Explanatory variable proposal device, explanatory variable proposal method, and program

Abstract

To propose an unknown explanatory variable effective for an objective variable.SOLUTION: An explanatory variable proposition device includes: a composite variable generation unit configured to calculate a word vector of a composite variable by adding the word vector of a variable regarding the composite variables obtained by combining a plurality of variables; a distance calculation unit configured to calculate an inter-vector distance between the word vector of an objective variable and the word vector of the composite variable; and an explanatory variable extraction unit configured to extract the composite variable similar to the objective variable based on the inter-vector distance.SELECTED DRAWING: Figure 3

Description

Technical Field

[0001] This invention relates to an explanatory variable proposal device, an explanatory variable proposal method, and a program.

Background Art

[0002] There is a technique for exploring factors related to a target physical property in order to predict the properties of a material. For example, Patent Document 1 discloses a technique for exploring physical property parameters that meet search conditions by constructing a graph having nodes of physical property parameters and edges between nodes corresponding to pairs of related physical property parameters.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, the prior art has a problem that it can only search for variables with known relationships. For example, even if there is an effective explanatory variable among compound words formed by combining words representing known variables, it cannot be searched by the prior art.

[0005] In view of the above technical problems, an object of the present disclosure is to propose an unknown explanatory variable effective for a target variable.

Means for Solving the Problems

[0006] The present disclosure has the following configuration.

[0007] [1] A composite variable generation unit configured to calculate a word vector of a composite variable by adding word vectors of the variables for a composite variable obtained by combining a plurality of variables. A distance calculation unit configured to calculate the vector distance between the word vector of the objective variable and the word vector of the composite variable, An explanatory variable extraction unit is configured to extract composite variables similar to the objective variable based on the distance between the vectors, An explanatory variable proposal device equipped with the following features.

[0008] [2] An explanatory variable suggestion device as described in [1] above, The aforementioned target variable is a composite target variable that combines multiple variables. The system further includes a target variable transformation unit configured to generate a word vector of the target variable by adding the word vectors of the morphological analysis results of the composite target variable. Explanatory variable proposal device.

[0009] [3] An explanatory variable proposal device as described in [2] above, A model storage unit configured to store a transformation model that converts words into word vectors, A similar variable extraction unit is configured to extract variables to be included in the composite variable from the transformation model based on the distance between the target variable and the word vector, An explanatory variable proposal device further equipped with additional features.

[0010] [4] The explanatory variable proposal device described in [2] above, A variable candidate receiving unit configured to accept input of multiple variable candidates, A similar variable extraction unit is configured to extract candidate variables to be included in the composite variable based on their distance from the word vector of the target variable, An explanatory variable proposal device further equipped with additional features.

[0011] [5] An explanatory variable suggestion device as described in [2] above, A model storage unit configured to store a transformation model that converts words into word vectors, A model retraining unit configured to retrain the conversion model using the training data including the morphological analysis result, An explanatory variable proposal device further including

[0012] [6] A computer A composite variable conversion procedure for calculating a word vector of a composite variable formed by combining a plurality of variables by adding word vectors of the variables; A distance calculation procedure for calculating a vector distance between a word vector of a target variable and the word vector of the composite variable; An explanatory variable extraction procedure for extracting the composite variable similar to the target variable based on the vector distance; An explanatory variable proposal method for executing [[ID=十五]]

[0013] [7] On a computer A composite variable conversion procedure for calculating a word vector of a composite variable formed by combining a plurality of variables by adding word vectors of the variables; A distance calculation procedure for calculating a vector distance between a word vector of a target variable and the word vector of the composite variable; An explanatory variable extraction procedure for extracting the composite variable similar to the target variable based on the vector distance; A program for causing the above to be executed

Advantages of the Invention

[0014] According to the present disclosure, an unknown explanatory variable effective for a target variable can be proposed.

