JP7531384B2 - Needs display system, trained model generation method, and control method - Google Patents

Description

Article 30, paragraph 2 of the Patent Act applies. June 17, 2020 Published in the paper "Proposal for a method of improving the efficiency of building facility management by linking humans and sensor information" published in Session 6A-2 of the DICOMO2020 Symposium held online at the website at the following address (http://dicomo.org/) (http://tsys.jp/dicomo/2020/program/program_abst.html) September 10, 2020 Published in the paper "Proposal for a method of improving the efficiency of building facility management by linking humans and sensor information" published in Session 6A-2 of the IWIN2020 Symposium held online at the website at the following address (http://tsys.jp/dicomo/2020/program/program_abst.html) Sense and Information of BEMS" (http://www.infsoc.org/conference/iwin2020/) (http://www.infsoc.org/conference/iwin2020/program)

This disclosure relates to a needs display system, a trained model generation method, and a control method.

Building owners sometimes survey the tenants of their buildings to find out their dissatisfaction and satisfaction with the building's facilities. In such cases, the survey is often conducted by interviewing the tenants during meetings or by conducting questionnaires.

However, interviews and surveys cannot be conducted frequently because it takes time to coordinate time with participants, record responses, and compile the recorded data. In addition, residents often have difficulty recalling past memories when answering interviews or surveys. In addition, they tend to answer about events that made a strong impression on them or events that are easy to remember, which can lead to bias in the survey data.

When investigating dissatisfaction and satisfaction with building facilities, it is possible to utilize historical data already accumulated in the management of building facilities. An example of a system for accumulating such historical data is JP 2019-121392 A (Patent Document 1).

JP 2019-121392 A

However, while the system in Patent Document 1 can search accumulated historical data, it cannot directly extract needs for building facilities, such as dissatisfaction or satisfaction with the facilities.

The present disclosure has been made to solve the above-mentioned problems, and its purpose is to provide a needs display system that can optimally extract the needs of building users from historical data related to the management of building facilities.

According to an aspect of the present disclosure, a needs display system aggregates and displays the needs of building users. The needs display system includes a control unit, a storage unit, and a display unit. The storage unit stores a word dictionary and a trained model. The word dictionary is a dictionary in which historical data related to management of building facilities, a word representative word representing a word used in the historical data, and a word-related word having a meaning related to the word representative word are associated with each other. The trained model is a model used to output an estimation result based on the historical data. The display unit displays the estimation result. The trained model is a model that has been subjected to machine learning processing using teacher data so that, when a word representative word is input as a feature, a need and a location or date and time when the need occurred are output as an estimation result. The control unit uses the word dictionary to extract a word representative word from the historical data. The control unit inputs the extracted word representative word as a feature into the trained model, and outputs a need and a location or date and time when the need occurred from the trained model as an estimation result. The control unit aggregates the estimation results for each need. The display shows the estimated results, aggregated by need.

According to another aspect of the present disclosure, the trained model generation method is a method for application to a needs display system that aggregates and displays the needs of building users. The needs display system includes a storage unit that stores historical data related to the management of building facilities, a word dictionary in which word representative words representing words used in the historical data are associated with word related words having meanings related to the word representative words. The trained model generation method includes a step of selecting training data from a training dataset, a step of inputting the word representative words included in the training data into an estimation model as features, thereby outputting the needs and the location or date and time when the needs occur as estimation results from the estimation model, and a step of updating parameters of the estimation model based on an error between the estimation result and ground truth data corresponding to the training data.

According to yet another aspect of the present disclosure, a control method controls a needs display system that aggregates and displays the needs of building users. The needs display system includes a storage unit that stores historical data related to management of building facilities, a word dictionary in which word representative words representing words used in the historical 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 historical data. The control method includes a step of extracting word representative words from the historical data using the word dictionary, a step of inputting the extracted word representative words as features into the trained model to output the needs and the location or date and time when the needs occur as estimation results from the trained model, a step of aggregating the estimation results for each need, and a step of displaying the estimation results aggregated for each need. The trained model is a model that has been subjected to machine learning processing using teacher data so that when the word representative words are input as features, the model outputs the needs and the location or date and time when the needs occur as estimation results.

According to the present disclosure, the needs of building users can be ideally extracted from historical data related to the management of building facilities. This makes it possible to extract latent needs of building users such as tenants that cannot be collected through interviews or questionnaires. In addition, because historical data recorded each time is used as an information source, it is possible to extract needs that building users have forgotten, and it is possible to collect needs data that exists only on-site and does not reach the building owner. This makes it possible to improve building operations and facilities for building users that have not been noticed until now.

1 is a schematic diagram showing an example of a hardware configuration of a search system and a needs display system according to an embodiment of the present invention. FIG. 2 is a block diagram showing an example of a functional configuration of a search system and a needs display system. FIG. 2 is a diagram illustrating an example of history data and classification of the history data. FIG. 1 is a diagram illustrating learning of an estimation model in a search system. 13 is a flowchart of a learning process (for history search). FIG. 10 is a diagram illustrating a procedure for generating a search database from history data. FIG. 4 is a diagram illustrating an example of a search database. 13 is a flowchart of a history search process. FIG. 11 is a diagram illustrating a history search process. FIG. 11 is a diagram illustrating a history search process. FIG. 13 is a diagram illustrating an example of a history search result. FIG. 13 is a diagram illustrating learning of an estimation model in the needs display system. 13 is a flowchart of a learning process (for extracting needs). 13 is a flowchart of a needs extraction process. FIG. 1 is a diagram illustrating a flow of needs extraction in a needs display system. FIG. 11 is a diagram showing an example of a needs extraction result.

The following describes the embodiment with reference to the drawings. In the following description, the same components are given the same reference numerals. Their names and functions are also the same. Therefore, detailed descriptions thereof will not be repeated.

[Configuration of search system and needs display system]
First, an overview of a search system 1a and a needs display system 1b according to the present embodiment will be described. Fig. 1 is a schematic diagram showing an example of a hardware configuration of the search system 1a and the needs display system 1b according to the present embodiment.

As shown in FIG. 1, in this embodiment, the search system 1a includes a server device 10 and a terminal 100. The server device 10 and the terminal 100 are configured to be able to communicate with each other. The search system 1a is a system that searches for historical data related to the management of building facilities.

