Visual Quantification, Spatial Distribution and Combination Association of Tourist Attractions in Qingdao Based on Social Media Images

Abstract

Focusing on the deficiencies of traditional tourism attraction survey methods in terms of accuracy, efficiency, and large-scale visual representation, this study selects Qingdao as the research case, collects tourism image data from the Weibo platform, applies a deep learning model to identify the visual elements of tourism images, and employs kernel density analysis and Apriori association analysis to clarify further the distribution characteristics and associated features of tourist attractions. Its core objective is to systematically reveal the visual composition, spatial distribution patterns, and related features of tourist attractions in the case study area by identifying and extracting tourist attraction elements from images, thereby providing a decision-making basis for effectively identifying tourism demands and their spatial distribution characteristics, as well as for tourism spatial planning. The findings are as follows: Buildings, sea, and other elements are the main components of tourist attractions in Qingdao. Regarding spatial distribution, tourist attractions in Qingdao exhibit the spatial characteristic of “distributed around the bay and converging towards the sea”, with a certain circular structure and multi-level core distribution pattern. Regarding associated features, tourist attractions in Qingdao form combinations centered on buildings, sea, and signs—such as building-centric, sea-centric, cityscape-centric, and sign-centric combinations—around elements including buildings, sea, and signs. The contribution and significance of this study lie in providing technical support for resolving the contradiction between traditional tourist attraction survey methods and precise demands, offering a scientific basis for decision-making in tourism spatial layout planning, and opening up a new path for the intelligent and refined development of tourism resources using massive visual data.

1. Introduction

Tourist attractions are the basic elements for carrying out tourism activities [1] and the core material carrier for developing tourism. With the gradual maturity of the tourism market, as tourism enters the stage of popularization and national participation, social production factors continue to evolve, and the tourism market becomes increasingly segmented and diversified. For instance, new tourism forms such as “Village BA”, Zibo Barbecue, and City Walk have gained widespread popularity. Driven by the awakening of social individualization, composite tourism demands have promoted the reshaping of forms and reorganization of types of tourist attractions [2], which in turn affects tourists’ demands and consumption decisions, solidifies the classes, types, and levels of tourism demands, and makes it difficult to achieve the goal of “host-guest sharing” in tourism development. There is an urgent need to update the new tourism resource concept—characterized by the shift from the logic of resource supply to that of market demand—and its “boundaryless” development concept [3].

Tourist attractions adopt tourists as the primary perspective, emphasizing the objective attributes of “objects” and the essential characteristics of their attractiveness. In tourism activities, the relationship between tourists and attractions is close [4], which helps to shift research from focusing solely on the tourism industry to considering the dual subjects of the tourism industry and tourists [5,6]. In the early stage of research, scholars mainly explored the element composition and categories of tourist attractions through methods such as field surveys and social statistics [7,8,9]. With the increasing improvement of the tourism economic system and the evolution and innovation of tourism demands, network information collection and spatial data analysis have provided strong support for tourism research under the fourth paradigm of scientific. Scholars have conducted in-depth studies on the spatial distribution and functional characteristics of tourist attractions based on POI geospatial data and user-generated information such as travel notes and reviews published by tourists [10,11].

In recent years, amid technological transformations such as digital twins, intellectualization, and informatization, images have become an important carrier of information dissemination, providing massive visual information. Images’ spatial, textual, and temporal information offer abundant content and diverse perspectives for in-depth exploration of tourist attractions and tourists’ demands. As a materialized result constructed under tourism gaze, tourism photos are a direct visual presentation of the “host-guest relationship” between tourist attractions and tourists [12], embodying tourists’ subjective choices and objective reflections on destination attractions [13]. Compared with traditional data types, they provide a unique perspective for interpreting the visual representation of tourist attractions and reflecting tourists’ demand preferences [14,15]. Some scholars have made preliminary attempts to identify tourist attractions in images through qualitative research methods such as coding and manual recognition [16]. However, due to technical analysis barriers, traditional manual recognition methods still have limitations, such as low efficiency in identifying tourist attractions and excessive subjective interference. The contradiction between large-scale samples and refined demands remains to be resolved. Moreover, existing studies focus on the basic attribute characteristics of tourist attractions, such as classification and evaluation, and rarely pay attention to higher-level attribute characteristics, such as the relationship construction of tourist attractions and tourists’ preferences. Empirical research on exploring tourist attractions through tourism images is still in the initial exploration stage. In the new era, with the digitalization of scenes and the renewal of digital technologies, there is an urgent need to strengthen the new cognition and reconstruction of the tourist attraction system through new technologies. Thus, against the backdrop of the rapid development of computer vision, how can we break the barriers of image language and technology to identify and quantify the visual representation of tourist attractions? On this basis, what co-occurrence laws and association rules exist among different tourist attractions? These questions have become urgent issues that need to be addressed in current tourist attraction research. Solving these problems will help effectively identify tourism market demands, bridge the gap between the given resources on the supply side and the experience demands on the demand side, and provide a decision-making basis for tourism spatial planning.

