EP1072982A2
Expired / Lapsed
EP
⏱️ 0.0 years remaining
Method and system for similar word extraction and document retrieval
Assignee: Panasonic Holdings Corp
•
Filing: 2000-07-28
•
Granted: —
•
Expires: 2020-07-28
Abstract
Abstract Similar words for a given sequence of letters are extracted from a document database of documents is provided. The occurrence pattern of the similar words is similar to that of the given sequence. A feature vector table which contains a first feature vector for each of words in the document database is prepared. Statistical information on the given sequence is obtained through a retrieval of the document database with respect to the given sequence. From the statistical information, a second feature vector with respect to the given sequence is calculated. A degree of similarity between the second feature vector and each of the first feature vectors is calculated Such words as yield degrees of similarity higher than a predetermined value is selected.
Inventors
- 👤 Mitsuhiro Sato
- 👤 Hayashi Ito
Bibliographic Details
Application Number
—
Jurisdiction
EP
Data Source
google_patents
Last Synchronized
2026-11-08 12:02
CPC Classifications
G; G06; G06F; G06F16/00; G06F16/30; G06F16/33; G06F16/3331; G06F16/334
Claims Analysis
Total Claims: 13
Claim 1
Independent Claim
121 words
A method of extracting similar words for a given queryfrom a
document database of documents, the occurrence pattern of the similar words
being similar to that of the given query, the method comprising the steps of: preparing a feature vector table which contains a first feature vector
for each of words in said document database; obtaining statistical information on said given query through a
retrieval of said document database with respect to said given query; calculating a second feature vector with respect to said given query
from said statistical information; calculating a degree of similarity between said second feature vector
and each of said first feature vectors; and selecting such words as yield degrees of similarity higher than a
predetermined value.
Claim 2
Dependent Claim
80 words
A method of claim 1, wherein said step of preparing a feature vector
table comprises the steps of: generating an index file from said document database, said index file
being such that statistical information for each of words used in said document
database can be obtained from said index file; and calculating each first feature vector from said index file, and wherein said step of obtaining statistical information comprising the
step of searching said index file for said given query.
Claim 3
Independent Claim
124 words
A method of extracting similar words for a given sequence of letters
from a document database of documents, the occurrence pattern of the similar
words being similar to that of the given sequence, the method comprising the
steps of: preparing a feature vector table which contains a first feature vector
for each of words in said document database; obtaining statistical information on said given sequence through a
retrieval of said document database with respect to said given sequence; calculating a second feature vector with respect to said given sequence
from said statistical information; calculating a degree of similarity between said second feature vector
and each of said first feature vectors; and selecting such words as yield degrees of similarity higher than a
predetermined value.
Claim 4
Dependent Claim
204 words
In a document retrieval system comprising a first document
database and a retriever, a method of expanding a given query to supply a
expanded query to the retriever, the method including the steps of: extracting similar words of said given query or a query word in said
given query from a second document database of documents, the occurrence
pattern of said similar words being similar to that of said given query or said
query word; and utilizing said similar words of said given query or said query word for
said given query, wherein said step of extracting said similar words comprises
the steps of: preparing a feature vector table which contains a first feature vector
for each of words in said second document database; obtaining statistical information on said query or said query word
through a retrieval of said second document database with respect to said
query or said query word; calculating a second feature vector with respect to said query or said
query word from said statistical information; calculating a degree of similarity between said second feature vector
and each of said first feature vectors; and selecting, as said similar words, such words as yield degrees of
similarity higher than a predetermined value.
Claim 5
Dependent Claim
16 words
A method of claim 4, wherein said first and second document
databases are the same thing.
Claim 6
Independent Claim
156 words
A method of receiving any given sequence of letters in a source
language and providing translations in a target language such that an
occurrence pattern of said given sequence in a source-language document
database being the same as that of translations in a target-language document
database wherein one of said two document databases is a translation of the
other, the method comprising the step of: preparing a feature vector table which contains a first feature vector
for each of words in said target-language document database; obtaining statistical information on said given sequence through a
retrieval of said source-language document database with respect to said given
sequence; calculating a second feature vector with respect to said given sequence
from said statistical information; calculating a degree of similarity between said second feature vector
and each of said first feature vectors; and selecting, as said translations, such target-language words as yield
degrees of similarity higher than a predetermined value.
Claim 7
Dependent Claim
82 words
A method of claim 6, wherein said step of preparing a feature vector
table comprises the steps of: generating an index file from said target-language document database,
said index file being such that statistical information for each of words used in
said target-language document database can be obtained from said index file;
and calculating each first feature vector from said index file, and wherein said step of obtaining statistical information comprising the
step of searching said index file for said given sequence.
