1 Research Background

This section elaborates on the various challenges faced by classroom teaching assessment in international Chinese language education under the backdrop of educational modernization policies, such as insufficient objectivity of evaluation, high costs, limitations of time and space, impact on teaching states, and information lag. To address these issues, the research team developed an artificial intelligence-based system for recognizing and analyzing classroom teaching behaviors. The system aims to automatically collect and analyze classroom teaching behavior data from international Chinese language teaching videos, establish a basic model for offline classroom teaching, and thereby promote the standardization and normalization of international Chinese language education.

2 Introduction of Research Theory and Tools

This section introduces the theoretical foundation for classroom teaching behavior analysis based on the Flanders Interaction Analysis System, including the key focus areas for verbal interaction behaviors, the method of coding and recording classroom segments in 3-second units, and the application of statistical processing. It details the seven modules of the traditional Flanders Interaction Analysis System, such as transition matrix analysis, classroom structure analysis, teacher tendency analysis, etc., and explains the specific analysis content and function of each module. Additionally, it mentions the 41 analysis items developed based on this analytical method, as well as the research objective of constructing a basic model through analyzing offline Chinese teaching videos.

3 Process and Methods for Constructing the Basic Teaching Model

This section elaborates in detail on the process and methods for constructing the basic model for offline classroom teaching in international Chinese language education. First, representative offline Chinese classroom teaching videos were selected as samples; these samples come from different countries and teaching contexts, covering various types of lessons. Next, the coding system of the Flanders Interaction Analysis System was employed to identify and code teaching behaviors, with adjustments and supplements made according to the characteristics of Chinese classrooms. Then, a large amount of real international Chinese language teaching case data was collected, and a predictive model for classroom teaching behavior recognition was obtained through manual annotation and deep learning training, achieving relatively high accuracy. Finally, based on the recognition results of classroom teaching behaviors, the basic model for offline classroom teaching in international Chinese language education was constructed.

4 Construction of the Basic Teaching Model

This chapter focuses on constructing the basic model for offline classroom teaching in international Chinese language education. Drawing on the concept of norms from educational measurement and referencing the norm construction procedures of the Flanders Interaction Analysis System (FIAS), it conducts a quantitative analysis of offline Chinese classroom teaching behaviors. The research involves 41 items across seven modules, covering data characteristics such as mean, standard deviation, and maximum frequency.

In the transition matrix analysis, the steady-state cell data for offline Chinese classrooms show an average of 168.16 instances of continuous teacher lecturing behavior and an average of 25.77 instances of continuous student expression behavior. The mean of the positive integration cell is 26.08, accounting for 3.48%, which is lower than the expected proportion, indicating that teachers' behaviors encouraging student insights are not prominent. The mean of the defect cell is 23.9, accounting for 3.2%, lower than the expected proportion, suggesting that teachers less frequently directly control the classroom or reprimand students.

The classroom structure analysis reveals that the mean ratio of teacher behavior in offline Chinese classrooms is 63.29%, higher than that in online classrooms but lower than the FIAS norm. The mean ratio of student behavior is 13.69%, lower than both online classrooms and the FIAS norm. The mean ratio of ineffective behavior is 22.85%, higher than both online classrooms and the FIAS norm. The mean ratio of teacher questioning is 16.75%, lower than both online classrooms and the FIAS norm. The mean ratio of teacher behavior driven by students is 56.86%, higher than both online classrooms and the FIAS norm. The mean ratio of student-initiated behavior is 38.01%, higher than the FIAS norm. Among the newly proposed indicators, the mean teacher encouragement ratio is 3.69%, the mean teacher opinion adoption ratio is 1.73%, the mean teacher lecturing ratio is 33.75%, the mean teacher criticism ratio is 0.05%, the mean teacher instruction ratio is 5.87%, the mean student response ratio is 8.84%, and the mean leading reading/echo reading ratio is 4.85%.

In the teacher tendency analysis, the mean ratio of indirect to direct influence is 14.64%, indicating that teachers tend to directly control teaching activities and students. The mean ratio of positive to negative influence is 172.44%, suggesting that teachers have a positive reinforcement tendency, but with a large standard deviation, indicating significant differences among teachers.

The classroom emotional atmosphere analysis shows that the mean ratio of the positive integration cell is 3.48%, lower than the expected proportion. The mean ratio of the defect cell is 3.2%, slightly lower than the expected proportion. The mean ratio of the content cross cell is 38.94%, higher than the expected proportion. Among the new indicators, the mean positive feedback ratio is 6.23%, the mean progressive positive emotion ratio is 1.75%, the mean negative feedback ratio is 5.92%, and the mean progressive negative emotion ratio is 0.

In the analysis of classroom behavior interaction patterns, within the "question and answer" pattern, the sequence pair (4,4) appears most frequently, reaching 9,032 times, indicating that continuous teacher questioning is common. Within the "creative inquiry and answer" pattern, the (9,9) pattern appears 51 times, the (3,3) pattern appears 44 times, the (8,9) pattern appears 14 times, while other patterns appear rarely or not at all. Among the new indicators, the mean questioning success rate is 29.77%, the mean questioning conciseness rate is 44.07%, the mean response feedback rate is 2.6%, the mean response continuity rate is 34.82%, the mean instruction feedback rate is 3.65%, the mean instruction conciseness rate is 49.54%, the mean void intervention rate is 38.27%, the mean void continuity rate is 61.73%, the mean criticism continuity rate is 35.97%, the mean criticism feedback rate is 7.02%, and the mean follow-up questioning behavior ratio is 9.05%.

The classroom behavior chart analysis, presented through dynamic characteristic curves, shows the dynamic changes in teacher and student behaviors. The mean number of times teacher behavior ratio exceeds 50% is 115.05, and the mean number of times student behavior ratio exceeds 50% is 21. The duration where teacher behavior ratio exceeds 50% accounts for 64.22% of the total class time, and the duration where student behavior ratio exceeds 50% accounts for 14.53%.

In the analysis of key classroom characteristics, the mean conciseness of teacher language is 39.18%, higher than the expected proportion. The mean speaking rate is 209.05 characters per minute, which is relatively slow. Among commonly used modal particles, "啊 (a)", "嗯 (en)", "呃 (e)", and "呀 (ya)" appear most frequently.

Different lesson types exhibit variations in teaching behaviors. For example, in comprehensive lessons, teacher lecturing and student echo reading behaviors are prominent, whereas in speaking lessons, student-initiated expression behaviors are more frequent.

5 Conclusions and Implications

This section summarizes the characteristics of teaching behaviors in offline Chinese classrooms, pointing out that teacher behavior ratios are high while student behavior ratios are low, teacher questioning ratios are low and language use needs optimization, leading reading/echo reading is prominent in comprehensive lessons, and interaction is frequent in speaking lessons. Classrooms have advantages in emotional communication, but the efficiency of teacher-student interaction and student participation need improvement. Future directions for teaching improvement are proposed, emphasizing balancing teacher-student relationships and utilizing technology to enhance efficiency. Simultaneously, future research directions are suggested, including refining the classification and calculation methods of teaching behavior variables, and exploring the application of the model in different teaching scenarios.