JP7745593B2 - Extraction system and extraction method - Google Patents

Description

This disclosure relates to an extraction system and an extraction method.

Building managers (such as resident managers who reside in buildings) need to master the tasks of managing buildings. For example, Japanese Patent Application Laid-Open Publication No. 2018-132556 (Patent Document 1) discloses a worker training device that trains workers. The worker training device improves knowledge by presenting trainees with questions that simulate work content corresponding to the work site and requiring the trainees to answer them.

The tasks performed by the building manager include, for example, tasks that are repeated periodically (routine tasks, etc.) and other tasks that are performed in response to incidents that arise in relation to building management, such as complaints or inquiries (non-routine tasks, etc.). Because routine tasks are repeated periodically, managers will first become proficient in routine tasks. After becoming proficient in routine tasks, managers will need to become proficient in non-routine tasks as the next step.

Japanese Patent Application Laid-Open No. 2018-132556

Training for managerial tasks is often done through on-the-job training (OJT). However, when learning non-routine tasks, it is often not clearly defined which items should be prioritized in training. As a result, the items taught depend on the experience of the OJT trainer, making them highly personal.

If trainers do not prioritize training items, employees will become proficient in their work through on-site experience whenever a complaint, inquiry, or other incident occurs. In other words, the order of training will be determined by the order in which the above incidents occur.

If a series of low-frequency events occur during OJT training, employees' proficiency in responding to those events will increase, but the ability to respond to high-frequency events, which should be a priority, will be put off, resulting in inefficient training.

On the other hand, among the procedures included in non-routine work, there are some procedures for which training is not necessary if the worker has already mastered routine work. Therefore, if training is conducted without taking into account procedures that overlap with routine work, the same procedures will be taught redundantly when training non-routine work, resulting in inefficient training.

This disclosure has been made to solve the above-mentioned problems, and its purpose is to provide a training item extraction system and control method that allows building managers to efficiently master their work.

According to one aspect of the present disclosure, the extraction system includes a storage device and a control device. The storage device stores a task database that records the procedures for tasks performed by a building manager. The control device extracts, as training items, the procedures for tasks that the manager should prioritize mastering. The tasks performed by the manager include a first task that the manager repeats periodically and a second task that is different from the first task. The control device acquires the procedures for the first task and the procedures for the second task from the task database. The control device extracts, from the second task, high-frequency tasks that occur at a frequency equal to or greater than a reference value. The control device extracts, from the procedures included in the high-frequency tasks, the procedures included in the first task, excluding the procedures included in the first task, as training items.

In accordance with another aspect of the present disclosure, the extraction method includes the steps of storing, in a storage device, a task database in which the procedures for tasks performed by a building manager are recorded, and extracting, as training items, the procedures for tasks that the manager should prioritize mastering. The tasks performed by the manager include a first task that the manager periodically repeats, and a second task that is different from the first task.

The extraction step includes the steps of acquiring the procedures for the first task and the procedures for the second task from the task database, extracting high-frequency tasks from the second task whose occurrence frequency is equal to or exceeds a reference value, and extracting, as training items, the procedures included in the high-frequency tasks, excluding the procedures included in the first task.

This disclosure allows building managers to efficiently learn their duties.

1 is a schematic diagram showing an example of the overall configuration of an education item extraction system according to an embodiment of the present invention. 1 is a schematic diagram illustrating an example of a hardware configuration of an education item extraction system according to an embodiment of the present invention. 10 is a flowchart of a main process. 10 is a flowchart of a main process. FIG. 2 is a diagram illustrating an example of an equipment daily report DB. FIG. 10 is a diagram illustrating an example of a high-priority education item DB. FIG. 2 is a diagram illustrating an example of a routine procedure DB. FIG. 2 is a diagram illustrating an example of an unstructured procedure manual DB. FIG. 10 is a diagram illustrating an example of a priority training procedure table. FIG. 10 is a diagram illustrating the learning of an estimation model. 10 is a flowchart of a learning process.

The following describes the embodiments with reference to the drawings. In the following description, identical components are designated by the same reference numerals. Their names and functions are also the same. Therefore, detailed descriptions of them will not be repeated.

First, an overview of the training item extraction system (also simply referred to as the "extraction system") 1 according to this embodiment will be described. Figure 1 is a schematic diagram showing an example of the overall configuration of the training item extraction system 1 according to this embodiment. The training item extraction system 1 is a system that extracts, as training items (hereinafter also referred to as "priority training procedures"), work procedures that a building manager 5 should be trained on a priority basis.

