Learning facility, administrative sheet creation support facility, program, learning method and administrative sheet creation support method

Abstract

An FMEA sheet creation support device (100) comprises: a past sheet storage unit (121) containing a past sheet created as an administrative sheet, which has a plurality of lines, each of the plurality of lines containing at least operational process information specifying an operational process contained in a plurality of operational processes, and a risk set specifying information about a risk in the one operational process;a training data generation unit (111) that generates correspondence relationship training data containing a positive example and a negative example, wherein the positive example is a combination of the operational process information contained in one of the plurality of rows in the former case sheet and the risk set contained in that one row, and wherein the negative example is a combination of the operational process information contained in that one row and the risk set contained in a row other than that one row; and a correspondence relationship learning unit (113) that generates a correspondence relationship model by using the correspondence relationship training data.

Description

AREA OF TECHNOLOGY

[0001] The present disclosure relates to a learning facility, an administrative document creation support facility, a program, a learning procedure and an administrative document creation support procedure. TECHNICAL BACKGROUND

[0002] Failure Mode and Effect Analysis (FMEA) is a quality management method that uses an FMEA sheet to predict potential failures in a design or execution phase of a process and to identify the necessary countermeasures.

[0003] FMEA sheets are often created based on the operator's knowledge or experience, or on past failures and similar data. However, relying too heavily on the operator's knowledge or experience can lead to variations in the FMEA sheet's content, and there's a risk of omitting failures the operator hasn't encountered. Furthermore, selecting a suitable document from multiple past cases to create the sheet can be challenging and a time-consuming and labor-intensive process.

[0004] Patent reference 1 discloses a support system designed to assist in the creation of FMEA sheets. When a user identifies a relevant portion of documents to be consulted in the FMEA sheet, the support system generates reference text data using text data entered into the identified portion and also generates reference feature data that maps the relationship between words in the text to their relevance level. The support system then generates similar feature data for the failure case documents to be searched, calculates the degree of similarity between the reference feature data and the feature data for each failure case document, and outputs a failure case document with the highest degree of similarity. REFERENCES ON THE STATE OF THE TECHNOLOGY PATENT REFERENCE

[0005] PATENT REFERENCE 1: Japanese patent application Publication No. 2011-8355 SUMMARY OF THE INVENTION; PROBLEM TO BE SOLVED BY THE INVENTION

[0006] However, conventional support systems have generated feature sets intended for text searches solely from text information on FMEA sheets, neglecting both the structural features of the FMEA sheets and previously created FMEA sheets. Consequently, there have been instances where no useful information could be provided to the user.

[0007] Therefore, one or more aspects of the present disclosure aim to provide more useful information when a user creates an administrative sheet. MEANS TO SOLVE THE PROBLEM

[0008] A learning device according to one aspect of the present disclosure comprises: a former case sheet storage unit that stores a previously created former case sheet as an administrative sheet, the administrative sheet containing a plurality of lines, each of the plurality of lines containing at least operational process information specifying an operational process contained in a plurality of operational processes, and a risk set specifying information about a risk in the one operational process;a training data generation unit that generates correspondence relationship training data, wherein the training data contains a positive example and a negative example, the positive example being a combination of the operational process information contained in one of the plurality of rows in the former case sheet and the risk set contained in that row, and the negative example being a combination of the operational process information contained in that row and the risk set contained in a row other than that row; and a correspondence relationship learning unit that generates a correspondence relationship model by learning a correspondence relationship between the operational process information and the risk set using the correspondence relationship training data.

[0009] An administrative sheet creation support device according to one aspect of the present disclosure comprises: a correspondence relationship model storage unit that stores a correspondence relationship model for a previously created former case sheet as an administrative sheet, wherein the administrative sheet contains a plurality of rows, each of the plurality of rows containing at least operational process information specifying an operational process contained in a plurality of operational processes, and a risk set specifying information about a risk in the one operational process, wherein the correspondence relationship model is generated by learning a correspondence relationship between the operational process information and the risk set using the correspondence relationship training data, which are data that contain a positive example and a negative example.wherein the positive example is a combination of the operational process information contained in one of the multitude of lines in the former case sheet and the risk set contained in that one line, wherein the negative example is a combination of the operational process information contained in that one line and the risk set contained in a line other than that one line; a document storage unit that stores a multitude of documents; an information retrieval unit that retrieves search operational process information, which is operational process information for the search; a correspondence relationship determination unit that generates a multitude of pieces of search sequence data by adding each of a multitude of sets contained in the multitude of documents to the search operational process information, summing a multitude of scores,which are obtained by inputting the multitude of pieces of search sequence data into the correspondence relationship model for each of the multitude of documents, each containing the multitude of sentences, and then specifying a document with the highest summed score as reference information; and a display processing unit that generates a screen image for displaying the reference information.

[0010] A program according to one aspect of the present disclosure causes a computer to function as: a past-case sheet storage unit that stores a past-case sheet created as an administrative sheet, the administrative sheet containing a plurality of lines, each of the plurality of lines containing at least operational process information specifying an operational process contained in a plurality of operational processes, and a risk set specifying information about a risk in the one operational process;a training data generation unit that generates correspondence relationship training data, wherein the training data contains a positive example and a negative example, the positive example being a combination of the operational process information contained in one of the plurality of rows in the former case sheet and the risk set contained in that row, and the negative example being a combination of the operational process information contained in that row and the risk set contained in a row other than that row; and a correspondence relationship learning unit that generates a correspondence relationship model by learning a correspondence relationship between the operational process information and the risk set using the correspondence relationship training data.

[0011] A program according to one aspect of the present disclosure causes a computer to function as: a correspondence relationship model storage unit that stores a correspondence relationship model for a previously created former case sheet as an administrative sheet, the administrative sheet containing a plurality of rows, each of the plurality of rows having at least operational process information specifying an operational process contained in a plurality of operational processes, and a risk set specifying information about a risk in the one operational process, the correspondence relationship model being generated by learning a correspondence relationship between the operational process information and the risk set using the correspondence relationship training data, which are data that include a positive example and a negative example.wherein the positive example is a combination of the operational process information contained in one of the multitude of lines in the former case sheet and the risk set contained in that one line, wherein the negative example is a combination of the operational process information contained in that one line and the risk set contained in a line other than that one line; a document storage unit that stores a multitude of documents; an information retrieval unit that retrieves search operational process information, which is operational process information for the search; a correspondence relationship determination unit that generates a multitude of pieces of search sequence data by adding each of a multitude of sets contained in the multitude of documents to the search operational process information, summing a multitude of scores,which are obtained by inputting the multitude of pieces of search sequence data into the correspondence relationship model for each of the multitude of documents, each containing the multitude of sentences, and then specifying a document with the highest summed score as reference information; and a display processing unit that generates a screen image for displaying the reference information.

