Textual forma mentis networks bridge language structure, emotional content and psychopathology levels in adolescents

TL;DR

A network-based AI framework for predicting dimensions of psychopathology in adolescents using natural language and the potential of cognitive network approaches to capture meaningful links between psychopathology and language use in adolescents is introduced.

Abstract

We introduce a network-based AI framework for predicting dimensions of psychopathology in adolescents using natural language. We focused on data capturing psychometric scores of social maladjustment, internalizing behaviors, and neurodevelopmental risk, assessed in 232 adolescents from the Healthy Brain Network. This dataset included structured interviews in which adolescents discussed a common emotion-inducing topic. To model conceptual associations within these interviews, we applied textual forma mentis networks (TFMNs)-a cognitive/AI approach integrating syntactic, semantic, and emotional word-word associations in language. From TFMNs, we extracted network features (semantic/syntactic structure) and emotional profiles to serve as predictors of latent psychopathology factor scores. Using Random Forest and XGBoost regression models, we found significant associations between language-derived features and clinical scores: social maladjustment (r = 0.37, p < .01), specific internalizing behaviors (r = 0.33, p < .05), and neurodevelopmental risk (r = 0.34, p < .05). Explainable AI analysis using SHAP values revealed that higher modularity and a pronounced core-periphery network structure-reflecting clustered conceptual organization in language-predicted increased social maladjustment. Internalizing scores were positively associated with higher betweenness centrality and stronger expressions of disgust, suggesting a linguistic signature of rumination. In contrast, neurodevelopmental risk was inversely related to local efficiency in syntactic/semantic networks, indicating disrupted conceptual integration. These findings demonstrated the potential of cognitive network approaches to capture meaningful links between psychopathology and language use in adolescents.

Figures (7)

• Figure 1: A,B,C: Algorithm steps for creating a textual forma mentis from an interview transcript. D: Two sample TFMNs illustrating differences in density, modular organization (more pronounced in b), and central node connectivity. Each TFMN represents the transcript text of a single interview provided by an individual in the dataset. In TFMNs, negative (positive) words are highlighted in red (cyan). Links indicate syntactic relationships and are cyan (red, purple) if between positive (negative, negative and positive) words.
• Figure 2: SHAP analysis of feature contributions for predicting Social Maladjustment with GBM. The dataset was scaled to $[-5, 5]$. (a) Beeswarm plot showing the relationship between observed feature values (y-axis) and their impact on the model output (x-axis). (b) Mean absolute SHAP values for each feature, indicating their average importance across the dataset. (c) Heatmap of SHAP values for all instances, with blue indicating negative contributions and red indicating positive contributions to the model output.
• Figure 3: SHAP analysis of feature contributions for predicting Specific Internalising with GBM. The dataset was scaled to $[-5, 5]$. (a) Beeswarm plot showing the relationship between observed feature values (y-axis) and their impact on the model output (x-axis). (b) Mean absolute SHAP values for each feature, indicating their average importance across the dataset. (c) Heatmap of SHAP values for all instances, with blue indicating negative contributions and red indicating positive contributions to the model output.
• Figure 4: SHAP analysis of feature contributions for predicting Neurodevelopmental Risk with GBM. The dataset was scaled to $[-5, 5]$. (a) Beeswarm plot showing the relationship between observed feature values (y-axis) and their impact on the model output (x-axis). (b) Mean absolute SHAP values for each feature, indicating their average importance across the dataset. (c) Heatmap of SHAP values for all instances, with blue indicating negative contributions and red indicating positive contributions to the model output.
• Figure S1: SHAP analyses of feature contributions for predicting Social Maladjustment with RFR. Model output is explained as an additive contribution of features.