JPH096796A - Method and device for generating collation pattern - Google Patents

Abstract

PURPOSE: To provide a method and device for generating collation pattern with which burden for a user to prepare a collation pattern is canceled, at the same time, the construction of huge knowledge for collation is unnecessitated as well and the pattern with high collation accuracy is generated. CONSTITUTION: This device is provided with a suitability discriminating means 1 for judging whether the user is suitable for an entire text or one part of the text as the result of collation due to text collating processing or not and a collation pattern generating means 6 for generating a collation pattern 10 only from suitability information 8 of the text, for which it is judged by the suitability discriminating means 1 that the user is suitable, and unsuitability information 9 of the text for which the user is judged unsuitable.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a collation pattern generation method and apparatus, and more particularly to a collation pattern formed by using an input pattern or a generated pattern when performing information retrieval or information extraction. The present invention relates to a collation pattern generation method and apparatus for collating and collating and generating an appropriate collation pattern from the collation result.

[0002]

2. Description of the Related Art Conventionally, relatively common proper nouns such as a person's name, a company name, a product name, etc. in a text and the contents of the text are used as clues for a text matching process necessary for information retrieval or information extraction. It uses relatively specialized common nouns or verbs, adjectives, and adjectives such as adjectives as keywords. In the text matching process, one or a plurality of these keywords are used to combine them to create a matching pattern called a search formula or an extraction pattern, and matching is performed using the pattern.

A conventional collation pattern generation device for generating a retrieval formula or an extraction pattern is largely classified into the following two types according to its input format.
Classified into one. The device in which the user directly inputs the collation pattern; the device in which the user inputs in natural language; and the device in which the user directly inputs the collation pattern are devices that directly perform the text collation processing using the user's input as the collation pattern. The user recalls one or a plurality of keywords by himself / herself and combines them to create a search expression or a matching pattern of an extraction pattern.

The user's device for inputting in natural language is a device for extracting collating information such as a keyword and a relationship between keywords from a user's natural language input or a limited one. This device builds in advance various collating knowledge such as knowledge held by individual users, knowledge held by the text, and knowledge regarding correspondence between the knowledge. Since this matching knowledge is associated with the keywords that the user may input and the vocabulary on the text side, the matching knowledge is referenced to refer to the vocabulary on the user side and the relationship between the vocabulary and the text. Correlate side vocabularies and relationships between them.

[0005]

However, the above conventional method has the following problems. The above-mentioned device in which the user generates a search expression or an extraction pattern and directly inputs the matching pattern requires a user to create the search expression or the extraction pattern by himself or herself, which imposes a heavy burden on the user. . Further, since the vocabulary on the text side to be matched and the relationship between the vocabularies are not clear to the user, an appropriate pattern cannot be generated and the matching accuracy becomes low.

Further, the above-mentioned apparatus for the user to input in the natural language only indirectly corresponds to the matching information extracted from the natural language input by the user by the matching knowledge built in advance, The user usually cannot refer to this matching knowledge. For this reason, when the user cannot obtain a satisfactory output result, it is not possible to specify where the cause is in the input, and it is difficult to change or improve the collation pattern.

The present invention has been made in view of the above points, solves the above-mentioned conventional problems, eliminates the burden of creating a collation pattern for a user, and at the same time eliminates the need for constructing a vast amount of collation knowledge. Moreover, it is an object of the present invention to provide a matching pattern generation method and device that can generate a pattern with high matching accuracy and at the same time make it possible to gradually improve the accuracy by repeating the matching process.

[0008]

FIG. 1 is a diagram for explaining the principle of the present invention. According to the present invention, in a collation pattern generation method for generating a collation pattern for text collation processing, when a collation result obtained by the text collation processing is input (step 1), the entire text or a part of the text which is the collation result is input by the user. Judges conformity or nonconformity (step 2), and if the user judges that it is conforming, acquires the conformance information consisting of the whole text or a part of the text (step 3). Acquires the incompatibility information consisting of the entire text or a part of the text (step 4), and generates a matching pattern represented as a keyword logical expression or a regular expression from only the compatibility information and the incompatibility information (step 5).