The tasks performed by Manager 5 include a first task that Manager 5 performs repeatedly on a regular basis, and a second task that is different from the first task. The second task refers to a task performed in response to events that occur in relation to building management, such as complaints or inquiries. In this embodiment, the first task is exemplified as a "routine task," and the second task is exemplified as a "non-routine task." Note that the "first task" may refer to a simple task that anyone can perform if performed according to a manual (procedure manual). On the other hand, the "second task" may refer to a task that involves complex procedures that are difficult to document in a manual.

The procedures for work performed by Manager 5 are recorded in a work procedure manual. The work procedure manual includes a routine work procedure manual 72 and an unroutine work procedure manual 73. The procedures for routine work are stored as document data in the routine work procedure manual 72. The procedures for unroutine work are stored as document data in the unroutine work procedure manual 73.

On the other hand, requests, inquiries, complaints (claims) made to Manager 5, as well as incidents that occur in relation to building management, such as equipment failures in the building, and responses to those incidents, are stored as document data in the form of equipment daily report 71. When Manager 5 responds to the above-mentioned incidents, he records the details in the form of a daily report. The work of responding to the events recorded in the equipment daily report 71 corresponds to non-routine work.

The data for the routine work procedure manual 72, non-routine work procedure manual 73, and equipment daily report 71 is recorded using, for example, spreadsheet software. This data is stored, for example, on a server installed in the building or on a server within the management company that manages the building, and can be viewed or updated by each manager 5.

The training item extraction system 1 stores an equipment daily report DB (database) 61, a routine procedure DB 62, an unroutine procedure DB 63, a high-priority training item DB 64, and a priority training procedure table 65. Details will be provided later, but the following describes first the processing flow executed by the training item extraction system 1.

The training item extraction system 1 generates an equipment daily report DB 61 (Figure 5) based on the equipment daily report 71, and further generates a high-priority training item DB 64 (Figure 6) based on the equipment daily report DB 61 (Step 1).

The training item extraction system 1 generates a routine procedure manual DB 62 (Figure 7) based on the routine work procedure manual 72 (Step 2). The training item extraction system 1 generates an unroutine procedure manual DB 63 (Figure 8) based on the unroutine work procedure manual 73 (Step 3).

The training item extraction system 1 generates a priority training procedure table 65 (Figure 9) in which priority training procedures are recorded based on the high-priority training item DB 64, the standard procedure manual DB 62, and the unstandard procedure manual DB 63 (Step 4).

The terminal 100 displays the priority training procedures recorded in the priority training procedure table 65 (step 5). This allows the building manager 5 and his/her trainer to confirm the priority training procedures. The trainer then uses the priority training procedures to train the manager 5 on the work procedures.

Figure 2 is a schematic diagram showing an example of the hardware configuration of the education item extraction system 1 according to this embodiment. The education item extraction system 1 includes a server device 10. The server device 10 is configured to be able to communicate with the terminal 100.

The server device 10 includes a control unit 21, a ROM (Read Only Memory) 22, a RAM (Random Access Memory) 23, a communication interface 24, and a storage unit 25. The control unit 21 is, for example, a CPU (Central Processing Unit).

The control unit 21 controls the entire server device 10. The control unit 21 loads programs stored in the ROM 22 or the memory unit 25 into the RAM 23 and executes them. The RAM 23 serves as a working area for the control unit 21 when executing programs, and temporarily stores programs, data used when executing programs, etc.

The communication interface 24 is an interface for communicating with the terminal 100. The memory unit 25 is a non-volatile storage device. The memory unit 25 may be, for example, a hard disk drive (HDD) or a solid state drive (SSD).

The memory unit 25 stores an estimation model (also referred to as a "trained model") 31, a training dataset 32, a word dictionary 34, an equipment daily report DB 61, a routine procedure DB 62, an unroutine procedure DB 63, a high-priority training item DB 64, and a priority training procedure table 65.

The trained model 31 and training dataset 32 are used for training to determine event classification names, as described below. Training will be explained in detail using Figures 10 and 11.

The terminal 100 includes a control unit 121, a ROM 122, a RAM 123, a communication interface 124, a storage unit 125, an input unit 126, and a display unit 127. The control unit 121 is, for example, a CPU. The terminal 100 may be a desktop computer, a laptop computer, or a tablet terminal.

The server device 10 may also be configured to operate as a web server and display the priority training procedure by accessing a specified URL from the terminal 100. Alternatively, software for displaying the priority training procedure may be installed on the terminal 100, and the priority training procedure may be displayed by running this software.

