Analysis on factors affecting tourist involvement in coffee tourism after the COVID-19 pandemic in Thailand

Related works

As stated in Ref. 6, the coffee tourism system in every tourist spot must be continually improved, and measures should be taken to share information about coffee growth in Northern Thailand’s tourism sustainability.

A comprehensive literature review technique for tea and coffee tourism has been proposed in Ref. 7, which aims to identify the opportunities and obstacles of expanding such niche products for locations. This study’s methodology uses an English-language literature review of the previous studies on tea and coffee tourism. 31 papers on coffee tourism and 33 papers regarding tea tourism have been examined in total. Many studies have used a single stakeholder’s viewpoint or the viewpoint of a particular coffee/tea tourism setting, like buyers’ or manufacturers’, according to the findings of this investigation. Because this research is the first of its kind, it showed potential suggestions for future work in coffee and tea tourism.

Community-based tourism (CBT) is a growing topic of study, and this study⁸ provides a framework to understand it grounded in experimentally verified socio-psychological, phenomenological, and cybernetic conceptions of behaviour. One of the most significant outcomes of this study is a more structured knowledge of how socio-psychological variables might support and enable a successful CBT business plan to achieve a desirable “sustainable” goal.

Analyzing the characteristics that influence the success of coffee agro-tourism and also designing an efficient mechanism for community-based coffee tourism have been the goals of this research. A total of 142 people from the Sukaratu District agrotourism sector took part in.⁹ Coffee agrotourism throughout the Galunggung tourism region was shown to be sustained by the coffee plantation, according to the study findings.

Using data gleaned from a number of researches on the Thai coffee market, the author of this thesis makes recommendations for resolving the market’s current issues. According to the limited capacities of Ref. 10, inadequate data and analysis of information, the facts and actual conditions could vary, therefore more complete inquiries and analyses of the current system and problems are required, so as to put up more effective remedies and proposals. Although the Thai coffee industry has been thoroughly examined, there are still issues that need to be addressed and solutions found.

Depending upon that extended TPB framework,¹¹ intends to analyze the factors. Using the TPB structure as a starting point, this study provides a preliminary examination of the elements that may influence Thai consumers’ purchase intentions for certified coffee. According to the findings, self-identification has a greater impact on attitude than social identity, whereas attitude has the greatest impact on consumer buying behavior consequently.

Based on an extensive theoretical map that shows how an unfunded community member had carefully conceptualized a system of actions and endeavors that have a significant impact on society,¹² provides a practical example of a conceptually rich strategic vision. It is feasible and increased by attempts to expand common values with buyers, communities, and other relevant parties to establish sustainability-oriented principles.

Another study¹³ found that customer happiness and consumer loyalty are directly linked to service quality. As a result, cost fairness and consumer loyalty are influenced positively by a company’s ability to keep its customers happy. Additionally, client satisfaction serves as a bridge connecting service quality and satisfaction. In addition, perceived price can operate as a bridge between a company’s ability to satisfy its customers and maintain their loyalty.

This research investigated the drivers of Norwegian senior tourists’ desire to visit Thailand and analyzed a path model of variables impacting their travel intention throughout the pre-visit stage. This research hypothesized that travel motivation, expectation, limitations, destination image, and electronics word of mouth (e-WOM) impact travel intention. The authors also explored participants’ travel sentiment about Thailand. Analysis was done on 500 samples of eligible responders. Using Linear structural relations (Lisrel) software version of 8.72, a structural equation model (SEM) was run to evaluate the factors affecting tourism. The impact investigation was carried out to see how study constructs had an impact. About 62% of such variation in tourist perception is explained by its predictors.¹⁴

It was the focus of Ref. 15 to examine the possibility and possibilities of coffee tourism throughout Ethiopia’s south-western highlands. Here, we are looking at how sustainable tourism may support rural community livelihoods while also helping to protect biodiversity, particularly the wild Arabica coffee gene pool, in the montane forest. According to the findings, a fragile livelihood strategy and a desire to increase revenue are to blame for expanding agriculture into the forest and destroying the unique wild Arabic coffee gene pool.

