The Gradual Cyclical Process in Adaptive Gamified Learning: Generative Mechanisms for Motivational Transformation, Cognitive Advancement, and Knowledge Construction Strategy

Abstract

The integration of gamification into digital learning environments is reshaping educational models, advancing towards more adaptive and personalized teaching evolution. However, within large Chinese corpora, the transition mechanism from passive participation to adaptive gamified learning remains underexplored in a systematic manner. This study fills this gap by utilizing LDA topic modeling and sentiment analysis techniques to delve into user comment data on the Bilibili platform. The results extract five major themes, which include multilingual task-driven learning, early-age programming thinking cultivation, modular English competency certification, cross-domain cognitive integration and psychological safety, as well as ubiquitous intelligent educational environments. The analysis reveals that most themes exhibit highly positive emotions, particularly in applications for early childhood education, while learning models that involve certification mechanisms and technological dependencies tend to provoke emotional fluctuations. Nevertheless, learners still experience certain challenges and pressures when faced with frequent cognitive tasks. In an innovative manner, this study proposes a theoretical framework based on Self-Determination Theory and Connectivism to analyze how motivation satisfaction drives cognitive restructuring, thereby facilitating the process of adaptive learning. This model demonstrates the evolutionary logic of learners’ cross-disciplinary knowledge integration and metacognitive strategy optimization, providing empirical support for the gamification learning transformation mechanism in China’s digital education sector and extending the research framework for personalized teaching and self-regulation in educational technology.

1. Introduction

China’s educational culture has traditionally focused on exam-oriented education, emphasizing the transmission of knowledge and mastery of skills. Under the influence of the “Double Reduction Policy” and the promotion of quality education, the limitations of traditional education models have gradually become apparent, particularly in terms of fostering students’ initiative and innovative abilities. Gamified learning, as a new approach that integrates technology and teaching, is gradually reshaping educational models and driving education toward a more adaptive and personalized direction [1]. With its ability to break the traditional constraints of time and space in teaching [2], as well as enhance interaction and immersion in learning [3], gamified learning is constructing a flexible, efficient, and engaging learning environment. A growing body of research has shown that gamified teaching can significantly stimulate learners’ intrinsic motivation [4,5], improving their autonomy in learning and learning outcomes [6,7]. However, despite the positive results brought about by gamified education, learners still face numerous challenges in the adaptive gamified learning process, particularly regarding the transition from passive participation to active adaptive learning [8]. This has attracted extensive attention and in-depth research from both scholars and educational practitioners [9,10].

Adaptive gamified learning, by combining gamification elements with personalized learning pathways, can dynamically adjust learning content and methods based on learners’ progress, interests, and needs, thereby enhancing the learning experience and outcomes [11]. In this process, learners not only enjoy the interaction and immersion brought by gamification, but also achieve self-regulated and goal-driven learning progress within personalized learning pathways [12]. The core of adaptive gamified learning lies in the integration of external motivational mechanisms from traditional gamified learning (such as points, badges, and leaderboards) with personalized learning systems [13,14]. Supported by big data and artificial intelligence technologies, it analyzes learners’ learning data in real-time and dynamically adjusts task difficulty and content [15], making the teaching process more flexible and personalized [16]. This transformation focuses not only on learners’ knowledge acquisition and skill enhancement but also emphasizes their psychological well-being and active engagement [17], reflecting a close integration of technology and educational philosophy.

In recent years, research on adaptive gamified learning has primarily focused on system function design, optimization of motivational mechanisms, and customization of learning pathways. For example, the role of personalized recommendation algorithms in improving learning efficiency [18], the positive impact of long-term motivational mechanisms on learning behaviors [19,20], and the proposal of embedding emotion recognition and behavioral feedback mechanisms into adaptive systems [21] to enhance learner engagement and interactivity. In the context of China’s exam-oriented education system, students are often accustomed to relying on teachers’ guidance and traditional learning methods, making it difficult for them to autonomously adjust learning strategies and manage learning progress [22]. Therefore, how to assist students in achieving cognitive development and psychological motivation mechanisms during the transition from “passively accepting game rules” to “actively adjusting learning strategies” within the unique educational and cultural context of China has become an urgent issue for both scholars and educational practitioners [23,24].

Moreover, the existing literature generally lacks an in-depth analysis of how adaptive gamified learners construct self-learning goals and motivation maintenance strategies, and has not developed a structural model to characterize the evolution of their motivation and self-regulation mechanisms [25,26]. This research gap limits educators’ systematic understanding of learners’ behavioral characteristics and the effectiveness of system design. Therefore, there is an urgent need to explore the intrinsic mechanisms and evolutionary paths of adaptive learning behaviors from the perspective of semantic analysis and user viewpoints, using Chinese learners as an example. This study aims to address this gap by combining corpus mining and topic modeling methods to develop an integrated cognitive framework of motivation, thereby deepening the theoretical foundation and practical guidance for adaptive gamified learning.

At the theoretical level, adaptive gamified learning demonstrates the potential for a deep integration of technology and educational philosophy in enhancing learning motivation and promoting knowledge construction [27,28]. Adaptive gamified learning aligns closely with Self-Determination Theory (SDT), which emphasizes three basic psychological needs: Autonomy, Competence, and Relatedness [29]. By providing a flexible learning environment, adaptive gamified learning allows learners to autonomously choose their learning pathways, thereby enhancing their sense of achievement and social connections, and improving the proactivity and sustainability of learning [30]. At the same time, Connectivism provides important theoretical support for knowledge construction [31]. In an adaptive gamified learning environment, learners do not simply master existing knowledge but instead build interdisciplinary information networks, extracting and reorganizing meaningful knowledge from the ever-expanding learning resources [32,33]. Through interaction with diverse information sources, learning communities, and virtual contexts, learners are able to promote cross-domain knowledge integration, forming a flexible and comprehensive knowledge system [34], which fosters critical thinking, while also stimulating sustained learning motivation and the development of self-regulation skills.