Qingdao has various tourist attractions, integrating natural landscapes such as mountains, seas, and beaches, as well as humanistic customs such as European-style buildings and characteristic streets. Urban landmarks, leisure formats, and tourism scenarios are constantly enriched, attracting tourists from home and abroad to visit and check in. In the “World Tourism and Leisure City Development Report” released by the China Tourism Academy in 2023, Qingdao was listed as one of the most beautiful domestic coastal leisure cities in the eyes of tourists. However, long-term reliance on established tourism resources, prominent homogeneous competition of products, insufficient resource integration and development, and inadequate market expansion have restricted its development space. How to break through the dilemma of homogeneous markets and explore the positioning of urban tourism development in line with market demands has become an urgent problem to be solved. Based on this, this study selects Qingdao as the case site, based on web data crawling, image semantic segmentation, and Apriori algorithm, takes the “host-guest relationship” of tourist attractions as the main line, integrates spatial information and image information, and explores issues such as which attractions attract tourists from tourism photos, what characteristics the spatial distribution differences in different tourist attractions present, and what combination relationships exist between attractions, to clarify the situation of tourist attractions and their associated combinations in Qingdao from the perspective of tourism demand. It is expected to expand the research boundary on identifying and applying tourist attractions through images, and provide guidance for tourism development, market demand matching, product design, and management optimization.

2. Materials and Methods

2.1. Study Area

Qingdao is located in the southern part of the Shandong Peninsula, bordering the Yellow Sea to the east. It has a total coastline of 905.2 km and a sea area of 11,700 square kilometers. The sea area is home to numerous bays, including 49, such as Fushan Bay and Taiping Bay. From west to east, six major bay areas have been formed: Guzhenkou-Dongjiakou Bay Area, West Coast Qianhai Bay Area, Jiaozhou Bay Area, East Coast Qianhai Bay Area, Laoshan Bay Area, and Dingziwan–Tianheng Bay Area (Figure 1). The eastern part of Qingdao is supported by the Laoshan Mountain Range, forming a semi-enclosed natural harbor, which provides superior natural environmental conditions for tourism development. Since implementing the “13th Five-Year Plan”, Qingdao’s tourism revenue has accounted for 20.82% of Shandong Province’s total and 4.6% of the national tourism revenue, respectively. As Qingdao enjoys high popularity and influence as an important coastal tourist resort city and a major central city in China. Selecting Qingdao as the research area is representative for exploring the identification and associated features of tourist attractions, which can provide references for research on tourist attractions.

2.2. Data Sources and Processing

With the iteration and update of mobile terminal devices and online social media, images have become an important carrier for the co-construction and sharing of tourist destinations. Weibo1 is a powerful social media platform that integrates users’ public self-expression, social interaction, and content aggregation. It has a large user base and can provide rich and timely tourism image data for the study. Compared with other mainstream social media platforms, Weibo allows users to add precise geographic coordinates to the pictures they post actively, and this function is widely used. This feature constitutes irreplaceable data support for the core foundation of spatial dimension exploration in this study. Therefore, this study selects the Weibo client as the source of tourism image data, crawls photos with Qingdao geographic information published by tourists from 1 January 2020 to 3 November 2023, using Python 3.10.12, and retrieves users’ attribute information on Weibo to obtain data such as users’ IP locations, age, gender, and credit status. On this basis, Qingdao’s tourism image information database and user attribute information database are established, respectively. Secondly, the data is processed and cleaned, and core inclusion criteria for data are established: Weibo content must contain at least one picture and be generated content with precise geographic coordinates. The user account status for publishing Weibo must be normally available. For data cleaning and exclusion, to ensure data reliability, users with official Weibo credit ratings marked as “poor” or “very poor” are excluded; user data with abnormal ages (under 18 or over 100 years old) are removed; and duplicate publications of the same pictures are deduplicated.

