Discriminative features and contextual cognitive bias effects between robotic and natural handwriting.

Abstract

This study aims to systematically identify the discriminative features between robotic and natural handwriting and explore how contextual conditions influence the classification accuracy of two groups of discriminators: forensic document examiners and nonexaminers. Repeated measures ANOVA was utilized to analyze the main effect of contextual conditions (blind classification, informed quantity context, paired comparison context) as well as the interaction effect between context and group. Bonferroni-corrected post hoc tests were employed for pairwise comparisons across different conditions, whereas correlation analysis was conducted to examine the relationship between confidence level and classification accuracy, with a significance level set at α = 0.05. The results indicate that robotic and natural handwriting exhibit significant differences in five core features: ink distribution, stroke stability, pressure marks, starting/ending stroke characteristics, and character consistency. Forensic document examiners achieved notably higher classification accuracy (ranging from 82.00% to 87.33%) compared to nonexaminers (ranging from 67.33% to 76.67%). This superior performance can be attributed to their reliance on objective professional features and their ability to resist contextual bias. Contextual cognitive bias exerted a significant adverse impact on nonexaminers, particularly in the form of confirmation bias within the informed quantity context, but had minimal influence on professional examiners. The paired sample comparison method effectively enhanced the classification accuracy of both groups by strengthening the perception of subtle discriminative features and reducing dependence on macroscopic cues that are vulnerable to contextual interference. These findings offer important theoretical support and practical guidance for forensic handwriting examination.