Content area
The soaring popularity of Ice-and-Snow Tourism (IST) and the growing number of tourists pose challenges for destination managers, particularly in balancing attraction development with the maintenance of high-quality tourism experiences. Understanding tourist behavioral preferences is crucial for the effective utilization of ice-and-snow resources and targeted marketing strategies. However, differences in IST participants" behavioral preferences and influencing factors remain unclear. Taking northeast China as an example, this paper applies text clustering and image recognition techniques to analyze 43,063 pieces of user-generated content (UGC) related to IST. Notable disparities are identified in destination image perception as well as in the temporal and spatial distribution patterns of IST participants, including both tourists and local residents. These findings reveal potential links between the variations and general regional attributes in 41 cities. Overall, tourists tend to develop diverse travel patterns but show similar preferences during specific holiday periods. They tend to concentrate in popular destinations, with stronger perception of distinctive architectural and natural landmarks. In contrast, local residents, with little travel time, exhibit a more dispersed travel pattern, favoring short trips to lesser-known surrounding attractions. In addition, unexpected insights emerge from the coupled analysis of spatiotemporal characteristics related to the destination image. Among 16 regional attributes, most socio-economic factors are positively correlated with tourist distribution, whereas physical geographic factors show relatively weak correlations. Although tourists show a greater preference for well-developed tourism infrastructure, diverse experience offerings, and enhanced safety, they also have a greater appreciation for natural environments than local residents. These findings offer valuable insights for more human-centered IST planning, supporting the optimization of existing attractions to enhance their competitiveness and sustainability.
Abstract The soaring popularity of Ice-and-Snow Tourism (IST) and the growing number of tourists pose challenges for destination managers, particularly in balancing attraction development with the maintenance of high-quality tourism experiences. Understanding tourist behavioral preferences is crucial for the effective utilization of ice-and-snow resources and targeted marketing strategies. However, differences in IST participants" behavioral preferences and influencing factors remain unclear. Taking northeast China as an example, this paper applies text clustering and image recognition techniques to analyze 43,063 pieces of user-generated content (UGC) related to IST. Notable disparities are identified in destination image perception as well as in the temporal and spatial distribution patterns of IST participants, including both tourists and local residents. These findings reveal potential links between the variations and general regional attributes in 41 cities. Overall, tourists tend to develop diverse travel patterns but show similar preferences during specific holiday periods. They tend to concentrate in popular destinations, with stronger perception of distinctive architectural and natural landmarks. In contrast, local residents, with little travel time, exhibit a more dispersed travel pattern, favoring short trips to lesser-known surrounding attractions. In addition, unexpected insights emerge from the coupled analysis of spatiotemporal characteristics related to the destination image. Among 16 regional attributes, most socio-economic factors are positively correlated with tourist distribution, whereas physical geographic factors show relatively weak correlations. Although tourists show a greater preference for well-developed tourism infrastructure, diverse experience offerings, and enhanced safety, they also have a greater appreciation for natural environments than local residents. These findings offer valuable insights for more human-centered IST planning, supporting the optimization of existing attractions to enhance their competitiveness and sustainability.
Keywords ice-and-snow tourism; behavioral preference; regional attribute; user-generated content
1. Introduction
Ice-and-snow tourism (IST), a form of ecological tourism that highlights the natural and cultural riches found in icy and snowy regions, encompasses activities such as sightseeing (of snowfields, glaciers, and ice/snow sculptures), sports (skating, skiing, and ice-fishing), and cultural experiences (winter festivals and local winter cuisine). Owing to its broad reach and economic potential, IST is emerging as a pivotal catalyst in regional efforts to foster high-quality, sustainable tourism. Despite its popularity, IST products in China remain largely unstandardized, leading to unchecked expansion and product homogeneity in many areas (Cai et al., 2023). Consequently, it is essential to examine tourists' behavioral preferences and take appropriate measures to improve tourism management. An analysis of their preferences enables a deeper understanding of tourists' decision-making processes, behavior patterns, and overall satisfaction (Djeri et al., 2014). This, in turn, supports the development of more targeted marketing and management strategies, Which are crucial for the rational planning, design, and construction of attractions, as well as future development plans for the preservation, utilization, and management of IST destinations.
Survey methods have been the predominant approach in studying tourist behavioral preferences to gather behavioral data from tourists in specific circumstances (Zhang and Hu, 2024). However, as these traditional methods like questionnaires are the most effective in contexts with relatively few tourists, limited activity types, and small spatial scales (Wei et al., 2024), they fail to conveniently and effectively uncover the behavioral mechanisms of different tourist categories across regions. This results in substantial gaps in understanding IST participants" behavioral characteristics.
Currently, a widely adopted approach 1s the analysis of original content actively shared on online platforms. This user-generated content (UGC) has attracted significant research interest for its accessibility, volume, and authenticity. Unlike traditional survey methods, UGC analysis offers a more cost-effective and rapid means for capturing diverse tourist experiences. Based on these data, destination management organizations (DMOs) can gain valuable insights into tourist preferences and improve the image of destinations. Existing UGC studies have focused on emotional experiences, spatial characteristics, satisfaction, recreation patterns, and behavioral trajectories. For example, a study on Peru found statistical differences between data from a DMO's website and Flickr across several dimensions (Stepchenkova and Zhan, 2013). To complement the geographic cues in Instagram photos, a landmark recognition method was proposed and shown to be feasible for extracting visitor footprints from UGC data (Deng et al., 2022). Collectively, these studies indicate that big data derived from social media can effectively reveal the spatiotemporal patterns of tourist behaviors and confirm the potential of UGC analysis as an evaluation tool for tourism management.
