Transforming Education in the AI Era: A Technology–Organization–Environment Framework Inquiry into Public Discourse

Abstract

The advent of generative artificial intelligence (GAI) technologies has significantly influenced the educational landscape. However, public perceptions and the underlying emotions toward artificial intelligence-generated content (AIGC) applications in education remain complex issues. To address this issue, this study employs LDA network public opinion topic mining and SnowNLP sentiment analysis to comprehensively analyze over 40,000 comments collected from multiple social media platforms in China. Through a detailed analysis of the data, this study examines the distribution of positive and negative emotions and identifies six topics. The study further utilizes visual tools such as word clouds and heatmaps to present the research findings. The results indicate that the emotional polarity across all topics is characterized by a predominance of positive emotions over negative ones. Moreover, an analysis of the keywords across the six topics reveals that each has its own emphasis, yet there are overlaps between them. Therefore, this study, through quantitative methods, also reflects the complex interconnections among the elements within the educational ecosystem. Additionally, this study integrates the six identified topics with the Technology–Organization–Environment (TOE) framework to explore the broad impact of AIGC on education from the perspectives of technology, organization, and environment. This research provides a novel perspective on the emotional attitudes and key concerns of the Chinese public regarding the use of AIGC in education.

1. Introduction

Undertaking a historical review of AI development reveals several phases of progress, interspersed with periods of stagnation commonly referred to as “AI winters” [1]. However, AIGC, exemplified by ChatGPT (v.3,3.5,4), has reached unprecedented heights. AIGC refers to the automatic creation of content by AI systems tailored to users’ personalized needs, such as images, text, and videos [2,3,4]. Owing to its powerful capabilities, diverse application scenarios, and immense potential, society has shown increasing interest in various content-generation products developed by major technology companies [5,6]. In just a few years, AIGC has not only gained prominence within the field of computer science [7], but has also demonstrated significant potential across numerous domains, including education, industry, agriculture, tourism, transportation, marketing, and finance [8,9,10,11]. Undoubtedly, in the face of such a profound impact, AIGC serves as a double-edged sword, offering boundless opportunities while posing substantial challenges [12].

Before the release of ChatGPT, many researchers had recognized the potential of AI applications to significantly advance education, and its strategic value gradually gained widespread recognition and dissemination [13]. Some scholars have argued that AI in education serves as a driving force for economic growth, future workforce development, and global competitiveness [14,15]. From the perspective of the teachers and students, the effective integration of AI tools in teaching is intricately linked to teachers’ attitudes [16]. Moreover, research by Celik et al. highlights that teachers play multi-faceted roles in AI development, not only as participants in AI algorithm training but also as critical contributors to improving AI systems by validating the accuracy of automated assessment tools [17]. From the perspective of the school, as the demand for AI-enhanced education grows, many schools have introduced AI-supported tools such as intelligent tutoring systems, teaching robots, and learning analytics dashboards [18]. Zafari et al. reviewed the current state of AI integration in K-12 education, discussing its application across different grades and subjects, as well as the technologies and environments conducive to AI adoption in education [19]. From the perspective of families and communities, the successful implementation of AI can reduce workloads and improve efficiency in tasks such as enrollment management, assessment, and plagiarism detection [20]. Yu et al. applied the theory of overlapping spheres of influence to examine the relationships between families, schools, and communities [21]. They proposed a collaborative education model for the AI era, emphasizing that such collaboration involves six key elements: educational goals, environments, subjects, processes, resources, and evaluation. In summary, the application of AI in education is deepening across various dimensions, including teachers, students, schools, and families, showcasing immense potential to drive transformations and optimize educational models.

Following the release of ChatGPT, breakthroughs and advancements in AIGC technologies have brought about profound changes in the field of education [22]. Compared to traditional AI technologies, ChatGPT offers more flexible and intelligent interactions in teaching and learning, enhancing personalized education and fostering the development of advanced cognitive skills [23]. Hence, AIGC technologies have become widely known and utilized, with discussions appearing across public platforms [5,6]. Researchers have found that AIGC, with its robust text generation and comprehension capabilities, has revolutionized pedagogy, learning experiences and personalized education [24]. Thus, AIGC creates unprecedented opportunities for the digital transformation of education [25]. However, AIGC also poses significant risks to education [26]. Concerns have emerged over issues such as ethics, copyright, bias, plagiarism, inaccurate content, improper citations, and cybersecurity [27]. Consequently, the continued improvement of AIGC may determine whether it serves as a positive force for innovation or becomes a cautionary tale of unchecked technological advancement [24].

As a large-scale language model-based AI tool, ChatGPT has transcended the limitations of traditional AI applications. Moreover, given the relatively short time since the emergence of the new AI wave, discussions on AIGC in education should not rely solely on academic analysis but should also consider public perspectives. Thus, researchers must approach the opportunities and challenges brought by AIGC from new perspectives and methodologies. Currently, social media platforms have become vital channels for information acquisition and opinion expression. Analyzing the extensive commentary and bullet screen text data generated on these platforms can provide a broader and more diverse understanding of public concerns and opinions formed through personal experiences or observations. Based on this, the following questions can be posed:

(1)

What are the public’s emotional attitudes toward the impact of AIGC on education? What are the focal points of their concerns?

(2)

How does the influence of the new AI wave differ from that of previous AI technologies? In which specific areas are these differences most evident?

