Research On Semantic Feature Based Search Scheme Over Encrypted Data In Cloud

With the increasing development of cloud computing, more and more customers (enterprises and individuals) intend to outsource their large dataset into the cloud to avoid the overhead of management and local-storage. The cloud storage provides customers with convenient and flexible usage pattern of data, with which the customers can access their data anywhere. However, the widespread adoption of the cloud storage is hindered by many reasons of which security issue is the increasingly important one. After the dataset is outsourced to the cloud, unauthorized operations on the outsourced data may exist since the Cloud Service Provider (CSP) possesses full control of the outsourced data. In this case, the security of the outsourced data cannot be guaranteed, which results in leakage of sensitive data and privacy of user.To combat unauthorized accesses and protect the privacy of user and sensitive data, the dataset should be encrypted by the data owner before outsourcing. However, encrypted data make the traditional and efficient plaintext keyword search technique useless. The simple and awkward method of downloading the whole dataset and decrypting locally is obviously impractical for the authorized data users. What’s more, considering the potentially huge number of outsourced data and great deal of data users, it is also difficult to meet both the requirements of system usability and performance. So, it is an especially important task to explore secure and effective search scheme over encrypted cloud data.Existing search schemes based on encrypted data mostly use keyword as the feature of documents and construct the searchable index based on the keyword semantic. These search schemes can provide different query capabilities, including single keyword search, multi-keyword search, ranked search, fuzzy keyword search, parallel search, similarity search, and so on. But among these schemes no one can supports both multi-keyword ranked search and parallel search. In addition, keyword-based search schemes completely depend on inputted keywords and do not take the semantic relations among words into consideration which may lead to inaccurate and incomplete search results. And existing semantic search schemes based on encrypted data are very few. So, based on the problems mentioned above, we propose two search schemes over encrypted data based on keyword semantic feature and concept semantic feature respectively. The main works of this paper are described as follows:(1) We propose a ranked parallel search scheme based on keyword semantic feature over encrypted data. In this scheme, we use keyword semantic as the feature of documents, to address multi-keyword search and result ranking, we use Vector Space Model (VSM) to build document index and search index. Then, we can evaluate the relevance between document and search query by calculating the similarity score between document index and search index. To improve search efficiency and perform parallel search, we use a tree-based index structure which is a balance binary tree. In the query phase, the procedure searches the index tree based on certain rules to get interested documents, which avoids computing the relevance between search query and each document in the outsourced dataset. We design two secure search schemes which meet different privacy requirements in two threat models. We make security analysis for our two secure search schemes which proves privacy guarantees. And experiments on the real-world dataset show that proposed schemes are indeed efficient.(2) We propose a search scheme based on concept semantic feature over encrypted data. In this scheme, we use concept semantic as the feature of documents, and we construct an extended concept hierarchy based on several semantic relations which are extracted from the outsourced dataset and domain concepts related knowledge. Based on the extended concept hierarchy, we use Vector Space Model (VSM) to build document index and search index. And in the query phase, we also extend the search words using the concept hierarchy to improve the search accuracy. The "host-attribute" relation is one of the semantic relations in our extended concept hierarchy, where each attribute is corresponding to a concept, called as attribute concept, and each attribute can possess different values. By using attribute concept, the range searching based on outsourced dataset is realized. A k-d tree-based searchable index is constructed to improve the search efficiency. Based on the existing threat model, we design a secure semantic search scheme. We make security analysis for the scheme which proves privacy guarantees. And experiments on the real-world dataset show that proposed scheme is indeed efficient.