[0012] A learning method according to one aspect of the present disclosure comprises: generating correspondence relationship training data for a previously created former case sheet as an administrative sheet, wherein the administrative sheet contains a plurality of lines, each of the plurality of lines having at least operational process information specifying an operational process contained in a plurality of operational processes, and a risk set specifying information about a risk in the one operational process; the correspondence relationship training data being training data that includes a positive example and a negative example, wherein the positive example is a combination of the operational process information contained in one of the plurality of lines and the risk set contained in the one line.where the negative example is a combination of the operational process information contained in one row and the risk set contained in a row different from that row; and generating a correspondence relationship model by learning a correspondence relationship between the operational process information and the risk set using the correspondence relationship training data.

[0013] An administrative sheet creation support procedure according to one aspect of the present disclosure comprises: obtaining search operational process information, which is operational process information for the search; generating a plurality of pieces of search sequence data by adding each of a plurality of sets contained in the plurality of documents to the search operational process information; generating a correspondence relationship model for a previously created former case sheet as an administrative sheet, wherein the administrative sheet contains a plurality of lines, each of the plurality of lines having at least operational process information specifying an operational process contained in a plurality of operational processes and a risk set specifying information about a risk in the one operational process; and generating the correspondence relationship model.by learning a correspondence relationship between the business process information and the risk set using the correspondence relationship training data, which is data that includes a positive example and a negative example, where the positive example is a combination of the business process information contained in one of the multitude of lines in the past case sheet and the risk set contained in that one line, where the negative example is a combination of the business process information contained in that one line and the risk set contained in a line other than that one line, subsequently summing a multitude of scores obtained by inputting the multitude of pieces of search sequence data into the correspondence relationship model for each of the multitude of documents, each containing the multitude of sets,and then specifying a document with the highest total score as reference information; and generating a screen image to display the reference information. EFFECTS OF THE INVENTION

[0014] According to one or more aspects of the present disclosure, more useful information can be made available to the user if the user creates an administrative sheet. BRIEF DESCRIPTION OF THE DRAWINGS Fig. Figure 1 is a block diagram that schematically represents a configuration of an FMEA sheet preparation support device according to a first embodiment. Fig. Figure 2 is a schematic representation of a FEMA sheet. Fig. 3A and Fig. Figure 3B are schematic representations, each showing an example of linked sequence data. Fig. 4A and Fig. Figure 4B are schematic representations, each showing an example of replaced concatenated sequence data. Fig. Figure 5 is a schematic representation to explain machine learning in an Integrated Feature Learning Unit 112. Fig. Figure 6 is a first schematic representation to explain machine learning in a correspondence relationship learning unit. Fig. Figure 7 is a second schematic representation to illustrate machine learning in a correspondence relationship learning unit. Fig. 8A and Fig. 8B are schematic representations to explain the processing carried out in a correspondence relationship determination unit. Fig. Figure 9 is a block diagram showing an example of a computer. Fig. Figure 10 is a block diagram that schematically represents a configuration of an FMEA sheet preparation support device according to a second embodiment. Fig. 11A and Fig. Figure 11B are schematic representations, each showing an example of expanding integrated feature training data. Fig. Figure 12 is a block diagram that schematically represents a configuration of an FMEA sheet preparation support device according to a third embodiment. Fig. Figure 13 is a schematic representation that provides an example of additional linked sequence data. Fig. Figure 14 is a block diagram that schematically represents a configuration of an FMEA sheet preparation support device according to a fourth embodiment. TYPE AND METHOD OF EXECUTION OF THE INVENTION First embodiment

[0015] Fig. Figure 1 is a block diagram which schematically represents a configuration of an FMEA sheet preparation support device 100 according to a first embodiment.

[0016] The FMEA sheet creation support device 100 comprises a preprocessing unit 110, a storage unit 120, a search processing unit 130, an input unit 140 and a display unit 150.

[0017] The preprocessing unit 110 operates as a learning unit, learning learning models that are used in the search processing unit 130.

[0018] The preprocessing unit 110 comprises a training data generation unit 111, an integrated feature learning unit 112 and a correspondence relationship learning unit 113.

[0019] The training data generation unit 111 generates training data used for learning. Specifically, the training data generation unit 111 generates both integrated feature training data, which is the training data for learning in the integrated feature learning unit 112, and correspondence relationship training data, which is the training data for learning in the correspondence relationship learning unit 113.

[0020] First, an FMEA sheet is described.

[0021] FEMA uses a tabular administrative sheet, the so-called FEMA sheet, for quality management purposes. The FEMA sheet categorizes the content of the various error types and enters them into numerous fields. In addition, one or more fields are created for specific solutions to the errors, and the necessary information is entered therein.

[0022] Fig. Figure 2 is a schematic representation of the FEMA sheet.

[0023] A FEMA sheet 101, as presented, consists of tabular data comprising a product column 101a, a function column 101b, a process column 101c, a risk column 101k, an impact column 101l, an occurrence column 101m, a detection column 101n, and an importance column 101o. Each of these columns contains one entry.

[0024] Product identification information is stored in column 101a, e.g. a product name to identify a product to be manufactured through an operational process.

[0025] Function column 101b stores function identification information, e.g., a function name to identify the function of the product.

[0026] As described above, product information for specifying the product is stored in the product column 101a and in the function column 101b.

[0027] The process column 101c stores operational process information that specifies an operational process for manufacturing the product.

[0028] Here, process column 101c is subdivided into a large process column 101d, a medium process column 101e, and a small process column 101f. The small process column 101f is further subdivided into a "Who" column 101g, a "Where" column 101h, a "What" column 101i, and a "How" column 101j.

[0029] In other words, the operational process is divided into large, medium, and small processes, with the small process managing a single operational process in which a person performs a specific activity at a specific location.

[0030] The risk column 101k stores a risk set, i.e., a set that shows information about a risk in a single operational process.

[0031] The degree of impact of the risk is stored in column 101l.

[0032] The degree of occurrence of the risk is stored in column 101m.

[0033] The degree of risk detection is stored in column 101n.

[0034] The importance level column 101o stores the degree of importance of the risk.

[0035] As described above, the respective evaluation values ​​resulting from the risk evaluation are stored in the impact level column 101l, the occurrence level column 101m, the detection level column 101n, and the importance level column 101o. The information stored in these columns will also be referred to as evaluation information.

