Segmenting tourists’ motivations via online reviews: An exploration of the service strategies for enhancing tourist satisfaction

Abstract

Tourism motivation and satisfaction are classic themes in tourism research. This study combines latent Dirichlet allocation (LDA) and the Censydiam motivation model to analyze online reviews of tourism in Qinghai, China. The aim of this research is to explore tourist motivation through online reviews and provide innovative service suggestions to improve tourist satisfaction. The LDA model initially extracts six main topics from online comments. Then, using the fuzzy analytic hierarchy process (FAHP), it maps the relationship between topics and tourism motivations to propose strategies for enhancing tourists' enjoyment, conviviality, and other motivating factors. Furthermore, we employ the Kano model to evaluate tourists' satisfaction levels regarding these strategies, demonstrating their positive evaluations. Hence, this study provides tourism industry professionals and service designers with an innovative method for understanding tourists' motivations through online reviews, enabling them to design specific services that enhance tourism experiences.

1. Introduction

The motivation to travel is a crucial aspect of research within the field of tourism and serves as a fundamental component of travel behavior. This topic has been extensively researched in the context of tourism marketing strategy [1]. Previous research has demonstrated that travel motivation precedes the experience but that satisfaction follows this experience [2]. By examining tourists' experiences, researchers can assess the degree to which tourists' motivations are fulfilled and thus guide the formulation of better tourism strategies. Cohen [3], for example, has proposed that different kinds of people may desire distinct tourist experiences. Hence, it is crucial to categorize tourists’ motivations and adopt appropriate positioning strategies for destinations, enabling them to differentiate themselves from similar destinations [4].

Analyzing tourists' motivations from a cognitive perspective has become a prevailing trend in contemporary tourism research. The aim is to understand the authentic demands of tourists. For instance, Hadinejad et al. [5] have thoroughly analyzed 162 journal articles published over 42 years and emphasized the importance of integrating contemporary approaches, including cognitive response theory [6], the heuristic-systematic model [7], and the self-validation hypothesis [8], to develop a more profound understanding of tourists' attitudes. On the other hand, with the development of artificial intelligence technology, the large amount of user-generated content (UGC) in tourism review data has gradually become an essential way for understanding the authentic experiences of tourists. Accordingly, applying theories related to cognitive science when analyzing UGC in tourism allows a more comprehensive understanding of tourists’ authentic travel experiences, facilitating the formulation of more reasonable tourism strategies.

According to Ritchie et al. [9], destinations offering unique environments that allow visitors to engage in lifestyles distinct from their usual habits have a significant competitive advantage. Furthermore, when these lifestyles are enhanced by historical settings unfamiliar to visitors, a destination can sharpen its competitive edge, and create enduring memories. Qinghai, for example, is a province in northwestern China, a renowned tourist destination with a primary emphasis on nature tourism. This multiethnic region features abundant tourism resources based in culture, history, and traditional customs. However, due to the lack of development of cultural tourism resources and insufficient exploration of its tourists' motivations, hampering the sustainable development of Qinghai's tourism industry. Hence, it is necessary to develop tourism service strategies that align with this region's characteristics based on diversified and genuine tourist motivations. Accordingly, we have selected Qinghai as the research subject in this study. From the cognitive perspective, this research seeks to explore the potential motivations of this region's tourists, to discern their needs and formulate tourism strategies, thereby enabling the enhancement of tourist satisfaction in Qinghai.

The rest of this research is organized as follows: Section 2 provides a literature review of the theoretical and methods utilized in this research. Studies concerning tourism motivation and experience, text mining and processing techniques, the Censydiam motivation model, and the Kano model for measuring satisfaction are reviewed. Section 3 then introduces the comprehensive research process and methodology, which include the acquisition of online travel reviews and subsequent data processing. Specifically, the Censydiam model is employed in qualitative analysis of the data processing results, while the fuzzy analytic hierarchy process (FAHP) is used in quantitative analysis. Based on these analyses, we propose innovative elements for tourism services and then measure the level of satisfaction of these proposed elements. In Section 4, the research results are analyzed and discussed to reveal the differences in the perceptions of various tourist groups by considering the characteristics of the sample group. Some suggestions for improvement are proposed, and the service strategies are optimized. Next, in Section 5, the limitations of this research are discussed, and some corresponding suggestions and potential research directions are elaborated. Finally, this study concludes with a summary of its contributions.

2. Literature review

2.1. Research on tourism motivation, behavior, and experience

Amid increasingly competitive conditions in tourism, effective tourism marketing is impossible without an understanding of tourists' motivations. A detailed analysis of tourists' motivations enables researchers to understand, predict and influence the relationship between tourists' motivations and behaviors [10]. Diverse research perspectives on the potential motivations and demands associated with the field of tourism have been proposed. For example, Cazorla-Artiles et al. [11] have estimated the potential travel demand among individual pairs of origins and destinations based on travel type and seasonality. The focal market was found to be segmented, whereby the target market was defined according to these needs. Su et al. [12] investigated the influence of the motivations underlying tourists' disclosures of tourism goals on social media and the corresponding impacts on Chinese tourists' goal-directed behaviors (GDB). Based on self-determination theory [13] and the broaden-and-build theory [14] of positive emotion, the researchers found that when tourists receive positive feedback after sharing their travel plans on social networks, they become more inclined to implement their plans, ultimately enhancing their GDB.

Sentiment analysis of travel experience has typically been critical in tourism service research [15]. Numerous scholars have thus employed diverse cognitive theories and associated methodologies to investigate emotional experiences in the context of tourism. For instance, Chen [16] used integrated travel motivation and expectation confirmation theory to study the factors that influence memorable experiences in coffee tourism in Vietnam. Rasoolimanesh [17,18] discussed the effects of a “memorable tourism experience” (MTE) of cultural heritage tourism on the behavioral desires of tourists. Alhothali [19] addressed the relationship between religious services and experiences associated with spiritual experience travel, showing that such service experiences positively influence tourists’ willingness to revisit. Kim [20] focused on the influence of tourists' negative sentiments on their future tourism and studied their emotional responses to the attributes of six tourist destinations. Examining the factors that influence tourists' revisit intentions and recommendations, Woyo [21] found that the hospitality and friendliness of tourist destination residents are the most important predictors of revisit. Zhang [22], Culic [23], and Ramesh [24] used structural equation modeling (SEM) to study tourist satisfaction with and loyalty to tourist destinations regarding different attraction factors pertaining to various tourist destinations. Finally, Bui [25] developed a structural model, evaluating the four dimensions of popularity, emotion, time, and location to calculate tourist destination image (TDI).

Hence, the research on tourism motivation and emotional experience has been widespread, primarily conducting qualitative and quantitative analyzes of customer satisfaction; however, few studies have mined the motivations and experiences of customers using the big data in online tourism reviews. Compared to traditional questionnaire or interview research methods, online reviews can rapidly generate a wealth of data related to customers' genuine experiences. Moreover, the unstructured nature of online customer review data places respondents outside the traditional, predefined structure of a survey-based approach and enables them to provide rich assessments of the product preferences [26]. Therefore, the intelligent acquisition and efficient utilization of online reviews have become a crucial research trend.

