Design and Evaluation of Computer Assisted Chinese Second Language Acquisition Teaching Model
Pubblicato online: 31 mar 2025
Ricevuto: 13 nov 2024
Accettato: 19 feb 2025
DOI: https://doi.org/10.2478/amns-2025-0820
Parole chiave
© 2025 Wang Difei, published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.
With the rapid furtherance of technology, new technologies such as artificial intelligence (AI), big data, cloud computing, et al. are affecting various industries at an astonishing speed and scale, and the education sector is also undergoing profound changes [1]. In today's era of deepening educational informatization, online education has ushered in an unprecedented period of rapid development with its characteristics of breaking time and space limitations, diversifying resources, and strong interactivity, becoming a key force driving educational modernization [2]. The widespread application of information technology not only broadens the boundaries of online education, but also opens up new paths for the innovation of learning modes, driving education towards a more open, flexible, and efficient direction [3]. In this process, CALL, as a key element of educational reform, has become increasingly important. CALL not only reflects the deep integration of information technology in education, but also plays an important role in driving fundamental changes in traditional learning and teaching methods [4].
The core of CALL lies not only in the simple application of information technology, but also in comprehensive innovation covering multiple dimensions such as teaching methods, teaching design, teaching environment, and teaching resource construction [5]. With intelligent and personalized teaching methods, it greatly enriches learners' learning experience, enhances the efficiency and fun of language learning, and achieves the diversification and efficiency of learning content [6]. In the current area of education, personalized learning, as a key path to promote students' personalized development, has become an important direction for educational reform and school construction [7]. The vigorous development of Internet technology has brought unprecedented opportunities for online education, with more diverse ways of knowledge acquisition and more convenient access and sharing of learning resources. In this context, CALL technology comprehensively analyzes multidimensional information such as students' learning history, interest preferences, and ability status, and uses advanced AI algorithms to accurately identify and meet students' personalized needs, customizing the most suitable learning path and resources for them, truly practicing the educational philosophy of "teaching according to students' aptitude".
In addition, CALL technology can also use AI algorithms to predict learning outcomes based on students' learning history, ability status, and learning trajectories, providing teachers with accurate teaching feedback and helping students adjust their learning strategies and optimize their personal learning plans in a timely manner. This data-driven teaching decision support not only enhances the precision and validity of teaching, but also brings new perspectives and tools for optimizing the education evaluation and feedback system. This article innovatively designs a Chinese language teaching system that integrates CALL technology. The system deeply integrates DL algorithms and can intelligently analyze students' learning habits, interests, and ability levels, thereby providing personalized learning resource recommendations for students. The system also has a learning achievement estimation function, which uses big data analysis technology to pre evaluate students' learning validity, providing data support for teachers' teaching strategy adjustments and students' learning planning optimization. The innovation points are as follows:
This system not only fully utilizes the powerful capabilities of DL in data processing, pattern recognition, etc., but also deeply mines and analyzes students' learning habits, interest preferences, and ability levels through algorithm models, thereby achieving accurate grasp of learners' personalized needs. Through learners' historical learning data, interest preferences, and other information, this system can intelligently recommend learning resources that meet their personalized needs, effectively improving the pertinence and validity of learning. This article uses intelligent and personalized teaching methods to not only help students master Chinese knowledge and skills faster, but also stimulate their enthusiasm and interest in learning Chinese, promoting the comprehensive development of language skills.
This article begins with an in-depth analysis of the macro background and importance of research, followed by a detailed discussion of the latest developments and cutting-edge trends of CALL technology in the field of Chinese L2 acquisition. The main body of the article comprehensively reveals the unique design concept and innovative functional characteristics of the Chinese language teaching system based on CALL technology. To verify the validity and practicality of the system, we carefully designed scientific experiments and conducted comprehensive and rigorous verification. In the conclusion section, we not only systematically summarized the main research results of this article, but also deeply reflected on the limitations and shortcomings of the research. On this basis, we have looked forward to future research directions and proposed improvement suggestions, aiming to contribute to the sustained development of the field of Chinese L2 acquisition.
At present, numerous scholars have conducted extensive research on CALL. Chen et al. [8] constructed a student writing performance prediction model that integrates PCA (Principal Component Analysis) and RBF (Radial Basis Function) networks. The model first uses PCA for data dimensionality reduction, and then uses RBF networks to accurately predict student performance. Wang [9] proposed a groundbreaking collaborative filtering recommendation algorithm that integrates learners' social network information to address the issue of learning resource recommendation. Jie et al. [10] established a student performance prediction model by applying the decision tree (DT) C4.5 algorithm, deeply analyzed and determined the critical factors and rules that affect student performance, and provided a solid theoretical basis for the formulation of learning plans and performance prediction. Wei [11] provides users with a diversified personalized recommendation list based on their browsing history and current context. Tang [12] proposed a hybrid prediction model combining DL and collaborative filtering. The model is first based on the collaborative filtering model of students, calculates the similarity of students' styles based on their performance information, and then accurately predicts students' performance in various courses through the hybrid prediction model.
