To provide teaching assistance to secondary school mathematics teachers in ethnic areas and improve the teaching effectiveness of mathematics courses, a hybrid teaching model based on deep learning algorithms is proposed. Convolutional neural networks and joint probability matrix decomposition are fused to design teaching resource recommendation methods, and hybrid teaching is carried out according to three stages: before, during, and after class. The overall mean of the questionnaire for the experimental class improved from 2.39 to 3.01, and the pass rate, merit rate, and mean score of mathematics scores increased by 13.79%, 9.02%, and 15.17 points, respectively. This method enables educational technology to be effective in mathematics curriculum and improve the quality and information of mathematics education in ethnic areas.