[0036] The FMEA sheet is set up as described above. Previously created FMEA sheets are stored in memory unit 120 as former drop sheets, as mentioned later.

[0037] It is assumed that an empty cell on the FMEA sheet has the same content as information that first appears in a cell at the top of the same column.

[0038] It is assumed that a large number of lines in the FMEA sheet are arranged in the order in which the operational processes are carried out.

[0039] As described above, the FMEA sheet functions as an administrative sheet containing a multitude of lines, with each of the multitude of lines containing at least the operational process information specifying an operational process contained within the multitude of operational processes, and a risk set specifying information about a risk in that one operational process.

[0040] Referring again to Fig. 1. The training data generation unit 111 generates training data from the FMEA sheets created in the past and stored in the memory unit 120. The FMEA sheets created in the past and stored in the memory unit 120 are also referred to as former case sheets.

[0041] The training data generation unit 111 creates a pair of input and output data as training data for learning the content of the former case study. The output data is also referred to here as ground truth data.

[0042] In the first embodiment, the training data generation unit 111 generates integrated feature training data for learning an ordering task as the first task and a word task as the second task, as well as correspondence relationship training data for learning correspondence relationships.

[0043] First, the integrated feature training data is described.

[0044] The training data generation unit 111 extracts at least the following as text: operational process information stored in process column 101c and a risk set stored in risk column 101k. This is done in units of two consecutive rows from the beginning of the former case sheet. Note that if cells are empty, the corresponding information is added and extracted as text. Product information stored in product column 101a and function column 101b is also extracted.

[0045] In the Fig. 2 FEMA sheet 101 shown, the training data generation unit 111 extracts, for example, as one unit the product information, the operating process information and the risk set stored in line 102a, and the product information, the operating process information and the risk set stored in line 102b.

[0046] The training data generation unit 111 then performs morphological analysis of the text extracted from a line to divide the text into tokens; each token represents the smallest unit of meaning. The training data generation unit 111 then arranges the divided tokens in the order in which they appear in the corresponding text to form a string as sequence data.

[0047] The training data generation unit 111 assigns a positive example label to the concatenated sequence data, i.e., the data provided by concatenating the sequence data contained in a unit in the order from top to bottom of the former case sheet. Meanwhile, the training data generation unit 111 assigns a negative example label to the concatenated sequence data, i.e., the data provided by concatenating the sequence data contained in a unit in the order from bottom to top of the former case sheet.

[0048] Fig. 3A and Fig. 3B are, for example, schematic representations, each showing an example of linked sequence data.

[0049] Fig. 3A and Fig. Figure 3B shows examples of the concatenated sequence data derived from lines 102a and 102b of the document in Fig. 2 FEMA sheet 101 shown were extracted.

[0050] Fig. Figure 3A illustrates the concatenated sequence data, where sequence data SDa, extracted from line 102a of FEMA sheet 101, and sequence data SDb, extracted from line 102b of the same, are concatenated in the order of line 102a followed by line 102b. In this case, a positive example label is assigned to this concatenated sequence data.

[0051] Meanwhile, illustrates Fig. 3B the concatenated sequence data, in which the sequence data SDa, extracted from line 102a of FEMA sheet 101, and the sequence data SDb, extracted from line 102b of the same, are concatenated in the order of line 102b followed by line 102a. In this case, a negative example label is assigned to this concatenated sequence data.

[0052] The concatenated sequence data with the label assigned to them in this way are also referred to as labeled concatenated sequence data.

[0053] Next, the training data generation unit 111 replaces, with a certain probability, some of the multitude of tokens within the chained sequence data with a mask token, which is a special token for learning, thus generating replaced chained sequence data.

[0054] Fig. 4A and Fig. Figure 4B are schematic representations, each showing an example of the replaced concatenated sequence data.

[0055] Fig. 4A and Fig. Figure 4B also shows examples of the concatenated sequence data derived from lines 102a and 102b of the document in Fig. 2 FEMA sheet 101 shown were extracted.

[0056] As in Fig. As illustrated in Figure 4A, the multiple tokens are replaced by the special tokens [MASK] in the concatenated sequence data, where the sequence data SDa, extracted from line 102a of FEMA sheet 101, and the sequence data SDb, extracted from line 102b of the same, are concatenated in the order of line 102a followed by line 102b.

[0057] In Fig. 4A In sequence data SDa, the data in which one or more tokens are replaced by the special tokens [MASK] are designated as SDa#1, while in sequence data SDb, the data in which one or more tokens are replaced by the special tokens [MASK] are designated as SDb#1. However, the replacement of the tokens by the special tokens is not limited to this example and only needs to be performed randomly.

[0058] Meanwhile, as in Fig. 4B shows multiple tokens replaced by the special tokens [MASK] in the concatenated sequence data, in which the sequence data SDa extracted from line 102a of FEMA sheet 101 and the sequence data SDb extracted from its line 102b are concatenated in the order of line 102b followed by line 102a.

[0059] In Fig. In sequence data SDa, data in which one or more tokens are replaced by the special tokens [MASK] are designated as SDa#2, while in sequence data SDb, data in which one or more tokens are replaced by the special tokens [MASK] are designated as SDb#2. However, the replacement of tokens by the special tokens is not limited to this example and only needs to be performed randomly.

[0060] The integrated feature training data is set up by pairing the above-replaced linked sequence data as input data with the above-labeled linked sequence data as output data.

[0061] In other words, the training data generation unit 111 extracts two consecutive rows from the multitude of rows on the FMEA sheet: a first row and a second row following the first. The training data generation unit 111 then specifies a multitude of first tokens by performing a morphological analysis of the operational process information and the risk set contained in the first row, and arranges these first tokens, thereby generating first sequence data. The training data generation unit 111 also specifies a multitude of second tokens by performing a morphological analysis of the operational process information and the risk set contained in the second row, and arranges these second tokens, thereby generating second sequence data.The training data generation unit 111 concatenates the first sequence data and the second sequence data in the order of the first sequence data, followed by the second sequence data, thus generating first concatenate sequence data. Additionally, the training data generation unit 111 concatenates the first sequence data and the second sequence data in the order of the second sequence data, followed by the first sequence data, thus generating second concatenate sequence data. The training data generation unit 111 modifies one or more tokens, randomly selected from the multitude of first and second tokens contained in the first concatenate sequence data, into a mask token(s) designed to obscure the meaning of the token(s), thereby generating first input data.Furthermore, the training data generation unit 111 modifies one or more tokens, randomly selected from the multitude of first and second tokens contained in the second chained sequence data, into a mask token(s), thereby generating second input data. Then, the training data generation unit 111 appends a positive example label to the first chained sequence data, generating first labeled chained sequence data as first output data, which is the output data for the first input data. Additionally, the training data generation unit 111 appends a negative example label to the second chained sequence data, generating second labeled chained sequence data as second output data, which is the output data for the second input data.Thus, the training data generation unit 111 generates the integrated feature training data, which consists of training data that includes the first input and output data as well as the second input and output data.

