Analyzing online public opinion on Thailand-China high-speed train and Laos-China railway mega-projects using advanced machine learning for sentiment analysis

The Belt and Road Initiative (BRI), announced by the Chinese government in 2013, is a massive infrastructure investment by China to connect Asia, Europe, and Africa. The BRI has more than 147 participating countries. Its long-term policies and investment plans present a highly interesting economic and geographic perspective. The BRI focuses on developing transportation infrastructure, through both land and sea routes, to link international markets and promote economic cooperation between various countries (Chirathivat et al. 2022). The railway linking China and ASEAN countries through Laos is a significant route in the BRI. Construction of this route began in 2016 and was completed in December 2021, covering a distance of approximately 417 km. It connects, Vientiane, the capital of Laos with southern China via the Laotian border town, Boten. This has enabled Laos to have railway infrastructure on a scale it has never had before, transforming it from a landlocked country to a land-linked country (Rowedder 2020). Additionally, it will also link up with the Thailand-China high-speed train in the future. The Thailand-China high-speed train project, a collaboration between the Thai and Chinese governments to develop a high-speed railway system linking the two regions, has been jointly agreed upon. A railway line from Bangkok to Nakhon Ratchasima will be built, covering a distance of 250 km, in Phase 1, before opening in 2026. The construction of the second phase of the railway from Nakhon Ratchasima to Nong Khai, covering a distance of 357 km, is scheduled to begin operations between 2029 and 2030. This railway is part of the BRI, aimed at extending a railway from southern China through Laos and Thailand to Malaysia (State Railway of Thailand 2022). Previously, Thailand's export trade relied mainly on sea and land transportation. This railway will serve as a strategic trade and investment connection between Thailand, Laos, and the People’s Republic of China. It is expected to increase the efficiency of product transportation, promote exports of Thai products to other countries, and foster further development of the ASEAN region. Notably, the China-Laos railway has commenced operations, while the Thailand-China railway is still under construction. Our social media analysis revealed that there is a considerable amount of interest in this topic among users, including content, news, with user opinions and criticism (Wei and Sukhotu 2021).

Social media can be regarded as a virtual community that enables users to communicate, exchange information, express positive and negative opinions, and reflect diverse perspectives of people worldwide through the internet (Hou et al. 2020). The use of social media platforms has significantly increased in recent years. According to research, there were more than 3.6 billion social media users in 2020, and this number is expected to increase to 4.41 billion by 2025 (Kemp 2023a). The global digital report by we are social, which surveys digital and social media behavior of consumers worldwide, reveals that in January 2023, the number of social media users reached 4.76 billion globally. This demonstrates a remarkably high growth rate of social media users, increasing by almost 30% since the outbreak of COVID-19. There are over 1 billion new users in the past 3 years (Kemp 2023a). The effectiveness of social media platforms for governments to connect with citizens in the digital world has been demonstrated (Yavetz and Aharony 2022; Wukich 2022). Many governments use social media platforms as tools to disseminate information on government projects, policies, measures to assist, natural disaster warnings, and pandemic situations, as well as to manage disinformation and provide trustworthy information and recommendations (Sesa et al. 2022; Yuan et al. 2023). They have become very important communication and marketing tools for global businesses, enabling entrepreneurs to explore new business opportunities, access advice from experts and a diverse range of knowledge sources, and serve as a marketing channel to promote products and services (Jiawen Chen, Linlin Liu, Social media usage and entrepreneurial investment: An information-based view, Journal of Business Research, Volume 155, Part B 2023). Conversations and interactions with target customers through social media have been found to increase customer engagement. Marketers generally believe that social media marketing is the most effective way for companies to promote their products (Eslami et al. 2022). Academics and researchers are increasingly using social media to share their research findings and disseminate scientific opinions to their followers and networks on various platforms in diverse formats and content. Not only does it enhance the visibility of academics in their fields, but it also opens up a wide range of ideas from diverse groups of people and academics in different fields, which can help improve research and development (Cao 2023).

Utilization of social media platforms for viewing video content has become increasingly prevalent among individuals globally. YouTube, in particular, ranks among the top social media platforms worldwide, with over 2 billion daily active users, making it one of the most widely used applications. Android users spend up to 23 h and 9 min per month on YouTube (Kemp 2023b; Mehta and Deshmukh 2022). YouTube was established in February 2005 and was acquired by Google in November 2006 (Yao et al. 2007). The platform, which offers not only entertainment but also educational, political, and business/marketing videos (Kang and Kim 2023), enables users to upload and share video content as well as express their opinions through comments (Mehta and Deshmukh 2022). In Thailand, there are approximately 52.25 million social media users, of which 43.90 million, or 84.2%, are YouTube users. From 2022 to 2023, the number of active users on YouTube increased to 1.1 million. Marketers are successful at advertising on YouTube Thailand with a 1.9 percent increase in ad spending at the end of 2022 (Kemp 2023a). Therefore, for the purpose of this study, we have selected and analyzed user comments from the YouTube platform to examine the attitudes and opinions of the public.

