JP7664585B2 - Machine learning device, data processing device, inference device, machine learning method, data processing method, and inference method - Google Patents

Description

The present invention relates to a machine learning device, a data processing device, an inference device, a machine learning method, a data processing method, and an inference method.

Conventionally, by using a user's personal information to provide information tailored to that user, convenience for the user and usefulness for the information provider have been improved. For example, Patent Documents 1 and 2 disclose systems that deliver advertising information to a user's terminal based on personal information including the user's name, sex, age, etc.

JP 2019-211851 A JP 2002-366843 A

The systems disclosed in Patent Documents 1 and 2 are premised on personal information being input by each user, and therefore are unable to provide appropriate information to users for whom no personal information has been input. Furthermore, when a user takes some kind of action, it is believed that the user has inherent behavioral factor indices (e.g., sensibility, etc.) that indicate the factors that motivate the user to take that action. However, in the systems disclosed in Patent Documents 1 and 2, such behavioral factor indices are not input as personal information, and therefore it is unable to provide appropriate information according to the behavioral factor indices inherent to each user.

The present invention has been made in consideration of the above-mentioned problems, and aims to provide a machine learning device, a data processing device, an inference device, a machine learning method, a data processing method, and an inference method for appropriately grasping the factors that motivate a person to take action from that person's behavior.

In order to achieve the above object, a machine learning device according to one aspect of the present invention comprises:
a machine learning unit that inputs a plurality of sets of learning data, each set being composed of person data that records a person's behavior based on one or more viewpoints and a behavioral factor index that indicates a factor that motivates the person to perform the behavior, into a learning model, and causes the learning model to learn a correlation between the person data and the behavioral factor index by machine learning;
and a learned model memory unit that stores the learned model in which the correlation has been learned by the machine learning unit.

According to a machine learning device according to one aspect of the present invention, it is possible to provide a learning model capable of estimating a behavioral factor index that indicates factors that motivate a person to take action, based on person data that records the person's behavior from one or more perspectives. Therefore, by using this learning model, it is possible to appropriately assign (infer) a behavioral factor index inherent to a person (user), for example, without requiring the person (user) to input personal information.

Other objects, configurations and effects will become apparent from the detailed description of the invention described below.

1 is an overall configuration diagram showing an example of a person information management system 1. FIG. FIG. 9 is a hardware configuration diagram showing an example of a computer 900. FIG. 2 is a data configuration diagram showing an example of a person database 20. FIG. 2 is a block diagram showing an example of a machine learning device 3. FIG. 3 is a diagram showing an example of determination rule data 310. 2 is a diagram showing an example of learning data 11 and a learning model 12. FIG. FIG. 3 is a functional explanatory diagram showing an example of a machine learning unit 301. FIG. 2 is a block diagram showing an example of a data processing device 4. FIG. 2 is a functional explanatory diagram showing an example of a data processing device 4. 4 is a flowchart showing an example of a machine learning method performed by the machine learning device 3. 11 is a flowchart (continuation of FIG. 10 ) showing an example of a machine learning method by the machine learning device 3. 4 is a flowchart showing an example of a data processing method by the data processing device 4.

The following describes an embodiment of the present invention with reference to the drawings. The following shows a schematic view of the scope of the description necessary to achieve the object of the present invention, and mainly describes the scope of the description necessary to explain the relevant parts of the present invention. The parts that are omitted are based on publicly known technology.

(Personal Information Management System 1)
1 is an overall configuration diagram showing an example of a person information management system 1. The person information management system 1 functions as a system that can provide each person's behavior factor index to various services (product planning, product development, production management, purchasing/order management, sales strategy formulation, marketing, advertising, etc.) by adding behavior factor indexes (tags) that indicate factors that motivate a person to take action to person data that records the person's actions.

Person data is data that records a person's behavior based on one or more perspectives. Perspectives for recording a person's behavior include, for example, purchasing behavior when a person purchases a product or service at a store or an e-commerce site, web browsing behavior when a person browses a website, web search behavior when a person searches a website, travel behavior when a person travels by any means of travel (walking, train, car, etc.), and survey response behavior when a person responds to a survey on a website or in print. Therefore, person data includes, for example, a purchase history that records purchasing behavior, a web browsing history that records web browsing behavior, a web search history that records web search behavior, a travel history that records travel behavior, and a survey response history that records survey response behavior.

A behavioral factor index indicates factors that motivate a person to take a particular action. There are multiple types of behavioral factor indexes, classified by, for example, sensibility, preference, attributes, etc. Examples of sensibility and preference include health-conscious, organic, particular about origin of produce, price (savings) conscious, luxury-conscious, safety and security, high sensitivity to beauty, cleanliness freak, Japanese food lover, Western food lover, sweet tooth, spicy food lover, strong flavor lover, sour flavor lover, carbonated food lover, indoor type, outdoor type, etc. Examples of attributes related to demographics and lifestyle include age, sex, single, married, with children, own home, rent, own a car, keep pets, on a diet, etc.

The behavioral factor index is assigned to the person data according to the variable value, and a variable value is assigned for each type of behavioral factor index. In this embodiment, the case where the behavioral factor index of "health-oriented" is defined by a two-class classification (binary classification of 0 or 1) is mainly described, so the variable values of the behavioral factor index of "health-oriented" are assigned by "0: not health-oriented" and "1: health-oriented". Note that the behavioral factor index may be defined by a multi-class classification (multi-value classification), and for example, when it is defined by a three-class classification, the variable values of the behavioral factor index of "health-oriented" may be assigned by "-1: not health-oriented", "0: neither", and "1: health-oriented". Furthermore, the definition of the behavioral factor index may be different for each type of behavioral factor index.

The main components of the person information management system 1 are a database device 2, a machine learning device 3, a data processing device 4, and a worker terminal device 5. Each of the devices 2 to 5 is configured, for example, as a general-purpose or dedicated computer (see FIG. 2 described below), and is connected to a wired or wireless network 6 so that various data can be transmitted and received between them. Note that the number of the devices 2 to 5 and the connection configuration of the network 6 are not limited to the example in FIG. 1, and may be changed as appropriate.

The database device 2 is linked to an external system (not shown) that records the daily actions of each person as a behavioral history based on one or more perspectives, and includes a person database 20 that can register behavioral history based on one or more perspectives as person data for each person. The external system is composed of, for example, a point of sale information management system (POS system), a computer operation log collection system, a navigation system, a questionnaire system, etc. The external system records purchase history, web browsing history, web search history, movement history, questionnaire response history, etc. as behavioral history according to the daily actions of each person, and provides this to the database device 2, for example, in response to a request from the database device 2.

The machine learning device 3 operates as the subject of the learning phase of machine learning, and performs machine learning of the learning model 12 by using person data registered in the person database 20 as learning data 11. The learned learning model 12 is provided to the data processing device 4 via the network 6, a recording medium, etc. Supervised learning is adopted as the machine learning method.

