JP7291347B2 - Drawing retrieval device, model generation device, drawing retrieval method, and model generation method - Google Patents

Description

The present invention relates to a drawing retrieval device.

For example, in the field of plant engineering, it is necessary to handle a large number of drawings (eg past drawings). Therefore, there is a demand for a technique for efficiently retrieving a desired drawing from a large number of drawings.

As an example, Patent Literature 1 discloses a technique related to a user interface of a drawing search system (drawing management system). The drawing search system of Japanese Patent Laid-Open No. 2002-200000 accepts a user's selection operation for each of a device name and a drawing type name, and searches for a group of drawings corresponding to the selection operation.

JP-A-2003-99484

According to the drawing search system of Patent Document 1, it is possible to present search results (a group of drawings) according to the selection operation without requiring the user to input a search word. However, as will be described later, the drawing search system of Patent Document 1 quickly finds a desired drawing corresponding to the target drawing (eg, a specific drawing considered to be highly related to the target drawing) from the search results. is not always easy for users.

As described above, there is still room for improvement in the conventional drawing retrieval technology in terms of ingenuity for enhancing the user's convenience. An object of one aspect of the present invention is to improve the user's convenience in drawing search more than before.

In order to solve the above problems, a drawing retrieval device according to one aspect of the present invention is a drawing retrieval device for retrieving at least one drawing corresponding to a target drawing from a plurality of retrieval target drawings, wherein the plurality of retrieval The target drawings are pre-sorted into P groups (P is an integer of 2 or more) from the first group to the P-th group based on search target drawing content parameters related to the description content of each of the plurality of search target drawings. P learning models corresponding to each of the P groups are generated in advance based on the search target drawing content parameters of the plurality of search target drawings belonging to the P groups. a target drawing content parameter acquiring unit for acquiring target drawing content parameters related to the description content of the target drawing by analyzing the target drawing; (i) a specific learning model, which is a learning model corresponding to the specific group among the P learning models; (ii) a search unit that receives the target drawing content parameter and searches for the at least one drawing based on the output of the specific learning model corresponding to the target drawing content parameter.

In order to solve the above problems, a model generation apparatus according to an aspect of the present invention provides a model generation apparatus for generating a learning model for retrieving at least one drawing corresponding to a target drawing from a plurality of search target drawings. a search target drawing content parameter obtaining unit for obtaining a search target drawing content parameter related to the description content of each of the plurality of search target drawings by analyzing the plurality of search target drawings; A group for classifying the plurality of drawings to be searched into P groups (where P is an integer of 2 or more) from a first group to a P-th group based on the search target drawing content parameter of each of the plurality of search target drawings. Learning to generate P learning models corresponding to each of the P groups based on the classifying unit and the search target drawing content parameters of the plurality of search target drawings belonging to the P groups, respectively. and

In order to solve the above problems, a drawing retrieval method according to an aspect of the present invention is a drawing retrieval method for retrieving at least one drawing corresponding to a target drawing from a plurality of retrieval target drawings, The search target drawings are divided into P groups (P is an integer of 2 or more) from the first group to the P group based on the search target drawing content parameters related to the description content of each of the plurality of search target drawings. Based on the search target drawing content parameters of the plurality of search target drawings classified in advance and belonging to each of the P groups, P learning models corresponding to each of the P groups are prepared in advance. a target drawing content parameter acquisition step of acquiring target drawing content parameters related to the description content of the target drawing by analyzing the target drawing generated and analyzing the target drawing; (i) identifying a learning model corresponding to the specific group from among the P learning models; inputting the target drawing content parameters into a learning model; and (ii) retrieving the at least one drawing based on the output of the particular learning model responsive to the target drawing content parameters. there is

In order to solve the above problems, a model generation method according to an aspect of the present invention provides a model generation method for generating a learning model for retrieving at least one drawing corresponding to a target drawing from a plurality of search target drawings. a search target drawing content parameter acquiring step of acquiring search target drawing content parameters related to description content of each of the plurality of search target drawings by analyzing the plurality of search target drawings; A group for classifying the plurality of drawings to be searched into P groups (where P is an integer of 2 or more) from a first group to a P-th group based on the search target drawing content parameter of each of the plurality of search target drawings. Learning to generate P learning models corresponding to each of the P groups based on the classification step and the search target drawing content parameter of each of the plurality of search target drawings belonging to each of the P groups. and a step.

According to one aspect of the present invention, the user's convenience in drawing search can be improved more than before.

It is a block diagram which shows the structure of the principal part of the information processing system of a reference form. It is a figure which shows an example of the 1st reference table in a reference form. It is a figure which shows an example of the specific character string extraction process with respect to a past drawing. It is a figure which shows an example of the 2nd reference table in a reference form. It is a figure which shows an example of the 3rd reference table in a reference form. FIG. 10 is a diagram showing an example of specific character string extraction processing for a new drawing; It is a figure explaining the derivation example of the relevance score in a reference form. It is a figure which shows an example of the drawing search result in a reference form. 2 is a block diagram showing the configuration of the main part of the information processing system of Embodiment 1; FIG. 4 is a diagram showing an example of a first table according to Embodiment 1; FIG. FIG. 10 is a diagram showing an example of a classification result of each past drawing based on a past drawing content parameter in Embodiment 1; FIG. 10 is a diagram showing an example of a second table according to the first embodiment; FIG. FIG. 10 is a diagram showing an example of a third table according to the first embodiment; FIG. FIG. 10 is a diagram showing an example of a selection result of a specific group (closest group) in Embodiment 1; FIG. 10 is a diagram illustrating an example of deriving a relevance score according to the first embodiment; FIG. FIG. 10 is a diagram showing an example of a drawing search result according to the first embodiment; FIG.

[Reference form]
Before describing the information processing system 100 of the first embodiment, an information processing system 100s will be described as a reference form. For convenience of description, members having the same functions as members described in the reference embodiments are denoted by the same reference numerals in the following embodiments, and description thereof will not be repeated. Also, for the sake of simplification, the description of the same matters as those of known technology will be omitted as appropriate.

It should be noted that each numerical value mentioned herein below is merely an example. As used herein, the description "A to B" for two numbers A and B shall mean "greater than or equal to A and less than or equal to B" unless otherwise specified.

(Overview of information processing system 100s)
FIG. 1 is a block diagram showing the configuration of main parts of an information processing system 100s. The information processing system 100 s includes an information processing device 1 s, a past property drawing DB (Database) 91 s, and a new property drawing DB 92 .

The information processing device 1 s includes a control device 10 s, an input section 71 , a display section 72 and a storage section 80 . The control device 10s includes a learning device 11s and a drawing search device 12s. The "property" in this specification means, for example, a "site" in plant engineering.

The information processing device 1s may be connected to the past property drawing DB 91s and the new property drawing DB 92 so as to be communicable. Therefore, unlike the example of FIG. 1, at least one of the past property drawing DB 91s and the new property drawing DB 92 may be provided inside the information processing device 1s.

Drawings to be searched by the information processing device 1s (more specifically, the control device 10s) (search target drawings) include, for example, specification drawings, design drawings, and production drawings. Also, the drawings to be searched may include specifications, design documents, and quotations. In this way, drawings to be searched are not limited to documents containing the name "drawing". As an example, the drawings to be searched include any type of document relating to project planning in the field of plant engineering.

The control device 10s centrally controls each part of the information processing device 1s. The storage unit 80 stores various data and programs used for processing of the control device 10s. As described below, the control device 10s uses machine learning to select at least one drawing corresponding to a target drawing (eg, drawing ND) from a plurality of search target drawings (eg, past drawings a1 to MN). search for.

The input unit 71 receives a user's operation (user operation). The display unit 72 displays various data. As an example, the display unit 72 may display data indicating search results by the control device 10s. Note that the input unit 71 and the display unit 72 may be provided integrally. For example, by using a touch panel, the input section 71 and the display section 72 can be integrated.

(Past property drawing DB91s)
The past property drawing DB 91s stores each drawing (strictly speaking, drawing data) relating to each past property (each existing property). In the following description, "drawing data of drawing A (a certain drawing)" will be simply referred to as "drawing A" for short. Also, the "drawing number of drawing A" will be abbreviated as "drawing A" as appropriate.