(3)

How can the new opportunities and challenges posed by AIGC be viewed in the context of the Technology–Organization–Environment (TOE) framework? What responses should be taken?

In light of the above questions, this study aims to explore the attitudes of the Chinese public toward the application of AIGC in education through sentiment analysis. Additionally, the study integrates the TOE framework to provide a more comprehensive analysis of the impact of AIGC on the education sector. This study selects relevant textual comments from popular social media platforms, utilizing Python 3.12 software for data mining. Methods such as sentiment analysis, public opinion topic analysis, and content coding are employed, with research findings presented through visualization tools including word clouds and heatmaps. This study aims to analyze, from multiple dimensions, the emotional attitudes, focal concerns, and perspectives of the Chinese public regarding the application of AIGC in education. By doing so, it seeks to foster positive interaction between the public and AIGC, thereby promoting the deep integration and sustainable development of AIGC in education and teaching.

The paper is organized as follows. In Section 2, we introduce the related work of TOE theory, topic modeling, and sentiment analysis. In Section 3, we introduce the research process of this study. In Section 4, we give the main research results. In Section 5, we give the discussion. In Section 6, we give the implications. In Section 7, we give the limitations. In Section 8, we give the conclusion.

2. Related Work

2.1. Research Related to Technology–Organization–Environment (TOE) Theory

The TOE framework, proposed by Tornatzky, Fleischer, and Chakrabarti in 1990, is designed to analyze the key determinants influencing the adoption and development of innovations [28]. This framework examines innovation adoption from three dimensions: technology, organization, and environment [29], and provides the systematic supervision of each component [30]. As a widely recognized organizational theory, TOE posits that the adoption of new technologies is influenced by the combined effects of technological, organizational, and environmental factors. Technological factors primarily encompass the inherent characteristics of the technology itself and its compatibility with the organization [31]. Organizational factors refer to internal elements within an organization, including managerial support, organizational size, readiness, communication processes, and resource availability. Additionally, the organizational context, another critical component of the TOE framework, encompasses factors such as organizational size, structure, culture, and resources [30,32]. Organizational culture, particularly the cultivation of a technology-friendly environment, the promotion of teacher training, and the capability to integrate AI-driven tools into curricula, plays a significant role in technology adoption [33,34,35]. Studies indicate that organizational culture impacts technology adoption across various contexts, particularly in educational institutions [35,36,37]. Environmental factors pertain to the external environment in which an organization operates, primarily including government policies, cultural considerations, and external pressures [28,30]. Moreover, this theoretical model has been applied in various fields to study the factors influencing the adoption of information technology innovations. In the field of education, researchers have also conducted empirical studies based on the TOE framework to examine the factors affecting the adoption and diffusion of cloud computing technologies. For instance, Hiran et al., using the TOE-DOI framework, analyzed the differential impacts of technological, organizational, environmental, and sociocultural factors on the adoption of cloud computing in the higher education sector [38]. Similarly, Singh et al. integrated four theoretical models (TOE, TAM, DOI, and HOT-fit) to explore the effects of factors such as relative advantage, compatibility, reliability, security, organizational size, and managerial support on the adoption of cloud computing technology within India’s school education system [39].

2.2. Research Related to Topic Modeling and Sentiment Analysis

The vast amount of user-generated text has brought significant attention to the extraction of critical information from large document collections within the (Natural Language Processing) NLP community. Sentiment analysis, also known as opinion mining, is an active area of research in NLP that aims to identify subjective information and determine the sentiment polarity (e.g., positive or negative) of a given text [40,41]. In other words, sentiment analysis is the computational study of people’s opinions, assisting users in gathering the information necessary for decision-making.

Recently, sentiment analysis has been applied across various domains, including business, politics, and social media. For example, Adeshola used the Latent Dirichlet Allocation (LDA) method to perform topic modeling on text reviews of ChatGPT in India, combining it with sentiment analysis using the Valence Aware Dictionary and Sentiment Reasoner (VADER) [42]. Yu conducted a comprehensive analysis of key themes in the application of NLP in personalized learning, based on the LDA testing of top educational technology journals published between 2014 and 2023 [43]. Tan proposed a LDA-based model, Foreground and Background LDA (FB-LDA), to distill foreground topics and filter out longstanding background topics [44]. In recent years, LDA has been widely applied by many scholars in various research fields, such as scientific topic discovery, source code analysis, and opinion mining [45,46,47]. By integrating probabilistic methods, the LDA model enhances the information obtained through traditional vector space models and captures the multitasking features of documents [5]. It is considered one of the simplest and most effective topic models for analyzing text documents [48]. Sentiment recognition, however, is an inherently complex problem, leading to substantial efforts to analyze and understand its various dimensions.

SnowNLP is a Python-based natural language processing (NLP) library designed for Chinese text processing. One of its core features is sentiment analysis, which categorizes text into positive or negative emotions using built-in models trained on Chinese data. Due to its high adaptability to the Chinese linguistic context, SnowNLP performs outstandingly in analyzing public opinions, customer reviews, and social media content (e.g., Weibo, e-commerce reviews), as demonstrated in the following studies. Wang’s study involved crawling tourist reviews from the Ctrip travel platform and collecting data on China’s 5A scenic spots, fine-tuning SnowNLP’s sentiment categorization thresholds, and analyzing the correlation between tourists’ mood changes and their geographical distribution using spatial-temporal data [49]. In Fu’s research on the Wuyi Mountain National Park project, key themes were extracted from online comments using the LDA model, and sentiment trends for each theme were analyzed using SnowNLP to reveal tourists’ sentiment preferences for different landscapes [50]. SnowNLP has demonstrated its value in Chinese sentiment analysis in both academia and practice, making it particularly suitable for analyzing social media and user-generated content.

