Word/phrase classification processing method and apparatus

Claim 2

A word/phrase classification processing method, comprising the steps of: performing a first clustering process by extracting “V” different words from text data as a one-dimensional sequence of a plurality of words, and separating a set of the “V” different words into “C” word classes; extracting a set of word class sequences in which all of the degrees of stickiness between contiguous word classes are equal to or more than a predetermined value, in a one-dimensional sequence of word classes concerning the text data generated based on the first clustering process; generating a one-dimensional sequence of words and tokens concerning the text data by corresponding a specific token to a word class sequence, searching a word sequence belonging to the word class sequence for the text data, and replacing the word sequence in the text data with a corresponding token; performing a second clustering process, by extracting different words and tokens from the one-dimensional sequence of words and tokens concerning the text data, and separating a set of extracted words and tokens into word/token classes; and performing a third clustering process by extracting a word sequence corresponding to a token as a phrase, from word sequences included in the text data, replacing the token in the word/token class with the phrase, and separating the set of extracted words and phrases into word/phrase classes.

Claim 3

The word/phrase classification processing method according to claim 1 , wherein: the first clustering process is performed based on an average mutual information between word classes.

Claim 4

The word/phrase classification processing method according to claim 1 , wherein: the second clustering process is performed based on an average mutual information between word/token classes.

Claim 5

A word/phrase classification processing method, comprising t he steps of: generating word classes into which words included in text data are classified; generating a one-dimensional sequence of word classes by mapping the word classes to a one-dimensional sequence of words included in the text data; extracting a word class sequence in which all of the degrees of stickiness between contiguous word classes are equal to or more than a predetermined value, in the one-dimensional sequence of word classes concerning the text data, from the one-dimensional sequence of word classes of the text data; classifying the words and the word class sequences, which are included in the text data, together; extracting a phrase by taking out respective words which exist contiguously in the text data, from respective word classes structuring the word class sequence; and replacing the word class sequence with a phrase belonging to the word class sequence.

Claim 6

A word/phrase classification processing method, comprising t he steps of: generating word classes into which words included in text data are classified; generating a one-dimensional sequence of word classes by mapping the word classes to a one-dimensional sequence of words included in the text data; extracting a word class sequence in which all of the degrees of stickiness between contiguous word classes are equal to or more than a predetermined value, in the one-dimensional sequence of word classes concerning the text data, from the one-dimensional sequence of word classes of the text data; classifying the words and the word class sequences, which are included in the text data, together; extracting a phrase by taking out respective words which exist contiguously in the text data, from respective word classes structuring the word class sequence; and replacing the word class sequence with a phrase belonging to the word class sequence.

Claim 7

A phrase extraction method, comprising the steps of: generating word classes into which words included in text data are classified; generating a one-dimensional sequence of word classes by mapping the word classes to a one-dimensional sequence of words included in the text data; extracting a word class sequence in which all of the degrees of stickiness between contiguous word classes are equal to or more than a predetermined value, in the one-dimensional sequence of word classes of the text data, from the one-dimensional sequence of word classes of the text data; and extracting a phrase by taking out respective words which exist contiguously in the text data, from respective word classes structuring the word class sequence.

Claim 8

A phrase extraction method, comprising the steps of: generating word classes into which words included in text data are classified; generating a one-dimensional sequence of word classes by mapping the word classes to a one-dimensional sequence of words included in the text data; extracting a word class sequence in which all of the degrees of stickiness between contiguous word classes are equal to or more than a predetermined value, in the one-dimensional sequence of word classes of the text data, from the one-dimensional sequence of word classes of the text data; and extracting a phrase by taking out respective words which exist contiguously in the text data, from respective word classes structuring the word class sequence.