2.2. Research based on online review mining

Many scholars have researched online reviews of services or products [[27], [28], [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], [39], [40]]. The existing literature has suggested that the research on online reviews can be categorized into two streams. The first involves the development of a mining algorithm for more accurate sentiment calculation of online reviews. Wang [41], for instance, used a bidirectional long short-term memory (BLSTM) model for sentiment classification to obtain dependencies in microblog conversations and automatically detect blog post sentiment. Wei [42] proposed a BLSTM model based on multiple orthogonal attention to identify the implicit feelings in reviews. In contrast to the traditional single-attention model, the fuzzy sentiment tendency of words can thereby be identified. Sharma [43] proposed the SentiDraw method to improve sentiment classification; it calculates the probability distribution of words in online reviews with different star ratings to obtain sentiment orientation (SO). Al-Obeidat [36] developed the Sentiminder online tool to analyze the sentiment of online reviews and thus assist online retailers in identifying positive and negative emotions as well as the primary concerns expressed in customer reviews, using LDA to cluster six different sentiment themes.

Another group of scholars has primarily focused on the consumer intention and emotion hidden in online reviews and the impacts thereof on purchasing behavior. For example, Seshadri Tirunillai et al. [44] have identified multiple potential dimensions of customer satisfaction by studying online reviews of 15 companies in five markets. Zhang et al. [45] used online reviews as their source of information when exploring how to improve hotel services, analyzing the likelihood that online ratings contain different emotional categories. Their goal was thus to extract these consumers' expectations and reference values from these discrete attributes. Jia [46] employed statistical data, LDA, and frequency analysis to compare online reviews written by Chinese and American tourists, identifying disparities in tourists’ perceptions of restaurant service between these two nations. Ding [47] used LDA and supervised LDA (sLDA) to process online reviews and investigate tourist satisfaction levels. Guo et al. [48] identified 19 dimensions of hotel-customer interaction and critical dimensions that affect service quality according to 266,544 online reviews from 16 countries. Lang et al. [49] used traditional sentiment analysis to analyze online reviews, comparing the evaluation data of three fashion rental companies to identify the motivations and challenges among online fashion renters. Hsiao et al. [50] used Kansei engineering to mine online reviews of logistics services and service elements connected to perceptual words to obtain customer-oriented design elements. Wang et al. [51] used heuristic learning and Kansei engineering to extract the emotional attributes of 7 pairs of perceptual words from mass online reviews to help customers and designers understand user emotions and make appropriate decisions. Qi et al. [52] proposed an automatic filtering model that predicts the usefulness of online reviews from the perspective of product designers, analyzed comments using the Kano model, and then generated product improvement strategies. Finally, Yang et al. [53] extracted and restored user experience elements in online product reviews from multiple perspectives, providing a method that completes product design and strategic planning automatically. However, the actual amount of potential information contained in UGC has yet to be explored.

2.3. LDA topic modeling

The above literature review indicates that LDA has gained popularity in topic classification text analysis. This method employs an unsupervised three-layer Bayesian generative model, which was originally proposed by Blei et al. [54]. It leverages prior knowledge during classification, as people use similar keywords when they express opinions on the same topic. Therefore, LDA can calculate the probability distribution of keyword relevance in each document for each topic. LDA can also address the problems caused by the high computational complexity of the SVD model [55] and its inability to update new texts. As LDA employs the Dirichlet distribution and is suitable for analyzing big data and sparse time-incoherent data [54], it is also ideal for analyzing scattered online reviews and has found many applications in medicine, informatics, and sentiment analysis. That is, it efficiently and quickly deciphers a text's central theme and a user's focus on the salient product.

LDA's calculation logic asserts that the probability of a keyword appearing in a document is determined by a joint probability distribution involving the likelihood of the word occurring within the same topic and the likelihood of the topic occurring within the same document. Utilizing online reviews as an illustration, M denotes the number of reviews to be analyzed, W represents the total number of keywords contained in them, T indicates the overall number of topics in the reviews, φ_t symbolizes the probability distribution of keywords within each topic θ_m represents the probability distribution of topics within each document, ω_m,n denotes the nth word in the mth review, and τ_m,n signifies the nth topic in the mth review (Fig. 1). Subsequently, the joint probability distribution of keywords under the corresponding topic of the review is calculated as follows:

$$\mathrm{p}\left(\theta ,\mathrm{\tau },\mathrm{\omega }|\alpha ,\beta \right)=p\left(\theta |\alpha \right)\prod _{n=1}^{N_m}p\left({\tau }_n|\theta \right)p\left({\omega }_n|{\tau }_n,\beta \right)$$

in this formula, α and β are the hyperparameters for extracting the topic and word. The extraction process is as follows:

• 1The topic probability distribution θ_m in document m is drawn from the Dirichlet distribution with α as the hyperparameter;
• 2The nth topic τ_m,n of the mth comment is obtained from θ_m;
• 3The word distribution φ_t for each topic is generated from the Dirichlet distribution of the hyperparameter β;
• 4ω_m,n is obtained from φ_t.

Fig. 1.

The generation of an LDA topic model.

Source: The author used travel reviews as an example to elucidate the topic generation via the LDA model.

2.4. The Censydiam model

The Censydiam model is theoretically grounded in the personality analyzes of Freud [56], Jung [57], and Adler [58]. Initially, for instance, Synovate utilized the Censydiam model to derive consumer sentiment measurements and insight motivations [59]. Subsequently, they proceeded to develop market segmentation. Unlike Maslow's hierarchy of needs theory [60], existence, relatedness, and growth (ERG) theory [61], Doyal and Gough's needs theory [62], attractive quality theory (the Kano model) [63], and other research models concerning user requirements, the Censydiam model focuses on people's behavioral motivations from the perspective of psychoanalysis. Callebaut et al. [64] initially developed a two-dimensional motivation framework comprising eight fundamental consumer motivations at the subconscious level. This framework aims to facilitate differentiated positioning and reveal deeper insights into consumer motivation. More than ten renowned multinational companies have therefore used this framework in their brand positioning and development. As shown in Fig. 2, the Censydiam model divides people's behavioral motivations into eight dimensions and explores the psychological characteristics underlying their behaviors.

Fig. 2.

The Censydiam model [59].

This model categorizes human behavioral motivations into two dimensions [65]: individual or social. When individuals interact with society, four motivations emerge: belonging, power, control, and enjoyment. At times, individuals strive to surpass others and seek feelings of superiority and leadership. Conversely, they feel an occasional desire to integrate with a group and blend in among others. These contrasting desires generate two motivations: power and belonging. Individuals may thus experience the inclination to follow their impulses and abandon restraint; at other times, they control these urges and perform appropriate behavior. These contrasting experiences provide the motivations for enjoyment and control, whereby the model derives four additional motivations: conviviality, recognition, vitality, and security. At times, individuals desire recognition for their achievements, skills, and knowledge. At others, they seek to share experiences and establish connections with others. These thoughts generate two motivations: recognition and conviviality. Individuals thus occasionally yearn for adventurous experiences to test their limits and explore the unknown. Conversely, they may sometimes feel the inclination to retreat, seek solitude, and feel secure. These circumstances introduce two additional motivations: vitality and security.

