The Media Spatial Diffusion Effect and Distribution Characteristics of AI in Education: An Empirical Analysis of Public Sentiments Across Provincial Regions in China

Abstract

With the rapid integration of artificial intelligence (AI) technologies in the field of education, public sentiment towards this development has gradually emerged as an important area of research. This study focuses on the sentiment analysis of online public opinions regarding the application of AI in education. Python was used to scrape relevant online comments from various provinces in China. Using the SnowNLP algorithm, sentiments were classified into three categories: positive, neutral, and negative. The study primarily analyzes the spatial distribution characteristics of positive and negative sentiments, with a visualization of the results through Geographic Information Systems (GIS). Additionally, Moran’s I and Getis-Ord Gi* are introduced to detect the spatial autocorrelation of sentiment attitudes. Furthermore, by constructing a multivariable geographical detector model and MGWR, the study explores the impact of factors such as the development of the digital economy, the construction of smart cities, local government policy attention, the digital literacy of local residents, and the level of education infrastructure on the distribution of sentiment attitudes. This research will reveal the regional disparities in AI and education-related online public sentiment and its driving mechanisms, providing data support and empirical references for optimizing the application of AI in education.

1. Introduction

The application of artificial intelligence (AI) technology in the field of education is currently undergoing rapid development. AI, through intelligent sensing, teaching algorithms, and data-driven decision-making, enables automated analysis and precise interventions for learners, teachers, and educational content. It provides timely and personalized guidance and feedback, supporting and fulfilling educational and learning needs. As a revolutionary technology, AI’s application in education is highly anticipated, not only having a profound impact on traditional teaching models and educational ecosystems but also sparking widespread discussions and attention across society [1]. The promotion and application of AI technology are closely tied to the public’s awareness and acceptance. Despite AI’s promising prospects in the educational field, public sentiment has a direct and far-reaching impact on the practical application and sustained development of AI technology. If the public’s attitude towards AI is predominantly negative, it may lead to rejection of the technology or hinder its widespread adoption, affecting its application scenarios, acceptance, and future development [2].

Attitude is defined as an individual’s positive or negative response to a specific object, ultimately forming a favorable or unfavorable psychological tendency [3]. Artificial intelligence (AI) has driven new learning models and improved teachers’ workflows. However, with the emergence of a new technology, not all users will correctly understand and apply it. The controversial nature of AI technology may also lead to polarization in public discourse. Therefore, when introducing AI technology, it is essential to investigate public attitudes towards the technology and analyze the factors influencing these sentiments [3]. Previous research has emphasized the importance of learners’ attitudes towards technology, highlighting that these attitudes play a crucial role in their ability to absorb or engage with technological tools. Notably, negative attitudes towards technology are associated with reduced interest in technology-based learning and negative perceptions of technological tools. Such emotions hinder the effective use of technology, thereby obstructing the learning process [4]. Negative attitudes towards technology may adversely affect users’ willingness and efficiency in using technological platforms [5]. Therefore, when introducing new technology, it is necessary to investigate public sentiment towards the technology and explore the potential factors influencing it.

In recent years, with the penetration of social media and the internet into public life, people tend to exchange and share various opinions on social networks, making online public opinion an important channel for reflecting public sentiment. A massive amount of information spreads rapidly and instantaneously through social networks, triggering widespread emotional resonance. Monitoring and analyzing AI-related public opinion on online social platforms can serve as a significant revelation of societal sentiment and collective consciousness [6,7]. Currently, research on the application of artificial intelligence technology in education mainly focuses on aspects such as the technical implementation of educational applications, the assessment of teaching effectiveness, and its impact on traditional education. For example, Cho et al. analyzed the application of AI intelligent teaching systems in personalized learning [8], Liu et al. studied the role of AI teaching assistants in improving classroom teaching quality [9], and Ayeni et al. explored the multiple roles of AI technology in personalized learning, adaptive learning, and other areas [10].

At present, there are two opposing viewpoints regarding the use of AI in education. Some view it as a valuable tool that can enhance learning and reduce teachers’ workload, adopting a positive attitude towards AI [11]. In contrast, others hold a negative attitude towards AI in education, believing it diminishes students’ ability to think independently and may facilitate cheating and plagiarism, posing a potential threat to academic integrity and teaching quality [12]. These contrasting viewpoints reflect the complex emotional attitudes of the public towards the application of AI in education [13]. Sentimental attitudes are not only influenced by individual experiences and values but may also be shaped by the surrounding social environment, educational systems, and policy orientations. For example, regions with more developed technology and abundant educational resources are more likely to accept and promote AI technology, believing it can effectively address the shortcomings of traditional education. However, in regions with weaker technological foundations or high sensitivity to academic norms, people may be more inclined to focus on its negative impacts, such as ethical issues and potential risks.

In addition, it is worth noting that the spread and distribution of emotional attitudes often exhibit significant clustering effects and diffusion characteristics [14]. Public attitudes towards AI technology are not entirely independent of individual emotions but are influenced by multiple factors such as group culture and the social environment [15]. Within the framework of group interaction and social identity, public attitudes towards AI in education stem not only from individual cognition and emotions but are deeply rooted in specific social and spatial environments [16]. A region’s collective emotional attitude tends to influence surrounding areas through social networks and group interactions. Factors such as local government support and the level of digital infrastructure construction also contribute to the spatial differentiation of group emotional attitudes across regions [17,18,19,20,21]. Existing studies on sentiment analysis of AI in education mainly use survey methods, which have limited sample sizes and insufficient geographic coverage, making it difficult to fully reflect the true attitudes of the public in different regions. Currently, comprehensive research combining sentiment analysis with spatial distribution characteristics remains insufficient. Additionally, some studies have pointed out that factors such as the development of the digital economy and the level of educational informatization may influence the public’s acceptance of new technologies. However, no research has yet explored the spatial impact mechanism of these factors on emotional attitudes towards AI in education [22]. Against this backdrop, studying the spatial distribution characteristics of emotional attitudes towards AI in education and their influencing factors is particularly important. On the one hand, it can reveal the public’s level of acceptance and key concerns regarding AI applications, providing valuable insights for policymakers. On the other hand, analyzing the spatial distribution and influencing mechanisms of these emotional attitudes can offer deeper insights into the interaction between technological diffusion and social environments, thereby providing a basis for the rational promotion of AI technology.