In the reference form, each drawing about each of a plurality of M different properties is stored in the past property drawing DB 91s. M is an integer of 1 or more. Hereinafter, the j-th property is also referred to as "property j". j is an integer greater than or equal to 1 and less than or equal to M;

In addition, the past object drawing DB 91s stores N different drawings (N kinds) for each of the objects 1 to M. N is an integer of 1 or more. Hereinafter, the i-th drawing in property j is also referred to as "drawing (i, j)". In addition, the i-th (i-th type) drawing in each property is also generically referred to as drawing i. i is an integer greater than or equal to 1 and less than or equal to N;

As described above, a total of T drawings are stored in the past property drawing DB 91s. In the example of the reference form, T=M×N. Assume that T is an integer of 2 or more. That is, at least one of M and N shall be 1 or more.

Specifically, in the past property drawing DB 91s, the drawing number of each type is listed for each property number in the form of the first reference table TB1s shown in FIG. Hereinafter, the first reference table TB1s is also abbreviated as "TB1s". Other elements are similarly abbreviated as appropriate. The TB1s is also referred to as a "previous property/drawing correspondence table sorted by property". The i-th row and j-th column cell of TB1s indicates the drawing number of the drawing (i, j).

In the example of FIG. 2, properties 1 to 3 are also denoted as properties A to C, respectively, for convenience. In the example of FIG. 2, the first type of drawing (drawing 1) is an outline drawing, the second type of drawing (drawing 2) is an assembly drawing, and the third type of drawing (drawing 3) is a basic drawing. . The N-th type of drawing (drawing N) is a configuration diagram.

In the following, for the sake of simplicity, the drawings (1, 1) to (N, 1) in the example of FIG. For example, drawings a1 to aN in the example of FIG. Each drawing for other properties is also described in the same way.

As described above, the past object drawing DB 91s stores a total of T drawings from the drawing a1 to the drawing MN. Hereinafter, the drawings a1 to MN are also collectively referred to as past drawings. A past drawing is an example of a search target drawing. Therefore, the drawings a1 to MN are also referred to as a search target drawing group.

Further, in this specification, any one of the plurality of past drawings (search target drawings) is also referred to as a candidate drawing. As an example, each processing will be mainly illustrated when drawing a1 (outline drawing of object A) is set as a candidate drawing. The processing for other drawings is the same as for the drawing a1, although the description is omitted as appropriate.

(Learning device 11s)
The learning device 11s includes a past drawing data acquisition unit 111, a past drawing data preprocessing unit 112 (candidate drawing content parameter acquisition unit, search target drawing content parameter acquisition unit), and a learning model generation unit 113s. The learning device 11s generates a learning model for drawing retrieval by the drawing retrieval device 12s based on the drawings a1 to MN. For this reason, the learning device 11s may be called a model generation device. An example of the processing flow of the learning device 11s will be described below.

(Acquisition of candidate drawings)
In the past object drawing DB 91s, the drawings a1 to MN are pre-sorted by drawing type according to the TB1s. Therefore, for example, the past drawing data acquisition unit 111 acquires each outline drawing from the past object drawing DB 91s in the order of "drawing a1→b1→...→M1". Subsequently, the past drawing data acquisition unit 111 acquires each assembly drawing from the past object drawing DB 91s in the order of "drawing a2→b2→...→M2". Ultimately, the past drawing data acquisition unit 111 acquires each configuration drawing from the past property drawing DB 91s in the order of "drawing aN→bN→...→MN".

In the case of the above example, the past drawing data acquisition unit 111 first refers to the cell in the first row and first column of TB1s. Then, the past drawing data acquisition unit 111 acquires the drawing (1, 1) corresponding to the cell, that is, the drawing a1 from the past object drawing DB 91s. The past drawing data acquisition unit 111 supplies the acquired drawing a1 to the past drawing data preprocessing unit 112 .

(Generation of preprocessing data corresponding to candidate drawings)
The past drawing data preprocessing unit 112 generates preprocessed data corresponding to the drawing a1 by analyzing the drawing a1. As an example, the past drawing data preprocessing unit 112 performs OCR processing on the drawing a1 to generate data after the OCR processing. In the data after OCR processing, each character expressed as image data in drawing a1 is converted into character code.

Then, the past drawing data preprocessing unit 112 extracts (detects) a specific character string preset for each drawing from the OCR-processed data. Hereinafter, the specific character string is referred to as a specific character string. In the first embodiment, it is assumed that L (L is an integer equal to or greater than 2) different specific character strings are set in advance.

The k-th specific character string is hereinafter referred to as the k-th specific character string. k is an integer greater than or equal to 1 and less than or equal to L; In the following description, the first specific character string is set as "voltage value", the second specific character string is set as "current value", the third specific character string is set as "OR", and the Lth specific character string is set as "open". I will give an example if there is.

However, the specific character string is not limited to the above examples. The specific character string may be words that are generally used to describe product specification information (technical information) in drawings in the engineering field, for example.

FIG. 3 is a diagram showing an example of specific character string extraction processing for drawing a1. In the example of FIG. 3, the character string "voltage value 90 V" is written in the area L1 of the drawing a1. In this case, the past drawing data preprocessing unit 112 extracts a series of character strings "voltage value 90 V" of L1 from the OCR processed data. The past drawing data preprocessing unit 112 further extracts the first specific character string “voltage value” from the extracted series of character strings.

Subsequently, the past drawing data pre-processing unit 112 applies a known parsing technique to the series of character strings, so that the character string "90V" following the first specific character string "voltage value" is It is judged that the attribute of "voltage value" is indicated. In this case, the past drawing data preprocessing unit 112 acquires the number "90" included in the character string "90V" as a value indicating the magnitude of the "voltage value". In this way, the past drawing data preprocessing unit 112 determines, through syntax analysis, that "a series of character strings described in L1 indicates specification information that the voltage value is 90 V". .

In addition, a series of character strings "current value 1.0 A" is written in area L2 of drawing a1. The past drawing data preprocessing unit 112 extracts the second specific character string "current value" from the series of character strings in the same manner as in the example of L1. Then, the past drawing data preprocessing unit 112 determines that the character string "1.0A" following the second specific character string "current value" indicates the attribute of "current value". Therefore, the past drawing data preprocessing unit 112 acquires the number "1.0" included in the character string "1.0A" as a value indicating the magnitude of the "current value". In this way, the past drawing data preprocessing unit 112 determines that "a series of character strings described in L2 indicates specification information that the current value is 2.0 A".

By the way, although the third specific character string "OR" is described in drawing a1, no character string follows the third specific character string "OR". In this case, the past drawing data preprocessing unit 112 detects the number of appearances of the third specific character string “OR” included in the post-OCR data. As a result of the detection process, the past drawing data preprocessing unit 112 determines that the number of third specific character strings "OR" included in drawing a1 is three (see area L3).

Similarly, although the L-th specific character string "open" is described in drawing a1, no character follows the L-th specific character string "open". Therefore, the past drawing data preprocessing unit 112 determines that the number of L-th specific character strings “open” included in the drawing a1 is one (see area L4), as in the case of L3. ).

Then, the past drawing data preprocessing unit 112 generates preprocessing data corresponding to the drawing a1 based on the above-described analysis result of the OCR-processed data. The preprocessed data is a data set indicating the k-th content parameter (hereinafter referred to as Ak) associated with the k-th specific character string. The first to Lth content parameters are also generically referred to as content parameters.

It can be said that the content parameter is one of the data obtained by digitizing (quantifying) the description content of the drawing (specifically, the description content related to the specific character string). Therefore, the content parameter is used as an index indicating the description content of the drawing. To distinguish from the target drawing content parameter, which will be described later, the content parameter of the search target drawing (past drawing) is also referred to as the search target drawing content parameter. Also, the k-th content parameter of the drawing to be searched is also referred to as the k-th content parameter of the drawing to be searched.

Below, Ak of drawing (i, j) is also written as Ak(i, j). As described above, the past drawing data preprocessing unit 112 calculates Ak(i, j).

As an example, the past drawing data preprocessing unit 112 sets A1=90 for the drawing a1 (see also the hatched portions in FIG. 4 described later). That is, the past drawing data preprocessing unit 112 sets the value "90" indicating the magnitude of the first specific character string "voltage value" described in drawing a1 as the first content parameter.