Brief Description of Drawings

[0015] [Figure 1] It is a block diagram showing an example of the overall configuration of an explanatory variable proposal system. [Figure 2] It is a block diagram showing an example of the hardware configuration of a computer. [Figure 3] It is a diagram showing an example of the functional configuration of an explanatory variable proposal system in the first embodiment. [Figure 4] It is a flowchart showing an example of the processing procedure of the explanatory variable proposal method in the first embodiment. [Figure 5] It is a flowchart showing an example of the processing procedure of the model relearning method in the first embodiment. [Figure 6] It is a diagram showing an example of the functional configuration of the explanatory variable proposal system in the second embodiment. [Figure 7] It is a flowchart showing an example of the processing procedure of the explanatory variable proposal method in the second embodiment.

Mode for Carrying Out the Invention

[0016] Hereinafter, each embodiment of the present invention will be described with reference to the accompanying drawings. In the present specification and drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant descriptions are omitted.

[0017] [Overview] Among the physical properties of materials, there are some that have not been elucidated as phenomena because the physical phenomena are complex. For example, it is known that the Charpy impact value observed in the Charpy impact test is difficult to predict, and research on prediction has not advanced.

[0018] Generally, in order to predict specific physical properties, machine learning models using blending amounts, physical properties, etc. as explanatory variables are used. However, it has been found that the Charpy impact value cannot be predicted with sufficient accuracy even using these machine learning models.

[0019] Patent Document 1 discloses a technique for searching for factors related to a target physical property by searching a graph that represents the relationships between physical property parameters. Specifically, Patent Document 1 constructs a graph from a database that stores pairs of physical property parameters that have relationships with each other, with physical property parameters as nodes and the nodes corresponding to pairs of physical property parameters having relationships as edges. Then, by searching this graph based on search conditions, factors related to the target physical property are searched. However, the technique disclosed in Patent Document 1 can only be used to search for physical property parameters whose relationships are known.

[0020] Reference 1 discloses a technique that uses a transformation model trained on a large number of papers collected in a particular technical field to convert material names and physical property names into word vectors, and then predicts unknown physical properties of a material based on the cosine similarity between the word vectors. However, the technique disclosed in Non-Patent Document 1 only targets known words for prediction, and cannot search for more effective unknown words even if they exist.

[0021] [Reference 1] Tshitoyan, V., Dagdelen, J., Weston, L. et al., "Unsupervised word embeddings capturent knowledge from materials science literature", Nature, vol. 571, pp. 95-98, 2019.

[0022] In one embodiment of the present invention, a composite variable similar to the target variable is extracted from a composite variable formed by combining multiple variables, and proposed as an effective explanatory variable for the target variable. According to one embodiment of the present invention, by combining known variables, it is possible to evaluate a composite variable whose relationship to the target variable is unknown, and thus it is possible to propose explanatory variables that could not be predicted by conventional techniques.

[0023] [First Embodiment] A first embodiment of the present invention is an explanatory variable suggestion system that proposes effective explanatory variables for an input target variable. In this embodiment, the explanatory variable suggestion system extracts words similar to the target variable (hereinafter also referred to as "similar variables") based on the distance between word vectors. Next, the explanatory variable suggestion device proposes a composite variable similar to the target variable from a composite word (hereinafter also referred to as "composite variable") formed by combining similar variables, based on the distance between word vectors, as an effective explanatory variable for the target variable (hereinafter also referred to as "proposed explanatory variable").

[0024] <Overall structure of the explanatory variable proposal system> First, the overall configuration of the explanatory variable proposal system in this embodiment will be described with reference to Figure 1. Figure 1 is a block diagram showing an example of the overall configuration of the explanatory variable proposal system in this embodiment.

[0025] As shown in Figure 1, the explanatory variable suggestion system 1 in this embodiment includes an explanatory variable suggestion device 10 and a user terminal 30. The explanatory variable suggestion device 10 and the user terminal 30 are connected via a communication network N1 such as a LAN (Local Area Network) or the Internet, enabling data communication.

[0026] The explanatory variable suggestion device 10 is an information processing device such as a PC (Personal Computer), workstation, or server that suggests explanatory variables for the target variable in response to a request from the user terminal 30. The explanatory variable suggestion device 10 receives the target variable to be suggested from the user terminal 30. The explanatory variable suggestion device 10 also extracts effective explanatory variables for the received target variable and transmits them to the user terminal 30 as suggested explanatory variables.