Combining sentiment analysis with spatial analysis provides a comprehensive understanding of the dynamic mechanisms underlying this phenomenon and offers theoretical support for promoting the sustainable application of artificial intelligence in education. As an important branch of natural language processing, sentiment analysis aims to identify and extract emotional associations from textual data [23,24]. In recent years, with the development of technologies such as deep learning, the accuracy and scope of sentiment analysis have been continually improving. In China, SnowNLP, a powerful Chinese natural language processing toolkit, has become widely used in the field of Chinese sentiment analysis. By categorizing sentiments in large volumes of online comments, it is possible to effectively assess public attitudes towards the application of AI in education. Geographic Information Systems (GIS), as a powerful spatial analysis tool, can spatially visualize sentiment analysis results, revealing the distribution characteristics and spatial correlations of emotional inclinations in different regions. Therefore, this study aims to apply a hybrid approach of sentiment and spatial analysis to transform social media comments into sentiment attitude values, visualizing group differences from different regions. Further, it will explore the influencing factors and investigate the interaction between socio-economic environments and the public’s acceptance of technological diffusion. Based on the interpretation of these feature values, the study will provide a basis for the rational promotion of AI technology and contribute to the development of AI applications in education.

2. Materials and Methods

2.1. Data Collection

This study focuses on the 31 provincial-level administrative regions of China (excluding Hong Kong, Macau, and Taiwan) and analyzes public sentiment towards AI applications in education, exploring regional differences in emotional attitudes towards AI and the influencing factors behind them. The study will first apply the SnowNLP algorithm to perform sentiment analysis on data from major Chinese social media platforms (such as TikTok) to assess the public’s emotional attitudes (positive or negative) regarding AI in education. Spatial analysis will then be conducted using ArcGIS 10.2 software to visualize the spatial distribution characteristics of emotional attitudes across different regions. Finally, the study will examine the socio-economic factors influencing these spatial characteristics, such as the level of digital economy, technological innovation capacity, regional development vitality, residents’ digital literacy, and policy environment, and their impact on emotional attitudes.

The sentiment data for this study is sourced from the SnowNLP algorithm, which processes social media discussions and online public opinion from major Chinese social platforms, identifying, extracting, quantifying, and performing thematic analysis to capture two main categories of emotional data: positive and negative sentiments. SnowNLP is a method specifically designed for processing Chinese text, giving it a significant advantage in Chinese sentiment analysis, and it has been widely applied in social media sentiment analysis. When using SnowNLP for sentiment analysis, the Chinese text to be analyzed is first passed into the module. SnowNLP scans the words in the text to check if they have matching entries in the sentiment dictionary. Once a match is found, the sentiment polarity (positive, neutral, or negative) of the word is determined, and the number of positive and negative sentiment words in the text is counted. Furthermore, to improve the accuracy of sentiment analysis, SnowNLP also considers the relationships between words and sentence structure, such as the influence of negation words on sentiment. Based on the number of sentiment words, their polarity, and contextual information, SnowNLP calculates a sentiment score to reflect the strength of the sentiment. Ultimately, based on the calculation results, SnowNLP outputs the sentiment analysis results [25].

During the data collection process for sentiment attitudes, this study applied precise filtering operations to the data. Before starting data scraping, the data content to be collected was manually selected based on the core topic of the study to ensure that the gathered data closely aligned with the thematic requirements. During the data scraping process, a batch-by-batch strategy with immediate verification was used to reduce error rates. Using a crawler tool based on a crawler program, approximately 40,325 comment data points were collected from platforms such as TikTok, Bilibili, and Weibo. This data collection was completed on 30 September 2024, and included information such as user ID, city, comment content, and comment time.

Since the data collected in the initial stage was mostly unstructured and varied in format, a detailed preprocessing process was implemented to ensure the completeness and accuracy of the data. This process involved manually removing null values, duplicate records, advertising content, and irrelevant information. To further improve the accuracy and reliability of subsequent data analysis, more meticulous measures were taken. First, Excel’s duplicate highlighting and string processing features were used to remove duplicate evaluation records and short texts containing fewer than five words. Second, programming techniques and the Jieba word segmentation tool were used to perform in-depth word segmentation of the document content, effectively removing redundant information such as English letters, numbers, and meaningless words. After this series of processes, a dataset containing 37,135 valid data points was constructed.

The data for the influencing factors were obtained from authoritative platforms such as the National Bureau of Statistics of China and the National Earth System Science Data Center—National Science and Technology Infrastructure Platform. A total of 13 indicators (listed in

Table 1. Influencing factor index of AI educational emotional attitude.

Indicators	Meaning of Indicators
F1	Population Density
F2	Urbanization Rate
F3	Nighttime Light Index
F4	Per Capita Income of Residents
F5	Technology and Software Service Revenue
F6	Technology Transfer Amount
F7	Patent Authorization
F8	Number of Information Service Enterprises
F9	Government Report Statistical Economic Indicators
F10	Education Expenses
F11	Mobile Internet Users
F12	Internet Broadband Users
F13	Rural Broadband Users

), including digital economy level, technological innovation capacity, regional development vitality, residents’ digital literacy, and policy environment, were included to showcase a more comprehensive potential causal relationship from different perspectives. Additionally, to eliminate population differences between regions and dimensional differences between indicators, standardization and classification processing of the research data were conducted using min–max normalization and natural breaks methods.

2.2. Methods

2.2.1. Imbalance Index

The imbalance index is an important indicator for studying the degree of distribution balance of the research object across different regions. The imbalance index (S) takes values between 0 and 1, with a higher S value indicating a greater imbalance in the distribution of AI sentiment attitudes across China [26]. Its calculation formula is as follows:

$$S={\displaystyle \frac{\sum _{i=1}^n Y_i-50(n+1)}{100n-50(n+1)}}$$

where n is the number of study regions, and ∑_i is the cumulative percentage value of the i-th ranked research unit attribute, ordered from largest to smallest.

2.2.2. Spatial Autocorrelation Analysis (Moran’s I)

Moran’s I index is a key indicator for analyzing the spatial correlation relationship between units in the research region and can be used to identify and measure clustering patterns in the distribution of AI sentiment attitudes across China. When Moran’s I > 0, it indicates a positive spatial correlation between units, suggesting a clustering trend; when Moran’s I < 0, it indicates a negative spatial correlation. Moran’s I index can be tested for significance using the Z-score. When the Z-score exceeds the critical value of 1.96, it indicates that the spatial autocorrelation is statistically significant, with a probability of less than 5% that this clustering pattern was randomly generated [27]. The calculation formula is:

$$I={\displaystyle \frac{N\sum _i \sum _j w_{ij}(X_i-\overline{X})(X_j-\overline{X})}{\sum _i \sum _j w_{ij}\sum _i (X_i-\overline{X}{)}^2}}$$

where X_i and X_j are the mean values of the sample points in regions i and j, w_ij is the spatial weight matrix, and n is the total sample size.