Overall, this study aims to systematically analyze the underlying mechanisms through which adaptive gamified learning promotes learning motivation and cognitive development. Using a sample of learners from certain regions in China, the research explores in depth how adaptive gamified learning, combined with personalized technologies, can stimulate students’ learning motivation. Particularly in the context of the “Double Reduction” policy, the study provides strong support for achieving efficient autonomous learning through the application of intelligent technologies. These explorations offer a rich theoretical foundation and practical guidance for optimizing educational technology, implementing personalized learning strategies, and fostering innovation in educational technology.

2. Research Design

This study adopts a hybrid methodological design combining sentiment analysis and Latent Dirichlet Allocation (LDA) topic modeling, integrating topic extraction and sentiment recognition to mine multidimensional information embedded in the text. The research aims to explore the structural mechanisms through which adaptive gamified learning promotes learning motivation and cognitive development. Sentiment analysis is a natural language processing (NLP) technique [35] that identifies emotional attitudes in gamified education by analyzing textual data [36]. This approach can reveal the emotional tendencies and underlying intentions of individuals or groups in specific contexts, thereby providing valuable information for decision-making processes [37,38]. SnowNLP is an NLP tool specifically designed for processing Chinese text. Leveraging its superior capability in Chinese semantic understanding, it is more suitable than other sentiment recognition tools for handling data closely related to the Chinese sociocultural context. Its advantages lie in its deep insight into Chinese linguistic structures, cultural background, and expressive habits, enabling it to excel in sentiment classification tasks and accurately capture emotional tendencies and subtle fluctuations in text. Consequently, it serves as a powerful tool for analyzing and understanding emotional landscapes in China, with applications in public opinion analysis, consumer feedback, and social media commentary.

Latent Dirichlet Allocation (LDA) is a probabilistic topic modeling method widely used to extract latent thematic structures from large-scale document collections [39]. The model assumes that each document is generated from a probability distribution over a set of topics, and that each topic is represented by a probability distribution over a set of words. To estimate the model parameters from a given dataset, LDA typically employs inference methods such as variational inference or Gibbs sampling [40]. These approaches can effectively derive the topic distribution for each document and the word distribution for each topic, thereby enabling the automatic extraction of thematic structures from document corpora. This process quantifies the weight of each document across different topics, providing robust support for text analysis and understanding [41].

Based on the keywords derived from the LDA (Latent Dirichlet Allocation) topic model, this study first conducted a multi-dimensional sentiment analysis of the topics to gain a deeper understanding of the public’s emotional tendencies. It evaluated the emotional tendencies of each comment across different topics. The sentiment score of each comment is typically associated with its probability weight on different topics. The probability of a comment’s association with a topic is denoted as p, and the sentiment score of the comment is denoted as s. The weighted sentiment score is the product of p and s. As the number of topics increases, the range becomes more detailed, but the relevance of each comment to the corresponding topics tends to decrease. The main process is illustrated in Figure 1. This analytical method helps to comprehensively understand the public’s emotional response to each topic’s relevance and sentiment. However, this method does not effectively mitigate the dilution effect caused by multi-topic interference on the analysis results. To further reveal the core issues that the public focuses on, this study combines a sentiment analysis method centered on prominent topics. Specifically, LDA is used to extract the latent topic keywords from the semantic level of the comment texts, then the relevance between each comment and the potential topics is calculated. The topic with the highest relevance is considered the dominant topic, and the comments are assigned to the corresponding topics to compute the weighted sentiment score. Subsequently, the sentiment analysis focuses on the dominant topic dimension and evaluates the sentiment tendency of the commenter within that topic. The main process is illustrated in Figure 2. This method, by focusing on the dominant topic of each comment, enables a more precise evaluation of the commenter’s emotional tendency toward the dominant topic, thus enhancing the specificity of sentiment analysis. The combination of these two methods helps to comprehensively understand the emotional distribution of the public across different dimensions, avoids overlooking any potential emotional tendencies, and delves deeper into the emotional attitudes of commenters regarding the most critical topics. This approach improves the accuracy and relevance of the analysis while maintaining its general applicability.

This study constructs a systematic theoretical model based on the integrated results of LDA topic modeling and sentiment analysis. Firstly, latent topics in the text were extracted through the LDA topic modeling method, and combined with the sentiment analysis results, the emotional tendencies of different topics were thoroughly analyzed. Based on this analysis, the constructed theoretical model not only provides a theoretical framework for sentiment analysis in the field of gamified learning but also further expands the potential and practical value of sentiment analysis in the application of educational technology. As shown in Figure 3, the overall process of the study and the logical relationships between various steps are presented. This study, through a detailed examination of key research steps such as data collection, sentiment analysis, and topic modeling, forms a complete theoretical support system.

