Youngju Cho

Ms. Youngju Cho

orcid

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
31 July 2023 in Journal of Retailing and Consumer Services
Verified
Journal Article
Closed access
JPJunsung Park
Junsung Park
JYJoon Woo Yoo
Joon Woo Yoo
YCYoungju Cho
Youngju Cho
HPHeejun Park
Heejun Park

14 January 2026 in IEEE Access
Verified
Journal Article
Open access
YCYoungju Cho
Youngju Cho
HPHeejun Park
Heejun Park
JPJunsung Park
Junsung Park
Evaluating the user experience of complex, large-scale sociotechnical systems is a significant data science challenge, as traditional survey methods often fail to capture authentic user perceptions. This study proposes and validates a novel computational framework that combines Structural Topic Modeling (STM) and SHapley Additive exPlanations (SHAP) to mine and interpret unstructured user-generated data at scale. We apply this framework to comparatively analyze consumer satisfaction with three dominant service modalities: AI agents, human service providers, and standardized self-service technology (SST). By analyzing a large-scale dataset of 260,577 user reviews, our approach first uses STM to discover 100 distinct dimensions of user experience, which are then mapped to an integrated Technology-System-User (TSU) model. Second, an XGBoost model was used to predict satisfaction, and SHAP analysis was used to identify the precise drivers of satisfaction for each modality. The findings reveal three divergent paths to a positive user experience: human-led services rely on interpersonal connections; AI services on a strong price-value proposition and a non-judgmental safe space; and standardized systems on gamified engagement. This study contributes a robust, interpretable, and scalable methodological framework for analyzing user feedback, offering actionable data-driven insights for the design and optimization of complex human-in-the-loop and AI-powered digital systems.

30 September 2026 in Sensors
Verified
Journal Article
Open access
PDF file
YCYoungju Cho
Youngju Cho
JHJiyoon Hong
Jiyoon Hong
HPHeejun Park
Heejun Park
Vision-based sensing is increasingly important for liquefied natural gas (LNG) tank manufacturing, where missed defects in cable-related components can affect monitoring, control, and safety functions. However, industrial image datasets often contain duplicate or near-duplicate imagery, operating thresholds can be optimistically tuned on held-out test data, and abnormal labels are costly to obtain. This study evaluates LNG tank cable defect detection under a duplicate-aware and calibration-controlled protocol that separates model training, threshold calibration, and held-out testing. Using the AI-Hub LNG tank quality inspection image dataset, we compare ResNet-18, EfficientNet-B0, and ConvNeXt-Tiny supervised classifiers; PatchCore, PaDiM, and STFPM normal-only training anomaly detectors; and ResNet-18 training-label-budget experiments that retain all normal training images while varying the number of labeled abnormal training images used for model fitting. Across 36 de-duplicated final runs, EfficientNet-B0 achieved the strongest fully supervised performance (AUROC 0.9999±0.0001, F1 0.9956±0.0029). PatchCore was the best normal-only training anomaly detector (AUROC 0.8909±0.0109, F1 0.8666±0.0228). In the percentage-budget experiment, 1% labeled abnormal training samples already exceeded PatchCore in AUROC, F1, and false-negative rate under the fixed labeled-calibration protocol; in a finer 4–64 abnormal-training-label sweep with 90 runs, the first training budget exceeding PatchCore across AUROC, F1, and false-negative rate with 16 labeled abnormal training images. These results quantify abnormal training-label efficiency under a fixed labeled-calibration protocol, rather than the total abnormal annotation burden required to deploy the complete operating classifier.

1 October 2026 in IEEE Transactions on Computational Social Systems
Verified
Journal Article
Closed access
YCYoungju Cho
Youngju Cho
HPHeejun Park
Heejun Park

YCYoungju Cho
Youngju Cho
JYJunyoung Yoo
Junyoung Yoo
JYJoon Woo Yoo
Joon Woo Yoo
HPHeejun Park
Heejun Park
Online reviews are increasingly pivotal in consumer decision-making, with platforms employing mechanisms such as badges to denote reviewer credibility. Although prior research has examined the influence of expert reviewers, it has typically treated badge holders as a homogeneous group, overlooking how variation in tenure within expert tiers shapes the way readers process review content. This article examines how Yelp Elite badge tenure, operationalized as Red (1–4 years), Gold (5–9 years), and Black (10+ years) tiers and treated as a proxy for accumulated platform-recognized expertise, moderates the effects of peripheral cues (Extremity, Length) and central cues (Readability, Subjectivity, and Plutchik’s eight emotions) on perceived helpfulness within an Elaboration Likelihood Model (ELM) framework. The analysis draws on the full population of 324,426 restaurant reviews authored by Yelp Elite badge holders between 2019 and 2021, using a pooled count-model specification with badge tier as a categorical moderator. The primary specification is estimated using Poisson quasi-maximum likelihood with HC1-robust standard errors, and full negative binomial estimation is reported as a robustness check. Wald tests indicate that badge tenure significantly moderates eight of twelve cue–helpfulness relationships (χ2(24)=4938, p<0.001). The effect of readability is monotonically positive and increases sharply with tenure, while the effect of joy varies across tenure groups. These findings suggest that reviewer expertise signals are not monolithic, refining theoretical insights on how tenure-based credibility cues moderate cue processing and offering practical implications for review platform management. The findings also indicate that platforms applying uniform ranking or surfacing rules across all Elite reviewers risk misallocating visibility, and that tenure-conditional weighting of textual cues warrants consideration.