There has been a rise in the popularity of agritourism in developing nations, which has been under-researched by academics.¹⁶ They are helping to further the conversation about the advantages and disadvantages of coffee tourism by conducting a study based on the perspectives of many stakeholders. According to the study, farmers reap the most benefits from empowering and cooperating, diversifying their businesses, and creating a more sustainable environment.

As per Ref. 17, coffee tourism can potentially improve locals’ lives. This is because there are new markets to enter and benefits to be gained. Local communities and the environment benefit from strong government responsibilities, which should be supported.

In Ref. 18, there are several ways in which innovation and coffee tourism are linked in Gangneung, South Korea. By using coffee tourism as an example of food and drink tourism, they examined how previous research has linked meals and drinks with creativity before explaining how tiny towns like Gangneung, South Korea, have been able to thrive as coffee tourism operators and how their use of creative thinking contributed to its success.

The Cibulao landscape/environment was positively impacted by tourism.¹⁹ Strong interactions between actors, resources, and the local governing structure led to sound economic and social performance. Enhancing the economy’s performance is essential as a vital component of sustainable use of resources led by good local governance regulation. According to the study’s findings, Community-based Coffee Tourism (CbCT) is further expanded as a green tourism technique to help the community’s economy while also improving the environment.

The purpose of Ref. 20 is to discover the most important factors influencing people’s desire to visit sustainable coffee and tea estates. In addition, the study examined the impact of travelers’ fears about contracting COVID-19 on their decision to visit coffee and tea tourism hotspots. A survey of 302 eco-conscious Gen Y and Z customers was conducted online using the idea of planned behavior as a foundation. The data were analyzed using partial least squares. Buyers’ opinions regarding healthy coffee/tea tourism have been influenced by their desire to learn and unwind. Regarding sustainable coffee/tea tourism, risk and mindset are the most critical factors. Consumers’ attitudes and behaviors toward a rapidly expanding type of tourism, which is taking place amid historically unprecedented conditions, can now be better captured by the planned behavior by including contemporary factors.

The article²¹ outlines Lviv’s coffee tourism growth. The intended market for coffee tourism has been determined. Characteristics and significant components of the resource basis for the growth and operation of coffee tourism in Lviv are described. The local community and business work together to build and promote specialized infrastructure, tourist products (goods and services), and coffee tourism events. It aims to establish a pleasant urban hospitality area for both locals and visitors and to exhibit Lviv’s 200-year-old multiethnic coffee culture and distinctive coffee urban environment to the globe. The article includes representative statistics, data on Lviv’s coffee tourist offer, and survey findings.²²

In the coffee business, a reengineering effort focuses on identifying current innovations, evaluating existing ones, and designing new ones that can be implemented in a contingency. Changes, production, and transmission of information necessitate structural reengineering. Depending on the number of assets and external pressures, this is a new bet that will be made to rebuild the coffee industry; new jobs must be created, including those of a leader, communicator, and strategist.

Researchers²³ are interested in learning more about how to boost revenue during the COVID-19outbreak era in Central Aceh Regency by increasing coffee tourism, both physically and non-physically, using existing local expertise as a basis. Researchers used a descriptive qualitative research design and gathered responses from 40 coffee shop owners and patrons in various tourist destinations. Results revealed that the Central Aceh Regency offers significant potential for developing a coffee tourism model depending on the local tradition and enhancing income, particularly for those actively engaged in ecotourism development and coffee growers.

The issues faced by Thailand’s coffee farmers are explored in this research. Weather variations, rising labor rates, and an increase in plant diseases have contributed to several failures throughout the decades. Comparing the state of coffee tourist experiences from both the supply and demand sides was the primary focus of the investigation. Tourists are unpredictable. A data-organized questionnaire was circulated to the respondents to analyze coffee farmers’ concerns. Our proposed model can handle numerical and category data and manage issues with several outputs. As more data points are utilized for training the tree, it increases the forecast data exponentially while comparing with other existing approaches.

