Work classification estimation method, work classification estimation program, and information processing device

Abstract

This work classification estimation method includes: a step for acquiring the likelihood of a candidate classification from a likelihood model that is a machine learning model; a step for acquiring the appearance probability of the candidate classification from a probability model that is a statistical model; a step for estimating the classification of work executed by a caregiver, on the basis of the likelihood and the appearance probability of the candidate classification; and a step for outputting the results of estimating the classification of the work.

Description

Work Classification Estimation Method, Work Classification Estimation Program, and Information Processing Apparatus

[0001] The present disclosure relates to a work classification estimation method, a work classification estimation program, and an information processing apparatus related to caregiving.

[0002] Conventionally, a method of aggregating work data collected by time study using a device has been known (see, for example, Patent Document 1).

[0003] Japanese Patent Application Laid-Open No. 2003-280726

[0004] There is a need to efficiently collect work data while realizing the maintenance or improvement of the estimation accuracy or precision of work data.

[0005] An object of the present disclosure is to provide a work classification estimation method, a work classification estimation program, and an information processing apparatus that can efficiently collect work data while realizing the maintenance or improvement of the estimation accuracy or precision of work data.

[0006] (1) A method for estimating work categories according to one embodiment of the present disclosure is a method for estimating work categories performed by a care worker in a care setting, comprising the steps of: an information processing device inputting information about the care worker into a likelihood model, which is a machine learning model, and obtaining the likelihood of each candidate category, which represents the likelihood that the work category performed by the care worker is one of a plurality of candidate categories; an information processing device inputting information about the care worker into a probability model, which is a statistical model, and obtaining the probability of occurrence of each candidate category, which represents the probability that the work category performed by the care worker is one of the plurality of candidate categories; an information processing device estimating the work category performed by the care worker based on the likelihood of each candidate category and the probability of occurrence of each candidate category; and an information processing device outputting the estimation result of the work category performed by the care worker. The likelihood model is a machine learning model generated by performing machine learning using training data that associates input data including information about the care worker with ground truth data representing the categories of work performed by the care worker. It is configured to output the likelihood of each candidate category representing the probability that the category of work performed by the care worker is one of several candidate categories when information about the care worker is input. The probability model is a statistical model generated by performing statistical processing on data that associates information about the care worker with the categories of work performed by the care worker. It is configured to output the probability of occurrence of each candidate category corresponding to the conditions that the information about the care worker satisfies when information about the care worker is input.

[0007] (2) In the step of obtaining the probability of occurrence of each candidate category in the work category estimation method described in (1) above, the information processing device may input each of the multiple conditions into the probability model and obtain the probability of occurrence of each candidate category corresponding to each of the multiple conditions from the probability model. In the step of estimating the work category in the work category estimation method, the information processing device may estimate the work category performed by the care worker based on the probability of occurrence of each candidate category corresponding to each of the multiple conditions.

[0008] (3) In the step of estimating the work classification of the work classification estimation method described in (2) above, the information processing device may generate an average occurrence probability of each candidate classification by performing an arithmetic mean or geometric mean of the occurrence probabilities of each candidate classification corresponding to each of the plurality of conditions, and estimate the work classification performed by the care worker based on the average occurrence probability of each candidate classification.

[0009] (4) In the work classification estimation method described in any one of (1) to (3) above, the probability model may be configured to output as the probability of occurrence of each candidate classification the proportion of times the work classification performed by the care worker was one of the candidate classifications when the information about the care worker satisfies at least one condition in the data relating the information about the care worker to the classification of work performed by the care worker.

[0010] (5) In the work category estimation method described in (4) above, the probability model may be configured to output as the probability of occurrence of each candidate category a proportion that is corrected to be small when the proportion of each candidate category is small compared to a correction threshold, not corrected when the proportion of each candidate category is equal to the correction threshold, and large when the proportion of each candidate category is greater than the correction threshold.

[0011] (6) In the work category estimation method described in (5) above, the probability model may be configured to output a normalized ratio as the probability of occurrence of each candidate category, which is corrected to 0 when the proportion of each candidate category is less than or equal to the correction threshold, and corrected to a normalized ratio when the proportion of each candidate category is greater than the correction threshold. The normalized ratio is the reciprocal of the number of candidate categories that satisfy the condition that the proportion of each candidate category is greater than the correction threshold.

[0012] (7) A work classification estimation program according to one embodiment of the present disclosure is a program that causes an information processing device to estimate the classification of work performed by a care worker at a care facility, and causes the information processing device to execute the following steps: input information about the care worker into a likelihood model, which is a machine learning model, and obtain the likelihood of each candidate classification, which represents the likelihood that the classification of work performed by the care worker is one of the multiple candidate classifications; input information about the care worker into a probability model, which is a statistical model, and obtain the probability of occurrence of each candidate classification, which represents the probability that the classification of work performed by the care worker is one of the multiple candidate classifications; estimate the classification of work performed by the care worker based on the likelihood of each candidate classification and the probability of occurrence of each candidate classification; and output the estimation result of the classification of work performed by the care worker. The likelihood model is a machine learning model generated by performing machine learning using training data that associates input data including information about the care worker with ground truth data representing the categories of work performed by the care worker. It is configured to output the likelihood of each candidate category representing the probability that the category of work performed by the care worker is one of several candidate categories when information about the care worker is input. The probability model is a statistical model generated by performing statistical processing on data that associates information about the care worker with the categories of work performed by the care worker. It is configured to output the probability of occurrence of each candidate category corresponding to the conditions that the information about the care worker satisfies when information about the care worker is input.

[0013] (8) An information processing device according to one embodiment of the present disclosure estimates the categories of work performed by a care worker in a care setting. The information processing device includes: a likelihood calculation unit that inputs information about the care worker into a likelihood model, which is a machine learning model, and calculates the likelihood of each candidate category representing the likelihood that the category of work performed by the care worker is one of the multiple candidate categories; a probability calculation unit that inputs information about the care worker into a probability model, which is a statistical model, and calculates the probability of each candidate category representing the probability that the category of work performed by the care worker is one of the multiple candidate categories; an estimation unit that estimates the category of work performed by the care worker based on the likelihood of each candidate category and the probability of each candidate category; and an output unit that outputs the estimation result of the category of work performed by the care worker. The likelihood model is a machine learning model generated by performing machine learning using training data that associates input data including information about the care worker with ground truth data representing the categories of work performed by the care worker. It is configured to output the likelihood of each candidate category representing the probability that the category of work performed by the care worker is one of several candidate categories when information about the care worker is input. The probability model is a statistical model generated by performing statistical processing on data that associates information about the care worker with the categories of work performed by the care worker. It is configured to output the probability of occurrence of each candidate category corresponding to the conditions that the information about the care worker satisfies when information about the care worker is input.

[0014] According to a work category estimation method, work category estimation program, and information processing device according to one embodiment of this disclosure, work data is efficiently collected while maintaining or improving the estimation accuracy or precision of the work data.

[0015] This is a block diagram showing an example of the configuration of the information processing system related to this disclosure. This is a block diagram showing an example of the configuration of the model generation unit. This is a block diagram showing an example of the configuration of the work category estimation unit. This is a table showing an example of the probability of occurrence of a task corresponding to condition A. This is a table showing an example of the probability of occurrence of a task corresponding to condition B. This is a table showing an example of the probability of occurrence of a task corresponding to condition C. This is a table showing an example of the probability of occurrence of a task corresponding to condition D. This is a graph showing an example of the probability of occurrence of a task label C06 by time period. This is a graph showing an example of the probability of occurrence of a task label G01 by time period. This is a graph showing an example of the probability of occurrence of a task label xS3 by time period. This is a block diagram showing an example of the process of narrowing down candidate categories by probability of occurrence. This is a graph showing an example of the likelihood of each of the multiple candidate categories of a task. This is a graph showing an example of narrowing down candidate categories by the narrowing down process. This is a flowchart showing an example of the procedure for generating a likelihood model. This is a flowchart showing an example of the procedure for generating a probability model. This is a flowchart showing an example of the procedure for estimating work categories. This is a diagram showing an example of the hierarchical structure of work categories. This is a diagram showing an independent process. This is a diagram showing a parallel process in which some actions are performed in parallel. This is a diagram showing a parallel process in which the first action is performed in parallel. This is a diagram showing a parallel process in which all actions are performed in parallel. This diagram represents parallel processes where the last action is performed in parallel. This diagram represents interrupt processes. This is a timeline showing the estimated time periods during which care workers performed tasks in each category. This is a timeline showing the time periods during which care workers stayed at each position. This is a bar graph showing the estimated length of time during which care workers performed tasks in each category.

[0016] In the field of long-term care services, a shortage of care workers is a problem. There is a need to improve productivity so that tasks can be performed efficiently even with fewer personnel. A common method for improving productivity in long-term care services is to understand the long-term care service process, which is the sequence of events involved, and to implement technology changes or process improvements.

[0017] To understand the process of care services, it is necessary to collect work data from care workers. Care workers' work can be categorized into several groups. Work data includes information about the location where care workers performed each category of work, or the time or frequency of their work. Work data can sometimes be collected through time studies, where workers visually observe the work. However, this data collection process is burdensome. Efficient data collection is therefore required.

[0018] On the other hand, a high degree of accuracy or precision in estimating work data is required. Therefore, it is necessary to efficiently collect work data while maintaining or improving the accuracy or precision in estimating work data.

[0019] Therefore, this disclosure describes an information processing system 1 (see Figure 1) developed to efficiently collect work data of care workers while maintaining or improving the estimation accuracy or precision of the work data. The information processing system 1 according to this disclosure estimates what tasks a care worker performed based on the care worker's actions or information about the care worker including their actions, and outputs the estimation result. The information processing system 1 according to this disclosure will be described below with reference to the drawings.

[0020] (Outline of Information Processing System 1) As shown in Figure 1, an information processing system 1 according to one embodiment comprises an information processing device 10, a sensor 40, an input device 50, a display device 60, and a database 70.

