Information processing program, information processing device, and information processing method

Abstract

A sentence (104) conforming to a specific part of speech is selected from sentences extracted from non-standard text data (101) in a manufacturing process of a product, and a machine learning model is trained. The machine learning model outputs a determination result (107) corresponding to an input feature amount and a feature amount (108) contributing to the determination using training data in which the input feature amount including the selected sentence (104) and the configuration information (105) of the product and tag information (106) indicating the determination result relating to the product are associated with each other.

Description

Technical Field

[0001] The present invention relates to an information processing program, an information processing device, and an information processing method. Background Art

[0002] To extract important sentences from non-standard text data such as memos, general machine learning, explainable AI (XAI), and natural language processing (ie, morpheme analysis) are sometimes used.

[0003] Explainable AI (XAI) is AI that can explain the process up to the prediction result or classification result. Natural language processing is the process of extracting words by segmenting the memo according to part of speech.

[0004] Patent Document 1: Japanese Patent Application Laid-Open No. 2022-70766

[0005] However, in general machine learning, the output is only a prediction, which can be difficult to accept in the field without the persuasiveness of the prediction. In the field, there is generally a accumulation of data such as memos that cannot be used for learning, and there is a concern that the input content cannot be formatted.

[0006] Although it is envisioned that the content recorded on site can be decomposed through natural language processing and absorbed into feature quantities to improve accuracy and show a basis for more detailed predictions, there is also the possibility of showing keywords that have no direct relationship as a basis, which may cause a concern that the readability of the description will be reduced. Summary of the Invention

[0007] In one aspect, the goal is to improve the readability of output data when non-standard text data is fed into a machine learning model.

[0008] On one aspect, the information processing program causes the computer to execute: selecting sentences that conform to specific parts of speech from sentences extracted from non-standard text data in the manufacturing process of a product; and training a machine learning model, which uses training data that establishes a correspondence between input feature quantities containing the selected sentences and product composition information, and label information representing judgment results related to the product, to process the output of judgment results corresponding to the input feature quantities and feature quantities that contribute to the judgment.

[0009] In one aspect, the readability of output data when non-standard text data is input into a machine learning model can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a graph illustrating the contribution of sentences to defect identification in the related examples and embodiments.

[0011] Figure 2 This is a block diagram schematically showing an example of the hardware configuration of the information processing device according to the embodiment.

[0012] Figure 3 It indicates schematically Figure 2 A block diagram showing an example of the software configuration of the information processing device shown.

[0013] Figure 4 It is an explanation Figure 2 FIG. 1 is a flowchart of a machine learning process in the learning phase of the information processing device shown.

[0014] Figure 5 It is an explanation Figure 2 FIG. 1 is a flowchart of a machine learning process in the prediction phase of an information processing device shown.

[0015] Figure 6 This is an example Figure 4 as well as Figure 5 Table showing regional knowledge in part-of-speech selection.

[0016] Figure 7 This is an example Figure 4 as well as Figure 5 Table of regional knowledge in keyword selection shown.

[0017] Figure 8 This is an example Figure 4 as well as Figure 5 Table of data other than the memo shown.

[0018] Figure 9 This is an example Figure 4 as well as Figure 5 A table containing data other than memos containing keywords is shown.

[0019] Figure 10 This is an example Figure 2 FIG. 1 is a diagram showing output data from the information processing device shown.

[0020] Figure 11 This is a diagram illustrating an illustrative AI model that directly outputs explanations in addition to the prediction results in the embodiment.

[0021] Figure 12 This is a diagram illustrating a model for explaining general AI in the embodiment.

[0022] Figure 13 This is a diagram illustrating a model for explaining general AI with respect to image data in the embodiment. DETAILED DESCRIPTION

[0023] [A] Implementation Methods

[0024] An embodiment is described below with reference to the accompanying drawings. However, the embodiments shown below are merely illustrative and do not exclude the application of various modifications or technologies not explicitly described in the embodiments. In other words, the present embodiment can be implemented with various modifications without departing from its main purpose. Furthermore, each figure does not necessarily represent only the components shown in the figure but may also include other functions.