Further, according to the present invention, the matching information of the whole text or a part of the text judged by the user to match is read, partial information common to the matching information is extracted, and ranking is performed according to the length and frequency of the partial information. Until the pattern corresponding to the matching information and the one-to-one correspondence is obtained according to the order, the appearance of the partial information in the matching information is patterned into a matching pattern, the pattern inclusion relationship between the matching patterns is specified, and stored in the storage means. The non-conformance information that has been accumulated and read by the user as non-conformance is read, and the non-conformance information is collated with the conformance pattern of the pattern inclusion relation stored in the storage means, and the conformance pattern that has been successfully collated is non-conformance. All patterns at the top of the inclusion relations from the storage unit that has been removed as a pattern from the storage means and the incompatible pattern has been removed. Extraction, repeated series of processes to output as the matching pattern bonded at selected format.

Further, the present invention uses a keyword logical expression or a regular expression as the matching pattern 10. FIG.
FIG. 3 is a principle configuration diagram of the present invention. According to the present invention, in a collation pattern generation device for generating a collation pattern of a text collation process, a conformity determination means 1 for a user to determine conformity or nonconformity with respect to the whole text or a part of the text which is the collation result of the text collation process. And the matching information 8 consisting of the whole text or a part of the text judged by the matching judgment means 1 to be suitable by the user, and the non-matching information 9 consisting of the whole text or a part of the text judged to be unmatched by the user, based on the keyword logic. It has a matching pattern generating means 6 for generating a matching pattern 10 represented as an expression or a regular expression.

Further, according to the present invention, by the conformity determining means 1,
Conformance information accumulating means for accumulating the conformity information 8 judged to be conforming, nonconformance information accumulating means for accumulating the nonconformance information 9 judged to be nonconforming by the conformity judging means 1, and conformance reading the conformance information 8 from the conformance information accumulating means. An information reading means, a conforming pattern forming means for extracting common part information from the conforming information 8 to form a pattern, and a pattern inclusion relation specifying means for specifying an inclusion relation between the matching patterns patterned by the matching pattern forming means, The pattern inclusion relationship accumulating unit that accumulates the inclusion relationship between the patterns specified by the pattern inclusion relationship specifying unit, the incompatibility information reading unit that reads the incompatibility information 9 from the incompatibility information storage unit, and the incompatibility information 9 that is read by the incompatibility information reading unit.
Is compared with the pattern stored in the pattern inclusion relation accumulating unit, and the unsatisfactory pattern removing unit that removes the successfully collated pattern from the pattern inclusion relationship accumulating unit and the incompatibility information accumulated in the pattern inclusion relationship accumulating unit. From the matching patterns that do not match, the top-level pattern extraction means that extracts all the top-level patterns in the pattern inclusion relationship between them and all the top-level patterns that are extracted by the top-level pattern extraction means are selected. And a collation pattern output means for combining and outputting as a collation pattern.

[0012]

The present invention focuses on the user's conformity / nonconformity judgment with respect to the collation result, and generates the collation pattern only from the conformity / nonconformity information. Therefore, when the user tries to perform the collation again, it is only necessary to determine the suitability or improperness of the collation result, and it is possible to eliminate the burden on the user of creating the collation pattern.

Further, the present invention enumerates the patterns to be matched with the matching information, and narrows down the unmatched patterns as patterns to be matched to the patterns not to be matched with the unmatched information. As a result, it is possible to cover the matching patterns with few matching errors. In addition, the pattern inclusion relation specifying unit specifies the inclusion relation of those matching patterns, and extracts all the top patterns in the pattern inclusion relation. This allows
It is possible to take out the pattern with the smallest matching omission among the matching patterns. Therefore, the collation pattern output unit can generate a highly accurate pattern.

Further, according to the present invention, the common part information is extracted from the matching information, the matching information is patterned, and the pattern to be matched with the unmatching information is removed, so that the matching pattern is generated only from the matching / non-matching information, and is huge. It is not necessary to build accurate matching knowledge. Further, according to the present invention, it is possible to gradually improve the matching accuracy by the generated pattern by repeating the generation of the matching pattern, the matching processing, and the processing of determining whether the matching result is appropriate or not.

[0015]

Embodiments of the present invention will now be described in detail with reference to the drawings. FIG. 3 shows the configuration of a text matching system to which the present invention is applied. The text matching system shown in the figure includes a matching pattern generation device of the present invention. The text collation system includes an existing text collation device 1, a collation pattern generation device 6, a user evaluation reading unit 7, a conformance information storage unit 8 and an incompatibility information storage unit 9.