Moreover, Geeroms [59] have validated and critically evaluated this motivation model to confirm its cross-validity, construct validity, and practical applicability, scientifically demonstrating its effectiveness. Liu et al. [66] analyzed smartwatch consumer data using online big data to identify product attributes of interest to users. They then performed sentiment analysis using the LDA model, the KJ method of classifying users' perceived needs, and the Censydiam model to cluster users' motivations, resulting in the establishment of users' profiles. This methodology facilitates the construction of a demand forecasting model for multiple user groups. Gao et al. [67] used the Censydiam model to investigate deviations in Roewe's market positioning and proposed a future design concept for its human‒machine interface (HMI) system. Chen et al. [68] initially categorized the tourism requirements of Chinese solo millennial tourists into seven objectives based on Maslow's hierarchy of needs. These authors then integrated these objectives into three motivation categories, analyzed the potential tourist demands using the Censydiam model, and identified the factors that affect solo tourists' tourism experiences. The literature has thus confirmed that the Censydiam model can effectively map user interest tags to emotional motivations, constituting a direct method for observing and measuring consumers' purchase decisions. Furthermore, research has shown that combining user-generated online data with the Censydiam model represents a viable approach to analyzing how users' habits influence their purchasing behavior.

Specifically, the advantage of the Censydiam model is its ability to reveal individuals’ underlying motivations through behavioral analysis and to categorize demand motivation. It serves as a valuable tool for understanding consumption motivations [65]. The above research has therefore supported the effectiveness of the Censydiam model in user research, particularly in user profiling, product market positioning, and the formulation of brand development strategies.

The use of the Censydiam model to study travel reviews remains, however, somewhat rare. The motivations behind tourists' travel decisions are complex and influenced by various factors, including population characteristics, social factors, tourism destinations, and social changes. Hence, it is useful to analyze travel motivations using the Censydiam model. However, given that it is a general psychoanalytic model, it is necessary to incorporate the context of tourism to explain travel behavior and thereby increase its effectiveness.

2.5. Theory of attractive quality and the Kano methodology

In the field of enterprise management, enhancing customer loyalty requires customer satisfaction improvement. Therefore, the factors that impact customer satisfaction have become an important research focus [69]. In 1984, Noriaki Kano proposed the Kano model [63], which is based on two-factor theory (hygiene-motivational factors). This theory thus discusses the nonlinear relationship between customer demand and satisfaction, balances supply and demand with users’ needs, determines the priority of demand satisfaction, and objectively and comprehensively evaluates customer satisfaction. Therefore, attractive quality theory divides user demand satisfaction into five class features: must-be (M), one-dimensional (O), attractive (A), indifferent (I), and reverse (R). Below, Fig. 3 illustrates the classical Kano model.

Fig. 3.

The Kano model [63].

The Kano model evaluates each requirement in a “two-way” manner [70]. Respondents must assess their degree of satisfaction when the service element is provided (positive) or not provided (negative). Each “one-way” question contains a set of options scored on a 5-point Likert scale that uses answers indicating “positive/negative” satisfaction and a “5 × 5” table to determine the satisfaction of each requirement [71].

The must-be requirements attribute is the most basic level of demand. It represents an essential service or product feature that exhibits a “nonlinear” relationship with user satisfaction. Satisfaction will not increase when this service is provided, but it will decrease when it is not provided. This one-dimensional feature reflects the “linear” relationship between users’ needs and satisfaction levels. Providing this service can improve satisfaction to a certain extent; if it does not, satisfaction will decrease. These requirements can only offer products or services that meet the anticipated needs of users. Attractive conditions also exhibit a “nonlinear” relationship, representing a high-quality feature that exceeds customer expectations, aligning with the desires of service designers and managers, but it must be achievable. Indifference requirements refer to whether a customer is satisfied but do not impact satisfaction. Reverse requirements are the opposite of must-be requirements. If provided with these requirements, the focal product or service causes customer disgust, and joy decreases “linearly.” According to research needs, the commonly used dimension of “importance” [72], regression analysis [73], and other methods have been typically used to evaluate and subdivide these requirements. In service design, designers want to offer more one-dimensional and attractive features to improve the service experience [74], enhance satisfaction, and increase brand stickiness to ensure sustainable service delivery.

2.6. The research framework

The final research framework, visualized in Fig. 4, was constructed based on the above analysis and research concerning the aforementioned theories and methods. This study thus examined various forms of tourism motivation via quantitative and qualitative research methods, as follows: Online reviews were analyzed using statistical methods, while the theoretical model was based on Censydiam motivation model. First, we used LDA topic models to extract topics from large datasets of tourism reviews of destinations. Next, industry experts utilized the FAHP to map the Censydiam model to the topics and gain insights into the underlying motivations of the relevant tourists. Finally, based on these results, we designed tourism service products suited to local, characteristic tourism resources and then evaluated them with the Kano model to propose some suggestions for improvement.

Fig. 4.

Research framework.

3. Method

All participants were informed of the content of this study, and their consent to participate was obtained. The research methods were approved by the Bioethics Committee of the East China University of Science and Technology (certificate no. ECUST-2022-063).

3.1. Using LDA to extract topics

(1) Data preprocessing and result visualization. We obtained 77,946 items from 6 of China's tourism product sales platforms, including Ctrip, Tongcheng, and Lvmama, using Octoparse. After data deduplication, stop words, and word segmentation, 63,513 records remained. Generally, a larger number of text documents is more conducive to LDA topic extraction [75]. Once generated, these extracted topics have uncertainty, and perplexity and consistency indicators are typically used to determine the optimal results. Perplexity is used to evaluate the uncertainty of the number of topics: the lower the value is, the smaller the uncertainty and the better the final clustering result [54]. However, reducing perplexity will overfit the model, which is not conducive to topic extraction. Reducing perplexity must be combined with consistency. Thus, in this study, by calculating and comparing the model's perplexity and consistency, six topics were extracted with a perplexity of −8.3309 and a consistency of 0.644, more reasonable values. Below, Fig. 5 shows the perplexity level of the different topics generated, while Fig. 6 is a visualization of the final topic classification, developed using LDA.

Fig. 5.

Perplexity levels corresponding to the different numbers of topics, calculated using the LDA topic model.

Source: Author's calculations.

Fig. 6.

Visualized results of the topics in online tourism reviews of Qinghai.

Source: Author's calculations.

