JP5312531B2 - Text association system and text correspondence program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a sentence associating system and a sentence associating program that execute, when association between sentences related to practical conduct of business is to be established, association between a plurality of sentences. <P>SOLUTION: A sentence associating system 100 associates a sentence for practical conduct of business contained in a practical conduct of business sentence DB22 with a core sentence contained in a core sentence DB21. In doing so, the system figures out a degree of sentence similarity as a measure of identity between words contained in the core sentence and words contained in the sentence for practical conduct of business. Furthermore, each sentence for practical conduct of business is associated with the core sentence that gives the maximum sentence similarity thereto. After that, sentences whose association has been completed are accumulated as part of the core sentence collection, from which learning is to be done. <P>COPYRIGHT: (C)2012,JPO&amp;INPIT

Description

  The present invention relates to a system for associating sentences, and more particularly to a system that handles sentences related to business.

In organizations such as local governments and companies, various methods are known as design methods for optimizing the contents and procedures of operations and the configuration of information systems. For example, EA (Enterprise Architecture) organizes the current status (AsIs) and ideal form (ToBe) of business and systems, and implements step-by-step measures to improve the business and systems aimed at realizing the ideal form By doing so, it aims at standardization and efficiency improvement of business.
For example, regarding the work of the local government, the Ministry of Internal Affairs and Communications, which is the organizing organization, creates a reference model of the local government EA as a base for unifying the business, and provides guidance to the local government to improve the business. The local government aims to unify the business flow by conducting EA analysis using this reference model as a model. This reference model format is disclosed in Non-Patent Document 1 on the Internet.

  In order to examine how to optimize the current state in such a method, it is necessary to clearly understand the correspondence between the current state and the way it should be. For this reason, for example, in the analysis of business in EA, the current business and the business as it should be are described in text, and the work of associating each text is performed.

An example of a technique applicable to such correspondence between sentences is a search system described in Patent Document 1. This search system associates an arbitrarily input example document with one of documents registered in a database.
A knowledge processing system described in Patent Document 2 is an example of a technique for classifying data representing general knowledge instead of sentences and learning the result of classification.

JP 2003-281186 A JP-A-5-143342

Ministry of Internal Affairs and Communications, “Local Government EA Business / System Renewal Guide”, [online], 2006, Ministry of Internal Affairs and Communications, [March 27, 2008 search], Internet <URL: http: //www.soumu.go. jp / denshijiti / system_tebiki / hyouki / gyomu / 2a-4-yokenteigi.html>

However, the conventional technique has a problem in that it is not possible to construct a system that executes correspondence between sentences related to business among a plurality of sentences.
For example, Patent Document 1 does not specify that a plurality of example documents are handled.
In the technique of Patent Document 2, it is necessary to input data in a suitable format as described in paragraph number [0054] and the like. That is, an administrator with advanced knowledge is required to convert a natural language sentence describing business contents into a format that eliminates ambiguity. For this reason, it is difficult to apply the technique of Patent Document 2 to the processing of a sentence representing the content of business.

  The present invention has been made in order to solve such problems, and provides a sentence association system and a sentence association program for executing association of sentences related to work among a plurality of sentences. For the purpose.

  In order to solve the above-described problems, a sentence association system according to the present invention is a sentence association system that associates sentences, and includes a plurality of basic sentences that serve as a basis of association, and a basic sentence. A storage means for storing a plurality of actual business sentences to be associated with each other, and a computing means for associating each of the actual business sentences with any of the basic sentences. Each sentence includes a work number for identifying the actual work sentence and an actual work character string indicating the processing contents of the work corresponding to the actual work sentence, and each of the basic sentences is a number for identifying the basic sentence. And a basic character string representing the processing content of the work corresponding to the basic sentence, the calculating means identifies the words included in each of the basic character string and the actual business character string, and the calculating means includes the basic character string Based on the degree to which the basic word included in the character string matches the actual business word included in the actual business character string, the business number is associated with the number for specifying the basic sentence, and the calculation means Indicates that data for output is generated, and the data for output indicates that the basic sentence and the actual business sentence associated with the basic sentence are output in parallel on the left and right.

The computing means may perform association based on weights defined for each word.
The computing means performs morphological analysis for each of the actual business character string and the basic character string to obtain the actual business word and the basic word, and the arithmetic means for each combination of the actual business character string and the basic character string, The number of matches between the actual business word and the base word or its synonym is calculated, and the computing means multiplies each combination by the number of matches and the weight defined for the matched base word, The degree of coincidence of the actual business word is calculated, and the computing means calculates the total sum of the coincidence degrees of all the actual business words for each combination, and calculates the degree of matching in the combination as the sentence similarity based on the total. Then, the computing means sets each of the actual business character strings with respect to the basic sentence giving the largest sentence similarity, or the sentence similarity is equal to or greater than a threshold value. For all axis sentence, it may be associated.
If the base word is a noun, the computing means sets the weight of the base word to 1, otherwise sets the weight of the base word to 0, and the storage means defines synonyms for the plurality of base words. You may memorize the dictionary file.
The computing means may perform association based on a character string representing the name of information to be input or output of a business and a clause number of a law related to the business.
The storage unit may accumulate and store one basic sentence and all actual business sentences associated with the basic sentence in the association in one basic sentence.
The actual business text may represent the content of the local government business.

  The sentence association program according to the present invention causes a computer to function as the above-described sentence association system.

