Privacy Preserving Semantic-Aware Multi-keyword Ranked Dynamic Search Over Encrypted Data

Abstract

Searchable encryption (SE) is a promising primitive which enables data owners to privately query and retrieve required documents from encrypted outsourced data collection in cloud servers. Most of the SE schemes in literature strive to achieve a balanced trade-off between efficiency, privacy and expressiveness of query. However, a vast majority of efficient and secure schemes lack searching capabilities beyond exact match of queries and ignore the semantic relationship between keywords and documents. Therefore, a new trend in SE research is to address the question: How to make search semantic-aware and context-based? Majority of the existing semantic search schemes are static, do not support dynamic update operations, suffer from leakage of significant information to server, and employ huge size vectors to represent documents, keywords and topics/features. We propose a privacy preserving, relevance ranking-based dynamic search scheme using a sublinear search time hash-based indexing mechanism and semantic-aware search supporting semantic expansion of user queries. Our scheme employs Word2Vec model to quantitatively evaluate semantic relationships between query and other keywords to expand queries to fulfill user’s search intentions. It also supports dynamic operations like insert and delete of documents on outsourced collection with less efforts to update relevance information which is calculated partially at client and server. Experiments on RFC dataset have demonstrated a time and space efficient non-interactive semantic search over encrypted data.