Claim 9

A word/phrase classification processing apparatus, comprising: word classifying means for generating word classes by extracting different words from a one-dimensional sequence of words included in text data, and separating a set of extracted words; word class sequence generating means for generating a one-dimensional sequence of word classes concerning the text data by replacing respective words structuring a one-dimensional sequence of words included in the text data with the word classes to which the different words belong; word class sequence extracting means for extracting a word class sequence in which all of the degrees of stickiness between contiguous word classes are equal to or more than a predetermined value, in the one-dimensional sequence of word classes concerning the text data, from the one-dimensional sequence of word classes concerning the text data; token attaching means for attaching a token to each word class sequence extracted by the word class sequence extracting means; word/token sequence generating means for generating a one-dimensional sequence of words and tokens concerning the text data by replacing a word sequence belonging to the word class sequence extracted by the word class sequence extracting means, from the one-dimensional sequence of words included in the text data, with the token; word/token classifying means for generating a word/token class by separating a set of words and tokens, which are included in the one-dimensional sequence of words and tokens concerning the text data; and phrase replacing means for generating a phrase by inversely replacing the token in the word/token class, with a word sequence which is replaced by the word/token sequence generating means.

Claim 10

A word/phrase classification processing apparatus, comprising: word classifying means for generating word classes by extracting different words from a one-dimensional sequence of words included in text data, and separating a set of extracted words; word class sequence generating means for generating a one-dimensional sequence of word classes concerning the text data by replacing respective words structuring a one-dimensional sequence of words included in the text data with the word classes to which the different words belong; word class sequence extracting means for extracting a word class sequence in which all of the degrees of stickiness between contiguous word classes are equal to or more than a predetermined value, in the one-dimensional sequence of word classes concerning the text data, from the one-dimensional sequence of word classes concerning the text data; token attaching means for attaching a token to each word class sequence extracted by the word class sequence extracting means; word/token sequence generating means for generating a one-dimensional sequence of words and tokens concerning the text data by replacing a word sequence belonging to the word class sequence extracted by the word class sequence extracting means, from the one-dimensional sequence of words included in the text data, with the token; word/token classifying means for generating a word/token class by separating a set of words and tokens, which are included in the one-dimensional sequence of words and tokens concerning the text data; and phrase replacing means for generating a phrase by inversely replacing the token in the word/token class, with a word sequence which is replaced by the word/token sequence generating means.

Claim 11

The word/phrase classification processing apparatus according to claim 6 , wherein: the word classifying means comprises: initialization class assigning means for extracting the different words from the one-dimensional sequence of words included in the text data, and assigning a specific word class to each of extracted words having a predetermined frequency of appearances; temporary merging means for extracting two word classes from a set of word classes, and temporarily merging extracted word classes; average mutual information amount calculating means for calculating an average mutual information about temporarily-merged word classes of the text data; and merging means for actually merging two word classes which maximize the average mutual information, which are included in the set of word classes.

Claim 12

The word/phrase classification processing apparatus according to claim 6 , wherein: the word class sequence extracting means comprises: word class extracting out means for sequentially taking out two word classes which exist contiguously, from the one-dimensional sequence of word classes concerning the text data; mutual information amount calculating means for calculating the amount of mutual information between two word classes taken out by the word class extracting means; and class chain linking means for linking two word classes whose amount of mutual information is equal to or more than a predetermined value, by using a class chain.

Claim 13

The word/phrase classification processing apparatus according to claim 6 , wherein: the word/token classifying means comprises: initialization class assigning means for extracting different words and tokens from the one-dimensional sequence of words and tokens concerning the text data, and assigning a specific word/token class to each word and each token having a predetermined frequency of appearances; temporary merging means for taking out two word/token classes from a set of word/token classes, and temporarily merging extracted word/token classes; average mutual information amount calculating means for calculating an average mutual information about temporarily-merged word/token classes of the text data; and merging means for actually merging two word/token classes which maximize the average mutual information, which are included in the set of word/token classes.

Claim 14

A word/phrase classification processing apparatus, comprising: a phrase extracting unit to extract a phrase formed of contiguous words from text data as a one-dimensional sequence of a plurality of words; and a word/phrase classifying unit to classify both a word included in the text data and the phrase extracted by said phrase extracting unit as one block, and generating a class in which the word and the phrase exist together.

Claim 15

A word/phrase classification processing apparatus, comprising: a phrase extracting unit to extract a phrase formed of contiguous words from text data as a one-dimensional sequence of a plurality of words; and a word/phrase classifying unit to classify both a word included in the text data and the phrase extracted by said phrase extracting unit as one block, and generating a class in which the word and the phrase exist together.