3. Research Methods, Research Platform, and Data Sources

Combining the LDA topic model and SnowNLP sentiment analysis, this study analyzes the public cognition of educational application of productive AI (AIGC) based on the TOE framework: Through social media data mining, three dimensions of technology, organization, and environment are extracted, emotional tendency is quantified, and multi-factor interaction is revealed, providing a strategic basis for AI education integration (see Figure 1).

3.1. Research Methods,

3.1.1. Sentiment Analysis

This study applied SnowNLP for sentiment analysis. When using SnowNLP, the Chinese text is first input into the module. SnowNLP then processes the words in the text sequentially, checking if they correspond to sentiment labels in its sentiment dictionary. If a match is found, the module determines the sentiment polarity of the word (e.g., positive, negative, or neutral). Based on this foundation, SnowNLP also analyzes the relationships between words and the sentence structure, such as the impact of negation words. By calculating the number of positive and negative words in the text and considering the context, SnowNLP generates a sentiment score that reflects the intensity of the sentiment. Finally, the system outputs the sentiment analysis results of the text based on these scores [51] (see Figure 2 for the detailed process).

In order to test the accuracy of the model’s sentiment classification, SnowNLP and SnowNLP supplemented by Jieba conducted sentiment classification tests. The classification samples used the same corpus, namely 500 manually labeled positive and 500 negative comments. The accuracy reflects the proportion of samples with true positive (negative) sentiment tendency among the positive (negative) sentiment tendency samples determined by the classifier; the recall rate reflects the proportion of correctly determined positive (negative) sentiment tendency samples to all positive (negative) sentiment tendency samples, accurately reflecting the ability of the classifier to determine the entire sample. From the final classification results, the optimized SnowNLP classification results reached 87%, which can basically correctly judge the sentiment polarity of the comments, and can perform sentiment classification on the comment corpus of this study (see

Table 1. Accuracy of sentiment classification.

Classification Method	Accuracy	Category	Precision	Recall	F1
SnowNLP with Jieba subscripts	0.87	Positive	0.88	0.89	0.88
		Negative	0.89	0.86	0.87

).

3.1.2. Topic Modeling

We employed the LDA model to uncover and describe the thematic content associated with the application of generative artificial intelligence in the field of education. Topic mining, also referred to as topic extraction, is a process that aims to map high-dimensional textual content into a lower-dimensional semantic space through specific methodologies [52]. The LDA topic model is one such tool used for topic mining, designed to extract latent thematic information from large document collections, thereby achieving the goals of topic clustering or text classification [53]. The LDA model generates results based on two probability distributions: the document–topic distribution, which reflects the degree of association between each document and various topics, with higher weights indicating stronger relevance to a specific topic; and the topic–word distribution, which reveals the internal structure of each topic, where higher probability values for specific words indicate their greater importance within the topic [54] (see Figure 3 for the detailed process).

3.1.3. TOE Theoretical Analysis

We applied the TOE framework to systematically analyze the impact of AIGC in education and extracted key thematic insights from public discourse in conjunction with LDA topic modeling, which facilitates topic clustering and categorization by mapping high-dimensional textual data into a low-dimensional semantic space to identify potential topics in large textual corpora. The extracted themes were categorized into three dimensions of TOE: technology, reflecting concerns about AIGC accuracy, personalization, and moral hazard; organization, highlighting discussions of institutional adoption, faculty adaptation, and infrastructure readiness; and environment, including policy regulations, social acceptance, and market impact. By combining these thematic findings with the TOE framework, we provide a structured view of AIGC integration with education in terms of technical feasibility, organizational challenges, and external factors (see Figure 4 for the theoretical framework).

3.1.4. Word Cloud Analysis

Word cloud analysis is a technique that visualizes word frequency statistics, providing an intuitive representation of the distribution of high-frequency words in a text. It effectively filters out irrelevant content, highlights differences between keywords, and emphasizes critical information within the text [55]. In a word cloud, the color of keywords typically represents different categories or topics, while the font size is proportional to the frequency of word occurrences. Thus, words that appear more frequently are displayed in larger fonts, emphasizing their dominance in the text and their representativeness of the overall theme.

3.2. Research Platform

This study selected Weibo 15.1.2, Bilibili 8.36.0, and Douyin 33.2.0 as data sources to analyze Chinese users’ comments on AIGC. These three platforms are well-known social media platforms in China, facilitating user interaction and information dissemination through their real-time information-sharing mechanisms. According to relevant statistics, Weibo’s monthly active users have reached 583 million, with daily active users totaling 256 million. In the fourth quarter of 2024, Bilibili recorded 340 million monthly active users and 102 million daily active users. Meanwhile, Douyin surpassed 700 million DAUs in 2024, with users spending over 60 min daily on average. The selection of these platforms as data sources is justified not only by their significant advantages in terms of user scale, activity level, and representativeness but also by the strong scientific basis and persuasive evidence underlying the data. This choice ensures the comprehensiveness, diversity, and representativeness of the research data, providing a robust foundation for analyzing public sentiment and opinion dynamics.