[0025] Figure 1 This is a graph illustrating the contribution of sentences to defect identification in the related examples and embodiments.

[0026] Non-standard text data, such as memos, is sometimes generated on-site at production lines, resulting in inconsistent data formats required for AI predictions. Due to defects occurring on the production line, text data, such as memos, often cannot be formatted. Furthermore, if solutions are not taken, data that cannot be used for learning may accumulate.

[0027] For example, suppose a memo is input as "There is damage in the body 2." If natural sentence processing (i.e., morpheme analysis) is applied to this memo, the result output is ""body" (noun), "2" (noun), "in" (particle), "damaged" (noun), "have" (verb)" as all the parts of speech included in the memo.

[0028] Furthermore, as shown in the example of reference numeral A1, all parts of speech included in the memo are input to the AI model, and even if outputs such as "2" or "important" are given regarding the contribution of each part of speech to fault determination, the persuasiveness is low.

[0029] Therefore, in the embodiment, by utilizing the regional knowledge that only nouns are used for the parts of speech included in the above-mentioned natural sentence processing result, ""main body"" (noun), "2" (noun), "broken" (noun)" can be output.

[0030] Furthermore, in the embodiment, for example, by excluding parts of speech consisting of only one character or only numbers, keywords that are effective for description can be narrowed down. Thus, ""main body" (noun), "damage" (noun)" can be output.

[0031] Therefore, as shown in the embodiment of the figure mark A2, only keywords that are effective for explanation are input into the AI model, and regarding the contribution of each part of speech to the determination of the defect, output data such as "main body" and "damage" are effective for a certain defect determination are generated, and the persuasiveness of the output data is improved.

[0032] Also, suppose that as a memo, there is an input such as "Deviation is large compared to the benchmark". In this case, Figure 1Similar to the example shown, if natural sentence processing is applied to the memo, the result output is ""deviation" (noun), "and (particle)", "benchmark" (noun), "compared to" (particle), "larger" (adjective)" as all the parts of speech included in the memo.

[0033] The parts of speech contained in the above natural sentence processing results can be used in addition to Figure 1 In addition to the above-mentioned nouns, regional knowledge such as adjectives is also used to output "deviation" (noun), "benchmark" (noun), and "large" (adjective)".

[0034] Furthermore, for example, by excluding parts of speech consisting of only one character or only numbers, it is possible to narrow down keywords that are effective in explaining the text. In this example, since the output results utilizing the aforementioned regional knowledge do not include parts of speech consisting of only one character or only numbers, "deviation" (noun), "benchmark" (noun), and "large" (adjective) are output directly.

[0035] Thus, only keywords that are effective for explanation are input into the AI model, and output data such as "deviation", "benchmark", and "large" are generated regarding the contribution of each part of speech to defect determination, thereby improving the persuasiveness of the output data.

[0036] Figure 2 It is a block diagram schematically showing a hardware configuration example of the information processing device 1 according to the embodiment.

[0037] like Figure 1 As shown, the information processing device 1 includes a CPU (Central Processing Unit) 11 , a memory unit 12 , a display control unit 13 , a storage device 14 , an input interface (IF) 15 , an external recording medium processing unit 16 , and a communication IF 17 .

[0038] The memory unit 12 is an example of a storage unit, and examples thereof include Read Only Memory (ROM) and Random Access Memory (RAM). Programs such as the Basic Input / Output System (BIOS) can be written to the ROM of the memory unit 12. The software programs in the memory unit 12 can be read and executed by the CPU 11 as appropriate. The RAM of the memory unit 12 can also be used as temporary storage or working memory.

[0039] The display control unit 13 is connected to and controls a display device 131. The display device 131 may be a liquid crystal display, an organic light-emitting diode (OLED) display, a cathode ray tube (CRT), an electronic paper display, or the like, and displays various information for operators and the like. The display device 131 may also be combined with an input device, such as a touch panel. The display device 131 displays various information for the user of the information processing device 1.

