Chair: De Battisti
Avellone, A., De Capitani, L., Fattore, M. Optimal bi-dimensional embedding and visualization of multi-dimensional ordinal binary data
Abstract
The paper develops an original procedure for the dimensionality reduction and the visualization of multi-dimensional ordinal binary data. Using tools from order and lattice theory, the procedure maps a set of k-dimensional binary vectors into a bi-dimensional grid, optimally approximating their component-wise order and providing insights into the distribution and the dependence structure of the input variables. The embedding procedure is implemented through an exhaustive search and is exemplified on data pertaining to life satisfaction in Italy.
Agati, P., Stracqualursi, L. From Words to Personality: Machine Learning for MBTI Profiling
Abstract
This study proposes a machine learning (ML) approach to classify users’ personalities based on their online comments, using the Myers-Briggs Type Indicator (MBTI) as the reference model. Unlike recent transformer-based models, our method offers a faster and more cost-effective solution without requiring substantial computational power or additional expenses. Initially, we trained an Extreme Gradient Boosting (XGBoost) classifier using the well-known MBTI Kaggle dataset and enhanced its performance with the Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance. We then applied the trained model to real-world data — specifically, YouTube comments related to travel and spirituality — to demonstrate its effectiveness in capturing personality traits. The results highlight the potential of our approach for practical applications in personalized content and behavioral analysis. This study underscores the viability of traditional ML techniques in personality classification and paves the way for further research to improve model robustness and scalability.
Brombin, C., Cugnata, F., Di Serio, C. Profiling adolescents through an integrated approach of model-based clustering and sparse graphical models: insights from selfie behavior, social media use and psychological traits
Abstract
Although standard clustering techniques, are widely used in psychological and social sciences to uncover hidden natural clusters in the data they provide little information about pivotal variables in each cluster and their interrelationships. To overcome these limitations, Mixtures of Gaussian Graphical Models have been introduced to integrate clustering and network statistical approaches. We applied this approach to evaluate the relationship among digital self-image, selfie behaviour, psychological constructs and the role of socialization agents in in a a sample of adolescents taking part to an online survey. A three-cluster solution, selected based on the Bayesian Information Criterion (BIC), provided the best fit to the data. Cluster profiles exhibited differences in online self-image management, psychological well-being, and perceived social support, as well as in the interrelationships among the collected variables, with some clusters displaying sparser connections, while others exhibiting higher network density. Integrating key features of model-based clustering with graphical models while allowing to uncover target groups of users with specific online behaviour allow to identify target variables as well as new data- driven relationships on which leverage to develop training programs enhancing digital wellbeing.