[0027] The user terminal 30 is an information processing terminal such as a PC, tablet, or smartphone operated by the user. The user terminal 30 accepts input of the target variable in response to the user's operation and transmits it to the explanatory variable suggestion device 10. The user terminal 30 also receives proposed explanatory variables from the explanatory variable suggestion device 10 and outputs them to the user.

[0028] The overall configuration of the explanatory variable suggestion system 1 shown in Figure 1 is just one example, and various system configurations are possible depending on the application and purpose. For example, the explanatory variable suggestion device 10 may be implemented using multiple computers, or it may be implemented as a cloud computing service. Alternatively, for example, the explanatory variable suggestion system 1 may be implemented using a standalone information processing device that combines the functions that the explanatory variable suggestion device 10 and the user terminal 30 should each have.

[0029] <Hardware configuration of the explanatory variable proposal system> Next, the hardware configuration of the explanatory variable proposal system 1 in this embodiment will be described with reference to Figure 2.

[0030] Computer Hardware Configuration In this embodiment, the explanatory variable suggestion device 10 and the user terminal 30 are implemented, for example, by a computer. Figure 2 is a block diagram showing an example of the hardware configuration of the computer 500 in this embodiment.

[0031] As shown in Figure 2, the computer 500 includes a CPU (Central Processing Unit) 501, ROM (Read Only Memory) 502, RAM (Random Access Memory) 503, HDD (Hard Disk Drive) 504, input device 505, display device 506, communication interface 507, and external interface 508. The CPU 501, ROM 502, and RAM 503 form what is known as a computer. Each piece of hardware in the computer 500 is interconnected via a bus line 509. The input device 505 and display device 506 may also be used by connecting them to the external interface 508.

[0032] The CPU 501 is a processing unit that controls and implements the overall functions of the computer 500 by reading programs and data from storage devices such as the ROM 502 or HDD 504 onto the RAM 503 and executing processing.

[0033] ROM502 is an example of non-volatile semiconductor memory (storage device) that can retain programs and data even when the power is turned off. ROM502 functions as the main memory, storing various programs and data necessary for the CPU501 to execute the programs installed on HDD504. Specifically, ROM502 stores boot programs such as BIOS (Basic Input / Output System) and EFI (Extensible Firmware Interface) that are executed when the computer 500 starts up, as well as OS (Operating System) settings, network settings, and other data.

[0034] RAM503 is an example of volatile semiconductor memory (storage device) whose programs and data are erased when the power is turned off. RAM503 includes, for example, DRAM (Dynamic Random Access Memory) and SRAM (Static Random Access Memory). RAM503 provides a working area that is expanded when various programs installed on HDD504 are executed by CPU501.

[0035] HDD504 is an example of a non-volatile storage device that stores programs and data. The programs and data stored in HDD504 include the operating system (OS), which is the basic software that controls the entire computer 500, and applications that provide various functions on the OS. Note that computer 500 may use a storage device that uses flash memory as its storage medium (e.g., SSD: Solid State Drive) instead of HDD504.

[0036] The input device 505 includes a touch panel used by the user to input various signals, operation keys and buttons, a keyboard and mouse, and a microphone for inputting sound data such as voice.

[0037] The display device 506 consists of a display such as a liquid crystal or organic EL (Electro-Luminescence) that displays a screen, and a speaker that outputs sound data such as audio.

[0038] Communication I / F 507 is an interface that connects to a communication network and allows computer 500 to perform data communication.

[0039] External I / F 508 is an interface for external devices. Examples of external devices include the drive device 510.

[0040] The drive device 510 is a device for setting the recording medium 511. The recording medium 511 here includes media that record information optically, electrically, or magnetically, such as CD-ROMs, flexible disks, and magneto-optical disks. The recording medium 511 may also include semiconductor memory that records information electrically, such as ROMs and flash memory. This allows the computer 500 to read and / or write to the recording medium 511 via the external I / F 508.