The text collating apparatus 1 receives a collation pattern 2 such as a keyword logical expression or a regular expression, and collates the text stored in the text base 3 with the collation pattern 2 and succeeds in the collation. The comparison result output unit 5 outputs the entire text or a part of the text.

The collation pattern generation device 6 constructed according to the present invention is a user evaluation reading unit 7 for reading the conformity / nonconformity judgment of the user with respect to the entire text output by the existing text collation device 1 or a part thereof.
And a matching information storage unit 8 that stores all or part of the text classified as matching information by the user evaluation reading unit 7, and all or part of text that is classified as unmatching information by the user evaluation reading unit 7. The existing text collation device 1 is combined by using the nonconformance information storage unit 9 that stores, and the collation pattern 10 is generated only from the conformance information / nonconformance information. The matching pattern 10 is output as an input to the existing text matching device 1,
The existing text matching device 1 performs matching again using this matching pattern 10.

FIG. 4 is a block diagram of a collation pattern generation processing apparatus according to an embodiment of the present invention. The matching pattern generation device includes a matching information reading unit 11 and a matching pattern forming unit 1.
2, the pattern inclusion relationship specifying unit 13, the pattern inclusion relationship storage unit 14, the incompatibility information reading unit 15, the incompatibility pattern removal unit 16, the highest level pattern extraction unit 17, and the collation pattern output unit 18.

The matching information reading unit 11 reads the entire text or a part of the information judged by the user as matching from the matching information storage unit 8 and transfers it to the matching pattern forming unit 12. The matching pattern forming unit 12 forms a pattern by extracting common part information from the matching information, and extracting the matching information and the common part information to form a pattern.

The pattern inclusion relation specifying unit 13 specifies the inclusion relation between the matching patterns patterned by the matching pattern forming unit 12 and transfers it to the pattern containing relation accumulating unit 14. The pattern inclusion relation accumulating unit 14 accumulates the inclusion relation between the matching patterns identified by the pattern inclusion relation identifying unit 13.

The incompatibility information reading unit 15 reads the entire text or a part thereof determined by the user as incompatible from the incompatibility information storage unit 9. Incompatible pattern removing unit 16
Compares the incompatibility information read from the incompatibility information reading unit 15 with the pattern stored in the pattern inclusion relationship storage unit 14, and removes the matched pattern from the pattern inclusion relationship storage unit 14.

The highest-level pattern extraction unit 17 selects all the patterns at the highest level among the pattern inclusion relationships between the matching patterns that do not match the mismatch information remaining in the pattern inclusion relationship storage unit 14. Take out.
The matching pattern output unit 18 combines all the uppermost patterns extracted by the uppermost pattern extracting unit 17 in a selection format and outputs them as a matching pattern.

Next, the operation of the above configuration will be described. FIG.
6 is a flowchart of a matching pattern generation process according to an embodiment of the present invention. Step 101) The matching information reading unit 11 reads the matching information from the matching information storage unit 8.

Step 102) Matching pattern construction unit 1
2 counts all the character strings appearing in the matching information, and obtains the overlap of these character strings (partial information) (the product of the length and frequency of the partial information). Among the partial information, those having a frequency of 2 or more (common partial information) are ranked according to the size of the overlap, and the leading common partial information is adopted as the pattern constituent element.

Step 103) Next, the conforming pattern construction unit 12 composes the appearance of the pattern constituent elements in the conforming information as a conforming pattern. (Step 104) Further, in the conforming pattern forming unit 12, it is checked whether or not the conforming pattern and the conforming information have a one-to-one correspondence.
To step 105, and if they do not correspond, step 105 is performed.

Step 105) Further, the conforming pattern construction unit 12 adopts the next candidate of the common part information as a pattern constituent element in the order of the common part information, and executes step 103 again. Step 106) The pattern inclusion relation specifying unit 13 specifies the pattern inclusion relation between the matching patterns.

Step 107) The pattern inclusion relation accumulating unit 14 accumulates the matching patterns and the inclusion relation between them. Step 108) The nonconformance information reading unit 15 reads the nonconformity information from the nonconformance information storage unit 9.