3.3. Data Crawling and Cleaning

This study utilized Python programming tools and the requests module to collect comment data from platforms such as Weibo, Bilibili, and Douyin. Approximately 40,325 comments were collected using Python web crawling programs on 30 September 2024. Keywords used in the search included “artificial intelligence”, “ChatGPT”, “Sora”, or “machine learning” in combination with “education”. The collected data included user IDs, cities, comment content, and timestamps.

Table 2. Selected comment data.

User Name	Timing	Regions	Commentaries
User 1	16 April 2023 00:15:41	Sichuan	It does work. I have a problem writing code, and then I copy it directly into chatGPT, and it just gives me the correct code, and it works.
User 2	16 April 2023 04:08:54	Guangdong	It is not perfect. There may be some small mistakes in various disciplines, such as mathematical calculations and chemical equations. Depending on gpt, you need to have some judgment to judge whether the answer is right or not.
User 3	19 January 2024 02:00:44	Guangxi	The teacher recommended us to use EmoAi.
User 4	15 April 2023 23:48:35	Shanxi	Can the teacher explain how to use GPT? How can you upgrade the quality of your questions? Really need.
User 5	17 August 2023 13:30:27	Zhejiang	Recently, I have been using the domestic version of GapAI, which has indeed improved a lot of efficiency.

provides several typical examples with distinct emotional features. During the initial data collection phase, it was observed that most of the data were unstructured and varied in format. To ensure data completeness and accuracy, empty values, duplicate records, advertisements, and other irrelevant information were manually removed, resulting in a final dataset of 29,135 valid entries. To improve the precision and reliability of subsequent analysis, the Jieba tokenizer was employed for meticulous text segmentation, while English letters, numbers, and other meaningless words were excluded [56].

According to the syntax rules of Chinese, the operation of dividing comment texts into independent words is referred to as segmentation. When filtering textual elements, the quality of segmentation directly affects the accuracy of emotional scoring for words, thereby significantly influencing the validity of the analytical conclusions [57]. Furthermore, the textual data from Weibo, Bilibili, and Douyin are characterized by their brevity, high degree of colloquialism, non-critical linguistic structures, and frequent use of internet slang. These features necessitate sentiment analysis methods distinct from traditional text sentiment analysis approaches [58]. Additionally, the diversity and complexity of Chinese expressions add further challenges and difficulties to the sentiment analysis of Weibo data.

The sentiment knowledge network dictionary used in this study is structured into six subcategories: “Positive Emotions”, “Negative Emotions”, “Positive Evaluations”, “Negative Evaluations”, “Academic Levels”, and “Needs Categories”, encompassing a total of 17,877 lexical entries. To enhance the precision and practicality of lexical filtering, low-frequency words were excluded during the initial stage. Subsequently, with the help of regular expression techniques, user comment content was matched against the sentiment dictionary to identify and retain emotional terms, resulting in a total of 1833 valid words. Among these, 20 positive sentiment words (e.g., “amazing”, “perfect,” “exciting”) and 20 negative sentiment words (e.g., “rigid”, “harsh”, “terrible”) that appeared more than 100 times were selected. Using “stiff”, “disappointed”, and “regret” as benchmark terms, the semantic orientation pointwise mutual information (SO-PMI) method was employed to calculate the pointwise mutual information values between these 1833 emotional terms and each benchmark term. Based on the obtained SO-PMI values, the emotional tendencies of the sentiment terms were categorized, resulting in the construction of a domain-specific sentiment lexicon.

Based on the segmentation results, the sentiment knowledge dictionary (including the degree adverb dictionary), and the domain-specific sentiment lexicon, the following scoring system was developed. In the initial phase, the annotated data underwent preprocessing, which included the detailed handling of degree adverb modifiers and the removal of irrelevant terms (stop words). Subsequently, a Python script was utilized in conjunction with the human emotion lexicon to determine the sentiment polarity of the terms. Specifically, positive sentiment words were assigned a score of +1, while negative sentiment words were given a score of −1. Furthermore, the script checked for the presence of degree adverbs preceding the sentiment words. If such adverbs were detected, their corresponding scores were multiplied by the weight coefficient assigned to the adverb. Conversely, if no degree adverbs were detected, the scores remained unchanged. Additionally, the script repeatedly verified whether a negation word appeared before the evaluated term. This process was iteratively executed until a comprehensive traversal of the entire annotation was completed, yielding a definitive score.

4. Research Results

4.1. Word Cloud Analysis Results

Word clouds visualize the frequency of keywords, with the size of each word corresponding to its overall frequency—the larger the word, the higher its frequency. These visualizations highlight the main themes that capture the audience’s attention, as shown in Figure 5 and Figure 6. In the thematic word cloud in Figure 5, terms such as “education”, “ChatGPT”, “robot”, “university students”, and “development” are prominently displayed. These terms represent key points of interest and concern for the audience, underscoring the relationship between education and artificial intelligence. In the sentiment word cloud in Figure 6, negative emotion words such as “useless”, “sense of crisis”, and “anxiety” appear relatively frequently, indicating that audience attitudes toward generative artificial intelligence are primarily marked by worry and skepticism, but not outright fear. At the same time, positive terms such as “empowered”, “hope”, and “outstanding” also appear, reflecting satisfaction and optimism among many users regarding the use of generative AI tools. This highlights the public’s strong interest in the potential of artificial intelligence to drive progress in the field of education.

