Provenance-aware multimodal resource organization for smart library wellbeing services: A data-driven study of bibliotherapy-informed recommendations

Abstract Purpose: This study examines how the openly shared tabular portion of a publicly archived corpus of bibliotherapy-informed recommendations can be independently re-annotated and reorganized into a provenance-aware and media-sensitive framework for smart library wellbeing services. Methods: The analytical corpus comprises 391 unique resources and 447 source-occurrence records, all traceable to the publicly archived data package of X. Wang et al. (2026b). Target groups, seven problem categories, media types, and provenance layers were analyzed using descriptive statistics, contingency analysis, adjusted standardized residuals with Holm correction, correspondence analysis, and robustness checks. Results: Problem category was associated with media type (chi-square = 167.37, df = 24, p < 0.001; Cramer's V = 0.327). The association persisted after excluding all 108 records recoded from pending or mixed categories (chi-square = 170.99; V = 0.389), after excluding 18 reliability-sample items for which both coders agreed with each other but differed from the archived analytic label (chi-square = 167.18; V = 0.335), and after replacing those 18 labels with the coders' consensus assignments (chi-square = 168.25; V = 0.328). The strongest stable positive configurations linked anxiety and stress with books and music, growth and life education with films, and self-identity and personal value with picture books. An 80-item independent audit showed high inter-coder agreement (90.0%; Cohen's kappa = 0.880), while agreement with the frozen analytic field was lower (72.5% and 70.0% for the two coders). Conclusion: The study demonstrates a transparent re-annotation workflow for organizing a non-probability tabular subset of bibliotherapy-informed resources while making provenance, classification uncertainty, and corpus-construction effects visible. The framework is intended for non-clinical, human-reviewed smart library wellbeing services rather than therapeutic decision making.