Early Linguistic Pattern of Anxiety from Social Media Using Interpretable Linguistic Features: A Multi-Faceted Validation Study with Author-Disjoint Evaluation

TL;DR

The paper tackles scalable anxiety screening from social media by proposing an interpretable set of 13 linguistic features trained with a transparent logistic regression model. It emphasizes strict author-disjoint evaluation, keyword robustness testing, cross-domain validation with clinical data (DAIC-WOZ), and early detection from minimal posting history. The findings show strong performance (F1 around 0.89, AUC ~0.95), dominance of self-referential language as a key signal, and substantial cross-domain consistency, supporting the generalizability and clinical relevance of the approach. With explicit ethical safeguards and careful interpretation, this framework has potential to complement traditional mental health screening and enable earlier intervention, while acknowledging limitations such as synthetic control usage and platform-specific validation.

Abstract

Anxiety affects hundreds of millions of individuals globally, yet large-scale screening remains limited. Social media language provides an opportunity for scalable detection, but current models often lack interpretability, keyword-robustness validation, and rigorous user-level data integrity. This work presents a transparent approach to social media-based anxiety detection through linguistically interpretable feature-grounded modeling and cross-domain validation. Using a substantial dataset of Reddit posts, we trained a logistic regression classifier on carefully curated subreddits for training, validation, and test splits. Comprehensive evaluation included feature ablation, keyword masking experiments, and varying-density difference analyses comparing anxious and control groups, along with external validation using clinically interviewed participants with diagnosed anxiety disorders. The model achieved strong performance while maintaining high accuracy even after sentiment removal or keyword masking. Early detection using minimal post history significantly outperformed random classification, and cross-domain analysis demonstrated strong consistency with clinical interview data. Results indicate that transparent linguistic features can support reliable, generalizable, and keyword-robust anxiety detection. The proposed framework provides a reproducible baseline for interpretable mental health screening across diverse online contexts.

Figures (9)

• Figure 1: Main classification performance metrics showing accuracy, precision, recall, F1 score, and ROC-AUC with 95% confidence intervals.
• Figure 2: ROC curves for the anxiety detection model showing high discriminative performance (AUC = 0.9500).
• Figure 3: Learning curves showing model performance as a function of training set size, demonstrating stable convergence and minimal overfitting.
• Figure 4: Confusion matrix showing the distribution of true positives, true negatives, false positives, and false negatives for the test set.
• Figure 5: Feature importance ranked by standardized logistic regression coefficients. First-person pronoun rate dominates as the strongest predictor of anxiety expression.