Ideological and political education behavior recognition by artificial intelligence transformers

Area of Science:

• Artificial Intelligence
• Educational Technology
• Computer Vision

Background:

• Traditional classroom observation for ideological and political education (IPE) is subjective, costly, and limited in analyzing multimodal temporal interactions.
• Existing methods struggle with scalability and objectivity in modeling complex classroom dynamics.

Purpose of the Study:

• To develop an objective and scalable artificial intelligence (AI) framework for behavior recognition in IPE classrooms.
• To enhance the modeling of multimodal temporal interactions for improved educational process evaluation.

Main Methods:

• Proposed a Transformer-based AI framework for behavior modeling and recognition in IPE settings.
• Conducted experimental evaluations comparing the model against Video Masked Autoencoders V2 (VideoMAE V2) and VideoMamba.
• Investigated robustness and generalization capabilities across various challenging conditions.

Main Results:

• Achieved high accuracy (ACC) of 0.905 and area under the curve (AUC) of 0.958 on Mixed-Condition Clips.
• Demonstrated robust accuracy (R-ACC) of 0.883, 0.869, and 0.875 on Low-Resolution, Viewpoint-Shift, and Background-Noise Clips, respectively.
• Exhibited reliable confidence estimates with a low Brier Score (BS) of 0.091.

Conclusions:

• The Transformer-based model effectively mitigates the impact of input quality degradation and data distribution shifts.
• The developed framework offers valuable insights for educational behavior recognition and intelligent evaluation in IPE.
• The study enhances objectivity and scalability in IPE classroom process evaluation through advanced AI.