The control unit 121 provides overall control over the entire terminal 100. The control unit 121 loads programs stored in the ROM 122 or the memory unit 125 into the RAM 123 and executes them. The RAM 123 serves as a working area when the control unit 121 executes programs, and temporarily stores programs and data used when executing programs. The memory unit 125 is a non-volatile storage device. The memory unit 125 may be, for example, an HDD or SSD. The communication interface 124 is an interface for communicating with the server device 10.

The input unit 126 accepts input from the administrator 5. The input unit 126 is, for example, a keyboard, mouse, or touch panel. The display unit 127 displays various images such as priority training procedures. The display unit 127 is, for example, an LCD display or a display.

The main processing executed by the control unit 21 will be explained below using a flowchart. Figures 3 and 4 are flowcharts of the main processing. The main processing is processing that executes steps 1 to 5 shown in Figure 1. In the main processing, the control unit 21 extracts, as training items (priority training procedures), the procedures for tasks that the manager 5 should prioritize mastering.

The main processing begins based on an execution instruction from the administrator 5 via the input unit 126 of the terminal 100. Hereinafter, "step" will also be referred to simply as "S."

Steps S101 to S106 shown in Figure 3 correspond to step 1 shown in Figure 1. In S101, the control unit 21 records the equipment daily report data of the equipment daily report 71 in the equipment daily report DB 61.

Figure 5 shows an example of the equipment daily report DB61. The equipment daily report DB61 records "event details" such as requests and complaints from building users, or malfunctions that have occurred, along with the corresponding "event classification names." These event details correspond to non-routine tasks.

For example, the content of the incident (non-routine work) is recorded as "A store employee contacted us to request that something be done about the high temperature in the hallway in front of the store."

Returning to Figure 3, in S102, the control unit 21 uses the word dictionary 34 to extract word representative words from the equipment daily report data (occurrence event content). The control unit 21 then inputs the extracted word representative words as features into the trained model 31, and outputs an event classification name from the trained model 31 as an estimation result. The equipment daily report data (occurrence event content) is a written record of events that have occurred in relation to building management. The control unit 21 then simplifies the text based on the text recorded in the equipment daily report data to output an event classification name that represents the content of the event.

The trained model 31 is a model that has undergone machine learning processing using training data so that when a word representative word is input as a feature, an event classification name is output as an inferred result. The trained model 31, word dictionary 34, etc. will be described later using Figures 10 and 11.

In the example shown in Figure 5, the event classification name "Indoor temperature response" is determined based on the event content "A store employee contacted us to request that something be done about the high temperature in the hallway in front of the store," and is recorded in the equipment daily report DB61.

In S104, the control unit 21 acquires data recorded in the equipment daily report DB 61. In S105, the control unit 21 tallies the number of cases for each event classification name. In S106, the control unit 21 ranks the event classification names in descending order of number of cases and records the top 10 event classification names in the high-priority training item DB 64.

Figure 6 shows an example of the high-priority training item DB 64. The high-priority training item DB 64 records the number of event classification names and their rankings.

For example, the event category name "Temperature change request" had 152 occurrences, ranking it first. Following that, in second place was "Bulb replacement" (53 occurrences). In third place was "Door malfunction" (51 occurrences). In fourth place was "Power supply work request" (23 occurrences). In fifth place was "Toilet clog" (20 occurrences). In sixth place was "Alarm" (20 occurrences). In seventh place was "Multiple equipment setting request" (7 occurrences). In eighth place was "Blind malfunction" (6 occurrences). In ninth place was "Uncomfortable indoor draft" (6 occurrences). In tenth place was "Public water heater malfunction" (6 occurrences). In this way, the control unit 21 calculates the frequency of occurrence (number of occurrences, ranking) of event category names corresponding to the equipment daily report data recorded in the equipment daily report DB 61.

Returning to Figure 3, S107 and S108 correspond to step 2 shown in Figure 1. In S107, the control unit 21 records the routine work procedure manual data of the routine work procedure manual 72 in the routine work procedure manual DB 62.

In S108, the control unit 21 determines a procedure classification name from the procedure content of each procedure in the routine procedure manual DB 62 and records it in the routine procedure manual DB 62. This procedure content is a sentence that explains the procedure of the routine task. Based on the sentence (procedure content) that explains the procedure of the routine task, the control unit 21 simplifies the sentence and outputs a first classification name (procedure classification name) that represents the content of the procedure of the routine task. The method for determining the procedure classification name will be described later using Figures 10 and 11.

Figure 7 shows an example of the routine procedure DB 62. The routine procedure DB 62 stores the task name, procedure number, procedure content, and procedure classification name. The routine procedure DB 62 stores routine procedures for the tasks "Building temperature inspection (daily)" and "Toilet inspection (daily)."

