Agent-based models capture how local decisions give rise to collective phenomena, but are rarely learnable from real data. This study introduces a differentiable framework that reconstructs individual behavioral rules while preserving interaction structure and stochasticity through the integration of graph neural networks and diffusion models. The approach bridges mechanistic modeling and machine learning, enabling empirical testing of theories about decentralized and emergent social and ecological systems.

Agent-based models of opinion formation often rely on repeated simulations to approximate observed collective outcomes, limiting both interpretability and efficiency. This work introduces a likelihood-based estimation approach that directly connects model parameters to data through probabilistic generative modeling, allowing opinions and interactions to be inferred from evidence rather than tuned by trial. By applying it to the bounded-confidence model, the study advances a data-driven understanding of how opinions evolve through social influence.

Explainability defines how people interpret and trust machine learning systems, making it central to the interaction between humans and algorithms. This study introduces an axiomatic benchmark to evaluate whether graph-based explainers faithfully reflect the decision process of the models they interpret. By systematically testing white-box classifiers and real-world networks, it exposes the conditions under which explainers deviate from model logic, offering a rigorous framework for assessing faithfulness and robustness in graph explainability.

Agent-based models explain how micro-level rules produce macro-level patterns, yet linking them to data remains a challenge. This paper presents a protocol for inferring latent agent variables through probabilistic modeling and gradient-based expectation maximization, demonstrated on a housing market simulation. The method enhances predictive accuracy and interpretability, promoting a data-driven way to calibrate generative models of collective behavior.

Language encodes the social processes behind persuasion and disagreement. Drawing from Habermas’ theory of communicative action, this study uses natural language processing to model how intent and tone influence opinion change in online debates. The results show that expressions of knowledge, empathy, and similarity are most effective at fostering change, revealing how communicative structure mediates social learning in digital spaces.

Online discussions evolve through cascades of interactions that mirror collective opinion dynamics. This work introduces a probabilistic generative model that infers latent communities based on their degree of echo-chamber behavior and ideological polarity. By learning from both network structure and information propagation, the model identifies how polarized subgroups emerge and interact, providing a scalable statistical tool for understanding political and social fragmentation online.

Understanding ideology in social media requires models that learn from how information propagates, not just what it contains. This paper introduces a stochastic generative model that infers multidimensional ideological embeddings by fitting the flow of politically salient content across users. By capturing alignment through diffusion dynamics rather than labels, the model learns the structural complexity of political belief systems, offering a principled framework for representing ideology in networked communication.

Political advertising on social media exposes how algorithmic targeting intersects with nationalist discourse and demographic bias. By analyzing over two thousand migration-related campaigns from the Facebook Ads Library, this study uncovers how anti-immigration messages reach audiences differentiated by gender, age, and location, amplifying visibility among groups aligned with nationalist sentiment. The results reveal how algorithmic mediation reinforces cultural divides, showing that persuasion in digital politics depends as much on audience composition as on ideology.

Agent-based models explain how opinions evolve through interaction, yet they often remain detached from real data. This work develops a probabilistic inference framework that learns latent opinions and model parameters directly from social traces, retaining the interpretability of theoretical models while gaining empirical grounding. Applied to political conversations on Reddit, it provides a data-driven test of opinion dynamics theories, revealing limited evidence for the backfire effect in online debate.

The rise of digital nationalism on social media reveals how collective identities form around feedback and recognition. By predicting early Trump supporters on Reddit, this work tests competing sociological mechanisms—homophily, influence, and feedback—and finds that community reinforcement and social mirroring outweigh direct persuasion. The analysis shows how far-right mobilization emerges from everyday interaction dynamics, where belonging and affirmation become drivers of political identity.

Understanding how attributes interact within networks is key to modeling complexity in social and informational systems. This work extends the MGJ model to estimate latent feature–feature interactions that drive or inhibit link formation in large graphs. By reformulating the problem as a perceptron-like learning task, it introduces scalable algorithms that infer structural regularities from high-dimensional relational data, bridging network analysis and machine learning foundations later applied to socio-technical behavior.

Discovering non-obvious relations in knowledge systems requires models that go beyond surface similarity. This paper presents an online margin-based learning algorithm that infers a latent matrix of category interactions to uncover hidden connections within Wikipedia’s hyperlink structure. The method efficiently scales to large graphs, revealing how semantic structures and user-generated organization jointly shape information discovery.

The rise of political alienation reflects growing detachment from institutions even within stable democracies. Using over 35 million Italian tweets, this study applies machine learning time-series models to track expressions of disaffection and compare them with survey indicators of public trust. The analysis demonstrates that collective sentiment on social media mirrors shifts in public opinion, revealing how online discourse can function as a real-time barometer of civic disengagement.