Nonetheless, the diversity of tourism products and the complexity of individualized tourist preferences make it difficult to classify data under the same tourism category. Previous research has often focused on data analysis within specific geographic regions or relied on keyword-based distinctions and filtering (Li et al, 2023), leading to relatively incomplete and less representative findings. Applying such data-filtering methods to tourism types like IST - which combine natural and cultural sightseeing - has notable drawbacks: overly stringent rules may cause an exponential reduction in available data, while overly lenient rules raise concerns about data validity. In both cases, the approach hinders a comprehensive understanding of the factors driving tourist behaviors. Moreover, extensive research has highlighted significant disparities in behavioral preferences across demographic groups (Plaza-Mejia et al., 2023). For IST in particular, its distinctive regional features mean that participants' geographic origins play a crucial role in shaping travel motivations. For DMOs, recognizing the different activities of key IST participants - both tourists and local residents - is crucial to revealing a destinations social, cultural, and functional attributes, which in turn explains its overall attractiveness (Hu et al., 2023). These insights can help refine service strategies for different groups, including visitors from different places of origin, which is a key aspect of local tourism management and resource development. However, despite the current iceand-snow trend, mainstream research has not sufficiently examined the differences between these two groups.
To address these research gaps and explore the diversified utilization of UGC data, this study employs text clustering and image recognition techniques to filter, verify, and analyze IST-related content. Simultaneously, geotagged data analysis enables a more quantitative interpretation of the relationship between regional attributes and behavioral preferences. Focusing on the behavioral preferences of IST participants, the study examines 43,063 geotagged photos shared by tourists and local residents in northeast China. Our study aims to answer three research questions that shed light on innovative and sustainable development of IST: @ How can IST-related UGC data be effectively identified? @ What behavioral preferences and differences exist among IST participants, particularly between tourists and local residents? 6) Do regional attributes contribute to these differences, and how are they interconnected?
2. Study area and data
2.1 Study area
Northeast China (aka the northeast region) comprises 41 cities 1n Liaoning, Jilin, and Heilongjiang provinces, along with five cities in eastern Inner Mongolia. Encompassing a land area of 1.45 million km? and a population of 98.51 million, this region is distinguished by high-latitude and cold-temperate to temperate continental monsoon climate. With long and cold winters, it possesses unique advantages in IST resources. In recent years, the region has embraced the concept that "lucid waters and lush mountains are invaluable assets and so are ice and snow" (Ministry of Culture and Tourism, National Development and Reform Commission, and General Administration of Sport of China, 2021). Guided by this vision, 1t has persistently advanced the strategies of building a "strong cultural & tourism region" and promoting "all-for-one tourism." These concerted efforts have ushered in a new era of IST development.
Of the 93 national IST attractions designated by the Chinese Ministry of Culture and Tourism, 30 are located in the northeast region. According to the China IST Consumption Report 2022, the Beijing Winter Olympics greatly boosted public enthusiasm for IST, positioning the northeast region as the leading destination for advancing the national vision of "300 million people engaging in ice and snow activities." Consequently, this region serves a representative case for IST research, given its distinctive advantages in climate, tourism resources, and policy support, among other factors. These strengths make it a valuable reference for the development of the IST industry in other parts of the world.
2.2 Data
The primary data were obtained from Mafengwo Travel (MT). a leading travel and entertainment platform often regarded as the Chinese equivalent of TripAdvisor. MT provides information and UGC on tens of thousands of global travel destinations. According to a report by the Social E-commerce Research Center (SERC),? the platform has over 50 million users, 80% of whom access it via mobile devices. It records approximately two million daily activities and 100,000 UGC posts per day, covering 95% of the world's attractions. Building on prior studies suggesting that photos shared on general social media platforms can serve as proxies and representative data sources in tourism research (Huang et al., 2022), this study collected data from 41 cities in the northeast region of China. Data entries containing the term "ice" or "snow" in either the title or body were extracted. Using Fiddler Classic software (version 5.0) and the Python programming language (version 3.9), a total of 118,777 geotagged photos were retrieved from 14.434 MT notes, covering the period from Dec. 27,2012, to Mar. 18, 2023.
To identify potential factors influencing tourists' behavioral preferences, this study collected data on population density, gross domestic product (GDP), road network, points of interest (POIs), digital elevation model (DEM), temperature, snowfall, and snow cover. The 2020 population data were obtained directly from the LandScan™ database of the Oak Ridge National Laboratory (ORNL) (via East View Cartographic). GDP data for 2020 were sourced from the Resource and Environment Science and Data Center. Road network data were derived from OpenStreetMap, an editable global map platform. POI data, including parking, shopping, food, and entertainment services, were obtained via the application programming interface (API) of the Gaode Open Platform," yielding a total of 626,891 entries. The 2020 DEM dataset was sourced from the General Bathymetric Chart of the Oceans," an organization dedicated to producing bathymetric datasets and products. Meteorological data, including temperature, snowfall, and snow cover, were drawn from the ERA5-Land dataset, published by the European Union and the European Centre for Medium-Range Weather Forecasts, among other organizations. All data share the same temporal scope as the UGC data, spanning the ice-and-snow seasons from 2012 to 2023. Administrative boundary and base maps were obtained from the China Standard Map Service (CSMS) and ArcGIS online," respectively.