Claim 16

The word/phrase classification processing apparatus according to claim 10 , wherein: the class is generated based on an average mutual information about classes.

Claim 17

A phrase extraction apparatus, comprising: word classifying means for classifying words included in text data, and generating word classes; word class sequence generating means for generating a one-dimensional sequence of word classes of the text data by replacing respective words structuring a one-dimensional sequence of words included in the text data, with the word classes to which the words belong; word class sequence extracting means for extracting a word class sequence in which the degrees of stickiness between contiguous word classes are equal to or more than a predetermined value, in the one-dimensional sequence of word classes concerning the text data, from the one-dimensional sequence of word classes concerning the text data; and phrase extracting means for extracting a phrase by taking out respective words which exist contiguously in the text data from respective word classes structuring the word class sequence.

Claim 18

A phrase extraction apparatus, comprising: word classifying means for classifying words included in text data, and generating word classes; word class sequence generating means for generating a one-dimensional sequence of word classes of the text data by replacing respective words structuring a one-dimensional sequence of words included in the text data, with the word classes to which the words belong; word class sequence extracting means for extracting a word class sequence in which the degrees of stickiness between contiguous word classes are equal to or more than a predetermined value, in the one-dimensional sequence of word classes concerning the text data, from the one-dimensional sequence of word classes concerning the text data; and phrase extracting means for extracting a phrase by taking out respective words which exist contiguously in the text data from respective word classes structuring the word class sequence.

Claim 19

The word extraction apparatus according to claim 12 , wherein: the word classes are generated based on an average mutual information about word classes.

Claim 20

A storage medium, comprising: a first storage area containing a data structure for word classes into which words included in text data as a one-dimensional sequence of a plurality of words are classified; a second storage area containing a data structure for phrases, each formed of contiguous words extracted from the text data based on mutual information amounts of the word classes; and a third storage area containing a data structure for word/phrase classes where the phrases are classified in the word classes.

Claim 21

A storage medium, comprising: a first storage area containing a data structure for word classes into which words included in text data as a one-dimensional sequence of a plurality of words are classified; a second storage area containing a data structure for phrases, each formed of contiguous words extracted from the text data based on mutual information amounts of the word classes; and a third storage area containing a data structure for word/phrase classes where the phrases are classified in the word classes.

Claim 22

A computer-readable storage medium storing a program which includes instructions for making a computer perform the functions of: generating word classes by extracting different words from a one-dimensional sequence of words included in text data, and separating a set of extracted words; generating a one-dimensional sequence of word classes concerning the text data, by replacing respective words structuring the one-dimensional sequence of words included in the text data with the word classes to which the different words belong; extracting a word class sequence in which all of the degrees of stickiness between contiguous word classes are equal to or more than a predetermined value, from the one-dimensional sequence of word classes concerning the text data; attaching a token to the word class sequence; generating a one-dimensional sequence of words and tokens concerning the text data by replacing a word sequence belonging to the word class sequence in the one-dimensional sequence of words included in the text data, with the token; generating a word/token class by separating a set of words and tokens in the one-dimensional sequence of words and tokens concerning the text data; and generating a phrase by inversely replacing the token in the word/token class with the word sequence included in the text data.

Claim 23

A computer-readable storage medium storing a program which includes instructions for making a computer perform the functions of: generating word classes by extracting different words from a one-dimensional sequence of words included in text data, and separating a set of extracted words; generating a one-dimensional sequence of word classes concerning the text data, by replacing respective words structuring the one-dimensional sequence of words included in the text data with the word classes to which the different words belong; extracting a word class sequence in which all of the degrees of stickiness between contiguous word classes are equal to or more than a predetermined value, from the one-dimensional sequence of word classes concerning the text data; attaching a token to the word class sequence; generating a one-dimensional sequence of words and tokens concerning the text data by replacing a word sequence belonging to the word class sequence in the one-dimensional sequence of words included in the text data, with the token; generating a word/token class by separating a set of words and tokens in the one-dimensional sequence of words and tokens concerning the text data; and generating a phrase by inversely replacing the token in the word/token class with the word sequence included in the text data.