JP6973782B2 - Standard item name setting device, standard item name setting method and standard item name setting program - Google Patents

Description

The present invention relates to a standard item name setting device, a standard item name setting method, and a standard item name setting program.

Many forms are used by local governments and companies. Although the form is generally a paper medium, it is desired to reduce the management cost of the form by using an input form in which the form of the paper medium is digitized.

For example, Patent Document 1 below discloses a system that determines the type of form and uses an input form according to the type of form to process the acceptance of the form.

Japanese Unexamined Patent Publication No. 2004-126910

However, even if the forms are of the same type, the names (item names) of the corresponding items may differ depending on the local government, company, or the like. Therefore, when trying to standardize the item names for many types of forms, the list of item names becomes enormous, and there is a problem that the labor is extremely large when trying to organize them manually.

The present invention has been made in view of the above problems, and an object thereof is a standard item name setting device capable of setting a standard item name for a corresponding item name in a plurality of forms, a standard item name setting method, and a standard item name setting method. It is to provide a standard item name setting program.

According to the standard item name setting device according to the present invention, the above-mentioned problem is to be assigned to an item name acquisition unit for acquiring a plurality of item names described in a plurality of forms and the plurality of item names in one of a plurality of clusters. Based on the clustering unit to be classified and the item name classified into the attention cluster among the plurality of clusters, the type determination unit for determining the type of the attention cluster from the plurality of types and the type determination unit determine the type. Based on the rule corresponding to the type, the item names classified in the attention cluster are subdivided into subclusters consisting of item names having a common character string, and the item names belonging to the subcluster are the same characters. The item name candidate generation unit that generates a plurality of item name candidates based on a character string other than a column, and the item name candidate selected from the plurality of item name candidates of the subcluster based on a predetermined criterion are selected as the sub. This is solved by providing a standard item name setting unit that is set as a standard item name corresponding to the cluster.

In the above problem, according to the standard item name setting method according to the present invention, a computer acquires a plurality of item names described in a plurality of forms and classifies the plurality of item names into one of a plurality of clusters. , The type of the attention cluster is determined from the plurality of types based on the item name classified into the attention cluster among the plurality of clusters, and the attention cluster is determined based on the rule corresponding to the determined type. The item names classified in are subdivided into subclusters consisting of item names having a common character string, and multiple item name candidates based on character strings other than the common character string are selected for the item names belonging to the subcluster. The problem is solved by setting an item name candidate that is generated and selected from the plurality of item name candidates of the subcluster based on a predetermined criterion as a standard item name corresponding to the subcluster.

According to the standard item name setting program according to the present invention, the above-mentioned problem is to be assigned to an item name acquisition unit for acquiring a plurality of item names described in a plurality of forms and the plurality of item names in one of a plurality of clusters. Based on the clustering unit to be classified and the item name classified into the attention cluster among the plurality of clusters, the type determination unit that determines the type of the attention cluster from among the plurality of types, and the type determination unit determines the type. Based on the rules corresponding to the types, the item names classified in the attention cluster are subdivided into subclusters consisting of item names having a common character string, and the item names belonging to the subcluster are the same characters. The item name candidate generation unit that generates a plurality of item name candidates based on a character string other than a column, and the item name candidate selected from the plurality of item name candidates of the subcluster based on a predetermined criterion are selected as the sub. This is solved by making the computer function as the standard item name setting unit that is set as the standard item name corresponding to the cluster.

According to the above-mentioned standard item name setting device, standard item name setting method, and standard item name setting program, one standard item name can be set for the corresponding item name in a plurality of forms. As a result, it is possible to reduce the labor of combining a plurality of corresponding item names into one item name.

The standard item name setting device includes a feature vector generation unit that generates a feature vector of each of the plurality of item names, and the clustering unit is based on the similarity of each feature vector of the plurality of item names. , It is preferable to classify the plurality of item names into the plurality of clusters.
By doing so, one standard item name can be set for a plurality of item names similar to each other described in the form. As a result, it is possible to reduce the labor of combining a plurality of similar item names into one item name.

The standard item name setting device includes a learning model storage unit that stores a learning model in which words appearing in one or more forms as learning data are machine-learned, and the feature vector generation unit decomposes the item names. It is preferable to synthesize the vectors of each word based on the learning model to generate the feature vector of the item name.
By doing so, it is possible to improve the accuracy of classifying similar item names together.

The standard item name setting device includes a matching pattern storage unit that stores matching patterns including at least one of a keyword and a regular expression in association with each other for each of the plurality of types, and the type determination unit is the attention cluster. It is preferable to determine the type of the attention cluster from the plurality of types based on the matching pattern matching the item names classified in.
By doing so, the accuracy of determining the type of cluster can be improved.

In the above-mentioned standard item name setting device, the item name candidate generation unit is from a character string other than the character string corresponding to the matching pattern stored in association with the type of the attention cluster among the item names classified into the attention cluster. It is preferable to set the common character string.
By doing so, it is possible to easily determine the criteria for classifying one cluster into one or more subclusters.