[0021] The information processing device 10 acquires information about care workers in care facilities and other care settings. The information about care workers may include information about the locations where care workers have moved in care facilities and other care settings, and information about the time spent at each location. Information about the location of care workers is also called location information. Information about the time spent at each location is also called time information. Location information may include locations where care workers have moved and stayed, or locations that care workers have passed through while moving. In other words, location information may include a history of care workers moving to multiple locations. Time information may include the length of time that care workers have stayed at each location in the past, or the time that care workers have started and ended their stay at each location in the past. The time that care workers started staying at a certain location corresponds to the time that care workers completed their movement to that location. The time that care workers ended their stay at a certain location corresponds to the time that care workers started moving away from that location. Location information and time information may be generated based on measurement results from the sensor 40, as described later. Information concerning care workers may further include the schedule of tasks related to caregiving or indirect caregiving tasks that care workers will perform in the caregiving setting. The schedule of tasks performed by care workers may also be simply referred to as the work schedule.

[0022] The information processing device 10 estimates the tasks performed by the care worker based on information about the care worker. Furthermore, as will be described later, the tasks performed by the care worker may be classified into multiple categories. The tasks performed by the care worker may include tasks performed while the care worker moves to multiple locations. In other words, one task performed by the care worker may be a task that combines the actions performed by the care worker at each of the multiple locations. One task performed by the care worker may be a task that combines the care worker moving between multiple locations and the actions performed by the care worker at each of the multiple locations. The information processing device 10 may generate likelihood models and probability models, as will be described later, for example by performing machine learning, and use the generated likelihood models and probability models to estimate the categories of tasks performed by the care worker.

[0023] The information processing device 10 may display the tasks in each category, which are estimated to have been performed by the care worker, as a timeline on the display device 60. The information processing device 10 may also aggregate the time or frequency of the tasks in each category, which are estimated to have been performed by the care worker, and display the aggregated results on the display device 60.

[0024] (Example of Information Processing System 1 Configuration) The following describes an example of the configuration of Information Processing System 1.

[0025] <Information Processing Device 10> As shown in Figure 1, the information processing device 10 comprises a work category definition unit 12, a location definition unit 14, a schedule management unit 16, a model generation unit 20, and a work category estimation unit 30. The work category definition unit 12 defines the categories of work performed by the care worker. The location definition unit 14 defines the locations where the care worker moved and stayed, or the locations where the care worker passed through while moving, or the locations where the care worker performed work. The schedule management unit 16 manages the care worker's work schedule. The model generation unit 20 generates likelihood models and probability models used to estimate the categories of work performed by the care worker. The work category estimation unit 30 estimates the categories of work performed by the care worker.

[0026] Each of the work category definition unit 12, the location definition unit 14, and the schedule management unit 16 may be configured to include one or more processors. The processors may include, but are not limited to, general-purpose processors or dedicated processors specialized for specific processing. Each of the work category definition unit 12, the location definition unit 14, and the schedule management unit 16 may be configured to include one or more dedicated circuits. The dedicated circuits may include, for example, FPGAs (Field-Programmable Gate Arrays) or ASICs (Application Specific Integrated Circuits). Each of the work category definition unit 12, the location definition unit 14, and the schedule management unit 16 may be configured to include a dedicated circuit instead of a processor, or to include a dedicated circuit together with a processor. Each of the work category definition unit 12, the location definition unit 14, and the schedule management unit 16 may be configured as separate components. At least a part of the work category definition unit 12, the location definition unit 14, and the schedule management unit 16 may be configured as a single unit.

[0027] Each of the work division definition unit 12, the location definition unit 14, and the schedule management unit 16 may include a storage unit. The storage unit may store programs executed by each of the work division definition unit 12, the location definition unit 14, and the schedule management unit 16, or information or data used in the processing of each of the work division definition unit 12, the location definition unit 14, and the schedule management unit 16. The storage unit may include, for example, semiconductor memory, magnetic memory, or optical memory. The storage unit may include an electromagnetic storage medium such as a magnetic disk. The storage unit may include a non-temporary computer-readable medium. The storage unit may function as, for example, main memory, auxiliary memory, or cache memory. The storage unit may function as the work memory of each of the work division definition unit 12, the location definition unit 14, and the schedule management unit 16. The memory unit may be configured integrally with the processors of the work category definition unit 12, the location definition unit 14, and the schedule management unit 16, or it may be configured separately from the processors of the work category definition unit 12, the location definition unit 14, and the schedule management unit 16.

[0028] <<Model Generation Unit 20>> As shown in Figure 2, the model generation unit 20 includes an action measurement unit 21, a position measurement unit 22, a time measurement unit 23, a likelihood model generation unit 27, and a probability model generation unit 28.

[0029] Each component of the model generation unit 20 may include one or more processors or one or more dedicated circuits, similar to the work category definition unit 12, the location definition unit 14, and the schedule management unit 16. The model generation unit 20 may also include a storage unit, similar to the work category definition unit 12, the location definition unit 14, and the schedule management unit 16.

[0030] <<Work Classification Estimation Unit 30>> As shown in Figure 3, the work classification estimation unit 30 includes a position measurement unit 31, a position recognition unit 32, a time measurement unit 33, a likelihood calculation unit 34, a probability calculation unit 35, an estimation unit 36, and an output unit 37.

[0031] Each component of the work category estimation unit 30 may include one or more processors or one or more dedicated circuits, similar to each of the work category definition unit 12, the location definition unit 14, and the schedule management unit 16, or each of the components of the model generation unit 20. The work category estimation unit 30 may also include a storage unit, similar to each of the work category definition unit 12, the location definition unit 14, and the schedule management unit 16, or the model generation unit 20.

[0032] The work category definition unit 12, the location definition unit 14, the schedule management unit 16, the model generation unit 20, and the work category estimation unit 30 may each be configured as separate units. At least a part of the work category definition unit 12, the location definition unit 14, the schedule management unit 16, the model generation unit 20, and the work category estimation unit 30 may be configured as an integrated unit.

[0033] <<Other Configurations of the Information Processing Device 10>> The information processing device 10 may include a communication module configured to communicate with other devices such as a sensor 40, an input device 50, a display device 60, or a database 70. The communication module may be configured to communicate with other devices via a network, or it may be configured to communicate with other devices via P2P (Peer to Peer) without a network. The communication module may be configured to communicate with other devices via wired or wireless connections. The communication module may support mobile communication standards such as 4G (4th Generation) or 5G (5th Generation). The communication module may support communication standards such as LAN (Local Area Network). The communication module may support wired or wireless communication standards. The communication module is not limited to these and may support various communication standards.

[0034] The information processing device 10 may be configured to include at least one server. The information processing device 10 may be implemented as a computer such as a desktop PC (Personal Computer), a notebook PC, or a tablet PC. The information processing device 10 may be implemented as a smartphone or a tablet. The information processing device 10 is not limited to these examples and may include various devices.

[0035] The information processing device 10 may be configured as a device equipped with a model generation unit 20 and a device equipped with a work category estimation unit 30. The information processing device 10 equipped with the model generation unit 20 is also referred to as a model generation device. The information processing device 10 equipped with the work category estimation unit 30 is also referred to as a work category estimation device. The information processing device 10 may not be equipped with a model generation unit 20 and may acquire likelihood models and probability models generated by an external device.

[0036] <Sensor 40> Sensor 40 may be configured to detect the location of a care worker in a care facility or other care setting as information about the care worker. Sensor 40 outputs the detection result of the care worker's location to the information processing device 10. Sensor 40 may include, for example, a location measurement device that uses radio waves of wireless communication installed in the care setting to detect the location information of the care worker. Sensor 40 may include a human presence sensor or camera installed in a location to be targeted for detecting the location of the care worker, so as to be able to detect the location of the care worker in the care setting. Sensor 40 may include a location measurement device that uses a satellite positioning system such as GPS (Global Positioning System). Sensor 40 may include a location measurement device that uses an autonomous behavior measurement system (PDR). Sensor 40 is not limited to these examples and may include various other devices.

[0037] Sensor 40 may be configured to detect the body movements or posture of a care worker as information about the care worker. Sensor 40 may include, for example, an acceleration sensor or an angular velocity sensor worn by the care worker. Sensor 40 may include, for example, a depth camera that photographs the care worker and generates point cloud data of the care worker. Sensor 40 is not limited to these examples and may include various other devices such as magnetic sensors, barometric pressure sensors, BLE (Bluetooth Low Energy) sensors, or UWB (Ultra-Wide Band) sensors.

[0038] The sensor 40 may be configured to measure time. The sensor 40 may output to the information processing device 10 the detection result of information about the care worker and the time when the information was detected, in association with each other.

[0039] <Input device 50> The input device 50 accepts information input. The input device 50 may include, for example, a pointing device such as a touch sensor or a mouse. The input device 50 may include physical keys. The input device 50 may include an audio input device such as a microphone. The input device 50 may include a display device that displays the content of the input. The input device 50 may be configured as a touch display in which the display device and the touch panel are integrated.

[0040] The input device 50 may be implemented as, for example, a computer such as a desktop PC, notebook PC, or tablet PC, or as a terminal such as a smartphone or tablet PC. The input device 50 is not limited to these examples and may be implemented as various devices.

[0041] <Display device 60> The display device 60 displays information generated by the information processing device 10. The display device 60 may be configured to include various displays such as a liquid crystal display, an organic EL (Electro-Luminescence) or inorganic EL display, or an LED (Light Emitting Diode) display, so that the information can be displayed as characters, symbols, or images.

[0042] The display device 60 may be implemented as, for example, a computer such as a desktop PC, notebook PC, or tablet PC, or as a terminal such as a smartphone or tablet. The display device 60 is not limited to these examples and may be implemented as various devices.

[0043] <Database 70> Database 70 may store information about care workers collected by the information processing device 10. Database 70 may store the estimated results of the care worker's work category estimated by the information processing device 10.

