Research on personalized recommendation of teaching resources based on joint probability matrix decomposition model and CNN improvement algorithm

Abstract

This study proposes a teaching resource recommendation method (TRRDLMF) based on deep learning and probabilistic matrix decomposition, aiming to improve teaching resource recommendation’s accuracy and personalization level. The study combines hybrid neural network feature extraction of teaching resources and user feature extraction based on extended noise-reducing self-encoder and extended probabilistic matrix decomposition-based teaching resource recommendation technique. The study validates the model’s effectiveness by conducting empirical analyses on four online learning platform datasets of different sizes. The recommendation system can effectively track learners’ knowledge mastery status and learning preferences, and provide corresponding teaching resources. Test question similarity and difficulty analysis results show that the model can accurately capture the correlation between test questions and provide complementary data in education. The personalized recommendation analysis reveals the learners’ knowledge preference states that change with learning, demonstrating the model’s ability to give explanations at the level of learners’ knowledge preferences. The analysis of the impact of data size on recommendation results shows that the recommender system can achieve up to 73.12% accuracy and 97.89% recall for different TOP-K recommendation lists, with the best F-value at TOP-6. The personalized recommendation system for teaching resources based on the joint probability matrix decomposition model and CNN improvement algorithm proposed by the research demonstrates significant effects in improving the accuracy and personalization level of teaching resources recommendation, which helps learners to acquire suitable learning resources more efficiently, and has high practical value and promotion potential.

Keywords

• deep learning
• probability matrix decomposition
• feature extraction
• similarity
• teaching resources recommendation
• 97m50