4.2. LDA Topic Modeling

Based on an evaluation of the results with varying numbers of topics and their associated terms, and in alignment with the subsequent educational ecology theory, six topics were determined to be the optimal choice. Figure 7 presents the interactive visualization of the LDA topic modeling. The left side displays the distribution of topics and their correlations, while the right side shows the top 30 most-relevant terms associated with each topic. In terms of distribution, Topic 1 significantly exceeds the other topics, while Topic 5 shows a slight overlap with Topic 6, indicating that certain parts of these two topics are similar or related. The remaining four topics have no overlapping sections.

Table 3. Topic content.

Topic	Feature Word
Topic 1	Humans, AI, ChatGPT, Robots, Artificial intelligence, Work, Development, Models, Tools, Stress, Data, Era, etc.
Topic 2	Children, Sense of Crisis, Feelings, Truly, Employment, Machines, Young children, Primary school, Society, Adults, Thoughts, Growing up, etc.
Topic 3	Education, Replacement, Impressive, China, World, Funding, Northbound, Software, Internet, Outflow, Translation, Awareness, Professions, etc.
Topic 4	Teacher, Download, Anxiety, Software, Industry, Mobile phone, Knowledge, Translation, University, Like, Maturity, Intelligent, College Students, etc.
Topic 5	School, Students, Learning, Anxiety, Profession, Dreams, Teacher, Ability, Knowledge, Intelligence, Logic, Meaning, etc.
Topic 6	Download, Excellent, Worry, English, Industry, Video, Country, Voice, Code, Mathematics, College Entrance Exam, Reality, etc.

records the top 10 most-relevant terms corresponding to each of the six topics.

4.3. Proportion Analysis of Sentiment Across Topics

Six themes were first extracted from the comments by LDA modeling, and then SnowNLP was used to analyze the sentiment distribution of each theme (see Figure 8 and Table 3). The public reaction to artificial intelligence in the field of education demonstrates a complex emotional distribution. Most topics exhibit a positive sentiment tendency, with Topic 3 showing the highest number of positive sentiments. This suggests that Topic 3 likely involves advancements or innovations in artificial intelligence or education, eliciting a positive attitude from the public. On the other hand, Topic 5 displays a higher level of negative sentiment, particularly the largest number of negative sentiments among all topics. This could be attributed to challenges or controversies related to artificial intelligence and education, such as privacy concerns, inequality in education, and other issues, which evoke strong negative reactions from the public. Additionally, several topics exhibit a certain proportion of neutral sentiments, particularly Topic 2 and Topic 4, indicating that the public holds a more rational perspective or has yet to form a clear opinion on these topics. Overall, the discussions on artificial intelligence and education reflect a coexistence of optimism about technological innovation and concerns about its applications. These concerns are especially prominent regarding the potential impacts of technology on educational equity, privacy protection, and job displacement. Future research could further investigate the root causes of these negative sentiments and explore ways to improve the application of educational technology through policies and practices, thereby reducing public negative reactions and enhancing acceptance.

When analyzing the word clouds of the six topics in Figure 9, it is essential to highlight how AI is reshaping human learning and educational environments in unprecedented ways. With the widespread application of advanced AI models such as ChatGPT, artificial intelligence has transitioned from theory to practice, becoming a pivotal tool in education, as detailed below:

The co-occurrence of “human” and “AI” in Topic 1 underscores the core of this transformation: the deep integration of humans and intelligent technologies. In educational contexts, terms like “robot” and “artificial intelligence” highlight their emergence as innovative teaching aids, gradually transforming traditional teaching methods to provide students with more personalized and efficient learning experiences [59,60]. Words like “model” and “tool” emphasize specific AI applications in education, such as intelligent recommendation systems and online learning platforms, which leverage data analysis and machine learning to deliver tailored learning pathways and resources. However, terms such as “pressure” and “data” reveal challenges in this transformation. While AI-driven education relies on data collection and analysis, it also raises concerns about data security and privacy. Additionally, both teachers and students face psychological and learning pressures in adapting to these new teaching approaches. On a positive note, terms like “development” and “era” convey an optimistic attitude, suggesting that despite challenges, AI offers immense potential for educational progress. With continuous learning and adaptation, humans can fully harness AI’s advantages to drive innovation in education.

Topic 2 focuses on children, particularly primary school students, who are the future backbone of society. The term “children” highlights the critical importance of their growth and educational environment. Words like “sense of crisis” and “feeling” reflect society’s perception of uncertainty and challenges brought about by rapid AI development. Many adults and educators’ express concerns that, with the widespread adoption of “machines” across industries, the traditional job market will undergo profound changes, raising urgent questions about how children will secure their place in the workforce. Terms such as “employment” and “society” directly link to children’s future livelihoods and development. As AI transforms societal demands for talent, children must cultivate collaborative skills, innovative thinking, and problem-solving abilities to adapt to future job markets [61]. Meanwhile, terms like “adults” and “thoughts” indicate the attitudes and actions of adults in this transformation. Adults need to update educational concepts and prioritize the development of children’s comprehensive qualities, such as emotional understanding and interpersonal skills, which are difficult for machines to replace.

Topic 3 explores how the rapid development of AI is reshaping career development paths in education. Keywords like “replace” and “powerful” highlight the significant potential of AI to substitute human roles in specific professions, sparking deep concerns about the future of human careers [62]. Both in China and globally, with continued investment and the proliferation of internet technologies, AI software and cutting-edge technologies are dramatically altering the job market landscape. The term “education” underscores its crucial role in cultivating future workforce talent amidst these changes. Additionally, terms like “translation” and “awareness” suggest that enhancing cross-cultural communication skills and self-awareness is essential for adapting to new career environments and expanding professional development opportunities. While AI introduces the pressures of job displacement [63], terms such as “funding” and “learning” offer positive signals, indicating that through continued investment in education and skill development, humans can find new career opportunities, achieving personal and societal progress.

