Perceived Information Assurance in Conversational Systems
Information assurance (IA) in AI-driven conversational systems like ChatGPT remains understudied from a user perception perspective. This study develops and validates a perception-based IA model grounded in key IA pillars - confidentiality, integrity, availability, authenticity, and non-repudiation...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | , , |
| Format: | article |
| Published: |
2026
|
| Online Access: | http://hdl.handle.net/11675/14439 https: |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Information assurance (IA) in AI-driven conversational systems like ChatGPT remains understudied from a user perception perspective. This study develops and validates a perception-based IA model grounded in key IA pillars - confidentiality, integrity, availability, authenticity, and non-repudiation - and examines the role of digital literacy as an antecedent within an AI conversational context. Using data from 924 participants, we employed a dual-method analytical approach: 1) Partial Least Squares Structural Equation Modeling (PLS-SEM) to test hypothesized relationships; and 2) Unsupervised Machine Learning (ML) with feature engineering for user segmentation and predictive modeling. The measurement model ensured acceptable reliability and validity, with all constructs exceeding acceptable thresholds. PLS-SEM results revealed that digital literacy significantly predicts integrity (? =.484 ), availability (? =.481 ), authenticity (? =.455 ), and non-repudiation (? =.392 ), which in turn positively influence IA through confidentiality (? =.425 ). Integrity emerged as the strongest predictor of IA, with the model explaining 52.9% of the variance in IA (R2 =.529 ). Mediation analysis confirmed significant indirect effects of digital literacy on IA through all mediators. Complementing this, unsupervised clustering identified three distinct IA perception segments with meaningful differences in usage intensity. Through advanced feature engineering (117 features) and nature-inspired optimization algorithms, we achieved 98.38% classification accuracy. Feature selection demonstrated that comparable performance (96.76% accuracy) could be maintained with only 19 optimally selected features, representing an 83.8% reduction in model complexity. These integrated findings highlight the multifaceted nature of IA perceptions, the central role of digital literacy, and provide a parsimonious framework for predicting user trust for designing visible IA mechanisms and digital literacy support in conversational AI systems. |
|---|