Contributions of this paper

Ethics and consent

The Human Ethics Committee of the Faculty of Tourism and Faculty Management Program at University of Phayao gave its permission to the research protocol approval number 2.2/005/64; the date of approval was 03/23/2021. All participants also gave oral consent to take part in the study and received information about the research’s methodology. This method of consent was approved by the Human Ethics Committee.

Data collection and processing

6,485,791 Thai and international visitors who traveled to the provinces of Chiang Rai, Chiang Mai, Mae Hong Son, and Lampang made up the population for this research. This study’s sample, chosen using accidental sampling and convenience sampling, consists of Thai and foreign tourists on vacation in coffee tourism destinations in Chiang Rai, Chiang Mai, Mae Hong Son, and Lampang provinces who participated in coffee-related tourism activities such as a coffee farm tour or coffee tasting. Yamane (1973) method was used to compute the sample size at a 95% confidence level with a 0.05 margin of error.^24,34

Although the result was 398, it was rounded up to 400 to make data gathering easier. Additionally, a percentage is used to show how many responders there were in each region.

Visitors from Thailand were asked to complete the questionnaire in Thai, while tourists from English-speaking countries were asked to complete the questionnaire in English. The questionnaire was in pen and paper format, and participants were required to complete the entire questionnaire. 2 participants who were approached to take part declined. Table 1 shows the sample groups in each province.

Research instruments

A survey was created to learn more about the behaviors, demands, and satisfaction levels of visitors who engaged in coffee tourism in the northern region of Thailand. The survey consists of closed-ended questions about the respondent’s general characteristics, behaviors, needs, and satisfaction with the destination(s) for coffee tourism in terms of those destinations’ potential, with an emphasis on the attractions, accessibility, amenities, available package, activities, and ancillary services as defined by Buhalis’ 6As framework. The questionnaire included many options (based on a five-point rating scale) to help respondents adequately answer the closed-ended questions.

Instrument validation

The questionnaire’s validity was thoroughly investigated. Each item on the questionnaire, primarily assessed by the thesis adviser, was scrutinized by five specialists. An Index of Item Objective Congruence (IOC) was used to measure the content validity and language suitability. The acceptable range is shown by the IOC score of 0.926.³² After this, the questionnaire was modified following the expert’s advice, before being delivered to the thesis advisor for final revisions. A total of 30 visitors who were not part of the sample group completed the finished questionnaire as a test run. Cronbach’s alpha coefficient was used to gauge reliability. The output was 0.948, regarded as statistically valid.³³

Procedures

The researcher used the validated questionnaire to gather data from the samples between April 2021 and March 2022. The information was gathered in several districts throughout the provinces of Chiang Rai, Chiang Mai, Mae Hong Son, and Lampang, where coffee tourism activities were planned.

Data analysis

Validity testing was performed on the completed questionnaires. Following that, the data were encoded and examined using statistical analysis tools. This study examined the conceptual model by collecting data from a self-administered questionnaire. For the purpose of certifying its validity, the survey instrument was pretested. For the purpose of ensuring content validity, 50 national and international tourists were tested. All constructs were tested for reliability (Cronbach’s alpha greater than 0.70) and survey questions were confirmed to be reliable.

Data preprocessing

Data preprocessing using stop word removal, stemming, term weighting, and dimension reduction are included.

Stop word removal

In many natural language processing (NLP) applications, stop word removal is among the most often utilized test sets. The idea is to effectively remove words that exist in all of the corpus papers. Stop words frequently include adjectives and pronouns. Stop words are often eliminated from the text before deep learning and machine learning models are trained since stop words are plentiful and provide little to no unique information for classification or clustering. Stop words may help to reduce the size of your index and query by removing them. When it comes to performance, fewer words are usually better. Because stop words are semantically empty, they have no effect on relevance rankings for data preprocessing.