3. Methodology

3.1. Platform Selection

The goal of this study is to collect authentic and representative comment data. Initially, three major social media platforms—Xiaohongshu, Weibo, and Bilibili—were considered. However, during the data collection process from Xiaohongshu and Weibo, several challenges emerged: First, posts and comments on these platforms exhibited a high degree of repetition, limiting the diversity and representativeness of the data. Second, a significant proportion of the comments consisted of advertising information, which severely interfered with capturing authentic emotional expressions and made it difficult to accurately reflect users’ emotional feedback. Additionally, content related to adaptive gamified education was relatively scarce, complicating the collection of relevant data. These issues rendered Xiaohongshu and Weibo less applicable to this study. After conducting a comparative analysis and data screening, this study ultimately selected Bilibili as the data source.

Bilibili, as a leading online video platform in China, has developed into a broad social and entertainment space with its high interactivity and unique community culture. The active participation of users on the platform has made the comment section a core space for discussion, communication, and emotional interaction. Users not only express personal opinions when watching and evaluating content but also engage in deeper emotional interaction and multi-level discussions through comments. It is important to note that Bilibili’s user base is primarily composed of a younger demographic, whose interests, emotional expressions, and behavior patterns differ from those on other platforms. Therefore, relying solely on data from Bilibili may have certain limitations. However, considering that the main users of adaptive gamified learning also belong to this younger group, Bilibili, as an important social platform for this demographic, provides a wealth of interactive data. By analyzing this data, researchers can gain deeper insights into the audience’s emotional responses and behavioral patterns, thereby providing strong support for the optimization of gamified education and helping researchers gain a more comprehensive understanding of audience needs, emotional trends, and community dynamics. To overcome the limitations of the data source, future research could expand the range of platforms and conduct comparative analysis using data from other platforms, thereby gaining a more complete understanding of the emotional responses and participation behaviors of different groups in gamified education.

3.2. Data Acquisition and Preprocessing

In strict compliance with Bilibili’s terms of service and data usage policies, this study collected approximately 40,000 user comments by using keyword searches such as gamified learning, autonomous gamified learning, and personalized learning platform. To ensure the validity and representativeness of the data, a dual-track strategy of “manual review + automated processing” was adopted, involving multiple rounds of screening and cleaning. First, manual sampling was conducted to assess the relevance of comments to learning behaviors, with advertising and invalid content being removed. Then, automated scripts were applied to standardize formatting, non-Chinese text was removed, redundant content was deduplicated, and noisy data (e.g., emojis, garbled text, and repetitive words) was filtered out. As a result, 19,466 high-quality comments with semantic integrity and discussion density were retained. Based on this refined dataset, the study utilized the jieba Chinese word segmentation tool to process the comments and employed a customized stopword list to eliminate meaningless or irrelevant words. This ensured textual clarity and analytical precision, providing a reliable data foundation for subsequent research.

3.3. Deep Data Analysis

3.3.1. Sentiment Analysis

To further explore the public’s emotional tendency towards gamified learning, this study utilized SnowNLP for sentiment analysis. The sentiment analysis tool, SnowNLP, represents the emotional tendency of the text through a sentiment score ranging from 0 to 1. To simplify the sentiment analysis process and adapt it to practical application scenarios, this study employed a three-category classification method based on the sentiment score: positive, neutral, and negative. Specifically, when the sentiment score is greater than or equal to 0.6, the text is classified as having a positive sentiment; when the score is between 0.4 and 0.6, the text is classified as having a neutral sentiment, indicating a relatively neutral emotional expression with no obvious tendency; and when the sentiment score is below 0.4, the text is classified as having a negative sentiment. This three-category classification method not only helps us efficiently understand and analyze the emotional tendencies of the text but also provides a convenient tool for sentiment tendency judgment in fields such as public opinion analysis and social media comment analysis. This classification standard further aids in revealing the emotional attitudes and trends of the public in the field of gamified learning. Therefore, based on the sentiment score range of SnowNLP, the three-category classification method was widely applied in this study, aiming to meet the analytical needs and provide clear data support. However, this classification method may be too coarse when handling complex emotional expressions, potentially failing to capture subtle emotional differences within the text. In future research, more refined classification methods, such as a five-category system, could be combined to allow for a more comprehensive analysis and understanding of the emotional tendencies in the text.

To determine the performance of SnowNLP, this study employed a data extraction approach for testing. In each round, 200 data points were extracted, with a total of 5 rounds, yielding 1000 data points for testing. Expert label judgments were saved as True_Label, while labels predicted by SnowNLP were recorded as Snow_Label. The accuracy, precision, recall, and F1 score of SnowNLP were then calculated using scikit-learn. As shown in

Table 1. SnowNLP performance test.

Group	Accuracy	Precision	Recall	F1 Score
1	0.76	0.75	0.75	0.74
2	0.77	0.79	0.77	0.77
3	0.75	0.76	0.73	0.75
4	0.80	0.83	0.78	0.77
5	0.76	0.72	0.74	0.72

, the results of the five rounds of testing were displayed. From the five rounds of results, it can be observed that the overall performance of SnowNLP is relatively stable, with minimal fluctuation in accuracy, precision, recall, and F1 scores across the groups, demonstrating a certain level of robustness. As shown in

Table 2. Performance comparison of SnowNLP models.

Settings	Accuracy	Precision	Recall	F1 Score
Full model	0.76	0.77	0.75	0.74
Remove the part of speech	0.69	0.71	0.68	0.69
Remove the sentiment dictionary match	0.64	0.67	0.62	0.64

, tests were conducted on the Full Model, Removal of Part of Speech, and Removal of Sentiment Dictionary Match. The results indicated that the full model performed best across all performance metrics, suggesting that the comprehensive use of various features positively influences the model’s performance. Both removing the part of speech and removing the sentiment dictionary match led to a decline in model performance, with the impact of removing the sentiment dictionary match being more significant. However, it is important to note that due to the large volume of data in this study and the random extraction method used for expert evaluation, there remains a certain degree of discrepancy between human ratings and model predictions, which may also affect the evaluation results.