2.2.3. Getis-Ord Gi*

Unlike the global Moran’s I index, the Getis-Ord Gi* index is an effective tool for exploring local spatial clustering distribution characteristics. It distinguishes the degree of spatial distribution clustering of variables through cold spots (low values) and hot spots (high values), reflecting the clustering and distribution of the research object in local spatial regions [28]. The calculation formula is:

$$G_i^{*}={\displaystyle \frac{\sum _{j=1}^n w_{i,j}{x}_j-\overline{X}\sum _{j=1}^n w_{i,j}}{S\sqrt{\frac{\left[n\sum _{j=1}^n w_{i,j}^2-{\left(\sum _{j=1}^n w_{i,j}\right)}^2\right]}{n-1}}}}$$

where a Z-score greater than 0 indicates tighter clustering of high-value attributes (forming a hot spot), and a Z-score smaller than 0 indicates tighter clustering of low-value attributes (forming a cold spot).

2.2.4. Geographical Detector (Geo-Detector)

Geo-Detector is a quantitative method used to reveal the driving factors behind the spatial distribution of certain phenomena. It can detect the influence of socio-economic factors on the spatial differentiation of AI sentiment attitudes and explore to what extent these factors explain the spatial variations in AI sentiment attitudes [29,30]. The calculation formula is:

$$q=1-{\displaystyle \frac{{{\displaystyle \sum }}_{h=1}^L{N}_h{\sigma }_h^2}{N{\sigma }^2}}=1-{\displaystyle \frac{SSW}{SST}}SSW={{\displaystyle \sum }}_{h=1}^L{N}_h{\sigma }_h^2,SST=N{\sigma }^2$$

where N is the total sample size, is the total variance, h = 1, 2, ..., L represents secondary regions, Nh is the sample size of region h, and is the variance of region

2.2.5. Multiscale Geographically Weighted Regression (MGWR)

MGWR estimates spatial bandwidths and local parameters for different variables, allowing the differentiation of whether a variable has a stable global effect or exhibits spatial heterogeneity. It is commonly used to explore the driving mechanisms of complex geographical phenomena [31].

$$y_i=b{w}_0\left({\beta }_{0i}\right)+b{w}_1\left({\beta }_{1i}{x}_{1i}\right)+b{w}_2\left({\beta }_{2i}{x}_{2i}\right)+\dots +b{w}_k\left({\beta }_{ki}{x}_{ki}\right)+{\epsilon }_i$$

(1)

where bw0, ..., bwk represent the optimized bandwidths for each variable, yi denotes the dependent variable at location i, β0i is the local intercept, xki refers to the observed value of the k-th variable, and εi is the random error term at location (i = 1, 2, 3, ..., n).

3. Results

3.1. Spatial Distribution Characteristics of AI Education Sentiment Attitudesn

In the context of sentiment analysis, the results of this study align with previous research, showing that users with positive emotions generally outnumber those with negative emotions [32,33]. The gap between positive and negative sentiments remains roughly three times larger in all 31 administrative regions. This suggests that users on major Chinese social media platforms generally exhibit positive reactions to the application of AI in education.

In terms of regional performance, the calculation of the imbalance index (S) (

Table 2. The imbalance index and Moran’s index of emotional attitude in AI education.

	Positive Emotional Attitude	Negative Emotional Attitude
S	0.26	0.29
Moran’s I (Z-Score)	1.99	2.29
Gi* (Z-Score)	2.33	2.35
Spatial Distribution Type	High-value aggregation	High-value aggregation

) shows that China’s positive and negative sentiment attitudes towards AI education are 0.26 and 0.29, respectively. This indicates significant regional disparities in both positive and negative sentiments towards AI in education, with the regional variation in negative sentiment being slightly larger than that of positive sentiment. To further explore this disparity, Moran’s I and Getis-Ord Gi* indices were introduced to measure whether the two sentiment attitudes exhibit global and local spatial autocorrelation, respectively.

Moran’s I is used to assess the overall pattern and trend of the data. If the values in the dataset tend to cluster spatially (i.e., high values cluster near other high values, and low values cluster near other low values), the Moran’s I index will be positive. Getis-Ord Gi* is used to assess each element within the context of neighboring features. When global spatial autocorrelation is not significant, local spatial autocorrelation can be identified by pinpointing locations where anomalies or strong influence points may be hidden.

The Moran’s I indices for China’s positive and negative sentiment attitudes towards AI education are 0.15 and 0.17, respectively, both greater than 0, with Z-Score results passing the significance test. This indicates that, from a spatial proximity perspective, both positive and negative sentiment attitudes exhibit imbalance and spatial aggregation, further confirming the spatial clustering of sentiment attitudes towards AI education in China.

The results from Getis-Ord Gi* detection show that this spatial clustering is characterized by high-value aggregation, meaning that regions with high positive sentiment scores are likely to have neighboring regions with similarly high scores. Additionally, in contrast to the overall sentiment trend, negative sentiment demonstrates more significant spatial autocorrelation than positive sentiment. This suggests that negative sentiment is more prone to diffusion and spread between groups and regions compared to positive sentiment.

The research results above reveal that sentiment attitudes towards AI education in China exhibit significant spatial clustering and regional differences. To present the distribution characteristics of sentiment attitudes across different administrative regions more intuitively, this study further employs visual methods to display the positive and negative sentiment attitudes of AI education across the 31 provinces of China through visual maps, as shown in Figure 1 (Positive Sentiment) and Figure 2 (Negative Sentiment). In the visual maps, warmer colors (such as red and orange) represent higher emotional intensity, while cooler colors (such as blue) indicate lower emotional intensity.

From the figures, we can observe significant spatial differences in sentiment attitudes towards AI education across the 31 administrative regions. Specifically, the eastern coastal regions (such as Beijing, Guangdong, Shanghai, and Zhejiang) show higher positive sentiment scores, indicating that users in these areas generally hold optimistic and supportive attitudes towards AI education functions and applications. In the central regions, such as Shanxi, Sichuan, Chongqing, and Hubei, the sentiment attitudes are also relatively positive. In contrast, some western provinces (such as Tibet, Gansu, and Qinghai) show lower positive sentiment scores, suggesting that users in these areas are more cautious and less accepting of AI education applications.

