In this paper, we propose a user-friendly encrypted storage scheme named EStore, which is based on the Hadoop distributed file system. Users can make use of cloud-based distributed file systems to collaborate with each other. However, most data are processed and stored in plaintext, which is out of the owner’s control after it has been uploaded and shared. Meanwhile, simple encryption guarantees the confidentiality of uploaded data but reduces availability. Furthermore, it is difficult to deal with complex key management as there is the problem whereby a single key encrypts different files, thus increasing the risk of leakage. In order to solve the issues above, we put forward an encrypted storage model and a threat model, designed with corresponding system architecture to cope with these requirements. Further, we designed and implemented six sets of protocols to meet users’ requirements for security and use. EStore manages users and their keys through registration and authentication, and we developed a searchable encryption module and encryption/decryption module to support ciphertext retrieval and secure data outsourcing, which will only minimally increase the calculation overhead of the client and storage redundancy. Users are invulnerable compared to the original file system. Finally, we conducted a security analysis of the protocols to demonstrate that EStore is feasible and secure.
In recent years, more and more data have been generated via news media on Internet. Especially, when tremendous notable events happen at home and abroad, which cases a large demand for event retrieval. Text retrieval is the fundamental and important issue to search information under the scenario of Big Data Some recent works have proposed neural information retrieval, which has a significant improvement in terms of some problems (e.g., semantic gap). However, such retrieval takes a general-purposed query rather than event-oriented query as the retrieval object, which fails to search events in the event-oriented text retrieval scenaiio. As such, in this paper, we first analyze the difficulties and investigate a large of recent deep neural methods that may be used for event-oriented text retrieval. After that, we propose a pipeline based on deep neural networks. Finally, we present several tasks at different pipeline stages for evaluating event oriented text retrieval.
The recommendation idea is inspired by the user's search habits and experience, there is a demand that the server should provide active recommendation. However, data encryption makes keywords lose their semantics, which hinders the analysis and recommendation function of the server. In this paper, we study the "search mode" and "access mode" of encrypted data and present a novel ciphertext search model supporting association and recommendation. Specifically, we construct the basic searchable encryption with access control, association analysis and recommendation method in detail. Finally, we conduct an extensive security and performance analysis to ensure ERecommend is secure and feasible.