(2) Filter keywords. For each topic, we initially extracted 30 keywords. To accurately represent the themes and facilitate subsequent analysis, we invited nine experts (five individuals with over eight years of travel and tour guiding experience and four design experts) to filter these keywords. Below, Table 1 lists the keywords and the number of documents in each topic. Topic 1 contains the highest number of reviews, indicating high attention among tourists toward drivers and driving techniques. Topics 2 and 3 have a similar number of ratings. These keywords indicate that the former is about cultural heritage and the latter about the natural landscape. Topic 4 comprises travel information suggestions. Topic 5 consists of experiences related to travel accommodations and hotels. Finally, Topic 6 conveys subjective evaluations of travel experiences with fewer occurrences, indicating their less prominent thematic relevance.

Table 1.

Number of keywords and reviews in each topic.

Topic	Keywords	p			Total
		(0.000 < p ≦ 0.004）	(0.004 < p ≦ 0.007）	(0.007 < p ≦ 1.000）
1	Driver/Itinerary/Special/Thanks/Scenery/Arrangement/Happy/Takephotos/Great/Northwest/Truly/Driving/Skills/Care for/Recommended/Next time/Partner/Warm/Cheerful	1326	12221	10581	24128
2	The Mogao Grottoes/Dunhuang/Danxia/Qinghai Lake/Saka Salt Lake/Kumbum Monastery/Prairie/Desert/History/Feeling/The Northwest/Culture/Lakes/Loop trip/Shock/Nature	914	7300	3493	11707
3	Scenery/Place/Snow Mountain/Cole flower/Sky/Lake/Truly/Worth/Weather/Beautiful/Feeling/Taking Pictures/Special	878	6872	3516	11266
4	Scenic spot/Carpooling/Time/Good/Tickets/Advice/Scenery/Accommodation/Feeling/Worth/Taking pictures/Truly/Weather/Train/In advance/Tourists/Choose/Evening	787	4448	3585	8820
5	Trip/Customer service/Satisfied/Driver/Recommended/Accommodation/Choose/Interpretation/Tour guide/Travel agency/Service/Patience/Professional/Enthusiasm/Next time/Cost performance/Clean	264	3129	3932	7325
6	Save money/Quiet/Unhindered/Think of/Long history/So happy/Summer/Attractive scenic/Traveling companion/Honest/After-sales/Feast for the eyes/Looking forward to/Along the line/Reluctant to separate	86	162	19	267

*p: The association probability of a review with this topic.

3.2. Mapping topics to motivation

(1) Censydiam-tourism motivation model. To better map the topics and the motivations in the Censydiam model, it was necessary to explain the various motivations in the focal context of tourism behavior. Censydiam is a general behavioral motivation model that can be applied in different fields, including tourism. As a general model of motivations, the Censydiam model requires a specific analysis based on the specific behavioral context when solving practical problems [65,76]. This study thus referred to the multiple, practical cases of Censydiam model application by Ipsos, e.g., “Chinese Views On Holidays Abroad,” published by the China Team in March 2017, which analyzes the various motivations of Chinese tourists traveling abroad via a qualitative Censydiam model [77]. Following such analyzes, in this study, nine experts explained the motivations for travel based on the results of focus groups. These results are presented in Table 2.

Table 2.

Censydiam-tourism motivation analysis.

Censydiam motivation	Psychological needs	Tourism motivation
Enjoyment	Self-release	Thoroughly enjoy travel and experience the fun and meaning of travel.
Control	Repressed desires	The spiritual shock that the travel process can bring inspires reflection.
Power	Self-actualization	The sense of superiority brought by travel.
Belonging	Into the collective	Build friendships with tour guides and teammates.
Recognition	Social identity	Sharing travel experience can gain recognition from others.
Conviviality	Communicative relationship	There are opportunities to meet Aboriginal people in an unfamiliar travel environment, meet new friends, etc.
Vitality	Satisfy curiosity	Everywhere in an unfamiliar environment is full of discoveries that can satisfy curiosity.
Security	Survival Safety	People will naturally exhibit increased fear in an unfamiliar environment, and their safety needs will increase accordingly.

(2) Fuzzy analytic hierarchy process evaluation. The FAHP has been widely used in assessments of multicriteria decision-making in fuzzy environments. Here, to quantitatively analyze the relationship between motivation and subject, we used the FAHP to establish a better relationship between the topics and Censydiam tourism motives. First, we constructed a fuzzy hierarchy model for topic and motivation mapping, as shown in Fig. 7.

Fig. 7.

FAHP model of Censydiam: Tourism motivation.

We invited 14 tourism and user research experts to conduct the FAHP. The topic with the highest weight for each motivation was selected, as shown in Table 3. These results are as follows: topic 1- security, topic 2- control, topic 3- vitality, topic 4- recognition, topic 5 -belonging, and topic 6- power.

Table 3.

Results of the FAHP regarding group decisions.

	topic1	topic2	topic3	topic4	topic5	topic6	WI	motivation	CI
topic1	0.5	0.3814	0.3105	0.4943	0.4838	0.3428	0.1342	Enjoyment	0.0131
topic2	0.6186	0.5	0.4876	0.62	0.6545	0.4767	0.1905
topic3	0.6895	0.5124	0.5	0.6511	0.6789	0.552	0.2056a
topic4	0.5057	0.38	0.3489	0.5	0.5584	0.4019	0.1463
topic5	0.5162	0.3455	0.3211	0.4416	0.5	0.3738	0.1332
topic6	0.6572	0.5233	0.448	0.5981	0.6262	0.5	0.1902
topic1	0.5	0.2708	0.35	0.5058	0.5074	0.3723	0.1338	Control	0.0168
topic2	0.7292	0.5	0.6427	0.7494	0.6715	0.6698	0.2308a
topic3	0.65	0.3574	0.5	0.6004	0.5793	0.5782	0.1843
topic4	0.4942	0.2506	0.3997	0.5	0.4512	0.4222	0.1345
topic5	0.4926	0.3285	0.4207	0.5488	0.5	0.4501	0.1494
topic6	0.6277	0.3302	0.4218	0.5778	0.5499	0.5	0.1672
topic1	0.5	0.3786	0.4071	0.4929	0.4929	0.2786	0.1367	Power	0.0212
topic2	0.6214	0.5	0.4786	0.5929	0.5643	0.3071	0.171
topic3	0.5929	0.5214	0.5	0.5857	0.5714	0.3286	0.1733
topic4	0.5071	0.4071	0.4143	0.5	0.4714	0.3571	0.1438
topic5	0.5071	0.4357	0.4286	0.5286	0.5	0.3643	0.151
topic6	0.7214	0.6929	0.6714	0.6429	0.6357	0.5	0.2243a
topic1	0.5	0.5421	0.5567	0.5716	0.3137	0.4863	0.1647	Belonging	0.016
topic2	0.4579	0.5	0.4853	0.4288	0.2713	0.4786	0.1415
topic3	0.4433	0.5147	0.5	0.4216	0.2861	0.4782	0.1429
topic4	0.4284	0.5712	0.5784	0.5	0.3286	0.4784	0.159
topic5	0.6863	0.7287	0.7139	0.6714	0.5	0.6713	0.2314a
topic6	0.5137	0.5214	0.5218	0.5216	0.3287	0.5	0.1605
topic1	0.5	0.5684	0.5263	0.3704	0.4156	0.5012	0.1588	Recognition	0.0166
topic2	0.4316	0.5	0.4533	0.293	0.3231	0.4465	0.1298
topic3	0.4737	0.5467	0.5	0.2857	0.36	0.4971	0.1442
topic4	0.6296	0.707	0.7143	0.5	0.6615	0.7193	0.2288a
topic5	0.5844	0.6769	0.64	0.3385	0.5	0.6073	0.1898
topic6	0.4988	0.5535	0.5029	0.2807	0.3927	0.5	0.1486
topic1	0.5	0.5714	0.5714	0.5357	0.4143	0.5643	0.1771	Conviviality	0.0123
topic2	0.4286	0.5	0.4786	0.4857	0.3357	0.4643	0.1462
topic3	0.4286	0.5214	0.5	0.4571	0.3	0.5	0.1471
topic4	0.4643	0.5143	0.5429	0.5	0.3357	0.5143	0.1581
topic5	0.5857	0.6643	0.7	0.6643	0.5	0.6214	0.2157a
topic6	0.4357	0.5357	0.5	0.4857	0.3786	0.5	0.1557
topic1	0.5	0.35	0.2857	0.4929	0.5429	0.4143	0.139	Vitality	0.0128
topic2	0.65	0.5	0.4286	0.6214	0.65	0.5643	0.1943
topic3	0.7143	0.5714	0.5	0.7643	0.7786	0.6643	0.2329a
topic4	0.5071	0.3786	0.2357	0.5	0.4929	0.4357	0.1367
topic5	0.4571	0.35	0.2214	0.5071	0.5	0.4143	0.13
topic6	0.5857	0.4357	0.3357	0.5643	0.5857	0.5	0.1671
topic1	0.5	0.7386	0.6842	0.6728	0.6123	0.7084	0.2278a	Security	0.0133
topic2	0.2614	0.5	0.4557	0.4137	0.3295	0.5051	0.131
topic3	0.3158	0.5443	0.5	0.4447	0.3423	0.5087	0.1437
topic4	0.3272	0.5863	0.5553	0.5	0.4719	0.5413	0.1655
topic5	0.3877	0.6705	0.6577	0.5281	0.5	0.6313	0.1917
topic6	0.2916	0.4949	0.4913	0.4587	0.3687	0.5	0.1403