The server device 10 includes a storage unit 25. The storage unit 25 stores an attribute dictionary 33, a word dictionary 34, and a search database 36 including history data. When a user performs an operation to search history on the terminal 100 of the search system 1a, a search for history data (also referred to as a "history search") is performed using the attribute dictionary 33, the word dictionary 34, and the search database 36, and as a result, a history search result 38 is displayed. The "history data" and "history search" will be described in detail with reference to Figures 3 to 11.

The needs display system 1b also includes a server device 10 and a terminal 101. The server device 10 and the terminal 101 are configured to be able to communicate with each other. The needs display system 1b is a system that collects and displays the needs of building users. In this embodiment, "needs" refers to needs based on events that users find unpleasant at the place or date and time when the needs occur, or events that users require from building facilities in order to use the building comfortably.

In this embodiment, "building users" refers to tenants who reside in the building, but may also refer simply to users (such as shoppers) who use a commercial building.

The terminal 101 of the needs display system 1b is capable of aggregating and displaying the needs of building users (also referred to as "needs extraction"). When a user performs an operation to extract needs on the terminal 101, needs extraction is performed using the history data, the word dictionary 34, and the trained model 41 for extracting needs, and as a result, the needs extraction result 48 is displayed.

In this embodiment, a common word dictionary 34 and history data are used as the history data and the data used for needs extraction. History data is essentially data that records the history of responses made by a building facility manager to complaints from building users. This history data includes the needs of building users. Focusing on this point, this embodiment is configured to enable needs extraction using a trained model 41 that has been trained to be able to extract needs from history data. In this embodiment, the "building facility manager" refers to a person in charge of managing the building facilities. Needs extraction will be described in detail using Figures 12 to 16.

Figure 2 is a block diagram showing an example of the functional configuration of the search system 1a and the needs display system 1b. As described above, the search system 1a includes a server device 10 and a terminal 100, and the needs extraction system 1b includes a server device 10 and a terminal 101.

As shown in FIG. 2, 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 the programs stored in the ROM 22 into the RAM 23 and executes them. The ROM 22 stores programs in which the processing procedures of the server device 10 are written. The RAM 23 serves as a working area when the control unit 21 executes the programs, and temporarily stores the programs and data used when executing the programs.

The communication interface 24 is an interface for communicating with the terminals 100 and 101. The storage unit 25 is a non-volatile storage device. The storage unit 25 may be, for example, a HDD (Hard Disk Drive).

The memory unit 25 stores a trained model 31, a training data set 32, a trained model 41, a training data set 42, an attribute dictionary 33, a word dictionary 34, and a search database 36. The trained model 31 and the training data set 32 are used for training for history search. The trained model 41 and the training data set 42 are used for training for needs extraction. These will be described in detail using FIG. 4 and subsequent figures.

The terminals 100 and 101 include 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 terminals 100 and 101 may be desktop computers, notebook computers, or tablet terminals. Furthermore, the system is not limited to one in which the history search is performed on the terminal 100 and the needs extraction is performed on the terminal 101, and may be configured so that the history search and needs extraction are performed on a single terminal.

The server device 10 may also be configured to operate as a web server and display history search results 38 and needs extraction results 48 by accessing a specified URL from the terminals 100 and 101. Alternatively, software for performing history searches and needs extraction may be installed on the terminals 100 and 101, and the software may be operated to display the history search results 38 and needs extraction results.

The control unit 121 controls the entire terminal 100, 101. The control unit 121 loads a program stored in the ROM 122 into the RAM 123 and executes it. The ROM 122 stores a program in which the processing procedures of the terminal 100, 101 are written. The RAM 123 serves as a working area when the control unit 121 executes a program, and temporarily stores the program, data when the program is executed, and the like.

The communication interface 124 is an interface for communicating with the server device 10. The storage unit 125 is a non-volatile storage device. The storage unit 125 may be, for example, a HDD.

The input unit 126 accepts inputs including instructions from the user to the information processing device 90. The input unit 126 is, for example, a keyboard, a mouse, or a touch panel. The display unit 127 displays various images such as the history search results 38 and the needs extraction results 48. The display unit 127 is, for example, a liquid crystal display or a display.

[Historical data and its classification in search systems]
3A and 3B are diagrams showing an example of history data and classification of the history data, Fig. 3A is a diagram showing an example of the history data, and Fig. 3B is a diagram showing an example of classification of the history data.

The building manager enters historical data related to the management of building facilities. Historical data is data that is recorded as a history in the form of a daily report when the building facility manager responds to inquiries or complaints.

Since there is no set way of writing words and sentences, different words are often used depending on the person entering the data. For this reason, when someone who does not enter sentences, such as an administrator, or a person other than the person in charge searches historical data, the search accuracy drops in conventional search systems because the appropriate search keywords are not known.

This has resulted in variations in the speed and accuracy of response depending on the person in charge of the work. In particular, in a system where staff rotate 24 hours a day, 365 days a year, the quality of response varies depending on the staff on duty at the time, making it impossible to maintain a consistent quality of building management services. In this embodiment, the search system 1a is equipped with a configuration to solve such problems.

Building equipment is typically replaced every 10 to 15 years, and the equipment replacement cycle is long. For this reason, it takes time to gain experience in building management, as knowledge of old equipment is required. In addition, it is necessary to learn industry terminology and the building's unique terminology, and the terminology used by building equipment managers often differs from that used by tenants. Due to these industry-specific circumstances, when performing searches using a search system, it is necessary to build a search system that effectively links the words used by the person performing the search, the words used by the person who registered the historical data, and the words used by building users.

In particular, in building management, there are many tasks that require a quick response, such as plumbing problems, so knowing how to quickly retrieve information from the search system is a key point.

An example of history data will be described below. As shown in FIG. 3A, for example, history data No. 1 (also referred to as "history data 1") contains the following entry: "Driver's mobile phone dropped into mechanical parking pit. The disaster prevention center has requested that we take action as the driver dropped his mobile phone into the pit of B3 parking space 188 (mechanical parking). We have confirmed that the mobile phone has fallen into the pit. As it may be necessary to arrange for a mechanical parking maintenance company to take action, we have decided to have the parking management take action. The parking lot manager and a disaster prevention center representative are present to confirm. 18:00 The parking lot manager has contacted us to say that the mobile phone has been retrieved from the pit."

