Collection of Foreign Language Learners’ Behavioral Data on Online Platforms and Construction of Teaching Feedback Mechanisms
Publié en ligne: 19 mars 2025
Reçu: 30 oct. 2024
Accepté: 11 févr. 2025
DOI: https://doi.org/10.2478/amns-2025-0538
Mots clés
© 2025 Zhilin Sun, published by Sciendo
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
The outbreak of the New Crown Pneumonia epidemic hit the normal order of teaching and learning in schools around the world, which led to the massive development of online learning during the period of the epidemic [1-2]. Together with the development of communication technology and the strong support of the state, online learning, which has already attracted much attention, has ushered in a new development peak. Nowadays, online learning occupies an important position in educational services, and is an important means and mode of learning for learners [3-5]. Foreign language online learning, as a segment of online learning that has been developed for a relatively long time, has a huge market size and is expanding at a high growth rate [6-7].
Among them, asynchronous online learning modes such as Moclass and Khan Academy are one of the main modes of online language learning, in which learners can independently choose learning contents and self-regulate learning time and progress according to their own life and learning situation, so they are favored by foreign language learners [8-10]. “China University Catechism” and “Wisdom Tree” are mainstream online learning platforms carrying a large number of language teaching videos with high click and participation rates [11-12]. Nowadays, the development of foreign language teaching and its better development in the future can not be separated from the support of network resources and information technology [13].
Feedback is one of the important measures to realize the “integration of learning and assessment”, and it is an indispensable means to ensure the quality of online learning. The implementation of online courses focuses on the participatory learning process of learners and the diversified evaluation mechanism supported by big data, rather than just assessing the learning results [14-15]. This kind of evaluation not only provides timely feedback to learners and improvement advice, but also helps students accumulate successful experiences and stimulate intrinsic learning motivation through communication and interaction between teachers and students and the role of incentives, thus cultivating students’ ability to self-monitor and reflect on their learning. More than twenty years of online education practice has also proved that the online teaching mode combining learning and evaluation can effectively ensure the effect of online learning [16-18].
In the face of the global impact of the new coronary pneumonia, the Ministry of Education has launched the initiative of “stopping classes and not stopping schools”, and online education has encountered unprecedented opportunities and challenges [19-20]. Due to the lack of planning for the implementation of online education in each school, the implementation of online education is significantly worse than that of traditional education. In addition to technical problems, teachers and students regard online education as an “online form” of traditional education, and online learning activities are not well-designed, resulting in insufficient or ineffective feedback [21-22].
Literature [23] summarized the research related to anxiety in online foreign language teaching, combined with the Anxiety Measurement Instrument to assess the anxiety of online foreign language learning students, and proposed some recommended techniques and interventions to alleviate students’ anxiety. Literature [24] conducted a controlled experiment to investigate how Google Classroom affects students’ foreign language learning, and based on the results of the experiment, it was found that students’ foreign language reading and writing scores improved to a certain extent, and at the same time, they showed positive attitudes towards Google Classroom. Literature [25] systematically reviewed the research literature on the development and practice of mobile language learning, and found that the number of related papers showed an upward trend, and the topics of the research were mainly students’ attitudes towards mobile learning, and online language learning technologies, including smart phones, smart assistants, etc. The research has deepened people’s knowledge and understanding of the online language learning model. Literature [26] examined English teachers’ perceptions of online English teaching modes from the dimensions of perceived usefulness, perceived ease of use, and attitudes toward online English learning modes, and found that although English teachers showed positive attitudes toward the usefulness and ease of use of online English teaching modes, more than half of them still did not recognize the effectiveness of online English teaching modes. Literature [27] based on GravCMS as a platform, researched the ADDIE teaching model introduced into the tour of Arabic learning application development for Arabic language learners to provide a new online learning experience, the study provides an important reference for the development of applications for online learning of other languages. Literature [28] conducted a comparative experiment to examine students’ perceptions of asynchronous tools (podcasts, video broadcasts, online tests, online glossaries, and forums) used for English language learning, and the results of the study showed that students believed that the asynchronous tools provided strong autonomy for students to meet the need for independent learning and to adapt to the pace of their individual learning. The research on online foreign language teaching covers a wide range of topics, and introduces and analyzes online language teaching models in a multidimensional and three-dimensional way. Important topics include teachers’ and students’ attitudes towards online language teaching models, the effectiveness of online teaching models, and an overview of cutting-edge research and the application of technological tools in online language teaching models.