[0041] The various programs to be installed on the HDD 504 are installed, for example, when the distributed recording medium 511 is set in a drive device 510 connected to an external I / F 508, and the various programs recorded on the recording medium 511 are read by the drive device 510. Alternatively, the various programs to be installed on the HDD 504 may be downloaded via the communication I / F 507 from a network other than the communication network and installed that way.

[0042] <Functional Configuration of the Explanatory Variable Proposal System> Next, the functional configuration of the explanatory variable suggestion system in this embodiment will be described with reference to Figure 3. Figure 3 is a block diagram showing an example of the functional configuration of the explanatory variable suggestion system 1 in this embodiment.

[0043] ≪Functional Configuration of the Explanatory Variable Proposal Device≫ As shown in Figure 3, the explanatory variable suggestion device 10 in this embodiment includes a model storage unit 100, a target variable reception unit 101, a target variable conversion unit 102, a similar variable extraction unit 103, a composite variable generation unit 104, a distance calculation unit 105, an explanatory variable extraction unit 106, an explanatory variable suggestion unit 107, and a model retraining unit 108.

[0044] The target variable receiving unit 101, target variable transformation unit 102, similar variable extraction unit 103, composite variable generation unit 104, distance calculation unit 105, explanatory variable extraction unit 106, explanatory variable suggestion unit 107, and model retraining unit 108 are implemented by a process in which the program, which is loaded from the HDD 504 shown in Figure 2 onto the RAM 503, is executed by the CPU 501.

[0045] The model storage unit 100 is implemented, for example, using the HDD 504 shown in Figure 2.

[0046] The model memory unit 100 stores the trained transformation model. The transformation model is a model that transforms words into word vectors. In this embodiment, for example, word2vec can be used as the transformation model. In this embodiment, in order to enable the transformation of as many words as possible, a transformation model is used that uses articles published on an online dictionary site (for example, the Japanese version of Wikipedia®) as training data.

[0047] The target variable receiving unit 101 receives the target variable to be proposed from the user terminal 30. In this embodiment, the target variable may be a composite target variable formed by combining multiple variables. The target variable receiving unit 101 sends the received target variable to the target variable conversion unit 102.

[0048] The target variable transformation unit 102 uses the transformation model stored in the model storage unit 100 to convert the target variable received by the target variable reception unit 101 into a word vector. If the target variable is a composite target variable, the target variable transformation unit 102 performs morphological analysis on the target variable and generates a word vector of the target variable by adding the word vectors of each morpheme included in the morphological analysis result.

[0049] The similar variable extraction unit 103 obtains word vectors of all variables that can be converted by the conversion model stored in the model storage unit 100. The similar variable extraction unit 103 also extracts similar variables that are similar to the target variable based on the distance between the word vectors.

[0050] The composite variable generation unit 104 generates multiple composite variables by combining multiple similar variables extracted by the similar variable extraction unit 103. Furthermore, for each composite variable, the composite variable generation unit 104 generates a word vector for the composite variable by adding the word vectors of the similar variables included in that composite variable.

[0051] The distance calculation unit 105 calculates the distance between the word vector of the target variable and the word vectors of each composite variable (hereinafter referred to as "inter-vector distance").

[0052] The explanatory variable extraction unit 106 extracts composite variables similar to the target variable based on the distance between vectors calculated by the distance calculation unit 105.

[0053] The explanatory variable proposal unit 107 transmits the composite variables extracted by the explanatory variable extraction unit 106 to the user terminal 30 as proposed explanatory variables.

[0054] If the model retraining unit 108 finds that there are words that cannot be converted by the conversion model stored in the model storage unit 100, it retrains the conversion model using documents containing those words as training data. Any type of document can be used as training data, but examples include patent documents or technical papers.

[0055] ≪Functional Configuration of User Terminal 30≫ As shown in Figure 3, the user terminal 30 in this embodiment includes an objective variable input unit 301 and an explanatory variable display unit 302.

[0056] The target variable input unit 301 and explanatory variable display unit 302 of the user terminal 30 are realized by a process in which a program loaded from the HDD 504 shown in Figure 2 onto the RAM 503 is executed by the CPU 501.

[0057] The target variable input unit 301 accepts input of the target variable in response to user operation. The target variable input unit 301 transmits the accepted target variable to the explanatory variable suggestion device 10.