The steps for "Temperature patrol within the building (daily)" are "Carry a thermometer" (Step 1), "Move to the temperature measurement location" (Step 2), "Measure the temperature" (Step 3), and "Record the measured temperature on the record sheet" (Step 4). In contrast, the procedure classification names for Steps 1 to 4 have been determined to be "Carry a thermometer," "Move to the temperature measurement location," "Measure the temperature," and "Record the temperature."

The procedures for "Toilet Patrol (Daily)" are "Move to the toilet." (Procedure No. 1), "Check the opening and closing operation of the toilet entrance door." (Procedure No. 2), "Check for damage to the urinal." (Procedure No. 3), "Check for damage to the toilet bowl." (Procedure No. 4), "Check for water leaks." (Procedure No. 5), and "Check for clogged toilets." (Procedure No. 6). In contrast, the procedure classification names for procedures No. 1 to 6 have been determined as "Move to the toilet," "Open/close the toilet door," "Check for damage to the urinal," "Check for damage to the toilet bowl," "Check for water leaks in the toilet," and "Check for clogged toilets."

Returning to Figure 3, S109 and S110 correspond to step 3 shown in Figure 1. In S109, the control unit 21 records the non-routine work procedure manual data of the non-routine work procedure manual 73 in the non-routine work procedure manual DB 63.

In S110, the control unit 21 determines a procedure classification name from the procedure content of each procedure in the non-routine procedure manual DB 63 and records it in the non-routine procedure manual DB 63. This procedure content is a sentence that explains the procedure of the non-routine task. Based on the sentence (procedure content) that explains the procedure of the non-routine task, the control unit 21 simplifies the sentence and outputs a second classification name (procedure classification name) that represents the content of the procedure of the non-routine task. The method for determining the procedure classification name will be described later using Figures 10 and 11.

Figure 8 shows an example of the non-routine procedure manual DB 63. The non-routine procedure manual DB 63 records the business name, procedure number, procedure content, procedure classification name, whether it is a high priority item, and whether it is a routine procedure.

The non-routine procedure manual DB63 contains non-routine procedures for the tasks "Temperature change request," "Shutter malfunction," and "Toilet clog."

The steps for a "temperature change request" are "Accept a request that the temperature is too hot/cold" (step 1), "Move to the temperature measurement location" (step 2), "Change the temperature setting" (step 3), and "Report the temperature change to the requester" (step 4). In contrast, the procedure classification names for steps 1 to 4 have been determined to be "Temperature abnormality request," "Move to temperature measurement," "Change temperature setting," and "Temperature change report."

The procedures for "shutter malfunction" are "Receive notification of shutter malfunction" (procedure number 1) and "Move to the location of the target shutter" (procedure number 2). In contrast, the procedure classification names for procedures 1 and 2 have been determined to be "Shutter malfunction request" and "Move to shutter."

The steps for "toilet clog" are "Accept report of toilet malfunction" (step 1), "Move to the affected toilet" (step 2), "Check the extent of the blockage in the affected toilet" (step 3), "Check for leaks" (step 4), "Flush the clogged toilet with a plunger" (step 5), and "Wipe up any leaking water around the toilet" (step 6). In contrast, the procedure classification names for steps 1 to 6 have been determined to be "Accept report of toilet malfunction," "Move to toilet," "Check for clogged toilet," "Check for toilet leak," "Pluggage work," and "Wipe up toilet leak."

In this way, the memory unit 25 stores the routine procedure manual DB 62 and the unroutine procedure manual DB 63 as business databases that record the procedures for the business performed by the administrator 5.

Steps S111 to S114 in Figure 4 correspond to step 4 in Figure 1. In S111, the control unit 21 acquires data recorded in the high-priority training item DB 64, routine task DB 62, and non-routine task DB 63. This acquires the occurrence frequency (number of occurrences, ranking) of event classification names recorded in the high-priority training item DB 64, the routine task procedures recorded in the routine procedure manual DB 62, and the non-routine task procedures recorded in the non-routine procedure manual DB 63.

In S112, if the event classification name recorded in the high priority training item DB 64 matches the task name in the non-routine task DB 63, the control unit 21 records a circle in the high priority item column of the non-routine task DB 63. In S113, if the procedure classification name with a circle in the high priority item column of the non-routine task DB 63 matches the procedure classification name in the routine task DB 62, the control unit 21 records a circle in the routine procedure column of the non-routine task DB 63. Specific examples are explained below.