Stemming

Stemming is the process of reducing words to their base by eliminating extraneous letters, generally a suffix, and removing inflection. Porter and Snowball are two examples of stemming models. The findings may be used to find similarities and links in massive datasets. Stemming is the text preparation normalization job associated with eliminating word affixes in natural language processing (prefixes and suffixes).

Min-max normalization

The dataset was preprocessed using min-max normalization to give each element in the dataset a value ranging between 0 and 1. It establishes a data range by subtracting the greatest and lowest values of each data element. The least value of data elements is subtracted from actual data element in the numerator. The normalization of data is carried out according to equation 1.

Where ${\mathrm{x}}_{\mathrm{i}}$ refers to the original value of input dataset, ${\mathrm{x}}_{\mathrm{i}}^{\prime }$ denotes the normalized value, ${\mathrm{x}}_{\max }$ and ${\mathrm{x}}_{\min }$ are the maximum and minimum values of the data respectively.

Feature extraction using linear discriminant analysis (LDA)

Data transformation into numerical features that may be handled while keeping the information in the original data set is known as feature extraction. LDA is one of the most widely used linear algorithms for feature extraction. LDA is used to discover a lower-dimensional space that best distinguishes between data from various categories. The optimal way to separate data from various categories is to utilise a lower-dimensional space found by LDA. With this approach, the Fisher criteria, which is an objective function, is aimed towards maximisation:

Each of these is referred to as an Inter- or Intraclass scattering matrix. ${\mathrm{t}}_{\mathrm{x}}={\mathrm{m}}_{\mathrm{x}}/\mathrm{m}$ is the prior probability of a sample belonging to class x with expectation W. E may be derived by solving ${\mathrm{E}}^{\ast }$= argmaxY(E) in solution space ${\mathrm{L}}^{\mathrm{s}\times \mathrm{t}}=\left\{\mathrm{E}\in {\mathrm{Q}}^{\mathrm{s}\times \mathrm{t}},{\mathrm{E}}^{\mathrm{P}}\mathrm{E}=\mathrm{X}\right\}$. This may be achieved by solving the following extended eigenvalues reduction issue: ${\mathrm{D}}_{\mathrm{v}}\mathrm{e}=\mathrm{\lambda }{\mathrm{D}}_{\mathrm{e}}\mathrm{e}$. The label information on the samples is used directly by LDA to solve classification issues.

Feature selection using Bat algorithm

A predictive model’s input parameters are narrowed down via a process known as ‘feature selection’. BAT algorithm is used for feature selection.

Bat algorithm²⁶

Researchers from different professions have been fascinated by bats, which have a unique capacity to echolocate. Bats, especially microbats, use echolocation as a kind of sonar: they generate a loud and brief pulse of sound, wait for it to strike an object, and then wait for the echo to return to their ears. In this way, bats may calculate their distance from an item. As a result of this remarkable orienting system, bats can distinguish a barrier from a prey item even in full darkness, enabling them to hunt. Bat algorithm is a novel meta-heuristic optimization approach based on observations of bat behaviour. To mimic the behaviour of a colony of bats that uses echolocation to locate prey and food, such a system has been devised. The following are some of the idealized principles used to represent this method:

While it’s possible for the loudness to fluctuate in numerous ways, imagine that the loudness ranges from a big (positive) $B_0$ to a minimal constant $B_{\mathit{min}}$.

The Bat algorithm is shown in Algorithm 1

The movement of the digital bats is determined by updating their velocity and location using equations 5, 6, 7 for every time step u, where U is the maximum number of iterations are as follows:

Beta$\left(\beta \right)$ - random integer created between the values zero and one.

$y_j^k\left(u\right)$ - at time step u, the value of the decision parameter k for the bat.

The results of $f_j$ are utilised to regulate the speed and range of bat movement. The variable ${\widehat{y}}^k$ reflects the current global best position (solution) for the decision variable k, which is produced by comparing all of the n bats’ answers.

To increase the range of feasible options, random walks are introduced. Firstly, one of the finest solutions currently available is chosen, and every bat that accepts the criteria has its own solution generated using a random walk.

