# US 8,751,531 B2 · Text mining apparatus, text mining method, and computer-readable recording medium

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 8,751,531 B2
- Document ID: US8751531B2
- Status: Granted
- Kind: B2
- Application number: 13/060,608
- Filed: 2009-08-28
- Published: 2011-06-30
- Granted: 2014-06-10
- Earliest claimed: 2008-08-29
- Assignees: NEC Corporation
- Applicants: Kai Ishikawa; Akihiro Tamura; Shinichi Ando
- Inventors: Kai Ishikawa; Akihiro Tamura; Shinichi Ando
- Web page: https://patentgrep.com/patent/US8751531B2

## CPC

- G06F 40/284: Lexical analysis, e.g. tokenisation or collocates
- G10L 15/26: Speech to text systems (G10L15/08 takes precedence)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

A text mining apparatus, a text mining method, and a program are provided that accurately discriminate inherent portions of each of a plurality of text data pieces including a text data piece generated by computer processing. A text mining apparatus 1 to be used performs text mining using, as targets, a plurality of text data pieces including a text data piece generated by computer processing. Confidence is set for each of the text data pieces. The text mining apparatus 1 includes an inherent portion extraction unit 6 that extracts an inherent portion of each text data piece relative to another of the text data pieces, using the confidence set for each of the text data pieces.

## Claims

### Claim 1

A text mining apparatus for performing text mining, comprising:
at least one input that receives a first text data and a second text data, each word of the first and second text data has a confidence; and
a processor configured to calculate a mutual information amount I(wi, wj) at least based on a frequency of appearance of word wi in the first text data N1st(wi), a frequency of appearance of word wj in the second text data N2nd(wj), and a frequency of co-appearance of wi and wj in both text data N1st,2nd(wi,wj) using the confidence of wj, and the confidence of and use the mutual information amount I(wi, wj) to calculate degrees to which wi and wj correspond to inherent portions of the first and second text data,
wherein
N
1
⁢
st
⁡
(
w
i
)
=
∑
l
⁢
R
1
⁢
st
⁡
(
w
i
,
l
)
,
[
Equation
⁢
⁢
A
]
N
2
⁢
nd
⁡
(
w
j
)
=
∑
l
⁢
R
2
⁢
nd
⁢
(
w
j
,
l
)
,
[
Equation
⁢
⁢
B
]
N
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
=
∑
l
⁢
R
1
⁢
st
⁡
(
w
i
,
l
)
⁢
R
2
⁢
nd
⁡
(
w
j
,
l
)
,
[
Equation
⁢
⁢
C
]
R1st (wi,l) indicates a confidence of wi for record l,
R2nd (wj,l) indicates a confidence of wj for record l.

### Claim 2 (depends on 1)

The text mining apparatus of claim 1, wherein one of the first and second text data is generated by computer processing.

### Claim 3 (depends on 2)

The text mining apparatus of claim 2, wherein the one of the first and second text data is generated by speech recognition.

### Claim 4 (depends on 3)

The text mining apparatus according to claim 3, the processor further configured to set confidences for the words of the one of the first and second text data generated by speech recognition, using a word graph obtained during the speech recognition.

### Claim 5 (depends on 1)

The text mining apparatus according to claim 1, the processor further configured to separately perform text mining of the inherent portions of the first and second text data.

### Claim 6

A text mining method for performing text mining on a first text data and a second text data, wherein each word of the first and second text data has a confidence, comprising
(a) calculating a mutual information amount I(wi, wj) at least based on a frequency of appearance of word wi in the first text data N1st(wi), a frequency of appearance of word wj in the second text data N2nd(wj), and a frequency of co-appearance of wi and wj in both text data N1st,2nd(wi, wj) using the confidence of wi and the confidence of wj; and
(b) using the mutual information amount I(wi, wj) to calculate degrees to which wi and wj correspond to inherent portions of the first and second text data,
wherein
N
1
⁢
st
⁡
(
w
i
)
=
∑
l
⁢
R
1
⁢
st
⁡
(
w
i
,
l
)
,
[
Equation
⁢
⁢
A
]
N
2
⁢
nd
⁡
(
w
j
)
=
∑
l
⁢
R
2
⁢
nd
⁢
(
w
j
,
l
)
,
[
Equation
⁢
⁢
B
]
N
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
=
∑
l
⁢
R
1
⁢
st
⁡
(
w
i
,
l
)
⁢
R
2
⁢
nd
⁡
(
w
j
,
l
)
,
[
Equation
⁢
⁢
C
]
R1st (wi,l indicates a confidence of wi for record l,
R2nd (wj,l) indicates a confidence of wj for record l.