^aWI: Weight index; CI: Consistency index.

(3)Topic analysis: Censydiam tourism motivation. The keywords mentioned in topic 1 are “driver,” “driving,” and “technology.” The analysis of the highly relevant topic reviews showed that extensive comments affirm the driver's skills, indicating that driving safety during travel is a primary issue for tourists. Therefore, topic 1 corresponds to the “security” motivation and is satisfied.

Topic 2 contains mentions of many cultural sites and emphases on “shock” and “sensation.” It thus expresses how cultural landscapes produce a spiritual shock in tourists that satisfies their desire for emotional introspection through travel, which is related to the “control” motivation.

Topic 3 discusses natural landscapes, highlighting the “snow mountains” of the Qinghai-Tibetan Plateau, the act of “taking pictures,” and the notion of being “special.” These observations therefore reveal tourists’ desires to explore and experience unique and extraordinary phenomena. People often capture photographs to document their encounters with novel scenery and to generate a sense of significance that aligns with their motivation for “vitality.”

The words “ticket,” “suggestion,” “advance,” “choice,” and other keywords appear in Topic 4, reflecting the experiences of tourists with food, accommodation, and transportation. By sharing this information, tourists aim to receive validation from others. Hence, this topic demonstrates the fulfillment of tourists' motivations for “recognition.”; Keywords such as “driver,” “tour guide,” “travel agency,” “service,” “patience,” and “enthusiasm” are mentioned in Topic 5. This topic signifies tourists' appreciation of a service provider and their sense of belonging. Therefore, these keywords align with the motivation of “belonging.”

The motivation underlying Topic 6, however, remains unclear. Nevertheless, the presence of keywords such as “smooth,” “thinking,” “pleasant scenery,” “feast for the eyes,” “look forward to,” “reluctant to separate,” and others signifies the subjective feelings of tourists. These expressions capture tourists' overall evaluations and can be perceived as individual experiences. Accordingly, Topic 6 is defined as the motivation of “power.”

3.3. Tourism service design

Evaluating the mapping results for tourism in Qinghai, we discovered that the region offers tourists excellent travel experiences and satisfies motivations such as “control,” “vitality,” and “recognition.” However, there we observed less discussion of “enjoyment,” “power,” and “conviviality,” indicating lower levels of satisfaction. Hence, there are opportunities to design additional tourism service elements. Moreover, these tourists have higher security needs due to the high-altitude environment and the ongoing pandemic. In this study, the tourism service design experts thus proposed specific recommendations for Qinghai tourism services, conducting qualitative analyzes by focus groups to address the unfulfilled motivations of tourists.

Below, Table 4 lists service elements S1–S6, specifically crafted to fulfill tourists' “enjoyment” motivation needs. S7–S11 are designed to cater to their “conviviality” motivation needs. The service elements S12–S13 aim to enhance the tourism experiences associated with the motivation for “power.” Finally, the safety needs of tourists amid plateau conditions are accounted for by service elements S14–S18.

• (1)Service design for enjoyment motivation. The service elements S1–S6 aim to enhance travel experiences in Qinghai by establishing an emotional connection between tourists and their destinations, thereby increasing their loyalty. One reason for the few mentions of the enjoyment motivation by tourists is that Qinghai tourism primarily focuses on natural landscapes and lacks entertainment projects that cater to the experiential need for emotional release. From the perspective of human emotional expression, the enjoyment motivation refers to people's desires to express their private emotions freely and without restrictions. Tourists' descriptions of events that evoke their feelings provide insights into their implicit emotional experiences. However, capturing every tourist experience through simple keywords is not easy. Consequently, dividing the emotions expressed in online reviews using only LDA is a flawed method.
• (2)Service design for conviviality motivation. Service elements S7–S11 can provide tourists with increased opportunities for social engagement with various individuals, including team members, indigenous peoples, and fellow tourists. Tourists inevitably encounter new individuals throughout their journeys. As a result, interactive activities can be carefully crafted at potential points of contact among, e.g., individuals, events, and objects, to encourage organic social interactions among tourists and foster positive social relationships. These activities help fulfill tourists' desires for conviviality and reinforce their sense of belonging.
• (3)Service design for power motivation. The S12 and S13 service elements address the power motivation. Tourists frequently document their travel experiences and share them with others, which can enhance their sense of superiority. Therefore, offering tourists a more convenient and innovative way to document their experiences can help fulfill their power motivation. Taking photos and sending texts are the most prevalent and straightforward means of describing and sharing travel experiences. However, postprocessing and filtering become more challenging as the number of photo posts increases. The complex text input when using mobile devices also impacts people's willingness to share. Consequently, these two service elements offer tourists faster and simpler sharing paradigms, increasing their desires for self-expression and timely sharing.

