Table 2 Vitality proxy data for this study

Pedestrian counts

In-situ videos

Directly captures human movement in the area.

Includes numerous residents and commuters unrelated to exterior appeal, thus introducing noise and biasing VV measurement. Numerous roads and alleys make it challenging to capture all street spaces, leading to potential gaps in data.

Geolocation footprints

Third-party data service

Provides large-scale movement data over time.

Lacks pedestrian group differentiation (e.g., visitor, resident, and commuter), leading to noisy data. Spatial inaccuracies due to poor satellite signals in narrow and dense alleys reduce reliability.

User-generated content on social medias

Meituan/Dianping

Large database of user reviews and images from restaurants, hotels, and entertainment venues.

Primarily focused on indoor activities (e.g., food, interiors), making it less relevant for exterior appeal analysis.

Tiktok

Users document a wide range of experiences in spaces, reflecting their preferences for specific locations.

Compared to taking and uploading photos, recording and sharing videos on TikTok is relatively less convenient, resulting in a significantly smaller volume of video content on Tiktok than images on other image-based social media platforms, which undermines the representativeness of the videos.

RedNote

Provides numerous user-generated content, including detailed images of building facades and storefronts, which can offer direct insights into visitor interests and footprints.

The platform is predominantly used by females, who account for 79.13% of the total user base⁹¹. In terms of age distribution, users are largely concentrated within the 18–34 age range, with the 18–24 and 25–34 groups comprising 39.21% and 38.65% of all users, respectively⁹². While these figures refer to gender and age independently, their combined dominance suggests a demographic skew that may limit the representativeness of the data and introduce potential bias.