[0044] The database 70 may be configured to include at least one server or storage device. The database 70 may be configured to include, for example, a semiconductor memory, a magnetic memory, an optical memory, or the like. The database 70 may also be configured to include an electromagnetic storage medium such as a magnetic disk. The database 70 may be configured to include a communication module that is communicably configured with other devices such as the information processing device 10.

[0045] The information processing system 1, the information processing device 10, or the database 70 described above may be realized using a cloud service or may be realized in an on-premises environment.

[0046] (Example operation of estimating the category of work) Hereinafter, an example operation in which the information processing device 10 estimates the category of work performed by a caregiver in the information processing system 1 will be described.

[0047] In the information processing device 10, the work category definition unit 12 defines a plurality of categories for classifying the work performed by the caregiver. The work performed by the caregiver may be classified, for example, by the type of care service or the like. Examples of the categories for classifying the work performed by the caregiver will be described later. The information processing device 10 may obtain the definition of the categories for classifying the work performed by the caregiver from an external device without including the work category definition unit 12.

[0048] As will be described later, the information processing device 10 estimates the category of work performed by the caregiver based on information about the caregiver by the work category estimation unit 30. The work category estimation unit 30 may select and output at least one category as an estimation result of the category of work performed by the caregiver from among the plurality of categories defined by the work category definition unit 12. The plurality of categories defined by the work category definition unit 12 are categories that are candidates for being selected as the estimation result of the category of work performed by the caregiver, and are also referred to as candidate categories.

[0049] Information regarding care workers includes the locations where care workers move. In the information processing device 10, the location definition unit 14 defines the locations where care workers move. The locations where care workers move may include locations where care workers stay or locations where care workers pass through. The locations where care workers move may be defined as places classified by the function of providing care services within the care facility. The locations where care workers move may be defined as metadata such as, for example, the rooms of care service users, dining rooms, bathrooms, toilets, recreation rooms, or corridors. The locations where care workers move may also be defined as coordinate information that can be associated with the coordinates of a floor plan within the care facility. The information processing device 10 may obtain the definition of the locations where care workers move from an external device without having a location definition unit 14.

[0050] Information regarding care workers includes their work schedules. In the information processing device 10, the schedule management unit 16 manages the care workers' work schedules. The care workers' work schedules may include information regarding the care workers' shifts. Care workers' shifts may be defined as early shifts or late shifts, or day shifts or night shifts, etc. The care workers' work schedules may include information that associates the category of the scheduled work with the time or place where the work is scheduled to be performed. The schedule management unit 16 may set work schedules based on input from the care workers themselves or from the managers or supervisors of the care site. The schedule management unit 16 may also set work schedules based on the work plan of the care site. The schedule management unit 16 may also set work schedules based on the actual work performed.

[0051] <Example of operation of the model generation unit 20> Below, an example of the operation in which the model generation unit 20 of the information processing device 10 generates a likelihood model and a probability model will be explained with reference to Figure 2. The likelihood model and the probability model are models used by the work category estimation unit 30 to estimate the categories of work performed by the care worker.

[0052] The likelihood model is configured to calculate the likelihood that the work performed by a care worker falls into one of several candidate categories, based on information about the care worker. The likelihood that the work performed by the care worker falls into one of several candidate categories is also called the likelihood for each candidate category. As will be described later, the likelihood model is a machine learning model generated by performing machine learning. The likelihood model is configured to output the likelihood for each candidate category when information about the care worker is input.

[0053] The probability model is configured to calculate the probability that the work performed by a care worker falls into one of several candidate categories, based on information about the care worker. The probability that the work performed by a care worker falls into one of several candidate categories is also called the probability of occurrence for each candidate category. As will be described later, the probability model is a statistical model generated by performing statistical processing on data that associates information about care workers with the categories of work performed by the care workers. The probability model is configured to calculate the probability of occurrence for each candidate category corresponding to specific conditions when the information about the care worker satisfies certain conditions. In other words, when information about a care worker is input, the probability model is configured to output the probability of occurrence for each candidate category corresponding to the conditions that the information about the care worker satisfies.

[0054] <<Acquiring Information Used for Model Generation>> An example of how to acquire information used to generate likelihood models and probability models is explained.

[0055] The behavior measurement unit 21 acquires information regarding the care worker's behavior. The care worker's behavior may be measured by the sensor 40. The sensor 40 may measure the care worker's behavior by detecting the movement of various parts of the care worker's body. When the care worker's behavior is measured by the sensor 40, the behavior measurement unit 21 may acquire information regarding the care worker's behavior from the sensor 40.

[0056] The behavior measurement unit 21 obtains the definition of the work category performed by the care worker from the work category definition unit 12. The behavior measurement unit 21 associates the information regarding the care worker's behavior with at least one of the multiple work categories performed by the care worker. The work category performed by the care worker may be associated with the information regarding the care worker's behavior by inputting it into the input device 50 by the care worker themselves or another worker.

[0057] The behavior measurement unit 21 outputs the classification of tasks performed by the care worker, which is associated with information about the care worker's behavior, as the correct behavior to the likelihood model generation unit 27 and the probability model generation unit 28.

[0058] The position measurement unit 22 obtains the definition of the location to which the care worker moves from the position definition unit 14. The position measurement unit 22 obtains the care worker's location information. This location information includes location information where the care worker has stayed, as described later. The position measurement unit 22 may obtain metadata defining the location to which the care worker moves as the care worker's location information. The position measurement unit 22 may obtain the care worker's coordinate information, convert the coordinate information into metadata, and obtain it as the care worker's location information. In other words, the care worker's location information may include metadata defining the location to which the care worker moves. The care worker's location information may include the care worker's coordinate information. The care worker's location information may include categories of work that can be performed at each location to which the care worker can move. The care worker's location information may include information on equipment installed at each location to which the care worker can move.

[0059] The position of the care worker may be measured by the sensor 40. The sensor 40 may measure the position of the care worker, for example, by detecting the position of the care worker as coordinates. The sensor 40 may also measure the position of the care worker by detecting the presence of the care worker at a specific location in a care facility or other care setting, and the location where the presence of the care worker is detected is taken as the position of the care worker. When the position of the care worker is measured by the sensor 40, the position measurement unit 22 may acquire the position information of the care worker from the sensor 40.

[0060] The position of the care worker may be measured by the care worker themselves or by another worker inputting the position into the input device 50. When the care worker's position is input into the input device 50, the position measurement unit 22 may acquire the care worker's position information from the input device 50.

[0061] The position measurement unit 22 outputs the location information of the care worker, specifically the location where the care worker moved or stayed, to the likelihood model generation unit 27 and the probability model generation unit 28.

[0062] Care workers may move to multiple locations. The position measurement unit 22 may output information about the multiple locations to which the care worker has moved as care worker position information to the likelihood model generation unit 27 and the probability model generation unit 28. The behavior measurement unit 21 acquires information about the care worker's behavior at at least one of the multiple locations to which the care worker has moved. When the behavior measurement unit 21 acquires information about the care worker's behavior at one location, it associates the information about the care worker's behavior at that location with the category of work performed by the care worker at that location. When the behavior measurement unit 21 acquires information about the care worker's behavior at two or more locations, it associates the information about the care worker's behavior at each location with the category of work performed by the care worker at each location. The behavior measurement unit 21 outputs information that associates the care worker's behavior at each location with the category of work performed by the care worker to the likelihood model generation unit 27 and the probability model generation unit 28.

[0063] The care worker's location information may include information about the location where the care worker was located before the location where the behavior measurement unit 21 acquired information about the care worker's behavior. The care worker's location information may include information about the location to which the care worker moved after the location where the behavior measurement unit 21 acquired information about the care worker's behavior. The care worker's location information may include categories of work that can be performed at at least one of the multiple locations to which the care worker has moved. The care worker's location information may include information about equipment installed at at least one of the multiple locations to which the care worker has moved.

[0064] The time measurement unit 23 acquires time information of the care worker. The time information includes the time the care worker stayed at a single location if the care worker moved to that location. The time information includes the time the care worker stayed at each location if the care worker moved to multiple locations. The time the care worker stayed at a particular location may be expressed as the length of time the care worker stayed at that location. The time the care worker stayed at a particular location may be expressed by at least one of the time the care worker started staying at that location or the time the care worker ended staying at that location, or a combination of these.

[0065] The time a care worker spends at a particular location may be measured by the sensor 40, as described later. When the time a care worker spends at a particular location is measured by the sensor 40, the time measurement unit 23 may acquire the care worker's time information from the sensor 40. Thus, the sensor 40 may be equipped with a time measurement function. The time measurement unit 23 may acquire the time a care worker spends at one location or at each of multiple locations using the sensor 40.

[0066] The time a care worker spends in a particular location may be measured by the care worker themselves or another care worker inputting the information into the input device 50. When the time a care worker spends in a particular location is input into the input device 50, the time measurement unit 23 may acquire the care worker's time information from the input device 50.

[0067] The time measurement unit 23 outputs the time spent at each of the one or more locations the care worker moved to, i.e., the time spent at each location, as time information for the care worker to the likelihood model generation unit 27 and the probability model generation unit 28.

[0068] When a caregiver moves to multiple locations, the time measurement unit 23 may output information regarding the time the caregiver spent at each of the multiple locations as time information to the likelihood model generation unit 27 and the probability model generation unit 28. The caregiver's time information may include information regarding the time the caregiver spent at the location where the caregiver was before the location where the behavior measurement unit 21 acquired information regarding the caregiver's behavior. The caregiver's time information may also include information regarding the time the caregiver spent at the location to which the caregiver moved after the location where the behavior measurement unit 21 acquired information regarding the caregiver's behavior.

[0069] The schedule management unit 16 outputs the care worker's work schedule to the likelihood model generation unit 27 and the probability model generation unit 28. The work schedule output from the schedule management unit 16 helps the model generation unit 20 to reflect the likelihood of correct behavior output from the behavior measurement unit 21 in the model when generating the likelihood model and the probability model.