Step 109) In the nonconforming pattern removing section 16, the nonconforming information is stored in the pattern inclusion relation storing section 1
The matching pattern stored in 4 is matched, and the matching pattern that has been successfully matched is removed from the pattern inclusion relationship storage unit 14 as an unmatched pattern. Step 110) The highest-level pattern extraction unit 17 extracts all the patterns at the highest levels of the respective inclusion relationships from the pattern inclusion relationship storage unit 14 from which the incompatible patterns have been removed.

Step 111) Collation pattern output unit 1
In 8, all the top patterns are combined in a selection format and output as a matching pattern. Hereinafter, a detailed description will be given with reference to FIGS. [First Specific Example] First, an example of collation when a keyword logical expression is used as a collation pattern and an inclusion relation between patterns will be described. An example in which the matching pattern of the specific example shown below is composed of a logical expression of keywords will be described. FIG.
It is a figure for demonstrating the pattern inclusion relationship of one Example of this invention.

[Example] [New vehicle sales in July, again below the previous year] The number of new vehicle sales in July (preliminary) excluding minicars summarized by the Association of ABC Motor Sales Association on the 3rd (preliminary report) decreased by 5.6% from the same month last year. 541,599 units. As new car sales in June exceeded the same month last year, we expected that the new car sales slumped to the bottom, and we hoped for the bonus sales in July .... ”In the above example, the keyword logical formula was used. Is a combination of arbitrary keywords such as “∧” (logical product), “∨” (logical sum), and “(”, “)”. For example, the keyword formula “New vehicle sales in July ∧ YoY decrease” matches information that includes both “New vehicle sales in July” and “YoY decrease”.

It is judged that this keyword formula is (a) "New car sales in July, again below the previous year", but (b) "The number of new car sales in July compiled by the Automobile Sales Association is , 541,599 units are not included in the information (b), which includes "new car sales in July" but does not include "decrease in previous year".

The keyword formula "car sales ∧ new car sales in July" includes the information (a) because the information (a) includes "new car sales in July" and does not include "car sales". Information (b) is the keyword "car sales" and "new car sales in July"
Since both and are included, it is checked against the information (b).

Here, A is a set of information to be matched with the keyword formula "New car sales in July ∧ YoY sales", and B is a set to be matched with the keyword formula "Car sales ∧ New cars in July". To do. In addition, the set of information to be matched with the keyword logical expression "new car sales in July" is U
Then, the relationship between the sets A, B and U is as shown in FIG. When the set U includes A (or B), that is, when they have the following relationship, A (or B) ⊆ U keyword formula "new car sales in July" is the keyword formula "July New car sales ∧ fall below the previous year (or “car sales ∧ new car sales in July”) are included.

An example of generating such a keyword logical expression as a matching pattern will be described with reference to FIGS. (1) The matching information reading unit 11 reads the matching information from the four matching information storage units 8 shown in FIG. 7 (step 1
01).

(2) The conforming pattern construction unit 12 obtains common part information and its overlap from the conforming information. FIG.
The example of the common part information extracted by the matching pattern structure part of one Example of this invention is shown. In the example of the figure, since "car sales" has the largest overlap, "car sales" is adopted as the pattern component (step 102).

(3) Next, the matching pattern construction unit 12 checks whether or not the matching information includes a pattern constituent element. All pattern components included in the conformance information are connected by "∧". Here, since there is only one type of pattern component, the appearance of the pattern component in the matching information is as follows: 1: car sales 2: car sales 3: car sales 4: It is represented as "car sales", and eventually one kind of matching pattern "car sales" is obtained (step 103).

(4) Furthermore, the conforming pattern forming unit 12
In step 1, the matching pattern “car sales” does not correspond to the matching information having four types of information on a one-to-one basis.
05 is executed (step 104). (5) Further, in the adaptive pattern configuration unit 12, the common portion information having the next largest overlap, “for used car sale” and “new car sale for July” are adopted as pattern components, and step 103 is performed again. Execute (step 10)
5).

