Dynamic User Tourism Interest Modeling through Domain Information Integration: A Hierarchical Approach

Abstract

With the exponential growth of online review platforms, understanding user preferences and interests in the tourism domain has become increasingly critical for businesses and service providers. However, extracting meaningful insights from the vast amount of available data poses a significant challenge. Traditional methods often struggle to capture the nuanced and hierarchical nature of user interests within the tourism domain. This paper pioneers the integration of domain information modeling technology into the realm of online review information mining, presenting a novel approach to constructing a user tourism interest model. Unlike existing methods, which primarily rely on flat or simplistic representations of user data, our approach leverages the hierarchical structure inherent in tourism domain information modeling. By harnessing big data within the tourism domain, we construct hierarchical tourism attributes and apply a conditional random field model along with an affective dictionary to facilitate the hierarchical mining of user travel interest information. This culminates in the establishment of a comprehensive user travel interest model using advanced information modeling techniques. Building upon this foundation, we further propose a dynamic user travel interest model, showcasing its adaptability and responsiveness to changing user preferences. Finally, we validate the accuracy and effectiveness of our model through simulation experiments within a user travel recommendation system, demonstrating significant improvements over traditional methods.

1. Introduction

Tourism has become a significant aspect of people’s lives, leading to the emergence of various highly credible tourism service platforms. Analysis of these platforms reveals that obtaining personalized tourism interest information is crucial to addressing the diverse needs of users. Despite this, the development of a comprehensive user tourism interest model remains incomplete, with several critical issues yet to be resolved [1,2,3,4]. Consequently, the methods for collecting tourism interest data and representing the model are pivotal challenges in constructing an effective user tourism interest mode [5,6].

Research Gap 1: Comprehensive User Tourism Interest Model

While multiple studies indicate that users’ evaluations of various product attributes contain richer information about their interests compared to single evaluations, there is still a lack of comprehensive user tourism interest models that accurately identify user needs based on these detailed evaluations [7,8]. This approach offers the most direct and authentic feedback on tourism products and services, enhancing the accuracy of users’ tourism interest information. It is a critical factor in ensuring the precision of tourism recommendation systems [9,10,11,12].

Research Gap 2: Effective Data Source Utilization

Currently, there is a need for effective methods to utilize online comments from tourism platforms as the data source for users’ tourism interest information. These platforms provide a wealth of user-generated content that, if properly harnessed, can significantly improve the modeling of user interests [13].

Research Gap 3: Hierarchical Information Representation

The information-based model can accurately depict user interest through the hierarchical structure of information among concepts, yet there is a need for a more robust implementation of this approach in the tourism domain [13,14]. Specifically, there is a hierarchical relationship between the overall attributes and local attributes of tourism products, as well as between the overall evaluation of tourism products and the evaluation of their attribute characteristics [15,16]. Therefore, the information-based model is more suitable for constructing users’ tourism interest models [17], but its application has been limited [4].

In summary, this paper addresses these research gaps by pioneering the integration of domain information modeling technology into the realm of online review information mining. By leveraging big data within the tourism domain and constructing hierarchical tourism attributes, we developed a comprehensive and dynamic user travel interest model. This model is validated through simulation experiments within a user travel recommendation system, demonstrating significant improvements over traditional methods [12,18,19].

2. Construction of Tourism Domain Information

The construction process is divided into two stages: the first stage involves creating a thesaurus of tourism, and the second stage involves building an information library of topic headings, transforming free tags into highly formalized information. The tourism theme words are sourced from major tourism platforms, ensuring that the thesaurus aligns with the expression habits of users’ online reviews. This alignment facilitates the subsequent mining of online tourism reviews.

The construction of the concept system is a crucial aspect of tourism information construction. It includes five parts: determining the core class of information, establishing the hierarchy of classes, defining the relationships between classes, defining the attributes of classes, and creating examples of classes.

