Improving file locality in multi-keyword top-k search based on clustering

Nowadays, fast growing number of users and business are motivated to outsource their private data to public cloud servers. Taking into consideration security issues, private data should be encrypted before being outsourced to remote servers, though this makes traditional plaintext keyword search rather difficult. For this reason, there exists an urgent need of an efficient and secure searchable encryption technology. In this paper, an affinity propagation (AP) K-means clustering method (CAK-means, a combination of AP and K-means clustering) is proposed to realize fast searchable encryption in Big Data environments. CAK-means clustering utilizes affinity propagation to initialize K-means clustering, thereby making the clustering process faster, stable and effectively improving the initial clustering center quality of the K-means. As the AP algorithm identifies the clustering center with much lower errors than other methods, it significantly improves the search accuracy. Simultaneously, the related files in one cluster are stored at the contiguous locality of disks which will substantially improve the file locality and speedup the read and write disk I/O. Additionally, the coordinated matching measure is utilized to support accurate ranking of search results. Experimental results show that the proposed CAK-means-based multi-keyword ranked searchable encryption scheme (MRSE-CAK) has higher search efficiency and accuracy while simultaneously ensuring equivalent security.