The data processing device 4 operates as the subject of the inference phase of machine learning, and assigns behavioral factor indices to the person data using the trained learning model 12 generated by the machine learning device 3, and provides the assignment results (in this embodiment, the variable values for each type of behavioral factor index) to the database device 2, the worker terminal device 5, etc.

The worker terminal device 5 is a terminal device used by a worker 10 who performs annotation work on the learning data 11 when performing machine learning of the learning model 12. The worker terminal device 5 accepts various input operations via a display screen such as an application program or a web browser, and displays various information (e.g., person data, behavioral factor indexes, etc.) via the display screen.

(Computer 900)
2 is a hardware configuration diagram showing an example of the computer 900. Each of the devices 2 to 5 of the person information management system 1 is configured by a general-purpose or dedicated computer 900.

2, the computer 900 includes, as its main components, a bus 910, a processor 912, a memory 914, an input device 916, an output device 917, a display device 918, a storage device 920, a communication I/F (interface) unit 922, an external device I/F unit 924, an I/O (input/output) device I/F unit 926, and a media input/output unit 928. Note that the above components may be omitted as appropriate depending on the application of the computer 900.

The processor 912 is composed of one or more arithmetic processing devices (such as a CPU (Central Processing Unit), MPU (Micro-processing unit), DSP (Digital signal processor), GPU (Graphics Processing Unit)) and operates as a control unit that controls the entire computer 900. The memory 914 stores various data and programs 930, and is composed of, for example, a volatile memory (DRAM, SRAM, etc.) that functions as a main memory, a non-volatile memory (ROM), a flash memory, etc.

The input device 916 is, for example, a keyboard, a mouse, a numeric keypad, an electronic pen, etc., and functions as an input unit. The output device 917 is, for example, a sound (audio) output device, a vibration device, etc., and functions as an output unit. The display device 918 is, for example, a liquid crystal display, an organic EL display, electronic paper, a projector, etc., and functions as an output unit. The input device 916 and the display device 918 may be integrated, such as a touch panel display. The storage device 920 is, for example, a hard disk drive (HDD), a solid state drive (SSD), etc., and functions as a memory unit. The storage device 920 stores various data necessary for the execution of the operating system and the program 930.

The communication I/F unit 922 is connected to a network 940 (which may be the same as the network 6 in FIG. 1) such as the Internet or an intranet by wire or wirelessly, and functions as a communication unit that transmits and receives data to and from other computers according to a predetermined communication standard. The external device I/F unit 924 is connected to an external device 950 such as a camera, printer, scanner, or reader/writer by wire or wirelessly, and functions as a communication unit that transmits and receives data to and from the external device 950 according to a predetermined communication standard. The I/O device I/F unit 926 is connected to an I/O device 960 such as various sensors and actuators, and functions as a communication unit that transmits and receives various signals and data, such as detection signals from sensors and control signals to actuators, between the I/O device 960. The media input/output unit 928 is composed of a drive device such as a DVD (Digital Versatile Disc) drive or a CD (Compact Disc) drive, and reads and writes data to media (non-transient storage media) 970 such as DVDs and CDs.

In the computer 900 having the above configuration, the processor 912 calls up the program 930 stored in the storage device 920 into the memory 914, executes it, and controls each part of the computer 900 via the bus 910. The program 930 may be stored in the memory 914 instead of the storage device 920. The program 930 may be recorded on the medium 970 in an installable file format or an executable file format, and provided to the computer 900 via the media input/output unit 928. The program 930 may be provided to the computer 900 by downloading it via the network 940 via the communication I/F unit 922. In addition, the computer 900 may realize various functions realized by the processor 912 executing the program 930 using hardware such as an FPGA (field-programmable gate array) or an ASIC (application specific integrated circuit).

The computer 900 may be, for example, a desktop computer or a portable computer, and may be any type of electronic device. The computer 900 may be a client computer, a server computer, or a cloud computer.

(Person Database 20)
3 is a data configuration diagram showing an example of the person database 20. The person database 20 is a database for managing person data based on person identification information (user ID) assigned to each person. The person database 20 is composed of, for example, a purchase history table, a web browsing history table, a web search history table, a movement history table, a questionnaire response history table, a behavioral factor index table, product master information, store master information, website master information, and map information.

The purchase history table has multiple records for recording purchasing behavior, and each record registers the date and time, product, price, store (which may be an e-commerce site), etc. A product ID is registered for a product, and the product is linked to product master information by the product ID, thereby associating the product with product attributes (product name, product category, etc.) registered in the product master information. A store ID is registered for a store, and the store is linked to store master information by the store product ID, thereby associating the store with store attributes (store name, store category, etc.) registered in the store master information. A product may be assigned a product index (behavioral characteristic index) corresponding to the behavioral factor index, and may be registered in the product master information as part of the product attributes, for example. A store may be assigned a store index (behavioral characteristic index) corresponding to the behavioral factor index, and may be registered in the store master information as part of the store attributes, for example.

The web browsing history table has multiple records for recording web browsing behavior, and each record is registered with the date and time, website, browsing time, etc. The web search history table has multiple records for recording web search behavior, and each record is registered with the date and time, search words, etc. A URL for accessing the website is registered with the website, and the URL is linked to the website master information, thereby associating the website with the website attributes (administrator name, website category, etc.) registered in the website master information. Note that the website may be assigned a website index (behavioral characteristic index) corresponding to the behavior factor index, and may be registered in the website master information as part of the website attributes, for example.

The movement history table has multiple records for recording movement behavior, and each record registers the date and time, location, means of transportation, etc. For example, the location may be a stay point or area where a user stayed for a specified period of time or more, and the location or area may be linked to map information, thereby associating the location attributes (location name, location category, etc.) and area attributes (area name, area category, etc.) registered in the map information. Note that the location or area may be assigned a location index (behavioral characteristic index) or area index (behavioral characteristic index) that corresponds to the behavioral factor index, and may be registered in the map information as part of the location attribute or area attribute, for example.

The survey response history table has multiple records for recording survey response behavior, and each record contains the date and time, question content, response content, etc.

The behavioral factor index table has multiple records for storing the behavioral factor indexes assigned by the data processing device 4, and each record registers the date and time, the variable values of the behavioral factor index, etc. In the example of Figure 3, three fields are shown corresponding to the three behavioral factor indexes "health-conscious," "sweet tooth," and "car ownership," but similar fields are provided for the other behavioral factor indexes as well.

