Emoji fingerprints: topic-independent authorship verification using 20 novel emoji-based stylometric features

Authorship verification in social media text is a growing challenge in digital forensics, yet existing stylometric approaches rely on textual features that degrade substantially when reference and questioned texts originate from different communicative contexts. This study investigates whether emoji usage patterns-a pervasive but underexplored dimension of digital communication-can serve as viable stylometric markers for forensic authorship verification. We propose 20 novel emoji-based features organized into six thematic categories and evaluate them on a large-scale dataset of 4697 Reddit authors comprising 1023,826 comments (19.9% emoji-bearing, 325,746 emoji instances across 2604 unique types). Three models-emoji-only (20 features), classic stylometric baseline (27 features), and hybrid (47 features)-are compared across 49 configurations varying profile and test sample sizes from 5 to 500 comments. The emoji-only model achieves a diagonal accuracy range of 55.8-94.9%, the classic model 71.8-96.3%, and the hybrid model 72.4-98.2%, with the hybrid significantly outperforming the classic baseline at moderate sample sizes (concordant on the Holm-corrected McNemar test, the paired t-test, and likelihood-ratio cost at P = 25-100). Crucially, cross-subreddit robustness analysis reveals that emoji features maintain virtually identical accuracy across topical boundaries (Δ = -0.009), whereas classic features suffer a 26.6 %age point degradation-demonstrating that emoji features capture topic-independent authorial behavior resistant to the domain shifts that undermine traditional stylometry. Ablation analysis indicates that the distributional emoji features-corpus-level frequency divergences between authors-contribute the greatest non-redundant information to the hybrid model. Recasting the verification scores as calibrated likelihood ratios yielded well-calibrated forensic evidence, with the hybrid model achieving the lowest empirical cost (Cllr) at profile sizes up to 100 comments. These findings establish emoji-based stylometry as a complementary forensic tool with a distinctive advantage in cross-domain scenarios and provide empirically grounded data sufficiency thresholds for forensic casework.