Two Papers at xAI 2025 🇹🇷

I am happy to share that two of our papers have been accepted at the Third World Conference on Explainable Artificial Intelligence — xAI 2025, held in Istanbul, Turkey 🇹🇷!

Interpretable Personality Detection in Online Social Networks

Our first paper, “Unsupervised and Interpretable Detection of User Personalities in Online Social Networks”, was developed together with Laura Pollacci and Riccardo Guidotti.

The work investigates how interpretable machine learning can be used to identify data-driven behavioural profiles of users in online social networks, with particular attention to differences between toxic and non-toxic users.

A major challenge in this setting is the lack of reliable ground-truth personality information. Instead of assuming predefined personality labels, our approach adopts an unsupervised and data-driven perspective: behavioural profiles are first identified through clustering, and an interpretable predictive model is then trained using transparent linguistic and affective features.

Through a case study on Reddit, we show that interpretable models can achieve competitive predictive performance while also providing meaningful insights into the characteristics that distinguish different user profiles.

This is particularly relevant for the development of personalized and transparent moderation strategies, where understanding why a user is assigned to a particular behavioural profile can be just as important as the prediction itself.

Fair and Interpretable Clustering

Our second paper, “Balancing Fairness and Interpretability in Clustering with FairParTree”, was developed together with Cristiano Landi, Marta Marchiori Manerba, and Riccardo Guidotti.

In this work, we introduce FairParTree, an approach designed to combine two properties that are particularly important when clustering is used in sensitive decision-making scenarios: fairness and interpretability.

Traditional clustering algorithms may produce groups that reflect or reinforce biases present in the data, while their assignments can also be difficult to understand.

FairParTree addresses these issues by integrating fairness constraints directly into the clustering process while exploiting a decision-tree-based representation to provide transparent and human-readable rules describing how instances are assigned to clusters.

Our experiments show that FairParTree can achieve a competitive balance between clustering quality, fairness, and interpretability, providing an alternative for scenarios in which clustering outcomes need to be both responsible and understandable.

Both works reflect different aspects of a common research direction: developing machine learning systems whose decisions are not only effective, but also interpretable, transparent, and responsible.

Looking forward to presenting these works at xAI 2025 and meeting the Explainable AI community in Istanbul.

See you in Istanbul! 🇹🇷