[0070] The information used for model generation may include information recording the actual duration or number of times a task was performed for each defined location in the care setting where the categories of tasks performed by care workers are estimated. The information used for model generation may include information recording the duration or number of times a planned task was performed for each defined location in the care setting. The information used for model generation may include information defining tasks that are expected to occur for each defined location in the care setting. The location-specific information used for model generation may include information recorded by time studies or information estimated by machine learning models. The locations defined in the care setting may include the target locations for time studies or indoor positioning, or the locations where the task is planned to be performed.

[0071] The information used for model generation may include information recording the actual duration or number of tasks performed for each time slot defined in the caregiving setting. The information used for model generation may include information recording the duration or number of tasks planned for each time slot defined in the caregiving setting. The information used for model generation may include information defining tasks expected to occur for each time slot defined in the caregiving setting. The information for each time slot used for model generation may include information recorded by time studies or information estimated by machine learning models. The time slots defined in the caregiving setting may include the time slots targeted by time studies or indoor positioning, or the time slots in which tasks are planned to be performed.

[0072] The information used to generate the model may include information recording the actual duration or number of tasks performed for each role category defined in the caregiving setting. The information used to generate the model may include information recording the duration or number of tasks planned for each role category defined in the caregiving setting. The information used to generate the model may include information defining tasks expected to occur for each role category defined in the caregiving setting. The information for each role category used to generate the model may include information recorded by time studies or information estimated by machine learning models. The role categories defined in the caregiving setting may include, for example, the scope of work that a caregiver is responsible for, or the job title of a caregiver.

[0073] <<Model Generation>> The following describes an example of how to generate likelihood models and probability models based on information about care workers.

[0074] The likelihood model generation unit 27 and the probability model generation unit 28 acquire the classification of tasks performed by care workers as correct actions from the action measurement unit 21.

[0075] The likelihood model generation unit 27 and the probability model generation unit 28 acquire information about the care worker. The likelihood model generation unit 27 and the probability model generation unit 28 may acquire the location where the care worker moved from the location measurement unit 22. The likelihood model generation unit 27 and the probability model generation unit 28 may acquire the time the care worker stayed at the location where they moved from the time measurement unit 23. In other words, the likelihood model generation unit 27 and the probability model generation unit 28 may acquire the location where the care worker moved and the time the care worker stayed at that location as information about the care worker. The likelihood model generation unit 27 and the probability model generation unit 28 may acquire the care worker's work schedule from the schedule management unit 16. The likelihood model generation unit 27 and the probability model generation unit 28 may acquire all or some of the above-mentioned information as information about the care worker.

[0076] The likelihood model generation unit 27 generates a likelihood model by using information about care workers as input data and performing learning using the tasks performed by care workers, i.e., correct actions, as correct data. For example, the likelihood model generation unit 27 may use data combining the location where the care worker moved and the time the care worker stayed at that location as input data. The data used as input data is not limited to this example and may be single data or various other combinations. The likelihood model generation unit 27 may generate a likelihood model by performing learning using data that associates location information and time information included in the input data with the categories of tasks performed by care workers included in the correct data, i.e., correct actions. The likelihood model generation unit 27 outputs the generated likelihood model to the task category estimation unit 30.

[0077] When a care worker moves to multiple locations, the likelihood model generation unit 27 may acquire location information from the location measurement unit 22, which includes information about the multiple locations the care worker has moved to, and time information from the time measurement unit 23, which includes information about the time the care worker spent at each of the multiple locations the care worker has moved to. The combined information of location information including information about the multiple locations the care worker has moved to and time information including information about the time the care worker spent at each of the multiple locations the care worker has moved to is also referred to as the care worker's movement history information. In other words, when a care worker moves to multiple locations, the likelihood model generation unit 27 can acquire the care worker's movement history information and the categories of work performed by the care worker at each location included in the movement history information.

[0078] The likelihood model generation unit 27 may perform learning using data that combines the care worker's movement history information and information regarding the time the care worker stayed at each location, i.e., time information, with the categories of work performed by the care worker at each location included in the movement history information, and generate a likelihood model. In this case, the generated likelihood model may be configured to accept the care worker's movement history information as input and output the likelihood of each candidate category at at least one of the multiple locations to which the care worker moved included in the care worker's movement history information. In other words, the likelihood model is configured to accept location information regarding the multiple locations to which the care worker moved and time information regarding the time the care worker stayed at each location as input, and to output the likelihood of each candidate category at at least one of the multiple locations.

[0079] The probability model generation unit 28 calculates a probability model by performing statistical processing on data that associates information about care workers with the categories of tasks performed by the care workers, i.e., correct actions. The data that associates information about care workers with the categories of tasks performed by the care workers, i.e., correct actions, is also called the data to be processed.

[0080] As described above, the probability model is configured to calculate the probability of occurrence of each candidate category corresponding to a specific condition, given that the information about the care worker satisfies that condition. The probability model generation unit 28 may use data as processing data that associates information indicating whether a specific condition has been met regarding the information about the care worker with information indicating the category of work performed by the care worker, i.e., the correct action. The information indicating whether a specific condition has been met regarding the information about the care worker is also called prior information. In other words, the processing data may include data that associates prior information with the correct action when the specific condition represented by that prior information is met.

[0081] When the probability model generation unit 28 uses prior information X as the data to be processed, it may use information representing a condition that corresponds to the logical negation of a specific condition represented by prior information X as prior information X' that is paired with prior information X.

[0082] The prior information may indicate whether one condition is met regarding the information of the care worker, or whether two or more conditions are met. For example, suppose conditions A to D are defined as conditions that must be met regarding the information of the care worker. The prior information may indicate whether one of conditions A to D is met, or whether two or more of conditions A to D are met. The prior information may indicate which conditions are met regarding the information of the care worker. The prior information may indicate which of conditions A to D are met and which are not.

[0083] The prior information may include multiple pieces of information corresponding to each of the multiple conditions defined as conditions that must be met in the information about care workers. The prior information may include multiple pieces of information corresponding to each of the combinations of multiple conditions. The data to be processed may include data that associates each of the multiple pieces of prior information with the correct behavior.

[0084] The probability model generation unit 28 may use data that associates prior information with correct actions in a time series as the data to be processed. The time-series associated data may be data that associates prior information with correct actions recorded for each time period defined in the care setting.

[0085] The probability model generation unit 28 calculates the number of correct actions associated with prior information indicating that a specific condition has been met in the data to be processed. The correct actions associated with prior information indicating that a specific condition has been met represent which of several candidate categories the task performed by the care worker when the specific condition was met falls into. The probability model generation unit 28 calculates the number of correct actions associated with prior information indicating that a specific condition has been met in the data to be processed for each category of correct actions, and by dividing this by the number of times the specific condition was met, it can calculate the probability that the task performed by the care worker when the specific condition was met falls into each of the several candidate categories, i.e., the probability of occurrence of each candidate category.

[0086] The probability model generation unit 28 may calculate the total time spent performing correct actions associated with prior information indicating that specific conditions have been met in the data to be processed. The probability model generation unit 28 calculates the total time spent performing correct actions associated with prior information indicating that specific conditions have been met in the data to be processed for each category of correct actions, and by dividing this by the total time spent meeting the specific conditions, it can calculate the probability that the task performed by the care worker when the specific conditions were met falls into one of several candidate categories, i.e., the probability of each candidate category occurring.

[0087] In other words, the probability model may be configured to output the probability of occurrence of each candidate category, which is the proportion of times the care worker performed a task in a given candidate category when the information about the care worker satisfies at least one condition, in data that associates information about the care worker with the categories of tasks performed by the care worker, i.e., the correct actions.

[0088] The probability model may be configured to output the probability of occurrence of each candidate category as a proportion that has been corrected so that the probability of occurrence of each candidate category corresponding to at least one condition is small when the proportion of each candidate category is small, uncorrected when the proportion of each candidate category is equal to the correction threshold, and large when the proportion of each candidate category is greater than the correction threshold.

[0089] The probability model may be configured to output the probability of occurrence of each candidate category as follows: for each candidate category corresponding to at least one condition, the probability is corrected to 0 if the proportion of each candidate category is below a correction threshold, and corrected to a normalized value if the proportion of each candidate category is greater than the correction threshold. The normalized value is the reciprocal of the number of candidate categories that satisfy the condition that the proportion of each candidate category is greater than the correction threshold. In other words, the normalized value is a coefficient that adjusts so that the sum of the probabilities of occurrence for each candidate category is 1. When the correction threshold is set to 0, the corrected probabilities of occurrence become uniform at the normalized value for candidate categories with a probability greater than 0.

[0090] The probability model generation unit 28 may assume multiple conditions as specific conditions and calculate the probability of occurrence of each candidate category for each assumed condition. The probability model generation unit 28 may generate a probability model as a database-type model that stores the probability of occurrence of each candidate category corresponding to each of the multiple conditions. The probability model may be configured to accept input information about care workers, extract the conditions that the information about care workers satisfies, and output the probability of occurrence of each candidate category corresponding to the extracted conditions. In other words, the probability model may be configured to output the probability of occurrence of each candidate category corresponding to each of the multiple conditions that the information about care workers satisfies when the information about care workers satisfies multiple conditions.

[0091] The probability model may be configured to accept input information about care workers, determine whether the information about care workers satisfies the conditions corresponding to the prior information, and output the probability of occurrence of each candidate category corresponding to the conditions that the information about care workers satisfies among the conditions corresponding to the prior information. For example, suppose conditions A to D are set as the conditions corresponding to the prior information. In this case, the probability model may be configured to determine whether the information about care workers satisfies each of conditions A to D, and output the probability of occurrence of each candidate category corresponding to the conditions that the information about care workers satisfies among conditions A to D.

[0092] The probability model generation unit 28 outputs the generated probability model to the work area estimation unit 30.

