as a case study, the approach delivers fine-grained, source-grounded insights while remaining adaptable to diverse domains. The framework achieves strong retrieval precision and stance accuracy, producing arguments judged relevant in 75% of pilot use cases. Beyond quantitative metrics, the tool has proven effective for operational use: helping project teams visualize networks of influence, identify emerging controversies, and support evidence-based decision-making.
Open AccessStakeholder Suite: A Unified AI Framework for Mapping Actors, Topics and Arguments in Public Debates
Public debates surrounding infrastructure and energy projects involve complex networks of stakeholders, arguments, and evolving narratives. Understanding these dynamics is crucial for anticipating controversies and informing engagement strategies, yet existing tools in media intelligence largely rely on descriptive analytics with limited transparency. This paper presents Stakeholder Suite, a framework deployed in operational contexts for mapping actors, topics, and arguments within public debates. The system combines actor detection, topic modeling, argument extraction and stance classification in a unified pipeline. Tested on multiple energy infrastructure projects as a case study, the approach delivers fine-grained, source-grounded insights while remaining adaptable to diverse domains. The framework achieves strong retrieval precision and stance accuracy, producing arguments judged relevant in 75% of pilot use cases. Beyond quantitative metrics, the tool has proven effective for operational use: helping project teams visualize networks of influence, identify emerging controversies, and support evidence-based decision-making.
- Open Access
- 24 November 2025
- Published by Elsevier in Data & Knowledge Engineering
- Vol. 162
“Detectors Lead, LLMs Follow”: Integrating LLMs and traditional models on implicit hate speech detection to generate faithful and plausible explanations
- 24 November 2025
- Published by Elsevier in Data & Knowledge Engineering
- Vol. 162
Open Access“Detectors Lead, LLMs Follow”: Integrating LLMs and traditional models on implicit hate speech detection to generate faithful and plausible explanations
- Open Access
- 1 September 2025
- Published by International Joint Conferences on Artificial Intelligence
- Vol. 12, 11114-11118
SAFE: Structured Argumentation for Fact-checking with Explanations
Explainable fact-checking plays a vital role in the fight against disinformation in today’s digital landscape. With the increasing volume of unverified content online, providing justifications for fact-checking has become essential to help users make informed decisions. While recent studies provide user-friendly explanations through abstractive or extractive summarization, they often assume the availability of human-written fact-checking articles, which is not always the case. This demo introduces SAFE, an argument-based framework designed to enhance both fact-checking and its justification. Specifically, SAFE offers three key features: i) producing argument-structured summaries of human-written fact-checking articles, ii) in the absence of human-written articles, generating structured summaries based on evidence retrieved from a corpus through a jointly trained summarization and evidence retrieval system, and iii) assessing the truthfulness of a claim by analyzing the structured summary.
- 1 September 2025
- Published by International Joint Conferences on Artificial Intelligence
- Vol. 12, 11114-11118
Open AccessSAFE: Structured Argumentation for Fact-checking with Explanations
Explainable fact-checking plays a vital role in the fight against disinformation in today’s digital landscape. With the increasing volume of unverified content online, providing justifications for fact-checking has become essential to help users make informed decisions. While recent studies provide user-friendly explanations through abstractive or extractive summarization, they often assume the availability of human-written fact-checking articles, which is not always the case. This demo introduces SAFE, an argument-based framework designed to enhance both fact-checking and its justification. Specifically, SAFE offers three key features: i) producing argument-structured summaries of human-written fact-checking articles, ii) in the absence of human-written articles, generating structured summaries based on evidence retrieved from a corpus through a jointly trained summarization and evidence retrieval system, and iii) assessing the truthfulness of a claim by analyzing the structured summary.
- 30 June 2025
- Published by SAGE Publications in Argument & Computation
- Vol. 17 (1) , 3-27
Mining implicit arguments for reasoning: A survey
Argumentation is the process of creating arguments for and against competing claims. Computational argumentation involves different ways of analyzing and reasoning upon arguments and their relations. More precisely, Argument Mining is the research field aiming at automatically identifying and classifying argument structures in text. The research field is mainly focussed on the extraction of explicit argument structures (i.e., claims and premises connected by support and attack relations). However, an even more challenging task consists in extracting implicit argument structures in text (e.g., enthymemes). These structures are particularly valuable to then address argument reasoning, e.g., on incomplete and uncertain information, to finally compute the set of acceptable arguments, i.e., argument justification and skepticism. In this paper, we present and compare current approaches and available datasets for the novel task of Implicit Argument Mining. Future work perspectives are discussed to pave the way to further studies in this direction.
- 30 June 2025
- Published by SAGE Publications in Argument & Computation
- Vol. 17 (1) , 3-27
Mining implicit arguments for reasoning: A survey
Argumentation is the process of creating arguments for and against competing claims. Computational argumentation involves different ways of analyzing and reasoning upon arguments and their relations. More precisely, Argument Mining is the research field aiming at automatically identifying and classifying argument structures in text. The research field is mainly focussed on the extraction of explicit argument structures (i.e., claims and premises connected by support and attack relations). However, an even more challenging task consists in extracting implicit argument structures in text (e.g., enthymemes). These structures are particularly valuable to then address argument reasoning, e.g., on incomplete and uncertain information, to finally compute the set of acceptable arguments, i.e., argument justification and skepticism. In this paper, we present and compare current approaches and available datasets for the novel task of Implicit Argument Mining. Future work perspectives are discussed to pave the way to further studies in this direction.