### Claim 7 (depends on 6)

The text mining method of claim 6, wherein one of the first and second text data is generated by computer processing.

### Claim 8 (depends on 7)

The text mining method of claim 7, wherein the one of the first and second text data is generated by speech recognition.

### Claim 9 (depends on 8)

The text mining method according to claim 8, further comprising setting confidences for the words of the one of the first and second text data generated by speech recognition, using a word graph obtained during the speech recognition.

### Claim 10 (depends on 6)

The text mining method according to claim 6, further comprising separately performing text mining of the inherent portions of the first and second text data.

### Claim 11

A non-transitory computer-readable medium comprising a program for performing operations on a first text data and a second text data, wherein each word of the first and second text data has a confidence, the program including instructions operable to, when executed by a computing device, cause the computing device to:
(a) calculate a mutual information amount I(wi, wj) at least based on a frequency of appearance of word wi in the first text data N1st(wi), a frequency of appearance of word wj in the second text data N2nd(wj), and a frequency of co-appearance of wi and wj in both text data N1st,2nd(wi,wj) using the confidence of wi and the confidence of wj; and
(b) use the mutual information amount I(wi, wj) to calculate degrees to which wi and wj correspond to inherent portions of the first and second text data,
wherein
N
1
⁢
st
⁡
(
w
i
)
=
∑
l
⁢
R
1
⁢
st
⁡
(
w
i
,
l
)
,
[
Equation
⁢
⁢
A
]
N
2
⁢
nd
⁡
(
w
j
)
=
∑
l
⁢
R
2
⁢
nd
⁢
(
w
j
,
l
)
,
[
Equation
⁢
⁢
B
]
N
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
=
∑
l
⁢
R
1
⁢
st
⁡
(
w
i
,
l
)
⁢
R
2
⁢
nd
⁡
(
w
j
,
l
)
,
[
Equation
⁢
⁢
C
]
R1st (wi,l) indicates a confidence of wi for record l,
R2nd (wj,l) indicates a confidence of wj for record l.

### Claim 12 (depends on 11)

The non-transitory computer-readable medium of claim 11, wherein one of the first and second text data is generated by computer processing.

### Claim 13 (depends on 12)

The non-transitory computer-readable medium of claim 12, wherein the one of the first and second text data is generated by speech recognition.

### Claim 14 (depends on 13)

The non-transitory computer-readable medium according to claim 13, wherein confidences are set for the words of the one of the first and second text data generated by speech recognition, using a word graph obtained during the speech recognition.

### Claim 15 (depends on 11)

The computer-readable medium according to claim 11, the program further including instructions operable to, when executed by the computing device, cause the computing device to separately perform text mining of the inherent portions of the first and second text data.

### Claim 16 (depends on 1)

The text mining apparatus of claim 1, wherein the first and second text data originate from a same event via different channels.

### Claim 17 (depends on 1)

The text mining apparatus of claim 1, wherein the processor extracts the inherent portions based on the calculated degrees.

### Claim 18 (depends on 6)

The text mining method of claim 6, wherein the first and second text data originate from a same event via different channels.

### Claim 19 (depends on 6)

The text mining method of claim 6, further comprising:
(c) extracting the inherent portions based on the calculated degrees.

### Claim 20 (depends on 11)

The non-transitory computer-readable medium of claim 11, wherein the first and second text data originate from a same event via different channels.

### Claim 21 (depends on 11)

The non-transitory computer-readable medium of claim 11, the program further including instructions operable to, when executed by the computing device, cause the computing device to:
(c) extract the inherent portions based on the calculated degrees.