[0040] The storage device 14 is a high IO performance storage device, and for example, Dynamic Random Access Memory (DRAM), SSD (Solid State Drive), Storage Class Memory (SCM), or HDD (Hard Disk Drive) can be used.

[0041] The input IF 15 can be connected to input devices such as a mouse 151 and a keyboard 152, and can control the input devices such as the mouse 151 and the keyboard 152. The mouse 151 and the keyboard 152 are examples of input devices, and an operator performs various input operations via these input devices.

[0042] The external recording medium processing unit 16 is configured to receive a recording medium 160. With the recording medium 160 loaded, the external recording medium processing unit 16 is configured to read information recorded on the recording medium 160. In this embodiment, the recording medium 160 is portable. For example, the recording medium 160 may be a floppy disk, an optical disk, a magnetic disk, a magneto-optical disk, or a semiconductor memory.

[0043] The communication IF 17 is an interface for communicating with an external device.

[0044] The CPU 11 is an example of a processor and is a processing device that performs various controls and calculations. The CPU 11 executes the Operating System (OS) and programs read into the memory unit 12. Figure 3 The various functions are realized as described below. In addition, the CPU 11 may be a multiprocessor including a plurality of CPUs, a multi-core processor including a plurality of CPU cores, or a configuration including a plurality of multi-core processors.

[0045] The device used to control the overall operation of information processing device 1 is not limited to CPU 11. For example, it can also be any one of an MPU, DSP, ASIC, PLD, and FPGA. Furthermore, the device used to control the overall operation of information processing device 1 can also be a combination of two or more of a CPU, MPU, DSP, ASIC, PLD, and FPGA. MPU stands for Micro Processing Unit, DSP stands for Digital Signal Processor, and ASIC stands for Application Specific Integrated Circuit. PLD stands for Programmable Logic Device, and FPGA stands for Field Programmable Gate Array.

[0046] Figure 3 It is a block diagram schematically showing an example of the software configuration of the information processing device 1 according to the embodiment.

[0047] The CPU 11 of the information processing device 1 can function as an analyzing unit 111 , a part-of-speech selecting unit 112 , a part-of-speech sentence extracting unit 113 , a valid part-of-speech extracting unit 114 , and an AI processing unit 115 .

[0048] The parsing unit 111 uses MeCab, an existing open source algorithm, to analyze the on-site memo data 101 (using Figure 4 、 Figure 5 etc.) to implement morpheme analysis and output the article segmentation result 102 (using Figure 4 、 Figure 5 etc. (described later).

[0049] The part-of-speech selection unit 112 is based on the regional knowledge 103 (using Figure 4 、 Figure 5 Selecting nouns and adjectives can indicate the parts of speech required for AI processing.

[0050] The part-of-speech sentence extraction unit 113 extracts the part-of-speech selected by the part-of-speech selection unit 112 from the sentence segmentation result 102 .

[0051] The effective part-of-speech extraction unit 114 extracts parts-of-speech effective for explanation from the parts-of-speech extracted by the part-of-speech sentence extraction unit 113 based on the regional knowledge 103 excluding parts-of-speech consisting of only one character or only numbers, and outputs the extracted parts-of-speech as a keyword list.

[0052] In other words, the part-of-speech sentence extraction unit 113 and the effective part-of-speech extraction unit 114 select sentences that conform to specific parts of speech from sentences extracted from non-standard text data in the product manufacturing process. When selecting sentences, the effective part-of-speech extraction unit 114 may further exclude at least one of sentences with a character length less than a specified length and sentences consisting solely of numerals. When selecting sentences, the part-of-speech sentence extraction unit 113 may select noun sentences and adjective sentences.

