With the rapid development of biosensing technology, the field of ideological and political education has also ushered in a new era of intelligent education. This paper proposes a method for evaluating the effectiveness of ideological and political education by combining facial expression and behavior recognition of college students. First, a student facial expression recognition model based on a deep attention network is proposed. This model learns the facial expression features of students, fuses multiple facial expression features, and classifies them. A student behavior recognition algorithm based on multi-task learning is proposed, using an object detector to extract data from videos as algorithm input. Through a multi-task heatmap network module, intermediate heatmaps are extracted and encoded into private heatmaps to obtain student joint position information. Subsequently, behavior vectors and metric vectors are introduced to model student classroom behavior separately. The obtained behavior states are combined with expression categories, and the two types of data are jointly input into the model for training, enabling dynamic evaluation of the effectiveness of ideological and political education for students. Experiments show that the evaluation model integrating expressions and behaviors achieves an accuracy rate of 85.44%, effectively overcoming evaluation biases caused by single-dimensional features. In practical applications, the model achieves an overall evaluation accuracy rate of 94%, comparable to manual detection levels, providing efficient and intelligent technical support for ideological and political classroom teaching.