The sentence association system according to the present invention is a sentence association system that associates sentences with each other, and is a target to be associated with a plurality of basic sentences serving as the basic axes of the association and the basic sentences. A storage means for storing a plurality of actual business sentences, and a computing means for associating each of the actual business sentences with any of the basic sentences, each of which identifies the actual business sentence A work number and an actual work character string indicating the processing contents of the work corresponding to the actual work sentence, and the basic sentence is a number for identifying the basic sentence and the work corresponding to the basic sentence, respectively. And a calculation means for identifying a word included in each of the basic character string and the actual business character string, and the calculation means includes a basic word included in the basic character string, Based on the degree of coincidence with the actual business word included in the business character string, the business number and the number for specifying the basic sentence are associated with each other, and the computing means creates output data, Since the output data indicates that the basic sentence and the actual business sentence associated with the basic sentence are output in parallel on the left and right, the actual business sentence that has been associated with the basic sentence is It is possible to accumulate and store as a part of the basic sentence, learn by this, and execute the correspondence of sentences related to the work of the organization among a plurality of sentences.
In addition, the sentence association program according to the present invention causes the computer to function as the above-described sentence association system, so that the association of sentences related to the work of the organization can be executed among a plurality of sentences.

It is a figure which shows the structure of the text matching system which concerns on this invention. It is a figure which shows the example of a structure of basic text DB of FIG. It is a figure which shows the example of the table | surface used as the origin of the basic text DB of FIG. It is a figure which shows the example of a structure of actual business text DB of FIG. It is a figure which shows the example of the table | surface used as the origin of actual business sentence DB of FIG. It is a figure which shows the example of a structure of the basic word DB of FIG. It is a figure which shows the example of a structure of the basic word word expansion DB of FIG. It is a figure which shows the example of a structure of actual business word DB of FIG. It is a figure which shows the example of a structure of the matching detail DB of FIG. It is a figure which shows the example of a structure of the matching detail DB of FIG. It is a figure which shows the example of a structure of the matching detail DB of FIG. It is a figure which shows the example of a structure of the matching detail DB of FIG. It is a figure which shows the example of a structure of total DB of FIG. It is a figure which shows the example of a structure of matching result DB of FIG. It is a figure which shows the correspondence of FIG. 14 roughly. It is a flowchart showing the flow of a process at the time of the text matching system of FIG. 1 creating a base word DB and a base word expansion DB. It is a flowchart showing the flow of a process at the time of the text matching system of FIG. 1 matching a basic text and a real business text. It is a figure which shows the example of the output of a matching result by the text matching system of FIG. It is a figure which shows the example of a structure of basic text DB updated as a result of the accumulation | storage of FIG. It is a figure explaining the outline | summary of operation | movement of the text matching system of FIG. It is a figure which shows the example of a structure of actual business regulation DB which concerns on Embodiment 2 of this invention. It is a figure which shows the example of a structure of basic axis input DB which concerns on Embodiment 2 of this invention.

Embodiments of the present invention will be described below with reference to the accompanying drawings.
Embodiment 1 FIG.
In FIG. 1, the structure of the text matching system 100 which concerns on this invention is shown. The sentence association system 100 is a system that associates sentences based on a plurality of sentences input from the outside.
The sentence association system 100 has a configuration as a well-known computer, and includes a calculation unit 10 that performs a calculation and a storage unit 20 that stores information. Although not shown, the sentence association system 100 includes an input unit that receives input of data from the outside and an output unit that outputs data to the outside. The computing means 10 includes a CPU (central processing unit), the storage means 20 includes a memory and an HDD (hard disk drive), the input unit includes a keyboard and a mouse, and the output unit includes a display and a printer. Moreover, the text matching system 100 includes a network interface (not shown) that serves as both an input device and an output device for a communication network.
Although not shown, the storage unit 20 stores a sentence association program that defines the operation of the sentence association system 100. The sentence association system 100, which is a computer, implements the functions described in this specification by executing this sentence association program.

  The storage means 20 stores and stores the basic sentence DB 21, the actual business sentence DB 22, the basic word DB 25, the basic word expansion DB 26, the actual business word DB 27, the matching detail DB 28, the aggregation DB 29, and the association result DB 30 as the DB (database). To do. These DBs are constructed, for example, as RDBs (relational database), but may be constructed in other formats.

FIG. 2 shows an example of the configuration of the basic sentence DB 21. Each line in FIG. 2 corresponds to one basic sentence. These basic texts represent the contents of the business of the organization. Moreover, these basic texts are used as the basic axes when the text matching system 100 performs matching between texts. That is, the sentence association system 100 associates sentences by determining which of the basic sentences the other sentences correspond to.
The basic sentence DB 21 is information represented by item names L1, L2, L3, SA1, SA2, SA3, NAIYOU, INPUT, OUTPUT, REF, METHOD, J1, J2, K1, K2, G1, and G2 for each basic sentence. Remember.

L1, L2, and L3 are items representing EA numbers as information for specifying the basic sentence. For example, L1 represents a large item classification, L2 represents a medium item classification, and L3 represents a small item classification. As an example, the basic text in the top row in FIG. 2 is specified by a combination of values L1 = 1, L2 = 1, and L3 = 0. In the following description, when specifying a basic sentence, three EA numbers are written in order and expressed as “base sentence (1-1-0)”.
SA1, SA2 and SA3 are items representing the outline of the work, that is, the main points of the basic sentence. These represent, for example, items corresponding to the EA number as large items, medium items, and small items, respectively.