Literature [29] describes the importance of learning feedback for learning scaffolding, and conceptualizes a centralized online teaching automatic feedback system to alleviate the workload of teachers, and reviews the relevant research literature on online learning feedback to improve the knowledge of the impact of online learning feedback mechanisms on students and teachers, as well as the difference between automatic and manual feedback. Literature [30] describes the operation logic and practical significance of online learning modes, and classifies and analyzes online learning technologies and tools based on machine learning algorithms, and classifies online learning into online learning modes with complete feedback, limited feedback, and no feedback based on whether online learning has a feedback mechanism, and discusses and compares them in detail. Literature [31] combined methodological tools such as questionnaires and mixed research method design to reveal that instructor’s feedback on students’ learning promotes students’ engagement in learning, while consulting with peers and reviewing relevant materials are the inferior alternatives for students to deal with feedback. Literature [32] investigated the integration path of virtual communities and online learning models and emphasized the importance of the application of feedback mechanisms after the integration of the two, concluding that feedback mechanisms help to enhance students’ motivation in learning. Scholars have revealed the positive role and operational logic of the learning feedback mechanism in online learning through literature review methods, empirical studies and other social practice approaches, and all believe that a reasonable and scientific feedback mechanism has positive significance for students’ online learning.
This paper proposes a method for processing and analyzing foreign language learners’ behavioral data on online platforms, develops a research context to provide environmental and technical support for the collection of complete data on foreign language learners’ behavior on online platforms, divides learners’ behavioral categories, and sets the research process. Data cleansing and data conversion methods are used to preprocess foreign language learner behavior data to obtain useful data for the study. The lagged sequence analysis method is proposed to create the sequence frequency conversion table, obtain the conditional probability table of sequence conversion, calculate the Z-score of sequence conversion, and finally create the sequence conversion graph for further inference and analysis of foreign language learners’ behavioral data. The foreign language learners’ behavioral data of “China University MOOC” online platform is taken as the research object to carry out in-depth data analysis, and the analysis results are used as the basis to put forward the construction strategy of the feedback mechanism for foreign language teaching on the online platform.
With the convergence and integration of Internet technology and information technology, which has triggered the rapid growth of data, the world has entered the era of big data. The flexible and changeable way of online learning gives foreign language learners greater freedom and choice, provides an effective way for lifelong learning, and at the same time accumulates a large amount of foreign language learner data. Accordingly, this paper proposes a targeted method for collecting and processing behavioral data from foreign language learners.
The design of a perfect data collection program is one of the key factors to ensure the completeness and reliability of research data collection. This study designs the research situation, the research object, the research data collection dimensions, the research process and methodology, with a view to providing environmental and technical support for the complete data collection of foreign language learners’ behaviors on the online platform of this study.
Instructional plot design is the primary issue in accomplishing this study. Through a complete instructional design, students are immersed in an instructional situation that realistically demonstrates their cognitive processes. The design of the research context in this study should take into account the stages of the students’ learning process, and at the same time, it should pay more attention to the fact that the context is helpful for understanding and differentiating students’ cognitive characteristics. Specifically, the learning process in the research context matches the cognitive process of students’ perception, information processing, memory, thinking and problem solving, and the design of the activities and learning materials can support the behavioral operations of students at each stage of the process and the demand for research data collection.
In this study, the practical research object is the foreign language learners on the “China University MOOC” online platform. In addition to the majority of students, there are also in-service front-line teachers and interactive whiteboard enthusiasts, distributed all over the country, and the diversity of learners in terms of gender, geography, education, and work is relatively obvious.