[0053] The AI processing unit 115 generates a list of keywords 104 output by the effective part-of-speech extraction unit 114 and data other than the memo 105 (using Figure 4 、 Figure 5 etc., described later), and the correct answer tag 106 (using Figure 4 、 Figure 5 In addition, the AI processing unit 115 uses the learned explainable AI model to perform the prediction result 107 of the cause of the defect (using Figure 4 、 Figure 5 etc.) and instructions 108 (using Figure 4 、 Figure 5 etc. (described later) output.

[0054] In other words, AI processing unit 115 trains a machine learning model using training data that associates input features, including selected phrases and product configuration information, with label information representing the judgment results related to the product. The model then outputs a judgment result corresponding to the input features and features contributing to the judgment. During the training of the machine learning model, AI processing unit 115 can output the cause of the product defect as the judgment result and the contribution of each selected phrase to the determination of the defect as the contributing feature.

[0055] according to Figure 4 The flowchart (steps S1 to S5) shown in FIG. Figure 2 The machine learning process in the learning phase of the information processing device 1 shown in FIG.

[0056] The analyzing unit 111 performs morphological analysis on the on-site memo data 101 using an existing algorithm such as MeCab, and outputs a sentence segmentation result 102 segmented into part-of-speech sentences (step S1 ).

[0057] The part-of-speech selection unit 112 selects parts of speech necessary for describing AI processing, such as nouns and adjectives, based on the regional knowledge 103 (step S2 ).

[0058] The part-of-speech sentence extraction unit 113 extracts the part-of-speech selected by the part-of-speech selection unit 112 from the sentence segmentation result 102 (step S3 ).

[0059] The effective part-of-speech extraction unit 114 extracts parts-of-speech effective for explanation from the parts-of-speech extracted by the part-of-speech sentence extraction unit 113 based on the regional knowledge 103 excluding parts-of-speech consisting of only one character or only numbers, and outputs the extracted parts-of-speech as a keyword list (step S4 ).

[0060] The AI processing unit 115 performs learning of the AI model based on the keyword list 104 output by the effective part-of-speech extraction unit 114, the data other than the memo 105, and the correct answer label 106 (step S5). Then, the machine learning process of the learning phase ends. Figure 8 As will be described later, the data 105 other than the memorandum may be a table (form) including the test date, test production line, and specifications of the product to be identified as the cause of the failure.

[0061] Next, according to Figure 5 The flowchart shown (steps S1 to S4, S51) is Figure 2 The machine learning process in the prediction stage of the information processing device 1 shown in FIG.

[0062] Processing and use in steps S1 to S4 Figure 4 The processing described above is the same.

[0063] The AI processing unit 115 uses the learned explainable AI model to output the prediction result 107 and explanation 108 of the cause of the failure (step S51). Then, the machine learning process of the prediction stage ends. Figure 10 Details of the failure cause prediction result 107 and explanation 108 will be described later.

[0064] Figure 6 This is an example Figure 4 as well as Figure 5 The table of local knowledge 103a in part of speech selection is shown.

[0065] The part-of-speech selection unit 112 can be based on Figure 6 The regional knowledge 103a in the illustrated part-of-speech selection can select the part-of-speech required for AI processing.

[0066] exist Figure 6 In the illustrated area knowledge 103a in the part-of-speech selection, nouns and adjectives are used (see the "0" mark), and verbs are not used (see the "x" mark).

[0067] Therefore, the part-of-speech selection unit 112 selects nouns and adjectives as the parts of speech necessary to describe the AI process.

[0068] Figure 7This is an example Figure 4 as well as Figure 5 Table of regional knowledge in keyword selection shown.

[0069] like Figure 7 As shown, the regional knowledge in keyword selection may include a conditional list 103b and an NG word list 103c.

[0070] The effective part-of-speech extraction unit 114 can extract a part of speech effective for explanation based on at least one of the condition list 103 b and the NG word list 103 c .

[0071] exist Figure 7 In the illustrated condition list 103 b , the condition “length(x)>1” at index 1 indicates that the length of the character is longer than 1, and the condition “x.isdecimal(x)==False” at index 2 indicates that the character is not a digit only character.