[0093] <Example of Operation of the Work Classification Estimation Unit 30> Below, an example of the operation of the work classification estimation unit 30 of the information processing device 10, which estimates the classification of work performed by a care worker using a likelihood model and a probability model, will be explained with reference to Figure 3.

[0094] The position measurement unit 31 obtains the definition of the location where the care worker moves or stays from the position definition unit 14. The position measurement unit 31 measures the location where the care worker moves or stays and outputs the measurement result of the location where the care worker moved or stayed to the position recognition unit 32. The position measurement unit 31 obtains the coordinate information of the care worker as the measurement result of the location where the care worker moved or stayed. Based on the coordinate information of the care worker obtained by the position measurement unit 31, the position recognition unit 32 recognizes the position of the care worker as metadata defined in the position definition unit 14. In other words, the position recognition unit 32 may generate metadata of the location where the care worker moved as the result of recognizing the care worker's position. The position recognition unit 32 outputs the recognition result of the care worker's position to the likelihood calculation unit 34 and the probability calculation unit 35. The position recognition unit 32 may output the recognition result of each of the multiple locations included in the care worker's movement history to the likelihood calculation unit 34 and the probability calculation unit 35. The position measurement unit 31 and the position recognition unit 32 may be configured as a single unit.

[0095] The time measurement unit 33 measures the time spent by the care worker at one location or multiple locations, i.e., the time spent at each location. The time measurement unit 33 outputs the measurement results of the time spent by the care worker at one location or multiple locations, i.e., the time spent at each location, to the likelihood calculation unit 34 and the probability calculation unit 35. The time measurement unit 33 may be configured identically to or similarly to the time measurement unit 23 of the model generation unit 20.

[0096] The schedule management unit 16 outputs the work schedules of care workers to the likelihood calculation unit 34 and the probability calculation unit 35. The work schedules output from the schedule management unit 16 help the work category estimation unit 30 estimate likelihoods and probabilities using the likelihood model and the probability model.

[0097] The likelihood calculation unit 34 operates the likelihood model generated by the model generation unit 20. As described above, the likelihood model is configured to accept input information about the care worker and output the likelihood that the work performed by the care worker is one of several candidate categories, i.e., the likelihood of each candidate category. The likelihood calculation unit 34 acquires information about the care worker from the position recognition unit 32, the time measurement unit 33, or the schedule management unit 16, and inputs the acquired information into the likelihood model. The likelihood calculation unit 34 may input all the information acquired as information about the care worker into the likelihood model, or it may input only some of the acquired information into the likelihood model. The likelihood calculation unit 34 calculates the likelihood of each candidate category by acquiring the likelihood of each candidate category output from the likelihood model. The likelihood calculation unit 34 outputs the calculated likelihood of each candidate category to the estimation unit 36.

[0098] The likelihood calculation unit 34 may calculate the likelihood of each candidate category by inputting information combining the care worker's location information and time information into the likelihood model as the care worker's movement history information. The likelihood calculation unit 34 may obtain the likelihood of each candidate category at at least one location included in the care worker's movement history information output from the likelihood model.

[0099] When a care worker moves between multiple locations, the care worker's location information included in the movement history information may include information about the locations where the care worker was located before the location where the care worker performed the task for which classification is to be estimated. Furthermore, the care worker's location information included in the movement history information may include information about the locations where the care worker was located after the location where the care worker performed the task for which classification is to be estimated. The care worker's time information included in the movement history information may include information about the time the care worker spent at the location where the care worker was located before the location where the care worker performed the task for which classification is to be estimated. Furthermore, the care worker's time information included in the movement history information may include information about the time the care worker spent at the locations where the care worker moved after the location where the care worker performed the task for which classification is to be estimated.

[0100] The probability calculation unit 35 operates the probability model generated by the model generation unit 20. As described above, the probability model is configured to accept input information about care workers and output the probability of occurrence of each candidate category corresponding to the conditions satisfied by the information about care workers. The probability calculation unit 35 acquires information about care workers from the location recognition unit 32, the time measurement unit 33, or the schedule management unit 16 and inputs it into the probability model. The probability calculation unit 35 may input all the information acquired as information about care workers into the probability model, or it may input only some of the acquired information into the probability model. The probability calculation unit 35 calculates the probability of occurrence of each candidate category by acquiring the probability of occurrence of each candidate category output from the probability model. The probability calculation unit 35 outputs the calculated probability of occurrence of each candidate category to the estimation unit 36.

[0101] The estimation unit 36 ​​obtains the likelihood of each candidate category calculated by the likelihood calculation unit 34 and the probability of occurrence of each candidate category calculated by the probability calculation unit 35. The estimation unit 36 ​​considers the likelihood of each candidate category as the prior probability P(H)_t and the probability of occurrence of each candidate category as the corrected probability P(D|H)_t, and calculates the posterior probability P(H|D)_t by performing Bayesian updating. The posterior probability P(H|D)_t represents the corrected likelihood of each candidate category, which is obtained by correcting the likelihood of each candidate category. The relationship between the prior probability, the corrected probability, and the posterior probability is expressed by the following equation (1): P(H|D)_t = {P(D|H)_t × P(H)_t} / P(D)_t (1) In equation (1), P(D)_t is the marginal likelihood. The marginal likelihood is a coefficient used to normalize so that the sum of the corrected likelihoods of each candidate category equals 1.

[0102] The estimation unit 36 ​​extracts the candidate category with the highest likelihood among the corrected likelihoods of each candidate category, and estimates the extracted candidate category as the category of work performed by the care worker.

[0103] The probability model may output the probability of occurrence of each candidate category corresponding to each of the conditions that the information about the care worker satisfies. The estimation unit 36 ​​may calculate the average probability of occurrence of each candidate category from the probability of occurrence of each candidate category corresponding to each of the conditions. For example, the estimation unit 36 ​​may calculate the average probability of occurrence of each candidate category by calculating the arithmetic mean or geometric mean of the probability of occurrence of each candidate category corresponding to each of the conditions. The estimation unit 36 ​​may consider the average probability of occurrence of each candidate category as a corrected probability and calculate the posterior probability, i.e., the corrected likelihood of each candidate category.

[0104] The output unit 37 outputs the results of the estimation unit 36's estimation of the care worker's work category, which can then be displayed on the display device 60 or stored in the database 70.

[0105] (Example of estimating work divisions) A specific example is described below.

[0106] <Example of the Probability Model> Figures 4A to 4D show examples of the probability of occurrence for each candidate category when the information about the care worker satisfies each of conditions A to D. The "Task Label" column in the table corresponds to the candidate category. The "Number" column in the table corresponds to the number of tasks for each task label performed by the care worker when conditions A to D are met. The "Percentage" column in the table is the value obtained by dividing the value in the "Number" column by the total number, and corresponds to the probability of occurrence. The probability model may be configured to output the probability of occurrence for each candidate category when each of conditions A to D is met.

[0107] Figure 4A is a table showing the probability of each candidate category occurring when the information about the care worker satisfies condition A. Condition A is the condition for determining whether the care worker was walking. The probability of each candidate category shown in Figure 4A corresponds to the proportion of correct actions when the condition that the care worker was walking was met when estimating the category of work.

[0108] Figure 4B is a table showing the probability of each candidate category occurring when the information about the care worker satisfies condition B. Condition B is the condition for determining the care worker's area. The probability of each candidate category shown in Figure 4B corresponds to the proportion of correct actions when the condition that the care worker was in a specific area was met when estimating the category of work.

[0109] Figure 4C is a table showing the probability of each candidate category occurring when the information about the care worker satisfies condition C. Condition C is the condition for determining the care worker's work schedule. The probability of each candidate category shown in Figure 4C corresponds to the proportion of correct actions when the care worker's work schedule, when estimating the work category, satisfies the condition that it was a specific task.

[0110] Figure 4D is a table showing the probability of occurrence for each candidate category when the information about the care worker satisfies condition D. Condition D is the condition for determining the time of day for the care worker. The probability of occurrence for each candidate category shown in Figure 4D corresponds to the proportion of correct actions when the time of day for estimating the work category satisfies the condition that it falls within a specific time period.

[0111] Figures 5A to 5C show examples of time-series data of the probability of occurrence of candidate categories by time of day. The horizontal axis represents the time from 0:00 to 24:00 within a day. The vertical axis represents the probability of occurrence in each time period. Figure 5A shows the probability of occurrence of the candidate category represented by work label C06 by time of day. Figure 5B shows the probability of occurrence of the candidate category represented by work label G01 by time of day. Figure 5C shows the probability of occurrence of the candidate category represented by work label xS3 by time of day.

[0112] <Example of Work Classification Estimation> An example of estimating work classification will be explained with reference to Figures 6 to 8.

[0113] As shown in Figure 6, the likelihood calculation unit 34 of the work category estimation unit 30 inputs care worker information into the likelihood model and calculates the likelihood of each candidate category. The calculation results of the likelihood of each candidate category are shown as a graph in Figure 7. The vertical axis represents the likelihood of each candidate category. As shown in Figure 7, the likelihood may be calculated for each of the 17 candidate categories from work A to Q. The bar graph for work E, which has the highest likelihood, is hatched with diagonal lines.

[0114] The probability calculation unit 35 of the work category estimation unit 30 calculates the probability of occurrence for each candidate category. In this embodiment, assume that the information regarding care workers satisfies conditions A to D. The probability model outputs the probability A corresponding to condition A, the probability B corresponding to condition B, the probability C corresponding to condition C, and the probability D corresponding to condition D.

[0115] The estimation unit 36 ​​of the work category estimation unit 30 performs an average processing of the occurrence probabilities A to D for each candidate category and calculates the average occurrence probability for each candidate category.

[0116] The estimation unit 36 ​​considers the likelihood of each candidate category as the prior probability, the average probability of occurrence of the task in each candidate category as the corrected probability, and calculates the posterior probability of each candidate category, i.e., the corrected likelihood of each candidate category, by performing Bayesian updating. The estimation unit 36 ​​may extract the candidate category with the highest posterior probability, i.e., corrected likelihood, as the estimated result of the task category performed by the care worker. The estimation unit 36 ​​may also extract multiple candidate categories in descending order of posterior probability, i.e., corrected likelihood, as the estimated result of the task category performed by the care worker.

