In the context of information security, the aesthetics and data privacy security of multinational children’s songs are studied. The research collects the data of foreign children’s songs, and on the basis of the collected data, the mathematical model of the children’s song text is constructed by using a random matrix. An industry consisting of various media entities that disseminate information is analyzed. In addition, vocal performance is used to improve children’s aesthetic ability and comprehensive quality. An efficient information hiding scheme based on the random matrix is proposed. Furthermore, by establishing a random matrix to optimize the information scheme hidden in the multinational children’s songs, the textual expression efficiency of children’s songs is improved. The results reveal that when the model is iterated for 200 times, the traditional algorithm has only a recognition accuracy of 0.73 for children’s songs, and the recognition accuracy of the convolution neural network (CNN) model can reach 0.76, while the proposed improved deep neural networks (DNN) algorithm model has the best recognition accuracy, which can reach 0.80. It explores how to better meet children’s aesthetic and cultural needs from the perspective of information security by optimizing the semantic indexing algorithm of random matrix mapping elements. Through the performance test of the algorithm, on account of reducing the time complexity of the algorithm, the classification accuracy of children’s song text can be effectively improved. This exploration has practical reference value for cultivating children’s comprehensive quality and aesthetic ability.