### Claim 22 (depends on 1)

The text mining apparatus of claim 1, wherein:
I
⁡
(
w
i
;
w
j
)
=
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
⁢
log
⁢
⁢
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
P
1
⁢
st
⁡
(
w
i
)
⁢
P
2
⁢
nd
⁡
(
w
j
)
+
[
P
1
⁢
st
⁡
(
w
i
)
-
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
⁢
log
⁢
[
P
1
⁢
st
⁡
(
w
i
)
-
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
P
1
⁢
st
⁡
(
w
i
)
⁡
[
1.0
-
P
2
⁢
nd
⁡
(
w
j
)
]
+
[
⁢
P
2
⁢
nd
⁡
(
w
j
)
-
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
⁢
log
⁢
⁢
[
P
2
⁢
nd
⁡
(
w
j
)
-
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
[
1.0
-
P
1
⁢
st
⁡
(
w
i
)
]
⁢
P
2
⁢
nd
⁡
(
w
j
)
+
[
⁢
1.0
-
P
1
⁢
st
⁡
(
w
i
)
-
P
2
⁢
nd
⁡
(
w
j
)
+
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
⁢
log
⁢
⁢
[
1.0
-
P
1
⁢
st
⁡
(
w
i
)
-
P
2
⁢
nd
⁡
(
w
j
)
+
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
[
1.0
-
P
1
⁢
st
⁡
(
w
i
)
]
⁡
[
1.0
-
P
2
⁢
nd
⁡
(
w
j
)
]
,
[
Equation
⁢
⁢
D
]
P
1
⁢
st
⁡
(
w
i
)
=
N
1
⁢
st
⁡
(
w
i
)
/
L
,
[
Equation
⁢
⁢
E
]
P
2
⁢
nd
⁡
(
w
j
)
=
N
2
⁢
nd
⁡
(
w
j
)
/
L
,
[
Equation
⁢
⁢
F
]
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
=
N
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
/
L
,
[
Equation
⁢
⁢
G
]
l is any natural number from 1 to L.

### Claim 23 (depends on 1)

The text mining apparatus of claim 1, wherein:
R
1
⁢
st
⁡
(
w
i
,
l
)
=
1.0
-
∏
m
⁢
⁢
(
1.0
-
R
1
⁢
st
⁡
(
w
i
,
l
,
m
)
)
,
[
Equation
⁢
⁢
H
]
R
2
⁢
nd
⁡
(
w
j
,
l
)
=
1.0
-
∏
m
⁢
⁢
(
1.0
-
R
2
⁢
nd
⁡
(
w
j
,
l
,
m
)
)
,
[
Equation
⁢
⁢
I
]
m is a time frame of record l.

### Claim 24 (depends on 6)

The text mining method of claim 6, wherein:
I
⁡
(
w
i
;
w
j
)
=
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
⁢
⁢
log
⁢
⁢
P
1
⁢
st
⁢
.2
⁢
nd
⁡
(
w
i
,
w
j
)
P
1
⁢
st
⁡
(
w
i
)
⁢
P
2
⁢
nd
⁡
(
w
j
)
+
[
P
1
⁢
st
⁡
(
w
i
)
-
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
⁢
⁢
log
⁢
⁢
[
P
1
⁢
st
⁡
(
w
i
)
-
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
P
1
⁢
st
⁡
(
w
i
)
⁡
[
1.0
-
P
2
⁢
nd
⁡
(
w
j
)
]
+
[
⁢
P
2
⁢
nd
⁡
(
w
j
)
-
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
⁢
⁢
log
⁢
⁢
[
P
2
⁢
nd
⁡
(
w
j
)
-
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
[
1.0
-
P
1
⁢
st
⁡
(
w
i
)
]
⁢
P
2
⁢
nd
⁡
(
w
j
)
+
[
⁢
1.0
-
P
1
⁢
st
⁡
(
w
i
)
-
P
2
⁢
nd
⁡
(
w
j
)
+
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
⁢
⁢
log
⁢
⁢
[
1.0
-
P
1
⁢
st
⁡
(
w
i
)
-
P
2
⁢
nd
⁡
(
w
j
)
+
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
[
1.0
-
P
1
⁢
st
⁡
(
w
i
)
]
⁡
[
1.0
-
P
2
⁢
nd
⁡
(
w
j
)
]
,
[
Equation
⁢
⁢
D
]
⁢
P
1
⁢
st
⁡
(
w
i
)
=
N
1
⁢
st
⁡
(
w
i
)
/
L
,
[
Equation
⁢
⁢
E
]
⁢
P
2
⁢
nd
⁡
(
w
j
)
=
N
2
⁢
nd
⁡
(
w
j
)
/
L
,
[
Equation
⁢
⁢
F
]
⁢
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
=
N
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
/
L
,
[
Equation
⁢
⁢
G
]
l is any natural number from 1 to L.