Similarly, the past drawing data preprocessing unit 112 sets A2=1.0. That is, the past drawing data preprocessing unit 112 sets the value "1.0" indicating the magnitude of the second specific character string "current value" described in the drawing a1 as the second content parameter.

On the other hand, the past drawing data preprocessing unit 112 sets A3=3. That is, the past drawing data preprocessing unit 112 sets the value "3" indicating the number of times (the number of appearances) of the third specific character string "OR" in the drawing a1 as the third content parameter.

Similarly, the past drawing data preprocessing unit 112 sets AL=1. That is, the past drawing data preprocessing unit 112 sets the value "1" indicating the number of times the L-th specific character string "OPEN" is described in the drawing a1 as the L-th content parameter.

As described above, the past drawing data preprocessing unit 112 sets A1 to AL for drawing a1. As described above, the past drawing data preprocessing unit 112 acquires the content parameters of the candidate drawing by analyzing the candidate drawing. For this reason, the past drawing data preprocessing unit 112 is also called a candidate drawing content parameter acquisition unit.

Also, the past drawing data preprocessing unit 112 performs similar processing for other past drawings. That is, the past drawing data preprocessing unit 112 sets A1 to AL for each of the drawings a1 to MN.

After that, the past drawing data preprocessing unit 112 generates a second reference table TB2s indicating A1 to AL of drawings a1 to MN, that is, A1(1,1) to AL(M,N). As described below, TB2s is an example of teacher data in the reference form. That is, the past drawing data preprocessing unit 112 also operates as a functional unit (teaching data generating unit) that generates teaching data.

TB2s includes the j-th sub-table TB2s-j in the second reference table. TB2s-j is a table showing each A1-AL of drawing j. FIG. 4 shows TB2s-1 as an example. TB2s-1 shows A1 to AL of drawings a1 to M1 (each drawing in property 1). TB2s is composed of a set of N sub-tables, TB2s-1 to TB2s-N. Thus, in the reference form, a sub-table is created for each drawing type.

(Generation of learning model in reference form)
The learning model generation unit 113s generates a learning model corresponding to the drawings a1 to M1 based on the preprocessed data corresponding to the drawings a1 to M1 (that is, A1 to AL in the drawings a1 to M1). A known machine learning algorithm may be used to generate the learning model. As an example, in Reference Form 1, the learning model generation unit 113s generates a learning model by multinomial logistic regression analysis.

As described above, TB2s shows the correspondence between the number (example of identifier) of each of a plurality of past drawings (eg, drawing a1) and A1 to AL of each of the plurality of past drawings. Therefore, the learning model generation unit 113s acquires TB2s as teacher data. Then, the learning model generation unit 113s generates a learning model in the reference form by performing multinomial logistic regression analysis using the teacher data.

As an example, the learning model generator 113s calculates a k-th weighting factor (hereinafter referred to as Bk) associated with the k-th specific character string in the drawing a1 based on each of A1 to AL of TB2s-1. The first to Lth weighting factors are also generically referred to as weighting factors. Also, Bk in the drawing (i, j) is also written as Bk(i, j). Bk(i,j) is a weighting factor corresponding to Ak(i,j). The weighting factor is used to calculate a relevance score, which will be described later.

For example, the learning model generation unit 113s calculates B1 to BL in the drawing a1 by the multinomial logistic regression analysis described above. In the reference embodiment, each weighting factor is calculated as B1=−0.5, B2=−1.2, B3=0.6, . 5).

The learning model generation unit 113s performs similar processing on other past drawings. That is, the learning model generation unit 113s sets B1 to BL for each of the drawings a1 to MN. In this way, the learning model generation unit 113s generates a learning model including B1 to BL of each of the drawings a1 to MN, that is, B1(1,1) to BL(M,N).

Then, the learning model generation unit 113s generates a third reference table TB3s indicating B1(1,1) to BL(M,N). TB3s includes the j-th sub-table TB3s-j in the third reference table. TB3s-j is a table showing each of B1-BL of drawing j. FIG. 5 shows TB3s-1 as an example. TB3s-1 shows B1 to BL of drawings a1 to M1, respectively. Thus, TB3s-1 indicates a weighting factor corresponding to the content parameter of TB2s-1. TB3s is composed of a set of N sub-tables, TB3s-1 to TB3s-N.

(New property drawing DB92)
In the new property drawing DB 92, each drawing (hereinafter also collectively referred to as a new drawing) related to a new property (eg, at least one property that is scheduled to be constructed in the future) is stored as a new property data set. there is One new property (property T) included in the new property data set will be described below.

As an example, the new property drawing DB 92 stores, for property T, N different drawings (outline drawing to configuration drawing) of the same type as the past property. Below, the external view of the object T is illustrated as an example of the new drawing. Also, the outline drawing of the object T is called a drawing ND (see also FIG. 6 described later).

(Drawing retrieval device 12s)
The drawing search device 12s includes a new drawing data acquisition unit 121, a new drawing data preprocessing unit 122 (target drawing content parameter acquisition unit), a score calculation unit 123s, and a search result data generation unit 124s. 123 s of score calculation parts and 124 s of search result data generation parts may be generically called a search part (more specifically, the search part in a reference form).

The drawing search device 12s searches for at least one drawing by matching the target drawing with each of the plurality of search target drawings using the learning model generated by the learning device 11s. Drawing ND of a reference form is an example of a target drawing. As described below, the drawing retrieval device 12s uses the learning model to compare the drawing ND with each of the drawings a1 to MN.

(Obtaining new drawings)
The new drawing data acquisition unit 121 is a functional unit paired with the past drawing data acquisition unit 111 . As an example, when the input unit 71 receives a predetermined user operation, the new drawing data acquisition unit 121 acquires a predetermined new drawing (eg, drawing ND). The new drawing data acquisition unit 121 supplies the acquired drawing ND to the new drawing data preprocessing unit 122 .

(Generation of preprocessing data corresponding to new drawings)
The new drawing data preprocessing unit 122 is a functional unit paired with the past drawing data preprocessing unit 112 . The new drawing data preprocessing unit 122 generates preprocessed data corresponding to the drawing ND by performing the same processing as the past drawing data preprocessing unit 112 . That is, the new drawing data preprocessing unit 122 performs OCR processing on the drawing ND and extracts a specific character string from the drawing ND. Then, the new drawing data preprocessing unit 122 sets the content parameters of the drawing ND based on the extraction result of the specific character string. Hereinafter, the k-th content parameter of drawing ND is also referred to as Ck.

That is, the new drawing data preprocessing unit 122 acquires the content parameters of the drawing ND by analyzing the drawing ND using the same analysis method as the past drawing data preprocessing unit 112 . Note that the content parameter of the target drawing (drawing ND) is also referred to as the target drawing content parameter in order to distinguish it from the search target drawing content parameter. The k-th content parameter of the target drawing is also referred to as the k-th content parameter of the target drawing.

FIG. 6 is a diagram showing an example of specific character string extraction processing for the drawing ND. FIG. 6 is a diagram paired with FIG. In the example of FIG. 6, the character string "voltage value 50 V" is written in the area R1 of the drawing ND. Therefore, the new drawing data preprocessing unit 122 sets C1=50 based on the extraction result of the first specific character string (see also FIG. 7 described later).

By the way, the position of region R1 in drawing ND is different from the position of region L1 in drawing a1 (see FIGS. 3 and 6). However, in the drawing ND, although the description position is different from that in the drawing a1, the same kind of specification information (voltage value) as in the drawing a1 is described. According to the method of setting the first content parameter by the new drawing data preprocessing unit 122, the drawing ND and drawing a1 can be quantitatively evaluated. Regarding this point, the same applies to C2 to CL (in other words, regions R2 to R4) described below.

In addition, in the example of FIG. 6, the character string "current value 2.5 A" is written in the region R2 of the drawing ND. Therefore, the new drawing data preprocessing unit 122 sets C2=2.5 based on the extraction result of the second specific character string (see FIG. 6).

Further, the new drawing data preprocessing unit 122 determines that the number of third specific character strings "OR" included in the drawing ND is three (see region R3 in FIG. 6). Therefore, the new drawing data preprocessing unit 122 sets C3=3.