The learner behaviors of online foreign language learners are divided into four categories, namely, “course learning - task completion - human-computer interaction - learning performance”, so how to characterize students’ online foreign language learning behaviors is the basis and the key of the next characterization based on the foreign language learning behaviors, and it is also the focus and difficulty of the research on the related categories. By analyzing the students’ actions recorded in the course log sheet, the study filtered out the types of episodic actions that occurred rarely, focused on 21 foreign language learning actions, and summarized them into 15 behaviors, as shown in Table 1.
Student’s behavior
| Dimension | Behavior | Coding |
|---|---|---|
| Man-machine interaction | Posting | F1 |
| Search | F2 | |
| Browse post | F3 | |
| Edit post | F4 | |
| Task completion | Browse test | Q1 |
| Making topic | Q2 | |
| Save the answer and check in | Q3 | |
| Submit questions | Q4 | |
| Check out the questions | Q5 | |
| Learning performance | Job score | A1 |
| Online duration score | A2 | |
| Post score | A3 | |
| Test score | A4 | |
| Course learning | Browse the content page | R1 |
| See the page documentation resources | R2 |
The design of this study focuses on solving the problem of collecting data on students’ cognitive characteristics characterizing their behaviors, while in the process of collection, attention must be paid to the problem of external factors interfering with the performance of students’ learning behaviors.
The design of the data collection process should be centered on the theory of humanism, so that students can learn independently without teacher intervention, and the process of students’ knowledge construction can be revealed naturally.On the one hand, students achieve behavioral data recording through independent learning, and on the other hand, they demonstrate the cognitive process in a self-directed context.
A total of 5,287 valid data were collected on the “China University MOOC” online platform. These data not only summarize students’ personal information and learning behaviors, but also include their total achievements in learning. In order to ensure the accuracy and usability of the data, all the collected data were organized and stored in the Exce1 form, which is convenient for subsequent data analysis and research work.
At this stage, the primary component is the pre-processing of the collected data to address issues such as data clutter, inconsistent formatting, and omissions to ensure the accuracy and usability of the data. Common data preprocessing methods include data integration, data cleansing, data normalization, and data transformation, which help to improve the overall quality and usability of data. Data integration involves the integration, merger, and storage of information from multiple data sources in a unified manner to achieve effective organization and management of data content.The main goal of data cleansing is to address missing and abnormal data and standardize data in different formats.The main task of data transformation is to reduce the quantitative inconsistencies between the data, so that the data can meet the standardization criteria.The so-called data constraint is designed to minimize the complexity of the data while ensuring its original state.
(1) Data cleaning [33]. Due to system aspects, some fields may be missing when collecting or transmitting data, or data for some parameters may be missing. In order to ensure the integrity of the data, data cleaning is performed on the basic information table and the training and learning behavior data table. Therefore, it is necessary to perform data cleaning on the basic information data and learning behavior data table obtained from students’ foreign language learning. After testing, it was found that the learning behavior data table did not show abnormal values, but it did have more null values existed, such as no learning and other cases, so choose to automatically fill the null value of 0 or the adjacent value to make up the processing mode of operation.
(2) Data conversion [34]. To avoid errors caused by differences in data units and to standardize the format of behavioral data, standardization is being performed for different types of data. This involves a linear transformation of the original data to map the result to the range of [0,1], i.e., the data related to online learning behaviors become values between [0,1]. The linear transformation strategy adopted here is to utilize the maximum-minimum normalization method, and the following are the related detailed formulas:
The final data preprocessing resulted in 5083 valid data, which can be used for subsequent studies.
Lagged series analysis aims to assess the probability of occurrence of serial behavior over time [35]. The method is mainly used to test the statistical significance of the occurrence of one state of an unknown system followed by the immediate occurrence of another state immediately after it. If the system is in state A at moment
The general steps in carrying out the lag sequence analysis method are as follows.
