Chico Q. Camargo

Aditi Dutta, Susan Banducci, Chico Q. Camargo

Several computational tools have been developed to detect and identify sexism, misogyny, and gender-based hate speech, particularly on online platforms. These tools draw on insights from both social science and computer science. Given the increasing concern over gender-based discrimination in digital spaces, the contested definitions and measurements of sexism, and the rise of interdisciplinary efforts to understand its online manifestations, a systematic literature review is essential for capturing the current state and trajectory of this evolving field. In this review, we make four key contributions: (1) we synthesize the literature into five core themes: definitions of sexism and misogyny, disciplinary divergences, automated detection methods, associated challenges, and design-based interventions; (2) we adopt an interdisciplinary lens, bridging theoretical and methodological divides across disciplines; (3) we highlight critical gaps, including the need for intersectional approaches, the under-representation of non-Western languages and perspectives, and the limited focus on proactive design strategies beyond text classification; and (4) we offer a methodological contribution by applying a rigorous semi-automated systematic review process guided by PRISMA, establishing a replicable standard for future work in this domain. Our findings reveal a clear disciplinary divide in how sexism and misogyny are conceptualized and measured. Through an evidence-based synthesis, we examine how existing studies have attempted to bridge this gap through interdisciplinary collaboration. Drawing on both social science theories and computational modeling practices, we assess the strengths and limitations of current methodologies. Finally, we outline key challenges and future directions for advancing research on the detection and mitigation of online sexism and misogyny.

Chico Q. Camargo

Most mathematical models of opinion dynamics treat attitudes as scalar quantities or positions on a low-dimensional ideological axis. Empirical attitudes, however, are bundles of positions across many policy issues, and the geometry of the resulting high-dimensional belief space is non-trivial. This paper develops a network-theoretic framework for analysing how individuals move through such a space between two measurement waves. Continuous attitude profiles in $[0,1]^n$ are discretised onto regular grids of resolution $k$, occupied positions form a network whose adjacency is defined by single-issue unit moves, and densely populated regions are interpreted as network-normative: empirically common configurations of attitudes in the population. We introduce a hierarchy of null models against which observed movement can be benchmarked: a closed-form coverage baseline requiring no behavioural parameters; a local random-walk that retains each respondent's baseline position and asks whether destinations are over-represented in occupied regions relative to a uniform 1- or 2-step move; and a marginal permutation null model that preserves per-issue change distributions while disrupting within-respondent cross-issue coupling. Applying the framework to a two-wave panel of $N=1194$ respondents on $n=10$ issues, we find that the observed inside rate exceeds the coverage baseline by a factor of 36 at the focal resolution $k=3$, exceeds the two-hop random-walk null model by $\sim 0.30$, and exceeds the perturbation null model by $\sim 0.04$; only the one-hop random walk is competitive. The perturbation gap grows from near zero at $k=2$ to $\sim 0.14$ at $k=5$, indicating that coupled cross-issue updating is detectable only at fine resolutions. Network-normative attraction is therefore real but representation-contingent: which null model is exceeded, and by how much, changes systematically with $k$.

Chico Q. Camargo, Scott A. Hale, Peter John et al.

Recent election surprises, regime changes, and political shocks indicate that political agendas have become more fast-moving and volatile. The ability to measure the complex dynamics of agenda change and capture the nature and extent of volatility in political systems is therefore more crucial than ever before. This study proposes a definition and operationalization of volatility that combines insights from political science, communications, information theory, and computational techniques. The proposed measures of fractionalization and agenda change encompass the shifting salience of issues in the agenda as a whole and allow the study of agendas across different domains. We evaluate these metrics and compare them to other measures such as issue-level survival rates and the Pedersen Index, which uses public-opinion poll data to measure public agendas, as well as traditional media content to measure media agendas in the UK and Germany. We show how these measures complement existing approaches and could be employed in future agenda-setting research.

Guillermo Valle-Pérez, Chico Q. Camargo, Ard A. Louis

Deep neural networks (DNNs) generalize remarkably well without explicit regularization even in the strongly over-parametrized regime where classical learning theory would instead predict that they would severely overfit. While many proposals for some kind of implicit regularization have been made to rationalise this success, there is no consensus for the fundamental reason why DNNs do not strongly overfit. In this paper, we provide a new explanation. By applying a very general probability-complexity bound recently derived from algorithmic information theory (AIT), we argue that the parameter-function map of many DNNs should be exponentially biased towards simple functions. We then provide clear evidence for this strong simplicity bias in a model DNN for Boolean functions, as well as in much larger fully connected and convolutional networks applied to CIFAR10 and MNIST. As the target functions in many real problems are expected to be highly structured, this intrinsic simplicity bias helps explain why deep networks generalize well on real world problems. This picture also facilitates a novel PAC-Bayes approach where the prior is taken over the DNN input-output function space, rather than the more conventional prior over parameter space. If we assume that the training algorithm samples parameters close to uniformly within the zero-error region then the PAC-Bayes theorem can be used to guarantee good expected generalization for target functions producing high-likelihood training sets. By exploiting recently discovered connections between DNNs and Gaussian processes to estimate the marginal likelihood, we produce relatively tight generalization PAC-Bayes error bounds which correlate well with the true error on realistic datasets such as MNIST and CIFAR10 and for architectures including convolutional and fully connected networks.