Returning to Figure 8, in the non-routine procedure manual DB 63, "Temperature change request" is ranked "1st" (within the top 10) in the high-priority training item DB 64. Also, in the non-routine procedure manual DB 63, "Toilet clog" is ranked "5th" (within the top 10) in the high-priority training item DB 64. On the other hand, "Shutter malfunction" is not ranked within the top 10 in the high-priority training item DB 64. For this reason, "Temperature change request" and "Toilet clog" are set as "high-priority items" in the non-routine procedure manual DB 63.

The "event classification name" in the high-priority training item DB 64 (or equipment daily report DB 61) corresponds to the "task name" in the non-routine task DB 63. Therefore, if a "task name" in the non-routine task DB 63 is listed in the "event classification name" in the high-priority training item DB 64, this indicates that the non-routine task occurs with a high frequency (within the top 10 in frequency of occurrence). In this way, the control unit 21 sets the occurrence frequency of non-routine tasks (sets high-priority items) based on the frequency of occurrence (number of occurrences, ranking) of the event classification name.

Furthermore, procedures with a circle in the high priority item column in the non-routine procedure manual DB 63 that overlap with procedure classification names in the routine procedure manual DB 62 are extracted. Since there is overlap between the non-routine procedure manual DB 63 and the routine procedure manual DB 62 for procedures with a circle in the high priority item column, such as "Change temperature setting," "Move toilet," and "Check for clogged toilet," these are set as "routine procedures."

Returning to Figure 4, in S114, the control unit 21 extracts procedures that have a circle in the high priority item column of the unstructured procedure manual DB 63 and no circle in the structured procedure column, and records these procedures in the priority training procedure table 65.

Figure 9 shows an example of the priority training procedure table 65. The priority training procedure table 65 records the task names for tasks that should be trained on as a priority, as well as the procedure content and procedure classification names.

In the non-routine procedure manual DB 63, items that are set as "high priority items" but not set as "routine procedures" are extracted and recorded in the priority training procedure table 65. As a result, procedure numbers 1 (temperature abnormality request), 3 (temperature setting change), and 4 (temperature change report) for "temperature change request" and procedure numbers 1 (toilet malfunction reception), 4 (toilet leak check), 5 (plunger cup work), and 6 (toilet leak wipe) for "toilet clog" are extracted and recorded in the priority training procedure table 65.

In this way, the control unit 21 extracts high-frequency tasks (tasks with a circle in the high priority item column) from among the non-routine tasks, whose occurrence frequency is equal to or greater than a reference value (ranking within 10). The control unit 21 determines that a first procedure and a second procedure match when a first classification name (temperature measurement movement) corresponding to a first procedure (procedure number: 2) included in a routine task (for example, indoor temperature patrol (daily)) matches a second classification name (temperature measurement movement) corresponding to a second procedure (procedure number: 2) included in the non-routine task (temperature change request). The control unit 21 extracts, from among the procedures included in the high-frequency tasks (tasks with a circle in the high priority item column), procedures included in routine tasks (tasks with a circle in the routine procedure column) as training items (priority training procedures).

Returning to Figure 4, S115 corresponds to step 5 shown in Figure 1. In S115, the control unit 21 reads the priority training procedure table 65 and generates display information as the priority training procedure. The generated display information is displayed on the terminal 100.

As explained above, the control unit 21 extracts frequently occurring non-routine tasks, compares the procedures for the extracted frequently occurring non-routine tasks with the procedures for routine tasks, and extracts procedures that cannot be mastered through routine tasks as items that should be trained on a priority basis (priority training procedures).In doing so, the comparison is made after classifying the procedures into comparable formats (classification names) based on the procedure manual text.

By learning the priority training procedures, Manager 5 can learn procedures for frequently occurring events (non-routine tasks) that are not routine tasks that they have already mastered. Therefore, no matter who becomes the OJT trainer, Manager 5, who manages the building, can efficiently learn the tasks (non-routine tasks).

The learning process for the estimation model in this embodiment will be described below using Figures 10 and 11. Figure 10 is a diagram illustrating the learning of the estimation model. Hereinafter, the "trained model 31" before learning is completed will also be referred to as the "estimation model 31."

As explained using Figure 5, in this embodiment, the "event classification name" is determined based on the "event content" (text from the equipment daily report) recorded in the equipment daily report DB61. In the example of Figure 5, the event classification name "Indoor temperature response" is determined based on the event content "A store employee contacted us to request that something be done about the high temperature in the hallway in front of the store."

In this embodiment, the equipment daily report DB61 is configured to record only incidents (complaints, requests, inquiries, etc.), but it may also be configured to record not only incidents but also the causes of the incidents and the results of the manager 5's response to the incidents.