In the current study, the researchers chose to use public sentiment data about the Thailand-China high-speed train project and Laos-China railway for analysis. Both projects have significant implications for cross-border transportation and other related aspects. Thailand is currently in the process of constructing a railway to connect to Laos, which is a highly specific and important project with massive investment and wide-ranging impacts. However, these data have not been previously utilized in research to the extent that they should be, especially in research in the field of Natural Language Processing (NLP) in the Thai language related to the analysis of government projects. To achieve the research objective of comparing and finding the best model from the popular models used for sentiment classification, the current study can be used as a guideline for applying technology to analyze Thai language data related to government projects for more effectiveness in the future. Therefore, sentiment analysis was performed utilizing various model architectures to analyze and determine the most effective model for predicting outcomes.

Both machine learning (ML) and deep learning (DL) models were employed to determine the most effective model for predicting the sentiment expressed in comments related to the Chinese-Laos and Chinese-Thai high-speed railway systems, with respect to their economic, transportation, and tourism impacts. Data were collected from online activity from October 2014 to May 2022. Six models, including a Logistic Regression (LR), Naive Bayes (NB), Random Forest (RF), Bidirectional Long Short-Term Memory (Bi-LSTM), Bidirectional Encoder Representations from Transformers (BERT) and a Thai Language model Trained specifically for the BERT architecture, BERT-Base-Thai and WangchanBERTa, were compared based on their predictive performance. This approach can provide insights into various aspects of human emotions, including feelings, opinions, and perceptions, and provide useful recommendations for decision-making, strategic planning, communicating news to the public, and promoting entrepreneurship in various industries, such as logistics and tourism. Effective utilization of appropriate technology can enhance the potential and increase an organization’s competitiveness. With automated tools, the government can save time and budget in monitoring public opinion and conducting public opinion surveys. More resources and budget can be allocated for planning development strategies.

Sentiment analysis is a popular research technique. It can be used to determine the hidden trends of emotions and feelings using social network data (Tam et al. 2021). Machine learning is one of the most used tools for analyzing large amounts of data in various fields. Researchers take advantage of the learning capabilities of machines to mine sentiments and opinions (Twitter data sentiment analysis of tourism in Thailand during the COVID-19 pandemic using machine learning 2022). For instance, it has been used to analyze public confidence in COVID-19 vaccines. People extensively share and discuss their opinions about the pandemic, vaccines, and vaccination processes. Users post news and comments, both positive and negative. Negative and false information about vaccines and vaccination processes may create vaccine hesitancy and distrust of the government and pharmaceutical companies, causing people to lose interest in receiving the vaccine and reduce vaccination rates in a country (Sabab Zulfiker et al. 2022; Vishwakarma and Chugh 2023). Therefore, the government can use data analysis as a tool for planning vaccine allocation, budgeting, and problem-solving, as well as to support policy and strategic decision-making (Villavicencio et al. 2021).

The analysis of political sentiments has incorporated public opinions from social media to examine feelings and attitudes toward political parties, their policies, and the popularity of individual candidates vying for the Prime Minister position. This information is then utilized to refine campaigns and vote-seeking strategies, amend party policies, devise communication strategies with the public, disseminate political messages, and also anticipate upcoming election results (Rita et al. 2023).

Analysis of student opinions and perspectives in online social media has been utilized to evaluate their perceptions of new teaching and learning formats. The content of the curriculum, the satisfaction level of students, and their opinions about their educational institution have been considered to assess the effectiveness of the education system. For instance, whether the shift to an online teaching and learning format has affected the quality of education has been examined (Imran et al. 2022). This approach aims to provide university administrators with reliable information on student viewpoints, which can then be utilized for the development and evaluation of new teaching and learning models that are sustainable and adaptable (He et al. 2022).

Investors, economists, and researchers have attempted to find ways to beat the stock market by analyzing data on emotions and their correlation with stock prices. It has been found that confidence data obtained from a large number of individuals can be used to make informed decisions about the stock market. It can even be used to make predictions, especially given that social media serves as a channel for news that can have a direct impact on stock prices. The stock market is greatly impacted when popular individuals share their thoughts about specific stocks (Hasselgren et al. 2023). Thus, the analysis of emotional data for market prediction and correlation with stock prices can significantly benefit investors in making informed decisions on whether to buy or sell stocks (Koukaras et al. 2022).