Topic 4 focuses on the potential transformations and profound impacts on human teachers’ careers due to AI’s growth. Keywords like “teacher” and “profession” directly address the theme: exploring how AI reshapes the occupational structure of education and redefines teachers’ roles [64]. The widespread use of intelligent tools such as “software” and “smartphones” has revolutionized knowledge acquisition and dissemination, imposing new demands on teachers’ professional expertise, teaching skills, and technological capabilities. Functions such as “download” and “translation” have made learning resources more accessible but also raised issues of information overload and student anxiety, necessitating teachers’ guidance in resource curation and utilization. In the university setting, terms like “university” and “students” point to direct recipients of education, whose preferences and adaptability influence the evolution of teaching methods. Words such as “like” and “mature” reflect students’ expectations for teaching styles, encouraging teachers to balance professionalism with emotional communication and personalized instruction [65]. Facing these challenges, teachers must demonstrate “intelligence” and “excellence” by continuously learning, innovating teaching methods, and leveraging their unique strengths in emotional understanding and humanistic care to lead the AI-driven educational era.

Topic 5 centers on how schools can promote students’ holistic development in the intelligent era [66], particularly in addressing challenges like academic anxiety and career planning. Keywords such as “school” and “student” clarify the focus on educational institutions and learners, while “learning” and “knowledge” highlight the core of educational activities: acquiring knowledge and enhancing capabilities. With the increasing integration of smart technologies, “intelligence” not only transforms teaching methods but also raises new demands for students’ learning approaches and cognitive skills, such as logical reasoning and comprehensive capability development. However, “anxiety” reflects the psychological challenges many students face when dealing with academic pressures and future career uncertainties, requiring collective attention from schools, teachers, and students. Terms like “career” and “dream” highlight the motivations driving students forward. Schools should guide students to align their interests with societal needs, plan their careers effectively, and pursue their dreams. In this process, “teachers” play a crucial role not only as knowledge transmitters but also as mentors who inspire students to develop the correct values and problem-solving abilities through personal example and moral guidance.

Topic 6 emphasizes the profound transformations and challenges brought by digital education resources, particularly the widespread use of “download” functions. The term “excellent” reflects the positive impact of digital resources on improving education quality and facilitating the sharing of high-quality resources [67]. Multimedia formats like “video” and “audio” have made learning more engaging and effective, greatly enhancing students’ interest in education. However, “concern” highlights associated worries, such as unequal access to educational resources, an overreliance on electronic devices, and information overload. In specific disciplines like “English”, the proliferation of digital resources has diversified learning methods but also imposed higher demands on teachers’ professional competencies and teaching approaches [68]. Moreover, the policy direction and regulatory efforts at the “industry” and “national” levels significantly influence the development and application of digital education resources. Additionally, digital teaching in STEM subjects like “coding” and “mathematics” showcases the immense potential of technology to drive educational innovation and prepare future talent.

5. Discussion

Based on the research findings, the following conclusions are drawn:

Given the widespread impact of AI on the field of education, China demonstrates a high level of attention toward the development of AIGC technologies, such as ChatGPT, in education. Despite the prohibition of non-China-based AIGC tools like ChatGPT in the country, the influence of AIGC on education remains undeniable, garnering significant attention. This observation is supported by the following three points:

(1)

Sentiment Distribution: Positive sentiments significantly outweigh negative ones, indicating public support for the disruption AIGC brings to the traditional education sector. This result is markedly different from previous work [5], where negative sentiments toward ChatGPT were found to be lower than positive sentiments.

(2)

Thematic Insights: Frequently appearing keywords, such as “education”, “ChatGPT”, “school”, “future”, and “work” suggest that Chinese people believe AIGC development will greatly impact the country’s educational landscape. This result is clearly more focused on the education sector, whereas previous work [6] was broader in scope, which hinders in-depth research specifically within the educational domain.

(3)

LDA Topic Analysis: Both societal and institutional dimensions, including schools, teachers, and students, have been influenced by AIGC. Different from the previous works [69,70], this study employs quantitative methods to derive the keywords and sentiment scores for the six topics, revealing that each emphasizes different aspects, yet they overlap with one another. This effectively demonstrates the complex interconnections among the various elements within the educational ecosystem.

Compared to earlier AI technologies, the new generation of AIGC is not only more advanced but also widely applied across various societal domains. This gives AIGC a more profound and far-reaching impact on education, touching nearly all aspects of society. To comprehensively examine this influence, a broader perspective and a more universal framework are required. Different from previous studies [5,6,42], this study provides a more comprehensive and in-depth analysis through the combination of qualitative and quantitative approaches. The TOE framework is employed to analyze the impact of AI on education across three dimensions: Technology, Organization, and Environment.