[0117] The estimation unit 36 ​​may perform a narrowing process using the average probability of occurrence for each candidate category, regardless of the likelihood of each candidate category. The narrowing process is a process of narrowing down the candidate categories to be targeted for estimation of work categories from among a plurality of candidate categories. In other words, at least one candidate category is excluded from the target of estimation of work categories by the narrowing process. The narrowing process may be performed without changing the likelihood of each candidate category. As a narrowing process, the estimation unit 36 ​​may perform a process to exclude, for example, a candidate category whose average probability of occurrence is 0. As a narrowing process, the estimation unit 36 ​​may perform a process to exclude, for example, a candidate category whose average probability of occurrence is less than a narrowing threshold.

[0118] Figure 8 shows the results of a filtering process performed on each candidate category shown in Figure 7, using the average probability of occurrence for each category. The vertical axis represents the likelihood of each candidate category. As shown in Figure 8, the bar graphs for tasks A to E and tasks O to Q are hatched with dots. These categories are candidate categories that have been excluded from the estimation of task categories by the filtering process in this embodiment. Conversely, tasks F to N, represented by bar graphs without dot hatching, remain as candidate categories for estimation of task categories.

[0119] The estimation unit 36 ​​extracts task L, which has the highest likelihood among the task F to N that remain as candidate task categories for estimation after the narrowing process. In the graph in Figure 8, the bar graph for task L is hatched with diagonal lines. The estimation unit 36 ​​estimates that the task category performed by the care worker is task L. The output unit 37 of the task category estimation unit 30 outputs task L as the task category estimation result.

[0120] (Flowchart example) The likelihood model generation unit 27 of the model generation unit 20 may generate a likelihood model by executing a model generation method that includes the steps of the flowchart illustrated in Figure 9. The probability model generation unit 28 of the model generation unit 20 may generate a probability model by executing a model generation method that includes the steps of the flowchart illustrated in Figure 10. The model generation method may be implemented as a model generation program to be executed by the processor constituting the model generation unit 20. The model generation program may be stored in a non-temporary computer-readable medium.

[0121] Referring to Figure 9, the likelihood model generation unit 27 acquires the location information, time information, and work schedule of the care worker (step S1). The likelihood model generation unit 27 acquires the categories of work performed by the care worker, i.e., the correct actions (step S2). The likelihood model generation unit 27 generates input data including information about the care worker, and correct data including the categories of work performed by the care worker at each location, i.e., the correct actions (step S3). The likelihood model generation unit 27 generates a likelihood model by performing learning using the input data and the correct data (step S4). The likelihood model generation unit 27 outputs the generated likelihood model to the work category estimation unit 30. After executing the procedure in step S4, the likelihood model generation unit 27 completes the execution of the procedure in the flowchart in Figure 9.

[0122] Referring to Figure 10, the probability model generation unit 28 acquires the location information, time information, and work schedule of the care worker (step S11). The probability model generation unit 28 acquires the category of work performed by the care worker, i.e., the correct action (step S12). The probability model generation unit 28 generates a probability model by performing statistical processing (step S13). The probability model generation unit 28 outputs the generated probability model to the work category estimation unit 30. After executing the procedure in step S13, the probability model generation unit 28 completes the execution of the procedure in the flowchart of Figure 10.

[0123] The work portion estimation unit 30 may execute a work portion estimation method that includes the steps of the flowchart illustrated in Figure 11. The work portion estimation method may be implemented as a work portion estimation program to be executed by the processor constituting the work portion estimation unit 30. The work portion estimation program may be stored on a non-temporary computer-readable medium.

[0124] The work category estimation unit 30 acquires location and time information of the care worker using the location measurement unit 31 and the time measurement unit 33, and acquires the work schedule from the schedule management unit 16 (step S21). The work category estimation unit 30 calculates the likelihood of each candidate category using a likelihood model with the likelihood calculation unit 34 (step S22). The work category estimation unit 30 calculates the probability of occurrence of each candidate category using a probability model with the probability calculation unit 35 (step S23). The work category estimation unit 30 estimates the work category performed by the care worker based on the likelihood of each candidate category and the probability of occurrence of each candidate category using the estimation unit 36 ​​(step S24). The work category estimation unit 30 outputs the estimated work category result with the output unit 37 (step S25). After executing step S25, the work category estimation unit 30 completes the execution of the procedure in the flowchart of Figure 11.

[0125] (Summary) As described above, the information processing device 10 relating to this disclosure can maintain or improve the accuracy of work classification estimation by estimating the classification of work performed by a care worker based on both the likelihood of the candidate classification and the probability of the occurrence of the candidate classification. As a result, the information processing device 10 can maintain or improve the accuracy of work data, including the classification of work performed by a care worker.

[0126] Furthermore, the information processing device 10 can collect work data from care workers without requiring human intervention such as observers by using likelihood models and probability models. As a result, work data from care workers is collected efficiently.

[0127] (Other embodiments) Other embodiments are described below.

[0128] <Example of classification of caregiving tasks performed by care workers> Caregiving tasks performed by care workers may be classified into multiple hierarchical levels, as illustrated in Figure 12. In the example in Figure 12, the classification of caregiving tasks performed by care workers is divided into three levels called major classification, medium classification, and minor classification. A major classification shall include at least one medium classification. That is, the number of classifications belonging to a major classification is less than or equal to the number of classifications belonging to a medium classification. A medium classification shall include at least one minor classification. That is, the number of classifications belonging to a medium classification is less than or equal to the number of classifications belonging to a minor classification. A major classification is also called the first level. A medium classification is also called the second level. A minor classification is also called the third level. A second level is included in the first level. That is, the second level is a lower level than the first level. A third level is included in the second level. That is, the third level is a lower level than the second level. The number of levels is not limited to three; there may be two or four or more levels.

[0129] In this embodiment, the major classification, or first level, is a classification that categorizes the work performed by care workers by the type of service. The major classification may include, for example, caregiving, nursing, rehabilitation, care support, meals and nutrition, indirect work, non-work-related work, common work, or other work. The major classification may also include measurement errors as a classification for when the work performed by care workers cannot be classified into any of the above categories. The major classification is not limited to these examples and may include various other classifications.

[0130] In this embodiment, the intermediate classification, or second level, is a classification that categorizes the tasks performed by care workers according to the purpose or circumstances of the task. The intermediate classification corresponds to the classification that categorizes actions into 35 types in the above-described embodiment of classification estimation.

[0131] The intermediate classification may include, as a category belonging to the major classification of care, for example, getting up and going to bed, changing positions, dressing and grooming, user mobility, excretion, meals, hygiene, environmental maintenance, laundry, going out, recreation, or transportation. The intermediate classification may also include, as a category belonging to the major classification of nursing, for example, medical procedures, medication management, vital sign measurement, physical examination, or direct intervention on physical function.

[0132] The subcategory may include, for example, physical therapy, occupational therapy, or speech-language therapy as a division belonging to the major category of rehabilitation. The subcategory may include, for example, care plans as a division belonging to the major category of care support. The subcategory may include, for example, nutritional management, meal service, or hygiene management as a division belonging to the major category of food and nutrition.

[0133] The subcategory may include, for example, the movement of care workers, meetings, or office cleaning, as it is a category belonging to the major category of indirect work. The subcategory may include, for example, breaks, as it is a category belonging to the major category of non-work-related work. The subcategory may include, for example, understanding users, information sharing, record keeping, verification, handling visitors, or other services, as it is a category belonging to the major category of common work.

[0134] The subcategory, or third level, is defined as a classification of tasks performed by care workers based on their content.

[0135] A subcategory may include, for example, assistance with getting up or assistance with going to bed, as a category belonging to the intermediate categories of getting up and going to bed. A subcategory may include, for example, assistance with changing body position or reclining adjustment, as a category belonging to the intermediate categories of dressing and grooming. A subcategory may include, for example, assistance with dressing or grooming, as a category belonging to the intermediate categories of user mobility. A subcategory may include, for example, wheelchair guidance, walking assistance, transfer assistance, or standing assistance, as a category belonging to the intermediate categories of excretion. A subcategory may include, for example, assistance with excretion, diaper changing, or handwashing assistance, as a category belonging to the intermediate categories of meals. The subcategory may include, for example, assistance with bathing, foot bathing, face washing, wiping, perineal cleansing, oral care, earwax removal, nail trimming, shaving, hair washing, hand washing assistance, or gargling assistance, as a subcategory of the hygienic category. The subcategory may include, for example, changing sheets, garbage collection, room cleaning, checking user's belongings, adjusting room temperature, humidity control, lighting control, or ventilation, as a subcategory of the laundry category, for example, washing or collecting laundry. The subcategory may include, for example, assistance with shopping or accompanying users on shopping trips, as a subcategory of the outing category. The subcategory may include, for example, exercise, oral exercises, music, picture storytelling, or cooking, as a subcategory of the recreation category. The subcategory may include, for example, assistance with transportation or getting on and off vehicles, as a subcategory of the transportation category.

[0136] The subcategory may include, as a division within the medium category of medical procedures, for example, tracheal suctioning, wound care, ointment application, disinfection, drug application, intravascular pressure monitoring, intravenous infusion, gastrostomy management, urinary management, enemas, or oxygen therapy. The subcategory may include, as a division within the medium category of drug management, for example, oral administration, enteral administration, dispensing of medications, eye drops, injections, suppository administration, or inventory checks. The subcategory may include, as a division within the medium category of vital sign measurement, for example, temperature measurement, SpO2 measurement, pulse measurement, blood pressure measurement, weight measurement, or vital sign measurement. The subcategory may include, as a division within the medium category of medical examinations, for example, attending medical examinations or accompanying patients to hospitals. The subcategory may include, as a division within the medium category of direct interventions on bodily functions, for example, massage or functional recovery.