In the standard item name setting device, the standard item name setting unit sets the item name candidate having the highest frequency of appearance in the subcluster as the standard item name among the plurality of item name candidates for the subcluster. Then it is suitable.
By doing so, the standard item name can be set based on the most commonly used expression among the item names classified into the same subcluster.

In the above standard item name setting device, for each of the plurality of clusters, processing by the type determination unit, the item name candidate generation unit, and the standard item name setting unit is executed to obtain a plurality of standard item names. It is suitable to set.
By doing this, standard item names can be set for various item names in the form. This can reduce the labor of setting the standard item name from the form.

According to the present invention, standard item names can be set for corresponding item names in a plurality of forms.

It is a figure which shows the whole structure of an information processing system. It is a figure which shows the relationship of the item name in a plurality of forms. It is a figure explaining the flow of the setting process of a standard item name. It is a functional block diagram of a standard item name setting device. It is a figure explaining the structure of a neural network. It is a figure explaining the feature vector of the item name. It is a figure which shows an example of a matching pattern table. It is a figure which shows an example of the item name candidate table. It is a flow diagram of the clustering process of the item name. It is a flow diagram of the setting process of a standard item name.

Hereinafter, the information processing system 1 including the standard item name setting device 10 according to the embodiment of the present invention (hereinafter referred to as the present embodiment) will be described with reference to FIGS. 1 to 10.
It should be noted that the embodiments described below are merely examples for facilitating the understanding of the present invention, and do not limit the present invention. That is, the system configuration, data, processing, etc. described below can be changed or improved without departing from the spirit of the present invention, and the present invention includes equivalents thereof.

[Configuration of information processing system 1]
As shown in FIG. 1, the information processing system 1 includes a standard item name setting device 10 and a form processing device 30. The standard item name setting device 10 and the form processing device 30 are communicably connected to each other via a network such as the Internet or an intranet (not shown).

The form processing device 30 is connected to the scanner 40.
The scanner 40 is a device that captures image information by optically scanning a paper medium. In the present embodiment, the scanner 40 outputs the scanned image (image information) obtained by scanning the form P to the form processing device 30.
Form P is a standard document such as a book, a slip, or an application form. In the present embodiment, many types of forms P are captured by the scanner 40 and output to the form processing device 30.

The form processing device 30 is a computer that processes the form P captured by the scanner 40. Specifically, the form processing device 30 executes OCR (optical character recognition) on the form P to acquire the character string described in the form P. Further, the form processing device 30 analyzes the table structure of the form P based on the arrangement of the ruled lines and the character strings. More specifically, the form processing device 30 divides the form P into an item field, an input field, and a fill-in-the-blank input field, and analyzes the information of the item name described in the item field (further, the fill-in-the-blank input field). ..
The item field is an area in which a character string as an item name is described, and the input field is an area in which a character string is not described and information corresponding to the item field is input. The fill-in-the-blank input field is an area in which a character string is described and information is input between the character strings.

In the present embodiment, the information of the plurality of types of forms P analyzed by the form processing device 30 is output to the standard item name setting device 10. Then, the standard item name setting device 10 specifies the corresponding item name in the plurality of types of forms P, and sets the standard item name standardized by the corresponding item name.

Next, the configuration of the standard item name setting device 10 will be described.
As shown in FIG. 1, the standard item name setting device 10 is a computer including a processor 11, a storage device 12, and a communication interface 13 as hardware.

The processor 11 is configured to include, for example, a central processing unit, executes various arithmetic processes based on programs and data stored in the storage device 12, and controls each part of the standard item name setting device 10. do.

The storage device 12 includes, for example, a memory and a magnetic disk device, stores various programs and data, and also functions as a work memory of the processor 11.

The communication interface has a communication interface such as a network interface card (NIC) and connects to the network via the communication interface. Then, the communication interface communicates with a device such as the form processing device 30 via the network.

[Outline of processing executed by the standard item name setting device 10]
Here, the outline of the process executed by the standard item name setting device 10 will be described with reference to FIGS. 2 and 3.

FIG. 2 shows the correspondence of item names in a plurality of forms (first form PA, second form PB, third form PC). Here, it is assumed that the first form PA, the second form PB, and the third form PC are forms related to the same procedure, but the formats are different.
Therefore, it is desirable to treat the first item name 50A, the second item name 50B, and the third item name 50C as the same item, but the item names are different from "name", "name", and "name". It is necessary to standardize these item names.
Therefore, in the standard item name setting device 10 according to the present embodiment, the standard item name in which the item name of the form is standardized is set by the following procedure. In the following, the outline of the process will be described, and the details of the process will be described later.

As shown in FIG. 3, first, the standard item name setting device 10 acquires an item name list 60 including a list of item names described in a plurality of forms P from the form processing device 30.