In history data No. 2 (also referred to as "history data 2"), the following is entered: "Replacing the bulbs of the indirect lighting displaying the company name at the reception on the 28th floor. Tenant A requested that we replace the bulbs of the indirect lighting displaying the company name at the reception on the 28th floor at night. 19:30 Two tenant-supplied bulbs (model number xxx) were replaced and the lights were confirmed to work. Tenant A was notified by email."

In history data No. 3 (also referred to as "history data 3"), the following is entered: "Tenant facilities (18th floor conference room light switch pushed in and collapsed) 4/2 14:40 Tenant B requested that we take action as the light switch in the conference room had been pushed in and collapsed. Situation upon arrival 14:15 Confirmed that the light switch in the conference room had been pushed in and collapsed. As the plastic mounting frame was unable to secure the switch in place, it was replaced with a spare metal mounting frame and the switch was secured in place. Tenant B was present to confirm. Reported to the owner via email."

In historical data No. 4 (also referred to as "historical data 4"), the following is entered: "Regarding the strange noise coming from the entrance door to the office on the 10th floor: An inspection discovered an strange noise coming from the entrance door to the office on the 10th floor. Condition at the time of inspection: A rubbing sound was heard between the electric lock code and guide on the side of the door when opening and closing the door. After adjusting the position of the code and applying grease, the strange noise was confirmed to have been resolved. The owner was notified via email."

Such historical data text can be divided and classified into three types of text. The first type is text about the content of an inquiry made by a building user about the building facilities (hereinafter simply referred to as "inquiry content"). "Inquiry content" can be any contact made by a building user to the building facilities manager (such as a complaint, consultation, or contact).

The second is a statement about the reason why the building user inquired about the content (hereinafter referred to simply as the "cause"). The "cause" corresponds to the reason that the "action" below was taken.

The third type is a sentence about the building facilities manager's response to an inquiry from a building user (hereinafter, simply referred to as "response"). "Response" also includes cases where the building facilities manager has performed some kind of work. The sentences in the historical data include at least one of the above three sentences.

For example, consider a case where some kind of malfunction occurs in a building facility, and a building user complains (makes an inquiry) to the building facility manager. In this case, the building facility manager will determine the "cause" of the malfunction (e.g., a component failure) and then take "action" to resolve the malfunction (e.g., replacing the failed component). The building facility manager will then compile a daily report on the "inquiry content," "cause," and "action," and store this in the memory unit 25 of the server 10 as historical data.

The sentences corresponding to "inquiry content," "cause," and "response" can be replaced with concise words. In this embodiment, they are replaced with a combination of two words as such concise words. It is appropriate that these two words are 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 history data 1, the "inquiry content" can be expressed in two words, "lost item" + "response," and the "response" can be expressed in two words, "maintenance company" + "arrangement." Note that there is no sentence equivalent to "cause" in history data 1.

For historical data 2, the "inquiry content" can be expressed in two words, "bulb" + "replacement," the "cause" can be expressed in two words, "bulb" + "out," and the "response" can be expressed in two words, "nighttime" + "replacement."

For historical data 3, the "inquiry content" can be expressed in two words, "lighting" + "malfunction," the "cause" can be expressed in two words, "switch" + "damage," and the "response" can be expressed in two words, "part" + "replacement."

In the case of history data 4, the "cause" can be expressed in two words, "part" + "rubbing," the "response" can be expressed in two words, "position" + "adjustment," and the "response" can be expressed in two words, "grease" + "application." Note that there is no sentence equivalent to the "inquiry content" in history data 4.

In this way, the historical data has a sentence structure that is specific to building facility management, and can be concisely expressed using characteristic words used in building facility management. As described above, the historical data can be categorized into sentences with three attributes: "inquiry content," "cause," and "response." Hereinafter, "inquiry content," "cause," and "response" will be referred to as "attribute representative words." In addition, a sentence that corresponds to each attribute representative word and is expressed in two words (such as "bulb replacement") will be referred to below as "attribute content."

From this perspective, the recorded history data is analyzed and classified into a table such as that shown in Figure 3 (B). As shown in Figure 3 (B), the most common attribute content corresponding to the "inquiry content" is "fluorescent light not working," with A1 occurrences. The second most common attribute content corresponding to the "inquiry content" is "suspicious object found," with A2 occurrences (A1 > A2). The third most common attribute content corresponding to the "inquiry content" is "toilet water leak," with A3 occurrences (A2 > A3).

"Fluorescent light not working" is composed of two words, "fluorescent light" and "not working." "Suspicious object found" is composed of two words, "suspicious object" and "found." "Toilet leak" is composed of two words, "toilet" and "leak." The contents of inquiries are tallied from 1st place to N1.

Similarly, "cause" and "response" are also tabulated. Causes are tabulated from 1st to N2nd place. Responses are tabulated from 1st to N3rd place. In this embodiment, a learning dataset for the trained model 31 used for history search is prepared based on the table in FIG. 3(B) in which the history data is classified. In the table in FIG. 3(B), the key points for improving the learning accuracy are what words are to be configured as combination attribute contents and up to what rank they are to be tabulated.

The learning process in the search system will be described in detail below. Hereinafter, the "trained model 31" before learning is completed will also be referred to as the "estimated model 31."

[Learning process in search system]
4 is a diagram illustrating the learning of the estimation model 31 in the search system 1a. In the learning process in the search system (also referred to as "learning process (for history search)"), the learning data set is a combination of data in which "words" are input and "attribute representative words" and "attribute contents" 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 historical data is input, one or more attribute representative words and attribute contents related to the attribute representative words are output. Here, a word having a meaning related to a word contained in the historical data is a word extracted from the words contained in the historical data using the word dictionary 34, and is a word that is characteristically used when managing a building (for example, a "word representative word" as described below).

As shown in FIG. 4, the learning dataset 32 includes a learning dataset consisting of word 01, attribute representative word 01, and attribute content 01, a learning dataset consisting of word 02, attribute representative word 02, and attribute content 02, and a learning dataset consisting of word 03, attribute representative word 03, and attribute content 03.

A "word" is one or more words extracted using the word dictionary 34. For example, word 01 is "repair." The attribute representative word 01 corresponding to "repair" is "response," and the attribute content 02 corresponding to "repair" is "part replacement."