In data processing, users with poor credit and abnormal ages are excluded to ensure data reliability. After data cleaning, 10,032 users and 59,175 images were obtained, with males accounting for 24.70% and females accounting for 75.51% (

Table 1. Description of Weibo Users and Image Quantity.

	Number of Users	Number of Photo Releases	Gender Distribution (%)
			Male	Female
Descriptive Statistics	10,032	59,175	24.70%	75.51%

). It is worth noting that female users in the sample are significantly more than male users, and this phenomenon is closely related to the user behavior characteristics of social media platforms: existing studies have shown that female users are usually more inclined to actively and frequently share life dynamics and visual content on platforms such as Weibo, while male users’ public sharing behavior is relatively less [17]. In addition, adding precise geographic tags to photos is a specific sharing behavior, which may be more in line with the habits of some user groups, especially those who focus on online image construction and visual expression. Therefore, this dataset can more accurately reflect the gaze focus and shooting behavior of user groups who actively and publicly share their Qingdao tourism experience on the Weibo platform.

2.3. Research Methodology

2.3.1. Image Semantic Segmentation

Image Segmentation is classifying image pixels to divide regions with the same properties. This study builds a Python environment, invokes models, imports images, sets labels for visual element types, and conducts numerical analysis on image content. With reference to existing representative research results shown in Figure 2, this study adopts the DeepLabv3+ model architecture and integrates the ResNeST-269 backbone network for the semantic segmentation of tourism images. After inputting the original photos, based on the pre-trained weights of the ADE20K dataset2. The model identifies element categories through pixel-by-pixel classification using an encoder–decoder structure. Its core atrous spatial pyramid pooling module uses multi-scale atrous convolution to capture targets of different sizes accurately, and the decoder fuses shallow boundary features to optimize segmentation accuracy, ultimately outputting pixel-level classification labels.

2.3.2. Association Rule Algorithm (Apriori)

Association rules, first proposed by Agrawal [18], are used to mine implicit co-occurrence relationships between items. This study regards “attraction elements identified in each image” as “items”, and “each image” as a “transaction”, to explore the combination rules of attractions. The core steps are simplified as follows:

• Data preparation: Construct a transaction database (59,175 transactions in total) where each transaction contains all attraction elements of one image.
• Threshold setting: Refer to field standards, set minimum support (minsup) = 1% (itemsets must appear in ≥592 images) and minimum confidence (minconf) = 50% (rules must be reliable in ≥50% of transactions).
• Frequent itemset mining: Generate 1-frequent itemsets (single attractions meeting minsup) → connect to generate candidate k-itemsets → prune non-frequent subsets → retain k-frequent itemsets.
• Association rule extraction: Screen strong association rules (meeting minsup and minconf) and interpret core combination patterns.

2.3.3. Kernel Density Analysis

Kernel density estimation reflects elements’ spatial distribution, agglomeration, and dispersion characteristics by calculating the quantity per unit area of geographical elements. To further clarify the spatial distribution of tourist attractions, kernel density analysis is used to explore their spatial distribution characteristics further, analyzing the overall spatial distribution of attractions, natural attractions, and humanistic attractions. Considering that visual elements such as sky, tree, person, ground, road, and rock are common in tourism scenes but not core attractions, the study defines them as environmental noise factors. Although these visual elements appear frequently, they are difficult to reflect the unique attractiveness of the destination, so they are excluded from spatial visualization analysis, focusing on more iconic core elements. The spatial distribution of iconic attractions such as sea, mountain, and sand in natural attractions, and buildings, signs, and skyscrapers in humanistic attractions is visualized.