Li et al. [13] used DT algorithm to predict the learning behavior and performance of online learners, and based on this, constructed an intervention model for adaptive learning system. Medetbayeva et al. [14] identified the key factors affecting academic performance by calculating correlation coefficients and information gains, and then used ensemble classifiers such as Bagging, Boosting, and RandomForest to efficiently predict student performance. Jin et al. [15] designed a two-stage recommendation method, with the first stage aimed at predicting and supplementing the sparsity of user ratings, and the second stage finely ranking the results of collaborative filtering recommendations. Yang [16] proposed a collaborative filtering recommendation method for teaching resources based on clustering algorithm, which groups users through clustering technology and significantly improves the scalability of the system. Jing et al. [17] used DTs, neural networks, support vector machines (SVM) and other methods to verify the feasibility and validity of predicting learning achievements based on college students' Internet use data.
Compared with existing research, the Chinese language teaching system based on CALL technology proposed in this article demonstrates significant innovation in the application of DL algorithm, personalized learning resource recommendation, and learning validity prediction functions. At the same time, it effectively promotes the improvement of learners' learning efficiency and interest, injecting new vitality and momentum into the reform and development of Chinese as L2 teaching.
CALL, as a major innovation in the area of educational technology, injects new vitality into language teaching by relying on advanced AI technology. In this learning process, learners are the core participants, who not only need to master the basic language knowledge, but also need to flexibly apply what they have learned in real communication environments. The traditional teaching mode often centers around the teacher and adopts classroom style teaching. However, in this mode, students' subjective initiative has not been fully utilized, and they are more passive in receiving knowledge rather than actively exploring, understanding, and applying it. In contrast, CALL, with its intuitive and vivid characteristics, makes language learning lively and interesting through diverse teaching methods such as audio, video, charts, pictures, etc., greatly enriching classroom activities, enlivening classroom atmosphere, promoting classroom interaction, and providing more possibilities for teachers' classroom arrangements.
More importantly, the leap of network technology has broken the constraints of time and space, opening a door for students to a vast world. They are able to frequently interact with the outside world and interact with individuals from diverse cultural backgrounds, promoting cross-cultural learning and communication. This cross temporal learning approach not only greatly broadens students' horizons, but also significantly enhances their language practical application abilities. With the continuous advancement of language learning theory and technology, the role of CALL in the area of Chinese language teaching is becoming increasingly prominent. With the interactivity of multimedia technology, teachers and students can directly communicate with the teaching content in both directions. This interactive mode not only stimulates students' interest in learning, but also gives them higher learning autonomy (as shown in Figure 1). This student-centered, interactive and practical teaching model undoubtedly injects new vitality and opportunities into the development of Chinese language education.
Multimedia teaching
CALL greatly enriches learners' experience by utilizing intelligent and personalized teaching strategies, making it more diverse and efficient, and significantly enhancing the efficiency and fun of language learning. Among them, accurate recommendation of personalized learning resources and accurate prediction of academic performance constitute the two core directions of CALL. In the area of AI, DL has become the mainstream method of research, especially in the processing of Euclidean data such as images, sequence data, and text, where significant achievements have been made. However, for the processing of graph structure representations (i.e. non Euclidean data), traditional DL techniques still have broad application space. This challenge has promoted the development of graph DL, especially graph neural networks (GNNs), which have demonstrated their position as the most successful learning framework in multiple application areas, especially in classification tasks. We have introduced the GCNTREC (Graph Convolution Neural Network Transformer based Recommendation Model) model for personalized learning resource recommendation. This model plays a key role in the recommendation matching stage by integrating the visual perspective information of user and item interactions, and introducing an efficient feature extractor Transformer structure, effectively improving the accuracy of item representation while possessing certain industrial practical application capabilities.
In terms of academic performance prediction, we have designed a prediction model that combines attention mechanism with relation graph convolutional neural network (AR-GCN) and gated recurrent unit (GRU). The model first constructs student and course information into a graph structure, where students and courses serve as nodes and their relationships serve as edges. By introducing attention mechanism R-GCN, we successfully extracted important attribute features of students, such as learning ability, interests and hobbies. Meanwhile, utilizing GRU to capture students' time-series behavioral characteristics. Subsequently, the key features output by R-GCN and GRU are fused in a fully connected layer and classified using a softmax function to ultimately obtain the student's grade prediction results. This article combines the GCNTREC model with the AR-GCN-GRU model and applies it to the system designed in this article, as shown in Figure 2. This combination not only enhances the accuracy of personalized learning resource recommendations, but also significantly improves the accuracy of academic performance prediction, providing learners with a more intelligent and efficient learning experience.