In the person database 20, the information in each table is associated with a user ID, so that the person data, which records the person's behavior based on multiple perspectives, and the behavior factor index are managed for each person. The information registered in the person database 20 is configured to be able to perform various statistical processes. The person data may be recorded in association with feature data (in this embodiment, a product, a store, a website, a location, or a region) to which a behavior feature index (in this embodiment, a product index, a store index, a website index, a location index, or a region index) that indicates the characteristics of the behavior and corresponds to the behavior factor index is assigned. In addition to the above, the behavior feature index may also be assigned to features such as a time period, a day of the week, or a season. The data configuration of the person database 20 may be changed as appropriate, and data other than the above may be registered, or some of the above data may be omitted.

(Machine learning device 3)
4 is a block diagram showing an example of the machine learning device 3. The machine learning device 3 includes a control unit 30 configured with a processor or the like, a storage unit 31 configured with a HDD, an SSD, a memory or the like, a communication unit 32 which is a communication interface with the network 6, an input unit 33 configured with a keyboard, a mouse or the like, and a display unit 34 configured with a display or the like. Note that the input unit 33 and the display unit 34 may be omitted.

The memory unit 31 stores the learning data 11, the learning model 12, the decision rule data 310, and the machine learning program 311, as well as an operating system, other programs, various data, etc. The memory unit 31 functions as a learning data memory unit that stores the learning data 11, and as a trained model memory unit that stores the learning model 12.

Figure 5 is a diagram showing an example of the determination rule data 310. In the determination rule data 310, a determination rule for determining whether or not a person matches a behavior factor index is registered for each type of behavior factor index. The determination rule data 310 is information that can be edited by the worker 10 via the worker terminal device 5.

The judgment rules are defined for person data based on at least a part of the viewpoint, and are defined by appropriately combining various numerical calculations with comparison, ratio, frequency, period, category, behavioral characteristic index (product index, store index, website index, location index, region index), etc. In addition, judgment rules are defined for each type of behavioral factor index, and in the example of FIG. 5, judgment rules for three behavioral factor indexes, "health-conscious," "sweet tooth," and "car ownership," are defined for "purchase history" based on the viewpoint of "purchase behavior." For example, the judgment rule for the behavioral factor index of "health-conscious" tallies up the purchase amounts of products that each user has purchased in the past year that are assigned the product category of "health food" or the product index of "health-conscious," and if the total purchase amount is X yen or more, the behavioral factor index of "1: health-conscious" is assigned to the user.

In FIG. 5, as the results of the judgment based on the judgment rules, the personal data of "User A" is assigned the behavioral factor indices "1: health-conscious", "0: not sweet tooth", and "1: car owner", the personal data of "User B" is assigned the behavioral factor indices "0: not health-conscious", "0: not sweet tooth", and "1: car owner", and the personal data of "User C" is assigned the behavioral factor indices "1: sweet tooth", and "0: not car owner". In addition, for the personal data of "User C", "N/A: unable to judge" is recorded as the judgment result based on the judgment rules for "health-consciousness" because there is not enough purchase history recorded to make a judgment based on the judgment rules for "health-consciousness".

The determination rules may be defined for person data based on multiple perspectives, for example, for "purchase history", "travel behavior", and "survey response history" based on multiple perspectives, i.e., "purchase behavior", "travel behavior", and "survey response behavior". The determination rules may also be defined for person data based on different perspectives for each type of behavior factor index, for example, a determination rule for "health-consciousness" may be defined for "purchase history" based on "purchase behavior", and a determination rule for "car ownership" may be defined for "travel history" based on "travel behavior".

As shown in FIG. 4, the control unit 30 functions as a data acquisition unit 300 and a machine learning unit 301 by executing the machine learning program 311 stored in the memory unit 41.

The data acquisition unit 300, for example, receives a request from the machine learning unit 301 or the worker terminal device 5, and accesses the person database 20 via the communication unit 32 and the network 6. The data acquisition unit 300 acquires person data for each person from the person database 20, which records the behavior of each of multiple people, and provides the data to the machine learning unit 301 or stores the data in the memory unit 31.

The machine learning unit 301 generates learning data 11 consisting of person data and behavioral factor indices, and inputs multiple sets of the data into the learning model 12, thereby causing the learning model 12 to learn the correlation between the person data and the behavioral factor indices through machine learning. The machine learning unit 301 then stores the learned learning model 12 (specifically, the adjusted parameter group) in the storage unit 31.

When inputting the person data constituting the learning data 11a to the input layer 120 of the learning model 12a, the machine learning unit 301 may perform pre-processing to convert the person data into predetermined features for input to the learning model 12 by any conversion method such as one-hot encoding or label encoding. In this case, the pre-processing may be, for example, converting the person data into features by appropriately combining various numerical calculations with comparison, ratio, frequency, period, category, behavioral feature index (product index, store index, website index, location index, region index), etc. for the purchase history, web browsing history, web search history, movement history, and survey response history contained in the person data.

In particular, when the person data is recorded in association with feature data (product, store, website, location, region) to which behavioral feature indices (product index, store index, website index, location index, region index) are assigned as shown in FIG. 3, the machine learning unit 301 may perform preprocessing to convert the person data into feature quantities based on the behavioral feature indices assigned to the feature data included in the person data, and input a plurality of sets of preprocessed learning data 11 consisting of the feature quantities converted from the person data in the preprocessing and the behavioral factor indices included in the person data to the learning model 12, thereby making the learning model 12 learn the correlation between the person data and the behavioral factor indices. In this case, the preprocessing may, for example, be based on the product indices assigned to products previously purchased by the user that are recorded as part of the purchase history, and may include tallying up the purchase amount of products to which a behavioral factor index of "health-consciousness" is assigned, the purchase amount of products to which a behavioral factor index of "sweet tooth" is assigned, the purchase amount of products to which a behavioral factor index of "car ownership" is assigned, and the like, and converting the tallying up purchase amounts into feature quantities by normalizing each of the tallying up purchase amounts to a value in the range of 0 to 1.

Figure 6 is a diagram showing an example of the learning data 11 and the learning model 12. The learning data 11 used for machine learning of the learning model 12 is composed of person data and behavioral factor indices. The learning data 11 is data used as teacher data (training data), verification data, and test data in supervised learning. In addition, the behavioral factor indices are data used as correct answer labels in supervised learning.

The person data constituting the learning data 11 is acquired from the person database 20. The behavioral factor indices constituting the learning data 11 are assigned based on the judgment results according to the judgment rules registered in the judgment rule data 310, or assigned by the worker 10 through annotation work.

The learning model 12, for example, employs a neural network structure and includes an input layer 120, an intermediate layer 121, and an output layer 122. Synapses (not shown) that connect each neuron are laid between each layer, and each synapse is assigned a weight.

The input layer 120 has a number of neurons corresponding to the person data (which may be features converted from the person data in preprocessing) as input data, and each value of the person data is input to each neuron. The output layer 122 has neurons corresponding to the behavior factor indexes as output data, and the result of assigning the behavior factor index to the person data (inference result) is output as output data (in this embodiment, a value in the range of 0 to 1). The behavior factor indexes (correct answer labels) constituting the learning data 11 are compared with the behavior factor indexes (inference result) output from the output layer 122, and a process (backpropagation, etc.) is performed to adjust a group of parameters such as the weights of each synapse, thereby performing machine learning of the learning model 12.