The core concepts of tourism domain information are utilized in mining online review information. Consequently, the top concepts of the tourism domain are extracted based on the content features of online review information and abstracted into four core concepts, as illustrated in Figure 1.

3. Tourism Online Review Information Mining

The research process for the hierarchical mining of tourism information in this paper is outlined as follows. The approach involves mapping feature words from online tourism reviews to different levels of concepts using domain information and leveraging the combinations of these feature words [9]. This method enables multi-level information mining with varying granularity.

3.1. Tourism Online Review Text Preprocessing

The general steps of tourism online review text preprocessing include the following:

• Spam review filtering: Tourism online review sets contain a large number of online reviews consisting of punctuation marks, emoticons, and single words, in which any useful information cannot be extracted. These spam reviews also affect the efficiency of the model, so these need to be filtered;
• Word segmentation: The attribute of tourism online review text word segmentation is extended by using the word segmentation system ICTCLAS2013 developed by the Institute of Computational Technology of the Chinese Academy of Sciences;
• Attribute extension: While the word segmentation software works well with regular text, tourism online reviews differ significantly from formal texts. They include many internet terms and specific combination words that are challenging for the software to accurately process. To address this, a two-dimensional sliding window is applied to the sequence of words after segmentation, allowing two adjacent words to form a new feature word. This process can be illustrated with an example: Original segmented text: [“beautiful”, “scenery”, “but”, “poor”, “service”], after applying the sliding window: [“beautiful scenery”, “scenery but”, “but poor”, “poor service”]. Here, the commentary with n words expands to n − 1 new feature words [13]. This approach helps in reducing dimensionality by extending online words and special combination words to feature attributes;
• Attribute dimension reduction: There are only a few dozens of words in tourism online reviews, but the whole attribute of a tourism domain can have thousands of attributes, which may reach tens of thousands after the attribute is extended. Tourism online reviews can be expressed as a sparse vector of high dimension. Tourism review information will be submerged in the process of mining, so it is necessary to reduce the dimension of feature attributes. In this paper, feature frequency and information gain are used to reduce the dimension of the attribute. Firstly, the word frequency which is less than the threshold λ is eliminated, then, discarding feature words with small information gain can achieve effective dimension reduction. The information gain calculation formula of the characteristic word tk is as follows:

  $$IG\left(t_k\right)=-{\displaystyle \sum _{i=1}^{n}P\left(c_l\right)}log\left(c_l\right)+P\left(t_k\right){\displaystyle \sum _{i=1}^{n}P\left(c_l|t_k\right)}•logP\left(c_l|t_k\right)+P\left(\overline{t_k}\right){\displaystyle \sum _{i=1}^{n}P\left(c_l|\overline{t_k}\right)}logP\left(c_l|\overline{t_k}\right)$$

  (1)

  P(ci) is the prior probability of ci class. P(tk) is the occurrence probability of characteristic attribute tk on the whole training set. $P\left(\overline{t_k}\right)$ is the probability of not appearing of characteristic tk. P(cl|tk) is the probability that the text belongs to cl class under the condition that the characteristic attribute tk exists. $P\left(c_l|\overline{t_k}\right)$ is the probability that the text belongs to cl class under the condition that the characteristic attribute tk does not exist;
• Removing useless words: Those words, such as prepositions, pronouns, and stop words, which appear frequently in the text but have no practical meaning and cannot represent the features of text information, need to be removed;
• Part-of-speech tagging: The part-of-speech tagging of feature words is completed based on the lexicon and tourism domain information. This can be illustrated more concretely as follows: Using a lexicon and tourism-specific terms, each word in the text is tagged with its part of speech. For example, “beautiful” (adjective), “scenery” (noun), “poor” (adjective), “service” (noun). This tagging helps in understanding the role each word plays in the text and aids in more accurate feature extraction and analysis.