1) First encode each behavior or event in the sequence and create a sequential frequency conversion table for each code by calculating the frequency of transitions between codes.
2) Use the sequential frequency conversion table obtained in the first step to compute the conditional probabilities of transitions between codes to obtain a table of conditional probabilities of sequence transitions.
3) Use the sequence frequency conversion table to compute the expectation table for the entire code conversion process.
4) Calculate the Z-score (adjusted residuals table) for the sequence conversions using the three tables listed earlier and determine whether the transfer continuity is significant for each sequence separately [36].
5) Finally, sequence transformation plots were created where each point represented a coding behavior and these points were connected with arrow lines for further inference and analysis.
In this chapter, the data collection, preprocessing and lag sequence analysis methods proposed in this paper will be used to explore and analyze the behavioral data of foreign language learners on the MOOC platform of Chinese universities, so as to provide reference and basis for the construction of the subsequent teaching feedback mechanism.
Individual learners vary greatly by gender, age, occupation, past experience, etc., which requires that individual learner profiles be analyzed with as much detailed data as possible to provide a detailed, three-dimensional, yet intuitive description of each individual.
Students’ learning activities are divided into four categories: “course learning - task completion - human-computer interaction - learning performance”, and in this section, in order to explore the significant correlation between the four, the results of the local analysis are shown in Figure 1. The “**” in the figure represents the significant correlation (Sig<0.01) between course learning, task completion, human-computer interaction and learning performance. Therefore, this study can analyze the group portrait of foreign language learners based on four dimensions of e-learning behaviors: course learning, task completion, interpersonal interaction, and learning achievement.
Significant correlation
To further analyze learner groups and study the characteristics of e-learning behaviors of different learner groups, this study uses a clustering algorithm to cluster learners. Due to the inconsistency of the dimensions of “course learning”, “task completion”, “interpersonal interaction” and grades, the SPSS tool was used to save the standardized scores as variables to standardize the data, and then run the clustering algorithm command to obtain the clustering analysis results, which are shown in Table 2. As can be seen from the table, learners can be effectively clustered into 3 classes according to course learning characteristics, interpersonal interaction characteristics, task completion characteristics, and academic performance. Class Cluster 1 and Class Cluster 3 have higher motivation for course learning and Class Cluster 2 has lower motivation; Class Cluster 3 was more active in human interaction, while Class Cluster 1 and Class Cluster 2 were less active; in terms of task completion, Class Cluster 3 was the highest, Class Cluster 1 was the next highest, and Class Cluster 2 was the least; Class Cluster 2 has relatively low performance and Class Cluster 1 & Class Cluster 3 have relatively high performance. Cluster analysis of the learner groups provides the following characteristics of the different learner groups.
Cluster result
| - | Clustering | ||
|---|---|---|---|
| 1 | 2 | 3 | |
| Zscore(course learning) | 0.52105 | -0.76607 | 1.97744 |
| Zscore(interpersonal interaction) | 0.08884 | -0.34337 | 1.45156 |
| Zscore(mission accomplished) | 0.19667 | -0.44222 | 1.53762 |
| Zscore(score) | 0.37158 | -0.385 | 0.62539 |
| Course learning | 233 | 74 | 415 |
| Interpersonal interaction | 50 | 25 | 126 |
| Task completion | 68 | 46 | 125 |
| Grade | 85 | 66 | 90 |
| N | 1945 | 2625 | 513 |
This group of learners, although not excellent, has a high frequency of course attendance and relatively high academic achievement. The provision of more instruction and intervention by teachers to this group of learners can lead to higher learning outcomes and a reduced risk of attrition.
Although this group of learners have passing grades, the per capita frequency of their course learning is much lower than that of the other two clusters, probably because of two reasons: one is that they choose the course in a curious mindset to see what it is like, without thinking of passing the course or getting a certificate, just for the sake of understanding; the other is that they may be at a higher level, and the content of the course is relatively simple to them, which can’t stimulate their interest in learning. This type of learner will be lost to a greater extent and needs to be emphasized by teaching stakeholders.