[0062] Next, the correspondence relationship training data will be described.

[0063] The training data generation unit 111 extracts at least the operational process information, which is the information stored in process column 101c, and the risk set, which is the information stored in risk column 101k, from the stored information in a row unit of the former case sheet. It should be noted that if cells are empty, the corresponding information is added and extracted as text. The product information stored in product column 101a and function column 101b is also extracted.

[0064] The training data generation unit 111 then performs morphological analysis of the text extracted from a line to divide the text into tokens; each token represents the smallest unit of meaning. The training data generation unit 111 then arranges the divided tokens in the order in which they appear in the corresponding text to form a string as sequence data.

[0065] The training data generation unit 111 sets a portion of the sequence data, with the exception of the risk sets, as leaf structure information.

[0066] The training data generation unit 111 then combines the leaf structure information in one sequence data set with a risk set in the other sequence data set to form combined sequence data. This process assumes that each of the risk sets from all other sequence data sets is combined with the leaf structure information from the one sequence data set.

[0067] The training data generation unit 111 provides the above-mentioned sequence data and combined sequence data as input sequence data, which are the input data for learning correspondence relationships.

[0068] The training data generation unit 111 attaches a positive example label to the sequence data and also attaches a negative example label to the combined sequence data.

[0069] Then the training data generation unit 111 sets the labeled sequence data and the labeled combined sequence data as output sequence data, which are the output data for learning correspondence relationships.

[0070] The input sequence data and output sequence data mentioned above constitute the correspondence relationship training data.

[0071] As described above, the training data generation unit 111 generates the correspondence relationship training data, i.e., the training data containing the positive example and the negative example. The positive example is a combination of the operational process information contained in one of the multiple rows in the former case sheet and the risk set contained in that single row. The negative example is a combination of the operational process information contained in that single row and a risk set contained in a row other than that single row.

[0072] The Integrated Feature Learning Unit 112 performs the learning process using the Integrated Feature Training Data generated by the Training Data Generation Unit 111 to create an Integrated Feature Model, which is a machine learning model. The generated Integrated Feature Model is stored in the Storage Unit 120.

[0073] For example, the Integrated Feature Learning Unit 112 generates the Integrated Feature Model by learning the tokens from the Integrated Feature Training Data before they are replaced by the mask tokens, and also learning the order of an arrangement of the first sequence data and the second sequence data.

[0074] For machine learning in the Integrated Feature Learning Unit 112, known methods described in the literature below can be used.

[0075] Literature: Jacob Devlin, Ming-Wei Chang, Kenton Lee, Kristina Toutanova, “Bert: Pre-training of deep bidirectional transformers for language understanding”, arXiv preprint arXiv: 1810. 04805, 2018

[0076] Fig. Figure 5 is a schematic representation to explain machine learning in the Integrated Feature Learning Unit 112.

[0077] As in Fig. As shown in Figure 5, the Integrated Feature Learning Unit 112, using the replaced linked sequence data InD#1 as input data and the labeled linked sequence data OuD#1 as output data, learns a word task that identifies original tokens for specific tokens in the replaced linked sequence data InD#1, enabling it to learn the features of the FMEA sheet in the row direction. If the order of the sequence data contained in the replaced linked sequence data InD#1 matches the order of the operating processes, the Integrated Feature Learning Unit 112 learns this as the positive example. If it does not match, the Integrated Feature Learning Unit 112 learns this as the negative example. Consequently, it can learn the features of the FMEA sheet in the column direction.

[0078] By machine learning these two tasks in a multitasking manner, the Integrated Feature Learning Unit 112 can procure the feature sizes as neural network parameters in which the structural features of the FMEA sheet and the linguistic features within the FMEA sheet are integrated.

[0079] Back to Fig. 1: The correspondence relationship learning unit 113 performs learning by using the correspondence relationship training data generated by the training data generation unit 111 to create the correspondence relationship model, which is a machine learning model. The generated correspondence relationship model is stored in the storage unit 120.

[0080] For example, the correspondence relationship learning unit 113 generates the correspondence relationship model by learning at least the correspondence relationship between the business process information and the risk set using the correspondence relationship training data.

[0081] Fig. 6 and Fig. Figure 7 are schematic representations to explain machine learning in the correspondence relationship learning unit 113.

[0082] As in Fig. As shown in Figure 6, the correspondence relationship learning unit 113 determines that if the input sequence data InD#2 matches the output sequence data OuD#2, which is labeled with a positive example, this is a positive example "1". In contrast, the correspondence relationship learning unit 113 determines, as shown in Fig. 7 is shown when the input sequence data InD#3 matches the output sequence data OuD#3, which is labelled with a negative example, this being represented as a negative example “0”.

[0083] Through the learning process described above, the correspondence between the structure of the FMEA sheet and the risk set can be learned. In particular, the correspondence between the operational process information and the risk set can be learned.

[0084] The correspondence relationship learning unit 113 can improve the learning accuracy of the correspondence relationship by using the parameters of the integrated feature model as initial value parameters of the neural network used in machine learning.

[0085] Back to Fig. 1: The storage unit 120 stores data and programs required for processing in the FMEA sheet preparation support unit 100.

[0086] The storage unit 120 comprises a former case sheet storage unit 121, an integrated feature model storage unit 122, a document storage unit 123, and a correspondence relationship model storage unit 124.

[0087] The Past Sheet Storage Unit 121 stores past sheets, i.e., FMEA sheets created in the past.

[0088] The Integrated Feature Model Storage Unit 122 stores the Integrated Feature Model generated by the Integrated Feature Learning Unit 112.

[0089] The document storage unit 123 stores a large number of documents that are searched for when creating the FMEA sheet.

[0090] The correspondence relationship model storage unit 124 stores the correspondence relationship model generated by the correspondence relationship learning unit 113.

[0091] The search processing unit 130 performs processing to search for information required to create the FMEA sheet.

[0092] The search processing unit 130 comprises an information retrieval unit 131, a correspondence relationship determination unit 132 and a display processing unit 133.

[0093] Information retrieval unit 131 procures search information, i.e., information for the search.

[0094] The information retrieval unit 131 obtains the search information by receiving input from a user via the input unit 140.