3. Methodology
The analytical framework comprises three parts: filtering ISTrelated UGC data, classifying scenes and conducting spatiotemporal analysis, and exploring the relationships between behavioral prefer ences and regional attributes. Its goal is to uncover differences in participants' behavioral preferences during IST and their underlying factors. Section 3.1 explains how to effectively identify IST-related UGC by proposing a filtering method combining text clustering and image recognition. Section 3.2 analyzes behavioral preferences and differences among IST participants, exemplified by tourists and local residents by classifying visitor footprint points and conducting descriptive and spatiotemporal analysis. Section 3.3 examines whether and how regional attributes contribute to these differences by performing a correlation analysis between the distribution of visitor footprints and regional attributes.
3.1 Filtering IST-related UGC
To ensure data accuracy and validity, the dataset underwent a twophase cleaning process. In the first phase, implemented in Python, incomplete records lacking essential information (e.g., geographical location tags), duplicate entries that could compromise analytical reliability, and non-standard data, such as malicious reviews, irrelevant texts, repeated recommendations, and content unrelated to attractions, were identified and removed. Data originating from official public websites that provided no insights into tourist behavior were also excluded. The second phase involved spatiotemporal screening. Temporally, the ice-and-snow season was defined as October to March, based on monthly average temperature data from the China Meteorological Administration and the China Tourism Industry Development Report. Spatially, the Gaode Open Platform API was used to verify geographic locations and exclude data points outside the northeast region.
To address the diversity of tourism forms and the complexity of individual tourist behaviors, the study employed a dual-filtering process combining text clustering and image recognition. Text clustering determined whether a note was tourism-related, while image recognition identified the presence of snow or ice scenes in the accompanying images. Specifically, a Latent Dirichlet Allocation (LDA)-based method was adopted to cluster and categorize note content. This approach uncovers latent topics within the text and provides more descriptive labels than traditional clustering algorithms (e.g., K-means, DBSCAN), making it well-suited for processing the unstructured text data used in this study (Chen et al., 2024). Based on the clustering results, non-tourism-related content was excluded.
For image recognition, the Aliyun Big Data Platform (ABDP)· was employed as both a data-filtering tool and the foundation for subsequent scene classification. The platform processed the retained UGC images depicting ice-and-snow scenery, portraits, travel records, and other subjects with clear focus, through localization, segmentation, feature recognition, and classification. Each image was annotated with five descriptive tags and corresponding high-level labels. A note was considered valid only if it contained at least one photo verified to depict ice or snow. Finally, the integration of the filtered datasets established the foundational database for IST analysis.
3.2 Scenes classification and spatiotemporal analysis
The IST data were first categorized into 53 scenarios based on photo labels obtained through image recognition. To generate more meaningful insights, these scenarios were further aggregated into 10 categories (see Table 1). This hierarchical scene classification enabled a comparative analysis between tourists and local residents. Temporal variations were examined at annual, monthly, and workdayweekend scales. Spatial analyses were conducted in ArcGIS (version 10.8) using kernel density analysis and tracking analyst tools. The kernel density analysis generated heat maps to visualize density distribution, revealing spatial hotspots, trends, and spatial patterns of tourists. Meanwhile, the tracking analyst tool examined tourist trajectories across time and space, providing insights into movement patterns, behavioral characteristics, and spatial correlations.
3.3 Exploring the relationship between behavioral preferences and regional attributes
Based on relevant literature (see Table 2), potential factors influencing tourists' behavioral preferences were identified, encompassing both socio-economic and physical geographic dimensions. The selected socio-economic factors include population density, GDP, road density, and POI density. Previous studies have demonstrated a linear correlation between population density and tourist distribution (Tenie and Fintineru, 2020; Huertas and Miguel, 2022), while GDP, road density, and POIs density have also been shown to influence tourist flows (Schein, 2016). GDP and road density serve as indicators of regional influence and accessibility, with areas exhibiting higher influence and accessibility often attracting more tourists (Mizzi et al., 2018). POIs were categorized into four types based on the services that attract tourists, including parking, shopping, food, and entertainment, each contributing to the understanding of tourists" behavioral preferences (Marrocu and Paci, 2013).
Regarding physical geographic factors, this study followed Steiger and other researchers in selecting variables relevant to IST infrastructure development. These include DEM data and meteorological data such as temperature, snowfall, and snow cover (Dawson and Scott, 2013; Steiger and Scott, 2020). Meteorological data were further divided into multi-year averages for January and multi-year averages for the entire ice-and-snow season (Steiger et al., 2019). All variables were examined to capture their spatial patterns and regional variations across northeast China.
The band collection statistics method was employed to quantify the correlation between tourists' behavioral preferences and regional attributes. Before conducting the correlation analysis, all variable distribution maps were resampled to ensure consistent pixel size and spatial alignment. Among the 16 variables and the tourist distribution maps, the ERAS Land dataset exhibits relatively coarse pixel resolution of 9 km. Therefore, all raster maps were resampled based on meteorological data. Statistical analyses in ArcGIS confirmed that all element categories were continuously distributed in geographic space without outliers. Given that the relationships between these regional attributes and tourist distribution have been discussed in previous studies, the premise for correlation analysis was satisfied. The band collection statistics method calculated the linear relationship between two variables, standardizing for variable scales and producing correlation coefficients ranging from -1 to 1.