(1)

From a technological perspective, AIGC offers students a more personalized and efficient learning experience, which is one of the reasons for the high levels of positive sentiment. From Topic 1, 3, and 6, it is evident that the advanced nature and maturity of AIGC influence its effectiveness in educational applications. The successful implementation and promotion of generative AI in education often depends on a comprehensive and structured framework [71,72]. This framework must address not only the capabilities of AI technology itself but also the readiness of educational institutions in terms of technological infrastructure. For effective AI implementation, educational institutions require adequate hardware and software support, including cloud computing, big data processing capabilities, the widespread use of intelligent devices, and robust network infrastructure. Many regions have already integrated ChatGPT into classrooms to enhance teaching effectiveness and student learning experiences. For instance, on 1 June 2023, the Hong Kong University of Science and Technology became the first institution in Hong Kong to officially provide a localized version of ChatGPT [73]. This move reflects the growing recognition of the importance of AI tools, especially generative AI, in enhancing learning experiences. ChatGPT and similar AI tools can provide instant learning assistance and recommend personalized learning paths based on individual student needs. However, complex AI tools may pose adaptation challenges for teachers and students, making usability and user-friendly interfaces critical for their success in education. Research indicates that easy-to-use AI tools significantly promote students’ willingness to adopt and engage with them. Empirical analyses using structural equation models have found that intuitive interfaces and straightforward operations increase students’ likelihood of continuing to use these tools for learning [74].

(2)

From an organizational perspective, schools and educational institutions must develop adaptive organizational cultures, management structures, and technological support, which is further corroborated by Topic 2, 4, and 5. The leadership of educational institutions plays a critical role in driving AI adoption. Particularly in higher education, strategic leadership is essential for promoting AI adoption and innovation. By formulating strategic plans tailored to educational needs, leaders can balance the potential benefits and risks of AI technology, ensuring its successful integration into education [75]. Unlike earlier concerns about AI threatening the balance of teaching professions [76], recent research has shifted to focus on whether teachers and educational administrators possess the skills and motivation to use AI tools. This shift highlights that while technology is transformative, its successful implementation depends on the proactive acceptance and adaptation by teachers and administrators. Effective training and capacity building for teachers have thus become urgent issues in the education sector [77]. Teachers need to embrace and apply AI tools in their teaching, such as automated grading systems and intelligent teaching assistants [78,79]. Educational institutions must integrate AI technologies with existing teaching methods, curriculum content, and assessment practices to maximize their value. Some scholars suggest that institutions can promote academic integrity by designing assessments that include limited AIGC and by establishing clear policies [42].

(3)

From an environmental perspective, the application of artificial intelligence in education is influenced by social and cultural contexts, legal frameworks, market demands, and competitive pressures, which is consistent with Topic 2, 3, and 4. National education policies vary in their promotion and use of AI technologies. For example, while some countries ban the use of ChatGPT, they simultaneously develop localized AI tools, such as China’s Wen Xin Yi Yan [5]. Other countries support AI adoption in education but impose strict regulations on data privacy and ethics. For instance, the EU’s General Data Protection Regulation (GDPR) establishes stringent compliance standards for handling personal data, ensuring students’ privacy rights and preventing the misuse of educational data. Regardless of these variations, AI use in education must adhere to data protection laws, safeguarding students’ personal information and learning data. Legislative efforts must keep pace with the rapid advancements in AI technologies [80]. Another critical factor is the collaboration between educational institutions and AI technology providers. Close partnerships between these entities are essential to align AI tools with educational needs [81]. Furthermore, international collaboration and the development of transnational educational ecosystems are crucial. As AI technology evolves globally, fostering international cooperation becomes increasingly important. Governments and educational institutions can advance the global application of AI in education by sharing data and conducting collaborative research, ultimately creating a more balanced and open technological ecosystem for education. Some scholars have highlighted the need for governments worldwide to enhance the oversight of data quality. Continuous updates to conversational AI databases are necessary to maintain system quality and capacity, ensuring the effective and ethical application of AI in education [82].

6. Implications

6.1. Theoretical Significance

This study advances the understanding of public attitudes and dominant sentiments regarding the application of AIGC in education. By carefully selecting a substantial number of text-based comments from well-known social media platforms as data samples, we directly captured diverse societal perceptions and insights into AIGC’ s role in education. Leveraging Python tools, we employed a combination of data mining techniques, sentiment analysis methods, and public opinion topic analysis strategies to conduct a comprehensive and in-depth examination of the collected data. The findings not only reveal the complex emotional attitudes of the public toward AIGC in education but also pinpoint key issues and trending topics that draw public attention.

Integrating the TOE framework, this study further uncovers the impacts of AIGC in the education sector and forecasts its developmental trajectory from three dimensions: technological, organizational, and environmental. For instance, the introduction of AIGC depends not only on technological advancement but also on whether the organizational structures of educational institutions can adapt to these new technologies and whether external social, cultural, and legal policies support its application. This multi-faceted analysis offers a holistic understanding of how AIGC can function effectively in educational settings and overcome potential barriers and challenges. Policy-makers, technology developers, and educators can use this framework to gain deeper insights into the technological, organizational, and environmental challenges associated with implementing AIGC, enabling them to design more precise strategies for advancing educational technologies and facilitating the intelligent transformation of education.

Moreover, by exploring the multi-dimensional impacts of AIGC in education, this study not only sheds light on the current application landscape but also provides strategic guidance for future educational technology innovations. The TOE framework allows us to predict future technological trends and potential challenges from the perspectives of technology, organization, and environment, offering strategic references for the sustainable development of educational technologies. The study also emphasizes the importance of fostering harmonious interactions between the public and generative AI, laying a theoretical foundation for promoting the deep integration and sustainable development of AIGC in education.