In other words, if the words extracted using the word dictionary 34 include "repair," the historical data will include a sentence with the attribute "response," and learning is performed so that the two-word expression of a sentence with the attribute "response" is "part replacement."

Figure 5 is a flowchart of the learning process (for history search). The control unit 21 executes the learning process (for history search). Hereinafter, "step" is also referred to simply as "S".

As shown in FIG. 5, when the learning process (for history search) is started, the control unit 21 selects learning data from the learning dataset 32 (for history search) in S11 and proceeds to S12. In S12, the control unit 21 inputs the word data of the selected learning data (for history search) to the estimation model 31 and proceeds to S13.

In S13, the control unit 21 performs an estimation process using the estimation model 31 to output an 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 learning 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 trained estimation model 31 as the trained model 31 and ends the learning process (for history search).

Figure 6 is a diagram explaining the procedure for generating a search database 36 from history data. As described above, the storage unit 25 stores an attribute dictionary 33, a word dictionary 34, and a search database 36.

The attribute dictionary 33 is a dictionary in which attribute representative words that represent attributes of historical data are associated with attribute related words that have meanings related to the attribute representative words. Attribute representative words include words that are higher-level concepts than the attribute related words, but they do not necessarily have to be words that are higher-level concepts as long as they have a related meaning.

The attribute representative words include "inquiry content," which is a word that represents the content of an inquiry made by a building user about building facilities, "cause," which is a word that represents the reason why the user inquired about the content, and "response," which is a word that represents the response of the building facilities manager to the user's inquiry.

As shown in FIG. 5, for example, the attribute related words corresponding to the attribute representative word "inquiry content" are "doubt," "question," "complaint," and "alarm." The attribute related words corresponding to the attribute representative word "cause" are "reason" and "basis." The attribute related word corresponding to the attribute representative word "response" is "handle."

The word dictionary 34 is a dictionary in which word representative words that represent words used in the history data are associated with word related words that have meanings related to the word representative words. The word representative words include words that are equivalent to superordinate concepts of the 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 for the word representative word "water supply" is "faucet." The attribute related words for the word representative word "lighting" are "light," "fluorescent lamp," "lamp," and "light." The attribute related words for the word representative word "desk" are "table" and "desk." There is no attribute related word for the word representative word "window." The attribute related word for the word representative word "entrance" is "entrance."

The above-described combinations of attribute related words corresponding to attribute representative words and word related words corresponding to word representative words are merely examples, and correspondence may be achieved using words other than those mentioned above. Selection of attribute related words for attribute representative words is performed manually, taking into account circumstances specific to the industry related to building facility management. However, this is not limiting, and attribute related words for attribute representative words may be selected by machine learning. Similarly, word related words for word representative words may be selected manually or by machine learning.

Next, a method for generating the search database 36 from history data will be described. The search database 36 is a database in which search data is registered in which one or more attribute representative words are associated with history data and attribute contents related to the attribute representative words are associated with the history data.

In the illustrated example, a search database 36 is generated that corresponds to history data 35a that reads, "Mr. A from the reception on the first floor contacted us to say that the light wouldn't turn on, and as a nighttime replacement was requested, the repair was carried out at 8 p.m."

The control unit 21 extracts words from the history data 35a. The word extraction is performed using the word dictionary 34. Here, three words, "electricity," "not turn on," and "repair," are extracted from the history data 35a. By extracting words using the word dictionary 34, only technical terms used in facility management and words suitable for extracting attribute content are extracted.

The control unit 21 inputs words having meanings related to words included in the history data into the trained model 31, and outputs one or more attribute representative words and attribute contents related to the attribute representative words from the trained model 31.

In this example, as shown in estimation result 35b, when the words "electricity" and "not turning on" are input to trained model 31, trained model 31 outputs "inquiry content" as the attribute representative word and outputs "lighting malfunction" as the attribute content. In addition, when the word "repair" is input to trained model 31, trained model 31 outputs "response" as the attribute representative word and outputs "part replacement" as the attribute content.

Attribute content is composed of a combination of two word representative words. For example, "lighting malfunction" is composed of a combination of two word representative words, "lighting" + "malfunction." For example, "part replacement" is composed of a combination of two word representative words, "part" + "replacement."

The attribute content is not limited to being composed of a combination of two word representatives, but may be composed of any other combination of words. It may also be composed of a combination of three or more words, or may be composed of a sentence.

The control unit 21 associates the output attribute representative word and attribute content with the history data and registers them in the search database 36 as search data.

In this example, the control unit 21 registers in the search database 36 the history data 35a in which the attribute content "lighting malfunction" for the attribute representative word "inquiry content" and the attribute content "part replacement" for the attribute representative word "response" are associated. Note that in this example, there is no attribute content for the attribute representative word "cause."

Figure 7 is a diagram showing an example of the search database 36. The search database 36 shown in Figure 7 is generated from the history data shown in Figure 3 (A).

As shown in FIG. 7, in the history data 1, search data (hereinafter referred to as "search data 1") that associates the attribute content "parts replacement" with the attribute representative word "inquiry content" and the attribute content "maintenance company arrangement" with the attribute representative word "response" is registered in the search database 36.

In the history data 2, search data (hereinafter referred to as "search data 2") that associates the attribute content "bulb replacement" with the attribute representative word "inquiry content," the attribute content "bulb out" with the attribute representative word "cause," and the attribute content "nighttime replacement" with the attribute representative word "response" is registered in the search database 36.

In the history data 3, search data (hereinafter referred to as "search data 3") that associates the attribute content "lighting malfunction" with the attribute representative word "inquiry content," the attribute content "switch damage" with the attribute representative word "cause," and the attribute content "part replacement" with the attribute representative word "response" is registered in the search database 36.

In the history data 4, search data (hereinafter referred to as "search data 4") that associates the attribute content "part chafing" with the attribute representative word "cause" and the attribute contents "position adjustment" and "grease application" with the attribute representative word "response" is registered in the search database 36.

[History search processing in search system]
Next, the history search process in the search system 1a will be described. Fig. 8 is a flowchart of the history search process. The control unit 21 executes the history search process. As a result of the history search process, the display unit 127 displays the search results obtained by the control unit 21 searching the search database 36 based on the keyword input by the user.