3.3.2. Identify and Analyze Potential Topics

In the process of Latent Dirichlet Allocation (LDA) topic modeling, selecting the appropriate number of topics is crucial [42]. If the number of topics is set too low, the model will be overly simplified and fail to capture the diversity within the text data; conversely, if set too high, it may lead to the generation of redundant or highly similar topics, thus affecting the model’s effectiveness [43]. To optimize the topic model and assess its performance, this study systematically validated the optimal number of topics using the CoherenceModel class from the Gensim library, which calculates the model’s coherence score via the c_v coherence measure. The validation range for the optimal number of topics was from 3 to 8, and eight experiments were conducted. The experimental results are shown in Figure 4. When the number of topics was set to 5, the model’s coherence score fluctuated around 0.63, demonstrating the best performance. This result indicates that with five topics, the model achieved a more ideal balance in terms of fitting the data and distinguishing between topics, ensuring the stability and interpretability of the analysis results.

After conducting LDA topic analysis, five potential topics and associated keywords were identified, and sentiment analysis was performed using two methods. Multi-dimensional topic analysis further illustrated the sentiment distribution across specific topics by incorporating sentiment scores for each comment obtained from SnowNLP. This approach provided more rich and accurate information for multi-layered sentiment interpretation. It helps to better understand the public’s emotional attitudes toward various topics. For example, reviewers may express positive sentiments about the learning outcomes, while negative emotions may be observed concerning interactivity or difficulty. The sentiment score for each comment not only reflects its overall emotional tendency but also displays the emotional distribution across multiple specific topics. Focused topic analysis uses SnowNLP to assign the dominant sentiment to the topic with the highest relevance in each comment for sentiment analysis. This process is beneficial for revealing the strength of the relationship between comments and specific topics, offering a deeper analysis of emotional attitudes toward each topic, and providing more detailed sentiment analysis results. Finally, based on the results, a theoretical model is constructed.

4. Results

4.1. Emotional Attitude

In the absence of thematic influence, sentiment analysis was conducted on 19,466 comment entries. As shown in Figure 5, the overall sentiment exhibited a predominantly positive trend; however, negative feedback and neutral sentiments still warrant attention. Specifically, in the context of adaptive gamified education, 63.8% of the comments expressed positive sentiments, indicating that most users hold a highly favorable attitude toward this personalized learning experience. Adaptive gamified education dynamically adjusts task difficulty based on learners real-time progress and competence level, thereby precisely matching their needs and providing opportunities that are both challenging and conducive to self-improvement. Nevertheless, 22.0% of comments with negative sentiment revealed skepticism and dissatisfaction among some users toward adaptive gamified education. Although the system demonstrated a certain degree of flexibility in adjusting task difficulty, for some learners the assigned challenges still appeared excessively difficult or misaligned with their actual competence, potentially leading to frustration. In addition, some users expressed a lack of trust in the system’s automatic adjustment mechanism, believing it failed to accurately identify individual differences and could not fully address their authentic learning needs. Meanwhile, 14.2% of comments with neutral sentiment reflected a wait-and-see attitude toward adaptive gamified education. For these users, the system’s tangible benefits had not yet been fully experienced. They neither fully embraced nor completely rejected this emerging educational approach, instead rationally assessing its potential value and practical effectiveness.

These findings suggest that the implementation of gamified education in contemporary learning environments produces significant emotional effects—both stimulating positive emotions among learners and exposing certain negative feedback and challenges [44]. This diversity of emotional attitudes provides valuable perspectives for further research on the effectiveness and optimization pathways of gamified education, particularly with regard to its adaptive capabilities.

4.2. LDA Topic Modeling

To gain deeper insights into the comment data, this study employed the Latent Dirichlet Allocation (LDA) topic modeling method, which revealed five themes and their associated keywords. The LDA-generated visualization (Figure 6) clearly demonstrates the model’s outstanding capability in distinguishing between different categories. On the left side of the figure, there is a clear separation among categories without any overlap, indicating the model’s high efficiency and precision in identifying discriminative features. On the right side, the most prominent and frequently occurring terms in the dataset are displayed, further confirming the reliability of LDA in feature extraction. Furthermore, as shown in Figure 7, the prominent keywords within each topic and their relative influence in the topic construction process not only reveal the underlying emotional and cognitive trends in the comment data, but also provide strong support for a more comprehensive understanding of users multidimensional feedback on adaptive gamified education. These analytical results reveal not only the potential structural differences embedded within the comment data but also establish a solid theoretical foundation and analytical framework for subsequent research. The five identified themes illustrate the diversity and innovativeness of modern educational models, highlighting the integrated roles of task-driven learning, autonomy, interdisciplinary integration, and psychological well-being in the educational process.

• Topic 1: Multilingual hierarchical construction: task-advanced language acquisition.

The frequent emergence of topic vocabulary “learn” and “language” shows that language learning is the core task of multilingual acquisition, while the emergence of “English” and “Japanese” reveals the complexity of cross-language learning. it shows that in the task-driven learning process, students should not only improve their language ability, but also choose the order and mode of language learning according to their own needs. The emergence of vocabulary “software” emphasizes the supporting role of modern learning tools for autonomous learning, especially through task-oriented learning software, students can learn independently according to their own learning rhythm and interests. The combination of these words reflects the combination of task-driven and autonomous learning in multilingual hierarchical construction. Students promote the process of language acquisition from simple to complex by choosing tasks and tools on their own.

