Dongkwon Seong Lab

The study introduces SOURCE-CT, a controlled data-source contamination benchmark built from 90 heritage cases across Korea, China, and Japan. It crossed two evidence conditions, six prompt variants, and three deployment conditions to produce 3,240 outputs, with 1,620 mixed-source outputs forming the primary inferential sample. Each contaminated packet contained verified contextual material plus one independently checked, plausible but unsupported claim. Outputs were adjudicated for false-claim acceptance, supported by a 20% blind reliability sample, and analysed using Bayesian hierarchical logistic regression with supplementary behavioural and robustness probes.

The paper reframes AI hallucination in heritage communication as a source-boundary problem and distinguishes three stages: source access, source processing, and source accountability. It defines a source-accountability gap when source-informed output fails to preserve evidentiary limits, and identifies three forms of boundary failure: source displacement, certainty inflation, and narrative absorption. SOURCE-CT operationalises claim-level evidentiary-boundary preservation as a measurable dimension of grounded heritage generation.

The study audited 15 initial design variables and retained four for computation: shell-edge radius (V06), pre-braking trigger distance (V09), translational speed (V11), and commanded deceleration (V12). A constrained Latin hypercube sampling procedure filtered infeasible combinations using a braking-distance relation, while the derived warning-margin proxy I5 was tested across fixed seeds for feasibility, distributional stability, four-dimensional coverage, cluster tendency, representative-coordinate stability, and distance-definition sensitivity.

The paper reframes product-level biomimetics as an evidence-tiered translation problem. Direct functional analogs, qualitative biological analogies, and ordinary engineering constraints are separated rather than collapsed into a single numerical ranking. By replacing unsupported multi-criteria optimization with constrained exploration and explicit diagnostic gates, the framework treats uncertainty and failed convergence as reportable knowledge rather than as defects to be hidden.

A within-participant paired survey design collected 891 valid responses evaluating two horological museums in Seoul and Yantai. The study applied exploratory and confirmatory factor analysis, paired-samples t-tests, and multiple regression to test the proposed Chrono-Spatial Experience Model (CSEM).

This study reconceptualizes exhibition space as an active organizer of subjective time rather than a passive container for artifacts, and proposes the Chrono-Spatial Experience Model (CSEM) as a design-scale framework linking circulation, wayfinding, and comfort to visitors' felt immersion in time.

A mixed exploratory design combining PLS-SEM and fsQCA analysed 728 valid questionnaires. Three purpose-built advertising stimuli — print, motion poster, and an AI-generated video — were produced in-house to avoid prior brand familiarity.

The study extends anthropomorphism theory into sustainable consumption and recasts perceived risk as a mediating mechanism rather than a standalone antecedent, offering a configurational account of why avoidance arises from combinations of conditions rather than any single factor.

A mixed-method approach combining structural equation modeling (SmartPLS) and fsQCA was applied to 497 participants, supported by AI-generated dynamic exhibits using deep learning, LoRA-based image synthesis, and voice cloning technologies.

This study reconceptualizes museum exhibits as active narrative subjects rather than passive objects and establishes a multimodal anthropomorphism framework integrating visual, linguistic, and auditory dimensions within narrative transportation theory.

The study employs a combined methodology of case study analysis and theoretical interpretation, examining representative examples across architecture, product design, fashion, and ecological systems to identify underlying biomimetic principles.

This study reconceptualizes biomimetic design as a holistic, system-level framework that unifies science, art, culture, and ecology — extending its scope beyond engineering into the domain of artistic and cultural production.