Model structure
The Chinese language teaching system based on CALL technology introduced in this article marks an important step towards intelligent and personalized language education. This system deeply integrates AI and big data technology, aiming to build an effectual and accurate Chinese learning environment. The user module is responsible for collecting students' basic information and learning behavior data, laying the foundation for personalized teaching. The learning resource module integrates a rich variety of Chinese learning materials, covering a comprehensive range of content from basic vocabulary to advanced dialogues, ensuring that students can gradually improve at a level suitable for their own level. The recommendation system module is the core highlight of the system, which uses machine learning algorithms to deeply analyze each student's learning habits, interests, preferences, and current ability level, intelligently push the most suitable learning resources and exercises, effectively stimulate learning motivation, and improve learning efficiency.
The performance prediction module utilizes big data analysis technology, combined with past learning data, to predict future learning outcomes, enabling teachers to adjust teaching strategies in a timely manner and strengthen guidance on weak areas; At the same time, it also allows students to clarify their learning progress, independently optimize their study plans, and ensure the achievement of learning goals. In summary, this Chinese language teaching system not only achieves personalized customization of teaching content, but also enhances the foresight and validity of teaching through predictive analysis, bringing disruptive changes to Chinese language education.
Data conversion is a process that involves transforming the representation of raw data into another representation. In this article, we specifically applied a technique in data transformation, data normalization, using the minimum maximum normalization method. This method adjusts the data values to the range of 0 to 1 through a specific conversion formula, ensuring the scale consistency of the data. The following is the mathematical formula for the conversion process:
Among them,
For learning resources that are difficult to classify, we introduce optimized GNN technology to deeply analyze their structure, achieving precise classification and efficient utilization. To address the issue of vague definition of GNN in complex knowledge domains, we adopt a metadata based ontology representation method to explore the essence of resources and construct a clear knowledge framework. Under this framework, we first carefully constructed a learning resource model.
In the formula,
The improvement of students' Chinese language grades is influenced by various complex factors such as learning conditions and personal performance. To accurately grasp the correlation characteristics between students, we adopted an attention mechanism. In this article, the eigenvectors of student node
In the GCNTREC model, we adopted Graph Convolutional Network (GCN) as the core graph neural network structure of the model. Each layer of GCN cleverly utilizes the adjacency matrix
Among them, the node hidden feature matrix
After understanding user preferences and learning resource details, we adopted a dual clustering method to optimize recommendation accuracy. This strategy utilizes a dual clustering algorithm to form user groups and resource groups separately, and pairs them to determine recommended content. In order to enhance the consistency between clustering results and user needs, we have developed clustering division principles based on resource evaluation. Here, we define the user set as
In this way, we obtained
In the AR-GCN-GRU model, by introducing the softmax function, we regularize the adjacent nodes
After dot product operation between attention mechanism ∂
In the activation function design of the GCNTREC model, we have improved the original ReLU activation function of GCN by introducing LeakyReLU. Unlike ReLU which directly discards all negative values, LeakyReLU assigns a non-zero slope to negative values. This change ensures that LeakyReLU can effectively capture and encode positive and negative signals during message transmission, thereby enhancing the model's expressive power. The mathematical expression is as follows:
To comprehensively evaluate the validity of our system, we conducted a series of comparative tests and compared it in detail with systems using traditional clustering algorithms and random forest (RF). Figure 3 visually presents the comparison between the two in terms of accuracy in recommending Chinese teaching resources, clearly demonstrating that our system has a higher accuracy. This outstanding performance is mainly attributed to the application of the GCNTREC model. The GCNTREC model not only inherits the advantages of GCN in processing graph structured data, but also significantly improves the model's ability to process complex information by innovatively introducing the LeakyReLU activation function. This change enables the model to encode positive and negative signals more accurately during information transmission, thereby enhancing the accuracy and credibility of recommendations. In addition, compared to traditional systems, this system cleverly integrates a dual clustering strategy of user preference information and learning resource ontology information, as well as a clustering division criterion based on resource evaluation. These measures collectively deepen the system's insight into user needs.