[0058] The explanatory variable display unit 302 receives proposed explanatory variables from the explanatory variable suggestion device 10. The explanatory variable display unit 302 outputs the received proposed explanatory variables to the display device 506 or the like.

[0059] <Processing procedure for the explanatory variable proposal system> Next, the processing procedure of the explanatory variable proposal method executed by the explanatory variable proposal system 1 in this embodiment will be described with reference to Figure 4. Figure 4 is a flowchart showing an example of the processing procedure of the explanatory variable proposal method in this embodiment.

[0060] In step S101, the target variable input unit 301 of the user terminal 30 receives input of the target variable to be proposed in response to the user's operation. Next, the target variable input unit 301 transmits the received target variable to the explanatory variable suggestion device 10.

[0061] In the explanatory variable suggestion device 10, the target variable receiving unit 101 receives the target variable from the user terminal 30. Next, the target variable receiving unit 101 sends the received target variable to the target variable conversion unit 102.

[0062] In step S102, the target variable conversion unit 102 of the explanatory variable suggestion device 10 reads the conversion model stored in the model storage unit 100. Next, the target variable conversion unit 102 inputs the target variable received from the target variable reception unit 101 into the read conversion model to generate a word vector of the target variable. Subsequently, the target variable conversion unit 102 sends the generated word vector of the target variable to the similar variable extraction unit 103.

[0063] Here, the objective variable transformation process in this embodiment (step S102 in Figure 4) will be explained in more detail with reference to Figure 5. Figure 5 is a flowchart showing an example of the objective variable transformation process in this embodiment.

[0064] In step S102-1, the target variable transformation unit 102 performs morphological analysis on the target variable. While known methods can be applied to morphological analysis, for example, a morphological analysis library such as Janome may be used.

[0065] In step S102-2, the target variable transformation unit 102 determines whether all morphemes of the target variable are included in the transformation model stored in the model storage unit 100. If all morphemes are included in the transformation model (YES), the target variable transformation unit 102 proceeds to step S102-3. If there are morphemes that are not included in the transformation model (NO), the target variable transformation unit 102 proceeds to step S108-1.

[0066] In step S108-1, the model retraining unit 108 collects training data containing morphemes that are not included in the conversion model. The training data to be collected is, for example, patent documents or technical papers containing the morphemes. The training data may be collected automatically, for example, by internet search, or it may be collected manually by the user by requesting the user to input training data from the user terminal 30.

[0067] In step S108-2, the model retraining unit 108 retrains the transformation model using the collected training data. Next, the model retraining unit 108 updates the transformation model stored in the model storage unit 100 with the retrained transformation model.

[0068] Next, the model retraining unit 108 returns to step S102-2. That is, the model retraining unit 108 repeats the collection of training data and retraining of the transformation model until all morphemes of the target variable are included in the transformation model.

[0069] In step S102-3, the target variable transformation unit 102 reads the transformation model stored in the model storage unit 100. Next, the target variable transformation unit 102 inputs each morpheme included in the morphological analysis result of the target variable into the transformation model to generate word vectors for each morpheme.

[0070] In step S102-4, the target variable transformation unit 102 generates a word vector for the target variable by adding the word vectors of each morpheme included in the morphological analysis results of the target variable.

[0071] Let's return to Figure 4 for explanation. In step S103, the similar variable extraction unit 103 of the explanatory variable suggestion device 10 reads the transformation model stored in the model storage unit 100. Next, the similar variable extraction unit 103 obtains word vectors of variables that can be transformed by the read transformation model. Variables that can be transformed by the transformation model are, for example, the vocabulary of word2vec. In this embodiment, the transformation model is a trained transformation model that uses articles from an online dictionary site as training data, so the obtained variables can be considered equivalent to the entire Japanese corpus.

[0072] Next, the similar variable extraction unit 103 calculates the vector distance between each word vector of the acquired variables and the word vector of the target variable. Subsequently, the similar variable extraction unit 103 extracts variables whose vector distance is smaller than a predetermined distance r as similar variables. Then, the similar variable extraction unit 103 sends the word vectors of the extracted similar variables to the composite variable generation unit 104.