Similarly, the new drawing data preprocessing unit 122 determines that the number of L-th specific character strings “open” included in the drawing ND is two (see region R4 in FIG. 6). Therefore, the new drawing data preprocessing unit 122 sets CL=2.

As described above, the new drawing data preprocessing unit 122 sets C1 to CL by performing the same processing as the past drawing data preprocessing unit 112 on the drawing ND. After that, the new drawing data preprocessing unit 122 generates a new drawing content parameter table TB-ND showing C1 to CL (see FIG. 7).

(Calculation of relevance score in reference form)
The score calculator 123s calculates a relevance score based on the learning model generated by the learning model generator 113 (more specifically, B1(1,1) to BL(M,N)). The relevance score is a score (index) indicating the degree of relevance of each past drawing (drawings a1 to MN) to the drawing ND. Hereinafter, the relevance score of drawing (i, j) is also written as SCORE (i, j). SCORE (i, j) is also abbreviated as SCORE.

Below, FIG. 7 (in particular, the hatched portions in FIG. 7) will be referred to, and a case of calculating the relevance score will be exemplified. In the following example, for convenience of explanation, a case will be described in which the relevance score is calculated for each of the past drawings of the same type as the drawing ND (that is, drawings a1 to M1 that are outline drawings). However, it should be noted that the relevance score is also derived for the past drawing of a type different from the drawing ND in the same manner as in the example of FIG.

FIG. 7 is a diagram illustrating an example of deriving a relevance score. FIG. 7 mainly describes an example of calculating the SCORE of drawing a1. The score calculation unit 123s uses (i) C1 to CL derived by the new drawing data preprocessing unit 122 and (ii) B1 to BL of the drawing a1 derived by the learning model generation unit 113s, The 1st to Lth raw scores of the drawing a1 are calculated.

Hereinafter, the k-th elementary point is also written as SSk. Note that the k-th elementary point in the drawing (i, j) is also written as SSk(i, j). In addition, the 1st to L-th raw scores are also generically referred to as raw scores. SSk corresponds to Bk (the k-th weighting factor).

Specifically, the score calculator 123 calculates the product of the k-th content parameter (Ck) of the drawing ND and the k-th weighting factor (Bk) of the drawing a1 as SSk. That is, the score calculation unit 123s
SSk=Ck×Bk (1)
Calculate

In the example of FIG. 7, C1=50 and B1=-0.5. Therefore, the score calculator 123s derives SS1 as SS1=50×−0.5=−25. Similarly, the score calculator 123s calculates subsequent raw scores. For example, the score calculator 123s calculates each raw score with SS2=−3, SS3=1.8, . . . , SSL=1.4.

Subsequently, the score calculation unit 123 calculates the relevance score (SCORE) of the drawing a1 based on the 1st to Lth raw scores of the drawing a1. Specifically, the score calculator 123s calculates the sum of the raw scores from the first raw score to the L-th raw score as SCORE. That is, the score calculation unit 123s
SCORE=ΣSSk (2)
Calculate The range of k on the right hand side of equation (2) is from 1 to L. This point also applies to formula (2A) and the like, which will be described later.

As described above, the score calculator 123s calculates SCORE using each Ck and each Bk. Hereinafter, the relevance score of drawing a1 is also written as SCORE(a1). The relevance scores of other past drawings are similarly indicated. In the example of FIG. 7, SCORE(a1)=-30.4 is obtained as a result of calculation by the score calculator 123s.

Subsequently, the score calculator 123 calculates the SCORE of each of the remaining past drawings (drawings b1 to M1 in the example of FIG. 7) by the same method. In the example of FIG. 7, SCORE(b1)=41.1, SCORE(c1)=36.5, .

Also, in the example of FIG. 7, it was confirmed that SCORE (b1) has the highest relevance score among M SCOREs SCORE (a1) to SCORE (M1). Also, it was confirmed that SCORE(c1) was the second largest among the M SCOREs. Furthermore, it was confirmed that SCORE (a1) has the lowest relevance score among the M SCOREs.

The score calculator 123s generates a fourth reference table TB4s indicating its own calculation results. TB4s includes the j-th sub-table TB4s-j in the fourth reference table. TB4s-j is a table showing respective SS1-SSL and SCORE of drawing j. FIG. 7 shows TB4s-1 as an example. TB4s-1 shows SS1-SSL and SCORE of drawings a1-M1, respectively. Thus, TB4s-1 is generated based on TB-ND and TB3s-1. TB4s is composed of a set of M sub-tables, TB4s-1 to TB4s-M.

Note that the above formula (1) is
SSk(i, j)=Ck×Bk(i, j) (1A)
can also be written as Thus, TB3s-1 calculates SSk(i,j) based on Bk(i,j).

Also, the above formula (2) is
SCORE(i,j)=ΣSSk(i,j)
=Σ{Ck×Bk(i, j)} (2A)
can also be written as Thus, the score calculator 123s calculates SCORE(i,j) based on each SSk(i,j) (that is, each Ck and each Bk(i,j)).

Also, the above formula (2A) is
SCORE(i,j)=Σ{Ak(i,j)×Bk(i,j)} (3)
can also be understood as an equation in which Ck is substituted for Ak(i,j). In this way, the score calculator 123s can calculate SCORE, which is a dependent variable (objective variable), by applying C1 to CL as independent variables (explanatory variables) in the learning model.
The right side of Equation (3) is an example of a linear model (multinomial logistic regression analysis model) in multinomial logistic regression analysis.

Note that SSk can also be expressed as the output of a learning model according to Ck. Therefore, it can be said that the score calculation unit 123s calculates SCORE based on the outputs (SS1 to SSk) of the learning model.

Alternatively, the learning model may be configured to output SCORE. In this case, the score calculator 123s acquires the SCORE as the output of the learning model and supplies the SCORE to the search result data generator 124s.

(Determination of Candidate Order in Reference Form)
The search result data generator 124s determines the candidate order (candidate ranking) corresponding to the drawing ND based on SCORE(a1) to SCORE(M1) calculated by the score calculator 123s. Specifically, the search result data generation unit 124s determines the order of candidates (first candidate to M-th candidate) for each of the drawings a1 to M1 (outline drawings) based on SCORE (a1) to SCORE (M1). .

The search result data generator 124s sorts SCORE(a1) to SCORE(M1) according to the value of each SCORE. As an example, the search result data generator 124s sorts SCORE(a1) to SCORE(M1) in descending order of value. Then, the search result data generation unit 124s determines the past drawing corresponding to the g-th largest SCORE as the g-th candidate (hereinafter also referred to as candidate g). In the reference form, g is an integer of 1 or more and M or less.

If the candidate number is set according to the SCORE in this way, it is expected that the smaller the candidate number (the higher the candidate), the higher the relevance between the past drawing and the drawing ND. For example, candidate 1 indicates the past drawing, among drawings a1 to M1, indicated by SCORE as having the highest relevance to drawing ND.

As described above, SCORE(b1) is the maximum relevance score (the highest relevance score) among SCORE(a1) to SCORE(M1). Therefore, the search result data generation unit 124s determines the drawing b1 as the candidate 1 (see hatched parts in FIG. 8). In this way, the search result data generation unit 124s selects the drawing b1 as the past drawing having the highest relevance to the drawing ND.

SCORE(c2) is the second largest relevance score among SCORE(a1) to SCORE(M1). Therefore, the search result data generation unit 124s determines the drawing c2 as the second candidate. In this way, the search result data generation unit 124s selects the drawing c2 as the past drawing having the second highest relevance to the drawing ND.

Furthermore, SCORE(a1) is the smallest relevance score (the M-th largest relevance score) among SCORE(a1) to SCORE(M1). Therefore, the search result data generation unit 124s determines the drawing a1 as the candidate M. FIG. In this way, the search result data generation unit 124s selects the drawing a1 as the past drawing having the lowest (Mth highest) relevance to the drawing ND.

The search result data generator 124s generates data indicating the candidate numbers of the drawings a1 to M1 for the drawing ND. In this data, the drawing numbers (identifiers) of drawings a1 to M1 are sorted according to SCORE. More specifically, in the data, the drawing numbers of drawings a1 to M1 are sorted in descending order of SCORE.