In building management, there is industry-specific terminology and building-specific terminology, and the terminology entered into the daily equipment report is often not standardized. For this reason, in this embodiment, the terminology is standardized using the word dictionary 34, and then the event classification name is determined using a trained model.

The sentence described as "Content of the Occurrence Event" has a sentence structure specific to building management, and can be expressed concisely using characteristic words used in building management. In the embodiment, this concise phrase is replaced with a combination of two words. It is appropriate for these two words to be a combination of a word corresponding to the subject and a word corresponding to the predicate, or a combination of a word corresponding to the predicate and a word corresponding to the object.

For example, in the example in Figure 5, this can be expressed using the two words "room temperature" + "response." Other examples include "bulb replacement" = "bulb" + "replace," and "toilet clog" = "toilet" + "clog."

The word dictionary 34 is a dictionary that associates word representative words that represent words used in the equipment daily report data (contents of occurrence events) with word related words that have meanings related to the word representative words. The trained model 31 is a model used to output estimation results based on the equipment daily report data (contents of occurrence events). Word representative words include words that are equivalent to superordinate concepts of word related words, but they do not necessarily have to be words that are equivalent to superordinate concepts as long as they have a related meaning.

For example, the attribute related word corresponding to the word representative word "water supply" is "faucet." The attribute related words corresponding to the word representative word "lighting" are "light," "fluorescent light," "lamp," and "light." The attribute related words corresponding to the word representative word "desk" are "table" and "desk." The attribute related word corresponding to the word representative word "correspondence" is "deal." There are no attribute related words corresponding to the word representative word "window."

The above combinations of word-related words corresponding to word representative words are merely examples, and other words may be used to associate them. The word-related words for a word representative word may be selected manually or by machine learning.

Below, we will explain how to output event classification names (hereinafter simply referred to as "classification names") from equipment daily report data (occurrence event details).

In the example of Figure 10, the control unit 21 extracts word representative words from the equipment daily report data (event content) using the word dictionary 34. Specifically, if the event content includes a word representative word or a word-related word, the control unit 21 extracts the corresponding word representative word. For example, if the event content includes any of the words "illumination," "light," "fluorescent lamp," "lamp," or "akari," the control unit 21 extracts "illumination" (word representative word).

In this example, the three words "corridor," "temperature," and "response" are extracted from the equipment daily report data 45a, "A store employee has reported that the temperature in the hallway in front of the store is high and would like you to take action..." By extracting words using the word dictionary 34, only specialized terms used in building management and words suitable for extracting classification names are extracted. Note that the above is merely an example, and other words (such as "store") may also be extracted.

The control unit 21 inputs words with meanings related to words contained in the equipment daily report data (event content) into the trained model 31, and outputs classification names from the trained model 31.

In this example, as shown in estimation result 45b, when the words "corridor," "temperature," and "response" are input to the trained model 31, the trained model 31 outputs "indoor temperature response" as the classification name. The control unit 21 registers the output event classification name (classification name) in the equipment daily report DB 61.

Note that the classification name is not limited to a combination of two representative words, but may be any other combination of words. It may also be a combination of three or more words, or may be a sentence.

The learning process is explained in detail below. In the learning process, the learning dataset is a combination of data in which "words" are input and "category names" are output.

The trained model 31 (estimation model 31) is a model that has undergone machine learning processing using training data so that when a word having a meaning related to a word contained in the equipment daily report data (event content) is input, it outputs an event classification name (classification name). Here, a word having a meaning related to a word contained in the event content is a word extracted from the words contained in the event content using the word dictionary 34, and is a word that is characteristically used when managing a building (in this embodiment, a "representative word").

The training dataset 32 includes a training dataset consisting of word 01 and classification name 01, a training dataset consisting of word 02 and classification name 02, and a training dataset consisting of word 03 and classification name 03.

A "word" is one or more words extracted using the word dictionary 34. For example, word 01 is "corridor, temperature, correspondence." The classification name 01 corresponding to "corridor, temperature, correspondence" is "indoor temperature correspondence." In other words, if the words extracted using the word dictionary 34 include "corridor, temperature, correspondence," learning is performed so that the classification name becomes "indoor temperature correspondence."

The above describes an example of generating an event classification name (classification name) based on equipment daily report data (occurrence event content). Using a similar method, a "procedure classification name (also referred to as a "classification name")" is generated from the "procedure content" in the routine procedure manual DB 62 in Figure 7, and a "procedure classification name (classification name)" is generated from the "procedure content" in the unroutine procedure manual DB 63 in Figure 8.