[0095] In the first embodiment, it is assumed that the search information includes at least operational process information, and it is described here as comprising both product information and operational process information. For this reason, the search information is also referred to as search operational process information, i.e., operational process information for the search.

[0096] The correspondence relationship determination unit 132 determines information required for the creation of the FMEA sheet by using the search information, the document stored in the document storage unit 123, and the correspondence relationship model stored in the correspondence relationship model storage unit 124.

[0097] For example, the correspondence relationship determination unit 132 generates a multitude of parts of search sequence data by concatenating the respective multitude of sentences contained in each of the documents stored in the document storage unit 123 with the search information.

[0098] Next, the correspondence relationship determination unit 132 obtains a score for each of the multitude of parts of the search sequence data by inputting each part of the search sequence data into the correspondence relationship model.

[0099] The Correspondence Relationship Determination Unit 132 then sums the obtained scores for each document and specifies a document with the highest summed value as reference information.

[0100] Fig. 8A and Fig. 8B are schematic representations to explain the processing carried out in the correspondence relationship determination unit 132.

[0101] As in Fig. As shown in Figure 8A, the correspondence relationship determination unit 132 generates search sequence data SID1, SID2 and SID3 by concatenating a multitude of records SE1, SE2 and SE3, which are contained in a document stored in the document storage unit 123, with search information SI.

[0102] As in Fig. As shown in Figure 8B, the correspondence relationship determination unit 132 next obtains scores for the respective search sequence data SID1, SID2 and SID3 by inputting each of these search sequence data into the correspondence relationship model.

[0103] The Correspondence Relationship Determination Unit 132 then sums the obtained scores for each document and specifies the document with the highest summed value as the reference information.

[0104] In other words, the correspondence relationship determination unit 132 generates the multitude of pieces of search sequence data by adding each of the multitude of sentences contained in the multitude of documents to the search operation process information. As a result, the correspondence relationship determination unit 132 sums a multitude of scores obtained by inputting the multitude of pieces of search sequence data into the correspondence relationship model for each of the multitude of documents, each containing the multitude of sentences. Subsequently, the correspondence relationship determination unit 132 specifies the document with the highest summed score as the reference information.

[0105] The display processing unit 133 generates a reference screen image that shows the reference information and displays the reference screen image on the display unit 150.

[0106] The input unit 140 receives input from the user.

[0107] The display unit 150 shows various screen images.

[0108] The FMEA sheet creation support device 100 described above can be implemented by a computer 15, as described in Fig. 9 is shown.

[0109] The computer 15 includes a memory 10, a processor 11, e.g. a central processing unit (CPU), an auxiliary storage device 12, e.g. a hard disk drive (HDD) or a solid-state drive (SSD), a display device 13, e.g. a display, and an input device 14, e.g. a mouse or a keyboard.

[0110] For example, part or all of the preprocessing unit 110 and the search processing unit 130 can consist of the main memory 10 and the processor 11, such as a central processing unit (CPU) that executes a program stored in the main memory 10.

[0111] Such a program can be delivered over a network or it can be delivered and recorded on a recording medium. That is to say, such a program can, for example, be delivered as a software product.

[0112] The storage unit 120 can be implemented by the auxiliary storage device 12.

[0113] The display unit 150 can be implemented by the display device 13.

[0114] The input unit 140 can be implemented by the input device 14.

[0115] As described above, according to the first embodiment, information about a document describing a useful set can be made available to a user when the user creates an FMEA sheet. Second embodiment

[0116] Fig. Figure 10 is a block diagram which schematically represents a configuration of an FMEA sheet preparation support device 200 according to a second embodiment.

[0117] The FMEA sheet creation support device 200 comprises a preprocessing unit 210, the storage unit 120, the search processing unit 130, the input unit 140 and the display unit 150.

[0118] The storage unit 120, the search processing unit 130, the input unit 140 and the display unit 150 in the FMEA page preparation support device 200 according to the second embodiment are each the same as the storage unit 120, the search processing unit 130, the input unit 140 and the display unit 150 in the FMEA page preparation support device 100 according to the first embodiment.

[0119] The preprocessing unit 210 operates as a learning unit, learning learning models that are used in the search processing unit 130.

[0120] The preprocessing unit 210 comprises the training data generation unit 111, the integrated feature learning unit 112, the correspondence relationship learning unit 113 and a training data extension unit 214.

[0121] The training data generation unit 111, the integrated feature learning unit 112 and the correspondence relationship learning unit 113 of the preprocessing unit 210 in the second embodiment are each the same as the training data generation unit 111, the integrated feature learning unit 112 and the correspondence relationship learning unit 113 of the preprocessing unit 110 in the first embodiment.

[0122] It should be noted that the training data generation unit 111 provides the generated integrated feature training data to the training data extension unit 214.

[0123] The Integrated Feature Learning Unit 112 performs the learning using the Integrated Feature Training Data, which has been augmented by the Training Data Extension Unit 214.

[0124] The training data extension unit 214 sets at least one token contained in the leaf structure information as a search query in the labeled linked sequence data contained in the integrated feature training data generated by the training data generation unit 111. The training data extension unit 214 then specifies a record containing such a token by searching the documents stored in the document storage unit 123.

[0125] The training data extension unit 214 then generates new integrated feature training data from the integrated feature training data by replacing the specified set with a risk set of leaf structure information, including the search query. By adding the newly generated integrated feature training data to the integrated feature training data generated by the training data generation unit 111, the training data extension unit 214 extends the integrated feature training data generated by the training data generation unit 111.

[0126] Fig. Figure 11 is a schematic representation illustrating an example of extending the integrated feature training data.

[0127] In Fig. Figure 11 shows an example where new replaced linked sequence data InD#3 and new labeled linked sequence data OuD#3 are derived from the data in Fig. The 5 shown replaced linked sequence data InD#1 and the labeled linked sequence data OuD#1 are generated.

[0128] As in Fig. Figure 11 shows that the risk set L1, which is contained in the replaced linked sequence data InD#1 and the labeled linked sequence data OuD#1, is shown in the new replaced linked sequence data InD#3 and the new labeled linked sequence data OuD#3. Fig. 5, replaced by a set L1T specified by searching the documents stored in document storage unit 123 using at least one of “A”, “A1”, “A11”, “#1”, “S1”, “%1” and “&1” as the search query.

[0129] In Fig. In case 11, the risk set L1T is replaced by the set L1T because the set L1T was specified by at least one search query of “$1”, “%1”, and “&1”, which are only present in the first part of the sequence data. However, if the set was specified by at least one search query of “A”, “A1”, “A11”, and “#1”, both of which are present in the sequence data, each of the risk sets L1 and L2 is replaced by the specified set.