[0137] The subcategory may include, for example, range of motion exercises, thermotherapy, gait training, standing training, stair climbing, exercise training, or physical therapy assistance, as it belongs to the intermediate category of physical therapy. The subcategory may include, for example, occupational tasks, calculation tasks, cognitive tasks, memory tasks, artistic tasks, play tasks, or occupational therapy assistance, as it belongs to the intermediate category of rehabilitation. The subcategory may include, for example, expression training, auditory comprehension training, swallowing function assessment, indirect swallowing training, or direct swallowing training.

[0138] A subcategory may include, for example, care plan creation, conferences, monitoring, or assessment, as a category belonging to the intermediate category of care plans. A subcategory may include, for example, nutrition care plan creation, conferences, monitoring, nutrition assessment, nutrition screening, or nutritional counseling, as a category belonging to the intermediate category of nutrition management. A subcategory may include, for example, menu planning, cooking, meal count management, or ingredient procurement, as a category belonging to the intermediate category of hygiene management, as a category belonging to the intermediate category of hygiene management, as a category belonging to hygiene management, ingredient management, or equipment management.

[0139] A subcategory may include, for example, movement, waiting, or transportation, as a category belonging to the intermediate category of movement of care workers. A subcategory may include, for example, morning meetings, evening meetings, or conferences, as a category belonging to the intermediate category of meetings. A subcategory may include, for example, cleaning of common areas or tidying up, as a category belonging to the intermediate category of office cleaning.

[0140] The subcategory may include, for example, verbal communication, attentive listening, monitoring, user condition check, or nurse call response, as it belongs to the intermediate category of understanding the user. The subcategory may include, for example, communication, handover, telephone, or fax sending and receiving, as it belongs to the intermediate category of information sharing. The subcategory may include, for example, record making, calculation, printing, and copying, as it belongs to the intermediate category of record making. The subcategory may include, for example, record checking, work checking, or equipment checking, as it belongs to the intermediate category of confirmation. The subcategory may include, for example, family interaction or visitor interaction, as it belongs to the intermediate category of handling visitors.

[0141] Subcategories may include common tasks such as preparation, cleanup, or handwashing.

[0142] The likelihood model generation unit 27 and the probability model generation unit 28 may set the categories belonging to each of the hierarchical classifications described above as candidate categories. Based on the categories belonging to each of the hierarchical classifications described above, the likelihood model generation unit 27 may identify the categories of work performed by care workers and generate correct answer data. As mentioned above, the number of categories belonging to major classifications is less than the number of categories belonging to intermediate classifications. Also, the number of categories belonging to intermediate classifications is less than the number of categories belonging to minor classifications.

[0143] Here, the likelihood model generation unit 27 and the probability model generation unit 28 assume that the number of pieces of information on the care worker's behavior that can be collected by the behavior measurement unit 21 is constant. When associating the collected information on the care worker's behavior with work categories, the number of pieces of information on the care worker's behavior that can be associated with each work category is inversely proportional to the number of work categories. In other words, the fewer the number of candidate categories, the larger the number of training data corresponding to each candidate category. The larger the number of training data, the higher the accuracy of calculating the likelihood or probability of occurrence for each candidate category. Therefore, the likelihood model generation unit 27 may generate multiple training data sets by combining input data and ground truth data with different numbers of work categories associated with the collected information on the care worker's behavior, and generate a likelihood model by performing training using each training data set. Alternatively, the probability model generation unit 28 may generate a probability model based on processing data with different numbers of work categories associated with the collected information on the care worker's behavior.

[0144] For example, the likelihood model generation unit 27 may generate a likelihood model capable of estimating the classification of care workers' tasks within the major classifications by classifying the collected information on the care workers' behaviors into categories belonging to major classifications, generating ground truth data, and performing learning using that ground truth data and the corresponding input data. The likelihood model generation unit 27 may generate a likelihood model capable of estimating the classification of care workers' tasks within the major classifications by classifying the collected information on care workers' behaviors into categories belonging to medium classifications, generating ground truth data, and performing learning using that ground truth data and the corresponding input data. The likelihood model generation unit 27 may generate a likelihood model capable of estimating the classification of care workers' tasks within the minor classifications by classifying the collected information on care workers' behaviors into categories belonging to minor classifications, generating ground truth data, and performing learning using that ground truth data and the corresponding input data.

[0145] The likelihood model generation unit 27 and the probability model generation unit 28 may generate likelihood models and probability models that can calculate the likelihood and probability of occurrence for each candidate category, not limited to major, medium, or minor categories, but including various other groups of categories. For example, the likelihood model generation unit 27 and the probability model generation unit 28 may generate likelihood models and probability models that calculate the likelihood and probability of occurrence for each candidate category, using either categories of tasks involving movement or tasks without movement as candidate categories for tasks performed by care workers.

[0146] The likelihood model generation unit 27 and the probability model generation unit 28 may generate likelihood models and probability models that can calculate the likelihood and probability of occurrence for each candidate category, using one of the major, medium, or minor categories as the candidate category. The likelihood model generation unit 27 and the probability model generation unit 28 may generate likelihood models and probability models that can calculate the likelihood and probability of occurrence for each candidate category, using two or more of the major, medium, or minor categories as the candidate categories. For example, the likelihood model generation unit 27 and the probability model generation unit 28 may generate likelihood models and probability models that can calculate the likelihood and probability of occurrence for each candidate category in both the major and medium categories. In this case, the likelihood model and probability model output both the likelihood and probability of occurrence for the candidate categories included in the major category and the likelihood and probability of occurrence for the candidate categories included in the medium category.

[0147] <Estimation using subsequent movement history> In the embodiment described above, the likelihood model was configured to calculate the likelihood of each candidate category based on location information regarding the locations the care worker moved to before the location where the work to be estimated as a category was performed, and time information regarding the time the care worker stayed at those locations. In other words, the likelihood that the category of work performed by the care worker at a certain location is one of the candidate categories was calculated based on the locations the care worker had moved to in the past.

[0148] The likelihood model may be configured to calculate the likelihood of a candidate category for a task to be estimated, based on location information regarding the location to which the care worker moved after the location where the task to be estimated was performed, and time information regarding the time the care worker stayed at that location. In other words, the likelihood that a task performed by a care worker at a certain location is one of the candidate categories may be calculated based on the future locations to which the care worker moves.

[0149] <Examples of other items included in the training data> The training data may include other items. The training data may include, for example, nurse call information, monitoring sensor information, care equipment logs, or work manual information. The training data may include seasonal variation or calendar information. The training data may include climate data such as weather or temperature. The training data may include information about care workers that can be obtained using IoT (Internet of Things) devices, other than the location and time information of care workers.

[0150] <Description of the Care Process> In care services, the tasks performed by care workers can be described as a care process, which is a combination of multiple actions. The care process can be described as an independent process in which a care worker performs multiple actions one by one in sequence, as shown in Figure 13A, for example. In Figure 13A, the care worker proceeds from (1) "Action 1" to "Action 2", and then from (2) "Action 2" to "Action 3".

[0151] The care process can be described not only as an independent process, but also as a parallel process in which a caregiver performs multiple actions simultaneously. A parallel process can be described as one in which some actions are performed in parallel, as shown in Figure 13B, for example. In Figure 13B, the caregiver (1) proceeds from "Action 1" to "Action 2", (2) and (3) performs "Action 2" and "Action 3" in parallel, and (4) proceeds from "Action 2" to "Action 4".

[0152] A parallel process can be described as having the first actions performed in parallel, for example, as shown in Figure 13C. In Figure 13C, the care worker (1) proceeds from "Action 1" to "Action 2" and "Action 3" in parallel, (2) performs "Action 2" and "Action 3" in parallel, and (3) proceeds from "Action 2" to "Action 4".

[0153] A parallel process can be described as all actions occurring in parallel, as shown in Figure 13D, for example. In Figure 13D, the care worker (1) proceeds from "Action 1" to "Action 2" and "Action 3" in parallel, performing "Action 2" and "Action 3" in parallel, and (2) proceeds from "Action 2" and "Action 3" to "Action 4".

[0154] A parallel process can be described as having parallel final actions, as shown in Figure 13E, for example. In Figure 13E, the caregiver (1) proceeds from "Action 1" to "Action 2", (2) performs "Action 2" and "Action 3" in parallel, and (3) proceeds from "Action 2" and "Action 3" to "Action 4" in parallel.

[0155] The care process can also be described as an interruption process, as illustrated in Figure 13F, in which a caregiver temporarily suspends a specific action to perform another action and then returns to the original action. In Figure 13F, the caregiver (1) proceeds from "Action 1" to "Action 2", (2) interrupts "Action 2" to perform "Action 3", (3) after completing "Action 3", returns to the interrupted "Action 2", and (4) proceeds from "Action 2" to "Action 4".

[0156] The likelihood model generation unit 27 may generate movement history information of care workers that takes into account the flow of actions in these care processes as input data, generate work categories corresponding to the input data as ground truth data, and generate a likelihood model by performing learning using training data that combines the input data and the ground truth data. The likelihood model may be configured to calculate the likelihood of candidate categories that include some of the actions included in the care process. The likelihood model may be configured to consider the entire care process as a single task and calculate the likelihood of candidate categories that correspond to that task.

[0157] The likelihood model generation unit 27 may generate a likelihood model by generating movement history information, including the order of movement to the locations where each action in the care process is performed, as input data when each action in the care process is performed at a different location, generating work categories corresponding to that learning data as ground truth data, and performing learning using the learning data, which is a combination of the input data and the ground truth data. In this case, the likelihood model is configured to consider the entire care process as a single task and calculate the likelihood of the candidate category corresponding to that task.