Table 4.

Service design elements of Qinghai tourism.

Number	Design element	Motivation
S1	Using local wool, tourists make wool-felt souvenirs independently under the guidance of local herders.	Enjoyment
S2	Hand-plant trees locally and give tourists regular feedback on tree growth.
S3	There are many stray cats and dogs in rural areas, and tourists can adopt them and receive photos of their growth.
S4	Tourists can personally feed and adopt yaks and receive dairy products from their yaks.
S5	Protecting the ecological environment of the plateau and picking up outdoor garbage for recycling in exchange for rewards.
S6	Purify the mind and experience a spiritual journey on the plateau.
S7	Team leaders can organize icebreaking activities, enabling tourists to integrate into the team naturally.	Conviviality
S8	Let tourists experience the real life of herders and their accommodations and work together with herders.
S9	Team members can understand their zodiac signs and establish communication quickly.
S10	Scenic spots provide visual information on the number of visitors, origin, gender, ethnicity, etc., which can help tourists know how many like-minded tourists are visiting simultaneously.
S11	After the trip, tourists receive a travel booklet from the travel team.
S12	The social media app can provide tourists with quick and straightforward templates and automatically generate a travel diary to record beautiful moments.	Power
S13	The social media app can record travel stories in audio recordings to avoid typing.
S14	Hotels for travel have visible disinfection information in public areas.	Security
S15	The restaurant servers provide tourists with their daily health code information at the travel dining location.
S16	Restaurants provide the nucleic acid test certificate for cold chain food so tourists can eat safely.
S17	The hotel provides a free disinfection service for tourists' luggage.
S18	The travel team provides oxygen saturation monitoring and anti-altitude sickness oxygen equipment and drugs.

In addition, considering the global COVID-19 pandemic and potential for hypoxia in high-altitude Qinghai, we have proposed S14–S18 service elements to address the safety requirements of tourists throughout their travels in the region.

3.4. Kano model evaluation of tourist service satisfaction

We adopted Kano questionnaire assessed tourists' satisfaction with a proposed service design element. The questionnaire consisted of two parts. The first part acquired the basic information of respondents including their age, gender, occupation, educational background, and travel mode. The second part was the Kano questionnaire, a 5-point evaluation of the positive and negative questions concerning the 18 service elements. These questionnaires were distributed electronically through an online survey platform from November 2021 to January 2022. The investigation resulted in a random collection of 158 questionnaire, with 150 of them being valid. This study also employed SPSSAU software [78] to assess the questionnaire's reliability and validity. Below, Table 5 shows the results of the demographic survey. Table 6 presents the results of the reliability and validity tests conducted on the Kano questionnaires. All the survey data passed the reliability and validity tests, enabling subsequent data analysis.

Table 5.

Survey results regarding population attributes.

Categories	Options	Frequency	Percentage (%)	Cumulative rate (%)
Age	born after the 00s	34	22.67	22.67
	born after the 90s	90	60	82.67
	born after the 80s	22	14.67	97.33
	born after the 70s	4	2.67	100
Education level	High school or below	27	18	18
	Undergraduate	113	75.33	93.33
	Graduate or above	10	6.67	100
Gender	Male	60	40	40
	Female	90	60	100
Monthly income	RMB 5000 or below	51	34	34
	RMB 5000-10000	79	52.67	86.67
	RMB more than 10000	20	13.33	100
Profession	Student	32	21.33	21.33
	Professional (teacher, doctor, etc.)	22	14.67	36
	Service industry	50	33.33	69.33
	Freelancer (painter, writer, etc.)	11	7.33	76.67
	Civil servant	9	6	82.67
	Other	26	17.33	100
Travel frequency	Every three months	27	18	18
	Once a year	33	22	40
	Every half-year	34	22.67	62.67
	Uncertain	56	37.33	100
Preferred way of traveling	Group tour (more than ten people)	12	8	8
	Private group tour(up to five people)	49	32.67	40.67
	Self-driving tour	89	59.33	100

Table 6.

Reliability and validity tests of the questionnaires.

	Cronbach's α	KMO	Barlett test
Kano questionnaire	0.776	0.825	0.000
Positive question	0.874	0.845	0.000
Inverse question	0.899	0.871	0.000

*KMO: Kaiser‒Meyer‒Olkin.

Table 7 presents the evaluation results of each design element based on the analysis of the Kano questionnaires. S1, S2, S4, S11, S17, and S18 are therefore attractive demands. The satisfaction influence index is used to segment satisfaction, i.e., the terms “better” (indicating satisfaction influence) and “worse” (indicating dissatisfaction influence) are utilized to determine user sensitivity to service level. This formula is as follows:

$$\text{Better}=\left(\mathrm{A}+\mathrm{O}\right)/\left(\mathrm{A}+\mathrm{O}+\mathrm{M}+\mathrm{I}\right)$$ (1)

$$\text{Worse}=-\left(\mathrm{O}+\mathrm{M}\right)/\left(\mathrm{A}+\mathrm{O}+\mathrm{M}+\mathrm{I}\right)$$ (2)

Table 7.

Kano model analysis.

Design elements	A	O	M	I	R	Q	CR	Better	Worse
s1	73	19	2	55	0	1	A	61.74 %	−14.09 %
s2	62	28	3	54	0	3	A	61.22 %	−21.09 %
s3	52	19	3	74	0	2	I	47.97 %	−14.86 %
s4	61	26	3	59	0	1	A	58.39 %	−19.46 %
s5	53	14	2	79	1	1	I	45.27 %	−10.81 %
s6	58	23	0	68	0	1	I	54.36 %	−15.44 %
s7	42	12	4	85	5	2	I	37.76 %	−11.19 %
s8	44	13	4	87	0	2	I	38.51 %	−11.49 %
s9	27	5	0	115	1	2	I	21.77 %	−3.40 %
s10	43	6	2	96	1	2	I	33.33 %	−5.44 %
s11	69	28	7	45	0	1	A	65.10 %	−23.49 %
s12	60	20	3	66	0	1	I	53.69 %	−15.44 %
s13	50	14	2	83	0	1	I	42.95 %	−10.74 %
s14	57	14	9	68	0	2	I	47.97 %	−15.54 %
s15	39	18	1	89	0	3	I	38.78 %	−12.93 %
s16	54	23	6	66	0	1	I	51.68 %	−19.46 %
s17	69	20	4	54	1	2	A	60.54 %	−16.33 %
s18	70	24	6	48	0	2	A	63.51 %	−20.27 %

*A: Attractive; O: One-dimensional; M: Must-be; I: Indifference; R: Reverse; Q: Questionable; CR: Classification results.

The index obtained by formula ① ranges between 0 and 1, with higher values indicating greater sensitivity and priority. The index obtained by formula ② ranges between −1 and 0, with smaller values indicating higher sensitivity and priority. Below, Fig. 8 displays a graph of the better-worse coefficients, illustrating the coordinates of each service design element. Priority should be given to elements in the first and second quadrants. The third quadrant represents attributes of indifference that do not require the provision, while the elements in the fourth quadrant are essential attributes that must be satisfied. Specifically, S1, S2, S4, S6, S11, S12, S14, S16, S17, and S18 exhibit higher satisfaction influence indexes, indicating that they should be prioritized.