4. Results
4.1 Descriptive analysis of behavioral preferences
People from diverse geographical, social, and cultural backgrounds exhibit distinct travel preferences, leading to variations in tourist behavior across destinations. To capture this diversity, IST participants were classified into two groups: tourists and local residents, based on their place of residence (some inferred from their most frequently used IP address) and travel destination. From a cultural geography perspective, northeast China occupies a relatively unique geographical position. The construction of the Chinese Eastern Railway endowed the region with distinctive historical and cultural characteristics, profoundly influencing local residents and fostering a shared sense of identity and cultural belonging. Particularly notable is the emergence of the "Northeast Culture Circle" in online public discourse, which has reinforced the perception of the region as a cohesive cultural entity. Accordingly, individuals residing in any of the 41 cities in the northeast region are broadly defined as local residents. The dataset comprises 43,063 IST-related photos extracted from 5,085 travel notes, including 21,567 photos from tourists and 21.496 from local residents. On average, tourists posted nearly twice as many photos per note as local residents (see Table 3), suggesting a greater propensity to share their IST experiences.
The distribution of scene categories reveals differences between tourists and local residents, with each category's percentage representing its share of the total dataset (see Figure 1). Overall, the top three preferred categories for both groups are outdoor scenes (24.22%), natural environments (20.43%), and activities (17.18%). Together, these account for over 60% of all tags, indicating that IST participants are particularly drawn to ice-and-snow scenery and outdoor activities. In contrast, character photos represent only a small fraction of the total photos taken by tourists (3.44%) and local residents (2.48%), indicating that such images were less prevalent than expected. Previous research has shown that photography-friendly landscapes or the availability of professional photography services can encourage tourists to take selfies, thereby stimulating tourismrelated consumption among online audiences (Sharma et al., 2022). This is an important consideration for local DMOs.
4.2 Spatiotemporal distribution of behavioral preferences
The annual and monthly variations in IST distribution are illustrated in Figure 2. From 2012 to 2023, an overall upward trend is observed in the number of IST participants and photo uploads in the northeast region. A slight decline occurs during the 2020 - 2021 ice-and-snow season due to the impact of COVID-19, followed by a rapid recovery (see Figure 2a). The blue dotted line represents a fitted prediction based on photo quantity trends, with an В? value of 0.967 for the prediction line, indicating a highly reliable exponential growth projection. December and January record the highest number of IST photos, accounting for 45% of the total during the entire ice-and-snow season, while March shows the lowest, comprising only 9% (see Figure 2b). Tourists exhibit a peak in photo-taking activity in October, while their patterns in other months align with those of local residents, displaying an increase from November to January, followed by a decrease from January to March.
The heatmap reveals distinct differences in the spatial distribution of tourists and local residents (sec Figure 3). Tourists" footprints are more concentrated, primarily around popular IST attractions such as Harbin Ice and Snow World, Snow Country Scenic Area, Yabuli Ski Resort, Changbai Mountain International Resort, Beidahu Ski Resort, and Beiji (Arctic) Village. The three cities with the densest tourist footprints - Harbin, Mudanjiang, and Changchun - also rank among the top ten most favored IST destinations according to the China Tourism Academy. In contrast, local residents are distributed across a broader geographical area. In Yichun, attractions like the Meihua River Ski Resort and Fenglin National Nature Reserve are particular popular with local visitors, while residents in cities like Jinzhou, Dalian, and Qiqihar tend to visit specific but relatively niche hotspots.
A tracking analysis tool was employed to examine the spatiotem poral trajectories of IST participants by linking note IDs with the timestamps of associated photos (see Figure 4). The lines represent participants' movement trajectories, with each line indicating a vec tor from the point of departure to the destination, as illustrated in the detailed diagram in the upper right corner. Tourists' trajectories are primarily concentrated within popular IST attractions, including Beiji Village, Changbai Mountain International Resort, and Harbin Ice and Snow World. Some trajectories exhibit repetitive patterns, such as northbound routes from Shenyang to Harbin, potentially with stops at Suihua and Qiqihar in the Greater Khingan Mountains region. Similarly, eastbound routes from Shenyang to Harbin may involve visits to Mudanjiang and Jixi. In contrast, local residents ex hibit sparser and predominantly short-distance travel paths.
4.3 Relationship between behavioral preferences and re gional attributes
The correlation analysis using band collection statistics yields the following results (see Table 4). All 16 regional attributes in northeast China show statistically significant correlations with the distribution of IST participants. Among the variables examined, socio-economic factors and certain physical geographic factors, such as temperature and multi-year average snowfall during the ice-and-snow season, show positive correlations, whereas the DEM displays a negative correlation. Interestingly, multi-year average snowfall in January is correlated with the distribution of tourists but not with that of local residents, whereas the opposite pattern is observed for multi-year average snowfall across the entire ice-and-snow season.
Overall, socio-economic factors exhibit stronger correlations with the distribution of IST participants than physical geographic factors. Additionally, regional attributes display higher correlations with the distribution of local residents compared to tourists. For tourists, among socio-economic factors, the density of parking service POIs shows the highest correlation (R = 0.586), while GDP exhibits the lowest (R = 0.121). For local residents, the densities of food and en tertainment POIs are particularly influential, with correlation coef ficients of 0.814 and 0.811, respectively. Among physical geographic factors, DEM, temperature, and multi-year average snowfall dur ing the ice-and-snow season show relatively stronger correlations, whereas the remaining factors display comparatively weaker corre lations.