3.2. Tourism Online Review Information Representation

Tourism online review information is mapped to concepts based on WordNet, allowing it to be expressed in the form of concepts. The feature vector dimension of tourism review information is reduced through the semantic relationships between concept nodes and concepts [14]. This information-based representation better reveals the semantic level of text knowledge and significantly reduces feature vector dimensions. However, current general information only captures simple relationships between concepts, failing to uncover deeper semantic relationships. Additionally, text mapping can only match product attributes explicitly mentioned in the text. Due to the randomness in tourism online reviews, reviewers often omit review objects, and some tourism review objects are not explicitly expressed. Consequently, implicit product attributes cannot be extracted through information mapping, leading to the loss of some text information.

This paper introduces a method of hidden product attribute extraction based on semantic web rule language (SWRL) rules and accomplishes the extraction of hidden product attributes in online reviews through the reasoning function of information and SWRL. Abstract syntax is used to represent the knowledge described in information, and information-based semantic rule reasoning is realized.

The vector space model (VSM) is used to express the text of a tourism online review as a vector composed of feature item c, information relation l, and weight value of feature item wc. The text is expressed as:

$$VC\left(D_i\right)=\left\{\left(c_1,l_1,w{c}_{i1}\right),\left(c_2,l_2,w{c}_{i2}\right),\dots ,\left(c_n,l_n,w{c}_{in}\right)\right\}$$

(2)

The concept-based text representation processing method is as follows:

• Concept mapping. Tourism online review text can be seen as a set of words after preprocessing. Through the matching between words and information concepts, words are converted into concepts, so that the text can be expressed as a concept-based vector model;
• Extraction of information relations. The information relations that need to be extracted include integral–partial relations and demonstrative relations. The integral–partial relation is used to map the hierarchy of concept to that of product attribute, and the indicator relation is used to preserve the relationship between product attribute and evaluation word;
• Calculation of feature weights (wc). Each product has many attributes, but the function of each attribute is different. In this paper, we use classical TF.IDF (term frequency–inverse document frequency) to calculate the weight of feature item.

fij is the number of times the word ci appears in document j. The frequency of term ci in document j:

$$T{F}_{ij}={\displaystyle \frac{f_{ij}}{{max}_k{f}_{kj}}}$$

(3)

If the word ci appears in ni of all online reviews, the IDF definition of the word is as follows:

$$ID{F}_i={log}_2{\displaystyle \frac{N}{n_i}}$$

(4)

Then,

$$w{c}_{ij}=T{F}_{ij}\times ID{F}_i$$

(5)

3.3. Hierarchical Mining of User Tourism Interest Information

Utilizing the integral–partial relation within information l, we can construct a hierarchical set of tourism product attributes. Leveraging the hierarchical structure of tourism domain information, we establish a hierarchical tourism product attribute set. The pivotal concept in mapping tourism domain information to product attributes is the “comment object”. The “comment object” refers to the specific aspect of the tourism product being reviewed, such as overall experience, location, cleanliness, or service. It serves as the root node in the hierarchical tourism product attribute set, accommodating diverse domain information requirements due to variations in comment objects, which determine domain distinctions.

Tourism product attributes encompass both explicit and implicit attributes. Explicit attributes can be directly extracted from the results of concept mapping. Conversely, implicit attributes require reasoning aided by SWRL rules and information to extract effectively.

To construct the “hierarchical user interest,” we map user comments to a structured set of attributes and sub-attributes. This hierarchical structure allows for a more detailed and nuanced understanding of user interests. For instance, a user’s review mentioning “excellent room service” would be mapped to the “service” attribute and further broken down into the sub-attribute “room service”. This hierarchical mapping helps in capturing both broad and specific user interests, enhancing the model’s ability to provide accurate recommendations.