• Topic 2: Sorted programming thinking: Cognitive leap in the creative environment of young age.

The emergence of the theme words “kindergarten” and “elementary school” shows that the main object is mainly young children. The emergence of “game” and “programming” emphasizes the combination of game-based learning and programming education, which can effectively promote children to explore and learn in a pleasant environment. In particular, the emergence of “scratch” highlights the core position of graphical programming tools in young education. These tools provide children with a stress-free learning experience through visual design interfaces. In the “kindergarten” and “primary school” stages, step-by-step teaching methods can gradually guide children to complete programming tasks from simple to complex according to their cognitive development level. The combination of these words shows that programming thinking can help children master skills and promote cognitive leap at the same time in game-based learning at a young age.

• Topic 3: Certification-driven chain learning: modular English proficiency and behavioral continuity.

The frequent appearance of thematic vocabulary “video” and “check-in” shows that learning behavior is gradually strengthened through visual forms and systematic feedback mechanisms, and English abilities are continuously accumulated through daily clock-in and watching teaching videos. In particular, the emergence of the term “winning streak” reveals that learners’ intrinsic motivation is motivated by completing tasks and achieving continuous success and certification, forming lasting learning behaviors. The emergence of the word “everyday” further emphasizes the dailyization of learning tasks. It can be seen that the user’s learning has formed a continuous and cumulative process, thereby effectively maintaining the learner’s learning motivation and promoting the modular improvement of English ability. Through this system, while learning can achieve the accumulation of English skills, they also gain dual support from internal driving force and external evaluation, effectively maintaining the consistency and sustainability of learning.

• Topic 4: Cross-domain cognitive fusion: a program learning space for mental resilience and psychological safety.

The frequent appearance of “programming” and “AI” shows that in the learning mode of cross-domain cognitive integration, learners make progress in multi-disciplinary expression and creation on the programming platform. The emergence of the word “history” shows that cross-domain learning is not only limited to technical skills, but also includes the comprehensive use of knowledge, thus promoting the expansion of learners’ thinking in a wider range of fields. At the same time, the emergence of the words “horror” and “mental health” implies that the challenge of emotion regulation may be involved in the learning process, especially in the aspect of mental health. How to ensure that safe and appropriate age content has a positive impact on learners’ psychology, so as to improve their cognition of mental resilience and mental security has become an important issue. The combination of these words reveals the procedural learning space of cross-domain cognitive integration, which not only pays attention to learners’ skill development, but also takes into account the needs of their mental health and emotional management, forming a healthy learning cycle driven by feedback.

• Topic 5: Universal educational environment: Innovative exploration of adaptive gamified learning strategies.

The frequent appearance of “education” and “systems” shows that this topic focuses on the deep integration of educational tools and learning systems, especially through the integration of platforms such as Khan Academy and Kahoot to build an adaptive gamified learning environment with personalized learning choices. In this system, the core of learning strategies is to stimulate the active participation of learners, and enhance intrinsic motivation through gamified elements and personalized recommendation mechanisms, thereby enhancing learning momentum. System design not only relies on the transmission of text knowledge, but also integrates multimedia resources, including videos, images, etc. Through the dual support of vision and hearing, learners can master knowledge more intuitively and comprehensively.

4.3. Potential Topic Analysis

4.3.1. Multidimensional Thematic Sentiment Analysis

Initially, a sentiment analysis was conducted on all the comments based on the five identified themes, as shown in Figure 8. The emotional expressions of each comment varied across the four themes. While each comment may emphasize one particular theme, it could still exhibit some degree of emotional tendency across each theme. For instance, the comment “Learning some languages as a hobby is quite interesting” scored 0.83, 0.04, 0.13, 0.04, and 0.72 across the five themes. This comment primarily expresses a positive attitude toward Topic 1, with minimal association to Topic 2, Topic 3, and Topic 4, but still shows a relatively positive sentiment toward Topic 5. From the overall sentiment distribution, there is a notable concentration of values between 0.0 and 0.2, which occurs frequently. This may stem from a lower degree of relevance between the comment content and the themes. However, some comments with more negative sentiment are included within this range, which reflects the authenticity and diversity of the data. The distribution of other sentiment values is more even, with some significant values appearing between 0.5 and 0.8, indicating a slight positive trend in sentiment. This distribution further reveals the general emotional tendency of the comment group, which leans toward positive evaluations of adaptive gamified education.

4.3.2. Highlight Thematic Sentiment Analysis

Topic 1: Multilingual hierarchical construction: task-advanced language acquisition. Positive emotion (57.41%) is higher than negative emotion (20.74%), neutral emotion (21.85%) is in the middle. The emotional tendency of this theme shows a more positive trend. It shows that the learning tasks involved in this topic are challenging and complex, but under the framework of task-driven and autonomous learning, it can provide students with space to choose their own learning style and schedule, and this flexibility helps to improve learning motivation and emotion. The emergence of neutral emotion and negative emotion shows that there are still some uncertainties and challenges in the process of implementation, which have an impact on the experience of some learners.