NAIYOU is an item representing the content of work corresponding to the basic sentence. This item includes, for example, a character string representing the processing contents of the municipal business.
INPUT and OUTPUT are items representing information for specifying materials to be input and output when performing the work. These include, for example, a character string representing the name of the material. In the example of FIG. 2, the basic text (1-1-0) is created by the person in charge of the local government based on the material named “Taxable Person Information”, for example, “Summary Table”. Indicates the work content of

REF is an item representing other information related to the basic sentence.
“METHOD” represents an execution method of the work, and is an item representing whether or not manual work is included, whether or not processing by a computer system is included, and whether or not outsourcing is included.
J1, J2, K1, K2, G1, and G2 are items that represent item numbers of laws and regulations related to work (related item numbers), and in this order, item numbers, item numbers, item numbers, item numbers, Corresponds to issue numbers and issue numbers. For example, in Article 317-6, first term, J1 = 317, J2 = 6, and K1 = 1. In this example, since all of the basic texts included in the basic text DB 21 relate to only the same law (for example, local tax law), the law name is not stored, but an item representing the law name may be additionally provided. Good.

The basic sentence DB 21 is created based on, for example, a table shown in FIG. Here, FIG. 3 is an example of a local government EA reference model defined by the Ministry of Internal Affairs and Communications. The format of this reference model is disclosed in Non-Patent Document 1 described above. In addition, the actual data created according to this model, that is, the contents of the table shown in FIG. 3 can be obtained by the local government inquiring individually from the Ministry of Internal Affairs and Communications.
After inputting this data into the DB, the basic sentence DB 21 can be created by assigning the item names in FIG. 2 to predetermined columns. The assignment of item names is performed by, for example, a specialized person in charge, but may be automatically performed by the text matching system 100 according to a predetermined rule.

  FIG. 4 shows an example of the configuration of the actual business text DB 22. Each line in FIG. 4 corresponds to one actual business sentence. These actual business sentences represent the contents of the business of the organization. In addition, these actual business sentences are objects to which the sentence association system 100 associates with the basic sentence. That is, the text association system 100 associates each of the actual business texts with any of the above-described basic texts.

The actual business text DB 22 stores information represented by item names L1, L2, L3, SA1, SA2, SA3, NAIYO, J1, J2, K1, K2, G1, and G2 for each basic text. These items represent the information (work number) for identifying the actual business text, the work outline of the actual business text, and the work content regarding the actual business text, in the same manner as the basic text DB 21 described above. In the following, in the same way as the basic text, when specifying the actual business text, three business numbers are written in order and expressed as “actual business text (1-1-0)”.
In addition to the items shown in FIG. 4, the actual business sentence DB 22 may include the same INPUT, OUTPUT, REF, and METHOD items as the basic sentence DB 21.

  The actual business sentence DB 22 is created based on, for example, a table shown in FIG. Here, FIG. 5 is a table describing actual work in the local government. After inputting this data into the DB, the actual business sentence DB 22 can be created by assigning the item names of FIG. 4 to predetermined columns. The assignment of item names is performed by, for example, a specialized person in charge, but may be automatically performed by the text matching system 100 according to a predetermined rule.

  FIG. 6 shows an example of the configuration of the basic word DB 25. The base word DB 25 represents information related to each word (hereinafter referred to as “base word”) included in the base sentence. The basic word DB 25 is created based on words included in predetermined items of the basic sentence DB 21, for example, sentences stored in the NAIYOU. In the example of FIG. 6, only the part corresponding to the work content of the basic sentence (1-1-0) and the basic sentence (1-2-0) of FIG. 2 is shown.

The basic word DB 25 stores information represented by item names L1, L2, L3, FUBAN, TANGO, and HINSI for each basic word.
L1, L2, and L3 are the same as the basic sentence DB 21. FUBAN is an item representing a word number as a number assigned to identify each of the basic words included in the same basic sentence (that is, the basic words having the same EA numbers L1, L2, and L3). is there. TANGO is an item representing a character string corresponding to the basic word. HINSI is an item representing the part of speech of the basic word.

FIG. 7 shows an example of the configuration of the basic word expansion DB 26. The basic word extension DB 26 is for associating additional information with each of the basic words included in the basic word DB 25. In the example of FIG. 7, as in FIG. 6, only portions corresponding to the work contents of the basic sentence (1-1-0) and basic sentence (1-2-0) in FIG. 2 are shown.
The base word expansion DB 26 stores information represented by item names L1, L2, L3, FUBAN, TANGO, and HINSI for each base word as in the base word DB25. Further, the basic word expansion DB 26 stores information represented by item names OMOMI and DOUGIGO in association with each basic word. OMOMI is an item that indicates what weight the core word has when the text matching system 100 determines text matching. DOUGIGO is an item representing a word (or a list of such words) having the same meaning or similar meaning to the basic word.
In the example of FIG. 7, a non-zero weight (item name OMOMI) value is set only for words whose part of speech (item name HINSI) is a noun, and all other words have zero weight.

FIG. 8 shows an example of the configuration of the actual business word DB 27. The actual business word DB 27 represents information related to each word (hereinafter referred to as “actual business word”) included in the actual business text. The actual business word DB 27 is created based on words included in a predetermined item of the actual business text DB 22, for example, a text stored in the NAIYOU. In this example, this item name “NAIYOU” is the same as the item name “NAIYOU” that is the basis of the above-mentioned basic word DB 25. In the example of FIG. 8, only the portions corresponding to the work contents of the actual business sentence (1-1-0) and the actual business sentence (1-2-0) of FIG. 4 are shown.
The actual business word DB 27 associates information represented by the item names L1, L2, L3, FUBAN, TANGO, and HINSI with each actual business word in the same manner as the basic word DB 25 described above.