In this embodiment, in order to explain a case where the machine learning unit 301 performs machine learning of the learning model 12 for each type of behavioral factor index, the storage unit 31 stores multiple learning models 12 corresponding to each type of behavioral factor index. The learning model 12 shown in FIG. 6 outputs a behavioral factor index of "health-oriented" for person data, and machine learning is performed using learning data 11 consisting of person data and the behavioral factor index of "health-oriented". Note that the learning model 12 may output multiple behavioral factor indexes for person data, and may output three behavioral factor indexes, "health-oriented", "sweet tooth", and "car ownership", for example, for person data. In that case, machine learning may be performed using the learning data 11 consisting of person data and the behavioral factor indexes of "health-oriented", "sweet tooth", and "car ownership".

The number and types of learning models 12 stored in the storage unit 31 after machine learning is performed by the machine learning unit 301 may be changed as appropriate. For example, multiple learning models 12 with different conditions, such as the machine learning method, the type of data included in the person data, the type of data included in the behavioral factor index, the method of preprocessing the person data, etc., may be stored. In that case, multiple types of learning data 11 having data configurations corresponding to multiple learning models 12 with different conditions may be used.

As shown in FIG. 4, the machine learning unit 301 includes a first generation processing unit 301a, a first learning processing unit 301b, a second generation processing unit 301c, and a second learning processing unit 301d as units that perform machine learning of the learning model 12.

Figure 7 is a functional explanatory diagram showing an example of the machine learning unit 301. Details of the machine learning unit 301 will be described below with reference to Figure 7, the decision rule data 310 shown in Figure 5, and the learning data 11 and learning model 12 shown in Figure 6.

The first generation processing unit 301a judges for each person whether or not the person whose behavior is recorded in the person data acquired by the data acquisition unit 300 matches the behavior factor index according to the judgment rules registered in the judgment rule data 310. The first generation processing unit 301a generates learning data 11a by assigning a behavior factor index to the person data based on the judgment result.

In the example of FIG. 7, the first generation processing unit 301a judges whether each person (users A to C) meets the criteria of “health-consciousness” according to the judgment rules (see FIG. 5) defined for the purchase records,
Based on the determination result, the learning data 11a is generated by assigning a behavioral factor index (provisional answer label) of "1: health-conscious" to the personal data of "user A" and "0: not health-conscious" to the personal data of "user B." Note that for the personal data of "user C" who cannot be determined according to the determination rule for "health-conscious" due to reasons such as insufficient recording of purchase history, a behavioral factor index (provisional answer label) is not assigned, and therefore learning data 11a based on the personal data of "user C" is not generated.

The first learning processing unit 301b uses the learning data 11a generated by the first generation processing unit 301a to train the learning model 12a on the correlation between the person data and the behavioral factor index. In this case, the person data constituting the learning data 11a includes not only purchase history but also web browsing history, web search history, movement history, and survey response history, as shown in FIG. 6. The first learning processing unit 301b may also use the preprocessed learning data 11a to train the learning model 12a on the correlation between the person data (feature amount) and the behavioral factor index.

The second generation processing unit 301c receives a correction instruction for the behavior factor index output from the learning model 12a by inputting person data (which may be a feature converted from the person data in preprocessing) to the learning model 12a in which correlations have been learned by the first learning processing unit 301b. Specifically, as described above, the second generation processing unit 301c displays the behavior factor index output from the learning model 12a for each person in association with the person data, and receives a correction instruction for the displayed behavior factor index by having the worker 10 perform annotation work using the worker terminal device 5. The second generation processing unit 301c then generates learning data 11b by assigning a behavior factor index to the person data based on the reception result.

In the example of FIG. 7, the second generation processing unit 301c inputs the person data of each person (users A to C) into the learning model 12a, and outputs "0.9" for "user A", "0.6" for "user B", and "0.8" for "user C" as behavioral factor indices (inference results). When the second generation processing unit 301c displays an evaluation list on the worker terminal device 5 in which the behavioral factor indices (inference results) output from the learning model 12a are associated with the person data, the worker 10 visually checks the person data of "user B" and, if the worker 10 determines that "1: health-conscious" is more appropriate than "0: not health-conscious" assigned as the provisional correct label by the first generation processing unit 301a, issues a correction instruction to assign "1: health-conscious" to "user B". Furthermore, if the worker 10 checks the person data of "user C" and determines that "0: not health-conscious" is more appropriate than "1: health-conscious," the worker 10 issues a correction instruction to assign "0: not health-conscious" to "user C." The second generation processing unit 301c then accepts these correction instructions and, based on the acceptance results, assigns "1: health-conscious" to the person data of "user B" and "0: not health-conscious" to the person data of "user C" as behavioral factor indices (corrected correct labels), thereby generating learning data 11b.

In addition, when performing machine learning of the learning model 12a corresponding to the preprocessed learning data 11a, the second generation processing unit 301c may input the feature amount (preprocessed person data) converted from the person data in the preprocessing to the learning model 12a in which the correlation has been learned by the first learning processing unit 301b, and accept a correction instruction for at least one of the behavior factor index output from the learning model 12a and the behavior feature index assigned to the feature data included in the person data. Specifically, as described above, the second generation processing unit 301c displays the behavior factor index output from the learning model 12a and the behavior feature index assigned to the feature data included in the person data for each person in association with the person data, and accepts a correction instruction for at least one of the displayed behavior factor index and the behavior feature index by the worker 10 performing annotation work using the worker terminal device 5. The second generation processing unit 301c may then generate learning data 11b by assigning behavioral factor indices to the person data based on the reception result, and may also modify the behavioral feature indices assigned to the feature data.

The second learning processing unit 301d uses at least the learning data 11b generated by the second generation processing unit 301c to make the learning model 12b learn the correlation between the person data and the behavioral factor index. In this case, the second learning processing unit 301d may use not only the learning data 11b generated by the second generation processing unit 301c, but also the learning data 11a generated by the first generation processing unit 301a (in the example of FIG. 7, the learning data 11a is based on the person data of "User A"). Note that the second learning processing unit 301d may use the preprocessed learning data 11b to make the learning model 12b learn the correlation between the person data (feature amount) and the behavioral factor index.

When the second learning processing unit 301d has the learning model 12b learn the correlation between the person data and the behavioral factor index, if the predetermined learning end condition is not met, the second generation processing unit 301c may repeatedly execute a process of inputting person data into the learning model 12b trained by the second learning processing unit 301d instead of the first learning processing unit 301b, accepting a correction instruction for the behavioral factor index output from the learning model 12b, and generating learning data 11b by assigning the behavioral factor index to the person data based on the acceptance result, and a process of the second learning processing unit 301d having the learning model 12b learn using at least the learning data 11b generated by the second generation processing unit 301c, until the learning end condition is met. Whether or not the learning end condition is met is determined, for example, depending on whether the accuracy of the learning model 12 has reached a predetermined accuracy or whether the number of learning data 11 has reached a predetermined upper limit.