Note that, even when the type of the new drawing is different from the outline drawing, the search result data generation unit 124s performs the same processing on the past drawing of the same type as the new drawing. For example, consider the case where the type of new drawing is an assembly drawing. In this case, the search result data generation unit 124s determines candidate numbers of drawings a2 to M2 (assembly drawings) for the new drawing based on SCORE(a2) to SCORE(M2). The same applies to other types of drawings.

As described above, the search result data generation unit 124s generates drawing (i, 1) to drawing (i, M) based on SCORE (i, 1) to SCORE (i, N) for drawing ND belonging to type i. ) can be set. Specifically, the search result data generation unit 124 identifies SCORE (i, g), which is the g-th largest SCORE among SCORE (i, 1) to SCORE (i, M). Then, the search result data generation unit 124s determines the past drawing corresponding to SCORE (i, g), that is, the drawing (i, g) as the candidate g.

As an example, the search result data generation unit 124s generates a fifth reference table TB5s as search result data. In TB5s, the past drawings are sorted in ascending order of the candidate number (in descending order of SCORE) for each type of drawing ND. FIG. 8 shows an example of TB5s. The search result data generation unit 124s supplies the TB5s to the display unit 72 and causes the display unit 72 to display the TB5s.

By displaying the TB5s on the display unit 72, the user can be presented with the search results of the drawing search device 12s (more specifically, the search result data generation unit 124s) in a highly browsable manner.

However, the search result data generation unit 124s may not necessarily present the entire TB5s as the search result data to the user. This is because, if the number of past drawings is too large, presenting the entire TB 5s to the user may rather reduce the user's convenience.

Therefore, the search result data should just be at least a part of TB5s. The search result data may be data indicating at least one drawing (at least one past drawing) corresponding to the target drawing (drawing ND).

By the way, as described above, the candidate 1 drawing is the drawing with the highest SCORE among the plurality of past drawings. Therefore, from the viewpoint of reducing the amount of search result data, it is conceivable to generate the search result data as data indicating only the drawings of candidate 1. FIG.

However, when designing equipment in the field of plant engineering, a designer needs to comprehensively judge various technical information. Therefore, when designing a new device described in drawing ND with reference to past drawings, the past drawing with the highest SCORE (drawing of candidate 1) is not necessarily the most suitable for design reference. For example, a Candidate 2 drawing (a drawing with a lower SCORE) may actually be a better design reference than a Candidate 1 drawing.

Therefore, it is preferable that the search result data is data indicating a plurality of top candidates (eg, candidates up to the γ-th candidate). That is, the search result data may be data indicating drawings of candidates 1 to γ. γ in the reference embodiment is an integer of 1 or more and g or less.

(Effect of reference form)
In plant engineering, it may be necessary to search for past drawings that are highly relevant to the drawing ND (target drawing) as the project related to property T progresses (see also Patent Document 1). In the above project, when various plans (e.g., production period forecast, production cost estimate) are made for the same type of equipment in property T, referring to the design and production results of a specific type of equipment in the past property. This is because there are many

Here, if it is a past drawing in which the same specification information as that of the drawing ND is described, it is expected that the drawing ND has a high relevance. However, in plant engineering, it is common that the number of drawings of past projects is enormous. Therefore, even if the drawing search system of Patent Document 1 is used, it is not necessarily easy for the user to quickly find a desired drawing (a past drawing highly related to the drawing ND) from the search results. This is because, in the drawing search system of Patent Document 1, if a keywordless search is performed, it is conceivable that a considerable number of past drawings will be hit as a search result.

Therefore, it is one idea to search for a desired drawing highly related to the drawing ND using a known drawing search system having a keyword search function. For example, it is conceivable to search past drawings using a specific character string described in the drawing ND as a keyword (search word). However, it is not always easy for the user to quickly find a desired drawing from the search results even when performing such a keyword search. This is because it is conceivable that a considerable number of past drawings will be hit as a search result even when a keyword search is performed. Moreover, if the user cannot determine an appropriate search word, it is difficult to use the keyword search in the first place.

As described above, it is difficult for a user (eg, a new designer) who does not have sufficient knowledge of plant engineering to quickly search for a desired drawing using a conventional drawing search system. Unless you are a user familiar with plant engineering (eg, a veteran designer), it is not necessarily easy to efficiently search for a desired drawing using a conventional drawing search system.

The inventors of the present application (hereinafter referred to as the inventors) newly created the information processing system 100s (more specifically, the information processing apparatus 1s) in view of such problems of the conventional technology. According to the information processing device 1s, a desired drawing corresponding to the drawing ND can be retrieved using a learning model obtained by machine learning for each past drawing.

First, in the learning device 11s of the information processing device 1s, a content parameter (search target drawing content parameter) related to a specific character string is set for each past drawing. Then, a weighting factor for each past drawing is set based on the search target drawing content parameter.

After that, the drawing search device 12s of the information processing device 1s uses the content parameter (target drawing content parameter) of the drawing ND and the weighting coefficient of each past drawing to increase the relevance between the drawing ND and each past drawing. A relevance score, which is an index indicating the degree of relevance, is calculated for each past drawing. Each past drawing is then sorted in descending order of relevance score. That is, based on the relevance score, at least one past drawing is presented to the user as a candidate for the desired drawing.

Thus, according to the drawing search device 12s, unlike a known drawing search system having a keyword search function, the user can efficiently search for a desired drawing through keywordless search. In particular, the drawing search device 12s allows the user to efficiently search for drawings even when the user cannot determine an appropriate search word. For this reason, the drawing retrieval device 12s can be said to be particularly suitable, for example, for improving the work efficiency of the above-described new designer.

Note that the drawing retrieval system of Patent Document 1 is common to the information processing device 1s (more specifically, the drawing retrieval device 12s) in that it is premised on keywordless retrieval. However, unlike the information processing device 1s, the drawing retrieval system of Patent Document 1 does not order the retrieval results. For this reason, in the drawing search system of Patent Document 1, the past drawings listed at the top of the search results do not necessarily have a high degree of relevance to the drawing ND.

On the other hand, in the information processing device 1s, data obtained by sorting past drawings in descending order of relevance score is presented to the user as a search result. Therefore, the user can confirm each past drawing presented as a search result in descending order of relevance score. In other words, it is possible to allow the user to preferentially check past drawings that are listed as higher-ranked candidates.

In this way, in the information processing device 1s, unlike the drawing search system of Patent Document 1, it is not necessary for the user to exhaustively check each past drawing presented as a search result (in a round-robin manner). . Therefore, according to the information processing apparatus 1s, even compared to the drawing search system of Patent Document 1, it is possible for the user to find the desired drawing corresponding to the drawing ND more easily than before. That is, according to the information processing device 1s, it is possible to further improve the user's convenience as compared with the drawing retrieval system of Patent Document 1.

By the way, in recent years, image matching technology using machine learning (eg, face recognition technology) has been widely used. It is also conceivable to search for a desired drawing using the image matching technology. However, in the above image collation technique, the relevance between the drawing ND and each past drawing is determined based on the "appearance" (eg, layout) of the image. Therefore, the drawing search method based on the image matching technology (hereinafter referred to as image matching-based drawing search) may not be suitable for drawing search compared to the information processing device 1s. This point will be described below.

For example, if the drawing description format is not standardized (unified), it is conceivable that the drawing description method differs for each drawing creator. Therefore, even if the same content (eg, specification information) is described in two drawings, the content may be described in different positions in each drawing. For example, refer to "area L1 in FIG. 3" (hereinafter also referred to as area A) and "area R1 in FIG. 6" (hereinafter also referred to as area B).

In such a case, in the image matching-based drawing search, since the positions of the area A and the area B are different, it can be evaluated that the relevance of the drawing a1 to the drawing ND is low. However, as described above, area A and area B contain the same type of specification information (specification information regarding voltage values). Therefore, drawing a1 can actually be considered to have a certain degree of relevance to drawing ND.

As described above, the image matching-based drawing search cannot evaluate the relevance between the drawing ND and the drawing a1 in consideration of the content of the specification information described in each drawing. Therefore, it is not possible to appropriately search for a past drawing (drawing b1 in the example of FIG. 8) in which although the appearance is different from the drawing ND, similar content itself is described.