Topic 2: Sorted programming thinking: Cognitive leap in the creative environment of young age. This theme exhibits a strong positive emotional tendency, with positive emotions (87.98%) dominating, negative emotions (5.52%) being relatively infrequent, and neutral emotions (6.50%) being relatively low. The combination of gaming and programming, especially with the support of tools such as Scratch, provides children with a fun and relaxed learning experience. This stress-free learning environment greatly enhances students’ positive emotional responses. The innovation and playfulness of the educational approach not only improve students’ cognitive abilities but also increase their emotional engagement. The occurrence of negative and neutral emotions may stem from some students’ confusion or challenges in adapting to the programming environment. Although the overall learning experience is positive, individual students may face difficulties in understanding or adjusting to new tools.

Topic 3: Certification-driven chain learning: modular English proficiency and behavioral continuity. This theme shows that positive emotions (53.30%) outweigh negative emotions (29.63%), with neutral emotions (17.07%) in between. The emergence of negative and neutral emotions primarily stems from some learners’ experience of accumulated pressure and anxiety in the pursuit of long-term learning goals, particularly when short-term results are not readily visible and immediate gratification is lacking, which prevents emotional experiences from reaching the expected positive state. This emotional response reflects learners’ internal confusion and unease when facing challenges. However, as learners progressively achieve success through certification and task completion, positive emotions are significantly stimulated. Notably, as learners make progress and achieve certification, their sense of accomplishment and self-confidence are enhanced, further boosting their learning motivation and emotional engagement.

Topic 4: Cross-domain cognitive fusion: a program learning space for mental resilience and psychological safety. This theme presents a generally positive emotional tendency, with positive emotions (70.20%) dominating, negative emotions (12.75%) being relatively few, and neutral emotions (17.05%) being low. The high proportion of positive emotions is attributed to the enjoyment and sense of achievement brought about by task unlocking and goal accomplishment. When learners experience satisfaction and self-affirmation, their emotions are stimulated, thus enhancing their mental resilience and self-confidence. Although negative emotions are less frequent, they should not be overlooked, as they mainly arise from emotional regulation and psychological pressure, especially during high-difficulty tasks or when there is a lack of timely feedback, which may trigger anxiety. The low occurrence of neutral emotions indicates that most learners do not show indifference, and when they do occur, they are typically associated with difficulties in understanding or slow progress.

Topic 5: Universal educational environment: Innovative exploration of adaptive gamified learning strategies. This theme is dominated by positive emotions (67.19%), with negative emotions (24.43%) being relatively notable and neutral emotions (8.38%) being less frequent. The relatively high proportion of positive emotions indicates a general public perception that adaptive gamified learning strategies can effectively stimulate learning motivation and enhance learning outcomes, particularly by providing positive experiences in the design of personalized learning pathways. However, the presence of neutral and negative emotions reflects a reliance on traditional learning methods. Some learners express concerns about difficulties in adapting to new technologies, the complexity of the system, and potential technical barriers, as well as apprehensions about the challenges and uncertainties that may arise during the implementation of adaptive learning systems.

For example, Figure 9 is the emotional distribution under five themes. The emotions of the five themes show that innovative educational models can generally bring about more positive emotional tendencies, but there are also certain emotional challenges. These challenges are mainly reflected in the complexity of task-driven and learning tools, as well as the stress and anxiety that may arise during the learning process.

5. Discussion and Contribution

Through sentiment analysis and theme modeling, this study identified five semantic themes and their emotional tendencies related to gamified education. In order to integrate the semantics and emotional tendencies of the topic, this study further constructs a theoretical model of adaptive gaming “Motivation–Cognition–Strategy” (M–C–S), as shown in Figure 10.

5.1. Topic Discussion

5.1.1. Multidimensional Motivational Fulcrum of Motivation Transformation Driven by Gamification: Autonomy, Relatedness, Competence

In the context of multilingual and cross-disciplinary learning, gamification mechanisms not only enhance learners’ sense of engagement but also provide crucial support for their cognitive development and psychological regulation [31]. Adaptive learning environments, through the flexibility of task selection, empower learners with greater autonomy, transforming them from passive recipients of task assignments into active participants with the ability to regulate learning strategies [33]. This process strengthens self-drive and goal orientation [45]. At the same time, social interaction, psychological safety education, and real-time feedback mechanisms increase the emotional connection between learners and the learning environment, fulfilling the need for relatedness, particularly in collaborative tasks and community interactions, effectively boosting situational engagement and sustained motivation [46].

Emotional analysis shows that most learning topics elicit highly positive emotions, especially in the context of Scratch programming for young children, indicating that gamified elements can significantly enhance learners’ emotional states. This is particularly evident in early childhood education, where emotion-driven motivation becomes a key factor in motivational transformation. However, excessive reliance on certification mechanisms and technology-driven learning models may lead to emotional fluctuations. For instance, technology-driven learning models place too much emphasis on quantitative results, such as scores and certificates, which may make learners feel that their efforts are not recognized, leading to a lack of achievement and increased anxiety [47]. Self-Determination Theory suggests that adaptive gamification, through progressively challenging tasks and the enhancement of competence [48], motivates learners by fostering a sense of achievement and self-efficacy in dynamic environments [49]. Reward systems and personalized feedback not only reinforce motivational transformation but also provide key pathways for learners to shift from “external rule followers” to “internal strategy creators” [50].

5.1.2. The Interactive Cycle Between Motivation Transformation and Cognitive Leap: Advanced Learning from the Perspective of Connectivism

Under the dual framework of Connectionism and Self-Determination Theory, learners gradually shift from extrinsic motivation to intrinsic motivation, which triggers a leap in their cognitive structure [51]. The transition from external rewards to internal drives, through the satisfaction of the three psychological needs (Autonomy, Relatedness, Competence), constitutes the intrinsic mechanism of motivational transformation, laying the foundation for the generation of learning strategies and the expansion of cognitive structures [52].