[0130] If the sentence is specified by at least one of the search queries “$2”, “%2” and “&2”, which are only contained in the last part of the sequence data, only the risk sentence L2 can be replaced by the specified sentence.

[0131] In other words, in the second embodiment, the training data extension unit 214 generates extended integrated feature training data from the integrated feature training data by performing at least one of the following processes: replacing the risk set contained in the first sequence data with the set obtained by searching the multitude of documents using the operational process information contained in the first sequence data, and replacing the risk set contained in the second sequence data with the set obtained by searching the multitude of documents using the operational process information contained in the second sequence data.

[0132] Then, the Integrated Feature Learning Unit 112 also learns the extended Integrated Feature Training data to generate the Integrated Feature Model.

[0133] As described above, according to the second embodiment, in addition to a combination of the sheet structure information and the risk set, a combination of the sheet structure information and a set contained in the document in question can be used as new training data, thereby extending the training data. Third embodiment

[0134] Fig. Figure 12 is a block diagram which schematically represents a configuration of an FMEA sheet preparation support device 300 according to a third embodiment.

[0135] The FMEA sheet creation support device 300 comprises a preprocessing unit 310, the storage unit 120, the search processing unit 130, the input unit 140 and the display unit 150.

[0136] The storage unit 120, the search processing unit 130, the input unit 140 and the display unit 150 in the FMEA page preparation support device 300 according to the third embodiment are each the same as the storage unit 120, the search processing unit 130, the input unit 140 and the display unit 150 of the FMEA page preparation support device 100 according to the first embodiment.

[0137] The preprocessing unit 310 operates as a learning unit, learning learning models that are used in the search processing unit 130.

[0138] The preprocessing unit 310 comprises the training data generation unit 111, the integrated feature learning unit 112, the correspondence relationship learning unit 113 and a sequence addition unit 315.

[0139] The training data generation unit 111, the integrated feature learning unit 112 and the correspondence relationship learning unit 113 of the preprocessing unit 310 in the third embodiment are each the same as the training data generation unit 111, the integrated feature learning unit 112 and the correspondence relationship learning unit 113 of the preprocessing unit 110 in the first embodiment.

[0140] However, the Integrated Feature Learning Unit 112 also performs learning by using as input the linked sequence data added by the Sequence Addition Unit 315.

[0141] The sequence addition unit 315 adds additional linked sequence data, i.e., linked sequence data that stores information specifying the contents of each position, as input data for the integrated feature training data generated by the training data generation unit 111.

[0142] Fig. Figure 13 is a schematic representation that provides an example of additional linked sequence data.

[0143] The additional linked sequence data stores information that specifies the content of each piece of information in the replaced linked sequence data, which is used as the input data for the integrated feature training data.

[0144] For example, in Fig. 13 “R1” is information that specifies “product identification information”, “R2” is information that specifies “functional identification information”, “R3” is information that specifies a “large process” that represents the classification of the operating process, “R4” is information that specifies a “medium process” that represents the classification of the operating process, “R5” is information that specifies a “small process” that represents the classification of the operating process, and “R6” is information that specifies a “risk set”.

[0145] In other words, in the third embodiment, the sequence addition unit 315 adds additional sequence data to each of the first and second input data, specifying the contents of the plurality of first and second tokens before they are changed into the mask tokens.

[0146] According to the third embodiment, the structural information on the FMEA sheet can be explicitly provided by adding the additional linked sequence data as the input data during the learning performed by the Integrated Feature Learning Unit 112, thereby improving the accuracy of the machine learning. Fourth embodiment

[0147] Fig. Figure 14 is a block diagram which schematically represents a configuration of an FMEA sheet preparation support device 400 according to a fourth embodiment.

[0148] The FMEA sheet creation support device 400 comprises a preprocessing unit 410, the storage unit 420, a search processing unit 430, the input unit 140 and the display unit 150.

[0149] The input unit 140 and the display unit 150 in the FMEA page preparation support device 400 according to the fourth embodiment are each the same as the input unit 140 and the display unit 150 in the FMEA page preparation support device 100 according to the first embodiment.

[0150] The preprocessing unit 410 operates as a learning unit, learning learning models that are used in the search processing unit 130.

[0151] The preprocessing unit 410 comprises a training data generation unit 411, the integrated feature learning unit 112, the correspondence relationship learning unit 113 and an evaluation learning unit 416.

[0152] The integrated feature learning unit 112 and the correspondence relationship learning unit 113 of the preprocessing unit 410 in the fourth embodiment are each the same as the integrated feature learning unit 112 and the correspondence relationship learning unit 113 of the preprocessing unit 110 in the first embodiment.

[0153] The training data generation unit 411 not only generates the integrated feature training data and the correspondence relationship training data as in the first embodiment, but also generates evaluation training data for learning in the evaluation learning unit 416.

[0154] For example, the training data generation unit 411 obtains product information, operational process information, a risk set, and evaluation information from each line of the former case sheet stored in the former case sheet storage unit 121, and it generates evaluation training data that has the product information, operational process information, and risk set as input data and also has the product information, operational process information, risk set, and evaluation information as output data.

[0155] The generated evaluation training data is provided to evaluation learning unit 416.

[0156] The evaluation learning unit 416 generates an evaluation model, which is a machine learning model, by learning using the evaluation training data generated by the training data generation unit 411. The generated evaluation model is stored in the storage unit 420.

[0157] The evaluation learning unit 416 can use the parameters of the integrated feature model as initial value parameters of the neural network used in machine learning, thus improving the learning accuracy of the correspondence relationship.

[0158] The storage unit 420 stores data and programs required for processing in the FMEA sheet preparation support unit 400.

[0159] The storage unit 420 comprises the Former Case Sheet storage unit 121, the Integrated Feature Model storage unit 122, the Document storage unit 123, the Correspondence Relationship Model storage unit 124, and an Evaluation Model storage unit 425.

[0160] The former flip-sheet storage unit 121, the integrated feature model storage unit 122, the document storage unit 123 and the correspondence relationship model storage unit 124 of the storage unit 420 in the fourth embodiment are each the same as the former flip-sheet storage unit 121, the integrated feature model storage unit 122, the document storage unit 123 and the correspondence relationship model storage unit 124 of the storage unit 120 in the first embodiment.

[0161] The evaluation model storage unit 425 stores the evaluation model generated by the evaluation learning unit 416.

[0162] The search processing unit 430 performs processing to search for information required to create the FMEA sheet.