(6) Next, whether or not the pattern information "car sales", "used car sales", and "new car sales in July" are included in the matching information in the matching pattern construction unit 12 In a pattern (step 10
3). As a result, the matching pattern is the matching information 1 in FIG.
For example, it is represented as follows: 1: Selling used cars 2: Selling cars ∧ Selling used cars 3: Selling new cars in July 4: Car sales ∧ Selling new cars in July

(7) Further, the adaptive pattern forming unit 12
In, the matching patterns “also used car sales” and “new car sales in July” correspond to two pieces of information in the matching information. Therefore, since the matching pattern and the matching information do not have a one-to-one correspondence, the process moves to step 105 (step 1
04).

(8) Furthermore, the conforming pattern forming unit 12
In step 2, the common part information having the next largest overlap, “monthly” and “less than the previous year” are adopted as pattern components, and step 103 is executed again. (9) Next, in the adaptive pattern configuration unit 12, the pattern components “car sales”, “used car sales also”, “new car sales in July”, “monthly”, and “year-on-year decrease” are applicable information. The pattern indicates whether it is included or not. As a result, the conformance patterns are as follows for the conformance information 1 to 4 in Fig. 1: 1: Used car sales ∧ Month ∧ YoY decrease 2: Car sales ∧ Used car sales 3: New car sales in July ∧ YoY decrease 4: Car sales ∧ New car sales in July are represented (step 103).

(10) Furthermore, the conforming pattern forming section 1
In 2, the matching pattern has a one-to-one correspondence with each piece of matching information, so step 106 is executed (step 104). (11) In the pattern inclusion relationship specifying unit 13, the matching patterns are arranged in an array including only different patterns in the order in which they are obtained.

"Car sales""Used car sales""Car sales ∧ Used car sales""July new car sales""Car sales ∧ July new car sales""Used car sales ∧ Month sales ∧ Decrease of previous year ”“ New car sales in July ∧ Decrease of previous year ”Identify the inclusion relationship between these conformance patterns (Step 1
06).

(12) In the pattern inclusion relation accumulating unit 14, the inclusion relation between the matching patterns as shown in FIG. 10 is temporarily accumulated (step 107). (13) The incompatibility information reading unit 15 acquires the incompatibility information shown in FIG. 8 from the incompatibility information storage device 9 (step 108).

(14) In the nonconforming pattern removing unit 16, the nonconforming information is collated with the conforming pattern accumulated in the pattern inclusion relation accumulating unit 14. As a result, it was found that two of the matching patterns “car sales” and “used car sales” succeeded in collating with the nonconformance information, and when these are removed from the pattern inclusion relationship storage unit 14, the pattern inclusion relationship storage unit 14 11 is as shown in FIG. 11 (step 109). As shown by being surrounded by a dotted line in the drawing of FIG. 11, three inclusion relationships a, b, and c are included. The top level pattern of the inclusion relationship a is "new car sales in July", the top level pattern of the inclusion relationship b is "even used cars ∧ months ∧ less than the previous year", and the top level pattern of the inclusion relationship c is "Car sales ∧ also used car sales".

(15) Top-level pattern extracting section 17
Is a pattern inclusion relationship accumulating unit 14 from which non-conforming patterns are removed, and all top-level patterns “cars” located at the top of each inclusion relationship shown in FIG. "Sale ∧ Used car sales""July new car sales""Used car sales ∧ Month ∧ YoY decrease" (step 110).

(16) In the collation pattern output section 18, the extracted highest-ranking patterns are combined by "∨" (logical sum) to obtain the final collation pattern (also for car sales ∧ used car sales) ∨ July New car sales ∨ (also used car sales ∧ month ∧ year-on-year decrease) is output (step 111).

By the above processing, the burden of creating a collation pattern on the user is eliminated, and at the same time, it is not necessary to construct a huge amount of collation knowledge, and at the same time a pattern with high collation accuracy is generated, the collation processing is repeated. A gradual improvement in accuracy is possible. A second specific example will be shown below.

[Second Specific Example] First, an example of collation when a keyword logical expression is used as a collation pattern will be described with respect to the inclusion relation between patterns. An example in which the matching pattern is represented by a regular expression is shown below.

The second specific example also has the same inclusion relation as that of the first specific example. The regular expression is
An arbitrary character string is formed by combining "." (Any one character), "*" (repeated zero or more times), "|" (selection), and "(", ")". For example, the regular expression "new car sales in July. * Less than last year" is matched with the matching information including "new car sales in July" and "lower than last year" in that order, or a part thereof.