The task challenges, immersive environments, and connections between knowledge nodes brought about by gamification encourage learners to form “point-to-point” cognitive links, prompting deep thinking across multidisciplinary and multidimensional learning tasks [53].Through various learning contexts and task-driven approaches, learners build cross-disciplinary knowledge connections and demonstrate strong cognitive restructuring abilities [54]. When facing challenges from different domains, the adaptive gamification learning mechanism enhances learners’ ability to integrate knowledge [55], enabling them to flexibly respond to challenges from different fields, thus fostering the fusion and innovation of cross-domain knowledge [56]. This mechanism responds to current literature’s calls for adaptive mechanisms to promote cognitive deepening and fills the gap in systematically explaining the path of motivational evolution [56].However, learning models that overly rely on external incentives or feedback mechanisms may provoke negative emotions, which, while diminishing learners’ autonomy and competence, also continuously reduce intrinsic motivation [57]. For example, an excessive focus on quantitative outcomes, such as scores or recognition, when learners feel unacknowledged, further impacts the expansion of cognitive structures and the integration of cross-domain knowledge [58]. While adaptive gamification learning mechanisms contribute to enhancing cognitive abilities, the accumulation of negative emotions still affects motivational transformation and cognitive deepening, which should be carefully considered in instructional design.

5.1.3. The Mechanism of Strategy Generation and the Integration of Ubiquitous Learning: Innovative Exploration of Adaptive Gamified Learning Strategies

The generation of cognitive networks is not an isolated process, but a dynamic mechanism evolved based on continuous motivational satisfaction and cognitive transfer. Connectivism theory provides a basic framework for understanding this process: the expansion and reconnection of knowledge nodes shifts the cognitive structure from linear evolution to multidimensional interaction, thereby promoting the development of strategic cognition.

With the integration and innovation of cross-domain knowledge, learners have effectively integrated knowledge in different fields and continuously promoted the integration and deepening of cognitive structures. In particular, by combining gamified elements with interdisciplinary education, learners can enhance their learning situational and creativity through cross-media narrative and task-driven.

More importantly, learners not only accept knowledge, but also reconstruct and transform in the learning process. The exploration of multidimensional motivation driven prompts them to actively formulate learning goals, reflect on learning paths, and dynamically adjust their strategies, thereby building a highly adaptable personalized cognitive system [59]. This cognitive reconstruction process not only echoes the expectations of personalized learning strategies in the adaptive gamification environment but also responds to the previous research call on “how to support learners to transform from passive acceptance to active construction”, showing the highly coupled relationship between motivation satisfaction, strategy regulation and knowledge reconstruction [60]. The cognitive mechanism revealed in this study not only expands the theoretical boundaries of existing gamified learning but also provides empirical support for learners’ psychological regulation and strategic evolution [61].

Particularly in the integration of gamified elements with interdisciplinary education, learners are able to enhance their situational awareness and creativity through cross-media storytelling and task-driven approaches [62]. Under positive emotional states, most learners not only acquire knowledge but also reconstruct and transform it during the learning process. Exploration driven by multidimensional motivation encourages them to proactively set learning goals, reflect on learning paths, and dynamically adjust strategies, thereby building an adaptive and personalized cognitive system [63]. However, the emergence of anxiety and frustration can weaken learners’ cognitive resources and psychological regulation abilities, reducing self-efficacy and hindering effective reflection and adjustment [64]. These emotions can limit the construction of personalized learning strategies, thus impacting the integration and innovation of cross-domain knowledge. This cognitive reconstruction process not only aligns with the expectations of adaptive gamified environments for personalized learning strategies but also responds to previous research calls on “how to support the transition from passive acceptance to active construction in learners,” demonstrating the high coupling relationship between motivation fulfillment, strategy regulation, and knowledge reconstruction [65]. Therefore, in the context of ubiquitous education, designing gamified mechanisms that support strategy generation, emotional regulation, and cognitive reconstruction becomes a critical issue in future instructional design.

The strategy generation mechanism revealed in this study further expands the theoretical depth of the “strategy” dimension in the M–C–S model, providing both empirical foundation and theoretical support for the personalized design of adaptive gamified learning.

5.2. Research Contributions and Suggestion

5.2.1. Research Contribution

A.

Detecting the heterogeneity of adaptive learning emotional motivation and acceptance with the LDA and sentiment analysis fusion model.

This study integrates Latent Dirichlet Allocation (LDA) topic modeling with sentiment analysis to construct a corpus-processing framework capable of capturing the interaction between thematic semantics and emotional dimensions. By performing semantic classification and sentiment quantification on nearly 20,000 high-quality user reviews, the analysis reveals significant heterogeneity in emotional tendencies across different topics. In the context of applying Scratch programming to early childhood education, the proportion of positive emotions is the highest, indicating that gamification elements exert a pronounced emotional stimulating effect on young learners. At the same time, certain negative emotional expressions also emerge within the topics, reflecting that when confronted with high-frequency cognitive tasks, learners may experience adverse emotional responses linked to challenges to their psychological resilience and cognitive abilities. This research addresses a gap in the existing literature, which has predominantly emphasized cognitive outcome assessments while overlooking the role of emotion as a transformative driver [66], thereby providing empirical support for constructing the emotion–motivation–behavior chain in educational design.

B.

Fusion of SDT and Connectivism to construct an adaptive gamification theoretical model: the progressive generation mechanism of motivation-cognition-strategy.