Fig. 8.

Quadrant diagram of better-worse coefficients.

Source: Author's calculations.

In summary, the research methodology of this research followed three steps. First, we employed the LDA model to extract the salient topics from the online reviews. Second, according to the Censydiam model, the FAHP was used to map the relevant motivations based on these topics. Finally, by utilizing the results of this correlation, tourism service designers were able to address any unmet motivations and evaluate the efficacy of these proposed service strategies, leading to further optimization.

4. Results

This study employed LDA topic modeling to analyze a dataset consisting of 63,513 tourism reviews in Qinghai and identified six primary topics. These topics were subsequently linked to various motivations using the Censydiam model, as presented in Table 12. Specifically, Topic 1 relates to security, Topic 2 to control, Topic 3 to vitality, Topic 4 to recognition, Topic 5 to belonging, and Topic 6 to power.

Table 12.

Results of the final mapping of topics and tourism motivations.

Topic	Keywords	Explanation	Motivation
1	Driver/Itinerary/Special/Thanks/Scenery/Arrangement/Happy/Takephotos/Great/Northwest/Truly/Driving/Skills/Care for/Recommended/Next time/Partner/Warm/Cheerful	The competence of the drivers is highly rated, which means that tourists can experience a safe journey.	Security
2	The Mogao Grottoes/Dunhuang/Danxia/Qinghai Lake/Saka salt lake/Kumbum Monastery/Prairie/Desert/History/Feeling/The Northwest/Culture/Lakes/Loop trip/Shock/Nature	The humanistic landscape elicits a profound sense of awe and satisfies the inner experiences of tourists.	Control
3	Scenery/Place/Snow Mountain/Cole flower/Sky/Lake/Truly/Worth/Weather/Beautiful/Feeling/Taking/Pictures/Special	Tourists can witness unavailable frontier scenery within the city and fulfill their innate desire for exploration and discovery.	Vitality
4	Scenic spot/Carpooling/Time/Good/Tickets/Advice/Scenery/Accommodation/Feeling/Worth/Taking pictures/Truly/Weather/Train/In advance/Tourists/Choose/Evening	Tourists share valuable travel information, which assists others and engenders a sense of validation and recognition.	Recognition
5	Trip/Customer service/Satisfied/Driver/Recommended/Accommodation/Choose/Interpretation/Tour guide/Travel agency/Service/Patience/Professional/Enthusiasm/Next time/Cost performance/Clean	Tourists derive satisfaction and a sense of belonging from the attentive and considerate services that drivers, guides, and others offer.	Belonging
6	Save money/Quiet/Unhindered/Think of/Long history/So happy/Summer/Attractive scenic/Traveling companion/Honest/After-sales/Feast for the eyes/Looking forward to/Along the line/Reluctant to separate.	The emotive expressions shared by tourists concerning their travel experience reflect the sense of superiority they derive from the process of tourism.	Power

However, we were unable to map the motivations of enjoyment and conviviality according to the above topics. This result therefore emphasizes that the optimization of tourism services is required to fulfill these two motivations. Hence, S1-6 were designed to address the enjoyment motivation, while services S7-11 were developed to target the conviviality motivation. S12-13 were designed to address power motivations, while S14-18 were designed to address security motivations. Moreover, the Kano model assessed the satisfaction level of each service element and identified ten highly satisfying service items: S1, S2, S4, S6, S11, S12, S14, S16, S17, and S18.

While the Kano model yielded satisfactory outcomes, we observed decreased satisfaction levels related to the service elements of conviviality. Hence, we conducted a correlation analysis between all design elements and demographic characteristics. The analysis revealed a negative correlation between income and the adoption of stray animals (S3) and engagement in spiritual journeys (S6). Furthermore, icebreaking activities (S7) were found to be associated with education level and the mode of travel, whereas experiencing residents' lifestyles (S8) exhibited a higher correlation with age.

Both S3 and S6 are service elements associated with the motivation of enjoyment. However, contrary to the results of satisfaction evaluation, as Table 8 reveals, individuals with low incomes demonstrate a high level of interest in these two activities. This finding suggests that income disparities can result in diverse emotional needs. Consequently, it is vital to understand the needs of tourists from different income groups to provide appropriate emotional-release tourism services.

Table 8.

Monthly income and S3/S6 factor analysis of variance.

	Monthly income (mean ± SD)			F·	p·
	5000 and below (n = 51)	5000-10000(n = 79)	More than 10000 (n = 20)
S3 positive	4.51 ± 0.70	4.27 ± 0.78	3.80 ± 0.89	6.15	0.003**
S6 positive	4.63 ± 0.63	4.35 ± 0.70	4.20 ± 0.89	3.51	0.032*

*p < 0.05; **p < 0.01.

The S7 element pertains to individuals' attitudes toward icebreaking activities while traveling. Table 9 thus illustrates a gradual decrease in demand for this element that occurs as respondents' educational level increases. Accordingly, further discussion is needed on interpersonal communication among individuals with diverse academic backgrounds. Additionally, Table 10 indicates that individuals engaged in group tours strongly desire team icebreaking activities. However, individuals who travel in small groups or embark on self-driving tours are often accompanied by acquaintances, making these activities less relevant to them.

Table 9.

Educational level and S7 factor analysis of variance.

	Educational level (mean ± SD)			F	p
	High school and below (n = 27)	Undergraduate (n = 113)	Graduate and above (n = 10)
S7 positive	4.37 ± 0.69	4.05 ± 0.95	3.40 ± 0.70	4.319	0.015*

*p < 0.05; **p < 0.01.

Table 10.

Travel mode and S7 factor analysis of variance.

	Travel mode (mean ± SD)			F	p
	Participate in a group tour (more than ten people) (n = 12)	Private small group (within five people) (n = 49)	Self-driving tour (n = 89)
S7 positive	4.75 ± 0.45	3.98 ± 0.92	4.02 ± 0.93	3.792	0.025*

*p < 0.05; **p < 0.01.

The S8 focuses on individuals' attitudes toward experiencing residents' lifestyles. Table 11 therefore reveals the relationship between the S8 and age. That is, individuals become less receptive to this experience as their age increases, while young people exhibit higher receptiveness. Further consideration should therefore be given to the conviviality motivation exhibited by individuals of diverse ages.

Table 11.

Age and S8 factor analysis of variance.

	Age (mean ± SD)				F	p
	After00(n = 34)	After90 (n = 90)	After80(n = 22)	After70(n = 4)
S8 positive	4.50 ± 0.75	4.02 ± 0.85	4.18 ± 0.91	3.25 ± 0.96	4.205	0.007**

*p < 0.05; **p < 0.01.