In addition, when performing machine learning of the learning model 12b corresponding to the preprocessed learning data 11b, if a predetermined learning end condition is not met when the second learning processing unit 301d has made the learning model 12b learn the correlation between the person data (feature amount) and the behavioral factor index, the machine learning unit 301 inputs the feature amount (person data after preprocessing) converted from the person data in the preprocessing to the learning model 12b trained by the second learning processing unit 301d instead of the first learning processing unit 301b, and generates the behavioral factor index output from the learning model 12b. and the behavioral feature index assigned to the feature data included in the person data, and based on the result of the reception, generate learning data 11b by assigning a behavior factor index to the person data, and correct the behavioral feature index assigned to the feature data. The second learning processing unit 301d may repeatedly execute the process of training the learning model 12b using at least the learning data 11b generated by the second generation processing unit 301c, which is the preprocessed learning data 11b, until a learning end condition is satisfied.

(Data Processing Device 4)
Fig. 8 is a block diagram showing an example of the data processing device 4. Fig. 9 is a functional explanatory diagram showing an example of the data processing device 4. The data processing device 4 includes a control unit 40 including a processor or the like, a storage unit 41 including a HDD, an SSD, a memory, or the like, a communication unit 42 which is a communication interface with the network 6, an input unit 43 including a keyboard, a mouse, or the like, and a display unit 44 including a display, or the like. Note that the input unit 43 and the display unit 44 may be omitted.

The storage unit 41 stores the learning model 12 and the data processing program 410, as well as an operating system, other programs, various data, etc.
1 functions as a trained model storage unit that stores a trained learning model 12.

The control unit 40 functions as a data acquisition unit 400, an index assignment unit 401, and an output processing unit 402 by executing a data processing program 410 stored in the memory unit 41.

The data acquisition unit 400, for example, receives a request from the index assignment unit 401 or the worker terminal device 5, and accesses the person database 20 via the communication unit 42 and the network 6. The data acquisition unit 400 acquires person data from the person database 20, provides it to the index assignment unit 401, or stores it in the memory unit 41.

The index assignment unit 401 assigns the behavioral factor index (inference result) output from the learning model 12 to the person data by inputting the person data acquired by the data acquisition unit 400 (which may be a feature converted from the person data in preprocessing) to the learned learning model 12 stored in the storage unit 41. At that time, as post-processing of the behavioral factor index (inference result) output as output data (in this embodiment, a value in the range of 0 to 1) from the learning model 12, the index assignment unit 401 compares the value of the variable of the behavioral factor index (in FIG. 9, "0.9") with a predetermined threshold (for example, "0.5"), and if it is equal to or greater than the threshold, assigns the behavioral factor index of "1: health-conscious" to the person data, and if it is less than the threshold, assigns the behavioral factor index of "0: not health-conscious" to the person data. When inputting person data to the input layer 120 of the learning model 12, the index assignment unit 401 may perform preprocessing similar to that performed by the machine learning unit 301, so as to input the preprocessed person data, i.e., the features converted from the person data in the preprocessing, to the learning model 12.

In this embodiment, the learning model 12 is generated by the machine learning device 3 for each type of behavioral factor index, and therefore the storage unit 41 stores multiple learning models 12 corresponding to each type of behavioral factor index. The learning model 12 shown in FIG. 9 outputs a behavioral factor index of "health-oriented" for person data. Therefore, the index assignment unit 401 assigns behavioral factor indices other than "health-oriented" to the person data by inputting the person data into the multiple learning models 12 corresponding to each type of behavioral factor index, as in FIG. 9.

The output processing unit 402 performs output processing to output the behavior factor index assigned by the index assignment unit 401. The output processing unit 402 may provide the behavior factor index to the database device 2, for example, to register it in a behavior factor index table of the person database 20, or may provide it to the worker terminal device 5 to present it to the worker. In this case, the output processing unit 402 may output the behavior factor index (a value in the range of 0 to 1) before post-processing is performed in addition to or instead of the behavior factor index (0 or 1) after post-processing by the index assignment unit 401 is performed.

The number of learning models 12 stored in the storage unit 41 is not limited to one, and multiple learning models 12 with different conditions, such as machine learning methods, types of data included in person data, types of data included in behavioral factor indicators, and methods of preprocessing of person data, may be stored and selectively available. The learning models 12 may also be stored in the storage unit of an external computer (e.g., a server-type computer or a cloud-type computer), in which case the data processing device 4 may access the external computer.

(Machine learning methods)
10 and 11 are flowcharts showing an example of a machine learning method by the machine learning device 3. Below, a case will be described in which the worker 10 performs a series of learning instruction operations including annotation work using the worker terminal device 5, and the machine learning device 3 performs machine learning of the learning model 12 by referring to the person database 20 and the decision rule data 310.

First, in step S100, the worker terminal device 5 receives a learning instruction from the worker 10, including learning conditions for performing machine learning on the learning model 12, and in step S101, transmits the learning instruction to the machine learning device 3. As the learning conditions, for example, person data acquisition conditions when acquiring person data for generating the learning data 11 from the person database 20 and learning end conditions are specified.

Next, in step S110, when the machine learning device 3 receives the learning instruction for the learning model 12 transmitted in step S101, the data acquisition unit 300 accesses the person database 20 to acquire person data that satisfies the learning conditions of the learning model 12 included in the learning instruction, and acquires multiple person data corresponding to multiple people.

Next, in step S120, the first generation processing unit 301a judges for each person, based on the multiple person data acquired in step S110, whether the behavior of each person matches the behavior factor index in accordance with the judgment rules registered in the judgment rule data 310. Then, in step S121, the first generation processing unit 301a generates multiple sets of learning data 11a by assigning a behavior factor index to the person data for each person based on the judgment result.

Next, in step S130, the first learning processing unit 301b causes the learning model 12a to learn the correlation between the person data and the behavioral factor index using the multiple sets of learning data 11a generated in step S121 (which may be preprocessed learning data 11a).

Next, in step S140, the second generation processing unit 301c inputs the multiple person data acquired in step S110 into the learning model 12a trained in step S130, and associates the behavioral factor indices output from the learning model 12a as inference results with the person data to create an evaluation list consisting of the person data and the behavioral factor indices. Then, in step S141, the second generation processing unit 301c transmits the evaluation list to the worker terminal device 5.

Next, in step S142, when the worker terminal device 5 receives the evaluation list transmitted in step S141, the worker terminal device 5 displays the evaluation list on the display screen of the worker terminal device 5. At that time, the worker terminal device 5 displays person data corresponding to the people displayed in the evaluation list on the display screen of the worker terminal device 5 as reference information for the worker 10 to perform annotation work.