On the other hand, in the information processing device 1s, unlike the image matching-based drawing search, a content parameter (a search target drawing content parameter and a target drawing content parameter) is set by focusing on a specific character string described in each drawing. be. That is, the information processing device 1s can consider the content of the specification information described in each drawing. Based on the content parameter, the relationship between the drawing ND and each past drawing is evaluated (eg, scored).

Therefore, according to the information processing device 1s, unlike the image matching-based drawing search, a past drawing (eg, drawing b1) in which similar content is described although it looks different from the drawing ND, Can be searched properly. In this manner, the information processing device 1s is more suitable for drawing retrieval than for image matching-based drawing retrieval.

In addition, as the number of past drawings increases, it becomes more difficult to appropriately search for past drawings highly relevant to a given new drawing using a conventional drawing search system (or image matching-based drawing search). Therefore, the greater the number of past drawings, the more useful the information processing apparatus 1 s is for improving user convenience.

[Modification]
(1) An information processing apparatus according to an aspect of the present invention is also applicable to drawing searches in fields other than plant engineering. The drawing according to one aspect of the present invention may be any drawing from which content parameters can be acquired by the drawing search device.

(2) In the reference form, the learning model generation unit 113s generates a learning model using multinomial logistic regression analysis. However, the learning model generation method is not limited to this.

The learning model generation unit 113s only needs to be able to generate a learning model by machine learning using the above-described teacher data (it is sufficient if each weighting factor can be calculated). For example, the learning model generation unit 113s can also generate a learning model using known neural network technology (eg, known deep learning technology).

(3) The search result data generator 124s may further acquire predetermined data corresponding to the drawing selected as the candidate g. In the example of FIG. 8, the search result data generation unit 124s may acquire the specification data of a predetermined device corresponding to the drawing b1 (candidate 1). This is because, in various plans of plant engineering, it is often the case that both past drawings and specifications of predetermined equipment described in the past drawings are referred to.

In this case, the search result data generator 124s can present the specification data to the user together with the search result data. Therefore, user convenience can be further improved. As an example, in the information processing device 1s, a table showing the correspondence between the drawings a1 to MN and the specification data may be prepared in advance. The search result data generation unit 124s may obtain specification data corresponding to the candidate g by referring to the table.

[Embodiment 1]
In the information processing system 100s (information processing device 1s) of the reference embodiment, each process is executed on the assumption that past drawings (drawings a1 to MN) are classified in advance by drawing type. However, in practice, there may be a case where past drawings are not classified in advance. Also, it may not always be easy to classify past drawings by drawing type depending on manpower. The inventors have created the information processing system 100 of the first embodiment in order to enable appropriate retrieval of past drawings even in such a situation.

FIG. 9 is a block diagram showing the configuration of the main part of the information processing system 100. As shown in FIG. An information processing device of the information processing system 100 is called an information processing device 1 . The past object drawing DB of the information processing system 100 is referred to as a past object drawing DB 91 . Similarly, the control device of the information processing device 1 will be referred to as a control device 10 . The learning device and the drawing retrieval device of the control device 10 are referred to as a learning device 11 (model generation device) and a drawing retrieval device 12, respectively.

(Learning device 11)
Unlike the learning device 11s, the learning device 11 includes a group classifying section 114. FIG. Also, the learning model generation unit of the learning device 11 is referred to as a learning model generation unit 113 (learning unit). In the past object drawing DB 91, unlike the past object drawing DB 91s, the drawings a1 to MN are not classified in advance by drawing type.

In the first embodiment, the past drawing data acquisition unit 111 acquires each past drawing from the past object drawing DB 91 in a predetermined order (eg, in order of the time stamp of each past drawing). For convenience of explanation, in the first embodiment, the past drawing data acquisition unit 111 acquires past drawings in the order of "drawing a1→a2→a3→...→MN".

The past drawing data preprocessing unit 112 derives a content parameter (search target drawing content parameter) for each past drawing acquired by the past drawing data acquisition unit 111 . Then, the past drawing data preprocessing unit 112 generates a table showing content parameters of each past drawing as a first table TB1. FIG. 10 shows an example of TB1. The l-th column of TB1 indicates the number of the l-th past drawing acquired by the past drawing data acquisition unit 111 . l is an integer greater than or equal to 1 and less than or equal to T; The k-th row of TB1 indicates the k-th content parameter of a certain past drawing.

Based on TB1 (more specifically, based on each content parameter shown in TB1), the group classification unit 114 classifies each past drawing into P groups from the first group to the P-th group. P is an integer greater than or equal to 2 and less than T; In the following description, the qth group is also called group q. q is an integer greater than or equal to 1 and less than or equal to P;

The group classification unit 114 classifies each past drawing into P groups using a known clustering method. As an example, the group classification unit 114 classifies each past drawing using the k-means method. FIG. 11 shows an example of the classification result of each past drawing by the k-means method.

In the classification by the k-means method, the coordinates of points associated one-to-one with each past drawing (for convenience, also referred to as "points of each past drawing") are stored in an L-dimensional data space ( L-dimensional space). The coordinates of a certain past drawing are defined by the content parameters (first content parameter to L-th content parameter) of the past drawing. In addition, in the example of FIG. 11, for convenience of illustration, the case where the points of each past drawing are plotted in a three-dimensional space (that is, the case of L=3) is illustrated. x, y, and z in FIG. 11 represent the first content parameter, the second content parameter, and the third content parameter, respectively.

For convenience of explanation, the k-th content parameter of drawing a1 in Embodiment 1 is written as a1(k). The group classification unit 114 acquires a1(1) to a1(L) from the past drawing data preprocessing unit 112 . Then, the group classification unit 114 stores the coordinates of the point P(a1) in the drawing a1 in the above L-dimensional space as follows:
P(a1)=[a1(1),a1(2),...,a1(k),...a1(L)]
plot as In this way, the group classification unit 114 acquires the coordinates of the point P(a1) from the past drawing data preprocessing unit 112 .

The group classification unit 114 also performs similar plotting for other drawings. Below, the case where the drawing a1 belongs to the group 1 is illustrated like the example of FIG. In addition, past drawings belonging to group q are also generically referred to as past drawings within group q.

Then, the group classification unit 114 generates a second table TB2 indicating the content parameters (A1 to AL) of each past drawing in the first embodiment. TB2 includes the q-th sub-table TB2-q in the second table. TB2-q shows the content parameters of each past drawing belonging to group q. FIG. 12 shows TB2-1 as an example. TB2 is composed of a set of P sub-tables, TB2-1 to TB2-P. Thus, in the first embodiment, sub-tables are created for each group classified by the group classification unit 114, unlike the reference embodiment.

Note that in the example of FIG. 12 , a plurality of past drawings of different types are classified into group 1 . As described above, by performing clustering based on the content parameter, past drawings can be classified with a higher degree of flexibility than when past drawings are classified by manpower.

(Generation of learning model in Embodiment 1)
As described above, the learning model generation unit 113s of the reference embodiment generates one (common) learning model based on the drawings a1 to MN. Unlike the learning model generation unit 113s, the learning model generation unit 113 of the first embodiment generates a plurality of learning models.

The learning model generation unit 113 generates a learning model corresponding to the group q (hereinafter referred to as q-th learning model ). Specifically, the learning model generation unit 113 generates the qth learning model using TB2-q as teacher data. The method of generating each learning model is the same as that of the learning model generation unit 113s.

As described above, the learning model generation unit 113 generates P different learning models from the first learning model to the Pth learning model. The q-th learning model includes weight coefficients of each past drawing belonging to group q.

Then, the learning model generating unit 113 generates a third table TB3 indicating the content parameters of each past drawing in the first embodiment. TB3 includes the q-th sub-table TB3-q in the second table. TB3-q shows the weighting factor of each past drawing belonging to group q. FIG. 13 shows TB3-1 as an example. TB3 is composed of a set of P sub-tables, TB3-1 to TB3-P.

(Drawing retrieval device 12)
The drawing retrieval device 12 includes a group selection unit 125, unlike the drawing retrieval device 12s. Also, the score calculation unit and the search result data generation unit of the drawing search device 12 are referred to as a score calculation unit 123 and a search result data generation unit 124, respectively. The score calculator 123 and the search result data generator 124 may be generically called a searcher.