[0072] exist Figure 7 The illustrated NG word list 103 c includes the NG word “tap” at index 1 and the NG word “position” at index 2.

[0073] Therefore, the effective part-of-speech extraction unit 114 removes the parts of speech with a character length of less than 1, or characters consisting only of digits, or parts of speech that match the NG word "tap" or "position" from the parts of speech extracted by the part-of-speech sentence extraction unit 113, and outputs them as the keyword list 104.

[0074] Figure 8 This is an example Figure 4 as well as Figure 5 A table of data 105 other than the memo is shown.

[0075] exist Figure 8 In the data other than the memo 105 shown as an example, the test date, test production line, CPU type, memory capacity, exterior color, and display type are registered for the product to be identified as the cause of failure.

[0076] The AI processing unit 115 can use the data 105 other than the memo to perform learning and prediction that can explain the AI model.

[0077] Figure 9 This is an example Figure 4 as well as Figure 5 The table shown is a table of data 105a other than memos containing keywords.

[0078] like Figure 9 As shown, Figure 8The data other than the memo 105 shown may have keywords extracted by the effective part-of-speech extraction unit 114 registered therein as data other than the memo 105 a including the keywords.

[0079] exist Figure 9 In the example shown, for data No. 0, the flags for the keywords "main body" and "damaged" are registered as "1," indicating that "main body" and "damaged" were extracted by the effective part-of-speech extraction unit 114. Furthermore, for data No. 1, the flags for the keywords "deviation," "baseline," and "large" are registered as "1," indicating that "deviation," "baseline," and "large" were extracted by the effective part-of-speech extraction unit 114. Furthermore, for data No. 2, the flags for all keywords are registered as "0," indicating that no keywords were extracted by the effective part-of-speech extraction unit 114.

[0080] Figure 10 This is an example Figure 2 FIG. 1 is a diagram showing output data in the information processing device 1 .

[0081] like Figure 10 As shown, the output data may include data 105 a (that is, input data) other than the memo containing the keyword related to the target of identifying the failure cause, the prediction result 107 , and the explanation 108 .

[0082] exist Figure 10 In the example shown, "main body" and "damaged" are registered as data 105a other than the memo containing the keyword, "damaged" is shown as the prediction result 107, and the contribution of "main body", "damaged", and "exterior = red" to the determination of the defect is shown in the form of a bar graph as an explanation 108.

[0083] Figure 11 This is a diagram illustrating an illustrative AI model that directly outputs explanations in addition to the prediction results in the embodiment.

[0084] Figure 3 The AI processing unit 115 shown is as follows Figure 11 The shown can function as an explainable AI model 211.

[0085] The AI model 211 can be explained to present a prediction result 107 and its basis (ie, explanation 108 ) for a certain instance based on the data 105 other than the memo and the keyword list 104 .

[0086] As the explainable AI model 211 , for example, Wide Learning (WL) can be applied.

[0087] Figure 12 This is a diagram illustrating a model for explaining general AI in the embodiment.

[0088] Figure 3 The AI processing unit 115 shown is as follows Figure 12 As shown, it can function as a general AI model 221 and a description-providing model 222.

[0089] The general AI model 221 outputs a prediction result 107 for a certain instance based on the data 105 other than the memo and the keyword list 104 .

[0090] The description adding model 222 adds a description 108 to a certain instance based on the data 105 other than the memo and the keyword list 104 .

[0091] As the general AI model 221 , for example, Deep Learning (DL) or Random Forest can be applied, and as the description assignment model 222 , for example, LIME, SHAP, or WL can be applied.

[0092] Figure 13 This is a diagram illustrating a model for explaining general AI with respect to image data in the embodiment.

[0093] Figure 3 The AI processing unit 115 shown is as follows Figure 13 As shown, it can function as a general AI model 231 and a description-providing model 232.