The distribution of negative sentiment exhibits a pattern that overlaps with positive sentiment in some areas but is more complex. Certain regions in northeast China (such as Heilongjiang and Jilin) and northwest China (such as Xinjiang) show higher negative sentiment scores. While the eastern regions have overall higher positive sentiment, some local areas also exhibit higher negative sentiment, indicating critical attitudes towards AI education applications in these regions. This may reflect a higher level of attention and discussion regarding the application of AI technology in these areas. However, despite the challenges, the overall intensity of positive sentiment remains significantly higher than negative sentiment, suggesting that the general attitude towards AI education in these areas remains optimistic.

The spatial distribution results of sentiment attitudes further confirm the findings from the imbalance index (S) and spatial autocorrelation analysis. In China, the eastern regions show the highest level of attention and discussion regarding AI education applications, followed by the central regions, with the western regions showing the least attention. Moreover, the spatial clustering pattern of sentiment attitudes is characterized by high-value aggregation, indicating that areas with stronger emotional intensity tend to influence their neighboring regions, forming a “word-of-mouth effect” or “homogenization” phenomenon. This leads to a stronger diffusion effect, resulting in spatial clustering.

The regional commonality of AI education sentiment attitudes in China may be shaped by similar environmental factors. Under the influence of factors such as similar levels of economic development, resource distribution, and informationization, different regions have formed relatively consistent attitudes towards AI education applications. However, this remains a hypothesis, and further attention should be paid to the socio-economic and environmental characteristics of these regions. Through empirical research, the study aims to further clarify any unexplained potential driving mechanisms.

3.2. Factors Influencing AI Education Sentiment Attitudes via Geo-Detector

To explore the extent to which different factors influence the sentiment attitudes of Chinese users towards AI education applications, 13 indicators were selected for this study, including economic development level (such as per capita disposable income, nighttime light index, urbanization rate, and population density), information infrastructure (such as educational funding, number of internet users, and local government focus on the digital economy), and technological innovation capabilities (such as patent grants, number of information service professionals, technical contract transaction amounts, and the proportion of information technology and software service revenue in GDP). These indicators were analyzed to assess their correlation with sentiment attitudes (q-values), revealing the impact of each factor on sentiment attitudes (

Table 3. Influencing factors of emotional attitude in AI education.

Indicator Code	Name of the Indicator	q of Positive Emotion	q of Negative Emotion
F4	Population Density	0.52	0.51
F2	Urbanization Rate	0.52	0.50
F1	Nighttime Light Index	0.43	0.41
F7	Per Capita Income of Residents	0.44	0.41
F8	Technology and Software Service Revenue	0.42	0.41
F6	Technology Transfer Amount	0.41	0.33
F10	Patent Authorization	0.24	0.25
F3	Number of Information Service Enterprises	0.18	0.15
F12	Government Report Statistical Economic Indicators	0.11	0.16
F13	Education Expenses	0.14	0.13
F5	Mobile Internet Users	0.08	0.08
F11	Internet Broadband Users	0.06	0.08
F9	Rural Broadband Users	0.03	0.03

).

The data from the table show that the most influential factors on sentiment attitudes are per capita disposable income (F4, q = 0.52, 0.51), urbanization rate (F2, q = 0.52, 0.50), and population density (F1, q = 0.43, 0.43). This suggests that individuals in economically developed, highly urbanized, and socially active regions tend to pay more attention to the development of AI education. Additionally, the key influencing indicators include the number of patent grants (F7), the number of information service professionals (F8), and the technical contract transaction amount (F6), which indicate that regions with active technological innovation are more supportive of AI education. Insufficient technological investment may result in the public having an incomplete understanding of the positive impacts of AI education, thus influencing sentiment attitudes. Overall, regions with high comprehensive development levels and active information service industries show greater attention and acceptance of AI education, which aligns with the characteristics of high development vitality and rich technological environments in the eastern coastal regions [34].

Other factors, such as mobile internet users (F11) and rural broadband users (F13), have lower q-values, suggesting that these factors have a weaker influence on sentiment attitudes. Additionally, the frequency of digital economy-related terms in government reports (F9) has a q-value of 0.03, indicating that policy promotion has limited direct influence on public sentiment attitudes. However, its impact may be exerted through indirect factors, such as economic development or technological progress.

Furthermore, within the factors themselves, there is no significant difference between the two sentiment attitudes. In the eastern regions, despite having higher positive sentiment, there are also higher levels of negative sentiment (e.g., Guangdong, Shanghai). This may be due to the rapid spread of technological and educational resources, which generates more attention and discussion on potential negative impacts. On the other hand, the western regions show lower levels of negative sentiment, but this is not necessarily due to a lack of technological acceptance. Instead, it may be because the public has less awareness and discussion of AI education, leading to overall lower emotional intensity.

3.3. Factors Influencing AI Education Sentiment Attitudes via MGWR

This study employs a multiscale geographically weighted regression (MGWR) with an adaptive bandwidth optimized using AICc, iterated 200 times until convergence (termination criterion: 1.0 × 10⁻⁵). Based on 13 standardized variables, the model analyzes the positive and negative emotional attitudes of the public across 31 observation points, exploring spatial heterogeneity and differences in driving mechanisms to reveal the spatial distribution characteristics of emotions.

The global regression results indicate that the MGWR model explains 96.7% of the variance, demonstrating a high goodness of fit (R² = 0.967, adjusted R² = 0.946). This suggests that the model effectively captures the relationships between variables. Additionally, the degree of dependence (DoD) is close to 1, indicating that the fitted model effectively explains the spatial dependence of emotional attitudes, validating its advantage in capturing spatial non-stationarity.

Bandwidth analysis results show that in the positive emotion model, except for F6 (bandwidth: 712 km, ENP = 4.059, β range [−0.102, 0.482], standard deviation 0.129), other variables such as F2, F8, and F7 have bandwidths exceeding 9976 km, with ENP values close to 1 and minimal parameter variation (e.g., F8 β = 0.673), exhibiting a global effect. In the negative emotion model, the bandwidth of F6 slightly increases to 729 km (ENP = 3.917, β range [−0.138, 0.341], standard deviation 0.141), still showing significant local variation. Meanwhile, F5 (bandwidth: 4417 km, β range [0.055, 0.061]) also reveals some spatial differences, whereas other variables maintain bandwidths exceeding 9975 km (e.g., F8 β = 0.702), exhibiting nearly uniform effects.

In comparison, F8 serves as the strongest positive factor in both emotion models (positive: β = 0.797; negative: β = 0.854), while F7 consistently acts as a strong negative factor (positive: β = −0.471; negative: β = −0.617). However, the effects of F4 and F6 exhibit significant differences: F4 is not significant for positive emotions but has a strong positive effect on negative emotions (β = 0.508, p = 0.002). In contrast, F2 and F6 are significant for positive emotions (β = 0.366, p < 0.001; β = 0.252, p = 0.017) but show weakened effects on negative emotions (β = 0.175, p = 0.062; β = 0.147, p = 0.170), highlighting their prominent local variability in both emotion models. Other variables, such as F3, F5, and F9–F13, do not exhibit significant effects in either model.

