Knowledge discovery and dissemination of text by mining with words
AssigneeThe United States of America as represented by the Secretary of the Navy
InventorStuart Rubin
Life of this patent
Estimating the term…
Abstract
The present invention enables the construction and deployment of semantic search engines, which learn. Here, the learning necessarily occurs in unsupervised mode given the huge amount of text being searched and the resultant impracticality of providing human feedback. That is, it acquires a lexicon of synonyms and the context for their proper application by searching a moving window of text and compiling the results. No human operator is involved. This allows for the practical search, using machine learning techniques, of very large textual bases.
Claims16
A computer-implemented iterative semantic search method for searching words on a document, the method comprising the steps of:
- scanning a document containing a sequence of document words broken into a sequence of moving document windows formed by contiguous sentences on the document;
- forming a query containing a sequence of query words broken into query windows representing queries about the contents of the scanned document;
- a normalizer step responsive to the query for converting the various forms of each word to its normal form, where its meaning is mostly preserved by the surrounding context;
- a lexicon step for providing synonyms defined by repetition and locality of reference for any normalized word;
- a repository step for storing and indexing each of the synonyms for a defined window size;
- a parser step for building undirected graphs represented by edges and vertexes using a specified number of sentences from the document;
- a heuristics step for taking as an input the current success in growing graphs and producing as an output the size of the moving window for use by the parser;
- a randomizer step for reducing required storage and search time by removing duplicate sub-graphs and embedded sub-graphs from the storage repository;
- a cache step for applying temporal and spatial locality for subsequent queries;
- and
- a graph search algorithm step, which rank orders matching graphs for the query sub-graphs, if any, on the basis of matching sub-graphs.
A computer-implemented iterative semantic search method for searching words on a document where the document contains a sequence of document words broken into a sequence of moving document windows formed by contiguous sentences on the document, the method comprising the steps of:
- a normalizer step responsive to a query about the contents of the document for converting the various forms of each word to its normal form, where its meaning is mostly preserved by the surrounding context;
- a lexicon step for providing synonyms defined by repetition and locality of reference for any normalized word;
- a repository step for storing and indexing each of the synonyms for a defined window size;
- a parser step for building undirected graphs represented by edges and vertexes using a specified number of sentences from the document;
- a heuristics step for taking as an input the current success in growing graphs and producing as an output the size of the moving window for use by the parser;
- a randomizer step for reducing required storage and search time by removing duplicate sub-graphs and embedded sub-graphs from the storage repository;
- a cache step for applying temporal and spatial locality for subsequent queries;
- and
- a graph search algorithm step, which rank orders matching graphs for the query sub-graphs, if any, on the basis of matching sub-graphs.
Dependencies are inferred from the claim text.