For example, an inference result 46b is obtained from procedure content 46a in the routine procedure manual DB 62. As with the equipment daily report data 45a, words (representative words) are extracted from procedure content 46a "move to the toilet" using the word dictionary 34. Here, the two words "toilet" and "move" are extracted from procedure content 46a "move to the toilet."

The control unit 21 inputs words (word representative words) that have meanings related to the words included in the procedure content into the trained model, and outputs a procedure classification name (classification name) from the trained model. The trained model used here, like trained model 31, is a model that has undergone a training process using a training dataset that is a combination of data in which "words" are input and "procedure classification names (classification names)" are output. For example, the training dataset includes word 01 "toilet, move" and its corresponding classification name 01 "toilet move." The method for obtaining a procedure classification name (classification name) from the procedure content in the unstructured procedure manual DB 63 is similar.

Figure 11 is a flowchart of the learning process. The control unit 21 executes the learning process. As shown in Figure 11, when the learning process starts, the control unit 21 selects learning data from the learning dataset 32 in S11 and proceeds to S12. In S12, the control unit 21 inputs the word data of the selected learning data into the estimation model 31 and proceeds to S13.

In S13, the control unit 21 performs an estimation process using the estimation model 31, outputs the estimation result, and proceeds to S14. In S14, the control unit 21 updates the parameters of the estimation model 31 based on the error between the estimation result and the ground truth data corresponding to the training data, and proceeds to S15.

In S15, the control unit 21 determines whether learning has been performed based on all of the learning data. If it is determined that learning has been performed based on all of the learning data (YES in S15), the process proceeds to S16. If it is determined that learning has not been performed based on all of the learning data (NO in S15), the process returns to S11. In S16, the control unit 21 stores the learned estimation model 31 as the learned model 31, and ends the learning process.

As described above, in this embodiment, the training item extraction system 1 includes a memory unit 25 and a control unit 21. The memory unit 25 stores a routine procedure manual DB 62 and an unroutine procedure manual DB 63 as task databases that record the procedures for tasks performed by the manager 5 who manages the building. The control unit 21 extracts, as training items, the procedures for tasks that the manager 5 should prioritize mastering. The tasks performed by the manager 5 include routine tasks as an example of a first task that the manager 5 periodically repeats, and unroutine tasks as an example of a second task that is different from the first task. The control unit 21 acquires procedures for routine tasks and procedures for unroutine tasks from the task databases (routine procedure manual DB 62, unroutine procedure manual DB 63). The control unit 21 extracts, from the unroutine tasks, high-frequency tasks that occur at a frequency equal to or greater than a reference value. The control unit 21 extracts, from the procedures included in the high-frequency tasks, procedures included in the routine tasks, excluding the procedures included in the routine tasks, as training items.

Manager 5 can learn procedures for frequently occurring events (non-routine tasks) that are not routine tasks that they have already mastered. This allows Manager 5, who manages the building, to efficiently learn non-routine tasks, regardless of who becomes the OJT trainer.

The memory unit 25 further stores an equipment daily report DB (database) 61, which stores equipment daily report data in which events that occur related to building management are recorded in text form. The control unit 21 outputs an event classification name that expresses the content of the event by simplifying the text recorded in the equipment daily report data. The control unit 21 calculates the occurrence frequency of the event classification name corresponding to the equipment daily report data recorded in the equipment daily report DB 61. The control unit 21 sets the occurrence frequency of non-routine tasks based on the occurrence frequency of the event classification name. This allows the manager 5 to prioritize learning non-routine tasks that actually occur frequently in the buildings he manages.

The storage unit 25 further stores a word dictionary 34 that associates word representative words representing words used in the equipment daily report data with word-related words that have meanings related to the word representative words, and a trained model 31 used to output estimation results based on the equipment daily report data. The trained model 31 is a model that has undergone machine learning processing using training data so that, when a word representative word is input as a feature, an event classification name is output as an estimation result. The control unit 21 uses the word dictionary 34 to extract word representative words from the equipment daily report data. The control unit 21 inputs the extracted word representative words as features into the trained model, and outputs an event classification name as an estimation result from the trained model. This makes it possible to classify the content of equipment daily report data input as text without human intervention, and to grasp the frequency of non-routine tasks based on equipment daily report data actually input in buildings managed by the manager 5.

Based on a sentence describing a procedure for a routine task, the control unit 21 simplifies the sentence and outputs a first classification name that represents the content of the procedure for the routine task. Based on a sentence describing a procedure for a non-routine task, the control unit 21 simplifies the sentence and outputs a second classification name that represents the content of the procedure for the non-routine task. The control unit 21 determines that the first procedure and the second procedure match when the first classification name corresponding to the first procedure included in the routine task matches the second classification name corresponding to the second procedure included in the non-routine task. This makes it possible to easily determine, without human intervention, whether the content of the procedure for a routine task entered in text matches the content of the procedure for a non-routine task entered in text.