[0094] The general AI model 231 outputs a prediction result 107 for a certain instance based on data 105 other than a memo, which is image data.

[0095] The description assignment model 232 assigns a description 108 to a particular instance based on the keyword list 104 .

[0096] As the general AI model 231 , for example, DL can be applied, and as the description provision model 222 , for example, WL can be applied.

[0097] In this way, if a memo is also attached to the image model, it can be used as input to the AI model.

[0098] 〔B〕Effect

[0099] According to the information processing program, the information processing device, and the information processing method in the above-described embodiments, for example, the following effects can be achieved.

[0100] The part-of-speech sentence extraction unit 113 and the effective part-of-speech extraction unit 114 select sentences that match specific parts of speech from sentences extracted from non-standard text data in the product manufacturing process. The AI processing unit 115 trains a machine learning model using training data that associates input features including the selected sentences and product configuration information with label information indicating a judgment result related to the product. The model then outputs a judgment result corresponding to the input features and features contributing to the judgment.

[0101] This improves the readability of the output data from machine learning models that incorporate non-standard text data. Furthermore, predictions using explainable AI are more convincing. Furthermore, by efficiently and automatically extracting necessary keywords based on local knowledge, such as on-site memos from production lines, the readability and persuasiveness of explanations can be further improved, automating the on-site learning and prediction cycle.

[0102] In the process of selecting sentences, the effective part-of-speech extraction unit 114 further excludes at least one of sentences having a character length of a predetermined length or less and sentences consisting only of numerals for selection.

[0103] This can prevent unimportant sentences such as one-character sentences and numbers from being output as feature quantities contributing to determination.

[0104] The part-of-speech sentence extraction unit 113 selects noun sentences and adjective sentences in the process of selecting sentences.

[0105] This makes it possible to output a part of speech that is effective as a feature quantity contributing to determination, such as a noun or an adjective.

[0106] In the process of training the machine learning model, the AI processing unit 115 outputs the cause of the product failure as a determination result, and outputs the contribution of each selected sentence to the failure determination as a contributing feature amount.

[0107] This makes it possible to output data suitable for identifying the cause of a product defect in a product manufacturing process.

[0108] 〔C〕Other

[0109] The disclosed technology is not limited to the above-described embodiment, and can be implemented with various modifications without departing from the spirit of the present embodiment. The various components and processes of the present embodiment can be selected or omitted as needed, or can be appropriately combined.

[0110] In the above embodiment, an example of extracting important words from non-standard text data in Japanese is shown, but the present invention is not limited to this. Important words may also be extracted from non-standard text data in languages other than Japanese, such as English.

[0111] For example, for a memo such as “Deviation is greater than the standard”, the analysis unit 111 applies natural sentence processing to output ““deviation” (noun), “is” (verb), “greater” (adjective), “than” (conjunction), “the” (article), “standard” (noun)”.

[0112] The part-of-speech sentence extraction unit 113 uses the area knowledge 103 that uses nouns and adjectives to output ““deviation” (noun), “greater” (adjective), and “standard” (noun)”.

[0113] Furthermore, for example, by excluding parts of speech consisting of only one character or only numbers, it is possible to narrow down keywords that are effective for explanation. In this example, since the output results using the aforementioned regional knowledge do not include parts of speech consisting of only one character or only numbers, the effective part of speech extraction unit 114 directly outputs "deviation" (noun), "greater" (adjective), and "standard" (noun)."

[0114] For example, for a memo such as "Body 2 damaged", the analyzing unit 111 applies natural sentence processing and outputs ""body" (noun), "2" (noun), "damaged" (verb)".

[0115] The part-of-speech sentence extraction unit 113 directly outputs ““body” (noun), “2” (noun), “damaged” (verb)” by utilizing the domain knowledge 103 that uses the past participle of verbs in addition to nouns and adjectives.

[0116] Furthermore, for example, by excluding parts of speech consisting of only one character or only numbers, keywords that are effective in explanation can be narrowed down. In this example, the effective part of speech extraction unit 114 outputs "body" (noun), "damaged" (verb)".