The emotion orientation analysis of tourism online reviews aims to assess the evaluation of the entire tourism product, its attribute classes, and the characteristics of each attribute. This analysis is conducted in three steps:

• Overall Evaluation Analysis: Traditionally, classification based solely on polarity fails to adequately capture users’ attitudes towards tourism products. Therefore, it is necessary to further analyze emotional intensity. Utilizing the standard five-star scoring system common on tourism platforms, evaluations are classified into {1, 2, 3, 4, 5}, corresponding to very dissatisfied to very satisfied, respectively.
• Attribute Class Evaluation Analysis: While overall evaluations provide insight into users’ responses to tourism products, individual users often have specific requirements for product details. Hence, it is essential to identify sentences in reviews related to each tourism product attribute class. Evaluations for various attribute classes are conducted using a five-classification process {1, 2, 3, 4, 5}, with each star representing very dissatisfied to very satisfied, respectively.
• Attribute-Level Evaluation Analysis: Each attribute class contains multiple features, representing the lowest-level characteristics of a tourism product. Sentences mentioning a specific attribute class are segmented into single tourism product attribute phrases. Subsequently, evaluations are conducted for each tourism product attribute using a five-classification process {1, 2, 3, 4, 5}, with each star indicating very dissatisfied to very satisfied, respectively.

By employing these steps, the evaluation process aligns closely with users’ general tourism preferences, providing nuanced insights into their sentiment towards tourism products and their attributes.

The first and second layers of emotion orientation analysis in this paper adopted the method of CRFs, and the third layer used the method based on emotion dictionary. CRFs excel in capturing context-dependent emotional cues and handling long sequences of text, providing precise emotion labeling. In contrast, the emotion dictionary method offers stability and detailed emotion classification through a predefined vocabulary. By combining these approaches, the paper leverages the strengths of both methods: CRFs for accurate context-based analysis, and emotion dictionaries for consistent and detailed classification. The basic analysis flow is shown in Figure 2.

So far, the hierarchical mining model of user tourism interest information has been completed. Using Xi’an as an example, the results of hierarchical mining of user tourism interest information from one online review are depicted in Figure 3. In this illustration, the value 4 in the first layer signifies that the user is satisfied with the tourism product. The first value of 5 in the second layer indicates that the user is very satisfied with the food in Xi’an, while the subsequent value of 5 in the last layer denotes the user’s high satisfaction with the product attribute of Chinese-style hamburgers.

4. Construction of Tourism Interest Model

The tourism interest model is built upon the foundation of tourism domain information, with user tourism interest information undergoing structured preprocessing obtained through mining online review data.

4.1. The Calculation of Users’ Final Hierarchical Interest Degree

Each user typically submits multiple online reviews on a tourism platform. To accurately capture a user’s interests, it is crucial to integrate all of their online reviews. Through analysis, it has been observed that when a user mentions a particular tourism product attribute class or feature across multiple reviews, it indicates a higher level of interest in that attribute class or feature.

Let us denote the following:

m as the total number of tourism product attribute classes or features mentioned by user A in all online reviews.

r as the total number of times attribute class or feature i of tourism products appears in all reviews by user A.

Then, user A’s initial interest degree in tourism product attribute class or feature i can be calculated as ci = r/m

$${\mathrm{C}}_{\mathrm{i}}={\displaystyle \frac{{\mathrm{c}}_{\mathrm{i}}}{{\displaystyle \sum _1^{\mathrm{n}}{\mathrm{c}}_{\mathrm{i}}}}}$$

(6)

Assuming that user A posts two online reviews, two hierarchical information mining results of online reviews are as follows, shown in Figure 4 and Figure 5.

According to Formula (6), we can obtain the final hierarchical interest degree of user A. However, considering that the same user is unlikely to travel to a place they have already visited, the interest degree of the first level, which pertains to the tourism product, is ultimately eliminated, as shown in

Table 1. The final hierarchical interest degree of user A (absolute value).

Tourism product attribute class	M			N		P			Q
Interest degree of the second level	2			1		2			1
Tourism product attribute feature	M1	M2	M3	N1	N2	P1	P2	P3	Q1	Q2

.