The display screen may display, for example, the purchase history, web browsing history, web search history, movement history, and survey response history included in the person data, and may further display feature data associated with each history (in this embodiment, product, store, website, location, and region) and behavior feature indexes assigned to the feature data (in this embodiment, product index, store index, website index, location index, and region index). For example, a list of products purchased by each person and product attributes of each product may be displayed, and the product index currently assigned to the product (for example, "luxury orientation") may be displayed as part of the product attributes. In addition, the display screen may display not only raw data of the person data, but also information obtained by performing a predetermined statistical process on the person data, and the display content and the statistical process content may be changeable based on the operation of the worker 10. In addition, the evaluation list on the display screen may display not only the inference results by the learning model 12a in step S140, but also the judgment results by the judgment rules in step S120.

Then, in step S143, when the worker terminal device 5 receives a correction instruction for the behavioral factor index from the worker 10 on its display screen, in step S144, the correction instruction is transmitted to the machine learning device 3. The example of FIG. 11 shows a case in which, as in FIG. 7, the worker terminal device 5 receives a correction instruction to assign "1: health-conscious" to "User B" and a correction instruction to assign "0: not health-conscious" to "User C" as annotation work by the worker 10.

Next, in step S145, the second generation processing unit 301c receives the modification instruction sent in step S144, and receives the modification instruction for the behavior factor index output in step S140, and generates multiple sets of learning data 11b by assigning the behavior factor index to the person data based on the reception result. The worker terminal device 5 may receive a modification instruction for the behavior characteristic index from the worker 10 on the display screen, and in that case, the second generation processing unit 301c may update (modify) the person database 20 by modifying the behavior characteristic index in the characteristic data registered in the person database 20 based on the reception result. The modification instruction for the behavior characteristic index (here, the product index) may, for example, be to additionally assign a product index of "healthy use" to a product currently assigned a behavior characteristic index of "luxury orientation", or to delete the product index of "luxury orientation".

Next, in step S150, the second learning processing unit 301d uses the learning data 11b generated in step S145 (which may be the preprocessed learning data 11b) to cause the learning model 12b to learn the correlation between the person data and the behavioral factor index. In this case, the second learning processing unit 301d may use not only the learning data 11b generated in step S145, but also the learning data 11a generated in step S121. Note that, if the person database 20 is updated based on a correction instruction for the behavioral feature index, preprocessing is performed on the learning data 11b with the updated person database 20 reflected.

Next, in step S160, the machine learning unit 301 determines whether a predetermined learning end condition is met when the second learning processing unit 301d causes the learning model 12b to learn the correlation between the person data and the behavioral factor index.

If the machine learning unit 301 determines that the predetermined learning end condition is not met ("No" in step S160), the process returns to step S140. At that time, in step S140, the second generation processing unit 301c inputs the multiple person data acquired in step S110 to the learning model 12b trained by the second learning processing unit 301d in step S150 instead of the learning model 12a trained by the first learning processing unit 301b in step S130, and the machine learning unit 301 repeatedly executes the subsequent processes (steps S140 to S150) until it determines in step S160 that the learning end condition is met.

On the other hand, if the machine learning unit 301 determines that the predetermined learning end condition is met ("Yes" in step S160), then in step S170, the learning model 12b at that time is stored in the storage unit 31 as a trained learning model 12. This ends the series of processes shown in Figures 10 and 11. In the machine learning method, step S110 corresponds to the data acquisition process, steps S120 to S160 correspond to the machine learning process, and step S170 corresponds to the trained model storage process.

In the above description, the series of processes shown in FIG. 10 and FIG. 11 are executed when the worker terminal device 5 receives a learning instruction from the worker 10.
10 and 11 may be executed when it is determined that a predetermined execution condition (for example, when a certain time has elapsed since the previous execution or when the person database 20 is updated) is satisfied. In the above description, when the machine learning unit 301 determines that a predetermined learning end condition is not satisfied ("No" in step S160), the process returns to step S140. However, the process may return to step S120, or the process may be switched between returning to step S120 or returning to step S140 based on an instruction from the operator 10 or a determination by the machine learning unit 301. When returning to step S120, the operator 10 may change, for example, a determination rule for the behavior factor index, and the machine learning unit 301 may execute the process from step S120 onward based on the changed determination rule.

As described above, the machine learning device 3 and machine learning method according to this embodiment can provide a learning model 12 capable of estimating a behavioral factor index that indicates factors that motivate a person to take action, based on person data that records the person's behavior from one or more perspectives. Therefore, by using this learning model 12 by the data processing device 4, it is possible to appropriately assign (infer) a behavioral factor index inherent to the person (user) without requiring the person (user) to input personal information.

The machine learning unit 301 performs machine learning using the learning data 11a generated based on the judgment result according to the judgment rules, and performs machine learning in stages using the learning data 11b generated based on the reception result of the correction instruction for the behavioral factor index, thereby making the learning model 12 learn the correlation between the person data and the behavioral factor index. Therefore, the machine learning of the learning model 12 is performed in stages through the automatic generation of the learning data 11a according to the judgment rules and the correction of the learning data 11b through the annotation work of the worker 10, so that it is possible to generate a highly accurate learning model 12 while reducing the workload of the annotation work of the worker 10.

Furthermore, the machine learning unit 301 updates the behavioral feature index assigned to the feature data in the person database 20 based on the result of receiving the instruction to modify the behavioral feature index, thereby performing machine learning in a state that reflects the updated person database 20. Therefore, through the modification of the behavioral feature index by the annotation work of the worker 10, the accuracy of the behavioral feature index assigned to the feature data included in the person data is improved, so that a more accurate learning model 12 can be generated.

(Data processing method)
12 is a flowchart showing an example of a data processing method by the data processing device 4. The following describes a case where the worker 10 performs a tag assignment instruction operation using the worker terminal device 5, and the data processing device 4 refers to the person database 20 and the trained learning model 12, and assigns behavioral factor indices to the person data.

First, in step S200, the worker terminal device 5 receives an assignment instruction including an assignment condition for assigning a behavioral factor index from the worker 10, and in step S201, transmits the assignment instruction to the machine learning device 3. As the assignment condition, for example, a person data acquisition condition for acquiring person data to which a behavioral factor index is to be assigned from the person database 20 is specified.

Next, in step S210, when the data processing device 4 receives the instruction to assign the behavioral factor index transmitted in step S201, the data acquisition unit 400 accesses the person database 20 to acquire person data that satisfies the conditions for assigning the behavioral factor index included in the assignment instruction, and acquires, for example, multiple person data corresponding to multiple people.

