PIANO is a research project funded by Next Generation EU as part of the 2022 edition of Italian Research Projects of National Relevance (PRIN 2022), via the initiative Italia Domani.
Motivation
The online spread of toxic speech hinders the exchange and fruition of information, increases radicalization, and may even cause offline harms. To mitigate toxicity, online platforms enforce moderation interventions, such as warning messages, to discourage toxic behaviors and the removal of toxic content and users. Until now, however, these have not significantly reduced online toxicity. So far, all moderation interventions have followed a “one-size-fits-all” approach, where each intervention is applied in the same way for all users. That is to say, personalization in online moderation remains an unexplored strategy.
Approach
PIANO proposes a pioneering approach to online moderation by designing personalized moderation interventions against online toxicity. PIANO will adopt a cross-disciplinary approach grounded on psychology and on machine learning and AI techniques to achieve the project’s Research Objectives (ROs). To this end, our multidisciplinary team will collaborate on user modeling, producing methods for inferring user personality traits based on public traces of online activity.
We will first use knowledge of the user traits and theories from personality psychology and past user behavior to describe users (RO1: User Modeling), which will then inform a process for crafting multiple text-based interventions tailored to the personality of the users to which it will be applied (RO2: Personalization). The end result of PIANO will be a set of theoretical and methodological models for suggesting the best intervention for any given toxic user.
Consortium
Our multidisciplinary consortium is composed of three research units:
- Institute of Informatics and Telematics, National Research Council (CNR)
- Department of Computer Science, University of Pisa (UNIPI)
- Department of Humanities, University of Foggia (UNIFG)
The CNR unit (project coordinator) has proven expertise in ICT research and technology transfer, with an outstanding and award-winning track record of results in social media analysis and mining, particularly about online harms detection and mitigation (e.g., toxicity, social bots, trolls, deepfakes). It also has extensive expertise in all core areas of the project: big data, machine learning, natural language processing, user interaction and persuasive technologies.
The UNIPI unit has a long and excellent experience in personal data analytics, user modeling, decision support systems, agent-based models, federated learning and ethical aspects of AI (explainability and fairness). The unit is part of the KDDLab, a joint research lab focusing on theories, methods and systems for extracting knowledge out of big data. It also has expertise in legal aspects of data protection, as part of DETECT (Centro Interdipartimentale Diritto e Tecnologie di Frontiera).
The UNIFG unit has leading expertise in multiple areas of social and personality psychology, including psychometrics, personality and technological addictions, predictive models for reducing maladaptive online interactions, identification of user profiles for personalized instructional contents, and self-regulation processes of cognitive and emotional aspects.
Workplan
The project started in October 2023, it has an expected duration of two years, and consists of five work packages (WPs).
WP1: Project Management
Leader: CNR
Ensure that the project meets its objectives and overarching goal within budget and schedule, by following R&D best practices and in compliance with all ethical and legal requirements.
- T1.1: Scientific and technical coordination
- T1.2: Quality assurance and project oversight
WP2: User Modeling for Social Media Toxicity
Leader: UNIFG
Fulfill RO1 by advancing models of personality theory related to online toxicity, as well as reactions to past interventions, which will be used for intervention personalization.
- T2.1: Selection of personality traits of intetest
- T2.2: Analysis of past interventions by social media platforms
- T2.3: Computational toxicity user model
WP3: Development of personalized interventions
Leader: CNR
Fulfill RO2 by developing models and a proof-of-concept system to deploy personalized moderation interventions against online toxicity, based on user traits and behavior.
- T3.1: Design of personalized interventions
- T3.2: Computational personalization intervention model
- T3.3: Development of the proof-of-concept system
WP4: Validation of Methods
Leader: UNIPI
Validation of intermediate results and the overall methodological framework developed during the project.
- T4.1: Validation via survey experiments
- T4.2: Validation via simulation experiments
- T4.3: Validation via field experiments
WP5: Dissemination and Exploitation
Leader: CNR
Communicate and make best use of project results with all stakeholders, including scholars, practitioners, and policymakers.
- T5.1: Dissemination to the scientific community
- T5.2: Dissemination to the media
- T5.3: Guidelines for stakeholders
Contact
- Stefano Cresci (Principal Investigator)
- Istituto di Informatica e Telematica (IIT)
- Consiglio Nazionale delle Ricerche (CNR)
- Area della Ricerca
- Via Giuseppe Moruzzi, 1
- 56124 Pisa – Italy