📄 social_science

Tourists' Spatial Perception of Traditional Villages Based on UGC Data: A Case Study of Heshun Ancient Town, China

This paper utilizes a quantitative framework integrating multimodal User-generated content (UGC) data with spatial kernel density estimation to map tourists' emotional and visual perceptions of Heshun Ancient Town, providing an empirical basis for targeted micro-updates and preventing excessive commercialization in traditional Chinese villages.

Original authors: Yun Zhang, Xixuan Fan, Dongqiang Zhang, Xiaodong Lu

Published 2026-06-26

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Yun Zhang, Xixuan Fan, Dongqiang Zhang, Xiaodong Lu

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you want to understand how people truly feel about a historic village, but instead of asking them to fill out a boring survey, you decide to listen to their digital "diaries." That is exactly what this paper does.

The researchers studied Heshun Ancient Town in China, a place full of old buildings, narrow streets, and local culture. They wanted to know: Where do tourists actually go? What do they like? And what makes them unhappy?

To find out, they built a "digital detective" system using User-Generated Content (UGC). Think of this as gathering thousands of photos and posts from social media (like Weibo and Xiaohongshu) where people shared their trips between 2021 and 2025.

Here is how they cracked the case, broken down into simple steps:

1. The Two-Pronged Detective Work (Text + Images)

Most studies only look at what people say (text) or only what they show (photos). This study looked at both, like checking a person's diary and their photo album.

The Text Detective: They used a computer program (SnowNLP) to read thousands of reviews. It acted like a mood ring, scoring every sentence from "Very Sad" to "Very Happy."
The Photo Detective: They used a super-smart AI (called CLIP) to look at 17,000+ photos. Instead of just counting pixels, this AI understood what was in the picture. Did it show a temple? A mountain? A coffee shop? It also figured out the specific details, like "white walls" or "wooden beams."

2. Mapping the "Heat" of the Village

Once they knew what people liked and where they took photos, they needed to put it on a map.

They took every photo and tagged it to a specific spot in the village (like a specific bridge or a specific alley).
Then, they used a technique called Kernel Density Estimation (KDE). Imagine dropping hot sauce on a map. Where the sauce is thickest, that's where the "heat" is—meaning, that's where the most people are going and taking pictures. Where the sauce is thin, it's a "cold spot" or a "blind zone" that tourists are ignoring.

3. What Did They Find?

The results revealed some interesting patterns, almost like a story about the village's personality:

The "Star" Attraction (Historic Character): The biggest "heat" was on the old buildings, ancestral halls, and stone bridges. People loved the authentic look of the place. The photos and the happy words matched perfectly here.
The "Supporting Cast" (Nature): Mountains, forests, and water systems were the second most popular. They acted like a beautiful backdrop for the main show.
The "Mismatch" (Shops and Cafes): This was the most surprising part. People talked a lot about guesthouses and coffee shops in their text (high frequency), but they didn't take many photos of them, and they weren't as happy about them as they were about the history. It's like saying, "I love the food," but then only taking pictures of the scenery. The researchers call this a "perceptual trap"—modern shops are there, but they don't quite fit the old vibe, so tourists don't feel the same connection.
The "Blind Zones": Because the village has one main entrance and a few main paths, tourists mostly stick to the center. The edges of the village, where there are quiet workshops or hidden cultural spots, became "perceptual blind zones." Tourists literally didn't see them because the path didn't lead them there.

4. The Big Picture

The paper concludes that to save and improve these ancient villages, you can't just guess what to fix. You need data.

Don't just build more shops: The data shows that over-commercializing the area actually lowers satisfaction because it breaks the "magic" of the old town.
Open the doors: The village needs new paths to guide people away from the crowded center and toward the quiet, hidden corners so they can appreciate the whole village, not just the main street.

In short, the researchers used AI to turn thousands of messy social media posts into a clear, colorful map. This map shows exactly where the village is shining and where it's in the shadows, helping managers fix the problems without guessing.

Technical Summary: Tourists' Spatial Perception of Traditional Villages Based on UGC Data

Problem Statement
Traditional Chinese villages (TCVs) are vital components of cultural heritage and drivers of regional economic revitalization. However, rapid tourism expansion poses significant threats to their spatial patterns and cultural ecosystems, leading to issues such as excessive commercialization, resource homogenization, and the destruction of traditional spaces. Existing evaluation systems often rely on macro-level policies, qualitative analyses, or static questionnaire data. These traditional methods suffer from high implementation costs, limited sample sizes, cognitive biases, and a lack of geospatial coordinates, making it difficult to accurately quantify tourists' real-time emotional fluctuations and attention distribution across specific physical spaces. There is an urgent need for an objective, data-driven framework that integrates multi-source data to capture authentic tourist perceptions and translate them into actionable spatial optimization strategies.