Furthermore, after refining the model at a local scale, the total residual sum of squares decreases to 0.577, and the coefficient of determination (R²) improves to 0.983. This further confirms that accounting for spatial locality significantly enhances the explanatory power of the model. Most variables, such as F2, F7, and F8, have bandwidths close to approximately 9976 km, suggesting that their influence remains relatively consistent over a large spatial extent. However, the bandwidths of F5 and F6 decrease to approximately 4417 km and 729 km, respectively, with a relatively high number of effective parameters (ENP = 4.059), indicating significant spatial variation in their effects across different regions.

4. Discussion

4.1. Research Conclusions

This study, through the integration of sentiment analysis and spatial analysis, reveals the regional distribution characteristics and spatial clustering patterns of public sentiment towards AI education applications in China. Specifically, the findings are as follows:

• Dominance of Positive Sentiment: In the 31 provincial-level administrative regions, positive sentiment significantly outweighs negative sentiment, maintaining a roughly three-fold difference. This indicates that the public generally holds an optimistic and supportive attitude towards the functionality and application of AI education technologies.
• Regional Variations in Sentiment: The imbalance index (S) and spatial distribution maps reveal significant regional differences in sentiment towards AI education. The eastern coastal regions (e.g., Beijing, Shanghai, Guangdong) show significantly higher positive sentiment scores compared to the central and western regions, while the western areas (e.g., Tibet, Gansu, Qinghai) exhibit lower sentiment scores.
• Spatial Diffusion of Sentiment: Moran’s I and Getis-Ord Gi* indices show that sentiment attitudes display a high-value clustering characteristic in the eastern coastal regions. These areas, with higher urban development levels, concentrate positive sentiment and spread to neighboring regions. Negative sentiment, in particular, exhibits stronger significance in terms of spatial diffusion.
• Influence of Regional Comprehensive Resources and Practical Experience: The comprehensive development level of a region serves as a fundamental factor influencing sentiment towards AI education. In economically developed and technology-rich regions, where practical applications of AI are more frequent, the public is more likely to develop trust in the technology and provide positive feedback. In contrast, areas with fewer resources and insufficient understanding tend to foster more negative sentiment due to the lack of adequate exposure and knowledge.

4.2. Distribution Patterns of Sentiment Attitudes

The study reveals that the public in China generally holds a positive emotional attitude towards the application of AI in education, but this attitude varies significantly across regions. The eastern coastal regions (e.g., Beijing, Guangdong, Shanghai, Zhejiang), characterized by high economic development, advanced information technology, and active technological innovation, exhibit consistently high positive sentiment scores. This result aligns with previous research indicating that economic and technological development levels enhance the acceptance of new technologies [35].

In the central regions (e.g., Shanxi, Sichuan, Hubei), although the development level is somewhat lower compared to the eastern regions, positive sentiment scores remain relatively high. This could be attributed to the rapid progress in informatization in recent years and increased local government investments in emerging technologies [36]. In contrast, the western regions (e.g., Tibet, Gansu, Qinghai) show lower positive sentiment scores, reflecting not only the inadequacy of economic development and technological dissemination but also potential limitations in public understanding of AI education. These areas’ public attitudes towards AI education are likely constrained by imbalanced resource allocation and insufficient technological application scenarios [34,37].

The regional differences in sentiment attitudes reflect varying levels of public understanding and discussion of AI education. The spatial distribution of negative sentiment shows a partially overlapping regional pattern with positive sentiment. Despite the dominance of positive sentiment in the eastern regions, certain provinces (e.g., Guangdong, Shanghai) exhibit relatively high levels of negative sentiment. The coexistence of high positive and certain high negative sentiments in the eastern regions indicates that public attitudes towards AI education are not merely supportive or opposed; instead, they reflect both recognition and anticipation of new technology as well as concerns about potential risks, showcasing a multifaceted and developmental approach. In the central and western regions, both positive and negative sentiments are weak, indicating insufficient public understanding and discussion of AI education. Therefore, the lower intensity of sentiment in these regions does not simply reflect negativity but rather a lack of engagement and discourse surrounding AI education compared to the eastern regions.

The “high-value clustering” phenomenon, where high-value areas (e.g., eastern coastal regions) influence surrounding areas, shows the interregional correlation of public sentiment. The hypothesis that public sentiment towards AI education has spatial diffusion properties is confirmed in this study. This diffusion pattern suggests that sentiment attitudes are more likely to resonate regionally through group interactions, reflecting the geographical proximity effect of similar social environments on regional development [38]. Additionally, negative sentiment demonstrates stronger spatial autocorrelation compared to positive sentiment, indicating that negative emotions spread and diffuse more effectively [9]

Social media platforms play a crucial role in this process. The herd mentality and emotional mimicry mechanisms amplify this effect, as individuals tend to align with and propagate negative views in the face of intense negative discussions within groups. This leads to regional emotional resonance [39,40]. Social media platforms’ algorithms tend to focus on highly discussed topics, and through multi-layered user interactions and information sharing, controversial negative emotions trigger stronger emotional resonance and widespread dissemination [41]. This phenomenon is particularly pronounced when the application of technology is controversial (e.g., ethics, fairness issues), highlighting the need to pay special attention to the risk of negative emotional diffusion when promoting AI education [42].

4.3. Factors Influencing of Sentiment Attitudes

This study employs the Multiscale Geographically Weighted Regression (MGWR) model and the Geo-Detector to thoroughly investigate the spatial heterogeneity and driving mechanisms of Chinese public sentiment towards AI technology applications.

4.3.1. Discussion Based on the Geo-Detector

The results of the Geo-Detector analysis reveal that urban development levels (such as per capita disposable income, urbanization rate, and population density) are core variables affecting public sentiment towards AI education. In regions with higher income and more advanced modernization, the public tends to exhibit higher levels of positive sentiment. This is closely related to their openness to new technological applications, their stronger ability to adapt to new changes, and their higher capacity to handle technological transformations. Technological innovation resources (such as the number of patents granted and the number of people employed in information services) also play an important role in shaping sentiment attitudes [20]. High sentiment-value regions are often accompanied by stronger community support, increased public interaction, and the sharing of experiences, which in turn creates a positive public opinion atmosphere. This attracts more users to participate in technology applications and develop positive experiences. Areas rich in technological innovation resources have a more timely and widespread understanding of AI technology, leading to a higher sensitivity and attention to the trends of new technology applications in the local public [43].