This group of learners has a high frequency of coursework, a lot of interpersonal interactions, and task completion. They have excellent grades, high motivation, and a strong willingness to communicate. This group of learners is a high-quality group of learners who serve as role models for other learners and are the backbone of the entire course, and the likelihood of attrition for this group of learners is very low. They will be even better if they receive guidance from their teachers.
In this study, SPSS linear regression analysis was adopted, with “course learning behavior”, “human-computer interaction behavior” and “task completion behavior” as independent variables and “learning performance” as the dependent variable, and the analysis results are shown in Table 3. From the table, it can be seen that in stepwise regression model 1, the independent variable (course learning behavior) can effectively predict the change of the dependent variable (academic performance) (R2=0.182, Sig<0.05); in stepwise regression model 2, the independent variables (course learning behavior, task completion behavior) can effectively predict the change of the dependent variable (academic performance) (R2=0.216, Sig<0.050). In addition, the results of the analysis showed that learners’ interpersonal interaction behaviors had no significant effect on academic performance.
Results of linear regression analysis
| Model | R | R2 | Sum of squares | freedom | Mean square | F | Sig. | |
|---|---|---|---|---|---|---|---|---|
| 1 | Regression | 0.415a | 0.182 | 439488.78 | 1 | 439488.7 | 1118.66 | 0.000b |
| Residual error | - | - | 1982801.27 | 5054 | 395.62 | - | - | |
| Total | - | - | 2422290.05 | 5055 | - | - | - | |
| 2 | Regression | 0.452b | 0.216 | 521282.68 | 2 | 260682.8 | 695.39 | 0.000c |
| Residual error | - | - | 1902227.56 | 5053 | 378.51 | - | - | |
| Total | - | - | 2423510.24 | 5055 | - | - | -- | |
Dependent variables-learning grades
Prediction variable- (constant), course learning behavior
Prediction variable- (constant), course learning behavior, task completion behavior
The results of the regression analysis with “course learning behavior” and “task completion behavior” as independent variables and “academic performance” as dependent variable are shown in Table 4. As can be seen from the table, in stepwise regression model 1, the standardized regression coefficient of the independent variable (course learning behavior) is greater than 0 (standardized regression coefficient=0.435> 0.00, Sig=0.000<0.050), which indicates that the learner’s course learning behavior has a positive impact on the learner’s performance; in stepwise regression model 2, the standardized regression coefficients of the independent variables (course learning behavior, task completion behavior) regression coefficient is greater than 0 (standard regression coefficient of course learning behavior=0.267> 0.000, standard regression coefficient of task completion=0.245>0.000, Sig=0.000< 0.050), which can be known that learner’s course learning behavior and task completion behavior have positively correlated influence on learner’s achievement.
Regression analysis based on the “study achievement”
| Model | Nonnormalized coefficient | Standard coefficient | T | Sig. | ||
|---|---|---|---|---|---|---|
| B | Standard error | |||||
| 1 | (Constant) | 64.826 | 0.456 | 138.758 | 0 | |
| Course learning behavior | 0.072 | 0.003 | 0.435 | 33.352 | 0 | |
| (Constant) | 61.472 | 0.525 | 119.348 | 0 | ||
| 2 | Course learning behavior | 0.052 | 0.002 | 0.267 | 17.262 | 0 |
| Task completion behavior | 0.136 | 0.009 | 0.245 | 14.747 | 0 | |
In the above paper, this paper classifies online platform foreign language learners’ behaviors into four dimensions: course learning, task completion, human-computer interaction, and learning performance, and proposes 15 specific behaviors corresponding to the dimensions. The study further utilizes lagged sequence analysis to construct a behavioral transformation model of online learning for student samples. In this section, the study is first based on generating behavioral sequence frequencies for the student samples to form a frequency table of conversion between the 15 learning behaviors, as shown in Table 5. The row behavior in the table represents the behavior that occurs first, the corresponding column behavior of the row represents the behavior that occurs immediately after the row behavior, and the data in the table represent the number of times the behavior sequence formed by converting the row behavior into the column behavior appears. From the table, it can be quickly found that the sequences of behaviors that occur more often are Q1Q2 (10671 times) and R1Q1 (7871 times).