9 to 12 show examples of the configuration of the matching details DB 28. The collation detail DB 28 represents the result of determining whether or not the actual business words match each of the basic words and whether or not they match, and the sentence similarity calculated based on the result.
The result of this match is stored in units of sentence combinations. As an example,
-FIG. 9 shows the result of matching the actual business sentence (1-1-0) with the basic sentence (1-1-0),
-FIG. 10 shows the result of matching the actual business sentence (1-1-0) with the basic sentence (1-2-0),
-FIG. 11 shows the result of matching the actual business sentence (1-2-0) with the basic sentence (1-1-0),
FIG. 12 shows the result of matching the actual business sentence (1-2-0) with the basic sentence (1-2-0).

The matching detail DB 28 is created based on each actual business word, and stores information represented by the item names L1, L2, L3, FUBAN, and TANGO in the same manner as the actual business word DB 27 described above. Furthermore, the matching detail DB 28 stores information represented by item names ITTIDO, ITTIDO2, GATTI in association with each actual business word.
ITTIDO and ITTIDO2 are items representing the degree of matching between words included in the base sentence and words included in the actual business sentence. Of these, ITTIDO represents the sum of the weights OMOMI of the base word that matches the actual business word. ITTIDO2 represents the total of the weights OMOMI of the base word when the actual business word matches the synonym of the base word, that is, matches the word included in the item DOUGIGO of the base word extension DB 26. GATTI represents a basic word that matches the actual business word, that is, a word number FUBAN of the basic word related to ITTIDO and ITTIDO2 (or a list thereof when there are a plurality of basic words).

  In FIG. 9, the word corresponding to the word number 3 of the actual business sentence (1-1-0), that is, the actual business word with L1 = 1, L2 = 1, L3 = 0, and FUBAN = 3 is “obligor”. However, the word “obligor” appears only once in the basic sentence (1-1-0) to be matched (L1 = 1, L2 = 1, L3 = 0, FUBAN = 28 in FIG. 7). ). The weight OMOMI is 2. Therefore, the coincidence degree ITTIDO of this actual business word is 2, and the coincidence number GATTI is 28. Furthermore, since the word “obligor” does not appear as a synonym of the base sentence (1-1-0) to be matched, the matching degree ITTIDO2 for the synonym is zero.

The actual business word corresponding to the word number 17 of the actual business sentence (1-1-0) is “Send”, but this word appears in the basic sentence (1-1-0) to be matched. Without it, ITTIDO becomes zero. However, the synonym “send” is associated with the base word “shipping” of the word numbers 22 and 33 of the base sentence (1-1-0), and they match. Each of these weights is 5. Therefore, ITTIDO2 of this word is 10, and the match number GATTI is “22, 33”.
The total sum of coincidences defined in this way, that is, the sum of the sum of ITTIDO and the sum of ITTIDO2, becomes the sentence similarity representing the degree of matching between the actual business sentence and the base sentence.

  Further, in FIG. 10, the “obligor” of the word number 3 of the actual business sentence (1-1-0) appears only once in the basic sentence (1-2-0) to be matched (FIG. 7). L1 = 1, L2 = 2, L3 = 0, FUBAN = 13). The weight OMOMI is 2. Therefore, the coincidence degree ITTIDO of this actual business word is 2, and the coincidence number GATTI is 13. Furthermore, since the word “obligatory” does not appear as a synonym of the base sentence (1-2-0) to be matched, the matching degree ITTIDO2 for the synonym is zero.

Further, “salary” of the word number 6 of the actual business sentence (1-1-0) appears only once in the basic sentence (1-1-0) to be matched (L1 = 1 in FIG. 7). L2 = 2, L3 = 0, FUBAN = 1). The weight OMOMI is 2. Therefore, the coincidence degree ITTIDO of this actual business word is 2, and the coincidence number GATTI is 1. Furthermore, since the word “salary” does not appear as a synonym of the basic sentence (1-2-0) to be matched, the matching degree ITTIDO2 for the synonym is zero.
In this way, the total degree of coincidence, that is, the sentence similarity is 2 + 2 = 4.

  Similarly, the basic sentence (1-1-0) and the actual business sentence (1-2-0) are collated, and as a result, the contents of FIG. 11 are created. Further, the basic sentence (1-2-0) and the actual business sentence (1-2-0) are collated, and as a result, the contents of FIG. 12 are created.

  FIG. 13 shows an example of the configuration of the summary DB 29. The tabulation DB 29 tabulates the sentence similarity between each basic sentence and each actual work sentence. For example, “4” is stored for the combination of the basic sentence (1-2-0) and the actual business sentence (1-1-0), and this is a sentence similarity of 4 in FIG. Corresponding to something.

FIG. 14 shows an example of the configuration of the association result DB 30. The association result DB 30 represents which actual business sentence is associated with each basic sentence. FIG. 15 schematically shows the correspondence of FIG. These correspondences are determined based on the tabulation DB 29 in FIG. 13 as will be described later. For each actual business sentence, the actual business sentence is associated with a basic sentence that gives the largest sentence similarity value (that is, the maximum value in each column).
In this example, for example, the basic sentence (1-1-0) is associated with a plurality of actual business sentences (1-1-0) and (1-2-0), and the basic sentence (1-0) No actual business text is associated with -5-0). Further, as in the basic sentence (1-3-0) and the basic sentence (1-4-0), the context in the basic sentence DB 21 and the context in the actual business sentence DB 22 of the corresponding business sentences are reversed. Correspondence is also included.

The operation of the sentence association system 100 configured as described above will be described using the flowcharts shown in FIGS. 16 and 17.
FIG. 16 shows a flow of processing when the sentence association system 100 creates the basic word DB 25 and the basic word expansion DB 26 based on the basic sentence DB 21. This process is executed every time the basic sentence DB 21 is created or changed, for example.
First, the computing means 10 of the text association system 100 reads the base text DB 21 and inputs all the base text (step S1).