Then, standard field name setting device 10 generates an item name included in the item name list 60 (item name _{I 1} ~ item name _{I N)} feature vectors for each of the (feature vector _V 1 ~V _N). In the above, N (an integer of 2 or more) represents the number of item names.

Then, standard field name setting device 10, and clustering feature vector (feature vector _V 1 ~V _N), classified into a plurality of clusters (clusters _C 1 _{~C M).} Here, the number of clusters is M (integer of 2 or more).

Then, standard field name setting device 10 determines the respective types of a plurality of clusters (clusters _C 1 _{~C M).} A plurality of types such as "name (name)", "date (date)", and "addless (address)" are defined in advance, and the standard item name setting device 10 corresponds to each cluster. Determine the type.

Next, the standard item name setting device 10 classifies the item names classified into the clusters into subclusters (for example, C1a, C1b, C1c) according to the subdivision rule based on the cluster type.
For example, subclusters are generated for each modifier that modifies a type of noun.

Next, the standard item name setting device 10 generates an item name candidate for each noun modified by the modifier of the subcluster. For example, when the modifier of the subcluster is "pregnant woman", the standard item name corresponding to the subcluster is "pregnant woman [name]". Here, if there are "name", "name", and "name" as candidates for [name] in the item name of the subcluster, the one with the highest frequency of appearance (for example, "name") is standard. Selected as the item name.
The standard item name setting device 10 executes the above processing for each cluster and each sub-cluster, and sets the standard item name.

[Functions provided in the standard item name setting device 10]
Hereinafter, the functions provided in the standard item name setting device 10 in order to realize the processing described above will be described.

FIG. 6 shows a functional block diagram of the standard item name setting device 10. As shown in FIG. 6, the standard item name setting device 10 has, as functions, a learning model storage unit 20, an item name acquisition unit 21, a feature vector generation unit 22, a clustering unit 23, a matching pattern storage unit 24, and a type determination unit. 25, an item name candidate generation unit 26, and a standard item name setting unit 27 are provided.

The functions of the above-mentioned parts provided in the standard item name setting device 10 are executed by the processor 11 operating each part of the standard item name setting device 10 according to the program (standard item name setting program) stored in the storage device 12. NS. The above program may be acquired by the standard item name setting device 10 via a communication network such as a network using a communication interface, or may be acquired by the standard item name setting device 10 read from a storage medium in which the program is stored. You may do it.
Further, the standard item name setting method according to the present invention is realized by operating the processor 11 according to the above standard item name setting program.
Hereinafter, the details of the functions of the above parts will be described.

[Explanation of learning model storage unit 20]
The learning model storage unit 20 stores a learning model in which words appearing in one or more forms as learning data are machine-learned.

The "learning data" is a data set used to make a learning model perform machine learning. In other words, "learning data" is a set of sample data used to make a learning model perform machine learning.
For example, the analysis data of the form P corresponds to an example of "learning data". Specifically, the above analysis data includes an item name obtained from the form P by optical character recognition and data obtained by decomposing the item name into words by morphological analysis.

The "word that appears in the form" is a word described in the form as learning data. In other words, it is a word that constitutes an item name obtained by optical character recognition from a form as learning data. A "word" is the smallest linguistic unit that has meaning and function.

"Machine learning" means inputting a certain number of sample data sets, performing analysis, extracting useful rules, rules, knowledge representations, judgment criteria, etc. from the data, and developing an algorithm.
In this embodiment, the method of unsupervised learning is used for "machine learning".

The "learning model" is a mathematical model that is the target of machine learning using learning data. For example, a neural network corresponds to an example of the above "learning model". Neural networks include two-layer perceptrons, three-layer hierarchical neural networks, and four-layer or higher multi-layer neural networks (deep neural networks).

FIG. 5 shows an example of the neural network 70 as a learning model. As shown in FIG. 5, the neural network 70 has an input layer 71, a hidden layer 72, and an output layer 73.

The input layer 71 is an input node whose dimension (number of elements) is the number of words extracted from the form as learning data.
The hidden layer 72 is a network configured as one layer or a plurality of layers that relays the input layer 71 and the output layer 73.
The output layer 73 is an output node whose dimension (number of elements) is the number of words extracted from the form as learning data.

Then, in machine learning using the learning data, a weight matrix from the input layer 71 to the hidden layer 72 and a weight matrix from the hidden layer 72 to the output layer 73 are learned.

The learning model storage unit 20 is mainly realized by the storage device 12 of the standard item name setting device 10.
Specifically, the storage device 12 includes a configuration of an input layer 71, a hidden layer 72, and an output layer 73 constituting the neural network 70, a weight matrix from the input layer 71 to the hidden layer 72, and a hidden layer 72 to the output layer. The data of the weight matrix to 73 is stored as the data of the learning model.
The machine learning process may be executed by the standard item name setting device 10, and the standard item name setting device 10 may use the data of the learning model obtained as a result of machine learning executed by another computer. It may be acquired via the communication interface 13.

