Vector Based Privacy-Preserving Document Similarity with LSA

Document similarity is the foundation of many intelligent data processing systems, such as information retrieval, text classification and clustering. However, traditional document similarity algorithms are challenged by the privacy-preserving problem. Recently, privacy-preserving document similarity approaches are provided to solve this problem and there are two kinds of approaches which are vector based protocols and set based protocols. Existing vector based protocols mainly use vector space model for document similarity computation. But vector space model has deficiencies to compute document similarity effectively and efficiently. In this paper, we focus on vector based document similarity and present a novel protocol with latent semantic analysis. Experimental evaluation shows that our protocol has better accuracy and performance than existing protocols.