Comparison of recommended accuracy
Figure 4 visually illustrates the comparison between our system and the traditional system in terms of time consumption for recommending Chinese teaching resources. Obviously, the system in this article requires less time, highlighting its significant advantage in improving recommendation efficiency. This achievement is largely attributed to the application of the GCNTREC model. GCNTREC, with its powerful GCN structure, accurately captures the complex connections between users and resources, and optimizes the model's information processing capabilities through the innovative LeakyReLU activation function. This improvement ensures that the model can complete recommendations more quickly while maintaining high accuracy, effectively shortening the system response time. In addition, the system in this article also adopts a dual clustering strategy, which clusters user preferences and resource ontology information, and formulates clustering standards based on resource evaluation results, further accelerating the recommendation process.
Time comparison
Figure 5 provides a detailed comparison of the recall rate of Chinese language teaching resource recommendations between our system and traditional systems. Recall rate reflects the proportion of relevant resources successfully recommended by the system to all relevant resources. The data in the figure clearly indicates that compared to traditional systems, our system performs better in recall rate, highlighting its significant advantage in improving recommendation quality. This significant achievement is largely attributed to the GCNTREC model adopted by our system. GCNTREC, with its advanced GCN structure, can deeply explore the potential associations between users and resources, thereby accurately identifying content that users may be interested in. In addition, this article also implements a dual clustering strategy, which involves clustering analysis of user preferences and resource ontology information, and developing clustering standards based on resource evaluation results. These strategies not only deepen the system's understanding of user needs, but also broaden the scope of related resource mining, thereby further enhancing the accuracy and coverage of recommendations and improving recall rates.
Comparison of recall rates
Figure 6 emphatically shows the comparison between our system and traditional systems in terms of academic performance prediction precision. The data in the figure intuitively indicates that compared to traditional systems, our system has shown significant advantages in prediction precision, indicating that our system has made key progress in improving the precision of academic performance prediction. The core of this outstanding achievement lies in the AR-GCN-GRU model adopted by the system in this article. This model cleverly combines AR-GCN with GRU to achieve deep analysis and efficient utilization of learner learning behavior data. Specifically, with the introduction of attention mechanism, AR-GCN can accurately capture the complex connections between learners and their learning resources, thereby gaining deeper insights into learning dynamics. The GRU part relies on its excellent sequence processing capability to effectively grasp the long-term dependency characteristics in the learning time series, thereby improving the precision of prediction.
Precision comparison
Figure 7 visually illustrates the comparison between the system proposed in this article and the traditional system in predicting F1 scores for academic performance. The data in the figure clearly shows that compared to traditional systems, the F1 value of our system has significantly increased, indicating that our system has made significant progress in improving the comprehensive performance of learning achievement prediction. The core of this outstanding achievement lies in the AR-GCN-GRU model adopted by the system in this article. This model innovatively integrates AR-GCN with GRU, achieving deep mining and efficient analysis of learner learning behavior data. AR-GCN utilizes attention mechanisms to accurately capture learners and their complex associations with learning resources, providing more detailed information for prediction. The GRU part relies on its excellent sequence processing ability to effectively grasp the long-term dependency features in the learning time series, further enhancing the accuracy and stability of prediction.
Comparison of F1 values
Figure 8 compares the performance of our system with traditional systems in terms of user satisfaction, and the results show that our system has higher satisfaction, indicating significant validity in improving user experience. This is due to the two major models of GCNTREC and AR-GCN-GRU. GCNTREC utilizes GCN to explore user resource connections and achieve personalized recommendations. AR-GCN-GRU combines attention mechanism and GRU to deeply process learning behavior data and accurately predict grades. The collaboration of the two models enables a more accurate understanding of user needs, provides personalized services, and improves the learning efficiency and interest of Chinese second language learners. This achievement helps optimize the Chinese language learning system and provides a new perspective for the growth of educational technology.
Satisfaction comparison
This article is based on CALL technology and innovatively develops a Chinese language teaching system. The system deeply integrates DL algorithm, comprehensively analyzes students' learning habits, interests and abilities, and provides personalized learning resource recommendations, effectively meeting students' diverse needs and greatly improving their learning efficiency and interests. In addition, the system also has the function of predicting academic performance, using big data analysis technology to predict learning outcomes, providing scientific basis for teachers and students to adjust teaching strategies and learning plans, and helping to accurately control learning progress and optimize learning strategies. Experimental results have shown that the system significantly promotes the comprehensive development of learning efficiency, interest, and language skills of Chinese second language learners.
Although the system has achieved certain results, it still faces challenges, such as the need to refine personalized recommendation algorithms to better fit student characteristics, and the need to incorporate more variables into academic performance prediction functions to improve prediction accuracy. Looking ahead to the future, we will make unremitting efforts to continuously optimize the system and strive to provide students with a more outstanding Chinese learning experience.