It is important to note that informatization infrastructure (such as the number of internet users and educational funding) has a relatively weaker impact on sentiment attitudes. This result suggests that improving informatization infrastructure alone may not be sufficient to significantly alter the public’s emotional orientation towards AI education. In contrast, the combined effects of economic development and technological innovation capacity play a more critical role. Additionally, government policies related to the digital economy have limited direct impact but may exert influence indirectly through economic development and technological progress.

The differences in AI education sentiment attitudes between the eastern coastal and western regions result from the interplay of multiple factors. The eastern regions typically exhibit higher economic development levels, urbanization rates, informatization levels, and technological innovation capabilities, all of which contribute to a more positive recognition and acceptance of AI education. Higher levels of urban development mean stronger technology acceptance, a smaller digital divide, and greater convenience in accessing information and services. The public in these areas has better access to information about AI education, fostering a favorable social interaction atmosphere that facilitates acceptance of new technology. In contrast, the lower sentiment intensity in the western regions reflects insufficient public understanding and discussion of AI education. This lack of recognition and discussion likely stems from limited channels for information acquisition. In areas with lower urban development levels, public exposure to emerging technologies is minimal, and local media coverage of relevant topics is scarce, leading to a lack of discussion spaces. This lack of awareness is a key factor in the formation of a relatively low emotional expression state, rather than complete rejection of technology [44].

4.3.2. Discussion Based on the MGWR

From the perspective of spatial heterogeneity in the MGWR model analysis, the driving mechanisms of positive and negative emotions exhibit significant differences. In the positive emotion model, except for technology transfer amount (F6), which shows local variation (bandwidth = 712 km, ENP = 4.059), other variables such as urbanization rate (F2), per capita disposable income (F7), and revenue from technology and software services (F8) have globally consistent effects on positive emotions. In contrast, in the negative emotion model, the bandwidth of technology transfer amount (F6) slightly increases to 729 km, while the bandwidth of mobile internet users (F5) is 4417 km, indicating a certain degree of local variation. This difference suggests that the formation of negative emotions is more susceptible to region-specific factors, such as the localized impact of technology transfer or the uneven distribution of internet access. Urbanization rate (F2) and technology transfer amount (F6) play significant roles in fostering positive emotions, while their effects diminish in the negative emotion model. This finding suggests that highly urbanized regions with higher levels of technology transfer, driven by two major socioeconomic transformation forces—technological innovation and urbanization—are more inclined to promote the widespread adoption of AI in education through positive effects such as improved infrastructure, increased employment opportunities, and enhanced convenience.

Notably, revenue from technology and software services (F8) emerges as the strongest positive driving factor in both emotion models, whereas per capita disposable income (F7) serves as a strong negative factor. This finding highlights the dual influence of economic development on emotional attitudes. The high positive effect of F8 suggests that the growth of technology and software service revenue not only reflects the modernization of regional economic structures but may also directly enhance public positive emotional experiences by increasing employment opportunities, improving living convenience, and fostering social recognition. Its global bandwidth (>9975 km) further indicates the widespread and consistent nature of this effect. The stable and significant role of F8 in both emotion models suggests that the prosperity of the technology industry can both reinforce public recognition of the benefits of AI applications in education and, to some extent, raise concerns about its potential risks.

The strong negative effect of F7 may reflect a form of “relative deprivation” or increased life pressure associated with economic growth. In particular, at the spatial scale, rising regional per capita income is often accompanied by higher living costs and intensified competition, with this negative association being more pronounced in the negative emotion model (β = −0.617). This suggests that the benefits of economic development have not been uniformly translated into emotional well-being. Additionally, population density (F4) has no significant impact on positive emotions but shows a strong positive effect on negative emotions (β = 0.508, p = 0.002), implying that high-density areas may be more prone to technology-related anxiety due to resource competition and disparities in technology accessibility.

4.4. Driving Mechanisms of Sentiment Attitudes

Regions with strong emotional attitudes (whether positive or negative) exert diffusion effects on surrounding areas through social networks and regional linkage mechanisms. The spatial diffusion of negative emotions is significantly stronger than that of positive emotions. The spatial spread of negative emotions is intertwined with multiple factors. As the core platform for information dissemination, social media plays a crucial role in the diffusion of negative emotions through its hotspot effect and user interaction mechanisms. Recommendation algorithms on these platforms, utilizing collaborative filtering and deep learning models, prioritize highly engaging content. Due to their higher arousal value (e.g., anger and anxiety), negative emotional content typically generates higher click-through rates and longer dwell times than neutral content. Moreover, sentiment dictionary algorithms on these platforms actively reinforce extreme labels (e.g., #AIUnemployment). Emotionally intense disclosures are more likely to be flagged by algorithms as “high-potential hotspots”, receiving higher content exposure weights. This algorithmic mechanism leads users to be continuously exposed to similar negative information, trapping them in a “negative information cocoon”, which forms an emotional resonance loop and a self-reinforcing transmission chain within social networks.

Through shared narratives (such as common stories, experiences, or events), groups concretize vague emotions, making individual emotional experiences more specific and distinct. For instance, an individual’s negative sentiment towards AI in education may initially be vague or uncertain. However, through discussions on social media, particularly the use of hashtags (#), these emotions become collectively expressed and clearly defined. When users adopt specific hashtags, they are not only expressing personal emotions but also integrating them into a broader group discourse. This process of concretization not only gives form to emotions but also fosters emotional resonance among group members, leading to unified emotional expressions.

Additionally, the adherence to collectivist cultural values amplifies psychological effects such as conformity, social norms, and social identity needs [45]. When individuals observe a widely recognized emotional state among their peers, they develop a subconscious motivation to align their own cognition with that of the group, in order to avoid the sense of exclusion and social pressure associated with being an “outlier” [46]. For example, the public may shift from a neutral or even positive attitude towards AI technology after witnessing a prevalent negative perception among those around them. This results in broader emotional transmission and attitude reinforcement, leading to regionally synchronized emotional diffusion. Furthermore, Hofstede’s cultural dimension theory suggests that in high power distance cultures (such as China), the public tends to rely more on authoritative interpretations when attributing technological risks [47]. In cases where there is a policy vacuum (e.g., delays in AI regulatory frameworks), negative emotions may spread more rapidly due to institutional trust deficits [48,49].