Conversion frequency table
| n | A1 | A2 | A3 | A4 | F1 | F2 | F3 | F4 | R1 | Q1 | Q2 | Q3 | Q4 | Q5 | R2 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| A1 | / | 627 | 3 | 16 | 0 | 0 | 90 | 0 | 175 | 124 | 6 | 0 | 0 | 1 | 5 |
| A2 | 135 | / | 0 | 235 | 0 | 0 | 38 | 0 | 88 | 48 | 1 | 0 | 0 | 2 | 1 |
| A3 | 20 | 0 | / | 0 | 0 | 0 | 3 | 0 | 12 | 1 | 0 | 0 | 0 | 0 | 0 |
| A4 | 262 | 3 | 0 | / | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 |
| F1 | 2 | 0 | 0 | 2 | / | 0 | 1078 | 0 | 8 | 2 | 2 | 0 | 0 | 0 | 0 |
| F2 | 4 | 0 | 0 | 0 | 5 | / | 166 | 0 | 28 | 3 | 1 | 0 | 0 | 0 | 0 |
| F3 | 132 | 11 | 3 | 1 | 1085 | 220 | / | 60 | 962 | 166 | 45 | 2 | 3 | 15 | 28 |
| F4 | 0 | 0 | 0 | 0 | 0 | 0 | 62 | / | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| R1 | 281 | 12 | 12 | 2 | 18 | 2 | 768 | 0 | / | 7871 | 213 | 8 | 10 | 30 | 385 |
| Q1 | 110 | 3 | 3 | 2 | 1 | 2 | 3 | 0 | 1985 | / | 10671 | 31 | 18 | 1315 | 65 |
| Q2 | 26 | 0 | 3 | 2 | 6 | 0 | 95 | 0 | 972 | 1350 | / | 8421 | 82 | 1635 | 46 |
| Q3 | 2 | 0 | 0 | 2 | 0 | 0 | 10 | 0 | 128 | 195 | 1132 | / | 7057 | 65 | 3 |
| Q4 | 0 | 0 | 0 | 0 | 0 | 0 | 6 | 0 | 9 | 36 | 25 | 20 | / | 7093 | 0 |
| Q5 | 60 | 5 | 2 | 0 | 4 | 0 | 265 | 0 | 1209 | 5578 | 1558 | 207 | 45 | / | 28 |
| R2 | 18 | 2 | 0 | 0 | 0 | 0 | 38 | 0 | 195 | 213 | 14 | 3 | 0 | 4 | / |
The Z-scores of the frequency of occurrence of each sequence were calculated on the basis of the behavioral transition frequency table thus obtaining the residual table as shown in Table 6. Only when the Z-score is greater than +1.96 does it mean that the frequency of occurrence of the corresponding behavioral sequence has reached the statistical significance level (p < 0.05). From the table, it can be seen that there are six behavioral sequences whose probability of occurrence reached the level of significance, namely R1Q1 (5.66), Q1Q2 (7.42), Q2Q3 (6.14), Q3Q4 (5.12), Q4Q5 (5.25), and Q5Q1 (3.82). Among them, R1Q1, Q1Q2, Q4Q5, and Q5Q1 reflect the gradual cognitive deepening of students experiencing the online learning process, and Q2Q3 and Q3Q4 reflect students’ management of their testing behavior for the purpose of course learning objectives or excellent grades, which is an indication of self-monitoring.