These findings thus underscore the significance of offering differentiated tourism experiences to distinct demographic groups in the provision of these service categories. The results of satisfaction evaluation also indicate that integrating qualitative and quantitative analysis of online reviews with the Censydiam motivation model can lead to service strategy designs that result in higher satisfaction levels.

5. Suggestions and discussion

5.1. Suggestions for optimizing the LDA model

LDA is an unsupervised classification model that employs word co-occurrence for topic mining in textual data. However, the existing rules for extracting topics in the LDA model may neglect crucial information in specific domains, resulting in overly broad issues. Hence, semi-supervised learning techniques can reduce noise and enhance the accuracy of topic extraction [79]. Formulating classification rules in text mining that align with diverse professional requirements is, then, imperative to attain enhanced precision in information extraction. Future research should therefore emphasize improvements to the LDA model to more effectively mine tourism motivations by integrating contextual information from comments, sentiment words, service features, and language patterns to construct a domain-specific dictionary. Nevertheless, they were not effectively extracted by the unmodified LDA model, resulting in the omission of essential information.

The LDA model employed in this study also inadequately captures specific keywords that reflect tourism motivation, impacting its overall thematic effectiveness. For example, some online reviews stated the following: “The famous Ta'er Monastery, the pious pilgrims, the solemn Buddhist temple, the place where Tibetan masters practice Dharma and chant sutras”; “The pious eyes of the grandmother at the Ta'er Monastery appeared before my eyes”; “… such as Chaka Salt Lake, Qinghai Lake, Qilian Mountain, Zhuoer Mountain, Ta'er Monastery, etc., what I deeply remember in every place is not how unique their shapes are, but how pious the people inside are … Although I am not a religious person, for those who are devout believers, I respect it from the bottom of my heart.” In these comments, the keyword “pious” plays a significant role in the tourists' control motivations, which current LDA models cannot extract. This challenge pertaining to extraction can be overcome by creating a domain-specific dictionary that incorporates emotional vocabulary labels relevant to the motivations proposed in the Censydiam model for automatic classification. Furthermore, this approach reflects a current research trend in the field of intelligent sentiment classification.

5.2. Discussion of the applicability of the method

Our combined use of LDA topic extraction and Censydiam motivation analysis is well suited to analyzing reviews of service products. Products are generally classified as either physical or abstract. When users evaluate physical products, they thus typically use simple terms such as “good,” “not easy to use,” or “bad review” to express their feelings. Due to the limited number of users willing to provide detailed product reviews, the LDA model often overlooks these few beneficial comments. Conversely, regarding abstract products such as services, individuals commonly employ more emotionally charged language to thoroughly describe their service experiences.

Through our analysis of the online reviews, we have found that when people evaluate a service, they provide detailed descriptions of their positive experiences and even more detailed reports of their negative experiences, rendering the latter more valuable for LDA extraction. For instance, when tourists are satisfied with the services a driver or tour guide provides, they tend to offer detailed positive feedback as follows: “The driver is perfect, familiar with the route, and the driving is safe. It is excellent! We felt the Northwest's beautiful scenery and the people's enthusiasm and civilization”; “Thanks to the two tour guides, Kaimin Yang and Tong Xu, for their warm service. It made us feel the enthusiasm and boldness of the people from the Northwest. It also involved no shopping during the whole process, and no products were sold in the car.” Here, the words “safe,” “enthusiasm,” and “civilization” represent practical information extracted regarding the topic.

In contrast, when tourists have unsatisfactory experiences, they provide statements similar to the following: “Although there are two trains, the travel route in the Northwest is still very tiring, causing us to leave early and return late … Difficulties occurred the entire way. Taking the national road and highway took at least 1 h longer by car!! That is the only source of dissatisfaction!”; “Chaka Salt Lake is the most unpleasant attraction I have ever been to, and it is wholly commercialized. Tickets for rides, sightseeing cars, miniature trains, shuttle buses, etc., must be paid separately. There are numerous hidden costs. Although they are all in the same lobby area, they claim they are not affiliated. Negative reviews!” The keywords such as “tiring,” “difficulties,” and “commercialized” in these comments thus represent critical, negative evaluations of travel experiences. These reviews indicate that people's comments on a service evidently contain more detailed and comprehensive information about negative experiences, the LDA model can be applied to extract valuable insights from these texts.

6. Contributions

6.1. Methodological contributions

This study has focused on Qinghai tourism, conducting a comprehensive analysis of UGC online travel reviews using the LDA model and Censydiam model. This approach has identified unmet travel demands, whereby a further exploration of potential travel motivations was conducted to propose improved strategies for tourism services. We can thus recommend various tourism products or services that enhance the overall enjoyment of travel experiences, e.g., offering tourists the opportunity to participate in the production of hands-on felt wool souvenirs, providing them with dairy products sourced from personally fed yaks, or creating opportunities for physical and mental relaxation. These experiences entertain and facilitate a deeper connection with one's destination. Tour groups can also enhance the travel experience by providing travel journals to their members, fostering social interaction and strengthening the bond between tourists and tourism providers. Additionally, ensuring the safety of tourists is paramount; this can be achieved by providing medications and facilities that alleviate altitude sickness. Indeed, the above results demonstrate the feasibility and practicality of this method, thereby constituting valuable references for service design across various domains.

6.2. Theoretical contributions

This study has validated, extended, and applied the Censydiam model theory of tourism motivation. The above evaluation of satisfaction has affirmed the model's effectiveness and enhanced its theoretical significance. Moreover, this study employed artificial intelligence methods to extract potential motivations from UGC on online platforms. Given the substantial and continuing growth of online reviews, using intelligent algorithms for automated analysis of their motivational factors has become an inevitable research trend. Hence, this study constitutes a valuable theoretical reference.

6.3. Practical contributions

We have proposed a comprehensive service design method, built upon tourists' motivations, for enhancing their satisfaction with travel experiences. In this study, we have analyzed the unmet tourist needs in Qinghai's tourism. According to the above results, we suggest that decision-makers prioritize improving insufficient tourism services by focusing on enjoyment and conviviality. By integrating local and distinctive tourist resources and diversifying the forms of available tourism, differentiated tourism services based on tourists' diverse motivations can be developed, improving their travel experiences. Hence, this study is significant; it can be used to resolve the tourism development issues specific to Qinghai and represents a valuable decision-making reference for other tourism destinations.

Data availability statement

The data supporting this study's findings are available from the corresponding author upon reasonable request.

CRediT authorship contribution statement

Xin Sun: Writing – review & editing, Writing – original draft, Visualization, Validation, Software, Investigation, Data curation, Conceptualization. Zhengyu Wang: Writing – review & editing, Investigation, Formal analysis, Data curation, Conceptualization. Meiyu Zhou: Writing – review & editing, Supervision, Resources, Funding acquisition, Conceptualization. Tianxiong Wang: Writing – review & editing. Hongying Li: Data curation.

Declaration of competing interest

We declare that we have no financial and personal relationships with other people or organizations that can inappropriately influence our work, there is no professional or other personal interest of any nature or kind in any product, service and/or company that could be construed as influencing the position presented in, or the review of, the manuscript entitled.