  Next, the computing means 10 performs a morphological analysis on the character string included in the item “NAIYOU” of each basic sentence, and creates a basic word DB 25 based on the result (step S2). At this time, the computing means 10 assigns word numbering based on the appearance order of words in each basic sentence. Note that the morphological analysis is a process of decomposing a sentence in a natural language such as Japanese into words and specifying the part of speech of each word. Since a technique for performing morphological analysis is known, detailed description thereof is omitted.

Next, the computing means 10 requests and accepts input of information relating to weights and synonyms for each of the basic words included in the basic word DB 25, and creates the basic word expansion DB 26 based on this input (step S3). This input is made, for example, by an administrator of the text matching system 100.
Here, the manager adds a greater weight to more important words in order to improve the validity of the match. For example, as shown in FIG. 7, the administrator assigns a weight 5 to the word “shipping” of the word number 22 of the basic sentence (1-1-0). In addition, in order to prevent words that are not directly related to the organization's business such as “te”, “ni”, “ha”, “ha”, etc. from affecting the association, The weight is set to 0 for parts of speech other than parts of speech to be used (for example, nouns). In this way, the administrator can input a parameter for designating a part of speech to be used in matching and association with the actual business sentence among the parts of speech output in the morphological analysis.
The manager also associates synonyms with important words to improve the validity of the match. For example, as shown in FIG. 7, the administrator associates the synonym “send” with the word “ship” of the word number 22 of the base sentence (1-1-0).

FIG. 17 shows the flow of processing when the text matching system 100 matches the basic text and the actual business text. This process is executed each time the actual business sentence DB 22 is created, for example.
First, the calculation means 10 of the text association system 100 reads the actual business text DB 22 and inputs all the actual business text (step S11). Next, the computing means 10 performs morphological analysis on the character string included in the item “NAIYOU” of each actual business sentence, and creates the actual business word DB 27 based on the result (step S12). This process is performed in the same manner as step S2 in FIG.

Next, the computing means 10 refers to the basic word extension DB 26 and the actual business word DB 27, performs a matching process for matching each basic sentence and each actual business sentence, and creates a matching detail DB 28 based on the results ( Step S13). Here, the computing means 10 treats a group of basic words having the same EA number as corresponding to one basic sentence, and corresponds a group of actual business words having the same business number to one actual business sentence. It is treated as a thing, and the matching process is performed in units of a combination of one basic sentence and one actual business sentence.
The computing means 10 repeats this matching process for all combinations of the basic text and all the actual business text. At this time, as shown in FIGS. 9 to 12, for each combination, the sum of ITTIDO that is the degree of coincidence of the words themselves and the sum of ITTIDO2 that is the degree of coincidence with the synonym are calculated. Also, the sentence similarity for the combination is calculated as the sum of these sums.
In the example of the basic word extension DB 26 of FIG. 7, since only a noun is given a non-zero weight, the computing means 10 determines the degree of coincidence between the noun included in the basic sentence and the noun included in the actual business sentence. The sentence similarity is calculated.

Next, the calculation means 10 totals the results of the matching detail DB 28 and creates a total DB 29 (step S14). For example, since the basic sentence (1-2-0), the basic sentence (1-1-0), and the sentence similarity shown in FIG. 10 are 4, the corresponding field in the tabulation DB 29, that is, the EA number “ The value of the field in the row of L1 = 1, L2 = 2, L3 = 0 and the column of the business number “L1 = 1, L2 = 1, L3 = 0” is 4.
In this way, for all the matching results, the results (sentence similarity) are totaled to create the tabulation DB 29.

Next, the computing means 10 associates sentences based on the summary DB 29 and creates an association result DB 30 (step S15). Here, the calculation means 10 obtains the largest sentence similarity value (that is, the maximum value in each column of the summary DB 29) for each actual business sentence, and the actual business sentence for the base sentence that gives the maximum value. Associate.
For example, in FIG. 13, if the value surrounded by the broken line is the maximum value of each column, the actual business sentence (1-1-0) and the actual business sentence (1-2-0) are both the base sentence (1-1). -0), the actual business sentence (1-3-0) is associated with the basic sentence (1-2-0). In this way, the calculation means 10 performs association based on the sentence similarity, thereby creating the association result DB 30.
Note that this association is not performed based on the maximum value of the sentence similarity, but may be performed based on whether the sentence similarity is equal to or higher than a predetermined threshold. That is, an actual business sentence may be associated with all of the basic sentences whose sentence similarity is equal to or higher than the threshold value, and if there is no basic sentence whose sentence similarity is equal to or higher than the threshold value, It may not be associated with a sentence.

  Next, the computing means 10 creates output data based on the association result, and controls the output unit of the sentence association system 100 based on this data. In response to this control, the output unit outputs the association result to the outside of the sentence association system 100 (step S16). This output is executed as a printing process by a printing apparatus such as a printer, but may be executed as a display process by a display apparatus such as a display.

FIG. 18 shows an example (EA material correspondence table) of output of the association result. In this example, the correspondence is represented in a tabular form, the basic sentence is displayed on the left side, and the corresponding actual business sentence is displayed on the right side. In order to output such a table, the calculation means 10 first calculates the basic sentence recorded in the basic sentence DB 21 based on the EA number based on the correspondence between the EA number recorded in the association result DB 30 and the business number. Extract the actual business text recorded in the actual business text DB 22 by the business number, and then associate the extracted basic document and the actual business document based on the correspondence between the EA number and the business number, and data Turn into. Alternatively, the calculation means 10 receives the designation of the EA number from the input unit, and similarly, based on the correspondence between the designated EA number and the business number recorded in the association result DB 30, the basic document and the business document May be converted into data. (In the example of FIG. 18, the basic document in which L1 = 1 for the EA number, that is, the basic document whose work summary is the personal resident tax is associated with the business document.) 10 indicates that the basic sentence and the actual business sentence associated with the basic sentence are to be output in parallel on the left and right, and creates data for output.
The output data format is, for example, a file format used by a relational database application, but this may be another format, for example, CSV format, HTML format, XML format, PDF format or the like. May be.