5. Discussion
5.1 Differences in the behavioral preferences of IST par ticipants
5.1.1 Differences in destination image perception between tourists and local residents
Leveraging image recognition techniques, the visual content in UGC data provides a more comprehensive understanding of the destination image perceived by IST participants - insights that were previously difficult to obtain through questionnaire-based research alone. This study finds that tourists post a higher proportion of images related to built environments, personal photos, transporta tion, and animals, whereas local residents exhibit a greater focus on indoor scenes and food. Regarding specific scene labels, tourists tend to highlight iconic landmarks, distinctive street views, and cultural activities. These elements, combined with the region's winter resources, contribute to the unique appeal of northeast China. However, several unexpected findings contradict initial assumptions. For instance, the hypothesis that tourists would show greater interest in food than local residents is not supported. Examination of posted user notes and profiles reveals that some local residents are active content creators who frequently share recommendations for nearby travel routes and popular attractions, with food being one of their focal points.
5.1.2 Differences in temporal distribution between tourists and local residents
Tourists exhibit pronounced monthly variations in photo numbers, consistently surpassing those of local residents. Analysis based on statistically classified image recognition results reveals that from December to February, tourists captured substantially more photos featuring natural elements, such as ice, snow, and lakes, as well as ice-and-snow activities like skiing and skating, compared to other months. Given the climate characteristics of the northeast region, this surge can be attributed to the availability of high-quality iceand-snow landscapes and related recreational activities facilitated by the extremely cold weather, which attract many tourists. The sharp decline in tourist number in March may result from the diminishing visual appeal of ice-and-snow landscape as temperatures rise from mid-to-late March. However, the concentrated distribution of tourists in October cannot be attributed to climatic alone. To investigate further, word separation and frequency analysis were conducted on user notes and shared content. Frequent mentions of terms such as "holiday." "golden week," and "early bird tickets," along with their associated tags, suggest that tourists are influenced by the clustering of holidays and early promotional campaigns for the upcoming iceand-snow season in October.
Differences between tourists and local residents are also evident in terms of workdays and rest days (see Table 5). The data indicates that local residents tend to share more photos on their rest days (7.65 vs. 6.17), but their destination image perception is lower than on workdays (2.35 vs. 2.74). In contrast, tourists show no significant differences between workdays and rest days. From the perspective of Expectancy Confirmation Theory (ECT), these differences might be shaped by socio-cultural and psychological factors (Tata et al., 2021; Singh et al., 2023). Although local residents have more leisure time on weekends, their destination image perception may be more focused and objective, lacking the heightened positivity or idealization associated with previous expectations on workdays. Tourists, on the other hand, appear to maintain a more consistent mindset and attitude, adopting an exploratory and enjoyment-oriented approach regardless of the day, resulting in minimal perceptual variation. Furthermore, within the category of natural environments, local residents seem more attuned on workdays to scenes such as the night sky and sunsets, which UGC text data indicates are valued for their role in promoting relaxation.
5.1.3 Differences in spatial distribution between tourists and local residents
Tourists and local residents in the northeast region exhibit distinct spatial distribution patterns, both overall and across different contexts. Tourists tend to concentrate in higher-ranked attractions, leading to significant variation in distribution density. In contrast, local residents do not necessarily gravitate toward the most famous attractions, often favoring lower-ranked IST attractions. For example, although not a top choice among tourists, Nianzishan International Ski Resort is one of the most popular IST destinations for residents of Qiqihar. These divergent preferences highlight the importance of tailoring services and promotional strategies to different visitor groups. Therefore, DMOs should consider these differences when managing tourist flows and incorporate tourists' intuitive preferences into attraction planning.
To enhance visualization clarity across the extensive geographical scope, trajectories shorter than 10 kilometers were excluded from the analysis (see Figure 4). Notably, 87% of local residents' trajectory data were removed, indicating that local residents are less inclined towards structured or checklist-based tourism and more likely to engage in short, spontaneous trips without predefined destinations. Analysis of trip origins and destinations reveals that local residents primarily travel within downtown areas, with some visits to IST scenic spots. In contrast, tourists exhibit more diverse patterns: some regions show high densities of either departure or arrival points, while others display a more balanced distribution. For example, attractions like Beiji Village and Changbai Mountain, located near provincial or national borders, often serve as ultimate destinations for tourists, whereas central urban areas of major cities commonly function as departure points. This pattern indicates a preference for an "urban-suburban tourism" strategy, in which cultural, historical, and architectural elements within cities are explored first, followed by excursions to nearby suburban areas for natural scenery, rural culture, and outdoor activities. Statistical analysis of publishers' places of residence further reveals that tourists from three provincial-level administrative areas - Guangdong, Beijing, and Shanghai - collectively contributed nearly half of all long-distance travel trajectories (19.17%, 16.79%, and 13.32%, respectively). Tourists from Guangdong recorded the highest average trajectory distance (approximately 28.98 kilometers), with two out of every five engaging in IST trips exceeding 500 kilometers. These findings suggest that tourists from economically developed regions invest more time and resources in IST and that those with less frequent exposure to ice-and-snow landscape are more receptive to long-distance travel experiences.
Kernel density analysis was applied to examine differences in the spatial distribution of IST participants across various scenes. The analysis revealed that the spatial distribution of tourists and local residents differs markedly across destination images. Consistent with the overall footprint distribution, tourists are concentrated in distinct hotspots for most categories, whereas local residents are more dispersed. Although tourists and local residents share some similarities in the distribution of preferred destination images, notable spatial differences are evident. Certain cities, such as Yichun and Qigihar, which are not regional hotspots, receive relatively few visits. Nonetheless, local residents still exhibit rich perception of the aforementioned IST elements. Regarding specific perception preferences, tourists show highly concentrated transportation-related behavior near the Snow Country Scenic Area, whereas local residents generally show less concern for transportation during the ice-and-snow season due to their preference for shorter trips. Similarly, tourists demonstrate greater interest in Yanbian's Korean cuisine, whereas local residents' culinary activities tend to concentrate around several major cities. These patterns reflect participants' feedback on destination images, encompassing both praise and criticism. DMOs should consider these distinctive spatial distribution features to better promote existing hotspots and address areas requiring improvement.