[0117] Description of Reference Numerals

[0118] 1...Information processing device, 11...CPU, 12...Memory unit, 13...Display control unit, 14...Storage device, 16...External recording medium processing unit, 101...Memo data, 102...Sentence segmentation result, 103, 103a...Regional knowledge, 103b...Condition list, 103c...NG word list, 104...Keyword list, 105, 105a...Data other than memo, 106...Correct answer label, 107...Prediction result, 108...Explanation, 111...Analysis unit, 112...Part-of-speech selection unit, 113...Part-of-speech sentence extraction unit, 114...Valid part-of-speech extraction unit, 115...AI processing unit, 131...Display device, 151...Mouse, 152...Keyboard, 160...Recording medium, 211...Description-capable AI model, 221, 231...General AI model, 222, 232...Description-assigning model, 15...Input interface, 17...Communication interface.

Claims

1. An information processing program, wherein: Causes the computer to perform the following processing: Selecting a sentence that matches a specific part of speech from sentences extracted from non-standard text data in a product manufacturing process; and A machine learning model is trained using training data that establishes a correspondence between input feature quantities containing the selected sentences and product composition information, and label information representing judgment results related to the product, to output judgment results corresponding to the input feature quantities and feature quantities contributing to the judgment.

2. The information processing program according to claim 1, wherein Make the above computer execute: In the process of selecting the above-mentioned sentences, the selection is performed by further excluding at least either one of sentences having a character length of a predetermined length or less and sentences consisting only of numerals.

3. The information processing program according to claim 1 or 2, wherein: Make the above computer execute: In the process of selecting the above-mentioned sentences, a process of selecting a noun sentence and an adjective sentence is performed.

4. The information processing program according to claim 1 or 2, wherein: Make the above computer execute: In the process of training the machine learning model, the cause of the product failure is output as the determination result, and the contribution of each selected sentence to the failure determination is output as the contributing feature amount.

5. An information processing device, wherein: A processor is provided, the processor performing: Selecting a sentence that matches a specific part of speech from sentences extracted from non-standard text data in a product manufacturing process; and A machine learning model is trained using training data that establishes a correspondence between input feature quantities containing the selected sentences and product composition information, and label information representing judgment results related to the product, to output judgment results corresponding to the input feature quantities and feature quantities contributing to the judgment. The information processing apparatus according to claim 5 , wherein: The processor further excludes at least one of sentences having a character length of a predetermined length or less and sentences consisting only of numbers from the selection process of the sentences.

7. The information processing device according to claim 5 or 6, wherein: The processor selects a noun sentence and an adjective sentence in the process of selecting the sentence.

8. The information processing device according to claim 5 or 6, wherein: The processor outputs the cause of the product failure as the determination result and outputs the contribution of each selected sentence to the failure determination as the contributing feature amount during the training of the machine learning model.

9. An information processing method, wherein: The computer performs the following processing: Selecting a sentence that matches a specific part of speech from sentences extracted from non-standard text data in a product manufacturing process; and A machine learning model is trained using training data that establishes a correspondence between input feature quantities containing the selected sentences and product composition information, and label information representing judgment results related to the product, to output judgment results corresponding to the input feature quantities and feature quantities contributing to the judgment.

10. The information processing method according to claim 9, wherein: The above computer executes: In the process of selecting the above-mentioned sentences, the selection is performed by further excluding at least either one of sentences having a character length of a predetermined length or less and sentences consisting only of numerals.

11. The information processing method according to claim 9 or 10, wherein: The above computer executes: In the process of selecting the above-mentioned sentences, a process of selecting a noun sentence and an adjective sentence is performed.

12. The information processing method according to claim 9 or 10, wherein: The above computer executes: In the process of training the machine learning model, the cause of the product failure is output as the determination result, and the contribution of each selected sentence to the failure determination is output as the contributing feature amount.

Images