[0073] In step S104, the composite variable generation unit 104 of the explanatory variable suggestion device 10 receives word vectors of similar variables from the similar variable extraction unit 103. Next, the composite variable generation unit 104 generates multiple composite variables by combining similar variables with a predetermined number of words n or less. Subsequently, for each composite variable, the composite variable generation unit 104 generates a word vector for that composite variable by adding the word vectors of the similar variables included in that composite variable. Then, the composite variable generation unit 104 sends the generated word vectors of the composite variables to the distance calculation unit 105.

[0074] The generation of composite variables is performed as follows: First, the composite variable generation unit 104 generates all composite variables consisting of k combinations of variables from similar variables, with k=2. Next, the composite variable generation unit 104 increments k and generates all composite variables consisting of k combinations of variables from similar variables. The composite variable generation unit 104 then repeats these processes until k=n.

[0075] In step S105, the distance calculation unit 105 of the explanatory variable suggestion device 10 receives the word vectors of the composite variables from the composite variable generation unit 104. Next, the distance calculation unit 105 calculates the vector distance between the word vector of the target variable and the word vectors of each composite variable. Subsequently, the distance calculation unit 105 sends the calculated vector distances to the explanatory variable extraction unit 106.

[0076] In step S106, the explanatory variable extraction unit 106 of the explanatory variable suggestion device 10 receives the vector-to-vector distance from the distance calculation unit 105. Next, the explanatory variable extraction unit 106 extracts composite variables whose vector-to-vector distance is smaller than a predetermined distance r as proposed explanatory variables. Then, the explanatory variable extraction unit 106 sends the extracted proposed explanatory variables to the explanatory variable suggestion unit 107.

[0077] In step S107, the explanatory variable suggestion unit 107 of the explanatory variable suggestion device 10 receives the suggested explanatory variables from the explanatory variable extraction unit 106. Next, the explanatory variable suggestion unit 107 sorts the received suggested explanatory variables in ascending order of the distance between vectors. Subsequently, the explanatory variable suggestion unit 107 transmits the sorted suggested explanatory variables to the user terminal 30.

[0078] On the user terminal 30, the explanatory variable display unit 302 receives proposed explanatory variables from the explanatory variable suggestion device 10. Next, the explanatory variable display unit 302 displays the received proposed explanatory variables on the display device 506.

[0079] <Effects of the First Embodiment> The explanatory variable suggestion device in this embodiment extracts suggested explanatory variables similar to the target variable from a composite variable which is a combination of similar variables similar to the target variable. This configuration makes it possible to evaluate unknown composite variables which are combinations of variables similar to the target variable. Therefore, according to the explanatory variable suggestion device in this embodiment, an unknown composite variable can be suggested as an explanatory variable similar to the target variable.

[0080] [Second Embodiment] In the first embodiment, the system is configured to extract variables similar to the target variable from the entire Japanese corpus and propose explanatory variables from composite variables created by combining these variables. In the second embodiment, the system is configured to accept variable candidates from the user and propose explanatory variables from composite variables created by combining these variable candidates.

[0081] The variable candidates entered by the user are, for example, a list of technical terms related to a specific technology field. The variable candidates may be multiple lists related to multiple technology fields. When generating composite variables from the entire Japanese corpus, explanatory variables combining terms from various fields will be proposed. If a user wants to search for explanatory variables from a combination of terms in a specific technology field, providing terms from that field as variable candidates allows for efficient searching of explanatory variables.

[0082] <Functional Configuration of the Explanatory Variable Proposal System> First, the functional configuration of the explanatory variable suggestion system in this embodiment will be described with reference to Figure 6. Figure 6 is a block diagram showing an example of the functional configuration of the explanatory variable suggestion system 1 in this embodiment.

[0083] ≪Functional Configuration of the Explanatory Variable Proposal Device≫ As shown in Figure 6, the explanatory variable suggestion device 10 in this embodiment, like the first embodiment, comprises a model storage unit 100, a target variable receiving unit 101, a target variable conversion unit 102, a similar variable extraction unit 103, a composite variable generation unit 104, a distance calculation unit 105, an explanatory variable extraction unit 106, an explanatory variable suggestion unit 107, and a model retraining unit 108. Furthermore, the explanatory variable suggestion device 10 in this embodiment further comprises a variable candidate receiving unit 111 and a variable candidate conversion unit 112.