3. Results

3.1. Tourist Attractions Element Extraction Results Under Image Segmentation

Based on the image segmentation results, the nodes, counts the nodes, frequency, frequency rate, and visual proportion in images of the extracted attractions, and sorts them according to their occurrence frequency and visual proportion. As shown in

Table 2. Frequency and visual proportion of attraction elements in travel images (top 20).

Rank by Frequency				Rank by Visual Proportion
Rank	Element Node	Frequency	Frequency (%)	Rank	Element Node	Image Ratio (%)
1	Building	54,043	91.33%	1	Sky	21.04%
2	Person	53,857	91.01%	2	Tree	10.41%
3	Sky	49,418	83.51%	3	Person	10.07%
4	Tree	41,654	70.39%	4	Building	8.68%
5	Rock	41,010	69.30%	5	Sea	7.73%
6	Sea	40,848	69.03%	6	Ground	5.67%
7	Ground	40,069	67.71%	7	Road	4.03%
8	Sign	34,316	57.99%	8	Plant	3.78%
9	Plant	32,372	54.70%	9	Mountain	2.85%
10	Mountain	31,776	53.70%	10	Rock	2.65%
11	Sand	28,658	39.29%	11	Sidewalk	2.16%
12	Chair	28,422	48.03%	12	Sand	2.08%
13	Grass	28,354	47.91%	13	Grass	1.90%
14	Windowpane	26,366	44.56%	14	Seat	1.77%
15	Sidewalk	23,255	39.30%	15	Table	1.59%
16	Table	25,625	43.30%	16	Bed	1.12%
17	Seat	25,312	42.77%	17	Sign	1.09%
18	Curtain	23,389	39.52%	18	Skyscraper	1.04%
19	Door	22,525	38.06%	19	Windowpane	0.82%
20	Bed	19,288	32.59%	20	House	0.48%

, regarding occurrence frequency, the extracted attractions in Qingdao’s tourism photos during the study period include humanistic and natural landscapes, with outdoor scenes as the main visiting scenes. Specifically, building, person, and sky rank the top 3 in occurrence frequency, with frequency rates above 80%. Among them, the occurrence frequencies of building and person reach 91.33% and 91.01%, respectively, indicating that tourists’ participation in tourist attractions is strong, and buildings in humanistic landscapes are more popular among tourists. In terms of the visual proportion of attractions in images, natural landscapes such as sky and trees rank the top two, becoming the main visual elements of tourism photos. This relates to the fact that sky, tree, etc., occupy a large visual area in the scene. Secondly, the person, building, and sea become the main visual elements of Qingdao’s tourism scenes.

To further clarify the composition of tourist attractions in Qingdao, referring to existing studies [19], the tourist attractions extracted from Qingdao’s tourism images during the study period are sorted and classified (

Table 3. Classification of tourist attraction elements in Qingdao.

Type of Tourist Attraction	Specific Elements	Examples
Natural Tourist Attractions	Geomorphological Element	Mountain, Rock, Sand, etc.
	Hydrological Element	Ocean, etc.
	Climatic Element	Sky, etc.
	Biological Element	Tree, Plant, Grass, etc.
Cultural Tourist Attractions	Outdoor Elements	Building, Ground, Road, Sidewalk, Sign, Skyscraper, etc.
	Indoor Elements	Seat (Chair), Table, Bed, Windowpane, House, etc.

Note: Elements are listed in order of their visual proportion ranking.

). It can be seen from the statistical classification of tourist attractions in Qingdao that, in terms of natural tourist attractions, sky, tree, sea, plant, mountain, rock, sand, grass, etc., have relatively large visual proportions, forming the main natural landscapes of tourist attractions in Qingdao. Humanistic attractions are mainly divided into indoor and outdoor elements. Among them, outdoor elements mainly include building, ground, road, sidewalk, sign, skyscraper, etc., as the main visual attraction elements. Indoors include seat, table, bed, windowpane, house, etc., as the main attractions.

