Research On Encrypted Cloud Data Retrieval Scheme Based On Text Semantics

In today's era of rapid development of cloud computing,massive data on the Internet explodes instantly,and the local storage of individuals and enterprises is facing unprecedented pressure.the local storage of individuals and enterprises is facing unprecedented pressure.As cloud computing and cloud storage services bring superior user experience in speed and economical cost,more people choose to upload their data to the cloud server for storage and management.However,the user data files stored in the cloud are not completely secure.The cloud will access the user data out of curiosity,and even attackers will attack the server to steal the data,which may eventually lead to the leakage of users' private information,making it impossible for data in the cloud to be completely secured.Due to these potential risks,data encryption is universally employed in cloud storage.Under these circumstances,ensuring efficient search of data in the encrypted environment has become one of the critical challenges we are facing at present.Therefore,the application of searchable encryption technique provides a promising solution for the confidentiality and efficient search of data in cloud storage.Most of the existing schemes based on keyword search also have some defects.On one hand,the semantic meaning of most keywords is weakly related to the overall content of the text,thus lacking semantic relevance.On the other hand,some schemes based on semantic search are quite inadequate in relevance and coherence of context semantics.Besides,there are also some troubles to be solved such as security defects caused by the leakage of search results.Aiming at the problems existing in the schemes available,this paper proposes a search scheme over encrypted data based on BERT model and a privacy security retrieval scheme based on differential privacy respectively.The specific contributions are as follows:1)A semantic retrieval scheme over encrypted cloud data based on BERT model is proposed.In this scheme,the extracted keywords are trained by the BERT model to generate semantic vectors rich in text semantic information,so that the vector can contain the context semantic information of the text,build an index,and the vectors are encrypted and stored in the cloud.As long as the users query trapdoor requests are received,the relevant files with similar semantic meaning are precisely matched and gave back,so that the retrieval results can better meet the actual needs of the user.Finally,the comparative experiment of the scheme also verifies the high efficiency and high precision of the scheme.2)A privacy security protection retrieval scheme based on differential privacy is proposed.In this scheme,the access mode obfuscation mechanism based on d-privacy is adopted to obfuscate index vectors with erasure codes,and then upload and encrypt to the cloud server to prevent the cloud server from obtaining the query result document when the data user retrieves it.When the user searches,the cloud returns the relevant encrypted document shards through the trapdoor matching,and the data user decrypts the encrypted document with the authorized key and decodes the decrypted document shards to obtain the correct retrieval results,so as to ensure the privacy security of the retrieval results.The experiment results show that the accuracy and efficiency of the scheme are guaranteed as well as the security of privacy.