Research on the Perception Mismatch of Multiple Stakeholders in Street Environments Based on Actor-Network Theory and the CLIP Model: A Case Study of the South Axis of Beijing

Resolving spatial perception mismatches among diverse stakeholders along historic linear heritage streets is essential for human-oriented urban renewal. However, traditional approaches often suffer from spatial discontinuity or fail to capture subjective–objective tensions. Under the lens of Actor-Network Theory (ANT), this study constructs a dual-track empirical framework integrating multi-modal deep learning with semantic differential (SD) questionnaires. Taking Beijing’s South Axis (5.9 km) as a case study, 140 continuous street-view sampling points were quantified using the Contrastive Language–Image Pre-training (CLIP) model, while 200 SD questionnaires were collected from tourists, residents, and commuters. Empirical results reveal a prominent mismatch characterized by strong macro-cultural resonance paired with suppressed micro-daily convenience. While “sense of culture” achieved strong alignment ( = +0.05), public convenience and physical walkability bottomed out. Inferential ANOVA tests (F = 31.50, p < 0.001) confirmed significant stakeholder divergences in spatial interest. Pearson tests showed street trees exerted the highest positive spillover on comfort (r = 0.535), whereas pedestrian density was the main negative variable (r = −0.614). Refined micro-renewal strategies focusing on facility patching and right-of-way protection are proposed for inclusive heritage governance.