Next, in step S220, the index assignment unit 401 inputs the multiple person data acquired in step S110 into the learning model 12, and assigns the behavioral factor index output from the learning model 12 as the inference result to each person data. In the example of FIG. 12, the index assignment unit 401 assigns "1: health-conscious" to "User P" and "User R" as the behavioral factor index of "health-consciousness" after post-processing based on the comparison of the variable value of the behavioral factor index with a predetermined threshold value, and assigns "0: not health-conscious" to "User Q". Note that the assignment list may include the behavioral factor index (value in the range of 0 to 1) before post-processing in addition to or instead of the behavioral factor index (0 or 1) after post-processing by the index assignment unit 401.

Next, in step S230, the output processing unit 402 provides the behavioral factor indexes (the assignment list in FIG. 12) assigned to the person data in step S220 to, for example, the database device 2, so that the behavioral factor indexes are registered in the behavioral factor index table of the person database 20. This ends the series of processes shown in FIG. 12. In the data processing method, step S210 corresponds to the data acquisition process, step S220 corresponds to the index assignment process, and step S230 corresponds to the output processing process.

In the above, the series of processes shown in FIG. 12 are described as being executed when the worker terminal device 5 receives an assignment instruction from the worker 10. However, the series of processes shown in FIG. 12 may be executed when the data processing device 4 determines that a predetermined execution condition is met (e.g., when a certain amount of time has elapsed since the previous execution or when the person database 20 is updated).

As described above, the data processing device 4 and data processing method according to this embodiment can appropriately assign (infer) behavioral factor indices inherent to a person (user) without requiring the person to input personal information. The behavioral factor indices assigned to each person can then be processed by predetermined statistical processing, visualization processing, etc., and used in various services (product planning, product development, production management, purchasing and ordering management, sales strategy formulation, marketing, advertising, etc.).

Other Embodiments
The present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit and scope of the present invention. All such modifications are included in the technical concept of the present invention.

In the above embodiment, the database device 2, the machine learning device 3, and the data processing device 4 are described as being configured as separate devices, but these three devices may be configured as a single device, or any two of these three devices may be configured as a single device. In addition, at least one of the machine learning device 3 and the data processing device 4 may be incorporated in the worker terminal device 5.

In the above embodiment, a case has been described in which a neural network is adopted as a learning model for realizing machine learning by the machine learning unit 301, but other machine learning models may be adopted. Examples of other machine learning models include tree models such as decision trees and regression trees, ensemble learning models such as bagging and boosting, recurrent neural networks, convolutional neural networks, and neural net models (including deep learning) such as LSTM, clustering models such as hierarchical clustering, non-hierarchical clustering, k-nearest neighbors, and k-means, multivariate analysis models such as principal component analysis, factor analysis, and logistic regression, and support vector machines.

(Inference device, inference method, or inference program)
The present invention can be provided not only in the form of the data processing device 4 (data processing method or data processing program) according to the above embodiment, but also in the form of an inference device (inference method or inference program). In this case, the inference device (inference method or inference program) can include a memory and a processor, and the processor executes a series of processes. The series of processes includes a data acquisition process (data acquisition step) for acquiring person data, and an inference process (inference step) for inferring a behavioral factor index for the person data when the person data is acquired in the data acquisition process.

By providing it in the form of an inference device (inference method or inference program), it can be easily applied to various devices compared to implementing a data processing device 4. It can be naturally understood by those skilled in the art that when the inference device (inference method or inference program) infers a behavioral factor index, the inference method implemented by the index assignment unit can be applied using the trained learning model generated by the machine learning device 3 and machine learning method according to the above embodiment.

1...person information management system, 2...database device, 3...machine learning device,
4... data processing device, 5... operator terminal device, 6... network,
10...worker, 11, 11a, 11b...learning data,
12, 12a, 12b... learning model, 20... person database,
30: control unit, 31: storage unit, 32: communication unit, 33: input unit, 34: display unit,
40: control unit, 41: storage unit, 42: communication unit, 43: input unit, 44: display unit,
120...input layer, 121...intermediate layer, 122...output layer,
300... data acquisition unit, 301... machine learning unit,
301a...first generation processing unit, 301b...first learning processing unit,
301c...second generation processing unit, 301d...second learning processing unit,
310 ... decision rule data, 311 ... machine learning program,
400: data acquisition unit, 401: index assignment unit, 402: output processing unit,
410...Data processing program

Claims (12)