- Open Access
- 22 April 2025
- Published by SAGE Publications in Argument & Computation
- Vol. 16 (3) , 405-424
When automated fact-checking meets argumentation: Unveiling fake news through argumentative evidence
The need for automated fact-checking has become urgent with the rise of misleading content on social media. Recently, Fake News Classification (FNC) has evolved to incorporate justifications provided by fact-checkers to explain their decisions. In this work, we argue that an argumentative representation of fact-checkers’ justifications can improve the precision and explainability of FNC systems. To address this challenging task, we present LIARArg, a novel linguistic resource composed of 2,832 news and their justifications. LIARArg extends the 6-label FNC dataset LIAR-PLUS with argumentation structures, leading to the first FNC dataset annotated with argument components (claim and premise) and fine-grained relations (attack, support, partial support and partial attack). To integrate argumentation in FNC, we propose a novel joint learning method combining, for the first time, Argument Mining and FNC which outperforms state-of-the-art approaches, especially for news with intermediate truthfulness labels. Besides, our experimental setting demonstrates that fine-grained relations allow an extra performance boost. We also show that the argumentative representation of human justifications can be exploited in a Chain-of-Thought manner both in prompts and model output, paving a promising avenue for research in explainable fact-checking. Finally, our fully automated pipeline shows that integrating argumentation into FNC is not only feasible but also effective.
- 22 April 2025
- Published by SAGE Publications in Argument & Computation
- Vol. 16 (3) , 405-424
Open AccessWhen automated fact-checking meets argumentation: Unveiling fake news through argumentative evidence
The need for automated fact-checking has become urgent with the rise of misleading content on social media. Recently, Fake News Classification (FNC) has evolved to incorporate justifications provided by fact-checkers to explain their decisions. In this work, we argue that an argumentative representation of fact-checkers’ justifications can improve the precision and explainability of FNC systems. To address this challenging task, we present LIARArg, a novel linguistic resource composed of 2,832 news and their justifications. LIARArg extends the 6-label FNC dataset LIAR-PLUS with argumentation structures, leading to the first FNC dataset annotated with argument components (claim and premise) and fine-grained relations (attack, support, partial support and partial attack). To integrate argumentation in FNC, we propose a novel joint learning method combining, for the first time, Argument Mining and FNC which outperforms state-of-the-art approaches, especially for news with intermediate truthfulness labels. Besides, our experimental setting demonstrates that fine-grained relations allow an extra performance boost. We also show that the argumentative representation of human justifications can be exploited in a Chain-of-Thought manner both in prompts and model output, paving a promising avenue for research in explainable fact-checking. Finally, our fully automated pipeline shows that integrating argumentation into FNC is not only feasible but also effective.
- Open Access
When automated fact-checking meets argumentation: unveiling fake news through argumentative evidence
The need for automated fact-checking has become urgent with the rise of misleading content on social media. Recently, Fake News Classification (FNC) has evolved to incorporate justifications provided by fact-checkers to explain their decisions. In this work, we argue that an argumentative representation of fact-checkers' justifications can improve the precision and explainability of FNC systems. To address this challenging task, we present LIARArg, a novel linguistic resource composed of 2,832 news and their justifications. LIARArg extends the 6-label FNC dataset LIAR-PLUS with argumentation structures, leading to the first FNC dataset annotated with argument components (claim and premise) and fine-grained relations (attack, support, partial support and partial attack). To integrate argumentation in FNC, we propose a novel joint learning method combining, for the first time, Argument Mining and FNC which outperforms state-of-the-art approaches, especially for news with intermediate truthfulness labels. Besides, our experimental setting demonstrates that fine-grained relations allow an extra performance boost. We also show that the argumentative representation of human justifications can be exploited in a Chain-of-Thought manner both in prompts and model output, paving a promising avenue for research in explainable fact-checking. Finally, our fully automated pipeline shows that integrating argumentation into FNC is not only feasible but also effective.
Open AccessWhen automated fact-checking meets argumentation: unveiling fake news through argumentative evidence
The need for automated fact-checking has become urgent with the rise of misleading content on social media. Recently, Fake News Classification (FNC) has evolved to incorporate justifications provided by fact-checkers to explain their decisions. In this work, we argue that an argumentative representation of fact-checkers' justifications can improve the precision and explainability of FNC systems. To address this challenging task, we present LIARArg, a novel linguistic resource composed of 2,832 news and their justifications. LIARArg extends the 6-label FNC dataset LIAR-PLUS with argumentation structures, leading to the first FNC dataset annotated with argument components (claim and premise) and fine-grained relations (attack, support, partial support and partial attack). To integrate argumentation in FNC, we propose a novel joint learning method combining, for the first time, Argument Mining and FNC which outperforms state-of-the-art approaches, especially for news with intermediate truthfulness labels. Besides, our experimental setting demonstrates that fine-grained relations allow an extra performance boost. We also show that the argumentative representation of human justifications can be exploited in a Chain-of-Thought manner both in prompts and model output, paving a promising avenue for research in explainable fact-checking. Finally, our fully automated pipeline shows that integrating argumentation into FNC is not only feasible but also effective.