5.2 Correlation between behavioral preferences and regional attributes
This study examines the relationship between behavioral preferences and regional attributes, finding that both regional socio-economic and physical geographic factors directly or indirectly influence participants' decision-making processes. The analysis reveals a strong correlation with socio-economic factors, suggesting that areas with better tourism infrastructure, richer tourism experiences, and higher levels of safety and comfort are generally more attractive. In contrast, the correlation with physical geographic factors is weaker. Nonetheless, statistically significant p-values indicate that these factors remain relevant to tourist behaviors. Their influence may be offset or moderated by other variables, thereby resulting in smaller correlation coefficients.
From a socio-economic perspective, the correlation analysis yields the following results. Population density, reflecting the concentration of permanent residents within a geographic area, accounts for the stronger correlation observed among local residents relative to tourists. GDP, which is closely related to population density, exhibits a similar pattern. These findings corroborate the conclusions of previous studies and provide additional empirical support for the spatial distribution patterns of IST participants discussed earlier (Peng et al., 2023; Wei et al., 2023). Road density likewise shapes tourist behaviors: tourists tend to cluster in areas with greater accessibility, whereas geographic accessibility exerts a comparatively weaker impact on local residents. Subtle variations are also evident across different types of POIs: tourists are more likely to be drawn to areas with abundant parking facilities, while local residents prefer locations offering more comprehensive services, such as food and retail. Collectively, these findings are consistent with survey-based research and highlight the urgency of IST infrastructure development, especially in light of regional disparities in tourism development (Zhao et al., 2024).
The heterogeneity of tourist preferences is examined due to significant fluctuations between physical geographic factors. The analysis of DEM and temperature suggests that tourists tend to adapt more readily to high-altitude and low-temperature geographical environments. A possible explanation is that they often have higher travel motivations and expectations, making them more willing to embrace challenges and seek diverse experiences. Additionally, such extreme environments often offer more attractive scenery that fulfills the psychological mechanism of "benign envy," recognized as one of travel motivations (Sharma et al., 2022). The two indicators, namely multi-year average snowfall in January and the whole ice-and-snow season, show contrasting correlation patterns. The distribution of local residents aligns more closely with overall snowfall during the ice-and-snow season, while the distribution of tourists is more strongly associated with January snowfall. In other words, local residents can take advantage of the extended ice-and-snow season to visit their desired destinations and experience diverse landscapes, while tourists are more inclined to plan their trips during the coldest month (January). The relatively low correlation coefficient for snow cover is somewhat unexpected, but it can be inferred from meteorological data that higher snow cover tends to be correlated with lower levels of urbanization (Wang, 2021). Although tourists are present in suburban areas, their numbers remain limited compared to IST sites and urban destinations. This suggests that homogeneous iceand-snow landscapes in rural areas lack sufficient attractiveness, underscoring the necessity of developing high-quality, integrated rural IST products.
5.3 Implementation paths for IST sustainability
In northeast China, challenges such as the disorderly development of IST products, chaotic attractions planning, and homogenized travel routes have become increasingly prominent, conflicting with growing aspirations for a higher quality of life. Analyzing tourist behavioral preferences through UGC data is crucial for enhancing tourist engagement, optimizing scenic area planning, and fostering sustainable tourism. The consistent pursuit of safety, happiness, and a sense of fulfillment highlights the importance of aligning people-centered tourism experiences with public needs. To promote the sustainable development of IST, the following recommendations are proposed:
(1) Different participants demonstrate varying destination image perceptions. The distinctive ice-and-snow landscape, coupled with the unique cultural atmosphere of northeast China, offer visitors with immersive experiential journeys. Promotion and planning centered on these well-regarded destination images can significantly enhance the visibility of attractions. Surveys completed by IST participants - covering demographic characteristics (e.g., age), adaptation to extreme cold, participation in iceand-snow activities, and related expectations - provide valuable insights for refining destination image strategies and informing local government decision-making. Equally important are the inclusive measures for vulnerable groups, such as children, older adults, and individuals with disabilities, by designing accessible opportunities for engagement with ice-and-snow landscapes and cultural offerings.
(2) Different participants exhibit variations in their temporal distribution. Tourists are more likely to travel during peak holiday periods, whereas local residents have more opportunities to explore even on workdays. Although the classification of visitors extends beyond the dichotomy of tourists and local residents, recognizing these tendencies provides a useful framework for segmenting and planning IST offerings. Diverse travel-time options can therefore be advantageous in meeting varying customer demands. Furthermore, this study reaffirms prior findings that destinations embodying strong local cultural attributes exert stronger appeal for tourists (Kucukergin et al., 2021). Accordingly, as an ecotourism model, the ice-and-snow tourism industry ought to prioritize the development and utilization of resources with distinct local characteristics. This involves the integration of local culture, traditions, cuisine, and folklore, which is essential for mitigating the impact of off-season downturns in ice-and-snow tourism.