Next, the computing means 10 adds each actual business sentence to the basic sentence DB 21 and accumulates it based on the correspondence between the sentences (step S17). This step corresponds to so-called “learning”.
Here, the computing means 10 adds each actual business sentence to the basic sentence associated with the actual business sentence and stores it in the basic sentence DB 21. That is, the storage means 20 of the sentence association system 100 stores each basic sentence and all the actual business sentences associated with the basic sentence in the association in the respective basic sentences. .

FIG. 19 shows an example of the configuration of the basic sentence DB 21 updated as a result of this accumulation. Of the information included in the actual business text DB 22, the item representing the work content, that is, the item NAIYOU, and the item representing the item number of the law related to the work, that is, the items J1, J2, K1, K2, G1, G2, New information is added to the base sentence DB 21. Here, in FIG. 19, the part newly added compared with FIG. 2 is shown enclosed with a broken line.
Note that new information may be added in other items. Further, a new item name may be added to the base sentence DB 21 in accordance with the addition of information.

Next, the computing means 10 performs the same process as in step S2 (FIG. 16) on the character string added to the base sentence DB 21 in step S17, and adds the word included in the added sentence to the base word extension DB 26. (Step S18). That is, the basic word expansion DB 26 is updated based on the added sentence.
Further, the computing means 10 performs the same processing as step S3 (FIG. 16) for the word added to the base word extension DB 26, and adds the weights and synonyms to update the base word extension DB 26 (step S19). .
In this way, each time the processing of FIG. 17 is executed, the sentence association system 100 learns and loads a new actual business sentence from the actual business sentence DB 22 and adds a new basic word to the basic word expansion DB 26. . As a result, the relationship between the basic sentence in the basic word expansion DB 26 and the actual business more accurately reflects the reality, and a more accurate result can be obtained in the next association processing.

The outline of the operation of the text association system 100 according to Embodiment 1 described as described above is summarized as follows using FIG.
First, the administrator of the sentence association system 100 converts the EA material (FIG. 3) of the Ministry of Internal Affairs and Communications into an RDB and creates a basic sentence DB 21 (this process may be automatically performed by the sentence association system 100). The sentence association system 100 creates the base word DB 25 by morphological analysis (steps S1 and S2). Further, the input of weights and synonyms is received, and the basic word expansion DB 26 is created (step S3).

Next, the administrator of the text association system 100 converts the actual business explanation document (FIG. 5) of the local government into an RDB and creates the actual business text DB 22 (even if this processing is automatically performed by the text association system 100). Good). The text association system 100 creates the actual business word DB 27 by morphological analysis (steps S11 and S12).
The sentence association system 100 collates and calculates the sentence similarity for each combination of the basic sentence and the actual business sentence (step S13), sums up the sentence similarity (step S14), and increases the sentence similarity. Then, the result of matching is judged and matched, and the matching result DB is stored (step S15). Further, the result of association is output as an EA material correspondence table (FIG. 18) (step S16), and the actual business sentences are accumulated and learned in the associated basic sentences (step S17). Then, the basic word DB 25 and the basic word expansion DB 26 are updated for the accumulated words (steps S18 and S19).

  As described above, the sentence association system 100 according to Embodiment 1 of the present invention is a sentence association system that associates sentences, and includes a plurality of actual business sentences representing the contents of business of an organization, A storage means for storing a plurality of basic texts that serve as a basis for correspondence, and a computing means for associating each of the actual business texts with any of the basic texts, and relating to the business of the organization Sentence association can be executed between a plurality of basic sentences and a plurality of actual business sentences. In addition, the actual business sentences that have been associated with the basic text can be accumulated as a part of the basic text and can be learned thereby. For this reason, the text matching system 100 can accurately execute text matching related to the work of the organization. In addition, the sentence association program according to the first embodiment causes the computer to function as the sentence association system 100, so that the association of sentences related to the work of the organization can be executed between a plurality of sentences.

Examples of effects obtained by learning will be described below.
In the basic sentence DB 21 of FIG. 2, the item (NAIYOU) representing the work content of the basic sentence (1-2-0) includes the term “payment”. This term is an abbreviation for “salary payment report”, but such abbreviations are not always properly handled in general morphological analysis by a computer. For example, in the basic word DB 25 of FIG. 6, this term is mistaken as a combination of the verb “Supply” in the word number 4 of the EA number (1-2-0) and the suffix “Information” in the word number 5 Have been analyzed.
Further, in the actual business text DB 22 of FIG. 4, the item (NAIYOU) representing the work content of the actual business text (1-4-0) includes the term “salary payment report”. This term indicates one document as a compound word and should be treated as a group of terms. However, in a general morphological analysis by a computer, such a compound word is not necessarily processed appropriately. For example, in the actual business word DB 27 in FIG. 8, this term is divided into three nouns of “salary”, “payment”, and “report” with the word numbers 5 to 7 of the business number (1-4-0). .
As described above, the basic text (1-2-0) and the actual business text (1-4-0) substantially include processing related to the same document as “salary payment report”. Even though this should be taken into account in the calculation of sentence similarity, appropriate processing is not performed in each morphological analysis, and as a result, this term does not match at the time of matching. Thus, in a state where learning is not performed, the value of the sentence similarity is not necessarily optimal.