[Explanation of item name acquisition unit 21]
The item name acquisition unit 21 acquires a plurality of item names described in a plurality of forms.

The "plurality of forms" is a form for which a standard item name is set. That is, a plurality of types of forms having different formats correspond to the above-mentioned "plurality of forms". Further, the above-mentioned "plurality of forms" may match the forms used as learning data.

The "plurality of item names" is an item name described in the "plurality of forms" to be processed. For example, all item names included in the "plurality of forms" to be processed correspond to the above "plurality of item names".

The item name acquisition unit 21 is mainly realized by the processor 11, the storage device 12, and the communication interface 13 of the standard item name setting device 10.
Specifically, the processor 11 acquires the analysis results of a plurality of forms to be processed from the form processing device 30 via the communication interface 13. Here, the analysis result of the plurality of forms includes the character string data of one or more item names obtained from the form by optical character recognition.

[Explanation of feature vector generation unit 22]
The feature vector generation unit 22 generates feature vectors for each of the plurality of item names.

The "feature vector" represents the feature of the item name as a vector. For example, the "feature vector" is a vector whose dimension number is the number of all words included in the form used as learning data. The feature vector of the item name is a vector obtained by synthesizing the vectors of the words constituting the item name.

Specifically, the feature vector generation unit 22 generates a feature vector of the item name by synthesizing a vector based on the learning model of each word obtained by decomposing the item name.
In the learning model, as a result of machine learning, vector information of each word is stored. Specifically, the word vector information is included in the weight matrix from the input layer 71 to the hidden layer 72. Then, the feature vector generation unit 22 acquires the vector of each word constituting the item name from the learning model, and generates the feature vector of the item name.

Here, the process of generating the feature vector of the item name will be described with reference to FIG.
In FIG. 6A, an example of the item name (“address before change of beneficiary”) is shown. On the other hand, the feature vector generation unit 22 performs morphological analysis on the item name and decomposes the item name into words as shown in FIG. 6 (b). In this example, the "recipient's address before change" is decomposed into "recipient", "before change", "address", and "no".

Here, the word vectors (v1 to v4) of "recipient", "before change", "address", and "no" are obtained from the weight matrix of the learning model stored in the learning model storage unit 20. ..
Then, by synthesizing the word vectors (v1 to v4) of "recipient", "before change", "address", and "no", the feature vector of "address before change of beneficiary" is generated. Will be done.

The feature vector generation unit 22 is mainly realized by the processor 11 and the storage device 12 of the standard item name setting device 10.
Specifically, the processor 11 generates a feature vector for each item name acquired by the item name acquisition unit 21 as follows. First, the processor 11 performs morphological analysis on the item name and decomposes the item name into words. Next, the processor 11 generates a feature vector of the item name by synthesizing the vector of the words constituting the item name based on the learning model stored in the learning model storage unit 20.

[Explanation of clustering unit 23]
The clustering unit 23 classifies a plurality of item names into one of a plurality of clusters.

The "cluster" is a classification to which a plurality of item names are distributed. In other words, it corresponds to each group in the case where items having similar feature vectors among a plurality of item names are grouped together.

The clustering unit 23 classifies the plurality of item names into a plurality of clusters based on the similarity of the feature vectors of the plurality of item names.

The "similarity of feature vectors" is an index showing the similarity between two feature vectors. Specifically, the cosine similarity between the two feature vectors corresponds to an example of the above-mentioned "similarity of feature vectors". In addition, Pearson's correlation coefficient, deviation pattern similarity, and the like correspond to an example of the above-mentioned "characteristic vector similarity".

The clustering unit 23 is mainly realized by the processor 11 and the storage device 12 of the standard item name setting device 10.
Specifically, the processor 11 calculates the similarity of the feature vectors of a plurality of item names generated by the feature vector generation unit 22, and based on the calculated similarity, a plurality of clusters of the feature vectors of the plurality of item names. Classify into.
For clustering, known methods such as the shortest distance method, the longest distance method, the group average method, and Ward's method can be used.

[Explanation of matching pattern storage unit 24]
The matching pattern storage unit 24 stores matching patterns including at least one of a keyword and a regular expression in association with each other for each of a plurality of types.

A "type" is a characteristic of a cluster. In other words, the "type" is data indicating the characteristics of the item names belonging to the cluster. More specifically, the type of the cluster is determined based on which of the matching patterns defined for each of the plurality of types the item name belonging to one cluster satisfies.

The "keyword" is a character string specified in advance. The character string may consist of one word or a plurality of words.
"Regular expression" is one of the methods of expressing a set of character strings with one character string. Specifically, the "regular expression" is represented by a combination of a character and a symbol string in which conditions such as wildcards are defined.

The "matching pattern" is a matching condition for a character string defined according to the type of cluster. The "matching pattern" may include one or more regular expressions and one or more keywords.

