Storage allocation based on secure data comparisons via multiple intermediaries

A method for data allocation based on secure data comparisons, the method comprising:

• receiving, from a data allocator and at each intermediary of a plurality of intermediaries, a request to identify a target storage container of a plurality of storage containers, the request including a sparsity number that is indicative of a truncated sparse binary vector representation of a truncated data term, wherein the truncated data term is a portion of a data term to be securely allocated to the target storage container, the truncated data term anonymizing the composition of the data term to the each intermediary and including information relevant to the each intermediary, and further wherein anonymizing is performed by removing portions of the sparse binary vector to generate the truncated sparse binary vector as a vector of smaller length that does not include all of the composition of the data term;
• comparing, at an intermediary and for a storage container, the truncated data term, represented by the sparse binary vector, with a collection of truncated candidate terms to select a representative term of the candidate terms, wherein the truncated candidate terms are truncated versions of candidate terms representative of each cluster of similar terms in the storage container;
• identifying, from the intermediary to the storage container, the selected representative term;
• receiving, at the intermediary from the storage container, a similarity profile representative of similarities between the truncated data term and terms in the storage container, wherein the similarity profile is determined via a similarity score between the truncated data term and the selected representative term;
• selecting, at the intermediary, a candidate target storage container based on similarity profiles received from each storage container;
• and
• providing from the intermediary and without knowledge of the data term or candidate target storage containers selected by other intermediaries, the selected candidate target storage container to the data allocator.

A non-transitory computer readable medium comprising executable instructions to:

• receive, from a data allocator and at each intermediary of a plurality of intermediaries, a request for identification of a target storage container of a plurality of storage containers to be associated with a data term, the request including a hash universe size and a sparsity number;
• generate a random permutation based on the hash universe size;
• receive, from the data allocator, a truncated version of the data term, the truncated version of the data term is based on a version of the sparsity number that is indicative of a sparse binary vector representation of the truncated version of the data term and the permutation, and wherein the truncated version of the data term is a portion of the data term, the truncated version of the data term anonymizing the composition of the data term to the each intermediary and including information relevant to the each intermediary, and further wherein anonymizing is performed by removing portions of the sparse binary vector to generate the truncated sparse binary vector as a vector of smaller length that does not include all of the composition of the data term;
• receive, from a storage container, truncated versions of a collection of candidate terms, wherein each candidate term represents a cluster of similar terms in the storage container, and wherein the truncated data terms of the candidate terms are based on the sparsity number and the permutation;
• compare, for the storage container, the truncated data term with the truncated candidate terms to select a representative term of the candidate terms;
• identify, the selected representative term to the storage container;
• and
• receive, from the storage container, a similarity profile representative of similarities between the truncated data term and terms in the storage container, wherein the similarity profile is determined via a similarity score between the truncated data term and the selected representative term.