4.2. Hierarchical Mining of User Tourism Interest Information

Based on the tourism online review of a certain user, the hierarchical interest information of the user on a certain tourism product can be mined. However, the evaluation of each user may be biased and cannot represent the final evaluation of the tourism product. In order to represent the final evaluation of a tourism product, the hierarchical interest information of all users should be integrated. If the interest evaluation of the ith online review (n online reviews in total) on the jth tourism product attribute class or feature is pi, then the final evaluation of the jth tourism product attribute class or feature is the following:

$${\mathrm{f}}_{\mathrm{j}}={\displaystyle \frac{{\displaystyle \sum _1^{\mathrm{n}}{\mathrm{p}}_{\mathrm{i}}}}{\mathrm{n}}}$$

(7)

The final evaluation of all the tourism product attribute classes or features constitutes the hierarchical evaluation of the tourism product.

Assuming that a tourism product B has two online reviews, the two hierarchical information mining results of the online reviews are shown in Figure 6 and Figure 7 as follows:

According to Formula (7), the final hierarchical evaluation of tourism product B can be obtained.

Table 2. Final hierarchical evaluation of tourism product.

Overall Evaluation of Tourism Products
3.5
Product attribute class	M			N			P		Q
Evaluation	3.5			4.5			3		3
Product attribute feature	M1	M2	M3	N1	N2	N3	P1	P2	Q1	Q2
Evaluation	4	3	3.5	5	5	3	3	2	4	3

shows the final hierarchical evaluation of tourism product.

5. Simulation Experiment

This paper employs a dataset comprising reviews and user information collected from the Ctrip tourism platform, focusing on a sample of 300 participants, including teachers and students from our city’s university. With informed consent, we used the Python Scrapy framework to systematically gather personal and review data from Ctrip. To protect privacy, all personal identifiers are replaced with aliases. Detailed user information is provided in

Table 3. The statistics table of user personal information.

User	Name	Gender	Age	Job	The Number of Online Reviews
1	Lili	woman	40	teacher	12
…	…	…	…	…
150	Jack	man	21	student	4
…	…	…	…	…
300	Lory	man	30	teacher	10

. Additionally, we ensure the validity of the thesaurus by cross-referencing it with actual user expressions from the collected reviews. This process involves analyzing common terminology and review patterns to refine the thesaurus, ensuring that it accurately reflects user expression habits. The alignment validation is supported by a comparative analysis of the thesaurus terms against user-generated content, demonstrating the effectiveness of our thesaurus in capturing relevant user interests and preferences.

The basic analysis flow involves mining each tourism online review information using tourism domain information. Subsequently, the final hierarchical interest degree of the user and the final hierarchical evaluation of the tourism product are calculated separately, utilizing Formulas (6) and (7), respectively. Finally, based on the final hierarchical interest degree of the user and the final hierarchical evaluation of the tourism product, personalized tourism product recommendations are made to each user.

Taking 100, 200, and 300 samples as units and limiting the number of online reviews posted by users to 1, 5, and 10, nine groups of experiments are carried out.

The accuracy of the user tourism interest model (A) is verified by comparing the recommended results (R) with the user’s expectations (E).

$$A={\displaystyle \frac{R\cap E}{E}}$$

(8)

The experimental results are compared and analyzed with the user–project evaluation matrix modeling method [8]. The results are shown in Figure 8 and Figure 9.

From Figure 8 and Figure 9, it is evident that regardless of the sample size, the accuracy of both modeling methods is nuanced when limited to one online review per user. This limitation arises because when each user posts only one online review, it is challenging to accurately gauge their interests. User expectations (E) were determined by aggregating the explicit preferences and implicit feedback from users’ online reviews. The accuracy of the model (A) was calculated by comparing the predicted user interests against these aggregated expectations using metrics such as precision, recall, and F1-score.

When the sample size is 100, and the number of online reviews increases from 1 to 5, the accuracy of recommendations does not show a significant improvement compared to when the sample size is 200 or 300. This discrepancy arises because, while the increased number of online reviews improves the accuracy of user interest information, the small number of users results in less realistic and one-sided evaluation information of tourism products. However, the method proposed in this paper demonstrates higher accuracy than the method based solely on the user–project evaluation matrix. This superiority can be attributed to the hierarchical user interest information based on domain information, which expands the scope of evaluation.

