Leveraging Deep Learning to Optimize AI Self-Efficacy Development through Tailored Prompt Engineering Interventions

Abstract

Artificial Intelligence (AI) self-efficacy is an essential component in the transformation of education, especially in improving the learning outcomes among students, as AI can do. Nevertheless, existing systems usually do not offer individualized learning opportunities, which restricts the potential of students. The proposed study will present a Position Attention -based Long Short-Term Memory (PA-LSTM) model, which identifies the level of AI self-efficacy in students based on the processing of real-time interaction information. The model is a combination of temporal modeling and position-based attention, which considers both behavioral and contextual aspects of interactions between students and AI. Such data preprocessing methods as cleaning, normalization, and text tokenization guarantee the provision of reliable and stable input to the model. The extraction of the features incorporates the student prompts and behavioral indicators including response time, edits and self-efficacy ratings. The proposed model has always shown better performance, with the accuracy of 97.5% to 99.5%. These findings underscore the strength of the model to give individualized feedbacks and effectively categorize learning outcomes, therefore, improving AI self-efficacy building in classrooms.