In order to improve the data embedding capacity and the communication efficiency of coverless information hiding algorithm, addressing the advantages of semi-structured coverless information hiding algorithm，this paper introduces a semi-structured coverless information hiding algorithm based on the behavioral habits of social platforms. The specific idea of the algorithm is to build a one-to-one mapping relationship between icons and secret messages in a small icon library. According to certain principles，some small icons are montaged a picture, the secret information can be expressed by the splicing picture, and the transmission of secret messages is realized by delivering the spliced pictures. In order to improve the recognition rate of small icons and the anti-interference ability of the whole hidden communication system, convolutional neural network is also introduced to train and classify the icons in the icon library, and the interference samples are introduced as training samples set. The experimental results show that the algorithm has good anti-attack ability and the hiding capacity can be improved, and therefore, the algorithm can be used in covert communication.