As shown in FIG. 8, when the history search process is started, the control unit 21 uses the attribute dictionary 33 in S21 to extract attribute representative words that have meanings related to the keyword, and proceeds to S22. If there are multiple keywords, this process is performed for all the keywords. In S22, the control unit 21 determines the search data associated with the extracted attribute representative words as the search target of the search database 36, and proceeds to S23.

In S23, the control unit 21 uses the word dictionary 34 to extract word representative words and word related words that have meanings related to the keyword, and proceeds to S24. Here, the process of S23 is not performed for keywords from which attribute representative words have been extracted, and the process of S23 is performed for keywords from which attribute representative words have not been extracted.

In S24, the control unit 21 determines whether the search data determined as the search target contains the extracted word representative word or word related word. If it is determined that the search data determined as the search target contains the extracted word representative word or word related word (YES in S24), the process proceeds to S25. If it is not determined that the search data determined as the search target contains the extracted word representative word or word related word (NO in S24), the process proceeds to S26.

In S25, the control unit 21 determines that the search data containing the extracted word representative word or word related word is the search result to be displayed by the display unit 127, and proceeds to S25. In S26, the control unit 21 causes the display unit 127 to display the history data and attribute contents as the search result, and ends the history search process.

A specific example of the history search process is described below. Figures 9 and 10 are diagrams for explaining the history search process. As shown in Figure 9, assume that a user performs a search on the search screen 37 using two keywords, "complaint" and "fluorescent light."

The control unit 21 searches the attribute dictionary 33 to see if the keyword "claim" exists. Since the attribute dictionary 33 contains "claim" as an attribute-related word that corresponds to the attribute representative word "inquiry content," the control unit 21 extracts the attribute representative word "inquiry content."

Next, the control unit 21 searches the attribute dictionary 33 to see if the keyword "fluorescent light" exists. Since "fluorescent light" does not exist in the attribute dictionary 33, the control unit 21 searches the word dictionary 34 to see if the keyword "fluorescent light" exists. In the word dictionary 34, "fluorescent light" exists as a word-related word that corresponds to the word representative word "lighting". The control unit 21 extracts the word representative word "lighting" and the corresponding word-related words "light," "fluorescent light," "lamp," and "light."

As shown in FIG. 10, the control unit 21 determines search data 1 to 3 associated with the extracted attribute representative word "inquiry content" as search targets in the search database 36.

Here, the user is searching using the keyword "claim." The attribute dictionary 33 defines "inquiry content" as a broad concept similar to the higher-level concept of the concept of "claim." In this embodiment, the search data 1 to 3 associated with "inquiry content" are determined as the search targets of the search database 36, thereby expanding the search targets to include broader concepts higher than the concept of "claim."

The control unit 21 further checks whether the search data 1 to 3 determined as the search target contain any of the extracted word representative words and word related words ("lighting," "light," "fluorescent lamp," "lamp," and "light"). In this example, since the search data 1 and 2 contain "lighting," the search data 2 and 3 are determined as the search results to be displayed by the display unit 127.

Here, the user is searching for the keyword "fluorescent light." The word dictionary 34 defines "illumination," "light," "lamp," and "light" as words related to the word "fluorescent light." When entering history data, the word "fluorescent light" is not necessarily entered, but a different word may be entered. Also, related information may be recorded in the history data as a word that is a higher or lower concept of the keyword. In this embodiment, it is possible to detect history data that includes such information.

The search database 36 is composed of the highest level concept "attribute representative words" (inquiry content, cause, response), the mid-level concept "attribute content" (lighting + malfunction, etc.), and lower level concept words (historical data), allowing not only simple searches using lower level concept words, but also searches that expand to include mid-level and higher level concepts. In order to build an optimal search system using these three layers of concepts, the key is how to extract industry-specific words and concepts using the word dictionary 34, and how to optimally learn combinations of words, attribute representative words, and attribute content.

Figure 11 is a diagram showing an example of a history search result. As shown in Figure 11, the history search result 38 shows the results of a search performed by a user using two keywords, "complaint" and "fluorescent light."

As described above, search data 2 and 3 contain "Lighting", and therefore search data 2 and 3 are displayed in history search result 38. History search result 38 also contains information that can identify representative words or related words contained in the search data. Specifically, history search result 38 displays the characters "Lighting" surrounded by a square frame so that the characters can be identified.

In this embodiment, by configuring as described above, it is possible to efficiently search for history data that contains words with meanings related to the input keyword. This makes it possible to search for personal sentences using words that other staff members normally use, allowing the experience of others to be utilized in one's own work.

[Learning process in needs display system]
Next, a learning process in the needs display system 1b (hereinafter also referred to as a "learning process (for needs extraction)") will be described. Fig. 12 is a diagram for explaining learning of an estimation model 41 in the needs display system 1b.

As mentioned above, a "need" is a need based on the occurrence of an event that causes discomfort to users or a requirement of building facilities to allow comfortable use of the building at the location or date and time when the need occurs.

Building owners sometimes survey the building's tenants (building users) about their dissatisfaction and satisfaction with the building's facilities. In such cases, the survey is often conducted by interviewing the tenants during meetings or by conducting questionnaires. However, interview and questionnaire surveys cannot be conducted frequently because it takes time to coordinate time with participants, record their responses, and compile the recorded data. In addition, tenants often have difficulty recalling past memories when answering interviews or questionnaires. Furthermore, because they tend to answer about events that made a strong impression on them or events that are easy to remember, the survey data can be biased.

When investigating needs such as dissatisfaction and satisfaction with building facilities, one possibility is to utilize historical data already accumulated in the management of building facilities. However, because historical data is a record of the work done by building managers in response to complaints, etc., the purpose of accumulating historical data does not match the purpose of conducting a needs survey. For this reason, it is not possible to directly extract needs for building facilities, such as dissatisfaction and satisfaction with building facilities, from the historical data.

In consideration of the above, in this embodiment, the needs display system 1b is configured to suitably extract the needs of building users from historical data related to the management of building facilities. The trained model 41 acts as a bridge to effectively extract only the needs from this historical data.

By extracting needs in this way, it is possible to uncover the latent needs of building users, and by improving the building's operations and facilities, it is possible to further increase the added value of the building. In addition, by aggregating such needs, it can also be used as marketing data.

A more detailed explanation is given below. As described with reference to FIG. 2 etc., the storage unit 25 stores historical data, a word dictionary 34, and a trained model 41. As described above, the word dictionary 34 is a dictionary in which historical data relating to the management of building facilities, word representative words that represent words used in the historical data, and word related words that have meanings related to the word representative words are associated with each other.

