Associative memory
View Patent ↗A system of retrieving documents comprising: coding a plurality of stored documents as a respective document feature vector; generating a query feature vector based on a query document; and performing one or more logical operations between the query feature vector and the document feature vector to obtain respective similarity measures.
1. A system of retrieving documents comprising the steps of:
coding a plurality of stored documents as respective document feature vector;
generating a query feature vector based on a query document; and
performing one or more logical operations between the query feature vector and said document feature vector to obtain respective similarity measures.
2. The system of claim 1 , further comprising a step of formulating a similarity score for respective ones of said stored documents, each similarity score representing how often a certain feature or a set of features defined by the logical 1's in the query vector occurs within a corresponding feature vector.
3. The system of claim 1 , wherein, based on said similarity measures, a set of candidate documents is chosen for further inspection, said set of candidate documents being further evaluated with respect to their similarity to the query document, said evaluation resulting in one or more further similarity measures reflecting the similarity between query and candidate under one or more aspects.
4. The system of claim 1 , further comprising a step selected from the set comprising:
returning results which lie above a certain similarity threshold, and/or retrieving a user definable number of documents which have the highest similarity.
5. The system of claim 1 , further comprising a step of a user defining a desired degree of similarity to be:
identity;
a settable degree of similarity or a similarity threshold; and/or
identity for certain features and a satiable degree of similarity for the other features.
6. The system of claim 1 , wherein features coded by a feature vector for their presence or absence comprise one or more of the following:
unigrams, digrams, multigrams, words, fragments of text, sentences, paragraphs, or a combination of them;
semantic classes reflecting a certain meaning which elements in the text to be represented may have;
concept classes reflecting an abstract concept to which elements in the text to be represented belong;
one or more classification classes classifying the document according to a classification scheme; and/or
one or more attributes which may be assigned to a document.
7. The system of claim 1 , wherein a matrix formed by the feature vector representing the documents to be searched is stored column-wise, and wherein said method further comprises:
for those columns of the matrix where the query feature vector bit is set, performing a processor supported bitwise operation between the columns of the matrix to obtain one or more result columns, the values in the result columns representing the similarity between the query and the stored documents.
8. The system of claim 1 , wherein there is a plurality of result columns such that the result columns form a matrix, and wherein a row of the thus formed result column matrix represents a degree of similarity between the query and the stored document corresponding to said row.
9. The system of claim 1 further comprising the steps of:
obtaining a set of examples for a certain classification class by carrying out a search for example documents belonging to a particular class; and
training a classification engine by said obtained set of examples as a learning input.
10. The system of claim 1 , wherein there is a plurality of result columns such that the result columns form a matrix, and wherein a row of the thus formed result column matrix represents a degree of similarity between the query and the stored document corresponding to said row.
11. A computer-implemented system of realizing an associative memory capable of storing a set of documents and retrieving one or more of said stored documents similar to an inputted query document, said method comprising:
coding each of said stored document or a part of it through a corresponding feature vector;
arranging said feature vector in a matrix; and
generating a query feature vector based on the query document.
12. The system of claim 11 , wherein there is a plurality of result columns which code a similarity score which is based on the following:
the number how often a certain feature or a set of features defined by the logical 1's in the query vector occurs within a row of the matrix.
13. The system of claim 11 , wherein based on said similarity measure a set of candidate documents is chosen for further inspection, said set of candidate documents being further evaluated with respect to their similarity to the query documents being further evaluated with respect to their similarity measures reflecting the similarity between query and candidate under one or more aspects.
14. The system of claim 11 , wherein results returned which lie above a certain similarity threshold, and/or
wherein a user definable number of documents are retrieved which have the highest similarity.
15. The system of claim 11 , wherein a desired degree similarity can be defined by the user to be:
identity;
a settable degree of similarity or a similarity threshold; and or
identity for certain features and a settable degree of similarity for the other features.
16. The system of claim 11 wherein features coded by a feature vector for their presence or absence comprise one or more of the following;
unigrams, digrams, multigrams, words, fragments of text, sentences, paragraphs, or a combination of them;
semantic classes reflecting a certain meaning which elements in the text to be represented may have;
concept classes reflecting an abstract concept to which elements in the text to be represented belong;
one or more classification classes classifying the document according to a classification scheme;
one or more attributes which may be assigned to a document.
17. The system of claim 11 , wherein the matrix formed by the feature vector representing the documents to be searched is stored column-wise, and wherein said method further comprises:
for those columns of the matrix where the query feature vector bit is set, performing a processor supported bitwise operation between the columns of the matrix to obtain one or more result columns, the values in the result columns representing the similarity between the query and the stored documents.
18. The system of claim 11 further comprising the steps of:
obtaining a set of examples for a certain classification class by carrying out a search for documents belonging to a particular class; and
training a classification engine by said obtained set of examples as the learning input.