This paper uses a deep learning algorithm to extract training samples for hybrid teaching resources in the context of Internet+. The support vector machine establishes a linear regression function, and a nonlinear function is used to map the samples to a high-dimensional feature space for solving. To optimize the training problem, a deep neural network is employed to construct a neuronal structure model and calculate the loss function. The text vectorization method was selected as the input layer processing of the convolutional neural network, and the index scores of the hybrid teaching model in the context of Internet+ were all 6 and above, with the highest score reaching 10. Therefore, online and offline hybrid teaching modes can provide a more flexible and optimized teaching method and promote resource sharing and optimization.