3.2. Spatial Distribution Characteristics of Tourist Attraction Elements

(1) Overall Pattern: The overall tourist attractions show circular multi-core structure and land–sea gradient of “distributed around bays and agglomerated towards the sea”

Overall, tourist attractions in Qingdao exhibit the spatial distribution characteristic of “distribution around bays and agglomeration towards the sea”, with a certain circular structure and multi-level core distribution pattern. Specifically, tourist attractions are distributed around Jiaozhou Bay, presenting a core–periphery structure pattern characterized by agglomeration along the coast and dispersion inland. A multi-core distribution pattern has been formed, where the first-level agglomeration cores include Qingdao Bay, Huiquan Bay, and Taiping Bay in the Eastern Coast Qianhai Bay Area of Shinan District, Maidao Bay in Laoshan District, and Tangdao Bay and Lingshan Bay in the Western Coast Qianhai Bay Area of Huangdao District; the secondary agglomeration cores are located in the centers of Chengyang District, Licang District, Jimo District, Jiaozhou City, Pingdu City, and Laixi City. It can be observed that the spatial distribution of tourist attractions in Qingdao shows an obvious sea-oriented feature: tourist attractions along the coastal line exhibit prominent spatial agglomeration, and the degree of agglomeration gradually decreases from the coast to inland areas. Among them, tourist attractions in county-level cities such as Pingdu City and Laixi City are mainly distributed in a scattered point-like manner. In general, the core attractiveness of marine resources is the fundamental reason for shaping this pattern. As a coastal city, Qingdao’s core tourism image and experience are highly dependent on the coastline. From the inland sea to the inland areas, the distribution of tourist attractions transitions from highly agglomerated cores to scattered secondary centers and sporadic point-like distributions, forming an obvious “land-sea gradient”.

(2) Natural Tourist Attractions: Significant differences between land and sea under coast dependence and geomorphic constraints, showing a distribution pattern of “small agglomeration and large dispersion”

Overall, natural tourist attractions in Qingdao generally exhibit the spatial distribution characteristic of “small-scale agglomeration and large-scale dispersion”, with agglomeration centers formed along the coastal zone and distributed in a point-like divergent spatial pattern. Specifically, the sea and sand elements show obvious coastal regional characteristics: the sea element has a core agglomeration area at the western end of the Eastern Coast Qianhai Bay Area, extending from Qingdao Bay in the west to Maidao Bay in Laoshan District in the east, mainly distributed in the southern parts of Shinan District, Shibei District, and Laoshan District, and scattered in a point-like manner around Jiaozhou Bay in Jiaozhou City, Licang District, Jimo District, and other areas. The spatial distribution of sand elements is relatively consistent with the sea elements, showing a distinct coastal distribution. They are concentrated in the western end of the Eastern Coast Qianhai Bay Area and the eastern end of the Western Coast Qianhai Bay Area, forming a spatial distribution pattern with Qingdao Bay, Huiquan Bay, and Taiping Bay in Shinan District as the main cores, and Tangdao Bay in Huangdao District and Maidao Bay in Laoshan District as secondary cores. It can be seen that sand elements are mainly distributed alongside sea elements; representative sand elements include the beaches of coastal bathing areas along the Qianhai coast, Huangdao Golden Beach, and Laoshan Yangkou Beach. Mountain elements show a spatial distribution characteristic of “coastal agglomeration and multi-core dispersion”: each county and district has formed obvious mountain agglomeration areas, with Shinan District, Shibei District, and Laoshan District as the first-level cores and other counties and districts as secondary cores. It can be observed that mountain elements in all counties and districts have been well developed. Among them, coastal mountain elements such as Fushan Mountain, Taiping Mountain, Qingdao Mountain, Signal Mountain, and Laoshan Mountain are more favored by tourists. The distribution of natural attractions such as the sea and sand is highly dependent on natural coastlines, with immobility and scarcity, which is also the fundamental reason for the formation of coastal agglomeration. At the same time, this also reflects tourists’ strong preference for Qingdao’s “coastal” image.