a machine learning unit that inputs a plurality of sets of learning data, each set being composed of person data that records a person's behavior based on one or more viewpoints and a behavioral factor index that indicates a factor that motivates the person to perform the behavior, into a learning model, and causes the learning model to learn a correlation between the person data and the behavioral factor index by machine learning;
A trained model storage unit that stores the trained model in which the correlation is trained by the machine learning unit ,
The machine learning unit is
a first generation processing unit that generates the learning data by determining, for each of the persons, whether or not the person matches the behavioral factor index in accordance with a determination rule defined for the person data based on at least a part of the viewpoint, and assigning the behavioral factor index to the person data based on the determination result;
a first learning processing unit that causes the learning model to learn the correlation by using the learning data generated by the first generation processing unit;
a second generation processing unit that receives a correction instruction for the behavior factor index output from the learning model by inputting the person data into the learning model in which the correlation has been learned by the first learning processing unit, and generates the learning data by assigning the behavior factor index to the person data based on the reception result;
A second learning processing unit that causes the learning model to learn the correlation by using at least the learning data generated by the second generation processing unit.
Machine learning device. The second generation processing unit,
inputting the person data into the learning model in which the correlation has been learned by the first learning processing unit, thereby causing the behavioral factor indexes output from the learning model to be associated with the person data and displayed for each person, and receiving the modification instruction for the displayed behavioral factor indexes;
The machine learning device according to claim 1 . The second generation processing unit,
The person data is input to the learning model in which the correlation is learned by the first learning processing unit.
and displaying the behavioral factor indexes output from the learning model for each person by inputting the data and corresponding to information obtained by performing a predetermined statistical process on the person data, and receiving the instruction to modify the displayed behavioral factor indexes.
The machine learning device according to claim 1 . The second generation processing unit,
the behavior factor index assigned to the person data by the first generation processing unit based on the determination result, and the behavior factor index output from the learning model in which the correlation has been learned by the first learning processing unit by inputting the person data into the learning model, are displayed for each person in association with the person data, and the modification instruction is received for the displayed behavior factor index;
The machine learning device according to claim 1 . The machine learning unit is
When the second learning processing unit has learned the correlation in the learning model, if a predetermined learning end condition is not satisfied,
the second generation processing unit inputs the person data into the learning model in which the correlation has been learned by the second learning processing unit instead of the first learning processing unit, accepts the correction instruction for the behavior factor index output from the learning model, and generates the learning data by assigning the behavior factor index to the person data based on the acceptance result; and the second learning processing unit repeatedly executes a process of having the learning model learn the correlation using at least the learning data generated by the second generation processing unit until the learning end condition is satisfied.
The machine learning device according to claim 1 . a machine learning unit that inputs a plurality of sets of learning data, each set being composed of person data that records a person's behavior based on one or more viewpoints and a behavioral factor index that indicates a factor that motivates the person to perform the behavior, into a learning model, and causes the learning model to learn a correlation between the person data and the behavioral factor index by machine learning;
A trained model storage unit that stores the trained model in which the correlation is trained by the machine learning unit ,
The person data is
a behavioral characteristic index corresponding to the behavior factor index is recorded in association with the assigned characteristic data,
The machine learning unit is
a pre-processing is performed to convert the personal data into features for input to the learning model based on the behavioral feature index assigned to the feature data included in the personal data, and a plurality of sets of pre-processed learning data, which are composed of the features converted from the personal data in the pre-processing and the behavioral factor index included in the personal data, are input to the learning model, thereby causing the learning model to learn the correlation;
The machine learning unit is
a first generation processing unit that generates the learning data by determining, for each of the persons, whether or not the person matches the behavioral factor index in accordance with a determination rule defined for the person data based on at least a part of the viewpoint, and assigning the behavioral factor index to the person data based on the determination result;
a first learning processing unit that causes the learning model to learn the correlation by using the learning data generated by the first generation processing unit and that is the learning data after the preprocessing;
The feature amount converted from the person data in the preprocessing is input to the learning model in which the correlation is learned by the first learning processing unit, and the behavior factor index output from the learning model and the preprocessing data assigned to the feature data included in the person data are output.
a second generation processing unit that receives a correction instruction for at least one of the behavioral feature index and the behavior factor index, and generates the learning data by assigning the behavior factor index to the person data based on the reception result, and corrects the behavioral feature index assigned to the feature data;
a second learning processing unit that causes the learning model to learn the correlation by using the learning data generated by at least the second generation processing unit, the learning data being the preprocessed learning data;
Machine learning device. The second generation processing unit,
inputting the feature amount converted from the person data in the preprocessing into the learning model in which the correlation has been learned by the first learning processing unit, the behavior factor index output from the learning model and the behavior feature index assigned to the feature data included in the person data are displayed for each person in association with the person data, and the modification instruction is received for at least one of the displayed behavior factor index and the behavior feature index.
The machine learning device according to claim 6 . The second generation processing unit,
the feature amount converted from the person data in the preprocessing is input to the learning model in which the correlation has been learned by the first learning processing unit, and the behavior factor index output from the learning model and the behavior feature index assigned to the feature data included in the person data are displayed for each person in association with information obtained by performing a predetermined statistical process on the person data, and the modification instruction is received for at least one of the displayed behavior factor index and the behavior feature index.
The machine learning device according to claim 6. The second generation processing unit,
the behavior factor index assigned to the person data by the first generation processing unit based on the determination result, and the feature amount converted from the person data in the preprocessing are input to the learning model in which the correlation has been learned by the first learning processing unit, and the behavior factor index output from the learning model and the behavior feature index assigned to the feature data included in the person data are associated with the person data and displayed for each person, and the modification instruction is received for at least one of the behavior factor index and the behavior feature index.
The machine learning device according to claim 6. The machine learning unit is
When the second learning processing unit has learned the correlation in the learning model, if a predetermined learning end condition is not satisfied,
the second generation processing unit receives the modification instruction for at least one of the behavior factor index output from the learning model, in which the correlation has been learned by the second learning processing unit instead of the first learning processing unit, by inputting the feature amount converted from the person data in the preprocessing, and the behavior feature index assigned to the feature data included in the person data, and generates the learning data by assigning the behavior factor index to the person data based on the reception result, and modifies the behavior feature index assigned to the feature data; and the second learning processing unit repeatedly executes a process of having the learning model learn the correlation using at least the learning data generated by the second generation processing unit and the learning data after the preprocessing, until the learning end condition is satisfied.
The machine learning device according to any one of claims 6 to 9 . a machine learning process for inputting a plurality of sets of learning data, each set being composed of person data that records a person's behavior based on one or more viewpoints and a behavioral factor index that indicates a factor that motivates the person to perform the behavior, into a learning model, thereby causing the learning model to learn a correlation between the person data and the behavioral factor index by machine learning;
A trained model storage step of storing the trained model in a trained model storage unit, the trained model having the correlation trained by the machine learning step;
The machine learning process includes:
a first generation process step of judging for each of the persons whether or not the person matches the behavioral factor index in accordance with a judgment rule defined for the person data based on at least a part of the viewpoint, and generating the learning data by assigning the behavioral factor index to the person data based on the judgment result;
a first learning process step of causing the learning model to learn the correlation by using the learning data generated by the first generation process step;
a second generation processing step of receiving a correction instruction for the behavior factor index output from the learning model by inputting the person data into the learning model that has learned the correlation by the first learning processing step, and generating the learning data by assigning the behavior factor index to the person data based on the reception result;
and a second learning process step of causing the learning model to learn the correlation by using the learning data generated at least by the second generation process step.
Machine learning methods. a machine learning process for inputting a plurality of sets of learning data, each set being composed of person data that records a person's behavior based on one or more viewpoints and a behavioral factor index that indicates a factor that motivates the person to perform the behavior, into a learning model, thereby causing the learning model to learn a correlation between the person data and the behavioral factor index by machine learning;
A trained model storage step of storing the trained model in a trained model storage unit, the trained model having the correlation trained by the machine learning step;
The person data is
a behavioral characteristic index corresponding to the behavior factor index is recorded in association with the assigned characteristic data,
The machine learning process includes:
a pre-processing is performed to convert the personal data into features for input to the learning model based on the behavioral feature index assigned to the feature data included in the personal data, and a plurality of sets of pre-processed learning data, which are composed of the features converted from the personal data in the pre-processing and the behavioral factor index included in the personal data, are input to the learning model, thereby causing the learning model to learn the correlation;
The machine learning process includes:
a first generation process step of judging for each of the persons whether or not the person matches the behavioral factor index in accordance with a judgment rule defined for the person data based on at least a part of the viewpoint, and generating the learning data by assigning the behavioral factor index to the person data based on the judgment result;
a first learning process step of causing the learning model to learn the correlation by using the learning data generated by the first generation process step and the learning data after the preprocessing;
The feature amount converted from the person data in the preprocessing is input to the learning model that has learned the correlation in the first learning processing step, and a correction instruction is received for at least one of the behavior factor index output from the learning model and the behavior feature index assigned to the feature data included in the person data, and the learning data is generated by assigning the behavior factor index to the person data based on the reception result.
and a second generation process step of correcting the behavioral feature index assigned to the feature data.
and a second learning process step of causing the learning model to learn the correlation by using the learning data generated by at least the second generation process step, the learning data being after the preprocessing.
Machine learning methods.

Images