Additionally, when the sample size is 300 and the number of online reviews posted by each user increases from 5 to 10, the accuracy of the method proposed in this paper shows significant improvement. The accuracy was calculated using a holdout validation method, where a portion of the data was used for training and another portion for testing, ensuring robust performance metrics. However, the accuracy of the method based on the user–project evaluation matrix improves only marginally. This difference is because as users post more online reviews, their interest degree based on the hierarchical user interest information becomes more accurate. This trend is observed across other sample sizes as well.

6. Comparison with Mainstream Algorithms

To objectively validate the performance of the proposed model, we compared it with several established models. Additionally, to address the issue of sample size, we collected data from multiple platforms to gather more user and review information. We performed cross-validation and applied statistical tests to ensure that the performance improvements of our model were statistically significant.

Below is a comparison of the tourism online review text preprocessing steps proposed in this paper with mainstream algorithms, including collaborative filtering, content-based filtering, matrix factorization-based recommender systems, deep learning-based recommender systems, natural language processing (NLP), time series analysis, geospatial analysis, and social network analysis. The table outlines the advantages, disadvantages, and applicable scenarios for each method, highlighting the advantages of the method proposed in this paper. The comparison results are shown in

Table 4. Comparison with mainstream algorithms.

Algorithm/Method	Advantages	Disadvantages	Applicable Scenarios	Advantages of Proposed Method
Collaborative Filtering	Easy to implement; no need for content information	Cold start problem; data sparsity; lack of explainability	Large-scale data recommendation	Improved adaptability and accuracy through multi-dimensional feature extraction and dimensionality reduction
Content-Based Filtering	Addresses cold start problem; suitable for new users and items	Requires extensive content description; cannot recommend items not yet encountered	Content-rich recommendation systems	Handles network terms and specific combination words using a two-dimensional sliding window and attribute extension
Matrix Factorization-Based Recommender Systems	Uncovers latent features; handles sparsity issues	High computational complexity; long training time	Recommendations for high-dimensional data	Effectively handles high-dimensional sparse data through frequency and information gain-based dimensionality reduction
Deep Learning-Based Recommender Systems	Handles complex nonlinear relationships; high precision	Requires large amounts of data and computing resources; complex training	Complex and dynamic recommendation scenarios	Performs well in efficiency and practical applications without requiring extensive computational resources
Natural Language Processing (NLP)	Extracts user sentiments and interests; fine-grained analysis	Complex processing of unstructured data	Text analysis and sentiment analysis	Enhances text understanding with part-of-speech tagging and domain-specific information
Time Series Analysis	Captures temporal trends and cyclic patterns	Not suitable for data without time dependency	Dynamic predictions	Outperforms in static recommendations
Geospatial Analysis	Analyzes geographic locations; identifies popular destinations	Requires geographic data	Geography-related recommendations	Suitable for location descriptions in online reviews
Social Network Analysis	Analyzes social relationships and influence	Requires social network data	Social network recommendations	Enhances recommendation accuracy by incorporating social evaluations

.

Unique Advantages of the Proposed Method:

• Multi-Dimensional Feature Extraction and Dimensionality Reduction: Using a two-dimensional sliding window and attribute extension, the proposed method effectively handles network terms and specific combination words, leading to more accurate feature extraction.
• Efficient Dimensionality Reduction: By leveraging word frequency and information gain, the proposed method significantly reduces the dimensionality of high-dimensional sparse data, improving model efficiency and performance.
• Domain-Specific Part-of-Speech Tagging: The use of part-of-speech tagging combined with domain-specific information enhances the accuracy of text understanding and analysis.
• Addressing Cold Start Problem: Through extensive attribute extension and feature extraction, the proposed method performs well in handling new users and items, overcoming the cold start problem associated with collaborative filtering.
• High Efficiency in Practical Applications: The proposed method does not require extensive computational resources, making it suitable for efficient processing of large-scale datasets.