The matching pattern storage unit 24 is mainly realized by the storage device 12 of the standard item name setting device 10. Specifically, the storage device 12 of the standard item name setting device 10 stores the matching pattern table T1 in which the matching pattern is stored.

FIG. 7 shows an example of the matching pattern table T1. As shown in FIG. 7, the matching pattern table T1 stores an item type (type), a keyword, and a regular expression in association with each other. For example, the "item type" includes "name (name)", "date (date)", "address (address)", and the like, and a keyword and a regular expression are associated with each item type.

Specifically, when the item type is "name", for example, "name", "furigana", "name" and the like may be associated as keywords and regular expressions.
When the item type is "date", for example, "* year * month * day", "* month * day", etc. may be associated as a keyword or a regular expression. Here, "*" is a wild card (arbitrary character string).
When the item type is "addless", for example, "address", "location", "* destination", etc. may be associated as keywords or regular expressions.

Further, the matching pattern storage unit 24 may store either a regular expression or a keyword for the item type.

[Explanation of type determination unit 25]
The type determination unit 25 determines the type of the attention cluster from the plurality of types based on the item names classified into the attention cluster among the plurality of clusters.

The “cluster of interest” is one cluster to be processed, which is selected from a plurality of clusters. In other words, a "cluster of interest" is any cluster selected from a plurality of clusters. That is, all of the plurality of clusters can be "attention clusters". For example, by sequentially selecting a cluster of interest from a plurality of clusters, it is possible to process each of the plurality of clusters as a cluster of interest.

The "plurality of types" is a plurality of predetermined types. Specifically, "name", "date", "addless" and the like correspond to the above "plurality of types".

The type determination unit 25 determines the type of the attention cluster from the plurality of types based on the matching pattern matching the item name classified into the attention cluster.

The "matching pattern that matches the item name" is a matching pattern that satisfies the item name classified into the cluster of interest among the plurality of matching patterns associated with each of the plurality of types.
For example, when the matching pattern contains a plurality of keywords and regular expressions, the predetermined number of item names included in the cluster of interest satisfies any of the keywords and regular expressions included in the above matching pattern. , It is determined that the item name belonging to the cluster of interest satisfies the above matching pattern.

The type determination unit 25 is mainly realized by the processor 11 and the storage device 12 of the standard item name setting device 10.
Specifically, the processor 11 sets one cluster out of a plurality of clusters as a cluster of interest, and executes the following processing.
First, the processor 11 searches for a matching pattern associated with the item type stored in the matching pattern table T1 stored in the matching pattern storage unit 24, whose matching pattern matches the item name classified in the cluster of interest. do. Then, the processor 11 sets the item type corresponding to the searched matching pattern as the item type of the attention cluster.

[Explanation of item name candidate generation unit 26]
The item name candidate generation unit 26 subdivides the item names classified into the attention cluster into subclusters consisting of item names having a common character string based on the rules corresponding to the types determined by the type determination unit. For item names belonging to the subcluster, a plurality of item name candidates based on character strings other than the common character string are generated.

The "rule corresponding to the type" is a subdivision rule of the cluster of interest defined in association with the type. For example, the above subdivision rule may be a rule that subdivides the attention cluster for each character string (for example, a modifier) other than the character string corresponding to the keyword or regular expression defined in the matching pattern.
Specifically, in the item names classified into the attention cluster of type "name", the modifier "1" (* of the applicant) is used as a character string other than the keywords such as "name", "furigana", and "name". , The modifier "2" (* for pregnant women) and the modifier "3" (* before the change of the beneficiary) are extracted. In this case, the item name candidate generation unit 26 subdivides the attention cluster into subclusters "1" to subclusters "3" having item names including the modifiers "1" to "3".

The "common character string" is a character string other than the character string representing the type of the attention cluster in the item name classified into the attention cluster, and is a character string common to a plurality of item names.
For example, the modifier in the item name classified into the cluster corresponds to the "common character string".

The "sub-cluster" is a cluster of interest subdivided by the above subdivision rule. That is, each of the subdivided sets of the clusters of interest is the above-mentioned "subcluster".

The "item name belonging to the subcluster" is an item name assigned to the subcluster. That is, the element of the subcluster as a set becomes the above-mentioned "item name belonging to the subcluster".

The "character string other than the common character string" is a character string other than the character string (for example, a modifier) used when subdividing the subcluster in the item name assigned to the subcluster.
For example, the modified word modified by the modifier corresponds to the above-mentioned "character string other than the common character string".

The "item name candidate" is the first when the item names classified into the subclusters are configured as the first character string (common character string) and the second character string (character string other than the common character string). It is a candidate for a combination of the first character string and the second character string. That is, if there is an item name in which the first character string is common and the second character string is different among the item names classified into the subclusters, these combinations are the above-mentioned "item name candidates". The variation of the second character string may be the above-mentioned "item name candidate".