Z-score
| Z | A1 | A2 | A3 | A4 | F1 | F2 | F3 | F4 | Q1 | Q2 | Q3 | Q4 | Q5 | R1 | R2 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| A1 | / | 0.022 | -0.21 | -0.21 | -0.24 | -0.25 | -0.16 | -0.25 | -0.13 | -0.17 | -0.24 | -0.23 | -0.24 | -0.24 | -0.22 |
| A2 | -0.14 | / | -0.24 | -0.05 | -0.26 | -0.26 | -0.18 | -0.25 | -0.15 | -0.17 | -0.24 | -0.25 | -0.23 | -0.25 | -0.26 |
| A3 | -0.21 | -0.22 | / | -0.27 | -0.23 | -0.24 | -0.24 | -0.27 | -0.21 | -0.22 | -0.25 | -0.25 | -0.23 | -0.26 | -0.27 |
| A4 | -0.07 | -0.21 | -0.26 | / | -0.23 | -0.23 | -0.24 | -0.26 | -0.23 | -0.21 | -0.23 | -0.22 | -0.24 | -0.21 | -0.26 |
| F1 | -0.24 | -0.22 | -0.21 | -0.21 | / | -0.25 | 0.56 | -0.23 | -0.26 | -0.23 | -0.27 | -0.27 | -0.25 | -0.27 | -0.24 |
| F2 | -0.22 | -0.24 | -0.25 | -0.21 | -0.23 | / | -0.13 | -0.21 | -0.24 | -0.21 | -0.22 | -0.23 | -0.22 | -0.25 | -0.25 |
| F3 | -0.16 | -0.24 | -0.27 | -0.26 | -0.56 | -0.05 | / | -0.17 | 0.51 | -0.15 | -0.23 | -0.23 | -0.26 | -0.23 | -0.23 |
| F4 | -0.22 | -0.26 | -0.25 | -0.23 | -0.25 | -0.24 | -0.22 | / | -0.25 | -0.24 | -0.25 | -0.21 | -0.22 | -0.21 | -0.25 |
| R1 | -0.02 | -0.23 | -0.23 | -0.25 | -0.26 | -0.24 | 0.37 | -0.21 | / | 5.66* | -0.1 | -0.24 | -0.24 | -0.18 | 0.08 |
| Q1 | -0.13 | -0.21 | -0.26 | -0.25 | -0.22 | -0.26 | -0.23 | -0.27 | 1.26 | / | 7.42* | -0.24 | -0.23 | 0.73 | -0.16 |
| Q2 | -0.22 | -0.24 | -0.2 | -0.26 | -0.21 | -0.26 | -0.19 | -0.24 | 0.49 | 0.74 | / | 6.14* | -0.19 | 1 | -0.22 |
| Q3 | -0.22 | -0.21 | -0.26 | -0.23 | -0.25 | -0.24 | -0.21 | -0.24 | -0.16 | -0.11 | 0.62 | / | 5.12* | -0.18 | -0.21 |
| Q4 | -0.24 | -0.23 | -0.27 | -0.25 | -0.26 | -0.21 | -0.24 | -0.26 | -0.21 | -0.23 | -0.24 | -0.23 | / | 5.25* | -0.23 |
| Q5 | -0.21 | -0.22 | -0.21 | -0.22 | -0.22 | -0.26 | -0.06 | -0.24 | 0.62 | 3.82* | 0.91 | -0.07 | -0.23 | / | -0.2 |
| R2 | -0.22 | -0.24 | -0.23 | -0.21 | -0.24 | -0.22 | -0.19 | -0.24 | -0.08 | -0.08 | -0.24 | -0.22 | -0.21 | -0.25 | / |
Based on the above six behavioral sequences that have reached the significance level, the online learning behavioral transition diagram is constructed as shown in Figure 2. The values above the arrows in the figure represent the Z value of each behavioral sequence, and the arrows represent the direction of behavioral conversion. From the figure, it can be seen that if students first browse the course content page after logging into the platform, then they are less likely to go to the forum to communicate afterward, and more likely to go to the test module to try the test tasks or simulated test questions. Students’ activities in the testing module usually follow progressive evaluation steps, such as browsing the questions and then doing them (Q1Q2), saving this attempt and checking the progress of answering the questions after doing them (Q2Q3), submitting the test questions after the last answer (Q3Q4), reviewing the test questions and checking the answers after submitting them (Q4Q5), and tending to browse the other test questions or the questions they have already done after checking the answers (Q5Q1). The course allows students to complete the test questions in multiple sessions, and the higher probability of occurrence of Q2Q3 and Q3Q4 suggests that students tend to complete a test task in multiple sessions, saving their answer progress after each answer but not submitting the test questions, and submitting the test questions after the last answer, when the student views their answer progress and finds that they have completed all the questions or have met the expectations.The presence of the Q4Q5 sequence in the figure suggests that students tend to test activities in which they evaluate their learning in a relatively complete manner, and they tend to know the correct answers and reinforce and refine what they have learned in the process of checking against the answers.The presence of the Q5Q1 sequence reflects the willingness of students to put more effort into getting better performance and meeting course objectives during the testing activity.