The embodiments disclosed herein are illustrative and are not limited to the above. The scope of the present invention is defined by the claims, and it is intended to include all modifications that are equivalent to and within the scope of the claims.

1. Training item extraction system, 5. Administrator, 10. Server device, 21, 121. Control unit, 22, 122. ROM, 23, 123. RAM, 24, 124. Communication interface, 25, 125. Storage unit, 31. Trained model, 32. Training dataset, 34. Word dictionary, 45a. Equipment daily report data, 45b. Estimation results, 46a. Procedure content, 46b. Estimation results, 61. Equipment daily report DB, 62. Routine procedure manual DB, 63. Non-routine procedure manual DB, 64. High-priority training item DB, 65. Priority training procedure table, 71. Equipment daily report, 72. Routine work procedure manual, 73. Non-routine work procedure manual, 100. Terminal, 126. Input unit, 127. Display unit.

Claims (5)

a storage device that stores a business database in which business procedures performed by a manager who manages a building are recorded, and an equipment daily report database in which equipment daily report data in which events occurring in relation to the management of the building are recorded in text form ;
a control device that extracts, as training items, work procedures that the manager should be prioritized to master;
The tasks performed by the administrator include a first task that the administrator periodically repeats and a second task that is different from the first task,
The control device
acquiring a first procedure included in the first task and a second procedure included in the second task from the task database;
extracting, from the second tasks, high-frequency tasks whose occurrence frequency calculated based on the event classification name corresponding to the equipment daily report data is equal to or greater than a reference value;
extracting, from among the procedures included in the high-frequency work, procedures excluding the first procedure by determining whether a first classification name corresponding to the first procedure matches a second classification name corresponding to the second procedure , as the training items ;
The event classification name is a name that represents the content of the event,
the first classification name is a name that represents the content of the procedure of the first business;
the second classification name is a name that represents the content of the second business procedure,
In the matching determination, when the first classification name and the second classification name match, it is determined that the first procedure and the second procedure match . The control device
Based on the sentence recorded in the equipment daily report data, the sentence is simplified and the event classification name is output;
Calculating the frequency of occurrence of the event classification name corresponding to the equipment daily report data recorded in the equipment daily report database;
The extraction system according to claim 1 , wherein the occurrence frequency of the second task is set based on the appearance frequency of the event classification name. The storage device further stores a word dictionary in which word representative words representing words used in the equipment daily report data are associated with word-related words having meanings related to the word representative words, and a trained model used to output an estimation result based on the equipment daily report data,
the trained model is a model that has been subjected to machine learning processing using training data so as to output the event classification name as the estimation result when the word representative word is input as a feature;
The control device
extracting the representative words from the daily equipment report data using the word dictionary;
The extraction system according to claim 2, wherein the extracted word representative words are input to the trained model as the features, and the event classification name is output from the trained model as the estimation result. The control device
based on a sentence explaining the procedure of the first business, simplifying the sentence and outputting the first classification name;
based on a sentence explaining the procedure of the second business, simplifying the sentence and outputting the second classification name;
An extraction system according to any one of claims 1 to 3, wherein the matching determination is performed to determine that the first procedure and the second procedure match when the first classification name and the second classification name match. 1. A computer-implemented extraction method comprising:
a step of storing in a storage device a business database in which business procedures performed by a manager who manages a building are recorded, and an equipment daily report database in which equipment daily report data in which events occurring in relation to the management of the building are recorded in writing ;
extracting, as training items, business procedures that the manager should be prioritized to master;
The tasks performed by the administrator include a first task that the administrator periodically repeats and a second task that is different from the first task,
The extracting step includes:
acquiring a first procedure included in the first task and a second procedure included in the second task from the task database;
extracting, from the second tasks, high-frequency tasks whose occurrence frequency calculated based on the event classification name corresponding to the equipment daily report data is equal to or greater than a reference value;
extracting, from among the procedures included in the high-frequency work, procedures excluding the first procedure by determining whether a first classification name corresponding to the first procedure matches a second classification name corresponding to the second procedure , as the training items ;
The event classification name is a name that represents the content of the event,
the first classification name is a name that represents the content of the procedure of the first business;
the second classification name is a name that represents the content of the second business procedure,
An extraction method , wherein, in the match determination, when the first classification name and the second classification name match, it is determined that the first procedure and the second procedure match .