The item name candidate generation unit 26 sets a common character string from the item names classified into the attention clusters other than the character strings corresponding to the matching pattern stored in association with the type of the attention cluster.

The "character string corresponding to the matching pattern stored in association with the type of the attention cluster" is a character string to which the matching pattern corresponding to the type of the attention cluster is satisfied. That is, when the matching pattern includes a plurality of keywords, those plurality of keywords correspond to the above-mentioned "character string corresponding to the matching pattern stored in association with the type of the cluster of interest".

The item name candidate generation unit 26 is mainly realized by the processor 11 and the storage device 12 of the standard item name setting device 10.
Specifically, the processor 11 groups the item names included in the attention cluster into subclusters according to modifiers other than the character string corresponding to the matching pattern according to the type of the attention cluster. That is, the processor 11 divides the item names included in the cluster of interest into subclusters for each modifier.
Next, the processor 11 extracts a character string different from the item name divided into the subclusters other than the modifier corresponding to the subcluster, and generates an item name candidate based on the extracted character string.

[Explanation of standard item name setting unit 27]
The standard item name setting unit 27 sets an item name candidate selected based on a predetermined criterion from a plurality of item name candidates of the subcluster as a standard item name corresponding to the subcluster.

The "predetermined standard" is an evaluation standard for a plurality of item name candidates of a subcluster. For example, the above evaluation criteria may be the frequency of occurrence in subclusters. That is, the standard item name setting unit 27 sets the item name candidate having the highest frequency of appearance in the subcluster as the standard item name corresponding to the subcluster among the plurality of item name candidates of the subcluster.

The "standard item name corresponding to the subcluster" is a standard item name derived from the subcluster. That is, the item name that combines the first character string (common character string) corresponding to the subcluster and the second character string corresponding to the subcluster is the standard item name corresponding to the subcluster.
Here, the standard item name is an item name used as standard for an item name corresponding to the meaning and use in different types of forms.

The standard item name setting unit 27 sets the item name candidate having the highest frequency of appearance in the subcluster as the standard item name among the plurality of item name candidates for the subcluster.

The "occurrence frequency" is the number of item names including the character string of the item name candidate among the item names belonging to the subcluster.

The standard item name setting unit 27 is mainly realized by the processor 11, the storage device 12, and the communication interface 13 of the standard item name setting device 10.
Specifically, the processor 11 counts the appearance frequency of each of the item name candidates generated by the item name candidate generation unit 26 for the subcluster, and sets the item name candidate having the highest appearance frequency as the standard item name. Hereinafter, a specific example of the standard item name setting process by the processor 11 will be described with reference to FIG.

FIG. 8 shows an example of the item name candidate table T2 generated by the processor 11. As shown in FIG. 8, the processor 11 counts the appearance frequency of each of the item name candidates (here, item name candidate “1” to item name candidate “3”).
Then, the processor 11 sets the item name candidate "1" having the highest appearance frequency as the standard item name. In the example shown in FIG. 8, the standard item flag is set to true (T) for the item name candidate “1”, and the standard item flag is set to false (F) for the other candidates.
The processor 11 may also set the item name candidates "3" and "2" as the first candidate and the second candidate that substitute for the standard item name.

Further, the standard item name setting device 10 executes processing by the type determination unit 25, the item name candidate generation unit 26, and the standard item name setting unit 27 for each of the plurality of clusters, and obtains a plurality of standard item names. Set.
That is, the standard item name setting device 10 sequentially sets one of the plurality of clusters to the attention cluster, and executes the standard item name setting process based on the set attention cluster.

[Process flow by standard item name setting device 10]
Next, the flow of processing executed by the standard item name setting device 10 will be described with reference to FIGS. 9 and 10.

[Item name clustering process]
First, the flow of the item name clustering process will be described with reference to the flow chart shown in FIG. In the following, the standard item name setting device 10, and it is getting an item name list 60 containing items name I _{1 ~} item name I _N from the form processing unit 30. In addition, N is the number of item names.

As shown in FIG. 9, the standard item name setting device 10 initializes the variable i to 1 (S1) _{and decomposes the item name I i} into words by morphological analysis (S2).
The standard field names set device 10 combines the vector of the words constituting the item name I _i, and generates a feature vector V _i of the item name I _i (S3).

Here, when the variable i has not reached N (S4: No), the standard item name setting device 10 adds 1 to the variable i (S5) and returns to S2.

On the other hand, if the variable i has reached N (S4: Yes), the standard item name setting device 10 calculates the similarity of the feature vector _{V 1} ~ feature vector _{V N} (S6).
Then, the standard field name setting device 10 based on the calculated similarity, clustering the feature vector _{V 1} ~ feature vector _{V N} to the cluster _C 1 _{~C M} (S7). Here, M is the number of clusters.
By the above processing, it can be classified based on the collection of mean item name I _{1 ~} item name I _N.