Regions with strong sentiment (whether positive or negative) exert a diffusion effect on neighboring areas through social networks and regional linkage mechanisms. The spatial diffusion of negative sentiment is significantly stronger than that of positive sentiment, which may be related to the characteristics of social media dissemination [50]. The diffusion of negative emotions towards AI education reflects a high level of public attention to technological applications, but it could also be driven by critical attitudes towards technology in regions with unequal resource distribution. This study further reveals the significant moderating effect of regional differences on sentiment attitudes, a topic that has been rarely discussed in the existing literature [51,52]. Previous studies have largely focused on individual-level factors of technology acceptance, while this research emphasizes the regional differences and spatial diffusion patterns at the macro level. This finding provides a more comprehensive perspective on understanding public sentiment towards AI technology applications and expands the research framework for sentiment attitude analysis.

4.5. Temporal Evolution of Public Sentiment and Event Tracking

This study analyzes the longitudinal sentiment data of the Chinese public towards AI applications in education from February 2023 to October 2024 (Figure 3). By integrating policy events and media reports, we further discuss the dynamic evolution of public sentiment. The findings reveal that negative events amplify risk perceptions, whereas positive events such as legislative policies and technological inclusivity can restore public sentiment. Although negative events can trigger short-term surges in negative emotions, positive sentiment remains the dominant emotional type in the long run. This trend is driven by multiple factors, including policy promotion, technological breakthroughs, and media influence.

Between February and May 2023, positive sentiment towards AI in education dominated, although negative sentiment showed a slight increase. From June to August 2023, positive sentiment experienced a significant rebound, reaching 85.7% in August. This peak was associated with several major events: in July 2023, seven government departments jointly released the “Interim Measures for the Management of Generative Artificial Intelligence Services”, implementing a prudent and classified regulatory approach to generative AI services. Additionally, the 2023 World Artificial Intelligence Conference (WAIC) was held in Shanghai, showcasing various AI-driven educational innovations.

However, in September and October 2023, negative sentiment saw a sharp increase, reaching 27.2% in October, while positive sentiment declined to 48.2%. This shift was likely driven by concerns over technological security and fairness, exacerbated by negative incidents. For example, in September 2023, Microsoft’s AI team leaked 38 TB of data, and in October 2023, China’s Internet Joint Rumor-Refuting Platform publicly addressed AI-generated misinformation cases.

In the first half of 2024, positive sentiment gradually rebounded and remained above 60%, indicating overall stability and a predominance of positive sentiment. Notably, in July and August 2024, positive sentiment surged again to 77.0% and 89.0%, respectively. Several key policy and regulatory developments contributed to this trend. In January 2024, the State Council’s executive meeting emphasized leveraging AI to enhance key industries, supporting the construction of a manufacturing powerhouse, a cyber powerhouse, and a digital China. In February 2024, the National Information Security Standardization Technical Committee released the “Basic Security Requirements for Generative Artificial Intelligence Services”, addressing long-term risks such as AI-induced cybersecurity threats and misinformation. Additionally, the Ministry of Education announced 184 AI education pilot bases for primary and secondary schools.

Further reinforcing this positive shift, in July 2024, the WAIC was held again, accompanied by the joint release of the “Generative AI Education Application Security Guidelines” by the Ministry of Education and the Cyberspace Administration. Moreover, on 3 April 2024, the National Cybersecurity Committee issued the “Information Security Technology: Security Standards for Pre-training and Fine-tuning Data in Generative AI”, helping to mitigate the impact of previous negative incidents.

The achievements showcased at the World Artificial Intelligence Conference, the introduction of generative AI management standards, the promotion of AI technology accessibility cases, and the directional signals from national policies—widely disseminated through mainstream media—have collectively created a positive emotional communication environment. This environment has facilitated the enhancement and stabilization of positive sentiment.

The dynamic evolution of public sentiment not only reflects optimism about the future development of AI technology but also validates the role of “perceived usefulness” and “perceived ease of use” in the Technology Acceptance Model (TAM). Specifically, policy safeguards reduce perceptions of technological risk, real-world application cases strengthen recognition of AI’s utility, and media dissemination achieves cognitive transformation through emotional mobilization.

This finding provides a new perspective for understanding sentiment-driven mechanisms and offers valuable policy insights for the future promotion of AI technology. The combination of technological breakthroughs and policy safeguards, supplemented by effective media campaigns to showcase AI’s societal value, is key to enhancing public emotional acceptance. At the same time, the temporary surges in negative sentiment highlight the importance of integrating public opinion monitoring with macro policies and technological events at different stages of AI development. Relevant authorities should proactively implement risk warnings, interventions, governance, and regulatory measures to prevent the further escalation of adverse public sentiment.

4.6. Differentiated Policy Design and Emotion Dissemination Governance: Enhancing the Acceptance of AI in Education

The promotion and application of AI technology in education are influenced by multiple factors, including regional economic levels, technological innovation capabilities, and public attitudes towards AI, all of which exhibit significant disparities. Therefore, formulating region-specific policies is crucial to improving AI education acceptance.

Geo-Detector analysis indicates that overall infrastructure development and technology-related variables (e.g., F8) significantly explain variations in public sentiment towards AI. MGWR analysis further confirms that these variables have a strong global positive effect. Based on this, policymakers should prioritize addressing public concerns in active regions by reinforcing positive narratives about the value and societal benefits of AI in education.

At the same time, MGWR reveals the localized effects of technology innovation-related variables (e.g., F6), indicating that the impact of technological advancements on sentiment varies across regions. This finding suggests that policy design should fully consider regional disparities and adopt flexible, differentiated implementation strategies, rather than a one-size-fits-all approach. In regions where technological innovation has a significant influence, targeted investments should be increased to maximize its positive effects.

In technology-active coastal regions, AI education technologies should be further integrated into local education systems. Hosting AI education forums and workshops can enhance public awareness and foster a supportive environment for AI adoption. In central regions, efforts should focus on improving AI literacy among teachers and students, ensuring the practical utility of AI applications, and gradually expanding AI adoption in education. In western regions, financial support from the government and private enterprises should be leveraged to provide free or low-cost AI education tools and software. Remote education technologies can help bridge the digital divide and provide equitable access to AI-powered learning resources.

Overall, policy guidance and regulation should avoid rigid “unified standards” or “one-size-fits-all” approaches. Instead, ensuring that AI technology yields positive benefits across all regions and progressively reduces regional disparities is essential to addressing existing imbalances in resource allocation and development.

In addition to differentiated policy design, the government can integrate public opinion platform data to monitor emotion dissemination patterns. Using spatial autocorrelation analysis, authorities can identify sentiment diffusion hotspots (e.g., regions where Moran’s I > 0.4) and develop early warning models to prevent negative sentiment outbreaks. For example, if negative sentiment exceeds **30%** for three consecutive days in a given region, rational discussion content (e.g., expert analysis videos) could be algorithmically prioritized to guide public discourse towards a more constructive direction.