Methodology
This study constructs a quantitative analysis framework based on User-Generated Content (UGC) multimodal data, utilizing Heshun Ancient Town in Yunnan, China, as a case study. The methodology integrates text mining, visual analysis, and spatial mapping through the following steps:

Data Acquisition: Multimodal UGC data (text and images) were collected from Weibo and Xiaohongshu (January 2021–December 2025), yielding 12,806 valid text-image samples and 17,684 filtered images. Point of Interest (POI) data were extracted from Amap, resulting in 1,496 spatial nodes covering historical features, natural landscapes, and tourism services.
Lexicon Construction: High-frequency words from UGC text were extracted and semantically merged to create a "location information word set" comprising 30 core elements, establishing a cognitive baseline.
Sentiment Quantification: The SnowNLP algorithm was employed to analyze text sentiment. Texts containing the location lexicon were processed to calculate sentiment scores ( $S_i$ ) mapped to a [-1, 1] interval. Mean sentiment scores ( $E_k$ ) were calculated for specific spatial nodes to classify emotional tendencies as positive, neutral, or negative.
Visual Preference Analysis: A cross-modal pre-trained large model, CLIP (Contrastive Language-Image Pre-training), was utilized for scene classification and label analysis.
- Scene Classification: Images were categorized into nine mutually exclusive scenes (e.g., Architecture, Streets, Water Systems) using prompt engineering and cosine similarity calculations.
- Feature Extraction: For confirmed scenes, the model performed multi-label feature reasoning to identify primary and secondary visual elements (e.g., "white walls and black tiles," "wooden components"), transforming unstructured images into structured feature matrices.
Spatial Mapping and Visualization: UGC images were mapped to corresponding POI nodes via similarity matching. Kernel Density Estimation (KDE) within the ArcGIS platform was applied to the POI frequency data to generate heatmaps, visualizing the spatial distribution patterns and density of tourist perceptions.
Validation: A reliability validation was conducted on 200 randomly selected image samples, where the CLIP model achieved 87% accuracy in scene classification, confirming its applicability without domain-specific fine-tuning.

Key Results

Perception Themes: Tourist perception is dominated by "Historic Character" (48.51% of visual data), followed by "Natural Landscape" and "Historical and Folk Culture." There is a structural divergence in "Tourism Services and Consumption": while these appear frequently in text (high attention), they are underrepresented in visual data (low recording), indicating a decoupling between experiential perception and visual documentation.
Sentiment Structure: A quadrant analysis revealed that "Historic Character" and "Natural Landscape" elements fall into the "Advantage Zone" (high frequency, high satisfaction). Conversely, commercial services and basic infrastructure often fall into the "Improvement Zone" (high frequency, low satisfaction) or "Marginal Zone" (low frequency, low satisfaction).
Spatial Distribution: The spatial perception follows a "single core, multiple nodes, and corridor connections" logic. High-density perception is concentrated in the north-central core (along the Sanhe River and Qiaoguang Square). Significant "perceptual blind zones" exist in peripheral areas and along secondary routes, where intangible cultural nodes are marginalized due to rigid circulation structures.
Visual vs. Emotional Divergence: Authentic architectural clusters and natural settings exhibit high visual documentation and positive emotional mapping. In contrast, symbolic commercial nodes show high textual attention but low visual recording rates and lower sentiment scores, suggesting a failure to integrate effectively into the historical fabric.

Significance and Contributions
This paper proposes a UGC-based analytical framework that addresses the limitations of traditional evaluation methods by objectively quantifying tourists' multidimensional perceptions. Its primary contributions include:

Methodological Innovation: It establishes a technical framework for cross-validating textual sentiment mining with visual feature analysis, linking digital features directly to physical POI nodes.
Empirical Insight: The study empirically demonstrates that authentic spaces with original morphological features possess robust "morphological resilience," serving as the core foundation for place attachment. It identifies "perceptual traps" where homogenized commercial interfaces capture attention but fail to sustain satisfaction or spatial continuity.
Practical Application: By revealing "perceptual blind zones" and spatial mismatches, the framework provides a scientific basis for targeted micro-updates, route restructuring, and the prevention of excessive commercialization. It enables heritage managers to move from homogenized development to differentiated, precise spatial interventions that balance heritage conservation with spatial vitality.

Limitations
The authors acknowledge that the study relies on data from out-of-town tourists active on social media, potentially underrepresenting local residents, the elderly, and children. Additionally, as a single-case study of Heshun Ancient Town, the findings may lack generalizability across different typologies of traditional villages (e.g., mountainous vs. plain, residential vs. commercial). Future research is suggested to include granular typological comparisons and broader demographic analysis.

1. The Two-Pronged Detective Work (Text + Images)

2. Mapping the "Heat" of the Village

3. What Did They Find?

4. The Big Picture

More like this