[Standard item name setting process]
Next, the flow of the standard item name setting process will be described with reference to the flow chart shown in FIG. The process described below is a process performed after the clustering process described above.

As shown in FIG. 10, the standard item name setting device 10 initializes the variable j and the variable l to 1 (S11), _{and sets the cluster C j} to the cluster of interest (S12).

Then, the standard item name setting device 10, to determine the item type of cluster _{C j} is the cluster of interest (S13). The _{item type determination process of the cluster Cj} is executed by the type determination unit 25.

Next, the standard item name setting device 10 _{subdivides the cluster C j into subclusters C j1 to} C jL according to the item type (S14). Here, L is the number of subclusters in the cluster C _j.

Next, the standard item name setting device 10 generates an item name candidate based on the item name classified into _{the subcluster C jl (S15).} Then, the standard item name setting device 10 counts the appearance frequency of the item name candidates (S16), and sets the item name candidate having the highest appearance frequency as the standard item name (S17).

Here, when the variable l has not reached L (S18: No), the standard item name setting device 10 adds 1 to the variable l (S19) and returns to S15.

On the other hand, when the variable l has reached L (S18: Yes), the standard item name setting device 10 further determines whether or not the variable j has reached M (S20).
Here, when the variable j has not reached M (S20: No), the standard item name setting device 10 adds 1 to the variable j (S21) and returns to S12.

On the other hand, when the variable j reaches M (S20: Yes), the standard item name setting device 10 outputs the data (standard item data) of the standard item name set in the above processing (S22). , End the process.
In S22, the standard item name setting device 10 may transmit, for example, data summarizing the information of the standard item name and its alternative candidate to the form processing device 30 as the above standard item data.

[summary]
The standard item name setting device 10 includes an item name acquisition unit 21 for acquiring a plurality of item names described in a plurality of forms, a clustering unit 23 for classifying a plurality of item names into any of a plurality of clusters, and a plurality of clusters. Based on the item name classified into the attention cluster among the clusters, the type determination unit 25 that determines the type of the attention cluster from among a plurality of types, and the rule corresponding to the type determined by the type determination unit 25. Then, the item names classified into the cluster of interest are subdivided into subclusters consisting of item names having a common character string, and character strings other than the common character string for the item names belonging to the subcluster (for example, modifiers). Item name candidate generation unit 26 that generates multiple item name candidates based on The standard item name setting unit 27, which is set as, is provided.

According to the standard item name setting device 10, one standard item name can be set for the corresponding item name in a plurality of forms. As a result, it is possible to reduce the labor of combining a plurality of corresponding item names into one item name.

The standard item name setting device 10 includes a feature vector generation unit 22 that generates each feature vector of a plurality of item names, and a clustering unit 23 is a plurality of clustering units 23 based on the similarity of each feature vector of the plurality of item names. Classify item names into multiple clusters.
By doing so, one standard item name can be set for a plurality of item names similar to each other described in the form. As a result, it is possible to reduce the labor of combining a plurality of similar item names into one item name.

The standard item name setting device 10 includes a learning model storage unit 20 that stores a learning model that machine-learns words appearing in one or more forms as learning data, and a feature vector generation unit 22 decomposes each item name. A feature vector of item names is generated by synthesizing vectors based on a word learning model.
By doing so, it is possible to improve the accuracy of classifying similar item names together.

The standard item name setting device 10 includes a matching pattern storage unit 24 that stores matching patterns including at least one of a keyword and a regular expression in association with each other for each of a plurality of types, and the type determination unit 25 is classified into a cluster of interest. The type of attention cluster is determined from among multiple types based on the matching pattern that matches the item name.
By doing so, the accuracy of determining the type of cluster can be improved.

In the standard item name setting device 10, the item name candidate generation unit 26 uses common characters other than the character strings corresponding to the matching pattern stored in association with the type of the attention cluster among the item names classified in the attention cluster. Set the column.
By doing so, it is possible to easily determine the criteria for classifying one cluster into one or more subclusters.

In the standard item name setting device 10, the standard item name setting unit 27 sets the item name candidate having the highest frequency of appearance in the subcluster as the standard item name among the plurality of item name candidates for the subcluster.
By doing so, the standard item name can be set based on the most commonly used expression among the item names classified into the same subcluster.

The standard item name setting device 10 sets a plurality of standard item names by executing processing by the type determination unit 25, the item name candidate generation unit 26, and the standard item name setting unit 27 for each of the plurality of clusters. ..
By doing this, standard item names can be set for various item names in the form. This can reduce the labor of setting the standard item name from the form.

[Other embodiments]
The present invention is not limited to the above embodiment.
The standard item name setting device 10 and the form processing device 30 may be configured as one device.
Further, the standard item name setting device 10 is not limited to one computer, and may be composed of a plurality of computers.

Further, the setting of the keyword for the type is not limited to the above example. For example, in the matching pattern storage unit 24, keyword_and may be defined in addition to keyword for the type. In this case, when the keyword included in the item name exists in both keyword_and and keyword, it is determined that the types corresponding to keyword_and and keyword match.