However, the sentence association system 100 according to Embodiment 1 accumulates the actual business sentences that have been associated with the base sentence as a part of the base sentence, and learns accordingly. In the above example, the basic sentence (1-2-0) and the actual business sentence (1-4-0) have a high degree of matching with words other than the term “salary payment report” (or implementation Since the degree of coincidence of any of the items representing the related law clause numbers, the items representing the input information, the items representing the output information, etc., which will be described later in the second embodiment, is matched, the key sentence (1 As shown in -2-0), it is accumulated as one basic sentence. As a result, the basic sentences after the accumulation are added to the item (NAIYOU) indicating the work content, the abbreviation “pay” (or the verb “pay” and the suffix “ho”), “salary” “payment” "It contains both three nouns" Report ". Therefore, even if a sentence that uses the abbreviation “payroll” is entered as a new actual work sentence, a sentence that uses the compound word “payroll report” is entered. Also, at least one of them will match, and the sentence similarity will be calculated more appropriately.
As described above, the sentence association system 100 can more appropriately calculate the sentence similarity by performing learning and repeating the learning.

In the first embodiment described above, the word weight is determined based on the part of speech, the noun weight is a non-zero value, and the non-noun weight is zero. As a variation, the weight of the word may be determined by other methods without being based on the part of speech.
Moreover, the text matching system 100 requires an input from an administrator when creating or updating the basic word expansion DB 26 in step S3 and step S19. As a modification, the sentence association system 100 may automatically create or update the basic word expansion DB 26.
In this case, for example, the sentence association system 100 may assign a weight “1” to a noun and not assign a weight to other parts of speech. Moreover, you may not give a synonym at all. In this way, since no external input is required, the work procedure can be simplified.
Moreover, the memory | storage means 20 of the text matching system 100 may store beforehand the dictionary file which defined the weight and synonym which should be provided for every word. In this case, the text association system 100 can automatically create or update the basic word expansion DB 26 based on the dictionary file.

  The basic text and the actual business text may be related to the business of an organization other than the local government, and may be related to the business of a company, for example. Further, the sentence association system 100 may be used for any sentence as long as it is not intended to be directly related to the business of the organization and is associated with a plurality of sentences.

  In the first embodiment described above, all the DBs are stored in the storage unit 20 of the sentence association system 100 that is a single computer. As a modification, the text association system may be configured by a plurality of computers, and each DB may be distributed and provided in a plurality of computers. For example, the morphological analysis process in step S2 and step S12 and the matching process in step S13 may be executed by different calculation means of different computers.

Embodiment 2. FIG.
The second embodiment has a configuration in which the texts are associated by using not only the items NAIYO of the basic text DB 21 and the actual business text DB 22 but also information included in other items in the first embodiment. . Hereinafter, differences from the first embodiment will be described.

In the first embodiment, a set including the basic word expansion DB 26 and the actual business word DB 27 is created based on words included in an item (NAIYOU) representing work content. In addition to these, the calculation means of the sentence association system (not shown) according to the second embodiment selects a set of DBs having the same configuration as these items (J1, J2, K1, K2, G1, G2) are also created, and the storage means stores them. That is, the storage means stores the basic law DB and the actual business law DB.
FIG. 21 shows an example of the configuration of the actual business law DB. The actual business law word DB represents a legal clause number related to each of the actual business texts. In this example, for example, Article 317-6 1st term is associated with the actual business sentence (1-1-0). The basic law DB has a similar configuration.

Similarly, a DB set is created for an item (INPUT) representing input information, and a DB set is created for an item (OUTPUT) representing output information. That is, the storage means stores a basic input DB, an actual business input DB, a basic output DB, and an actual business output DB. Note that, unlike the first embodiment, the actual business sentence DB 22 also includes the same items INPUT and OUTPUT as the basic sentence DB 21.
FIG. 22 shows an example of the configuration of the base axis input DB. The base axis input DB is created based on a character string extracted according to a predetermined rule from the item INPUT of the base axis text DB 21 of FIG. For example, from the item (INPUT) representing the input information corresponding to the basic sentence (1-2-0), a character string “Salary payment report” surrounded by a symbol “·” and information representing a line feed, Based on the character string “resident tax return” and the character string “proxy letter”, three lines indicated by the EA number (1-2-0) in FIG. 22 are created. The actual business input DB, the basic output DB, and the actual business output DB have the same configuration.
Further, in the first embodiment, the basic word input DB is created for the basic input DB in the second embodiment so that the basic word extended DB is created for the basic word DB. Thus, the base axis output extension DB is created. The base input expansion DB stores information represented by item names OMOMI and DOUGIGO in association with each character string of the base input DB. OMOMI is an item that represents what weight the character string has when the text matching system determines text matching. DOUGIGO is an item representing a character string (or a list of such character strings) having the same or similar meaning as the character string. The basic output extension DB has the same configuration. In step S3, the computing unit according to the second embodiment further accepts and sets information regarding weights and synonyms in the base axis input extension DB and base axis output extension DB. Alternatively, the calculation means may set a predetermined value for all weights of the base axis input extension DB and the base axis output extension DB, and may not set all synonyms.

  When the calculation unit according to the second embodiment creates the matching detail DB 28 in step S13, the calculation means matches the items (NAIYOU) representing the work content, that is, the basic word expansion DB 26 and the actual business word that are performed in the first embodiment. Not only the matching with DB27, but also the matching of the items (J1, J2, K1, K2, G1, G2) representing the relevant laws and regulations, the items representing the input (INPUT), and the items representing the output (OUTPUT). Do.