The proposed tourism online review text preprocessing method excels in handling network terms, specific combination words, reducing the dimensionality of high-dimensional sparse data, and incorporating domain-specific information for part-of-speech tagging. These capabilities address the limitations of many mainstream algorithms, offering broad applicability and high practicality.

7. Impact of Climate Change on Tourist Preferences

Due to the significant impact of the COVID-19 pandemic on the tourism industry, our ability to collect tourism data from after 2019 has been greatly affected. The pandemic has caused substantial disruptions in travel patterns, leading to a dramatic decline in tourism activities worldwide. Consequently, this has created challenges in obtaining accurate and representative data for the post-pandemic period.

In light of these challenges, we must also consider other factors that can influence tourism data, such as natural and climatic conditions. For instance, climate factors can affect travel behavior, tourism demand, and the attractiveness of destinations. Extreme weather events, seasonal variations, and long-term climate changes can all play significant roles in shaping tourist preferences and activities.

Therefore, in the following sections, we will discuss how these uncontrollable factors, including climate conditions, have impacted tourism data collection and analysis. We will explore the following:

Climate change is increasingly influencing where tourists choose to travel. The rise in global temperatures has led to more extreme weather events such as hurricanes, floods, and heatwaves, which can disrupt travel plans and pose safety risks. Tourists are becoming more cautious and selective about destinations, preferring those with stable and predictable climates. This preference ensures they can enjoy their travel experiences without significant weather-related disruptions.

Moreover, tourists are increasingly aware of their environmental footprint. They favor destinations that demonstrate a commitment to sustainability and climate resilience. This includes implementing sustainable infrastructure, reducing carbon emissions, and preserving local ecosystems. Destinations that align with these values not only attract environmentally conscious travelers but also enhance their long-term reputation and attractiveness.

Tourists are increasingly seeking destinations that demonstrate climate resilience and sustainable practices. According to the World Tourism Organization (UNWTO) data from 2023, destinations that are certified with eco-labels or demonstrate active sustainability efforts have seen a 20% increase in visitor satisfaction and repeat visits.

The adoption of sustainable tourism practices is not only a response to environmental concerns but also a key factor in attracting eco-conscious travelers. Data from a global survey conducted by Sustainable Tourism Insights in 2023 revealed that 85% of respondents prioritize staying at accommodations with eco-friendly initiatives such as renewable energy use and water conservation practices, as shown in

Table 5. Consumer preferences for sustainable accommodations.

Year	Survey Question	Response (%)
2023	Importance of Eco-Friendly Accommodations	85%
2023	Preference for Renewable Energy Use	78%
2023	Support for Local Community Initiatives	72%

.

The adoption of sustainable practices in tourism is crucial for shaping consumer interests. Travelers today prioritize destinations that prioritize sustainability and environmental stewardship. This includes initiatives like eco-friendly accommodations, use of renewable energy, waste reduction programs, and community engagement in conservation efforts. Destinations that embrace sustainable tourism not only appeal to eco-conscious travelers but also build resilience against climate change impacts. By preserving natural resources and minimizing environmental degradation, these destinations ensure sustainable tourism growth while benefiting local economies.

Based on the above data and discussions, we can conclude that sustainable practices of tourist sites and their adaptation to climate change have a significant impact on tourist preferences. Combining the above model allows for more accurate conclusions to be drawn. Similarly, utilizing this model to derive tourist preferences also benefits local planning efforts aimed at achieving environmental conservation and sustainable energy development.

8. Relationship between Tourism Interests and Public Health

The COVID-19 pandemic has heightened awareness of public health considerations in tourism. Travelers now prioritize destinations that prioritize safety, hygiene standards, and health protocols. A study by Health Tourism Trends in 2023 found that 67% of tourists consider destination hygiene as a critical factor when making travel decisions. Beyond safety measures, tourists are increasingly seeking destinations that offer opportunities for recreational activities promoting physical and mental well-being. Activities such as hiking, wildlife observation, and nature-based therapies have seen a surge in popularity as they provide avenues for stress reduction, physical fitness, and immersion in natural environments.