This regular expression matches (a) "new car sales in July, again below the previous year", but (b) "new car sales in July compiled by the Automobile Sales Association is 540,000. It became 1599 units. Even if the information (b) includes "new vehicle sales in July", it does not include "new year sales" and is not checked.

The regular expression "car sales. * New car sales in July" includes information "a new car sales in July" and does not include "car sales". Although not collated, the information (b) is collated with the information (b) because both the character strings “car sale” and “new car sale in July” are included in that order.

Here, let A be the set of information that is matched with the regular expression "new car sales in July. * YoY decrease", and B be the set that is matched with the regular expression "car sales. * New car sales in July". . Further, if the set of information to be matched with the regular expression "new car sales in July" is U, the relationship between sets A, B and U is as shown in FIG. When the set U includes A (or B), that is, when they have the following relationship, A (or B) ⊆ U The regular expression "new car sale in July" is the regular expression "new car in July". Sales. * YoY decrease (or "car sales. * New car sales in July") are included.

An example of generating such a regular expression as a matching pattern is shown in FIGS. 7 to 9 and FIGS.
This will be described with reference to FIG. (1) Compliance information reading unit 11
Reads the information in the matching information storage unit 8 shown in FIG. 7 (step 101).

(2) As shown in FIG. 9, the conforming pattern forming section 12 obtains common part information and its overlap from the conforming information. First, "car sales" is adopted as a pattern component (step 102). (3) Next, the matching pattern forming unit 12 represents the distribution of the pattern constituent elements on the matching information by a regular expression.
Here, since there is only one type of pattern constituent element, the appearance of the pattern constituent element in the matching information is as follows with respect to the matching information 1 to 4 shown in FIG. * Car sales. * 2 :. * Car sales. * Car sales * 3: * Car sales. * 4 :. * Car sales. It is represented as * car sales * and two types of matching patterns are obtained (step 103).

(4) Furthermore, the conforming pattern forming unit 12
In the above, since the matching patterns “. * Car sales. *” And “car sales. *” Do not correspond to the matching information having four types of information on a one-to-one basis, step 105 is executed (step 104). (5) Further, in the adaptive pattern configuration unit 12, the common portion information having the next largest overlap, “for used car sale” and “new car sale for July” are adopted as pattern components, and step 103 is performed again. Execute (step 10)
5).

(6) Next, the matching pattern construction unit 12 represents the distribution of the pattern components "car sales" and "used car sales" and "new car sales in July" on the matching information as patterns. . As a result, the conforming pattern corresponds to the conforming information 1 to 4 in FIG. * Also used car sales. * 2 :. * Car sales. * Also used car sales. * 3 :. * New car sales in July. * 4 :. * Car sales. * New car sales in July. It is expressed as * (step 103).

(7) Further, the adaptive pattern forming unit 12
In, the conformance patterns “. * Used car sales. *” And “. * New car sales in July. *” Correspond to two pieces of information in the conformance information. Therefore, since the matching pattern and the matching information do not have a one-to-one correspondence, step 105 is executed (step 104).

(8) Further, the adaptive pattern construction unit 12
In, the common part information having the next largest overlap, “monthly” and “less than the previous year” are adopted as pattern components, and step 103 is executed (step 105). (9) Next, in the conforming pattern configuration unit 12, the pattern constituent elements “car sales”, “used car sales also”, “new car sales in July”, “monthly” and “year-on-year decrease” on the conformance information. The distribution of "" is represented in a pattern. As a result, the matching pattern is as follows for the matching information 1 to 4 in FIG. * Of the month. * YoY change 2: Car sales. * Used car sales also 3: New car sales in July. * YoY decline 4: Car sales. * Expressed as new car sales in July (step 103).

(10) Furthermore, the conforming pattern construction unit 1
In 2, the matching pattern has a one-to-one correspondence with each piece of matching information, so step 106 is executed (step 104). (11) In the pattern inclusion relationship specifying unit 13, the matching patterns are arranged in an array including only different patterns in the order in which they are obtained.

“. * Car sales. * ”“. * Car sales. * Car sales. * ”“. * Also used car sales. * ”“. * Car sales. * Also used car sales. * ”“. * New car sales in July. * ”“. * Car sales. * New car sales in July. * "" For used car sales. * Of the month. * YoY lower ”" New car sales in July. * YoY decrease. * ”The inclusion relation between these matching patterns is specified (step 1
06).

