Interior style generation requires expressive visual transformation while preserving spatial layout, object boundaries, and semantic relationships. Existing generative and style-transfer methods have improved indoor scene synthesis, but they often suffer from boundary drift, furniture deformation, texture leakage, and unstable style representation when stronger style signals are introduced. To address this problem, this study proposes a semantic-conditional generative adversarial network framework for structure-preserving and controllable interior style generation. The framework constructs semantic masks, boundary maps, and semantic embeddings from indoor images and injects these semantic priors into the generator feature stream through a multi-scale conditional mechanism. A style encoding network is further introduced to represent interior style characteristics and to modulate image-level visual appearance features, including texture patterns, color tones, material-like surface appearance, and illumination-related visual cues, in a controllable manner. The model was trained and evaluated using 5000 selected indoor images from SUN RGB-D and ADE20K, together with a self-constructed style reference set of 600 images covering modern minimalist, Nordic, industrial, and neoclassical interiors. Across baseline comparisons and ablation analyses, the proposed framework achieved a semantic region consistency score of 0.91, maintained higher semantic boundary consistency under increasing style intensity, obtained style consistency scores ranging from 0.88 to 0.90, and achieved an LPIPS score of 0.128. A subjective evaluation with 20 participants also showed higher ratings for realism, style expression, and structural consistency. These results provide quantitative and perceptual evidence that semantic-conditional injection can improve the balance between spatial-semantic preservation and controllable image-level style expression in AI-assisted interior visualization.

Sustainability literacy is increasingly invoked in construction and planning research, yet it is most often framed as an educational construct concerned with awareness, knowledge, and attitudes. This framing provides limited explanatory power for understanding how sustainability values are translated into in real-world planning decisions, particularly under conditions of uncertainty and value conflict. In parallel, artificial intelligence (AI) has been introduced into planning practice largely as an optimization-driven analytical tool, reinforcing instrumental conceptions of rationality. This study reconceptualizes sustainability literacy as a decision capability and develops an AI-enabled theoretical framework that positions AI as a cognitive partner in sustainability-oriented construction planning. Methodologically, the study adopts a conceptual research design grounded in a systematic interdisciplinary literature synthesis spanning planning theory, circular economy, social sustainability, and AI-enabled decision support, combined with theory-building and framework development procedures. The proposed framework clarifies how human judgment can be cognitively augmented through AI-supported interpretation, trade-off exploration, and value-informed deliberation, thereby reframing sustainability as an internal driver of planning judgment rather than an external performance criterion. By conceptualizing human–AI collaboration as an iterative, reflective process, the study establishes a coherent theoretical basis for context-sensitive sustainability planning in the built environment, with implications for decision-support system design, planning practice, and professional education.

Conservation interventions in historic buildings require quantitative assessment to balance preservation needs and minimal intervention principles. This study presents a quantitative framework for minimum intervention assessment of heritage conservation, based on 133 documented interventions at Caishen Temple in Xinzhou across four periods: 1992, 2006, 2016, and 2017. Four dimensions were defined: Extent of Intervention (EI), Reversibility (RE), Information Loss (IL), and Necessity (NE). Expert scoring on a five-point Likert scale yielded high inter-rater reliability (ICC: 0.79–0.88). The entropy weight method was then used to derive data-driven weights from the expert-scoring matrix, $\mathit{IL}=0.3149$, $\mathit{EI}=0.3025$, $\mathit{NE}=0.2572$, $\mathit{RE}=0.1253$, and the Intervention Intensity Index ($\mathit{III}$) was calculated for each intervention. A random forest model was further developed as an exploratory cross-check, with 19 problematic interventions labelled as $\mathit{y}=1$ and 114 normal interventions as $\mathit{y}=0$. Target labels were defined through a dual-source procedure combining SSIM- and HSV-based image-similarity assessment for interventions with paired pre- and post-restoration photographs and explicit textual evidence from conservation records for interventions without paired photographs. Feature importance and SHAP analyses indicated that reversibility and necessity had the greatest discriminative importance for distinguishing problematic from normal interventions in this dataset, whereas the entropy-based ranking assigned higher weights to information loss and extent of intervention. This descriptive contrast, based on only four dimensions, highlights the distinction between data dispersion and discriminative capacity. Spatiotemporal analysis indicated that roof and rafter components exhibited the highest intervention intensity, while 1992 interventions contained the largest proportion of problematic practices. This study combines entropy-derived weighting with a machine-learning-based exploratory cross-check, providing a transparent, case-based framework for evidence-based conservation decision-making at historic heritage sites.