(3) Different participants exhibit distinct spatial distribution patterns. Tourists tend to leave concentrated footprints, whereas residents' movements are more dispersed, extending into areas less frequented by tourists. For long-distance travel products, routes like the Harbin-Heihe-Yichun-Mohe can combine highly popular destinations with unique, lesser-visited ice-and-snow sites. Trajectory analysis also offers valuable insights for short-term travel planning, such as the continued application of the "urban-suburban tourism" strategies in cities like Harbin and Changchun. Moreover, both tourists and local residents show preferences for specific scenes, ranging from leisure sightseeing to participatory ice-andsnow sports. Given the scattered demand and dispersed distribution of IST sites, Heilongjiang Province holds potential for additional sightseeing-oriented attractions. Meanwhile, optimizing the layout of sports facilities near Changbai Mountain can further attract ice-and-snow sports enthusiasts.
(4) There is a clear correlation between regional attributes and the distribution of IST participants. Analyzing the factors driving tourist behaviors is essential for optimizing the design and planning of tourist areas. Certain niche locations, which lose appeal due to limited accessibility, could benefit from improved transportation and leisure facilities, especially in ice-abundant rural areas of high-latitude regions like Heihe, to attract a larger influx of visitors. Densely populated areas require enhanced infrastructure, including parking, dining, and shopping facilities, to deliver diverse and customized services and address the current supplydemand gap. The finding that highly snow-covered areas may be less attractive than expected underscores the problem of product homogeneity in northeast China's IST sector and highlights the critical need for greater landscape diversity. For instance, expanding tourism services in rugged terrains would align with the preferences of adventure-seeking tourists while promoting a more effective utilization of natural resources.
5.4 Future perspectives
This study has several limitations that should be acknowledged. First, reliance on a single UGC data source (MT) constrains the representativeness of the findings, as not all travel platforms provide accessible geotagged photos. Second, the absence of complete user profiles in the collected online data may result in bias towards younger and more educated user groups, potentially overlooking a broader range of tourists. Third, due to data limitations, the study is unable to perform more in-depth statistical analyses, such as regression modeling, which would offer valuable insights for quantitatively assessing the factors influencing tourist behaviors.
To address these limitations and broaden the applicability of this methodology, future research should integrate multiple data sources and adopt diverse analytical approaches from related fields. Incorporating diverse UGC data and employing advanced machine learning techniques to analyze specific tourist behaviors would enable the identification of different social groups, a wider range of behavioral categories, and individual interest preferences. Such strategies would enhance both the granularity and the practical relevance of future research in this domain.
6. Conclusions
This study explores how and why IST participants differ in their behavioral preferences and influencing factors, addressing existing gaps in understanding their diversity. Using text clustering and image recognition methods, it identifies distinct behavioral patterns and examines their potential links to regional attributes. The results reveal marked disparities in participants' perception of destination images and their spatiotemporal distribution. Both socio-economic conditions and physical geographic features significantly influence tourist behaviors. These findings provide valuable insights for human-centered IST planning, offering guidance for rethinking tourism management, refining product design, and optimizing existing attractions to enhance competitiveness, visitor experience, and sustainability.
(This study was supported by the National Natural Science Foundation of China (No. 52408012); the Philosophical and Social Science Research Planning Project of Heilongjiang Province (No. 23GLA044); the Postdoctoral Funding Project of Heilongjiang Province (No. LBH-Z23189); the Natural Science Foundation of Heilongjiang Province, China (No. LH2024E050).)
References:
Cai W, Yu J, and Yang W (2023). The Spatial Synergy of the Ice-Snow Tourism Industry and Its Related Industries in Jilin Province. Sustainability; 15, pp. 1-15.
Chen Y, Zhang L, and Dong Q (2024). Using Natural Language Processing to Evaluate Local Conservation Text: A Study of 624 Documents from 303 Sites of the World Heritage Cities Programme. Journal of Cultural Heritage, 70, pp. 259-270.
Dawson J and Scott D (2013). Managing for Climate Change in the Alpine Ski Sector. Tourism Management, 35, pp. 244-254.
Deng №, Qu Y, Cheng X, and Qin J (2023). Seeing Is Visiting: Discerning Tourists" Behavior from Landmarks in Ordinary Photos. Current Issues in Tourism, 26 (15), pp. 2494-2512.
Djeri L, Armenski T, Tesanovic D, Bradic М, and Vukosav S (2014). Consumer Behaviour: Influence of Place of Residence on the Decision-Making Process When Choosing a Tourist Destination. Economic Research-Ekonomska Istrazivanja, 27, pp. 267-279.
Dong W, Liu Y, Dong Y, Wu W, and Chen Y (2024). How the Perceived Built Environment Affects Pedestrian Well-Being: Evidence from Three Types of Neighborhoods in Harbin, China. China City Planning Review, 33, pp. 65-76.
Hu Y, Liu Y, Chen P, and Zhang М (2023). The Impact of Residents" Perceptions of Urban Shrinkage on Overall Life Satisfaction: The Case of Yichun, China. Cities, 141. DOT: 10.1016/j.cities.2023.104445.
Huang Y, Li Z, and Huang Y (2022). User Perception of Public Parks: A Pilot Study Integrating Spatial Social Media Data with Park Management in the City of Chicago. Land, 11 (2). DOI: 10.3390/land11020211.
Huertas A and Miguel O M (2022). Do Tourists Seek the Same Information at Destinations? Analysis of Digital Tourist Information Searches According to Different Types of Tourists. Furopean Journal of Tourism Research, 32. DOI: 10.54055/ejtr.v321.2492.
Kucukergin К С, Caliskan С, Dedeoglu В В, and Birinci M С (2021). Analyzing the Role of Constraints and Motivations Behind Traveling in the Prediction of Destination Choice: Evidence from PLS-SEM and fsQCA. International Journal of Tourism Research, 23, pp. 1191-1209.