[0158] <Example of displaying estimation results> As described above, the output unit 37 of the work category estimation unit 30 may display the estimated results of the work categories performed by the care worker on the display device 60. The display device 60 may display the estimation results in the form of a timeline corresponding to the time period in which the care worker is estimated to have performed each work category, as illustrated in Figure 14. The horizontal axis in Figure 14 represents the passage of time from left to right. The time period in which each work category was performed is shown as a black rectangle. Displaying it in this way makes it easier to understand the estimated results of the care worker's work categories.

[0159] The display device 60 may display the time periods spent by care workers at each location in the form of a timeline, as illustrated in Figure 15. The horizontal axis in Figure 15 represents the passage of time from left to right. The time periods spent by care workers at each location are shown as black rectangles.

[0160] According to the timeline illustrated in Figure 14 or Figure 15, it is possible to investigate the ratio of direct tasks performed by care workers as caregiving tasks for the person receiving care, i.e., the user, to indirect tasks other than direct tasks for the person receiving care, such as tidying up the office, or to compare the time spent on work tasks with the care worker's own break time.

[0161] The display device 60 may display the total estimated time spent by the care worker on each task category as a bar graph, as illustrated in Figure 16. In the graph in Figure 16, the horizontal axis corresponds to the task category. The vertical axis represents the estimated time spent by the care worker on each task category. If a task category is performed in multiple sessions, the estimated time spent on that task category is the sum of the time spent on each session. The display device 60 may also display the frequency of each task category estimated to have been performed by the care worker as a histogram. In other words, the display device 60 may display statistical data on the task categories estimated to have been performed by the care worker. Displaying the data in this way makes it easier to understand the estimation results of the care worker's task categories.

[0162] When the work category estimation unit 30 performs estimation in real time using the estimation unit 36, the display device 60 may change the displayed content in real time according to the real-time estimation results. The display device 60 may display the estimation results of work categories performed by multiple care workers in a comparable format. The display device 60 may display in a comparable format the work categories performed by care workers estimated based on location and time information acquired before the procedure of the work performed by the care worker was changed, and the work categories performed by care workers estimated based on location and time information acquired after the procedure of the work performed by the care worker was changed. The display device 60 is not limited to these examples and may display the estimation results of work categories performed by care workers in various other forms. The output unit 37 of the work category estimation unit 30 may output the content to be displayed on the display device 60 as data or information to another device such as a database 70.

[0163] (Summary) As described above, the information processing device 10 according to this embodiment can input information about care workers into a likelihood model and a probability model to obtain the likelihood and probability of occurrence of each candidate category, and estimate the category of work performed by the care worker based on the likelihood and probability of occurrence of each candidate category. By estimating the category of work performed by the care worker based on both the likelihood and probability of occurrence of the candidate category, the information processing device 10 can maintain or improve the accuracy of work category estimation. As a result, the information processing device 10 can maintain or improve the accuracy of work data, including the category of work performed by the care worker.

[0164] Furthermore, the information processing device 10 can collect work data from care workers without requiring human intervention such as observers by using likelihood models and probability models. As a result, work data from care workers is collected efficiently. By efficiently collecting work data from care workers, improvements to care services using this data can be efficiently supported.

[0165] While embodiments relating to this disclosure have been described based on the drawings and examples, it should be noted that those skilled in the art can make various modifications or alterations based on this disclosure. Therefore, it should be noted that these modifications or alterations are included within the scope of this disclosure. For example, the functions included in each component or step can be rearranged in a logically consistent manner, and multiple components or steps can be combined into one or divided. While embodiments relating to this disclosure have been described primarily in terms of apparatus, embodiments relating to this disclosure can also be realized as methods including steps performed by each component of the apparatus. Embodiments relating to this disclosure can also be realized as methods, programs, or storage media recording programs executed by a processor in the apparatus. These should also be understood to be included within the scope of this disclosure.

[0166] 1. Information Processing System 10. Information Processing Device (12: Work Classification Definition Unit, 14: Location Definition Unit, 16: Schedule Management Unit) 20. Model Generation Unit (21: Action Measurement Unit, 22: Location Measurement Unit, 23: Time Measurement Unit, 27: Likelihood Model Generation Unit, 28: Probability Model Generation Unit) 30. Work Classification Estimation Unit (31: Location Measurement Unit, 32: Location Recognition Unit, 33: Time Measurement Unit, 34: Likelihood Calculation Unit, 35: Probability Calculation Unit, 36: Estimation Unit, 37: Output Unit) 40. Sensor 50. Input Device 60. Display Device 70. Database

Claims

1. A method for estimating the classification of tasks performed by a care worker in a care setting, comprising: a step of an information processing device inputting information about the care worker into a likelihood model, which is a machine learning model, and obtaining the likelihood of each candidate classification, which represents the likelihood that the classification of tasks performed by the care worker is one of the multiple candidate classifications; a step of the information processing device inputting information about the care worker into a probability model, which is a statistical model, and obtaining the probability of occurrence of each candidate classification, which represents the probability that the classification of tasks performed by the care worker is one of the multiple candidate classifications; a step of the information processing device estimating the classification of tasks performed by the care worker based on the likelihood of each candidate classification and the probability of occurrence of each candidate classification; and a step of the information processing device outputting the estimation result of the classification of tasks performed by the care worker. The likelihood model is a machine learning model generated by performing machine learning using training data that associates input data including information about the care worker with ground truth data representing the categories of work performed by the care worker, and is configured to output the likelihood of each candidate category representing the likelihood that the category of work performed by the care worker is one of several candidate categories when information about the care worker is input, and the probability model is a statistical model generated by performing statistical processing on data that associates information about the care worker with the categories of work performed by the care worker, and is configured to output the probability of occurrence of each candidate category corresponding to the conditions that the information about the care worker satisfies when information about the care worker is input, in a method for estimating work categories.

2. The method for estimating work categories according to claim 1, wherein the information processing device, in the step of obtaining the probability of occurrence of each of the candidate categories, inputs each of a plurality of conditions into the probability model and obtains the probability of occurrence of each of the candidate categories corresponding to each of the plurality of conditions from the probability model, and in the step of estimating the work category, the information processing device estimates the work category performed by the care worker based on the probability of occurrence of each of the candidate categories corresponding to each of the plurality of conditions.

3. The work classification estimation method according to claim 2, wherein the information processing device, in the step of estimating the classification of the work, generates an average occurrence probability for each of the candidate classifications by performing an arithmetic mean or geometric mean of the occurrence probabilities of each of the candidate classifications corresponding to each of the plurality of conditions, and estimates the classification of the work performed by the care worker based on the average occurrence probability for each of the candidate classifications.

4. The method for estimating work categories according to any one of claims 1 to 3, wherein the probability model is configured to output, in data relating information about the care worker to the categories of work performed by the care worker, the proportion of times when the information about the care worker satisfies at least one condition and the category of work performed by the care worker was one of the candidate categories as the probability of occurrence of each candidate category.

5. The work category estimation method according to claim 4, wherein the probability model is configured to output a percentage that has been corrected so that it becomes smaller when the percentage of each candidate category is smaller than a correction threshold, is not corrected when the percentage of each candidate category is equal to the correction threshold, and becomes larger when the percentage of each candidate category is greater than the correction threshold, as the probability of occurrence of each candidate category.

6. The method for estimating work categories according to claim 5, wherein the probability model is configured to output a normalized ratio as the probability of occurrence of each candidate category, which is corrected to 0 when the proportion of each candidate category is less than or equal to the correction threshold, and corrected to a normalized ratio when the proportion of each candidate category is greater than the correction threshold, and the normalized ratio is the reciprocal of the number of candidate categories that satisfy the condition that the proportion of each candidate category is greater than the correction threshold.

7. A work category estimation program that causes an information processing device to estimate the categories of work performed by a care worker in a care setting, comprising the steps of: inputting information about the care worker into a likelihood model, which is a machine learning model, and obtaining the likelihood of each candidate category, which represents the likelihood that the work performed by the care worker is one of the multiple candidate categories; inputting information about the care worker into a probability model, which is a statistical model, and obtaining the probability of occurrence of each candidate category, which represents the probability that the work performed by the care worker is one of the multiple candidate categories; estimating the category of work performed by the care worker based on the likelihood of each candidate category and the probability of occurrence of each candidate category; and outputting the estimation result of the category of work performed by the care worker, the information processing device is made to execute these steps. The likelihood model is a machine learning model generated by performing machine learning using training data that associates input data containing information about the care worker with ground truth data representing the categories of work performed by the care worker, and is configured to output the likelihood of each candidate category representing the likelihood that the category of work performed by the care worker is one of several candidate categories when information about the care worker is input, and the probability model is a statistical model generated by performing statistical processing on data that associates information about the care worker with the categories of work performed by the care worker, and is configured to output the probability of occurrence of each candidate category corresponding to the conditions that the information about the care worker satisfies when information about the care worker is input, in a work category estimation program.

8. An information processing device for estimating the classification of tasks performed by a care worker in a care setting, comprising: a likelihood calculation unit that inputs information about the care worker into a likelihood model, which is a machine learning model, and calculates the likelihood of each candidate classification representing the probability that the task performed by the care worker is one of the multiple candidate classifications; a probability calculation unit that inputs information about the care worker into a probability model, which is a statistical model, and calculates the probability of each candidate classification representing the probability that the task performed by the care worker is one of the multiple candidate classifications; an estimation unit that estimates the classification of tasks performed by the care worker based on the likelihood of each candidate classification and the probability of each candidate classification; and an output unit that outputs the estimation result of the classification of tasks performed by the care worker, wherein the likelihood model is a machine learning model generated by performing machine learning using training data that associates input data including information about the care worker with ground truth data representing the classification of tasks performed by the care worker, and is configured to output the likelihood of each candidate classification representing the probability that the task performed by the care worker is one of the multiple candidate classifications when information about the care worker is input, The aforementioned probability model is a statistical model generated by performing statistical processing on data that associates information about the care worker with the categories of work performed by the care worker, and is configured to output the probability of occurrence of each candidate category corresponding to the conditions satisfied by the information about the care worker when information about the care worker is input.

Images