Further, keyword_common may be used instead of keyword. This is a function that defines a set of some keywords in advance as information common to all procedures and refers to it in the definition for each procedure.
For example, keyword_common: child, and the common definition is child. When keyword: [child, child, child, child], keyword_common: child is synonymous with keyword: [child, child, child, child].
In this way, various methods can be used to specify the keyword.

1 Information processing system 10 Standard item name setting device 11 Processor 12 Storage device 13 Communication interface 20 Learning model storage unit 21 Item name acquisition unit 22 Feature vector generation unit 23 Clustering unit 24 Matching pattern storage unit 25 Type determination unit 26 Item name candidate Generation unit 27 Standard item name setting unit 30 Form processing device 40 Scanner 50A 1st item name 50B 2nd item name 50C 3rd item name 60 Item name list 70 Neural network (learning model)
71 Input layer 72 Hidden layer 73 Output layer A1 Item name candidate A2 Item name candidate AL Item name candidate C1 cluster C2 cluster CM cluster P form PA 1st form PB 2nd form PC 3rd form RI standard item name T1 Matching pattern table T2 Item name candidate table

Claims (9)

Item name acquisition unit that acquires multiple item names described in multiple forms, and
A clustering unit that classifies the plurality of item names into one of a plurality of clusters,
A type determination unit that determines the type of the attention cluster from among the plurality of types based on the item names classified into the attention cluster among the plurality of clusters.
Based on the rules corresponding to the types determined by the type determination unit, the item names classified into the attention cluster are subdivided into subclusters consisting of item names having a common character string, and items belonging to the subcluster. About the name The item name candidate generation unit that generates multiple item name candidates based on character strings other than the common character string, and
A standard item name setting unit that sets an item name candidate selected based on a predetermined criterion from the plurality of item name candidates of the subcluster as a standard item name corresponding to the subcluster.
A standard item name setting device characterized by being equipped with. It is provided with a feature vector generation unit that generates a feature vector for each of the plurality of item names.
The standard item name setting according to claim 1, wherein the clustering unit classifies the plurality of item names into the plurality of clusters based on the similarity of the feature vectors of the plurality of item names. Device. It is equipped with a learning model storage unit that stores a learning model that machine-learns words that appear in one or more forms as learning data.
The standard item name according to claim 2, wherein the feature vector generation unit synthesizes a vector based on the learning model of each word obtained by decomposing the item name to generate a feature vector of the item name. Setting device. Each of the plurality of types is provided with a matching pattern storage unit that stores matching patterns including at least one of a keyword and a regular expression in association with each other.
The type determination unit according to claim 1 to 3, wherein the type determination unit determines the type of the attention cluster from the plurality of types based on a matching pattern matching the item name classified into the attention cluster. The standard item name setting device described in any of them. The item name candidate generation unit sets the common character string from the item names classified into the attention clusters other than the character strings corresponding to the matching pattern stored in association with the type of the attention cluster. The standard item name setting device according to claim 4. The standard item name setting unit is characterized in that, among the plurality of item name candidates for the subcluster, the item name candidate having the highest frequency of appearance in the subcluster is set as the standard item name. The standard item name setting device according to any one of 5 to 5. A claim characterized in that processing is executed by the type determination unit, the item name candidate generation unit, and the standard item name setting unit for each of the plurality of clusters to set a plurality of standard item names. The standard item name setting device according to any one of 1 to 6. The computer
Get multiple item names listed in multiple forms,
Classify the multiple item names into one of multiple clusters and classify them into one of multiple clusters.
Based on the item name classified into the attention cluster among the plurality of clusters, the type of the attention cluster is determined from the plurality of types.
Based on the rule corresponding to the determined type, the item names classified into the attention cluster are subdivided into subclusters consisting of item names having a common character string, and the item names belonging to the subcluster are the same. Generate multiple item name candidates based on character strings other than the character string of
A standard item name setting method, characterized in that an item name candidate selected from the plurality of item name candidates of the subcluster based on a predetermined criterion is set as a standard item name corresponding to the subcluster. Item name acquisition unit that acquires multiple item names described in multiple forms, and
A clustering unit that classifies the plurality of item names into one of a plurality of clusters,
A type determination unit that determines the type of the attention cluster from among the plurality of types based on the item names classified into the attention cluster among the plurality of clusters.
Based on the rules corresponding to the types determined by the type determination unit, the item names classified into the attention cluster are subdivided into subclusters consisting of item names having a common character string, and items belonging to the subcluster. About the name The item name candidate generation unit that generates multiple item name candidates based on character strings other than the common character string, and
To make the computer function as a standard item name setting unit that sets an item name candidate selected based on a predetermined criterion from the plurality of item name candidates of the subcluster as a standard item name corresponding to the subcluster. Standard item name setting program.

Images