Ecotourism has emerged as a preferred choice for many travelers seeking meaningful and responsible travel experiences post-pandemic. This form of tourism encourages visits to natural areas while supporting conservation efforts and benefiting local communities. The pandemic has highlighted the importance of preserving natural environments and biodiversity, making ecotourism a compelling option for travelers looking to reconnect with nature in a sustainable manner. The benefits of ecotourism extend beyond personal health to encompass broader economic and social impacts. Destinations that embrace ecotourism practices not only protect natural resources but also generate income for local communities through guided tours, artisanal crafts, and eco-friendly accommodations. For instance, initiatives promoting sustainable agriculture and community-based tourism have flourished in regions committed to environmental stewardship and cultural preservation.

Therefore, leveraging the predictive power of the above model, integrating sustainable tourism practices with robust public health measures will be crucial for rebuilding and sustaining the tourism industry post-pandemic. Collaborative efforts among stakeholders, including governments, businesses, and local communities, will play a key role in enhancing destination resilience and ensuring traveler confidence. Initiatives such as eco-certifications, carbon-neutral initiatives, and health-focused tourism campaigns will be pivotal in shaping a more sustainable and resilient tourism landscape globally.

Combining the insights above, we have found that the current model still lacks adequate adaptability when faced with special conditions and extreme climate events. However, since climate conditions can be predicted using scientific methods, we plan to enhance our model through interdisciplinary collaboration in future research. Specifically, we will integrate advanced climate prediction models by leveraging state-of-the-art forecasting techniques from meteorology and environmental sciences to improve the model’s ability to handle extreme weather and unusual conditions. This will involve incorporating detailed climate models and predictive data to better understand how climate variations may affect tourism behaviors and trends. We will also engage in interdisciplinary collaboration by working closely with experts from various fields, including meteorologists, environmental scientists, and tourism analysts, to integrate climate prediction insights into our tourism models. This collaborative approach will help us develop a more robust and adaptive model that accounts for a broader range of environmental factors. Additionally, we will enhance data integration and analysis by combining climate prediction data with tourism behavior data to refine analytical methods and improve the model’s responsiveness to climate impacts. This will involve integrating these diverse data sources to capture the complex interactions between climate conditions and tourism dynamics. Finally, we will conduct comprehensive validation by performing extensive testing of the model under different climate scenarios and special conditions to ensure its accuracy and reliability. This approach aims to provide more precise and actionable insights for tourism planning and management, ultimately leading to more effective strategies in responding to climate-related challenges.

9. Conclusions

This paper introduces domain information technology into the research on mining online review information, proposing a user tourism interest model based on tourism domain information. It constructs tourism domain information and develops a hierarchical mining model for online review information based on this domain information. Finally, it presents a user tourism interest method utilizing tourism domain information and emphasizes the calculation process of the final hierarchical interest degree of the user and the final hierarchical evaluation of tourism products. Simulation experiments demonstrate that the user tourism interest model based on tourism domain information achieves superior recommendation accuracy in the tourism recommendation system across various experimental environments. The integration of domain information technology enhances the relevance of recommendations, while the hierarchical mining model captures nuanced user preferences. The proposed user tourism interest model considers individual preferences and broader domain-specific factors, resulting in more personalized recommendations. Emphasis on the calculation process ensures transparency and reproducibility, while validation through simulation experiments confirms the model’s effectiveness and robustness. Overall, the extensions presented in this paper contribute to advancing the theoretical understanding and practical implementation of tourism recommendation systems. By focusing on these interconnected themes, tourism stakeholders can better adapt to evolving consumer needs in a changing global landscape. Emphasizing sustainability, resilience to climate change, and public health enhances the attractiveness of destinations while ensuring the long-term viability of tourism industries worldwide.