Parasocial relationships in informal digital beauty education: explainable machine learning evidence from YouTube beauty influencers

Social media beauty influencers increasingly function as informal sources of procedural, product, and aesthetic knowledge, yet the relational mechanisms connecting audiences to these digital sources remain underexplored. This study examines parasocial relationship (PSR) strength in the Beauty & Fashion subset of a multimodal YouTube influencer dataset and introduces RAPID-PSR, a relational-aware, pseudo-labeled, interpretable ordinal ensemble framework. The dataset comprised 3,329 observations, including 819 usable human-annotated cases categorized as absent, weak, moderate, or strong PSR, and 2,483 application observations. RAPID-PSR combines behavioral and creator-context variables, engineered relational features, TF-IDF-based text representations, cumulative ordinal learning, an adaptive XGBoost-LightGBM-CatBoost ensemble, confidence-based pseudo-labeling, and ordinal SHAP explanations. An independent test partition (n = 158) was retained for evaluation. RAPID-PSR achieved 66.46% accuracy, macro-F1 = 0.553, quadratic weighted kappa = 0.647, ordinal MAE = 0.380, macro ROC-AUC = 0.810, and macro PR-AUC = 0.590. It produced the highest accuracy, macro-F1, QWK, and lowest ordinal error, while CatBoost achieved the strongest ROC-AUC and log-loss performance. SHAP analysis identified textual semantics, interaction depth, reciprocal engagement, interaction regularity, relational warmth, and sentiment as major predictors of PSR strength. Survey-linked validation further showed substantial convergence between computational and self-reported PSR scores (Pearson r = 0.666; Spearman ρ = 0.644). The findings highlight relational continuity and responsive interaction as important signals in creator-led informal beauty learning while emphasizing the distinction between interpersonal trust and epistemic reliability.

Authors

Institutions

Publication Details

Journal
Interpersona An International Journal on Personal Relationships
Published
2026-11-13
DOI
https://doi.org/10.54899/ijpr.v20i4.2297
Primary Topic
Digital Marketing and Social Media
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Parasocial relationships in informal digital beauty education: explainable machine learning evidence from YouTube beauty influencers

Trisnani Widowati, Septian Eko Prasetyo, Anik Maghfiroh, Clarita Aprilliani et al.
Interpersona An International Journal on Personal Relationships
Digital Marketing and Social Media
article

Parasocial relationships in informal digital beauty education: explainable machine learning evidence from YouTube beauty influencers

Trisnani Widowati, Septian Eko Prasetyo, Anik Maghfiroh, Clarita Aprilliani, Ade Novi Nurul Ihsani
article en

Abstract

Social media beauty influencers increasingly function as informal sources of procedural, product, and aesthetic knowledge, yet the relational mechanisms connecting audiences to these digital sources remain underexplored. This study examines parasocial relationship (PSR) strength in the Beauty & Fashion subset of a multimodal YouTube influencer dataset and introduces RAPID-PSR, a relational-aware, pseudo-labeled, interpretable ordinal ensemble framework. The dataset comprised 3,329 observations, including 819 usable human-annotated cases categorized as absent, weak, moderate, or strong PSR, and 2,483 application observations. RAPID-PSR combines behavioral and creator-context variables, engineered relational features, TF-IDF-based text representations, cumulative ordinal learning, an adaptive XGBoost-LightGBM-CatBoost ensemble, confidence-based pseudo-labeling, and ordinal SHAP explanations. An independent test partition (n = 158) was retained for evaluation. RAPID-PSR achieved 66.46% accuracy, macro-F1 = 0.553, quadratic weighted kappa = 0.647, ordinal MAE = 0.380, macro ROC-AUC = 0.810, and macro PR-AUC = 0.590. It produced the highest accuracy, macro-F1, QWK, and lowest ordinal error, while CatBoost achieved the strongest ROC-AUC and log-loss performance. SHAP analysis identified textual semantics, interaction depth, reciprocal engagement, interaction regularity, relational warmth, and sentiment as major predictors of PSR strength. Survey-linked validation further showed substantial convergence between computational and self-reported PSR scores (Pearson r = 0.666; Spearman ρ = 0.644). The findings highlight relational continuity and responsive interaction as important signals in creator-led informal beauty learning while emphasizing the distinction between interpersonal trust and epistemic reliability.

Interpersona An International Journal on Personal RelationshipsVol. 20(4)
State University of Semarang (ID)
Digital Marketing and Social Media
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon