Davide Bassi

The prevalence of controversies in online interactions has recently led researchers to develop diverse computational methods for controversy detection, reflecting the many ways controversies manifest across contexts. This paper presents a systematic literature review of computational methods for Controversy Detection to provide a unified framework for this rapidly evolving field. To handle the diversity of approaches, we introduce an inclusive operational definition that accommodates various manifestations while maintaining analytical precision. Building on this foundation, we gathered 47 studies spanning 2010–2023. Our analysis yields three main outcomes: (1) a detailed inventory of datasets used in controversy detection; (2) a six-dimensional taxonomy of linguistic and interactive features for operationalizing the task; and (3) an evaluation of computational models based on these features, showing that deep learning models integrating linguistic attributes with graph-based user interaction representations achieve superior performance. Despite these advances, real-world deployment remains limited by several gaps: conceptualization issues, an over-reliance on Twitter at the expense of other platforms, and technical challenges particularly around explainability. The paper concludes by identifying future research directions to address these shortcomings.

As artificial intelligence (AI) and its applications continue to reshape our world, understanding its historical roots becomes increasingly crucial. However, this historical perspective, focused on documenting what actually happened from the most objective point of view, is incomplete if we do not capture how different groups remember those events. For this reason, we argue in this paper in favor of developing a collective memory of AI in three different lines (as a body of knowledge, as attributes, and as a process) which will complement the historical task. As a result, we will obtain a more comprehensive vision of the development of AI, incorporating a more humanistic perspective, which will be crucial to understanding how society assesses AI.

Social dialogue is a cornerstone for political decision-making and maintaining mutual understanding between diverse societal groups when addressing collective challenges. However, this dialogue is increasingly strained in digital environments where users regularly encounter opposing viewpoints. While research has examined how political actors strategically leverage divisive rhetoric, less attention has been paid to how ordinary users utilize these devices in everyday online interactions. This study investigates how users employ divisive rhetorical strategies across social networks, examining the relationships between topic controversiality, user stance, and interactive patterns. Through a large-scale analysis of 146K YouTube comments on immigration and climate change discussions—two highly polarizing topics in contemporary discourse. The research combines computational methods for rhetoric mining with network analysis to track patterns of user interaction and manifestation of divisive rhetoric. Our analysis reveals three key findings: (1) Controversial topics elicit significantly higher frequencies of divisive rhetorical strategies compared to non-controversial ones, with distinct patterns across topics; (2) Users demonstrating strong stance commitment (Pro and Contra) use significantly more divisive rhetoric with parallel patterns, regardless of ideological position, distinguishing them from neutral users; (3) Users strategically adapt their rhetorical behavior to their interlocutor’s stance, suggesting that stance intensity rather than specific ideological content drives rhetorical similarity. Framed through Social Identity Theory, we conceptualize these wedge rhetorical devices as an interactive toolkit that users deploy to navigate social positioning in deindividualized discussions, either reinforcing solidarity among users sharing similar positions or creating distinctions from those holding opposing views. This study shows how computational methods can effectively track and analyze the ways citizens strategically navigate social positioning on sensitive issues, contributing to our understanding of online political discourse dynamics.

Hyperpartisan news consists of articles with strong biases that support specific political parties. The spread of such news increases polarization among readers, which threatens social unity and democratic stability. Automated tools can help identify hyperpartisan news in the daily flood of articles, offering a way to tackle these problems. With recent advances in machine learning and deep learning, there are now more methods available to address this issue. This literature review collects and organizes the different methods used in previous studies on hyperpartisan news detection. Using the PRISMA methodology, we reviewed and systematized approaches and datasets from 81 articles published from January 2015 to 2024. Our analysis includes several steps: differentiating hyperpartisan news detection from similar tasks, identifying text sources, labeling methods, and evaluating models. We found some key gaps: there is no clear definition of hyperpartisanship in Computer Science, and most datasets are in English, highlighting the need for more datasets in minority languages. Moreover, the tendency is that deep learning models perform better than traditional machine learning, but Large Language Models’ (LLMs) capacities in this domain have been limitedly studied. This paper is the first to systematically review hyperpartisan news detection, laying a solid groundwork for future research.

Michele Joshua Maggini, Davide Bassi, Pablo Gamallo. Proceedings of the 2nd Workshop on Natural Language Processing Meets Climate Change (ClimateNLP 2025). 2025.

Davide Bassi, Erik Bran Marino, Renata Vieira, Martin Pereira. Proceedings of the 12th Argument mining Workshop. 2025.

Davide Bassi, Dimitar Iliyanov Dimitrov, Bernardo D’Auria, Firoj Alam, Maram Hasanain, Christian Moro, Luisa Orrù, Gian Piero Turchi, Preslav Nakov, Giovanni Da San Martino. Findings of the Association for Computational Linguistics: ACL 2025. 2025.

The proliferation of digital communication has profoundly transformed the landscape of persuasive discourse. Online platforms have amplified the reach and impact of persuasive techniques. However, they have also enabled the rapid spread of manipulative content, targeted propaganda, and divisive rhetoric. Consequently, a wide range of computational approaches has emerged to address the multifaceted nature of digital persuasion, to detect and mitigate its harmful practices. In light of this, the paper surveys computational methods for detecting persuasive means in digital communication, focusing on how they integrate humanistic knowledge to operationalize this construct. Additionally, special emphasis is placed on models' explainability, a pivotal aspect considering these models are used by institutions to influence societal interactions. For the analysis, two primary perspectives in persuasion are defined: linguistic and argumentative. The linguistic approach analyzes specific textual features, allowing for highly accountable algorithms based on explicit rules. The argumentative approach focuses on broader persuasive mechanisms, offering greater scalability but often resulting in less explainable models due to their complexity. This tension between model sophistication and explainability presents a key challenge in developing effective and transparent persuasion detection systems. The results highlight the spectrum of methodologies for studying persuasion, ranging from analyzing stylistic elements to detecting explicitly propagandist messages. Our findings highlight two key challenges in using these algorithms to tackle societal issues of persuasion misuse: the opacity of deep learning models and the absence of a theoretically grounded distinction between vicious and virtuous persuasion. To address these challenges, we propose integrating social sciences and humanities theories to enhance the effectiveness and ethical robustness of persuasion detection systems. This interdisciplinary approach enables a more nuanced characterization of text, facilitating the differentiation between vicious and virtuous persuasion through analysis of rhetorical, argumentative, and emotional aspects. We emphasize the potential of hybrid approaches that combine rule-based methods with deep learning techniques, as these offer a promising avenue for implementing this interdisciplinary framework. The paper concludes by outlining future challenges, including the importance of multimodal and multilingual analysis, ethical considerations in handling user-generated data and the growing challenge of distinguishing between human and AI-generated persuasive content.