Information retrieval system and method that generates weighted comparison results to analyze the degree of dissimilarity between a reference corpus and a candidate document

Claim 3

A method as in claim 1 wherein: a first of said analysis algorithms is arranged to generate and compare word frequency lists and to produce a comparison result; and said predetermined algorithm includes forming from the reference corpus separate first and second parts, performing a comparison between said first and second parts in accordance with said first analysis algorithm, the resulting comparison result constituting a measure of the homogeneity of said first and second parts.

Claim 4

A method as in claim 2 wherein: a second of said analysis algorithms is arranged to generate and compare word-level n-gram frequency lists and to produce a comparison result; and said predetermined algorithm includes calculating a confidence value by multiplying the said measure of the homogeneity by the total number of words of the word frequency lists generated for obtaining said measure of homogeneity, and, if this confidence value is less than a predetermined threshold, setting to substantially zero the value of the weight corresponding to said second analysis algorithm.

Claim 5

A method as in claim 1, including the steps of: presenting to the user, for each of a plurality of candidate documents stored in the retained text store, the respective dissimilarity measures and respective links to said plurality of stored retrieved texts, receiving from the user an allocated relevance value R in respect of a presented dissimilarity measure, and modifying said weights by multiplying each weight (Wi) by a respective weight factor (1+ki), where ki is a function of the contribution that the respective weighted comparison result makes to the dissimilarity measure, and the value of R-Rmean, where Rmean is the mean of lowest and highest possible relevance values.

Claim 6

A method as in claim 2, including the steps of: presenting to the user, for each of a plurality of candidate documents stored in the retained text store, the respective dissimilarity measures and respective links to said plurality of stored retrieved texts, receiving from the user an allocated relevance value R in respect of a presented dissimilarity measure, and modifying said weights by multiplying each weight (Wi) by a respective weight factor (1+ki), where ki is a function of the contribution that the respective weighted comparison result makes to the dissimilarity measure, and the value of R-Rmean, where Rmean is the mean of lowest and highest possible relevance values.

Claim 7

A method as in claim 3, including the steps of: presenting to the user, for each of a plurality of candidate documents stored in the retained text store, the respective dissimilarity measures and respective links to said plurality of stored retrieved tests, receiving from the user an allocated relevance value R in respect of a presented dissimilarity measure, and modifying said weights by multiplying each weight (Wi) by a respective weight factor (1+ki), where ki is a function of the contribution that the respective weighted comparison result makes to the dissimilarity measure, and the value of R-Rmean, where Rmean is the mean of lowest and highest possible relevance values.

Claim 8

A method of generating a language model, said method comprising: (i) providing a reference corpus; (ii) forming from said reference corpus a training portion and a development portion; (iii) generating initial values of respective weights corresponding to a plurality of analysis algorithms by processing the training portion in accordance with a predetermined algorithm; (iv) performing a comparison between the training portion and the development portion in accordance with a first of said analysis algorithms and producing an initial comparison result, said first analysis algorithm being arranged to generate and compare word-level n-gram frequency lists having n-grams from unigrams up to m-grams, where m is a predetermined integer, the word-level n-gram frequency list generated from the training portion constituting a language model; (v) performing information retrieval in accordance with steps (d) to (h) of the method of claim 1, using the above mentioned plurality of analysis algorithms, and using the training portion as the reference corpus; (vi) retrieving a document from the retained text store; (vii) modifying the training portion by combining it with the document retrieved from the retained text store; (viii) repeating step (iv) in respect of the modified training portion to produce a further comparison result; and (ix) modifying the weights in accordance with a weight modifying function of said initial and further comparison results.

Claim 9

A method as in claim 7 wherein step (v) is stopped when a first candidate document is stored in the retained text store.

Claim 10

A method as in claim 7 wherein step (vi) comprises selecting the document for retrieval on the basis of the least degree of dissimilarity.

Claim 11

A method as in claim 9 wherein steps (vi) to (ix) are iteratively performed in respect of respective documents successively selected by increasing degree of dissimilarity.

Claim 12

A method as in claim 10 wherein the modified training portion produced by step (vii) of one iteration of steps (vi) to (ix) constitutes the training portion of step (vii) of the next iteration thereof.

Claim 13

A method as in claim 7, wherein the weight modifying function comprises multiplying each weight (Wi) by a respective weight factor (1+ki), where ki is a function of the contribution that the respective weighted comparison result makes to the dissimilarity measure, and the difference between the initial comparison result and the further comparison result.