(3) Humanistic Tourist Attractions: Strong participation of tourists as the main body, forming coastal agglomeration groups

Overall, the spatial distribution of humanistic tourist attractions in Qingdao is mainly agglomerated at the western end of the Eastern Coast Qianhai Bay Area, with secondary cores scattered in a point-like manner from the coast to inland areas across various counties and districts. In terms of building elements, they mainly present a “core area-isolated point” distribution pattern: specifically, a core agglomeration area is formed in Shinan District. In contrast, other regions are distributed in isolated points. It can be seen that building elements in Shinan District have been developed and utilized to a large extent; representative ones include modern Western-style building clusters (such as the Badaguan Building Complex) and characteristic former residences of celebrities (such as Kang Youwei’s Former Residence and Lao She’s Former Residence), which are popular among tourists. Other counties and districts mainly develop in isolation and manner and have not yet formed planar agglomeration areas. Sign elements are primarily distributed around bays and form coastal agglomeration clusters, located in Shinan District, Shibei District, the eastern part of Huangdao District, and other counties and districts. In terms of skyscraper elements, their distribution characteristics are consistent with those of humanistic tourist attractions, showing an overall bay-surrounding distribution; among them, southern coastal urban areas of Qingdao, such as Shinan District and Laoshan District, have become popular check-in spots for tourism. Shinan District, as the historical birthplace of Qingdao, concentrates many modern historical buildings, former residences of celebrities, and characteristic blocks, serving as the core historical foundation of humanistic attractions. Tourists’ perception of Qingdao’s humanistic landscapes is highly focused on the historical urban style of “red tiles, green trees, blue sea, and blue sky” and the image of a modern coastal city. However, the inland areas of Qingdao lack comprehensive humanistic attraction clusters of the same scale and popularity (Figure 3).

3.3. Association Analysis of Tourist Attractions

3.3.1. Types of Tourist Attraction Combinations

This study uses the Apriori association algorithm in SPSS Modeler 18.0 to analyze the association characteristics of attractions. Referring to existing research threshold standards [20], and combining the distribution characteristics of tourism image data, the minimum support is set to 10% and the confidence to 80% to balance the significance of rules and data coverage, resulting in a total of 785,161 association rules. Frequent itemsets refer to combinations of items that appear frequently in the data under the minimum support and minimum confidence thresholds; mining frequent itemsets of tourist attractions helps to clarify the association and combination characteristics between attractions. To further understand the combination of tourist attractions in Qingdao, this study statistically classifies the top 10 frequent itemsets with more than 4 items.

As shown in

Table 4. Top 10 frequent itemsets of tourist attraction elements in Qingdao.

Rank	Itemset	Frequent	Support (%)	Confidence (%)	Combination Type
1	{Building Sky Person Tree}	51,998	87.88%	87.21%	Building-Centric Type
2	{Sea Sky Person Tree}	38,372	64.85%	93.94%	Sea-Centric Type
3	{Building Sea sky person}	38,372	64.84%	94.94%	Building-Sea Combination
4	{Rock Building Sky person}	38,075	64.34%	92.81%	Cityscape-Centric Type
5	{Building Sky Rock person}	33,214	56.13%	81.00%	Cityscape-Centric Type
6	{Building Tree Rock person}	34,013	57.48%	82.94%	Cityscape-Centric Type
7	{Tree ground rock person}	32,572	55.04%	79.41%	Natural Landscape Type
8	{Signboard sky building person}	31,238	52.79%	91.02%	Sign-Centric Type
9	{Sky railing person building}	28,564	48.27%	69.66%	Cityscape-Centric Type
10	{Tree road building person}	27,821	47.01%	67.84%	Cityscape-Centric Type

, generally speaking, after excluding environmental noise such as sky and tree, tourist attractions in Qingdao take the sea and buildings as the main elements, and are divided into four combination types (building-core, sea-core, cityscape-core, and sign-core) according to the core nature of the dominant elements in the frequent itemsets and the theme characteristics of the scenes. Specifically, {Building, Sky, Person, Tree}, {Sea, Sky, Person, Tree}, and {Building, Sea, Sky, Person} rank among the top 3 frequent itemsets, forming important attraction combination types for tourists’ visits. Among them, {Building, Sky, Person, Tree} has the highest number of instances (51,998) and a support of 71.96%, indicating that buildings are the primary element of tourist attractions in Qingdao, which is consistent with the results of the frequency and visual proportion of tourist attractions. In addition, around building and sea elements, other itemsets are generated with visual elements such as rock and sign; for example, {Rock, Building, Sky, Person} and {Building, Sky, Rock, Person} form the cityscape-core type. This is related to Qingdao’s granite building appearance and the urban style integrating the sea and the city; the unique architectural style and building materials attract tourists. Overall, the combination types formed by the frequent itemsets of major attractions are building-centric, sea-centric, cityscape-centric, and sign-centric. It can be seen that urban buildings, the sea, signs, and other elements constitute the main components of tourist attraction combinations in Qingdao.