[0163] The search processing unit 430 comprises the information retrieval unit 131, a correspondence relationship determination unit 432 and a display processing unit 433.

[0164] The information retrieval unit 131 of the search processing unit 430 in the fourth embodiment is the same as the information retrieval unit 131 of the search processing unit 130 in the first embodiment.

[0165] The correspondence relationship determination unit 432 specifies as the reference information a document that best matches the search information by using the search information, the document stored in the document storage unit 123, and the correspondence relationship model stored in the correspondence relationship model storage unit 124, as in the first embodiment.

[0166] In addition, the correspondence relationship determination unit 432 generates evaluation search sequence data by concatenating one or more sentences contained in the document specified as reference information with the search information.

[0167] The correspondence relationship determination unit 432 can link all sentences contained in the document specified as reference information with the search information. Alternatively, a predefined number of sentences, selected in order of highest score, are linked with the search information to generate the evaluation search sequence data.

[0168] The correspondence relationship determination unit 432 then determines the evaluation information in the evaluation search sequence data by entering the evaluation search sequence data into the evaluation model storage unit 425. The determined evaluation information is made available to the display processing unit 433.

[0169] The display processing unit 433 generates a reference screen image that shows the reference information and the evaluation information and causes the display unit 150 to display the reference screen image.

[0170] In other words, in the fourth embodiment, the training data generation unit 411 generates the evaluation training data, which consists of the operational process information contained in one of the multiple rows in the former case sheet as input data and the operational process information and the risk set contained in that single row as output data. The evaluation learning unit 416 then generates the evaluation model by learning the risk set from the operational process information using the evaluation training data.

[0171] Furthermore, the correspondence relationship determination unit 432 generates evaluation determination sequence data by adding one or more records selected from the multitude of records contained in the document specified as the reference information to the search operation process information. It determines the evaluation for the evaluation determination sequence data by inputting the generated evaluation determination sequence data into the evaluation model. The display processing unit 433 then also displays the determined evaluation on the screen.

[0172] As described above, according to the fourth embodiment, the evaluation information can be determined using the document specified as reference information.

[0173] In the first through fourth embodiments described above, the correspondence relationship learning model or the evaluation model is learned by using the parameters of the integrated feature model, learned in Integrated Feature Learning Unit 112, as initial parameters. However, the first through fourth embodiments are not limited to this example. For instance, if a sufficient amount of training data can be provided to learn the correspondence relationship model or the evaluation model, the correspondence relationship model or the evaluation model can be learned without the integrated feature model.

[0174] Although the FMEA sheet-making support devices 100 to 400 described above have both learning and inference functions, the first to fourth embodiments are not limited to these examples.

[0175] A learning unit (not shown) may, for example, consist of a part that performs a learning function and includes, for example, preprocessing units 110-410, storage units 120, 420, input unit 140, and display unit 150. An inference unit or administrative sheet creation support unit (not shown) may consist of a part that performs an inference function and includes, for example, storage units 120, 420, search processing units 130, 430, input unit 140, and display unit 150.

[0176] The 120, 420 storage unit can be contained in an external device. REFERENCE MARK LIST 100, 200, 300, 400 FMEA sheet preparation support device 110, 210, 310, 410 Pre-processing unit 111,411 Training Data Generation Unit 112 Integrated Feature Learning Unit 113 Correspondence Relationship Learning Unit 214 Training Data Extension Unit 315 Sequence Addition Unit 416 Evaluation learning unit 120, 420 storage unit 121 Former Fall-Sheet Storage Unit 122 Integrated Feature Model Storage Unit 123 Document storage unit 124 Correspondence Relationship Model Storage Unit 425 Evaluation Model Storage Unit 130,430 search processing unit 131 Information gathering unit 132,432 Correspondence Relationship Determination Unit 133,433 Display processing units 140 input unit 150 display units QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] JP 2011-8355

[0005]