(12) The pattern inclusion relation accumulating unit 14 forms an inclusion relation between the matching patterns shown in FIG.
(Step 107). (13) The incompatibility information reading unit 15 reads out the incompatibility information shown in FIG. 8 from the incompatibility information storage unit 9 (step 108).

(14) In the nonconforming pattern removing unit 16, the nonconforming information is collated with the conforming pattern accumulated in the pattern inclusion relation accumulating unit 14. Thereby, in the matching pattern, ". * Car sales. * ”“. * Also used car sales. It can be seen that the two matching patterns “*” succeed in matching with the non-matching information. When these are removed from the pattern inclusion relation accumulating unit 14, the contents of the pattern inclusion relation accumulating unit 14 become as shown in FIG. 13 (step 109). FIG. 13 shows three inclusion relations d, e, and f as surrounded by a dotted line in the figure.
It is included. The top level pattern of the inclusion relationship d is “. * New vehicle sales in July. *”, And the top level pattern of the inclusion relationship e is “For used cars.
The highest-level pattern of the inclusion relation f is “. * Car sales. * Car sales. *”.

(15) Top-level pattern extracting section 17
In order to minimize the omission of collation from the pattern inclusion relationship storage unit 14 in which the nonconforming pattern is removed,
All the topmost patterns located at the top of the respective inclusion relations shown in FIG. * Car sales. * Also used car sales. * ”“. * New car sales in July. * "" For used car sales. * Of the month. * Under the previous year '.

(16) In the collation pattern output unit 18, these are combined by "|" (selection), and the final collation pattern "(car sales. * Used car sales also) | new car sales in July |
(For used car sales. * Month. * YoY decrease)

By the above processing, the burden of creating a user collation pattern is eliminated, at the same time it is not necessary to build a huge amount of collation knowledge, and a pattern with high collation accuracy is generated, and at the same time, the accuracy of the collation processing is repeated. It becomes possible to gradually improve. It should be noted that the present invention is not limited to the above-described embodiment, but can be variously modified and applied within the scope of the claims.

[0066]

As described above, according to the matching pattern generation method and apparatus of the present invention, the user determines whether the acquired matching result is the matching information or the non-matching information and accumulates it. The partial information common to the matching information is extracted, the appearance situation of the partial information is patterned until the pattern corresponding to the matching information and the one-to-one correspondence is obtained, and the pattern inclusion relationship between the matching patterns is determined. Accumulate. Therefore, the nonconformance information is collated with the conforming pattern in the pattern inclusion relation, the successful pattern is removed from the pattern inclusion relation, and the pattern at the top of the finally remaining pattern is set as the collation pattern. Pattern is never used as a matching pattern.

The present invention also collates nonconformance information,
Matched matching patterns are regarded as non-conforming patterns,
Since the collation patterns are acquired from the patterns that are finally deleted and are deleted from the stored pattern inclusion relationships, the burden of creating the collation patterns for the user is reduced, and the collation accuracy can be improved gradually. A matching pattern can be generated.

Further, since the matching pattern is generated only from the matching information and the mismatching information judged by the user, a huge amount of matching knowledge is unnecessary.

[Brief description of drawings]

FIG. 1 is a diagram for explaining the principle of the present invention.

FIG. 2 is a principle configuration diagram of the present invention.

FIG. 3 is a configuration diagram of a text matching device to which the present invention is applied.

FIG. 4 is a configuration diagram of a matching pattern generation device according to an embodiment of the present invention.

FIG. 5 is a flowchart of matching pattern generation processing according to an embodiment of the present invention.

FIG. 6 is a diagram for explaining a pattern inclusion relationship according to an embodiment of the present invention.

FIG. 7 is a diagram showing an example of matching information according to an embodiment of the present invention.

FIG. 8 is a diagram showing an example of incompatibility information according to an embodiment of the present invention.

FIG. 9 is a diagram showing an example of common part information extracted by a matching pattern forming unit according to an embodiment of the present invention.

FIG. 10 is a diagram showing an example of a matching pattern (keyword logical expression) inclusion relationship according to an embodiment of the present invention.