First, the new drawing data acquisition unit 121 acquires the drawing ND, as in the reference embodiment. Subsequently, the new drawing data preprocessing unit 122 acquires the content parameters of the drawing ND (target drawing content parameters). Below, for convenience of explanation, the k-th content parameter of the drawing ND is also written as ND(k).

(selection of specific groups)
The group selection unit 125 acquires ND1(1) to ND1(L) from the new drawing data preprocessing unit 122. FIG. Then, the group selection unit 125 stores the coordinates of the point P(ND) on the drawing ND in the above L-dimensional space as follows:
P(ND)=[ND(1),ND(2),...,ND(k),...ND(L)]
plot as In this way, the group selection unit 125 acquires the coordinates of the point P(ND) from the new drawing data preprocessing unit 122 . In FIG. 14 described below, the point ND is indicated by an asterisk.

Subsequently, the group selection unit 125 acquires information indicating the classification result of the group classification unit 114 from the group classification unit 114 . Based on the point P(ND) (in other words, based on the target drawing content parameter), the group selection unit 125 selects the group having the highest relevance to the drawing ND among the first to P-th groups (hereinafter referred to as specific group). The point P(ND) may be referred to as the target point because it is the point on the L-dimensional space corresponding to the target drawing.

In the first embodiment, the group selection unit 125 selects the group closest to the point P(ND) from among the first to P-th groups in the L-dimensional space as the specific group. For this reason, the particular group may be referred to as the closest group. FIG. 14 shows an example of the closest group selection result by the group selection unit 125 . In the example of FIG. 14, group 1 is selected as the closest group. In this case, the group selection unit 125 acquires the first learning model from the learning model generation unit 113 as the learning model corresponding to the specific group (hereinafter referred to as the specific learning model).

The "distance between the point P (ND) and the group q" (hereinafter referred to as the distance D) is, more specifically, the "distance between the point P (ND) and the representative point of the group q (more strictly , norm)”. As an example, the representative point of group q may be the center of gravity of group q. However, the representative point of group q is not limited to the center of gravity of group q. Any point located inside or on the boundary of group q in the L-dimensional space may be selected as the representative point of group q.

Note that the coordinates of the representative point of group q may be calculated in advance by the group classification unit 114 . In this case, the group selection unit 125 may obtain the coordinates of the representative point of the group q from the group classification unit 114 . Alternatively, the group selection unit 125 may calculate the coordinates of the representative point of the group q.

In the first embodiment, the group selection unit 125 calculates the distance D as the Euclidean distance between the coordinate P(ND) and the representative point of the group q. However, the distance D is not limited to the Euclidean distance. The distance D may be any known kind of distance (norm) definable in L-dimensional space. For example, the distance D may be calculated as Mahalanobis distance, Manhattan distance, or Chebyshev distance.

(Calculation of relevance score in Embodiment 1)
The score calculator 123 acquires the specific learning model from the group selector 125 . The score calculator 123 calculates the SCORE of the drawing ND using the specific learning model. In the example of the first embodiment, the score calculator 123 calculates the SCORE of the drawing ND using the first learning model as the specific learning model. In Embodiment 1, by calculating SCORE using a specific learning model, the calculation cost for calculating SCORE can be reduced compared to the reference embodiment.

FIG. 14 is a diagram illustrating an example of deriving SCORE according to the first embodiment. Note that the target drawing content parameters in the example of FIG. 14 are different from those in the example of FIG. In the example of FIG. 14, the score calculation unit 123 uses (i) C1 to CL derived by the new drawing data preprocessing unit 122 and (ii) B1 to BL of drawing a1 included in the specific learning model. Then, the 1st to Lth raw scores of the drawing a1 are calculated. That is, the score calculator 123 acquires each raw score as the output of the specific learning model. Then, the score calculator 123 calculates the sum of the first to L-th raw scores of the drawing a1 as the SCORE of the drawing a1, as in the reference embodiment.

Similarly, the score calculator 123 uses the specific learning model to calculate the SOCRE of other past drawings belonging to the specific group (group 1). Then, the score calculator 123 generates a fourth sub-table TB4-1 indicating its own calculation result. TB4-1 shows SS1-SSL and SCORE of Group 1 respectively.

(Determination of Candidate Order in Embodiment 1)
The search result data generation unit 124 determines the candidate order corresponding to the drawing ND based on the SCORE of each past drawing belonging to the specific group (group 1). In the following description, the number of past drawings belonging to group 1 is denoted as GN1. Based on the SCORE of each past drawing belonging to group 1, the search result data generation unit 124 determines the candidate order (first candidate to GN1 candidate) of each past drawing.

As an example, the search result data generation unit 124 generates a fifth sub-table TB5-1 as search result data. In TB5-1, past drawings belonging to group 1 are sorted in descending order of candidate number (in descending order of SCORE). FIG. 16 shows an example of TB5-1.

In the example of FIG. 15, SCORE(d2) is the maximum relevance score and SCORE(a1) is the minimum relevance score among GN1 SCOREs calculated by the score calculation unit 123 . Therefore, the search result data generation unit 124 determines the drawing d2 as the candidate 1 and the drawing a1 as the candidate GN1.

In the first embodiment, candidate ranking is determined only for each past drawing belonging to a specific group. Therefore, the calculation cost for generating search result data can be reduced compared to the reference embodiment. It should be noted that the search result data in the first embodiment may also be data indicating drawings of candidates 1 to γ. γ in Embodiment 1 is an integer of 1 or more and GN1 or less.

(supplement)
In the above example, the k-means method is used as a clustering method for classifying each past drawing based on the content parameter. However, other clustering techniques can also be used. For example, a fuzzy c-means method may be used as the clustering method. When the fuzzy c-means method is used, it can be expected that each past drawing can be classified more flexibly than when the k-means method is used.

(Experimental example)
The inventors conducted further experiments to verify the effectiveness of the information processing system 100 . Specifically, the inventors performed the same processing as in the first embodiment when about 1600 past drawings were stored in the past property drawing DB 91 . In other words, for the case of T≈1600, the inventors (i) generate the first to Pth learning models, and (ii) search past drawings using a specific learning model, as in the first embodiment. did

In this experimental example, the fuzzy c-means method was used as the clustering method. Also, P=96 was set. In this experimental example, as in the example of the first embodiment, it was confirmed that past drawings corresponding to new drawings can be appropriately retrieved. That is, it was confirmed that past drawings belonging to a specific group can be presented in descending order of relevance score.

(Effect of Embodiment 1)
Similar to the reference embodiment, the information processing system 100 (information processing apparatus 1) can improve the user's convenience in searching for drawings. In addition, according to the information processing apparatus 1, unlike the reference embodiment, even when past drawings are not classified in advance (or when it is difficult to classify past drawings by manpower), past drawings can be appropriately classified. can be searched for. Therefore, it is possible to further improve user convenience.

In particular, the learning device 11 of the information processing device 1 can classify the past drawings based on the content parameters of the past drawings (past drawing content parameters). Therefore, a plurality of past drawings with high content similarity can be classified into the same group. Subsequently, the learning device 11 can generate an individual learning model for each group after classification. More specifically, the learning device 11 can generate the q-th learning model based on the content parameters of each past drawing belonging to the group q.

Thus, in the first embodiment, common parameters called "content parameters" are used for different processes of "clustering" and "generation of learning models". A learning model is generated for groups with similar "content parameters." Therefore, it is possible to generate a learning model that accurately distinguishes similar drawings. The drawing search device 12 uses common parameters called "content parameters" to search for drawings by changing the viewpoint between "clustering" and "search by learning model". Therefore, the drawing search device 12 can extract search target drawings that are more similar to the target drawing. According to the flow of processing described above, it is possible to generate a plurality of suitable learning models (first to P-th learning models) by improving the accuracy of drawing search.

Further, the drawing search device 12 of the information processing device 1 can select a specific group based on the content parameter of the drawing ND (target drawing content parameter). Subsequently, in the drawing search device 12, based on the specific learning model, a search process can be performed only for past drawings belonging to a specific group.

According to the above-described process flow, the calculation cost for drawing search can be reduced, so that drawing search can be performed at a higher speed. Also, since the specific group is the group that has the highest relevance to the drawing ND, the specific learning model is expected to be the most suitable learning model for searching the drawing ND. By performing a drawing search using the specific learning model, the accuracy of the drawing search can be further improved.

