Large Language Models (LLMs) are becoming increasingly integrated into our lives. Hence, it is important to understand the biases present in their outputs in order to avoid perpetuating harmful stereotypes, which originate in our own flawed ways of thinking. This challenge requires developing new benchmarks and methods for quantifying affective and semantic bias, keeping in mind that LLMs act as psycho-social mirrors that reflect the views and tendencies that are prevalent in society. One such tendency that has harmful negative effects is the global phenomenon of anxiety toward math and STEM subjects. In this study, we introduce a novel application of network science and cognitive psychology to understand biases towards math and STEM fields in LLMs from ChatGPT, such as GPT-3, GPT-3.5, and GPT-4. Specifically, we use behavioral forma mentis networks (BFMNs) to understand how these LLMs frame math and STEM disciplines in relation to other concepts. We use data obtained by probing the three LLMs in a language generation task that has previously been applied to humans. Our findings indicate that LLMs have negative perceptions of math and STEM fields, associating math with negative concepts in 6 cases out of 10. We observe significant differences across OpenAI’s models: newer versions (i.e., GPT-4) produce 5× semantically richer, more emotionally polarized perceptions with fewer negative associations compared to older versions and $N=159$ high-school students. These findings suggest that advances in the architecture of LLMs may lead to increasingly less biased models that could even perhaps someday aid in reducing harmful stereotypes in society rather than perpetuating them.

Attribute-aware community discovery aims to find well-connected communities that are also homogeneous w.r.t. the labels carried by the nodes. In this work, we address such a challenging task presenting Eva, an algorithmic approach designed to maximize a quality function tailoring both structural and homophilic clustering criteria. We evaluate Eva on several real-world labeled networks carrying both nominal and ordinal information, and we compare our approach to other classic and attribute-aware algorithms. Our results suggest that Eva is the only method, among the compared ones, able to discover homogeneous clusters without considerably degrading partition modularity.We also investigate two well-defined applicative scenarios to characterize better Eva: i) the clustering of a mental lexicon, i.e., a linguistic network modeling human semantic memory, and (ii) the node label prediction task, namely the problem of inferring the missing label of a node.

The mental lexicon is a complex cognitive system representing information about the words/concepts that one knows. Over decades psychological experiments have shown that conceptual associations across multiple, interactive cognitive levels can greatly influence word acquisition, storage, and processing. How can semantic, phonological, syntactic, and other types of conceptual associations be mapped within a coherent mathematical framework to study how the mental lexicon works? Here we review cognitive multilayer networks as a promising quantitative and interpretative framework for investigating the mental lexicon. Cognitive multilayer networks can map multiple types of information at once, thus capturing how different layers of associations might co-exist within the mental lexicon and influence cognitive processing. This review starts with a gentle introduction to the structure and formalism of multilayer networks. We then discuss quantitative mechanisms of psychological phenomena that could not be observed in single-layer networks and were only unveiled by combining multiple layers of the lexicon: (i) multiplex viability highlights language kernels and facilitative effects of knowledge processing in healthy and clinical populations; (ii) multilayer community detection enables contextual meaning reconstruction depending on psycholinguistic features; (iii) layer analysis can mediate latent interactions of mediation, suppression, and facilitation for lexical access. By outlining novel quantitative perspectives where multilayer networks can shed light on cognitive knowledge representations, including in next-generation brain/mind models, we discuss key limitations and promising directions for cutting-edge future research.

Unveiling the homophilic/heterophilic behaviors that characterize the wiring patterns of complex networks is an important task in social network analysis, often approached studying the assortative mixing of node attributes. Recent works have underlined that a global measure to quantify node homophily necessarily provides a partial, often deceiving, picture of the reality. Moving from such literature, in this work, we propose a novel measure, namely Conformity, designed to overcome such limitation by providing a node-centric quantification of assortative mixing patterns. Different from the measures proposed so far, Conformity is designed to be path-aware, thus allowing for a more detailed evaluation of the impact that nodes at different degrees of separations have on the homophilic embeddedness of a target. Experimental analysis on synthetic and real data allowed us to observe that Conformity can unveil valuable insights from node-attributed graphs.

Multi-scale strategies to estimate mixing patterns are meant to capture heterogeneous behaviors among node homophily, but they ignore an important addendum often available in real-world networks: the time when edges are present and the time-varying paths that edges form accordingly. In this work, we go beyond the assumption of a static network topology to propose a multi-scale, path- and time-aware node homophily estimator specifically tied for feature-rich stream graphs: $$\Delta $$Δ -Conformity. Our measure can capture the homogeneous/heterogeneous tendency of nodes’ connectivity along a period of time $$\Delta $$Δ starting from a given moment in time. Results on face-to-face interaction networks suggest it is possible to track changes in social mixing behaviors that coincide with contextually reasonable everyday patterns, e.g., medical staff disassortative behavior when exposed to patients. In a different domain, that of the Bitcoin Transaction Network, we capture relationships between the quantity of money sent from (and to) different categories/continents and their respective mixing trends over time. All these insights help us to introduce $$\Delta $$Δ -Conformity as a suitable solution for understanding temporal homophily by capturing the mixing tendency of entities embedded in fine-grained evolving contexts.

Conceptual associations influence how human memory is structured: Cognitive research indicates that similar concepts tend to be recalled one after another. Semantic network accounts provide a useful tool to understand how related concepts are retrieved from memory. However, most current network approaches use pairwise links to represent memory recall patterns (e.g. reading “airplane” makes one think of “air” and “pollution”, and this is represented by links “airplane”-“air” and “airplane”-“pollution”). Pairwise connections neglect higher-order associations, i.e. relationships between more than two concepts at a time. These higher-order interactions might covariate with (and thus contain information about) how similar concepts are along psycholinguistic dimensions like arousal, valence, familiarity, gender and others. We overcome these limits by introducing feature-rich cognitive hypergraphs as quantitative models of human memory where: (i) concepts recalled together can all engage in hyperlinks involving also more than two concepts at once (cognitive hypergraph aspect), and (ii) each concept is endowed with a vector of psycholinguistic features (feature-rich aspect). We build hypergraphs from word association data and use evaluation methods from machine learning features to predict concept concreteness. Since concepts with similar concreteness tend to cluster together in human memory, we expect to be able to leverage this structure. Using word association data from the Small World of Words dataset, we compared a pairwise network and a hypergraph with $N= 3586$N = 3586 concepts/nodes. Interpretable artificial intelligence models trained on (1) psycholinguistic features only, (2) pairwise-based feature aggregations, and on (3) hypergraph-based aggregations show significant differences between pairwise and hypergraph links. Specifically, our results show that higher-order and feature-rich hypergraph models contain richer information than pairwise networks leading to improved prediction of word concreteness. The relation with previous studies about conceptual clustering and compartmentalisation in associative knowledge and human memory are discussed.