The historical data is data that records the content of inquiries made by users regarding building facilities, the reason why the users inquired about the content, and the building facilities manager's response to the user's inquiry.

The trained model 41 is a model used to output an inference result based on historical data. The trained model 41 is a model that has been subjected to machine learning processing using training data so that, when a word representative word (hereinafter also simply referred to as a "word") is input as a feature, the trained model 41 outputs a need and the location or date and time when the need occurred (hereinafter also simply referred to as the "location and date and time") as an inference result.

As shown in FIG. 12, in the learning process in the needs display system 1b, a combination of data is prepared as a learning dataset, in which "words" are input and "needs" and "location/date and time" are output.

In this embodiment, the system is configured to output "needs" and "location/date and time", but this is not limiting. As with the learning process for history search, the learning dataset may contain attribute representative words. In other words, attributes such as "inquiry content" and "cause" may be assigned to "needs" and "location/date and time".

The training data set 42 includes a training data set consisting of words 01, needs 01, and location/date/time 01, a training data set consisting of words 02, needs 02, and location/date/time 02, and a training data set consisting of words 03, needs 03, and location/date/time 03.

A "word" is one or more words that have a meaning related to a word contained in the history data. Here, a word that has a meaning related to a word contained in the history data is a word extracted from the words contained in the history data using the word dictionary 34, and is a word that is characteristically used when managing a building (for example, a "word representative word").

For example, word 01 is a plurality of words including "umbrella" and "rent". The corresponding need 01 is "It would be nice to have an umbrella to lend", and place/time 01 is "entrance".

In other words, if the historical data contains words such as "umbrella" and "lend," the system will learn to recognize that there is a need for "it would be nice to have an umbrella to lend" at the location "entrance."

For example, suppose there is a request to "raise the temperature in the conference room by 2 degrees because we are having an employee information session." In this case, the direct request is to "raise the temperature by 2 degrees." However, there is a hidden need, such as "I want a comfortable place to hold the employee information session." The key to eliciting such needs is how to prepare a learning dataset. If such needs can be elicited, for example, the building owner will be able to understand that many companies that hold company information sessions occupy the building and therefore need a space for such sessions. This will lead to improvements in the building's services.

The search system is a system that searches for events that focus on the work performed by the building manager, while the needs extraction system is a system that extracts events that focus on the emotions of users. For this reason, unlike estimation model 31, estimation model 41 learns to effectively extract human emotions (when, where, who, and what is wanted to be done). Time and place are also linked to the user's purpose. For example, if a location called "corridor" is extracted, it is also possible to elicit a latent need such as "poor flow."

Figure 13 is a flowchart of the learning process (for extracting needs). The control unit 21 executes the learning process (for extracting needs).

As shown in FIG. 13, when the learning process (for needs extraction) is started, the control unit 21 selects learning data from the learning dataset 42 (for needs extraction) in S31 and proceeds to S32. In S32, the control unit 21 inputs the word data of the selected learning data (for history search) to the estimation model 31 and proceeds to S33. The words input are representative words included in the learning dataset 42.

In S33, the control unit 21 performs an estimation process using the estimation model 41 to output an estimation result, and proceeds to S34. The output estimation result is the need and the location or date and time when the need occurs. In S34, the control unit 21 updates the parameters of the estimation model 41 based on the error between the estimation result and the ground truth data corresponding to the learning data, and proceeds to S35.

In S35, 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 S35), the process proceeds to S36. If it is determined that learning has not been performed based on all of the learning data (NO in S35), the process returns to S31.

In S36, the control unit 21 stores the trained estimation model 41 as the trained model 41 and ends the learning process (for needs extraction).

[Needs extraction process in needs display system]
Next, a needs extraction process in the needs display system 1b will be described. Fig. 14 is a flowchart of the needs extraction process. The control unit 21 executes the needs extraction process. As a result of the needs extraction process, the display unit 127 displays the estimation result.

As shown in FIG. 14, when the needs extraction process is started, the control unit 21 uses the word dictionary 34 to extract word representative words from the history data in S41, and proceeds to process S42. In S42, the control unit 21 inputs the extracted word representative words as features into the trained model 41, and outputs the needs and the location or date and time when the needs occur as estimation results from the trained model 41, and proceeds to process S43. The processes of S41 and S42 are performed for all history data.

In S43, the control unit 21 tally up the inference results by need and proceeds to S43. In S44, the display unit 127 displays the inference results tallied by need and ends the needs extraction process. A specific example will be described below.

Figure 15 is a diagram explaining the flow of needs extraction in the needs display system 1b. The word dictionary used in needs extraction is the same as the word dictionary used in history search.

As shown in FIG. 15, an inference result 45b is generated corresponding to the history data 45a, "I asked if I could temporarily borrow an umbrella..."

The control unit 21 uses the word dictionary 34 to extract word representative words from the history data 45a. In this example, word representative words such as "umbrella" and "lend" are extracted from the history data 45a.

Next, the control unit 21 inputs the extracted words "umbrella," "lend," etc., into the trained model 41, and outputs as inferred results from the trained model 41 the need "It would be nice to have an umbrella to lend," and the location or date and time (location and date and time) when the need occurs "entrance."

In this example, the words "umbrella" and "lend" indicate a potential need for "it is convenient to have an umbrella available for loan." And because the entrance is a suitable place for lending umbrellas, the system is trained to output the entrance as the "location."

In this way, estimation results are output from all historical data, and the output estimation results are tallied by need. Figure 16 shows an example of the needs extraction results.

For example, suppose that the need "It would be nice to have an umbrella available for borrowing" at the "entrance" is counted 123 times from historical data. As shown in Figure 16, there are 123 instances of the need "It would be nice to have an umbrella available for borrowing" at the "entrance".

The display unit 127 displays the needs, the location or date and time when the needs occur, and the number of needs that have been tallied, in descending order of the number of needs that have been tallied.

In this example, the needs displayed in descending order are "It would be nice to have umbrellas available to borrow" for the "entrance" (1st place, 123 cases), "I don't like the sink being wet" for the "restroom" (2nd place, 112 cases), "I want to know how to enter the building on holidays and at night" for "outside regular weekday hours" (3rd place, 77 cases), and "1st floor emergency exit" for "The door is hard to close" (4th place, 33 cases).