[Modification]
(1) P (the number of groups of past drawings) may be set by the user of the information processing apparatus 1, or may be set using a known algorithm. Pseudo F is known as one of indices for evaluating the results of clustering. Therefore, P may be set based on Pseudo F, for example.

(2) In the first embodiment, the case where clustering is performed in the L-dimensional space was exemplified. However, a known dimensionality reduction method may be used to reduce the calculation cost. An example of a dimensionality reduction technique is principal component analysis.

Another example of a dimensionality reduction technique is
・PCA (Principal Component Analysis)
・Kernel PCA
・Manifold Learning
・LDA (Linear Discriminant Analysis)
・NMF (Non-negative Matrix Factorization)
・SVD (Singular Value Decomposition)
can be mentioned.

As an example, dimensionality reduction may be performed before clustering. In this case, clustering and learning model generation are performed based on the content parameters after dimensionality reduction.

As another example, dimensionality reduction may be performed after clustering. In this case, clustering is performed based on the content parameters before dimensionality reduction. On the other hand, a learning model is generated based on the content parameters after dimensionality reduction. Thus, the dimensionality may be different in clustering and learning model generation.

(3) In the information processing device according to one aspect of the present disclosure, the content parameter may be corrected (eg, normalized). Based on the content parameters after normalization, each of the processes described above (eg, clustering and learning model generation) may be performed.

(4) In one aspect of the present disclosure, the drawing search device and the model generation device may be connected so as to be able to communicate with each other. Therefore, the drawing retrieval device and the model generation device can be provided in separate information processing devices.

[Example of realization by software]
The control block (especially the control device 10) of the information processing system 100 may be implemented by a logic circuit (hardware) formed in an integrated circuit (IC chip) or the like, or may be implemented by software.

In the latter case, the information processing system 100 comprises a computer that executes instructions of programs, which are software that implements each function. This computer includes, for example, one or more processors, and a computer-readable recording medium storing the program. In the computer, the processor reads the program from the recording medium and executes it, thereby achieving the object of the present invention. As the processor, for example, a CPU (Central Processing Unit) can be used. As the recording medium, a "non-temporary tangible medium" such as a ROM (Read Only Memory), a tape, a disk, a card, a semiconductor memory, a programmable logic circuit, or the like can be used. In addition, a RAM (Random Access Memory) for developing the above program may be further provided. Also, the program may be supplied to the computer via any transmission medium (communication network, broadcast wave, etc.) capable of transmitting the program. Note that one aspect of the present invention can also be implemented in the form of a data signal embedded in a carrier wave in which the program is embodied by electronic transmission.

[Additional notes]
The present invention is not limited to the above-described embodiments, but can be modified in various ways within the scope of the claims, and can be obtained by appropriately combining technical means disclosed in different embodiments. is also included in the technical scope of the present invention.

1 information processing device 10 control device 11 learning device (model generation device)
12 drawing search device 100 information processing system 111 past drawing data acquisition unit 112 past drawing data preprocessing unit (search target drawing content parameter acquisition unit)
113 learning model generation unit (learning unit)
114 group classification unit 121 new drawing data acquisition unit 122 new drawing data preprocessing unit (target drawing content parameter acquisition unit)
123 Score calculation unit (search unit)
124 search result data generation unit (search unit)
125 Group selection part a1 to MN Drawings (past drawings, search target drawings)
ND drawing (new drawing, target drawing)

Claims (9)

A drawing retrieval device for retrieving at least one drawing corresponding to a target drawing from a plurality of retrieval target drawings,
The plurality of search target drawings are P (P is an integer of 2 or more) from the first group to the P group based on the search target drawing content parameter related to the description content of each of the plurality of search target drawings pre-sorted into groups,
P learning models corresponding to each of the P groups are generated in advance based on the search target drawing content parameter of each of the plurality of search target drawings belonging to each of the P groups,
a target drawing content parameter acquiring unit for acquiring target drawing content parameters related to the description content of the target drawing by analyzing the target drawing;
a group selection unit that selects, from among the P groups, a specific group, which is the group most relevant to the target drawing, based on the target drawing content parameter;
(i) inputting the target drawing content parameter to a specific learning model, which is a learning model corresponding to the specific group, among the P learning models; and a search unit that searches for the at least one drawing based on the output of the specific learning model. The search unit calculates a relevance score indicating a degree of relevance of each of the plurality of search target drawings belonging to the specific group to the target drawing based on the output of the specific learning model. 2. The drawing retrieval device according to 1. The search unit generates search result data indicating an identifier of the at least one drawing belonging to the specific group,
3. The drawing retrieval device according to claim 2, wherein in said retrieval result data, identifiers of said at least one drawing belonging to said specific group are sorted according to said relevance score. In the above drawing search device, a specific character string related to drawings is set in advance,
The target drawing content parameter acquisition unit
Detecting the above-mentioned specific character string contained in the above-mentioned target drawing,
4. The drawing retrieval device according to claim 1, wherein said target drawing content parameter is set based on a detection result of said specific character string in said target drawing. In the drawing retrieval device, L (L is an integer of 2 or more) different character strings from a first specific character string to an L-th specific character string are set in advance as the specific character strings,
The target drawing content parameters include L parameters from the first target drawing content parameter to the Lth target drawing content parameter,
The search target drawing content parameter includes L parameters from the first search target drawing content parameter to the Lth search target drawing content parameter,
The above group selection part is
In an L-dimensional data space showing the distribution of the target drawing content parameter and the search target drawing content parameter, (i) the coordinates of the target point corresponding to the target drawing, and (ii) the first group to the Pth group. Get the coordinates of the representative points of each group of
5. The drawing retrieval apparatus according to claim 4, wherein a group having the shortest distance between the target point and the representative point in the L-dimensional data space is selected from among the groups as the specific group. A model generation device for generating a learning model for retrieving at least one drawing corresponding to a target drawing from a plurality of search target drawings,
a search target drawing content parameter acquisition unit that acquires a search target drawing content parameter related to the description content of each of the plurality of search target drawings by analyzing the plurality of search target drawings;
Based on the search target drawing content parameter of each of the plurality of search target drawings, the plurality of search target drawings are classified into P groups (P is an integer of 2 or more) from a first group to a P group. a group classifier;
a learning unit that generates P learning models corresponding to each of the P groups based on the search target drawing content parameters of each of the plurality of search target drawings belonging to each of the P groups; A model generator. In the above model generation device, a specific character string related to the drawing is set in advance,
The search target drawing content parameter acquisition unit
Detecting the specific character string contained in each of the plurality of search target drawings,
7. The model generation device according to claim 6, wherein said search target drawing content parameter for each of said plurality of search target drawings is set based on the detection result of said specific character string in each of said plurality of search target drawings. A drawing retrieval method for retrieving at least one drawing corresponding to a target drawing from a plurality of retrieval target drawings,
The plurality of search target drawings are P (P is an integer of 2 or more) from the first group to the P group based on the search target drawing content parameter related to the description content of each of the plurality of search target drawings pre-sorted into groups,
P learning models corresponding to each of the P groups are generated in advance based on the search target drawing content parameter of each of the plurality of search target drawings belonging to each of the P groups,
a target drawing content parameter acquiring step of acquiring target drawing content parameters related to the description content of the target drawing by analyzing the target drawing;
a group selection step of selecting, from among the P groups, a specific group, which is the group most relevant to the target drawing, based on the target drawing content parameter;
(i) inputting the target drawing content parameter to a specific learning model, which is a learning model corresponding to the specific group, among the P learning models; a retrieving step of retrieving the at least one drawing based on the output of a particular learning model. A model generation method for generating a learning model for retrieving at least one drawing corresponding to a target drawing from a plurality of search target drawings,
a search target drawing content parameter obtaining step of obtaining search target drawing content parameters related to the description content of each of the plurality of search target drawings by analyzing the plurality of search target drawings;
Based on the search target drawing content parameter of each of the plurality of search target drawings, the plurality of search target drawings are classified into P groups (P is an integer of 2 or more) from a first group to a P group. a group classification step;
a learning step of generating P learning models corresponding to each of the P groups based on the search target drawing content parameters of each of the plurality of search target drawings belonging to each of the P groups; Including, model generation methods.

Images