Claims

[1] Learning institution, comprising: a former case sheet storage unit configured to store a former case sheet created in the past as a management sheet, the management sheet including a plurality of rows, each of the plurality of rows including at least operation process information indicating one operation process included in a plurality of operation processes and a risk set indicating information about a risk in the one operation process; a training data generation unit configured to generate correspondence relationship training data that is training data including a positive example and a negative example, wherein the positive example is a combination of the operation process information included in one of the plurality of rows in the former case sheet and the risk set included in the one row, wherein the negative example is a combination of the operation process information included in the one row and the risk set included in a row different from the one row; and a correspondence relationship learning unit configured to generate a correspondence relationship model by learning a correspondence relationship between the operation process information and the risk set using the correspondence relationship training data. [2] Learning device according to claim 1, wherein the plurality of rows are arranged in the order in which the plurality of operating processes are performed, wherein the training data generation unit is configured to: extracting two consecutive lines from the plurality of lines, the two lines being a first line and a second line following the first line; Specifying a plurality of first tokens, which are a plurality of tokens, by performing a morphological analysis of the operation process information and the risk set included in the first row, and then arranging the plurality of first tokens, thereby generating first sequence data; Specifying a plurality of second tokens, which are a plurality of tokens, by performing a morphological analysis of the operation process information and the risk set included in the second line, and then arranging the plurality of second tokens, thereby generating second sequence data; concatenating the first sequence data and the second sequence data in the order of the first sequence data followed by the second sequence data, thereby generating first concatenated sequence data; concatenating the first sequence data and the second sequence data in the order of the second sequence data followed by the first sequence data, thereby generating second concatenated sequence data; Changing one or more tokens randomly selected from the plurality of first and second tokens included in the first concatenated sequence data into a mask token configured to obscure the meaning of the token, thereby generating first input data; Changing one or more tokens randomly selected from the plurality of first and second tokens included in the second concatenated sequence data into the mask token, thereby generating second input data; appending a positive example label to the first concatenated sequence data, thereby generating first labeled concatenated sequence data as first output data, which is output data for the first input data; appending a negative example label to the second concatenated sequence data, thereby generating second labeled concatenated sequence data as second output data that is output data for the second input data; and Generating integrated feature training data, which is training data having the first input data and the first output data and the second input data and the second output data, wherein the learning device further comprises an integrated feature learning unit configured to generate an integrated feature model by learning the token before replacement by the mask token from the integrated feature training data and also learning the order of an arrangement of the first sequence data and the second sequence data, and wherein the correspondence relationship learning unit is configured to learn the correspondence relationship model by using a parameter of the integrated feature model as an initial parameter. [3] Learning device according to claim 2, further comprising: a document storage unit configured to store a plurality of documents; and a training data expansion unit configured to generate expanded integrated feature training data from the integrated feature training data by performing at least one of two processes, the two processes comprising replacing the risk set included in the first sequence data with a set acquired by searching the plurality of documents using the operation process information included in the first sequence data, and replacing the risk set included in the second sequence data with a set acquired by searching the plurality of documents using the operation process information included in the second sequence data, wherein the integrated feature learning unit is also configured to learn the extended integrated feature training data to generate the integrated feature model. [4] Learning device according to claim 2, further comprising: a sequence addition unit configured to add additional sequence data indicating contents of the plurality of first and second tokens to each of the first and second input data before being changed into the mask tokens. [5] The learning device according to any one of claims 1 to 4, wherein the training data generation unit is configured to generate evaluation training data which is training data including the operation process information included in one of the plurality of rows in the former case sheet as input data and the operation process information included in the one row and the risk set as output data, the learning device further comprising an evaluation learning unit configured to generate an evaluation model by learning the risk set from the operation process information using the evaluation training data. [6] Administrative sheet preparation support facility, comprising: a correspondence relationship model storage unit configured to store a correspondence relationship model for a former case sheet created in the past as a management sheet, the management sheet including a plurality of rows, each of the plurality of rows including at least operation process information indicating one operation process included in a plurality of operation processes and a risk set indicating information about a risk in the one operation process, the correspondence relationship model being generated by learning a correspondence relationship between the operation process information and the risk set using the correspondence relationship training data, which is data including a positive example and a negative example,wherein the positive example is a combination of the operational process information contained in one of the plurality of rows and the risk set contained in the one row, wherein the negative example is a combination of the operational process information contained in the one row and a risk set contained in a row different from the one row;, a document storage unit configured to store a plurality of documents; an information acquisition unit configured to acquire search operation process information, which is operation process information for the search; a correspondence relationship determination unit configured to generate a plurality of pieces of search sequence data by adding each of a plurality of sentences included in the plurality of documents to the search operation process information, sum a plurality of scores obtained by inputting the plurality of pieces of search sequence data into the correspondence relationship model for each of the plurality of documents each containing the plurality of sentences, and then specify a document having a highest summed score as reference information; and a display processing unit configured to generate a screen image for displaying the reference information. [7] The management sheet creation support device according to claim 6, further comprising: an evaluation model storage unit configured to store an evaluation model that is a learning model generated by learning the risk set from the operation process information using evaluation training data that is training data containing the operation process information contained in the one row as input data and containing the operation process information contained in the one row and the risk set as output data, wherein the correspondence relationship determination unit is configured to generate evaluation determination sequence data by adding to the search operation process information one or more sentences selected from the plurality of sentences included in the document specified as the reference information, and to determine an evaluation for the evaluation determination sequence data by inputting the evaluation determination sequence data into the evaluation model, and wherein the display processing unit is also configured to display the determined evaluation on the screen image. [8] Program that causes a computer to act as: a former case sheet storage unit configured to store a former case sheet created in the past as a management sheet, the management sheet including a plurality of rows, each of the plurality of rows including at least operation process information indicating one operation process included in a plurality of operation processes and a risk set indicating information about a risk in the one operation process; a training data generation unit configured to generate correspondence relationship training data that is training data including a positive example and a negative example, wherein the positive example is a combination of the operation process information included in one of the plurality of rows in the former case sheet and the risk set included in the one row, wherein the negative example is a combination of the operation process information included in the one row and the risk set included in a row different from the one row; and a correspondence relationship learning unit configured to generate a correspondence relationship model by learning a correspondence relationship between the operation process information and the risk set using the correspondence relationship training data. [9] Program that causes a computer to act as: a correspondence relationship model storage unit configured to store a correspondence relationship model for a former case sheet created in the past as a management sheet, the management sheet including a plurality of rows, each of the plurality of rows including at least operation process information indicating one operation process included in a plurality of operation processes and a risk set indicating information about a risk in the one operation process, the correspondence relationship model being generated by learning a correspondence relationship between the operation process information and the risk set using the correspondence relationship training data, which is data including a positive example and a negative example,wherein the positive example is a combination of the operational process information contained in one of the plurality of rows and the risk set contained in the one row, wherein the negative example is a combination of the operational process information contained in the one row and the risk set contained in a row different from the one row;, a document storage unit configured to store a plurality of documents; an information acquisition unit configured to acquire search operation process information, which is operation process information for the search; a correspondence relationship determination unit configured to generate a plurality of pieces of search sequence data by adding each of a plurality of sentences included in the plurality of documents to the search operation process information, sum a plurality of scores obtained by inputting the plurality of pieces of search sequence data into the correspondence relationship model for each of the plurality of documents each containing the plurality of sentences, and then specify a document having a highest summed score as reference information; and a display processing unit configured to generate a screen image for displaying the reference information. [10] Learning methods, including: Generating correspondence relationship training data for a former case sheet created in a past as a management sheet, wherein the management sheet includes a plurality of rows, each of the plurality of rows comprising at least operation process information indicating an operation process included in a plurality of operation processes and a risk set indicating information about a risk in the one operation process, wherein the correspondence relationship training data is training data including a positive example and a negative example, wherein the positive example is a combination of the operation process information included in one of the plurality of rows in the past case sheet and the risk set included in the one row,where the negative example is a combination of the operational process information contained in one row and the risk set contained in a row different from the one row; and, Generating a correspondence relationship model by learning a correspondence relationship between the operation process information and the risk set using the correspondence relationship training data. [11] Administrative sheet preparation support procedure, comprising: Obtaining search operation process information, which is operation process information for search; Generating a plurality of pieces of search sequence data by adding each of a plurality of sentences included in the plurality of documents to the search operation process information; Generating a correspondence relationship model for a former case sheet created in the past as a management sheet, the management sheet including a plurality of rows, each of the plurality of rows including at least operation process information indicating one operation process included in a plurality of operation processes and a risk sentence indicating information about a risk in the one operation process; wherein the correspondence relationship model is generated by learning a correspondence relationship between the operation process information and the risk sentence using the correspondence relationship training data, which is data including a positive example and a negative example;wherein the positive example is a combination of the operation process information contained in one of the plurality of rows in the past case sheet and the risk sentence contained in the one row, wherein the negative example is a combination of the operation process information contained in the one row and the risk sentence contained in a row different from the one row, then summing a plurality of scores obtained by inputting the plurality of pieces of search sequence data into the correspondence relationship model for each of the plurality of documents each containing the plurality of sentences, and then specifying a document with a highest summed score as reference information; and, Generate a screen image to display the reference information.

Images