Learning behavior transformation diagram
Aiming at the problem of teaching feedback reflected in the analysis of foreign language learners’ behavioral data on the online platform, an effective Internet-based teaching feedback system is established before, during, and after class.
1) Setting up a feedback incentive mechanism for extracurricular interaction and classroom interaction.
Out-of-class interaction incentive. On the interactive communication network platform, students can raise questions or reply to questions at any time and any place through computers, cell phones or IPADs and other user terminals in the process of learning outside the classroom, and teachers regularly evaluate students’ questions and replies.
Classroom interaction incentives. In order to promote student participation in the classroom, encourage students to interrupt the teacher in the classroom process, at any time to put forward the problems and errors in the course of the lecture and participate in the discussion, for valuable interruptions, or lead to meaningful classroom discussion, according to the quality of different bonus points. This requires the teacher of the class to prepare some traps in the preparation of the class to guide students to ask questions.
2) Establishment of assessment and evaluation mechanism.
The method of “strengthening the usual assessment and downplaying the final examination” has been formulated, so that students can constitute the final grade by a certain proportion of the teacher’s bonus points obtained through the online interactive communication platform, the bonus points obtained in the classroom and the students’ question points, and if the score is more than 70, the students can obtain the qualification of exempting from the final examination. Similarly, exemptions can be obtained through pre-course test scores. For students who qualify for exemptions, if they take the final exam, the higher their exam score or exemption score will be the student’s final grade.
Aiming at the complicated data of foreign language learners’ learning behaviors on the online platform, this paper proposes the corresponding data collection, preprocessing and lag sequence analysis methods, and takes the foreign language learners’ behavioral data on the online platform of “MOOC for Chinese Universities” as the object of the study to carry out an in-depth data analysis. The four categories of foreign language learner behavior, namely, course learning, task completion, human-computer interaction, and learning performance, are all significantly correlated (Sig<0.01), and three major groups of learners, namely, high immersion learners, low immersion learners, and high immersion learners, can be obtained through cluster analysis. Through linear regression analysis, it can be learned that learners’ interpersonal interaction behaviors have no significant effect on academic performance (Sig>0.05), while course learning behaviors, task completion behaviors, and course learning behaviors all have a positive correlation effect on learners’ performance (Sig<0.05). While in the lagged behavioral sequence analysis, the behavioral sequences with more occurrences were Q1Q2 (10671 times), R1Q1 (7871 times), and in the z-score of the frequency of occurrence of each sequence, R1Q1 (5.66), Q1Q2 (7.42), Q2Q3 (6.14), Q3Q4 (5.12), Q4Q5 (5.25), and Q5Q1 (3.82) were all greater than + + 1.96, and the probability of occurrence of the behavioral sequences reaches the significance level. Combined with the results of foreign language learners’ behavioral data analysis on the “China University MOOC” online platform, we propose a strategy for the construction of a feedback mechanism for foreign language teaching on the online platform in terms of interactive feedback incentive mechanism and assessment and evaluation mechanism.