  For example, in the collation of items representing related law clause numbers, the collation is performed using the basic law DB and the actual business law DB. In this collation, it is determined whether or not the related law clause numbers completely match for each combination of the basic sentence and the actual business sentence. For example, if the basic text is associated with Article 317-6, Paragraph 2 and the actual business sentence is associated with Article 317-6, Paragraph 1, only a part of the relevant law clause numbers match. Therefore, it is determined that the basic sentence and the actual business sentence are not similar, and the sentence similarity of the combination is 0. When the related law clause numbers completely match, it is determined that the basic sentence and the actual business sentence are similar, and the sentence similarity of the combination is a predetermined value other than zero.

  In addition, in the matching of items (INPUT) representing input information, matching is performed using the basic input expansion DB and the actual business input DB. In the second embodiment, in the same manner as the method for determining the word matching degree for each sentence combination, in the second embodiment, the character string matching degree is determined for each sentence combination. In the same manner as in the first embodiment, the sentence similarity is calculated from the total sum of matching degrees in the second embodiment as well. The same applies to the matching of items (OUTPUT) representing output information.

As described above, a match of items (NAIYOU) representing work contents, a match of items (J1, J2, K1, K2, G1, G2) representing related law clause numbers, a match of items (INPUT) representing inputs, And the item (OUTPUT) showing an output is collated and the sentence similarity between sentences is calculated in each. The sentence similarity is summed up for each combination of sentences, and is then used for final association similar to step S15 (FIG. 17) of the first embodiment.
As described above, the computing unit according to the second embodiment is not limited to the character string that represents the processing contents of the business, but the character string that represents the name of the information that is input or output of the business and the item number of the law related to the business. Based on the above, the association is performed.

As described above, in the second embodiment, since the association is performed based on various items including various information, the accuracy of the association can be further improved.
In Embodiment 2, the accuracy of association can be further improved by changing the weight for each item according to the nature of the organization or the nature of the business. For example, for a business of an organization that is closely related to the law, such as a local government, for the words (ie, item numbers) included in the items (J1, J2, K1, K2, G1, G2) representing the relevant law item numbers. Thus, it is possible to further improve the accuracy by assigning more weight than words included in other items.
Note that the item number is simply a numerical value and does not have the ambiguity that the sentence expression generally includes, so the accuracy may be further improved.

In the second embodiment, the association may be performed based on other items. For example, items that represent the names of laws and regulations related to operations, items that represent the names of external organizations related to operations (such as "Tax Office"), and internal department names related to operations ("Residents Division", "Tax Affairs Section", etc.) ), Items indicating the implementation time and month (“April”, etc.), etc., are considered. In particular, when assuming a tax service of a local government, if the association is performed based on an item representing the name of the tax related to the service, the accuracy may be further improved.
Further, the second embodiment can be modified in the same manner as in the first embodiment.

  10 calculation means, 20 storage means, 21 basic sentence DB, 22 actual business sentence DB, 25 basic word DB, 26 basic word expansion DB, 27 actual business word DB, 28 collation detail DB, 29 aggregation DB, 30 result DB, 100 Sentence matching system.

Claims (8)

A text matching system for matching texts,
Storage means for storing a plurality of basic sentences serving as the basic axes of the association and a plurality of actual business sentences to be associated with the basic sentences;
Computation means for associating each of the actual business texts with any of the basic texts,
Each of the actual business sentences includes a business number for identifying the actual business text, and an actual business character string representing the processing content of the work corresponding to the actual business text,
Each of the basic texts includes a number for identifying the basic text and a basic character string representing the processing content of the work corresponding to the basic text,
The computing means identifies words included in each of the basic character string and the actual business character string,
The calculation means includes the number for identifying a base sentence and the business number based on the degree of coincidence between the basic word included in the basic character string and the actual business word included in the actual business character string. To perform the matching,
The computing means creates output data,
The output data is a sentence association system that indicates that the basic sentence and the actual business sentence associated with the basic sentence are output in parallel on the left and right.   The sentence association system according to claim 1, wherein the computing unit performs the association based on a weight defined for each of the basic words. The arithmetic means performs a morphological analysis for each of the actual business character string and the basic character string to obtain the actual business word and the basic word,
The computing means calculates the number of times the actual business word and the basic word or a synonym thereof match for each combination of the actual business character string and the basic character string,
The computing means multiplies the number of matches by the weight defined for the matched key word for each of the combinations to calculate the degree of match of each actual business word,
For each of the combinations, the calculation means calculates the sum of the matching degrees of all actual business words, and calculates the matching degree in the combination as a sentence similarity based on the sum.
The calculation means associates each of the actual business character strings with the basic sentence that gives the largest sentence similarity, or with all the basic sentences that have the sentence similarity equal to or higher than a threshold value, The sentence matching system according to claim 2. The computing means sets the weight of the base word to 1 when the base word is a noun, and sets the weight of the base word to 0 otherwise.
The sentence storage system according to claim 2 or 3, wherein the storage unit stores a dictionary file that defines synonyms for a plurality of basic words. The computing means is
A character string representing the name of information to be input or output of the business;
The sentence matching system according to any one of claims 1 to 4, wherein the matching is performed based on item numbers of laws and regulations related to the business.   6. The storage unit according to claim 1, wherein the storage unit stores one basic sentence and all actual business sentences that are associated with the basic sentence in the association in the one basic sentence. The sentence matching system according to claim 1.   The sentence correspondence system according to any one of claims 1 to 6, wherein the actual business sentence represents a content of business of a local government.   The text matching program for functioning a computer as a text matching system as described in any one of Claims 1-7.

Images