$\overline{B}\left(u\right)$ - overall bat noise level at t.

For each iteration of the algorithm, the loudness $B_j$ and the emission pulse rate $s_j$ are updated, as follows

Every iteration of the process, the $B_j$ and $s_j$ parameters are modified, as follows:

$s_j\left(0\right)$ and $B_j\left(0\right)$ are frequently chosen randomly.

In general, $B_j\left(0\right)\in \left[1,2\right]$ and $s_j\left(0\right)\in \left[0,1\right]$.

Data analysis using stochastic neuro fuzzy decision tree (SNF-DT)

Data analysis is the act of analysing, cleaning, manipulating, and modeling data in order to identify usable data, inform conclusion, and assist decision-making. Stochastic neuro fuzzy decision tree (SNF-DT) is used for data analysis.

Stochastic neuro fuzzy decision tree (SNF-DT)

Incorporating neural learning algorithms into the feedback loop of hierarchical FDT (fuzzy decision tree) is the goal of stochastic neuro fuzzy decision tree (SNF-DT). The approach considerably increases the accuracy of classification of FDT without sacrificing the comprehensibility of the FDT structure. By returning from each leaf node to the root node, the back propagation learning method may be directly applied to the SNF-DT structure. A single forward cycle of FDT induction precedes numerous rounds of back propagation to fine-tune the FDT parameters in SNF-DT (membership functions and leaf certainties). Because of this approach, the hierarchical structure of the FDT tree isn’t disrupted, and the tree parameters may be tuned efficiently while still being interpretable. Figure 3 depicts the structure of SNF-DT.

Figure 3. Structure of SNF-DT.

Two summing nodes have been added to the basic SNF-DT structure in Figure 3 in order to perform inference. To determine$z_1$, the certainty factors for class 1 (Yes) from each leaf node are added together. In the same approach, to determine$z_2$., the certainty factors corresponding to class 2 (No) are added together. To determine the strength of the $\mathit{mth}$ class discharging at the $\mathit{nth}$ leaf node in any pattern, use the formula

On the basis of the input characteristics accessible in traveling from the root node to the nth leaf node, each $p{\mathit{ath}}_n\left(n=2,\dots ,6\right)$ is created. ${\mu }_{\mathit{pat}{h}_n}^j$is the $\mathit{pat}{h}_n$ membership degree. Which may be estimated as

The degree of confidence with which $\mathit{pat}{h}_n$ can categorise class m is ${\beta }_{\mathit{nm}}\left(0\le {\beta }_{\mathit{nm}}\le 1;,=1,2\right)$.

Firepower of all leaf nodes belonging to a specific class is combined to determine the prediction confidence $z_m^j\left(m=1,2\right)$ of the $\mathit{jth}$ pattern using FDT

Predictions may be made by using the formula:

Segmentation to a unique class requires selecting the classes with the greatest degree of membership, such as classifying a given pattern to categorise $m_0$.

The class with the most accurate predictions will be chosen, $m_0=\mathit{arg}\underset{m=1,2}{max}\left\{z_1^j,z_2^j\right\}$

To fuzzify input attributes, the method selects Gaussian membership (GM) functions out of many alternatives due to its differentiable property. For $\mathit{jth}$ pattern membership degree of $\mathit{pat}{h}_n$ can be calculated by

Where the centre and standard deviation of GM of the $\mathit{kth}$ input attribute on the$\mathit{nth}$ route of $G_{\mathit{kn}}$ is represented by $D_{\mathit{kn}}$ and ${\sigma }_{\mathit{kn}}$, respectively.

The technique specifies that the FDT’s error function is a differentiable function like the mean square error F.

In SNF-DT, n is the number of training patterns, $d_m^j$ and $z_m^j$ are required classes for the jth training set.

For the error to be minimised, all variations about the parameter’s Gaussian centre locations, widths, and confidence factors must disappear. The parameter update rule is a consequence.

FDT structures with Gaussian membership functions are updated using the following update rules: centers, widths, and confidence factors.