Method for accelerated long document search using Hilbert curve mapping

A method for searching a textual document database with a query document, comprising:

• loading an index table into a memory, wherein the index table Ψ={ζ,η,ζ} is a triple attribute table where ζ is an embedding vector number in respect to all embedding vectors of document D, η is a document number that contains embedding vector ζ, and ζ is a Hilbert number that corresponds to a vector of number ζ;
• inputting, into the memory, a query document, which has a plurality of embedding vectors corresponding to sentences in the query document, wherein the query embedding vectors are mapped into Hilbert numbers using Hilbert curve transformation;
• searching, via a processing circuitry, the index table using the Hilbert numbers and retrieving candidate documents that are similar to the query document based on the Hilbert numbers;
• inputting, into the memory, a plurality of documents,
• wherein each document of the plurality of documents has a plurality of sentences, and each sentence of each document has a plurality of semantic embedding vectors;
• mapping, via the processing circuitry, the semantic embedding vectors from m-dimensions to 1-dimension vectors of Hilbert numbers using the Hilbert curve transformation;
• constructing, via the processing circuitry, an index table with the 1-dimensional vectors;
• storing the index table in the memory;
• assigning, via the processing circuitry, the Hilbert numbers to sentences in the query document;
• determining, via the processing circuitry, semantic similarity between the Hilbert numbers for the sentences in the query document and the sentences in each document;
• and
• outputting, via the processing circuitry, a predetermined number of candidate documents based on highest to lowest semantic similarity.