By implementing such a spatially differentiated policy framework, regional development dynamics can be respected while leveraging sentiment dissemination mechanisms for positive reinforcement. It is recommended to pilot this strategy in the Chengdu-Chongqing region, which serves as a transitional zone between eastern and western China. Monitoring how policy interventions reshape public sentiment in this area can provide empirical insights for nationwide AI policy expansion and widespread adoption.

5. Contributions

As artificial intelligence (AI) technology continues to be applied in the education sector, understanding public perceptions and emotional responses has become increasingly important. This study addresses a gap in the literature by investigating the regional differences and influencing factors of sentiment attitudes towards AI in education. The data collected from major Chinese social media platforms over a year provides a deeper insight into how user groups from different regions perceive this technology, particularly because most existing studies treat sentiment data as a whole, rather than broken down by region. Therefore, this study offers a unique opportunity to combine sentiment analysis with spatial analysis, enabling macro-regional monitoring and analysis, and offering a more comprehensive examination of user opinions across various social media platforms.

The study highlights the significant moderating role of regional factors in sentiment attitudes, which has not been sufficiently discussed in existing research. Additionally, the study uncovers a new relationship between residents’ income levels and sentiment attitudes, emphasizing the role of the socio-economic environment in shaping public perceptions of AI applications.

In this study, the SnowNLP algorithm is first applied to classify online comments from various provinces in China into positive, neutral, and negative categories, providing foundational data for subsequent spatial analysis. GIS technology is then used to visualize the spatial distribution characteristics of positive sentiment, reflecting regional differences in public sentiment. Moran’s I and Getis-Ord Gi* are employed to test the spatial autocorrelation of both positive and negative emotions, revealing patterns of emotional clustering. Furthermore, a multi-index spatial weighted regression model and geographical detector are used to systematically analyze the impact of factors such as digital economy development, urban vitality, policy attention, digital literacy, and infrastructure development on the distribution of positive sentiment, exploring their underlying driving mechanisms.

This study employs two analytical methods—Geo-Detector and Multiscale Geographically Weighted Regression (MGWR)—to examine the influencing mechanisms of public sentiment towards AI. The Geo-Detector method quantifies the explanatory power of various factors (e.g., per capita disposable income, urbanization rate, and number of authorized patents) on sentiment attitudes by computing the q value, identifying economic development level and technological innovation capacity as the primary driving factors.

The MGWR analysis not only validates the global effects of these variables based on the Geo-Detector results but also further identifies their local spatial influence and variation in intensity across regions. This approach provides a refined understanding of how different factors contribute to sentiment formation in diverse geographic contexts.

From a theoretical perspective, the socio-technical systems theory posits that the introduction and effective operation of new technologies are not merely technical issues but also involve the coordinated optimization of human, organizational, and societal factors [53]. This study corroborates the existing theoretical perspective that technology, society, and individuals interact and mutually influence one another. Moreover, it reveals significant spatial heterogeneity in the effects of influencing variables across different regions, emphasizing the importance of spatial nonstationarity in sentiment analysis.

Ultimately, the integration of these two methods provides mutual validation and effective complementarity, offering a multidimensional perspective from both impact intensity and spatial heterogeneity. This joint approach enhances the understanding of the spatial distribution patterns and formation mechanisms of public sentiment towards AI at the provincial level, providing robust empirical support for further research and policymaking.

6. Limitations

By integrating Multiscale Geographically Weighted Regression (MGWR) and Geo-Detector, this study successfully unveils the spatial distribution characteristics of public sentiment at the provincial level and its association with socio-economic factors, providing a solid quantitative foundation for understanding the acceptance of AI in education. However, several limitations remain, warranting further refinement in future research.

Although the spatial analysis framework and driving mechanism exploration employed in this study exhibit strong theoretical generalizability and can serve as a methodological reference for other countries—especially in examining the relationship between technology adoption and sentiment—the cultural dimensions theory suggests that power distance and collectivist cultural traits are critical factors in the study of social behavior and governance. Therefore, it is important to acknowledge that this study is grounded in China’s socio-political environment, where technological advancements shape public sentiment and its spatial distribution under strong governmental influence. Given China’s highly centralized political system and rapid economic transformation, public perceptions are often significantly shaped by government-led policy initiatives (e.g., the enactment of relevant legislation) and mainstream official media narratives. In contrast, in culturally diverse or more individualistic societies (e.g., Western democracies), public acceptance of AI in education may be driven more by concerns over privacy ethics and technological fairness, rather than direct policy advocacy. This suggests that the evolution patterns and sentiment distribution may follow different trajectories across regions. Future studies should incorporate cross-cultural data from multiple countries to examine moderating effects of cultural, economic, and technological contexts, thereby testing the model’s adaptability and boundary conditions and deepening the understanding of AI’s global impact.

Second, this study captures regional sentiment variations and key driving factors based on a limited temporal scope, focusing primarily on cross-sectional spatial analysis rather than longitudinal changes in public sentiment. The static nature of this design restricts insights into temporal trends and diffusion patterns of emotions. For instance, public acceptance of AI in education may fluctuate with technological advancements or policy shifts, and the limited timeframe of the dataset may introduce potential biases, affecting the interpretation of long-term trends. Additionally, while the study aggregates sentiment scores from multiple platforms, it does not provide a comparative analysis across platforms, nor does it differentiate specific AI education applications (e.g., personalized learning, intelligent tutoring, or automated assessment). Although many scholars recognize the value of social media data in the absence of large-scale, high-quality datasets, such data still fail to fully represent all social groups, particularly those with weaker socio-economic and technological capital, thus limiting the generalizability of the findings. The inherent biases in the dataset should therefore be acknowledged as a potential concern.

Furthermore, while this study primarily employs quantitative analysis to quantify spatial effects with high explanatory power, it lacks qualitative components (e.g., thematic analysis of social media comments or case studies). Future research could integrate qualitative coding techniques to further investigate the underlying reasons behind specific public sentiments (e.g., concerns about AI ethics).

This spatial-statistics-centered approach establishes a foundation for understanding the spatial patterns and evolution of public sentiment towards AI in education. However, these limitations suggest that the interpretation of findings should consider the study’s temporal scope, cultural context, and technological application scenarios. Future research should adopt a more comprehensive longitudinal, multidimensional, and domain-specific approach, incorporating qualitative analysis (e.g., thematic extraction or case coding) to enrich quantitative findings and enhance the generalizability of the results.