Claim 14

A method as in claim 8, wherein the weight modifying function comprises multiplying each weight (Wi) by a respective weight factor (1+ki), where ki is a function of the contribution that the respective weighted comparison result makes to the dissimilarity measure, and the difference between the initial comparison result and the further comparison result.

Claim 15

A method as in claim 9, wherein the weight modifying function comprises multiplying each weight (Wi) by a respective weight factor (1+ki), where ki is a function of the contribution that the respective weighted comparison result makes to the dissimilarity measure, and the difference between the initial comparison result and the further comparison result.

Claim 16

A method as in claim 10, wherein the weight modifying function comprises multiplying each weight (Wi) by a respective weight factor (1+ki), where ki is a function of the contribution that the respective weighted comparison result makes to the dissimilarity measure, and the difference between the initial comparison result and the further comparison result.

Claim 17

A method as in claim 11, wherein the weight modifying function comprises multiplying each weight (Wi) by a respective weight factor (1+ki), where ki is a function of the contribution that the respective weighted comparison result makes to the dissimilarity measure, and the difference between the initial comparison result and the further comparison result.

Claim 18

An information agent for use in a communications network including a plurality of databases, the agent comprising; means for generating initial values of respective weights corresponding to a plurality of analysis algorithms by processing an identified reference corpus in accordance with a predetermined algorithm; means for retrieving from storage a text document as a candidate document; means for performing respective comparisons between the candidate document and the reference corpus in accordance with each of said analysis algorithms and producing respective comparison results; means for generating corresponding weighted comparison results by multiplying each said comparison result by its respective weight; means for summing the weighted comparison results to produce a dissimilarity measure that is indicative of the degree of dissimilarity between the retrieved reference corpus and the retrieved candidate document, and means for storing the candidate document in a retained text store if said sum is indicative of a degree of dissimilarity less than a predetermined degree of dissimilarity.

Claim 19

An information agent for use in a communications network including a plurality of databases, the agent comprising; means for generating initial values of respective weights corresponding to a plurality of analysis algorithms by processing an identified reference corpus in accordance with a predetermined algorithm; means for retrieving from storage a text document as a candidate document; means for performing respective comparisons between the candidate document and the reference corpus in accordance with each of said analysis algorithms and producing respective comparison results; means for generating corresponding weighted comparison results by multiplying each said comparison result by its respective weight; means for summing the weighted comparison results to produce a dissimilarity measure that is indicative of the degree of dissimilarity between the retrieved reference corpus and the retrieved candidate document, and means for storing the candidate document in a retained text store if said sum is indicative of a degree of dissimilarity less than a predetermined degree of dissimilarity.

Claim 20

A document access system, for accessing documents stored in a distributed manner and accessible by means of a communications network, the access system comprising at least one software agent for use in accessing documents by means of the network, wherein the agent comprises: means for generating initial values of the respective weights corresponding to a plurality of analysis algorithms by processing an identified reference corpus in accordance with a predetermined algorithm; means for retrieving from storage a text document as a candidate document; means for performing respective comparisons between the candidate document and the reference corpus in accordance with each of said analysis algorithms and producing respective comparison results; means for generating corresponding weighted comparison results by multiplying each said comparison result by its respective weight; means for summing the weighted comparison results to produce a dissimilarity measure that is indicative of the degree of dissimilarity between the retrieved reference corpus and the retrieved candidate document; and means for storing the candidate document in a retained text store if said sum is indicative of a degree of dissimilarity less than a predetermined degree of dissimilarity.

Claim 21

A document access system, for accessing documents stored in a distributed manner and accessible by means of a communications network, the access system comprising at least one software agent for use in accessing documents by means of the network, wherein the agent comprises: means for generating initial values of the respective weights corresponding to a plurality of analysis algorithms by processing an identified reference corpus in accordance with a predetermined algorithm; means for retrieving from storage a text document as a candidate document; means for performing respective comparisons between the candidate document and the reference corpus in accordance with each of said analysis algorithms and producing respective comparison results; means for generating corresponding weighted comparison results by multiplying each said comparison result by its respective weight; means for summing the weighted comparison results to produce a dissimilarity measure that is indicative of the degree of dissimilarity between the retrieved reference corpus and the retrieved candidate document; and means for storing the candidate document in a retained text store if said sum is indicative of a degree of dissimilarity less than a predetermined degree of dissimilarity.