On Private Scalar Product Computation for Privacy-Preserving Data Mining

Abstract

Abstract. In mining and integrating data from multiple sources, there are many privacy and security issues. In several different contexts, the security of the full privacy-preserving data mining protocol depends on the security of the underlying private scalar product protocol. We show that two of the private scalar product protocols, one of which was proposed in a leading data mining conference, are insecure. We then describe a provably private scalar product protocol that is based on homomorphic encryption and improve its efficiency so that it can also be used on massive datasets.

Conditional Encrypted Mapping and Comparing Encrypted Numbers

“…The immediate uses lie in appointment scheduling, flexible timestamp verification, biometrics, etc. Certain kinds of set intersection problems, as studied in [7,9], can be represented succinctly as GT instances, resulting in more efficient solutions using our constructions.…”

Privacy-Preserving Record Linkage

“…Furthermore this definition has no mention of information leakage due to the protocol itself, however it could be amended so that the definition must hold for the intermediate messages as well as the final output. An analysis of some weakly secure inner product protocols is given by [15], who conclude that the weaker model presents a far greater prospect of information leakage than does the cryptographic model.…”

“…It appears that despite the elegance of this approach, the third party would still be able to mount a frequency based attack on these embeddings. Nevertheless the metric embedding idea is compelling since it results in lowdimensional vectors [33], and so in principle it allows reduction of string edit distance computation to secure inner products which are already well-studied in the literature (e.g., [15]). …”

“…One way, if the similarity scores are computed using a cryptographic protocol, would be to threshold it before it is allowed to be decrypted. For example reducing similarity to the inner product and using [15] results in an encrypted value held by one party, where it may only be decrypted by the other. In this case the holding party may apply a certain sequence of operations to the ciphertext in order to reduce it to a binary flag corresponding to thresholding against some constant value.…”

Secure computations on non-integer values

“…For example, several solutions have been proposed that use SMC to enhance privacy in auctions [12], data clustering [13], [14] or filtering [15]. Recently, there is also an increased interest in combining SMC with methods of signal processing in order to be able to privately analyze signals; examples are the analysis of medical signals [16] or the evaluation of biometrics on encrypted data [2].…”