Li J, Meng L, Zhang Z, and Yang K (2023). Low-Frequency, High-Impact: Discovering Important Rare Events from UGC. Journal of Retailing and Consumer Services, 70. DOT: 10.1016/j.jretconser.2022.103153.
LiR, Lu Z, and Li J (2012). Quantitative Caculation of Tourist's Landscape Perception: Strength, and Spatial Variation Within Ecotourism Destination. Ecological Informatics, 10, pp. 73-80.
Marrocu E and Paci R (2013). Different Tourists to Different Destinations. Evidence from Spatial Interaction Models. Tourism Management, 39, pp. 7183.
Ministry of Culture and Tourism, National Development and Reform Commission, and General Administration of Sport of China (2021). Notice on Printing and Distributing the Ice-Snow Tourism Development Action Plan (2021 - 2023).
Mizzi С, Fabbri A, Rambaldi $, Bertini F, Curti N, Sinigardi S, Luzi R, Venturi G, Davide M, Muratore G, Vannelli A, and Bazzani A (2018). Unraveling Pedestrian Mobility on a Road Network Using ICTs Data During Great Tourist Events. EPJ Data Science, 7. DOT: 10.1140/epjds/s13688-018-0168-2.
Peng R, Huang J, and Deng X (2023). Spatiotemporal Evolution and Influencing Factors of the Allocation of Social Care Resources for the Older Adults in China. International Journal for Equity in Health, 22. DOI: 10.1186/s12939023-02007-0.
Plaza-Mejia M A, Haldon-Hermoso J M, and Porras-Bueno N (2023). Residents as Gratuitous Referrals at Destination: An Integrative Model from Altruistic Values to Pro-Tourism Behavior. Infernational Journal of Tourism Research, 25, pp. 359-371.
Sai Vijay T, Prashar S, and Parsad C (2021). Examining the Influence of Satisfaction and Regret on Online Shoppers' Post-Purchase Behaviour. Benchmarking-an International Journal, 28, pp. 1987-2007.
Schein A (2016). The Growth of Foreign Tourists to Israel and Their Impact on Israel's GDP. In: Vrontis D, Weber Y, and Tsoukatos E (eds.), Innovation, Entrepreneurship and Digital Ecosystems. Cyprus: Euromed Press, pp. 16411652.
Sharma R, Singh G, and Pratt S (2022). Exploring Travel Envy and Social Return in Domestic Travel: A Cross-Generational Analysis. Journal of Travel & Tourism Marketing, 39, pp. 58-72.
Singh M, Tandon U, and Mittal A (2023). Modeling Users' and Practitioners' Intention for Continued Usage of the Internet of Medical Devices (IoMD): An Empirical Investigation. Information Discovery and Delivery, 51, pp. 306321.
Steiger R (2012). Scenarios for Skiing Tourism in Austria: Integrating Demographics with an Analysis of Climate Change. Journal of Sustainable Tourism, 20, pp. 867-882.
Steiger R and Scott D (2020). Ski Tourism in a Warmer World: Increased Adaptation and Regional Economic Impacts in Austria. Tourism Management, 77, 104032. DOI: 10.1016/j.tourman.2019.104032.
Steiger R, Scott D, Abegg B, Pons M, and Aall C (2019). À Critical Review of Climate Change Risk for Ski Tourism. Current Issues in Tourism, 22, pp. 1343-1379.
Stepchenkova S and Zhan F (2013). Visual Destination Images of Peru: Comparative Content Analysis of Dmo and User-Generated Photography. Tourism Management, 36, pp. 590-601.
Tenie B and Fintineru G (2020). What Attracts Tourists in Rural Areas? An Analysis of the Key Attributes of Agritourist Destinations That May Influence Their Choice. Agrolife Scientific Journal, 9, pp. 324-333.
Wang S (2021). Perception of Indigenous People of Climate Change and Its Impact on the Everest National Nature Preserve. Meteorological Applications, 28. DOI: 10.1002/met.1987.
Wei H, Zhan T, Yi Z, Shuo W, and Yan L (2023). À Study on the Drivers of Carbon Emissions in China's Power Industry Based on an Improved PDA Method. Systems, 11 (10). DOI: 10.3390/systems11100495.
Zhang W and Hu Y (2024). Tourist Perception and Development Strategies on Culture and Commerce in a Historic Area: A Case Study of Dayan Ancient Town in Lijiang. China City Planning Review, 33, pp. 62-74.
Zhao J, Liu P, and Li Z (2024). Exploring the Impact of Trip Patterns on Spatially Aggregated Crashes Using Floating Vehicle Trajectory Data and Graph Convolutional Networks. Accident Analysis and Prevention, 194. DOI: 10.1016/j.aap.2023.107340.
Notes:
(1) Available at: <https://www.mafengwo.cn/>.
(2) Available at: <https://www. 100ec.cn/>.
(3) Available at: <https://landscan.ornl.gov/>.
(4) Available at: <https://www.resdc.cn/>.
(5) Available at: <https://Www.openstreetmap.org/>.
(6) Available at: <https://Ibs.amap.com/>.
(7) Available at: https://www. gebco.net/>.
(8) Available at: <https://cds.climate.copernicus.eu/>.
(9) Available at: <http://bzdt.ch.mnr.gov.en/>.
(10) Available at: <https://www.arcgis.com/>.
(11) Available at: <https://Www.aliyun.com/>.
Copyright Urban Planning Society of China 2025