3.3.2. Association Characteristics of Tourist Attractions

Clarifying the association characteristics of tourist attractions helps to understand the association features of different attractions and tourists’ demand preferences. This study analyzes the internal association characteristics of tourist attractions through strong association rules centered on core attractions. Based on the proportion of visual elements in the images, emphasis is placed on analyzing the strong association rules of the sea, mountains, and sand in natural tourist attractions, and buildings, signs, and skyscrapers in humanistic tourist attractions. As shown in

Table 5. Strong association rules of tourist attraction elements in Qingdao.

Type of Attraction Element	Rank	Association Rule		Example	Support	Confidence
		Antecedent	Consequent
Natural Attraction Elements	1	Sea	{Sand Mountain Rock Sky}	8391	14.17%	20.54%
	2	Mountain	{Sea Rock Building Sky}	15,441	26.09%	48.59%
	3	Sand	{Sea}	23,249	39.29%	81.13%
Cultural Attraction Elements	1	Building	{Person}	53,857	91.01%	93.80%
	2	Sign	{Pedestal Windowpane Floor}	14,793	25.00%	43.12%
	3	Skyscraper	{Sky Person}	4043	6.83%	80.10%

, generally speaking, there are different association rules between different attractions: in terms of natural tourist attractions, the sea element has a strong correlation with natural elements such as sand, mountain, rock, and sky; the mountain element has a strong correlation with the sea, rock, building, sky, and other elements, jointly forming the characteristic “mountain-sea-city” tourism landscape of Qingdao; the sand element has a high correlation with the sea element—as a naturally formed landscape on the seaside, sand has a natural connection with the sea. Regarding humanistic tourist attractions, there are two groups of elements with strong correlations: The association rule between building and person, the association rule between skyscraper and person. On the one hand, tourists in urban buildings become part of the overall landscape; on the other hand, buildings are favored by tourists, and there are frequent interactions with buildings, so tourists have a closer connection with building elements. Sign elements strongly correlate with indoor elements such as windows and floors.

4. Discussion and Conclusions

4.1. Discussion

Current research focusing on extracting tourist attractions based on big image data is insufficient. This study aims to identify the attribute characteristics of tourist attractions in tourism photos through the emerging method of image semantic segmentation to deeply explore tourists’ preferences for attractions, breaking through the efficiency bottleneck of traditional manual coding. This is consistent with the direction of “automated analysis of big tourism visual data” advocated by Zhan [21], providing an extensible technical path for quantitatively representing tourist attractions. The study found that “buildings” and “sea” are the core visual elements of tourist attractions in Qingdao, which confirms Lew’s proposition that “destination image is embodied in iconic landscapes” [22], and also echoes Orams’ conclusion on the “sea-dominated visual narrative” in coastal cities [23]. However, it should be noted that over-reliance on visual proportion may ignore symbolic meanings. There are still shortcomings in the research on the value orientation and functional demands of tourist attractions: first, in the identification of tourist attractions, this study uses image segmentation technology to identify tourist attractions, mainly through image proportion and occurrence frequency for analysis, but has not discussed the impact of visual attention; in addition, this study only discusses fixed tourist attractions through image segmentation, while lacking discussion on non-material tourist attractions. Therefore, in the future, text mining should be combined to explore tourist attractions implied beyond the images.

The circular spatial structure of Qingdao’s tourist attractions with “bay-surrounding and sea-oriented” characteristics confirms McKercher’s core–periphery theory [24], while the land–sea differentiation of natural tourist attractions highlights the rigid constraints of resource endowment. Coastal mountains, as popular spots, echo the value of “landscape viewing points” proposed by Oku & Fukamachi [25]. The “isolated point-like” distribution of humanistic attractions in inland areas exposes the insufficient regional coordination in Qingdao’s current tourism development, reflecting the “fragmented development” problem referred to by Dredge [26]. This contrasts with international coastal cities such as Barcelona and Sydney, which balance tourist flows through multi-center networks [27]. However, Qingdao, as a coastal tourist destination, has distinct seasonal characteristics; due to the fluctuation of Weibo data over time with user changes, this study has not further analyzed seasonal variations, and empirical analyses such as comparative studies on differences between multiple case destinations, data sources, and temporal continuity need further verification. Finally, the gaze process in the construction of tourist attractions needs to be clarified, and the symbolization of the tourist attraction system needs to be further constructed to expand the breadth and depth of the research.