FIG. 11 is a diagram showing an example of a conforming pattern (keyword logical expression) inclusion relation in which an unconforming pattern is removed according to an embodiment of the present invention.

FIG. 12 is a diagram showing an example of a conforming pattern (regular expression) inclusion relationship according to an embodiment of the present invention.

FIG. 13 is a diagram showing an example of a conforming pattern (regular expression) inclusion relation in which the nonconforming pattern of the present invention is removed.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 text collation device, conformity determination means 2 collation pattern 3 text base 4 text collation unit 5 collation result output unit 6 collation pattern generation device, collation pattern generation unit 7 user evaluation reading unit 8 conformity information accumulation unit, conformance information 9 nonconformance information accumulation Part, non-conformance information 10 Matching pattern 11 Matching information reading unit 12 Matching pattern forming unit 13 Pattern inclusion relationship specifying unit 14 Pattern inclusion relationship accumulating unit 15 Non-matching information reading unit 16 Non-matching pattern removing unit 17 Top-level pattern extracting unit 18 Matching pattern output unit

Claims (6)

[Claims] 1. A collation pattern generation method for generating a collation pattern of text collation processing, wherein the user judges conformity or nonconformity with respect to the entire text or a part of the text which is the collation result by the text collation processing, The matching pattern is generated only from the matching information consisting of the entire text or a part of the text judged by the user and the unmatching information consisting of the whole text or a part of the text judged to be unsuitable by the user. Matching pattern generation method. 2. The matching information of the entire text or a part of the text that the user has determined to be matching is read, partial information common to the matching information is extracted, and ranking is performed according to the length and frequency of the partial information. The appearance of the partial information in the matching information is patterned into a matching pattern until a pattern corresponding to the matching information in a one-to-one correspondence is obtained according to the order, and a pattern inclusion relationship between the matching patterns is specified. , The non-conformance information accumulated in the storage means and read by the user as the non-conformance of the whole text or a part of the text, and the non-conformance information is collated with the conformity pattern of the pattern inclusion relation stored in the storage means. , The matching pattern that has been successfully matched is removed from the storage means as a non-matching pattern, and the non-matching pattern is removed. The collation pattern generation method according to claim 1, wherein all patterns at the highest level of the inclusion relation are taken out from the storage unit, combined in a selected format, and output as a collation pattern. 3. The matching pattern generation method according to claim 1, wherein a keyword logical expression or a regular expression is used as the matching pattern. 4. A matching pattern generation device for generating a matching pattern of a text matching process, wherein the user judges whether the whole text or a part of the text that is the matching result of the text matching process is matched or unmatched. Means, and matching information consisting of the whole text or a part of the text judged by the user to be matching by the matching determining means,
A collation pattern generation device comprising: a collation pattern generation means for generating a collation pattern from only the nonconformance information consisting of the whole text or a part of the text judged by the user as nonconformity. 5. The conformity information accumulating means for accumulating the conformity information judged to be conforming by the conformity judging means, the nonconformity information accumulating means for accumulating the nonconformity information judged to be nonconformity by the conformity judging means, Between the matching information reading unit that reads the matching information from the matching information storage unit, the matching pattern forming unit that extracts common part information from the matching information to form a pattern, and the matching pattern that is patterned by the matching pattern forming unit Pattern inclusion relation specifying means for specifying an inclusion relation, pattern inclusion relation storing means for storing the inclusion relation between the patterns specified by the pattern inclusion relation specifying means, and nonconformance information reading for reading the nonconformance information from the nonconformance information storage means Means and the incompatibility read by the incompatibility information reading means Information is collated with the pattern stored in the pattern inclusion relation accumulating means, and the pattern that has been successfully collated is removed from the pattern inclusion relation accumulating means, and the nonconforming pattern accumulating means is accumulated in the pattern inclusion relation accumulating means. From the matching patterns that are not matched with the non-matching information, the highest-level pattern extracting unit that extracts all the highest-level patterns in the pattern inclusion relationship between them, and the highest-level pattern extracting unit that extracts all the highest-level patterns. The collation pattern generation device according to claim 4, further comprising collation pattern output means for combining the upper patterns in a selection format and outputting as a collation pattern. 6. The collation pattern generation device according to claim 4, wherein a keyword logical expression or a regular expression is used as the collation pattern.