6.2. Practical Significance

AIGC brings teaching innovation and personalization breakthroughs to the education field. Based on the Technology–Organization–Environment framework, this paper analyzes the strategic path of systematic application of AIGC in educational institutions from the three dimensions of technological adaptation, organizational transformation, and external environment response.

(1)

Guiding the comprehensive implementation and adaptation of educational technologies. AIGC opens up new possibilities for education, with its powerful capabilities in text generation and comprehension revolutionizing various aspects such as pedagogy, learning experiences, and personalized education [24]. Through the technological dimension of the TOE framework, the education sector can better understand the potential applications and implementation conditions of AIGC. Technological innovation is not merely about introducing new tools or systems; it requires a consideration of the educational needs and AI acceptance levels of key stakeholders, such as teachers and students, who are the direct users of these technologies [83]. This approach helps educational institutions to identify technological maturity, implementation requirements, and potential challenges, providing clear guidance for adopting and implementing educational technologies. Especially in the context of rapid technological advancements, such analysis offers theoretical support for evaluating the feasibility of educational technologies and reduces the risk of blindly following trends.

(2)

Driving organizational transformation and innovation in education. Although AI has the potential to transform education [84], successful educational outcomes are not solely achieved through the use of advanced technologies [85]. The effective utilization of technology requires organizational support and adaptation to truly enhance educational goals. The organizational dimension emphasizes the internal adaptability issues faced by educational institutions during AIGC implementation. For example, factors such as management structures, leadership decision-making capabilities, and the acceptance levels of teachers and students significantly influence the practical effectiveness of AIGC applications. By examining the relationship between organizational structure and technology adoption, educational administrators and decision-makers can better address internal reforms and adjustments. This ensures that educational organizations can efficiently absorb and utilize new technologies, improving the quality of education and management efficiency.

(3)

Addressing changes and challenges in the external environment. The environmental dimension of the TOE framework is particularly important, encompassing factors such as policies and regulations, social culture, and market demand, which may affect the implementation and dissemination of AIGC technologies. The application of AI technologies in education must consider the differences in domestic and international legal policies (e.g., data privacy protection regulations) as well as varying levels of social and cultural acceptance and ethical debates. For instance, while some countries and regions demonstrate high acceptance of AI technologies, others remain skeptical. By incorporating the environmental dimension, educational policy-makers can anticipate and address these challenges, ensuring that the promotion and implementation of technologies align with societal expectations and requirements.

The TOE framework reveals that the application of AIGC in education requires Technology–Organization–Environment synergy: technology matches demand, organization optimizes decision-making, and environment anticipates policy. The study shows that educational innovation needs to break through the tool thinking and establish a multi-dimensional dynamic adaptation mechanism to avoid technology bubbles and realize the value leap from proof of concept to in-depth landing.

7. Limitations

Firstly, this study primarily focuses on the emotions and attitudes of the Chinese public, lacking comment data from social media platforms in other countries. This limitation may affect the generalizability of the findings. In fact, significant differences exist among countries in terms of policies, regulations, and public sentiments toward AIGC. Therefore, developing AI education policies tailored to the specific contexts of different countries is a critical issue for future research and policy-making. To gain a more comprehensive understanding of public emotions and attitudes toward AIGC in education across various countries, future studies should expand the sample scope to include more nations and cultural contexts, facilitating cross-national comparative research. Secondly, to further explore the factors influencing AIGC’s application in education, future research could adopt more refined quantitative analysis methods, such as empirical studies based on structural equation modeling or experimental designs to validate educational outcomes in real-world settings. These methods could more clearly reveal the key variables influencing AIGC applications in education and their underlying mechanisms. Lastly, future research should strive to provide a more comprehensive, forward-looking, and balanced perspective, thoroughly considering the profound impact of AIGC on the educational landscape and the evolving public emotions and attitudes toward technological advancements. Such efforts would not only offer policy-makers more precise decision-making support but also provide educators and technology developers with more effective guidance, fostering positive interactions and mutual progress between AIGC and education.

8. Conclusions

This study uncovers the underlying public sentiments regarding the impact of AIGC on the education sector. Specifically, it conducts an in-depth analysis of public opinion from multiple social media platforms in China, examining perspectives and attitudes toward AIGC in educational contexts through the three dimensions of technology, organization, and environment. By incorporating the TOE theoretical framework, this study offers a more comprehensive perspective on the application of AIGC in education and provides valuable theoretical support for its further integration into the field. This research not only enriches the theoretical understanding of the role of artificial intelligence in education but also delivers actionable insights for policy-makers, educators, and technology developers.

As GAI technologies continue to advance and gradually permeate the field of education, scholars and educational practitioners face dual challenges. On the one hand, they must fully explore the potential of AI to drive educational innovation and progress. On the other hand, they must effectively address the ethical, social, and economic issues arising from its application. In this complex context, educators, senior leaders, educational institutions, and society as a whole need to adopt a positive and open mindset to navigate the continuous development and challenges of AI technologies. Educators should actively embrace AIGC technologies, exploring their potential applications in teaching while maintaining a focus on students’ individual needs and ensuring educational equity. Senior leaders must provide strategic guidance for the adoption of educational technologies, ensuring the effective implementation of policies and the allocation of resources to achieve a deep integration of technology and education. Educational institutions should promote interdisciplinary collaboration, build an innovative ecosystem that aligns with AI developments, and cultivate future-ready students and teachers equipped to thrive in an AI-driven world.