Leveraging AI Technology for Coding the Classroom Observation Record Form of Flanders Interaction Analysis

This paper addresses several challenges in classroom interaction analysis, including incomplete data collection, coarse-grained behavior classification, subjectivity in analysis, and the difficulty of aligning traditional methods with the evolving demands of modern education for precise and efficient teaching evaluations. We propose an automatic coding method for the Flanders Interaction Analysis (FIA) classroom observation record form (CORF) by integrating various artificial intelligence technologies to tackle these issues. Initially, audio is extracted from recorded classroom videos to isolate the speech of both teachers and students. Next, we employ the Paraformer model for speaker separation and speech recognition, converting the extracted audio into corresponding text and timestamp information for each speaker. Simultaneously, a pre-trained BERT module is fine-tuned on all spoken text. Finally, this fine-tuned BERT model is used to classify the text into ten distinct categories of interaction behavior, effectively generating the coding necessary for the CORF of FIA. Experimental results reveal that our proposed method achieves an accuracy of 94.63% in speaker separation, 95.32% in speech recognition, and 87.95% in agreement with manual annotations during speech behavior analysis. These findings indicate that our method is highly effective for automating the coding process of the CORF for the FIA system.