[0084] The variable candidate receiving unit 111 receives multiple variable candidates from the user terminal 30. The variable candidate receiving unit 111 sends the received variable candidates to the variable candidate conversion unit 112.

[0085] The variable candidate conversion unit 112 uses the conversion model stored in the model storage unit 100 to convert the variable candidate received by the variable candidate reception unit 111 into word vectors.

[0086] In this embodiment, the similar variable extraction unit 103 receives word vectors of variable candidates from the variable candidate conversion unit 112. The similar variable extraction unit 103 also extracts variable candidates similar to the target variable as similar variables based on the distance between the word vectors.

[0087] ≪Functional Configuration of User Terminal 30≫ As shown in Figure 6, the user terminal 30 in this embodiment includes a target variable input unit 301 and an explanatory variable display unit 302, similar to the first embodiment. Furthermore, the user terminal 30 in this embodiment also includes a variable candidate input unit 303.

[0088] The variable candidate input unit 303 accepts input of variable candidates in response to user operations. The variable candidate input unit 303 transmits the accepted variable candidates to the explanatory variable suggestion device 10.

[0089] <Processing procedure for the explanatory variable proposal system> Next, the processing procedure of the explanatory variable proposal method executed by the explanatory variable proposal system 1 in this embodiment will be described with reference to Figure 7. Figure 7 is a flowchart showing an example of the processing procedure of the explanatory variable proposal method in this embodiment.

[0090] The following will focus on explaining the differences from the explanatory variable proposal method in the first embodiment.

[0091] In step S111, the variable candidate input unit 303 of the user terminal 30 receives input of multiple variable candidates in response to user operations. Next, the variable candidate input unit 303 transmits the received variable candidates to the explanatory variable suggestion device 10.

[0092] In the explanatory variable suggestion device 10, the variable candidate receiving unit 111 receives variable candidates from the user terminal 30. Next, the variable candidate receiving unit 111 sends the received variable candidates to the variable candidate conversion unit 112.

[0093] In step S112, the variable candidate conversion unit 112 of the explanatory variable suggestion device 10 reads the conversion model stored in the model storage unit 100. Next, the variable candidate conversion unit 112 inputs the variable candidates received from the variable candidate reception unit 111 into the read conversion model to generate word vectors of the variable candidates. Subsequently, the variable candidate conversion unit 112 sends the generated word vectors of the variable candidates to the similar variable extraction unit 103.

[0094] If the transformation model does not contain any candidate variables, it is not possible to generate word vectors for the candidate variables. In this case, the model retraining unit 108 can simply repeat the collection of training data and retraining of the transformation model until the candidate variable is included in the transformation model.

[0095] In step S103, the similar variable extraction unit 103 of the explanatory variable suggestion device 10 receives the word vectors of the variable candidates from the variable candidate conversion unit 112. Next, the similar variable extraction unit 103 calculates the vector distance between each of the word vectors of the variable candidates received from the variable candidate conversion unit 112 and the word vector of the target variable received from the target variable conversion unit 102. Subsequently, the similar variable extraction unit 103 extracts variable candidates for which the vector distance is smaller than a predetermined distance r as similar variables. Then, the similar variable extraction unit 103 sends the extracted word vectors of the similar variables to the composite variable generation unit 104.

[0096] The similar variable extraction unit 103 may, in addition to extracting similar variables from variable candidates, also extract similar variables from variables obtained from the transformation model (i.e., the entire Japanese corpus). In this case, a flag is assigned to the word vector of the similar variable and sent to the composite variable generation unit 104. This flag indicates whether the similar variable is based on a variable candidate or on a transformation model.

[0097] In step S104, the composite variable generation unit 104 of the explanatory variable suggestion device 10 generates multiple composite variables by combining similar variables received from the similar variable extraction unit 103. At this time, the composite variable generation unit 104 may generate composite variables by combining only similar variables extracted from the variable candidates, or it may generate composite variables by combining similar variables extracted from the variable candidates and similar variables extracted from the entire Japanese corpus.