The control unit 21 also tally up the estimation results by need for a specific period of the year. In this example, any of "January" to "December" can be selected as the specific period. When any of the links for "January" to "December" shown in the figure is clicked, tallying is performed for the clicked month.

For example, if historical data is recorded from 2018 to 2020, clicking "April" will result in only historical data recorded in April 2018, April 2019, and April 2019 being tabulated. The results tabulated for April only will be displayed on the screen. It is also possible to display the tabulated results for April 2018, April 2019, and April 2019 separately. Alternatively, a graph may be displayed to show the changes in needs for April 2018, April 2019, and April 2019.

This allows you to understand the needs that are likely to arise in April. Note that the data is not limited to being tabulated by month, but may be tabulated by any period (for example, by season).

In this embodiment, by configuring as described above, it is possible to preferably extract the needs of building users from historical data related to the management of building facilities. This makes it possible to extract latent needs of building users such as tenants that cannot be collected through interviews or questionnaires. In addition, since the historical data recorded each time is used as an information source, it is possible to extract needs that building users have forgotten, and it is possible to collect needs data that exists only on-site and does not reach the building owner. This makes it possible to improve building operations and facilities for building users that have not been noticed until now.

[Main components and effects]
The main configurations and effects of the above-described embodiments will be described below.

(1-1) The search system 1a searches for history data related to the management of building facilities. The search system 1a includes a control unit 21, a storage unit 25, and a display unit 127. The storage unit 25 stores an attribute dictionary 33, a word dictionary 34, and a search database 36. The attribute dictionary 33 is a dictionary in which attribute representative words representing attributes of the history data are associated with attribute related words having meanings related to the attribute representative words. The word dictionary 34 is a dictionary in which word representative words representing words used in the history data are associated with word related words having meanings related to the word representative words. The search database 36 is a database in which search data in which one or more attribute representative words are associated with history data and attribute contents related to the attribute representative words are registered. The display unit 127 displays the search results obtained by the control unit 21 searching the search database 36 based on the keywords entered by the user. The control unit 21 uses the attribute dictionary 33 to extract attribute representative words having meanings related to the keywords. The control unit 21 determines the search data associated with the extracted attribute representative word as the search target of the search database 36. The control unit 21 uses the word dictionary 34 to extract word representative words and word related words that have meanings related to the keyword. If the extracted word representative word or word related word is included in the search data determined as the search target, the control unit 21 determines the search data as the search result to be displayed by the display unit 127. With this configuration, it is possible to efficiently search history data that contains words that have meanings related to the input keyword. This makes it possible to search for personal sentences using words that other staff members normally use, allowing the experience of others to be utilized in one's own work.

(1-2) The display unit 127 displays the history data and attribute contents as search results. The search results include information that can identify word representative words or word related words contained in the search data. This configuration makes it easier to recognize word representative words or word related words contained in the search data.

(1-3) The control unit 21 inputs words contained in the history data to the trained model 31, and outputs one or more attribute representative words and attribute contents related to the attribute representative words from the trained model 31. The control unit 21 associates the output attribute representative words and attribute contents with the history data and registers them in the search database 36 as search data. With this configuration, data including the attribute representative words and attribute contents can be targeted for search, and searches can be performed with words of a broader concept, making it possible to efficiently search for history data that includes words with meanings related to the input keyword.

(1-4) The attribute content is composed of a combination of two word representative words. By configuring it in this way, words that more accurately and concisely indicate the content of the history data are targeted for search, making it possible to efficiently search for history data that contains words with meanings related to the input keyword.

(1-5) The attribute representative words include words that represent the content of an inquiry made by a building user about building facilities, words that represent the reason why the user inquired about the content, and words that represent the response of the building facilities manager to the user's inquiry. By configuring it in this way, it is possible to narrow down the historical data using a method optimized for managing building facilities.

(1-6) The trained model 31 is a model that has undergone machine learning processing using training data so that, when a word contained in history data is input, one or more attribute representative words and attribute contents related to the attribute representative words are output. By configuring it in this way, it is possible to efficiently search for history data that contains words with meanings related to the input keyword.

(1-7) The control method controls a search system 1a that searches for historical data related to the management of building facilities. The search system 1a includes a memory unit 25. The memory unit 25 stores an attribute dictionary 33 in which attribute representative words representing attributes of the historical data are associated with attribute-related words having meanings related to the attribute representative words, a word dictionary 34 in which word representative words representing words used in the historical data are associated with word-related words having meanings related to the word representative words, and a search database 36 in which search data in which one or more attribute representative words are associated with historical data and attribute contents related to the attribute representative words are registered. The control method includes the steps of: using the attribute dictionary 33 to extract attribute representative words having a meaning related to the keyword input by the user; determining search data associated with the extracted attribute representative words as a search target in the search database 36; using the word dictionary 34 to extract word representative words and word related words having a meaning related to the keyword; determining the search data as a search result if the extracted word representative word or word related word is included in the search data determined as the search target; and displaying the search result. This configuration makes it possible to efficiently search history data containing words having a meaning related to the input keyword. This makes it possible to search for personal sentences using words that other staff members normally use, allowing the experience of others to be utilized in one's own work.

(2-1) The needs display system 1b tally and displays the needs of building users. The needs display system 1b includes a control unit 21, a storage unit 25, and a display unit 127. The storage unit 25 stores a word dictionary 34 and a trained model 41. The word dictionary 34 is a dictionary in which historical data related to the management of building facilities, word representative words representing words used in the historical data, and word related words having meanings related to the word representative words are associated with each other. The trained model 41 is a model used to output an estimation result based on the historical data. The display unit 127 displays the estimation result. The trained model 41 is a model that has been subjected to machine learning processing using teacher data so that, when a word representative word is input as a feature, a need and the location or date and time at which the need occurred are output as an estimation result. The control unit 21 uses the word dictionary 34 to extract the word representative word from the historical data. The control unit 21 inputs the extracted word representative words as features to the trained model 41, and outputs the needs and the location or date and time when the needs occur as estimation results from the trained model 41. The control unit 21 tallies the estimation results for each need. The display unit 127 displays the estimation results tallied for each need.