Finally, the strong correlation between “buildings and the sea” obtained in this study confirms Qingdao’s tourism image of coordinated narrative integrating mountains, sea, and city. At the same time, the combination of signs and indoor elements points to the functional dependence of commercial spaces. The high-frequency co-occurrence of “person and building” empirically supports Willson’s “embodied gaze” theory—that the value of attractions is activated through tourists’ behaviors [28]. However, although the Apriori algorithm mines combination rules, it does not distinguish between active selection and background coexistence; in the future, it is necessary to introduce further visual saliency models such as eye tracking to assist in interpretation.

4.2. Conclusions

With the emergence of large-scale image data, visual scene recognition has become an important way to explore tourist attractions. This study takes images crawled from Qingdao’s Weibo check-ins as the data source, parses the visual representation of tourist attractions under tourism gaze through image semantic segmentation, and analyzes the composition, spatial distribution, and association characteristics of tourist attractions in Qingdao based on kernel density analysis and the Apriori association algorithm. The purpose is to expand the application boundary of research on tourist attraction identification through images, provide practical guidance for tourism development, and have certain practical significance for exploring tourist attraction identification through images. It also provides a standardized analysis framework for tourist attraction research and deepens the attraction mechanism of “co-creation by hosts and guests” in tourism. The main conclusions are as follows:

(1)

From the results of tourist attractions extracted based on image segmentation, tourist attractions in Qingdao include humanistic and natural landscapes, with outdoor elements as the main visiting scenes. In terms of occurrence frequency, elements such as buildings, persons, sky, trees, and rocks appear frequently in the images. In terms of the visual proportion of attractions in the images, after excluding environmental noise such as sky, trees, and persons, buildings and the sea are the main visual elements of tourist attractions in Qingdao. Through further induction and classification of tourist attractions, tourist attractions in Qingdao can be divided into natural tourist attractions and humanistic tourist attractions: natural tourist attractions mainly include geomorphic, hydrological, climatic, and biological elements, while humanistic tourist attractions consist of outdoor elements and indoor elements.

(2)

From the perspective of the spatial distribution of tourist attractions, tourist attractions in Qingdao generally present a spatial pattern of “distribution around bays and agglomeration towards the sea”, showing a core–periphery structure pattern of agglomeration along the coast and dispersion inland. Different tourist attraction elements show different spatial distribution characteristics: natural tourist attractions form agglomeration centers along the coastal zone and spread in a point-like manner, showing distinct coastal regional characteristics—among them, coastal and sand elements are distributed alongside each other, concentrated at the western end of the Eastern Coast Qianhai Bay Area; mountain elements show the characteristic of “coastal agglomeration and multi-core dispersion”. Humanistic tourist attractions are mainly agglomerated at the western end of the Eastern Coast Qianhai Bay Area, with secondary cores scattered in a point-like manner from the coast to inland areas across various counties and districts—among them, building elements are mainly distributed in a planar agglomeration along the coast and in an isolated point-like manner in inland areas; sign elements are distributed around bays and form coastal agglomeration clusters; skyscraper elements are basically consistent with the overall distribution of humanistic tourist attractions.

(3)

From the perspective of the association characteristics of tourist attractions, tourist attractions in Qingdao form frequent itemsets of attractions such as {Building, Sky, Person, Tree}, {Sea, Sky, Person, Tree}, and {Building, Sea, Sky, Person} around elements such as buildings, the sea, and signs, which can be summarized into attraction combinations such as building-centric, sea-centric, cityscape-centric, and sign-centric. In terms of association rules, among natural tourist attractions, the sea has a strong correlation with natural environmental elements such as sand and mountains; among humanistic tourist attractions, persons has a strong correlation with buildings and skyscrapers, forming strong association rules; and sign elements mainly have a strong correlation with indoor elements.