### Claim 25 (depends on 6)

The text mining method of claim 6, wherein:
R
1
⁢
st
⁡
(
w
i
,
l
)
=
1.0
-
∏
m
⁢
⁢
(
1.0
-
R
1
⁢
st
⁡
(
w
i
,
l
,
m
)
)
,
[
Equation
⁢
⁢
H
]
R
2
⁢
nd
⁡
(
w
j
,
l
)
=
1.0
-
∏
m
⁢
⁢
(
1.0
-
R
2
⁢
nd
⁡
(
w
j
,
l
,
m
)
)
,
[
Equation
⁢
⁢
I
]
m is a time frame of record l.

### Claim 26 (depends on 11)

The non-transitory computer-readable medium of claim 11, wherein:
I
⁡
(
w
i
;
w
j
)
=
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
⁢
⁢
log
⁢
⁢
P
1
⁢
st
⁢
.2
⁢
nd
⁡
(
w
i
,
w
j
)
P
1
⁢
st
⁡
(
w
i
)
⁢
P
2
⁢
nd
⁡
(
w
j
)
⁢
⁢
+
[
P
1
⁢
st
⁡
(
w
i
)
-
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
⁢
⁢
log
⁢
⁢
[
P
1
⁢
st
⁡
(
w
i
)
-
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
P
1
⁢
st
⁡
(
w
i
)
⁡
[
1.0
-
P
2
⁢
nd
⁡
(
w
j
)
]
+
[
⁢
P
2
⁢
nd
⁡
(
w
j
)
-
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
⁢
⁢
log
⁢
⁢
[
P
2
⁢
nd
⁡
(
w
j
)
-
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
[
1.0
-
P
1
⁢
st
⁡
(
w
i
)
]
⁢
P
2
⁢
nd
⁡
(
w
j
)
⁢
+
[
⁢
1.0
-
P
1
⁢
st
⁡
(
w
i
)
-
P
2
⁢
nd
⁡
(
w
j
)
+
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
⁢
⁢
log
⁢
⁢
[
1.0
-
P
1
⁢
st
⁡
(
w
i
)
-
P
2
⁢
nd
⁡
(
w
j
)
+
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
]
[
1.0
-
P
1
⁢
st
⁡
(
w
i
)
]
⁡
[
1.0
-
P
2
⁢
nd
⁡
(
w
j
)
]
,
[
Equation
⁢
⁢
D
]
⁢
P
1
⁢
st
⁡
(
w
i
)
=
N
1
⁢
st
⁡
(
w
i
)
/
L
,
[
Equation
⁢
⁢
E
]
⁢
P
2
⁢
nd
⁡
(
w
j
)
=
N
2
⁢
nd
⁡
(
w
j
)
/
L
,
[
Equation
⁢
⁢
F
]
⁢
P
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
=
N
1
⁢
st
,
2
⁢
nd
⁡
(
w
i
,
w
j
)
/
L
,
[
Equation
⁢
⁢
G
]
l is any natural number from 1 to L.

### Claim 27 (depends on 11)

The non-transitory computer-readable medium of claim 11, wherein:
R
1
⁢
st
⁡
(
w
i
,
l
)
=
1.0
-
∏
m
⁢
⁢
(
1.0
-
R
1
⁢
st
⁡
(
w
i
,
l
,
m
)
)
,
[
Equation
⁢
⁢
H
]
R
2
⁢
nd
⁡
(
w
j
,
l
)
=
1.0
-
∏
m
⁢
⁢
(
1.0
-
R
2
⁢
nd
⁡
(
w
j
,
l
,
m
)
)
,
[
Equation
⁢
⁢
I
]
m is a time frame of record l.