Abstract
Social networking services have become deeply embedded in modern society. By January 2024, an estimated 5.04 billion individuals (approximately 62.3% of the world’s population) were active users of social media. Over the previous year alone, these platforms gained 266 million new users, representing a 5.6% growth rate, or about 8.4 new users every second. More than 84% of users are adults aged 18 or older. On average, individuals interact with nearly 6.7 social platforms per month and devote around 2 hours and 23 minutes each day to creating, consuming, or sharing content, close to 15% of their waking hours. While platforms such as Facebook and Twitter serve as central channels for information dissemination and public debate, they also facilitate the rapid and large-scale spread of misleading or harmful content, commonly known as disinformation. The societal impact of disinformation can be profound: it erodes trust in democratic institutions, distorts public opinion, and in extreme cases contributes to social instability or violence. Consequently, ensuring timely and accurate detection of fake news has emerged as a pressing challenge in the digital age. Existing AI-driven fake news detection methods predominantly focus on semantic or linguistic features, attempting to identify irregularities in writing style or textual content. However, such approaches often neglect a critical dimension of the problem: disinformation propagates through social networks in patterns that are markedly different from those of reliable news. The diffusion process is shaped by users’ subjective interpretations and reactions to competing narratives, making propagation behavior a crucial signal. To overcome this limitation, the DISTORT project proposes a fundamentally novel perspective. Rather than concentrating exclusively on content, it employs logic-based and explainable AI techniques to study the dynamics of information diffusion and to identify inconsistencies that arise as content spreads across social networks.
Results achieved
The DISTORT project has successfully achieved all its scientific objectives, delivering the expected outcomes and, in several cases, advancing the state of the art beyond the initial plans. The research activities carried out across the different work packages have jointly contributed to the development of a comprehensive framework for the analysis, reasoning, and explanation of disinformation dynamics in social media. The DISTORT project pursued three core objectives: • Obj1: Logic-based frameworks to analyze fake and true stories diffusion patterns in dynamic disinformation networks • Obj2: Semantics for efficiently reasoning over dynamic inconsistent knowledge • Obj3: Explanations of reasoning over dynamic and/or inconsistent knowledge With respect to Objective 1, the project delivered novel logic-based frameworks for modeling and analyzing the diffusion of both fake and verified news in dynamic social networks. In particular, the introduction of temporal logic formalisms tailored to news propagation enabled the formal characterization of diffusion patterns, capturing temporal, structural, and behavioral aspects that distinguish disinformation from legitimate information. These models were further operationalized through algorithms capable of automatically inferring diffusion rules from real propagation data, thus providing interpretable and data-driven tools for early fake news detection. Regarding Objective 2, the project established solid theoretical foundations for reasoning over dynamic and inconsistent social media data. A key outcome is the definition of inconsistency-tolerant semantics based on customizable repair processes, allowing flexible control over the degree of tolerated inconsistency. The introduction of probabilistic repair mechanisms, formalized via repairing Markov Chains, enabled reasoning with confidence-aware answers. This work was complemented by a thorough complexity analysis and by the identification of fragments admitting efficient approximation algorithms, culminating in the development of novel algorithmic solutions for inconsistency-tolerant reasoning. Concerning Objective 3, the project produced original contributions in the area of explainable reasoning over dynamic and inconsistent knowledge. New explanation frameworks were defined to clarify why specific conclusions are drawn, even in the presence of conflicting or biased information. In addition, the project investigated explanations for machine-learned classifiers, introducing innovative notions of global explanations that go beyond instance-based approaches. These results strengthen the interpretability of both symbolic and hybrid (logic–ML) reasoning systems applied to disinformation analysis.Project details
Unibo Team Leader: Enrico Malizia
Unibo involved Department/s:
Dipartimento di Informatica - Scienza e Ingegneria
Coordinator:
Universita' Della Calabria(Italy)
Total Unibo Contribution: Euro (EUR) 79.658,00
Project Duration in months: 24
Start Date:
30/11/2023
End Date:
29/11/2025