Utilization of customer feedback and criticism in business practices has had significant economic repercussions, particularly in the transportation and tourism service sectors, following disease outbreaks. Analyzing customer satisfaction, service experiences, factors influencing satisfaction and customer emotions can help evaluate customer experiences and determine areas that require the most improvement, thereby leading to increased revenue generation (Ramos et al. 2023). Research has highlighted the importance of analyzing customer feedback data to understand the feelings about products. This provides in-depth business information that reveals customer attitudes and behaviors that reflect real consumer demands and highlight areas for product development to best meet customer needs. These data can facilitate strategic decision-making and sustainable business growth (Iqbal et al. 2022; Nandwani and Verma 2021). Sentiment analysis is applied in many areas and domains. Various ML algorithms have been used in sentiment analysis to classify polarity in different languages. However, research on Thai sentiment analysis is still limited (Pasupa et al. 2022). In particular, domains related to government projects have not been researched before. The comparison of suitable models in each domain is therefore essential.

Nine different ML algorithms have been implemented, including Support Vector Machine (SVM), Bernoulli Naïve Bayes (BNB), Ridge Regression (RR), Logistic Regression (LR), Random Forest (RF), Stochastic Gradient Decent (SGD), Passive Aggressive (PA), Decision Tree (DT), and AdaBoost (ADA). These were evaluated to find the most suitable model for creating a classification of opinions of hotel service reviews in Thailand from two online agencies, Agoda.com and Booking.com in Thai, comparing the values of recall, precision, F1 score, and accuracy (Khamphakdee and Seresangtakul 2021). Moreover, the research compared the performance of nine DL models (CNN, LSTM, Bi-LSTM, GRU, Bi-GRU, CNN-LSTM, CNN-Bi-LSTM, CNN-GRU, and CNN-Bi-GRU) with different numbers of layers. The underlying architecture for all models is similar, i.e., FastText and BERT pretrained models were used to perform sentiment polarity classification. The model demonstrated the potential of WangchanBERTa to improve accuracy. Factors like hyperparameter tuning, embedding dimensions, and the number of layers in the DL model affect the efficiency of sentiment classification. These research findings can be applied as a tool for Small and Midsized Enterprises (SME) to analyze the sentiment of the Thai language in the hotel domain. Therefore, comparison of suitable models in each domain is crucial (Khamphakdee and Seresangtakul 2023).

Recently, analyzed opinions were presented with agricultural technology (AgriTech) startups in Thailand on Facebook. This work analyzes sentiments using binary classification and categorizes posts and comments as positive and negative. A Naive Bayes classifier was used to determine the sentiments and attitudes of people and investors. The objective of this research is to reflect the awareness rate of agritech startups in Thailand as agricultural startups are one of Thailand’s 4.0 policies. They can be used to develop future agricultural startups and provide interesting points to investors and government sectors (Kewsuwun and Kajornkasirat 2022).

This section describes collection of public opinion data. The processes and methods used for sentiment analysis are presented. Classification of individual comments on YouTube in Thai was done using machine learning techniques and deep learning models to compare the accuracy of each model. The results of the two highest-precision performance models are compared for each class in a tabular and confusion matrix format, with the highest-precision model presented in a Word Cloud, as demonstrated in Fig. 1.

This research presents a methodology for sentiment analysis of Thai language in the context of the Thailand-China high-speed train project and the Laos-China Railway, utilizing algorithms based on ML and DL techniques. The study collected 10,582 comments from the YouTube Data API in the Thai language, which were classified as positive, neutral, or negative sentiments by experts. Six different algorithms, LR, NB, RF, Bi-LSTM, BERT-Base-Thai, and BERT: WangchanBERTa, were employed to classify the sentiment of the comments. The performance of each algorithm was evaluated and compared based on the accuracy of sentiment classification using various feature extraction techniques. The results indicate that the BERT: WangchanBERTa model achieved the highest accuracy, 94.57%, followed by LR with 90.27% accuracy. This indicates that DL performs better in sentiment classification than traditional ML when used in conjunction with an appropriate language model. The development of sentiment analysis models specifically for the Thai language, trained on large amounts of diverse data sources, significantly improves the accuracy of sentiment prediction. Overall, the feelings of each class can be categorized as moderate. This shows that deep learning algorithms can effectively predict the sentiment of textual data. Due to the robustness of the model, this confirms that the proposed approach can be applied to analyze a much larger scale of data for public opinion analysis.

However, this study has some limitations. While data from YouTube are useful in drawing on public opinion for analysis, some people share their opinions on other social media platforms, and there are limits to how much data can be extracted for analysis. Therefore, in future research endeavors, researchers propose incorporation of additional keywords for information retrieval and utilization of social media platforms from other frameworks to increase the diversity and quantity of available data. Even though the best model obtained from this study is quite stable with high accuracy, much effort is still required in data interpretation. If stories cannot be derived from the analysis, the process is incomplete, as strategic planners usually expect actionable results. We plan to alleviate this issue by integrating sentiment classification with topic extraction techniques to analyze, cluster, and find hidden topics by sentiment. This approach is expected to yield clearer results with greater insights. With the emphasis on delivering actionable results, the framework can be applied to other projects of the Thai government.