Claim 8
Dependent Claim
188 words
In a document retrieval system comprising a first document
database and a retriever, a query expanding method of converting a source-language
query into a target-language query to feed the retriever, the method
including the steps of: extracting a translation of each of query words in said source-language
query from a second document database of target-language documents, the
occurrence pattern of said translation being similar to that of each query word;
and replacing each query word in said source-language query with said
translation of said query word, wherein said step of extracting said translation
comprises the steps of: preparing a feature vector table which contains a first feature vector
for each of words in said target-language document database; obtaining statistical information on said query word through a
retrieval of said source-language document database with respect to said
query word; calculating a second feature vector with respect to said query word
from said statistical information; calculating a degree of similarity between said second feature vector
and each of said first feature vectors; and selecting, as said translations, such target-language words as yield
degrees of similarity higher than a predetermined value.
Claim 9
Independent Claim
201 words
A method of receiving a given letter sequence and providing bisected
letter sequences with a sufficient propriety of division, the method comprising
the steps of: forming a retrieval expression for retrieving documents which contains
all of the letters of said letter sequence but does not contain said letter
sequence itself; obtaining statistical information by executing a retrieval based on said
retrieval expression; generates a feature vector for said retrieval expression from said
statistical information, said feature vector being referred to as "mask feature
vector"; dividing said letter sequence into two subsequences shifting the
division point one by one to obtain N-1 sets of bisected letter sequences, where
N is the number of letters in said letter sequence; calculate a feature vector for each of said bisected letter sequences of
each of said N-1 sets; masks each of said calculated feature vectors with the mask feature
vector and normalizing the masked calculated feature vector; calculating a degree of similarity for each of said N-1 sets, said degree
of similarity being defined as an inner product of said bisected letter sequences
of each of said N-1 sets; and outputting such bisected letter sequences as have a degree of
similarity larger than a threshold.
Claim 10
Dependent Claim
262 words
In a system provided with translation means for receiving a given
word in a first language and providing translations in a second language such
that an occurrence pattern of said given word in a first-language document
database being the same as that of translations in a second-language document
database wherein one of said two document databases is a translation of the
other, wherein said translation means comprise a feature vector table which
contains a first feature vector for each of words in said second-language
document database; means for obtaining statistical information on said query
word through a retrieval of said first- language document database with
respect to said query word; means for calculating a second feature vector with
respect to said query word from said statistical information; means for
calculating a degree of similarity between said second feature vector and each
of said first feature vectors; and means for selecting, as said translations, such
second-language words as yield degrees of similarity higher than a
predetermined value, a method of finding, for a given word, at least one variant
including the given word, the method comprising the steps of: obtaining said translations for said given word by using said
translation means; reconfiguring said translation means by replacing said feature vector
table with a feature vector table which contains a third feature vector for each
of words in said first-language document database and by exchanging said
first-language document database and said second-language document
database; and obtaining, as said at least one variant, second translations for each of
said translations by using reconfigured translation means.
Claim 11
Independent Claim
129 words
A system of extracting similar words for a given query from a
document database of documents, the occurrence pattern of the similar words
being similar to that of the given query, the system comprising: means for preparing a feature vector table which contains a first
feature vector for each of words in said document database; means for obtaining statistical information on said given query
through a retrieval of said document database with respect to said given query; means for calculating a second feature vector with respect to said
given query from said statistical information; means for calculating a degree of similarity between said second
feature vector and each of said first feature vectors; and means for selecting such words as yield degrees of similarity higher
than a predetermined value.
Claim 12
Dependent Claim
212 words
In a document retrieval system comprising a first document
database and a retriever, a system for expanding a given query to supply a
expanded query to the retriever, the system comprising: means for extracting similar words of said given query or a query word
in said given query from a second document database of documents, the
occurrence pattern of said similar words being similar to that of said given
query or said query word; and means for utilizing said similar words of said given query or said query
word for said given query, wherein said means for extracting said similar words
comprises: means for preparing a feature vector table which contains a first
feature vector for each of words in said second document database; means for obtaining statistical information on said query or said query
word through a retrieval of said second document database with respect to said
query or said query word; means for calculating a second feature vector with respect to said
query or said query word from said statistical information; means for calculating a degree of similarity between said second
feature vector and each of said first feature vectors; and means for selecting, as said similar words, such words as yield degrees
of similarity higher than a predetermined value.
Claim 13
Dependent Claim
16 words
A method of claim 12, wherein said first and second document
databases are the same thing.
References Cited (Backward Citations)
No citation records indexed.
USPTO Annuity & Maintenance Windows
Maintenance schedule is calculated for granted US utility patents.