[0098] <Effects of the second embodiment> The explanatory variable suggestion device in this embodiment generates composite variables using variable candidates input from a user terminal and extracts suggested explanatory variables similar to the target variable from these composite variables. Therefore, according to the explanatory variable suggestion device in this embodiment, effective explanatory variables for the target variable can be suggested based on variable candidates narrowed down by the user's knowledge.

[0099] [supplement] Each of the embodiments described above can be implemented by one or more processing circuits. Hereinafter, "processing circuit" as used herein includes processors programmed to execute each function by software, such as processors implemented by electronic circuits, as well as devices such as ASICs (Application Specific Integrated Circuits), DSPs (Digital Signal Processors), FPGAs (Field Programmable Gate Arrays), and conventional circuit modules designed to execute each of the functions described above.

[0100] Although embodiments of the present invention have been described in detail above, the present invention is not limited to these embodiments, and various modifications or changes are possible within the scope of the gist of the present invention as described in the claims. [Explanation of symbols]

[0101] 1. Explanatory Variable Proposal System 10 Explanatory Variable Proposal Device 30 User terminals 100 Model Memory Unit 101 Target Variable Reception Section 102 Target Variable Transformation Unit 103 Similar Variable Extraction Unit 104 Composite Variable Generation Unit 105 Distance calculation section 106 Explanatory Variable Extraction Unit 107 Explanatory Variable Proposal Section 111 Variable Candidate Reception Unit 112 Variable Candidate Conversion Unit 301 Target Variable Input Section 302 Explanatory Variable Display Section 303 Variable Candidate Input Section

Claims

1. A composite variable generation unit is configured to calculate the word vector of a composite variable by adding the word vectors of the variables, for a composite variable that is a combination of multiple variables. A distance calculation unit configured to calculate the vector distance between the word vector of the objective variable and the word vector of the composite variable, An explanatory variable extraction unit is configured to extract composite variables similar to the objective variable based on the distance between the vectors, An explanatory variable proposal device equipped with the following features.

2. An explanatory variable proposal device according to claim 1, The aforementioned target variable is a composite target variable that combines multiple variables. The system further includes a target variable transformation unit configured to generate a word vector of the target variable by adding the word vectors of the morphological analysis results of the composite target variable. Explanatory variable proposal device.

3. The explanatory variable proposal device according to claim 2, A model storage unit configured to store a transformation model that converts words into word vectors, A similar variable extraction unit is configured to extract variables to be included in the composite variable from the transformation model based on the distance between the target variable and the word vector, An explanatory variable proposal device further equipped with additional features.

4. The explanatory variable proposal device according to claim 2, A variable candidate receiving unit configured to accept input of multiple variable candidates, A similar variable extraction unit is configured to extract candidate variables to be included in the composite variable based on their distance from the word vector of the target variable, An explanatory variable proposal device further equipped with additional features.

5. The explanatory variable proposal device according to claim 2, A model storage unit configured to store a transformation model that converts words into word vectors, A model retraining unit configured to retrain the transformation model using training data that includes morphemes from the morphological analysis results that are not included in the transformation model, An explanatory variable proposal device further equipped with additional features.

6. Computers A composite variable transformation procedure for a composite variable that combines multiple variables, which involves calculating the word vector of the composite variable by adding the word vectors of the variables, A distance calculation procedure for calculating the vector distance between the word vector of the objective variable and the word vector of the composite variable, A procedure for extracting explanatory variables, which extracts composite variables similar to the objective variable based on the distance between the vectors, A method for proposing explanatory variables to perform this operation.

7. On the computer, A composite variable transformation procedure for a composite variable that combines multiple variables, which involves calculating the word vector of the composite variable by adding the word vectors of the variables, A distance calculation procedure for calculating the vector distance between the word vector of the objective variable and the word vector of the composite variable, A procedure for extracting explanatory variables, which extracts composite variables similar to the objective variable based on the distance between the vectors, A program to execute.

Images