A Model of Precise Delivery of Advertising based on Encrypted Transmission in Network

Telecommunications(Telelcom) operators and Internet companies can effectively improve the advertising delivery effectiveness and user experience by using their own user basic figure data, as gathering and calculating with the deeply transformed link data (such as paid information) owned by the advertising data parties. However, there are challenges for the large data volume, high feature dimensions, and the inclusion of user privacy information in the data to meet such needs. Trusted execution environment (TEE) is used to prevent information leakage when data operates in memory, and machine learning algorithm is used to find and extract hidden information from a large number of high-dimensional data. In this work, encryption technology and machine learning are combined to establish a well-trained machine learning model by using data encryption transmission, calculation, and storage. The average accuracy of the model for the validation set was $\mathbf{9 9 . 9 \%}$ and the AUC value was 0.99. The accuracy of the model for the test set was $\mathbf{7 4 . 0 \%}$ and AUC value was 0.62. It showed that the generalization of model was general, and the prediction results of the model can be used as a reference. It showed that the model training with high-security requirements and large quantities of data can run in TEE. However, the current model needs to be further improved the generalization. Simultaneously, the loss performance ratio was $6.98 \%$ in the SGX virtual machine inference time compared to the general virtual machine time. In conclusion, this work has taken advantage of the technology to better solve the privacy protection facing in the precise delivery of advertising, as providing the solid evidence for future work.