This study is based on Self-Determination Theory and Connectivism to construct a theoretical model covering external incentives, internal motivation and cross-domain knowledge reconstruction. Through topic modeling and behavioral feature extraction, we explore how learners can gradually transition from passively accepting game rules to actively adjusting learning strategies in a gamified learning environment.

This model reveals the following three core mechanisms:

• Psychological needs progressive: autonomy → relatedness → competence, which constitutes the fulcrum logic of motivation transformation.
• Cognitive structure expansion: knowledge modularization → cross-domain connection → post-set strategy optimization to drive learners’ cognitive reconstruction.
• Strategy evolution systemization: feedback loop → situational adjustment → personalized path generation to achieve dynamic coupling of motivation-cognition.

Distinct from existing studies that tend to examine gamification mechanisms in isolation or focus primarily on optimization from a technical perspective, this study proposes an integrated three-dimensional theoretical model encompassing social–psychological factors, cognitive development, and strategic regulation. This framework offers a novel and systematic perspective for educational gamification. It further explores the interplay among learners’ motivation, cognition, and strategic regulation, thereby advancing theoretical development in the field and addressing the systematic gap in educational gamification research concerning the mechanisms of motivational evolution and the pathways of cognitive generation [52].

5.2.2. Research Suggestion

A. Improve the mechanism of technical support and emotion regulation. The adaptive game-based learning system should intelligently adjust the task difficulty and feedback mode according to the learner’s learning progress and emotional state through personalized technical support [67] so as to ensure that learners maintain a balance between challenges and abilities [68], and to enhance their enthusiasm and sense of participation in learning. At the same time, the construction of emotional regulation mechanism should be based on emotional calculation and psychological theory, and design intervention measures for different emotional needs [69]. For example, real-time positive feedback, virtual social interaction, and emotional support system [70] can effectively regulate learners’ emotional response and promote their long-term learning motivation and emotional input [71].

B. Balancing the relationship between academic stress and learning outcomes. In an adaptive gamified learning environment, academic pressure should be designed reasonably by designing the challenges and progress of tasks to ensure that learners can gain a full sense of accomplishment when facing adaptive challenges, rather than falling into excessive anxiety [72]. At the same time, the evaluation of learning outcomes should focus more on process feedback rather than simple result orientation [73], and encourage learners to pay attention to self-growth and ability improvement. This suggestion helps promote learners’ self-cognition and skills development, achieve a benign interaction between academic stress and learning outcomes, and promote the healthy development of the adaptive learning process [74].

C. Cognitive network design to promote cross-domain knowledge construction. Based on the principle of multidisciplinary integration, the construction of a cross-domain cognitive frame is the key to promote knowledge construction [75]. Through interdisciplinary learning tasks and situational design, learners are inspired to establish connections between different fields, so as to promote the integration and transfer of knowledge [76]. When designing cognitive networks, we should pay attention to the hierarchy and relevance of knowledge, through visualization tools and dynamic network graphs [77] to help learners intuitively understand the relationship between knowledge in various fields and reveal their internal relations.

6. Conclusions

Based on self-determination and connectionism, this study constructed the “Motivation–Cognition–Strategy” (M–C–S) theoretical model of adaptive gaming learning and systematically analyzed the transformation mechanism of learners from passive acceptance to active regulation of learning strategies. The research results show that the progressive satisfaction of autonomy, relatedness and competence constitutes the core fulcrum of motivation transformation; at the same time, real-time feedback and task ladder design promote the expansion of cognitive structure, while cross-domain knowledge integration further promotes the generation and optimization of post-set strategies.

Through the integration of semantic theme modeling and emotion analysis, this study identified five major semantic theme groups and their emotional tendencies, and then revealed the key role of emotional motivation in motivation transformation. The results show that in the field of language learning and young children’s program education, gamified design can significantly promote positive emotional activation and learning participation; while in the learning scenarios driven by technology and certification intensive, attention should be paid to the learners’ emotional burden and psychological resilience regulation. The above findings extend the cognitive understanding of adaptive gamified learning by existing research, emphasizing the key role of emotional regulation and strategic evolution in personalized teaching design.

Theoretically, the M–C–S model proposed in this study integrates three-dimensional logic of psychological dynamics, cognitive generation and strategic construction, providing a new explanatory architecture and theoretical support for gamified learning research; in this method, the corpus processing method of LDA and emotion analysis is integrated to provide a specific path for the mining of learners’ semantic behavior characteristics; in practice, this study emphasizes the impact of multidimensional motivational support points and cross-media interaction design on promoting learners’ long-term learning motivation and knowledge construction.

However, this study still has certain limitations, including a lack of long-term tracking data and empirical support from real-world validation settings, limitations in platform data sources, and the granularity issues in sentiment classification. Future research on the current M–C–S conceptual model can focus on experimental design and validation, with potential deepening in the following three directions: First, optimizing topic modeling techniques for other educational fields, such as applying BERT and large language models (LLM) to areas like cultural heritage [78], educational psychology [79], and public health, to uncover more cross-domain knowledge construction mechanisms; Second, incorporating multimodal data, including literature semantics [80], image visualizations, and speech emotions [81], to enhance the model’s generalization capability and contextual adaptability; Third, conducting longitudinal empirical designs to observe how learners from different cultural regions evolve their strategic regulation at various time points under the same gamified design. This would provide a more comprehensive understanding of emotional responses and engagement behaviors across different groups in gamified education, further optimizing adaptive mechanisms in educational technology.