Homomorphic Encryption-driven AI through Text Mining in The Cloud

Abstract

In the current era, there is increasing interest in data security, especially in cloud computing. Homomorphic Encryption (HE) supported by Artificial Intelligence (AI) technology offers a promising solution in this field. Homomorphic encryption ensures that computations are performed on encrypted data without decryption, thus ensuring privacy. However, the integration of AI algorithms and text mining in the cloud environment is still a challenging topic. The study aims to develop a framework for partial homomorphic encryption combined with a deep learning algorithm for text mining in the cloud environment. The aim of the proposed approach is to evaluate the trade-offs between security and computational performance through deep learning to ensure the highest accuracy.The proposed method uses frequency coding and combines it with the developed deep learning algorithm, which is based on the dynamic change of the weights accompanying the neural network. The text mining model is integrated by multiplying the encrypted frequency by the factor derived from the weight in the neural network iterations. The model was trained on data in two standard datasets and the model was tested afterwards. The computational overheads were evaluated as the text size before and after encryption, the use of computing resources, and the amount of noise generated. Using HE allowed for successful text mining on encrypted data, with minimal impact on accuracy. The ciphertext size was 3.3x larger than plaintext, with increased computational overhead. The computational resource utilization was balanced in an acceptable manner for cloud storage, with noise growth not exceeding 31% while accuracy remained at 98%. In this study, the feasibility of using homomorphic encryption on texts supported by deep learning technology in a cloud environment was concluded. This provides a solution for computational operations on data while preserving privacy. The framework provides a balance between security and computational efficiency and is important for applications that require high levels of security, despite some challenges that may be solved in the future by machine learning and working on larger texts.