With this configuration, the needs of building users can be ideally extracted from historical data related to the management of building facilities. This makes it possible to extract the latent needs of building users such as tenants that cannot be collected through interviews or questionnaires. In addition, because historical data recorded each time is used as an information source, it is possible to extract needs that building users have forgotten about, and it is possible to collect needs data that exists only on-site and does not reach the building owner. This makes it possible to improve building operations and facilities for building users that have not been noticed until now.

(2-2) The display unit 127 displays the needs, the location or date and time when the needs occur, and the number of needs that have been tallied in descending order of the number of needs that have been tallied. By configuring in this way, the needs can be grasped in an appropriate manner.

(2-3) The control unit 21 tally up the estimation results for each need for a specific period of the year. By configuring in this way, it is possible to grasp needs that occur only in a specific period and changes in needs from year to year for a specific period.

(2-4) The historical data is data that records the content of inquiries made by users regarding building facilities, the reason why the users inquired about the content, and the response of the building facilities manager to the user's inquiry. By configuring it in this way, it is possible to preferably extract needs by utilizing the historical data managed in the building facilities.

(2-5) The needs are based on events that users find unpleasant at the place or date and time when the needs occur, or on the requirements of building facilities to use the building comfortably. By configuring in this way, the needs of building users can be suitably extracted from historical data on the management of building facilities.

(2-6) The trained model 41 generation method is a method for applying to the needs display system 1b that tally and displays the needs of building users. The needs display system 1b includes a storage unit 25 that stores historical data related to the management of building facilities, a word dictionary 34 in which word representative words representing words used in the historical data and word related words having meanings related to the word representative words are associated. The trained model 41 generation method includes a step of selecting training data from a training data set, a step of inputting the word representative words included in the training data as features to an estimation model 41, thereby outputting the needs and the location or date and time when the needs occur as estimation results from the estimation model 41, and a step of updating the parameters of the estimation model 41 based on the error between the estimation result and the correct answer data corresponding to the training data. With this configuration, the needs of building users can be suitably extracted from historical data related to the management of building facilities.

(2-7) The control method controls a needs display system 1b that aggregates and displays the needs of building users. The needs display system 1b includes a storage unit 25 that stores historical data related to the management of building facilities, a word dictionary 34 in which word representative words representing words used in the historical data and word-related words having meanings related to the word representative words are associated with each other, and a trained model 41 used to output an estimation result based on the historical data. The control method includes a step of extracting word representative words from the historical data using the word dictionary 34, a step of inputting the extracted word representative words as features to the trained model 41 to output the needs and the location or date and time when the needs occur as estimation results from the trained model 41, a step of aggregating the estimation results for each need, and a step of displaying the estimation results aggregated for each need. The trained model 41 is a model that has been subjected to machine learning processing using teacher data so that when the word representative words are input as features, the trained model 41 outputs the needs and the location or date and time when the needs occur as estimation results. This configuration makes it possible to optimally extract the needs of building users from historical data related to the management of building facilities.

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 is intended to include all modifications within the scope and meaning equivalent to the claims.

1a Search system, 1b Needs extraction system, 10 Server device, 21, 121 Control unit, 22, 122 ROM, 23, 123 RAM, 24, 124 Communication interface, 25, 125 Storage unit, 31, 41 Trained model, 32, 42 Training data set, 33 Attribute dictionary, 34 Word dictionary, 35a, 45a History data, 35b, 45b Estimation result, 36 Search database, 37 Search screen, 38 History search result, 48 Needs extraction result, 100, 101 Terminal, 127 Input unit, 128 Display unit.

Claims (7)

A needs display system that collects and displays the needs of building users,
A control unit;
Historical data on building facility management and
a word dictionary in which a word representative word that represents a word used in the history data is associated with a word-related word having a meaning related to the word representative word;
a memory unit that stores a trained model used to output an estimation result based on the history data;
A display unit that displays the estimation result,
the trained model is a model that has been subjected to machine learning processing using training data so as to output, when the word representative word is input as a feature, the need and the location or date and time when the need occurred as the estimation result;
The control unit is
extracting the word representative words from the history data using the word dictionary;
The extracted word representative words are input as the features to the trained model, and the need and the location or date and time when the need occurs are output from the trained model as the estimation result;
aggregating the estimated results by each of the needs;
The display unit displays the estimation results tabulated for each need. The needs display system according to claim 1, wherein the display unit displays the needs, the location or date and time when the needs occur, and the number of the needs that have been collected in descending order of the number of the needs that have been collected. The needs display system according to claim 1 or claim 2, wherein the control unit aggregates the estimation results for each need for a specific period of the year. The needs display system according to any one of claims 1 to 3, wherein the history data is data that records the content of the user's inquiry about the building facility, the reason why the user inquired about the content, and the response of the manager of the building facility to the user's inquiry. The needs display system according to any one of claims 1 to 4, wherein the needs are based on events that the users find unpleasant or that the users require from the building facilities in order to use the building comfortably at the place or date and time when the needs occur. A trained model generation method for application to a needs display system that aggregates and displays the needs of building users, comprising:
The needs display system includes:
Historical data on building facility management and
a storage unit for storing a word dictionary in which a word representative word representing a word used in the history data is associated with a word related word having a meaning related to the word representative word;
The trained model generation method includes:
selecting training data from a training dataset;
a step of inputting the word representative words included in the learning data into an estimation model as features, and outputting the need and a place or date and time when the need occurs from the estimation model as an estimation result;
A method for generating a trained model, comprising: updating parameters of the estimation model based on an error between the estimation result and ground truth data corresponding to the learning data. A control method for controlling a needs display system that collects and displays the needs of building users, comprising:
The needs display system includes:
Historical data on building facility management and
a word dictionary in which a word representative word that represents a word used in the history data is associated with a word-related word having a meaning related to the word representative word;
a learned model used to output an estimation result based on the history data; and a storage unit that stores the learned model.
The control method includes:
extracting the word representative words from the history data using the word dictionary;
A step of inputting the extracted word representative words as features into the trained model, and outputting the need and the location or date and time when the need occurs from the trained model as the estimation result;
a step of aggregating the estimation results for each of the needs;
and displaying the estimation results tabulated for each of the needs;
A control method, wherein the trained model is a model that has been subjected to machine learning processing using training data so that, when the word representative word is input as the feature, the need and the location or date and time when the need occurred are output as the estimated result.

Images