NLP-Based Trust Signal Detection from 360-Degree Narrative Feedback Using Fine-Tuned BERT and Behavioral Authenticity Analysis

Natural Language Processing, BERT, Trust Assessment, 360-Degree Feedback, Sentiment Analysis, LDA, Behavioral Authenticity, DEEP Framework, Competency Mapping.

This paper outlines a comprehensive NLP pipeline for extracting trust-relevant signals from unstructured 360-degree narrative feedback using the DEEP (Drive, Engage, Empower, Purpose-align) Leadership Framework. The pipeline integrates three modules: (1) a fine-tuned BERT-based trust-specific sentiment classifier achieving 89.4% accuracy, outperforming generic sentiment by 16.0 percentage points; (2) guided LDA topic modeling mapping narrative themes to 53 DEEP competencies with 0.834 mean alignment; and (3) a novel Behavioral Authenticity Detection (BAD) module differentiating genuine from performative behaviors through six linguistic markers (ACS correlation with multi-rater consensus: r=0.71, p<.001). Validated on 24,800 comments from N=12,400 assessments across 14 organizations, NLP integration improves Trust & Leadership Index prediction from R²=0.921 to 0.943 (ΔR²=0.022, p<.001). Three algorithms are presented: DEEP-NLP Pipeline, Trust-BERT Fine-Tuning Protocol, and BAD Score Computation. Results show that domain-specific NLP notably advances leadership trust assessment exceeding traditional quantitative methods.