Design And Implementation Of Encrypted File Retrieval System Based On Neural Network In Cloud Computing Environment

In cloud computing,many current schemes for retrieving encrypted information through outsourced data use the TF-IDF model based on the data statistical characteristics to generate vectors,which cannot mine the deeper semantics of the data and cannot effectively obtain the context from a word information.A query has to be restricted to a specified keyword set and these schemes cannot support the retrieval of arbitrary words.In addition,with the expansion of the most frequent keywords,the retrieval efficiency is greatly reduced.Therefore,it is of great significance to establish a more accurate,safe and efficient encryption retrieval scheme.This thesis focuses on the problems of searchable encryption accuracy and low retrieval efficiency and the main work is as follows:(1)Aiming at the problems of low accuracy and slow retrieval speed of ciphertext retrieval,a Convolutional Neural Network-based Encrypted Document Sorting and Retrieval System(CNRSE)in cloud environment is proposed.A neural network vector generation model is proposed.The document vector and query vector are generated by a convolutional neural network vector generation model(CVGM)that conforms to the ciphertext retrieval characteristics,and encrypted by the secure KNN algorithm.The CVGM model is trained using a private cloud server combined with federated learning to protect the user’s sensitive data from being leaked during model training.As the model parameters in the private cloud are continuously updated,the CVGM is also updated periodically until the model converges.By improving the index tree structure based on clustering tree,and using the inner product calculation of document vector,the documents with similar semantics are gathered together,and a bottom-up index tree is established to improve the retrieval efficiency.(2)The ciphertext retrieval system is designed according to CNRSE scheme.According to the program development process,first of all to make a requirement analysis and outline design for the system implementation,including the design goal of the system,and determine its dependencies according to the functional characteristics of the core modules.Then a detailed design of a neural network encryption file sorting and retrieval system is made.(3)The CNRSE system is realized correctly,and the system is tested and analyzed from the two directions of system function test and performance test.Analysis and experiments based on real datasets show that the proposed scheme can perform expected functions and run stably in the retrieval process.The retrieval accuracy remains around 95%,and the retrieval time is reduced by at least 30%compared to other schemes.Tests show that the CVGM model can mine the deep semantics of articles.Combined with federated learning,it can ensure the security of ciphertext retrieval and improve the retrieval accuracy.The index tree structure based on clustering tree can also improve retrieval efficiency.