Incident prediction program for welfare support facilities

Abstract

The system will allow users to identify their risk of incidents occurring based on electronically stored support record document data. [Solution] To predict incidents occurring among users of welfare support facilities, the present invention provides an incident prediction program for welfare support facilities characterized by the following functions: a support record acquisition means that acquires support record text data from a support record database where support records of users of welfare support facilities are stored; a risk value derivation means that extracts entries correlated with incident occurrence (incident correlation entries) from the support record text data acquired by the support record acquisition means, scores each extracted incident correlation entry to derive an incident risk value for each user; and a risk value transmission means that transmits the risk value derived by the risk value derivation means to a user terminal.

Description

Technical Field

[0001] The present invention relates to an incident prediction program for welfare support facilities that predicts the occurrence of incidents by users of welfare support facilities.

Background Art

[0002] Welfare support facilities are facilities for disabled people and the elderly who have difficulty living on their own. By entering a welfare support facility, users (residents) can live their daily lives while receiving support from facility staff. However, the number of users of welfare support facilities is on the rise, and combined with the recent shortage of workers, the shortage of facility staff is becoming more serious. When there is a shortage of facility staff, it becomes difficult to keep an eye on each individual user, and troubles such as near misses are likely to occur. When this type of trouble occurs, facility staff have to spend time on aftercare such as cleaning up and caring for the parties involved, and as a result, they are unable to provide the services that they should have been able to provide. In order to maintain the quality of services in welfare support facilities, it is necessary to reduce the burden on facility staff.

[0003] In view of such a situation, in welfare support facilities, a program for welfare support facilities that enables clerical work (such as filling in support records for users) performed by facility staff to be carried out on a computer is also being used (see, for example, Patent Document 1). This can reduce the labor of facility staff compared to the case of writing support records by hand. In addition, daily support records can be accumulated as electronic data. Therefore, when it becomes necessary to check the past support records of users, the target support records can be easily extracted. In addition, there is also the merit that information about users can be easily shared among staff.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

[0005] However, conventional welfare support facility programs, while reducing administrative work for facility staff, do not prevent problems caused by users from occurring in the first place, and their effect on reducing the burden on facility staff is limited. In this respect, if it is possible to predict in advance that "this user is likely to cause trouble in the near future," then appropriate support can be provided to that user, preventing the trouble from occurring in the first place. Therefore, by allocating the time previously spent on post-incident cleanup to core duties, it is possible to improve services even with limited personnel.

[0006] This invention was made to solve the above problems and provides an incident prediction program for welfare support facilities that not only enables the input of support records for users of welfare support facilities to be done on a computer, but also allows for the identification of the risk that a user may cause trouble (incident) in the near future from the electronically stored support records (support record document data). [Means for solving the problem]

[0007] The above issues are, In order to predict the occurrence of incidents involving users of welfare support facilities, Computers, A means for obtaining support record text data from a support record database containing support records of users of welfare support facilities, A risk value derivation means extracts entries that correlate with incident occurrence (hereinafter referred to as "incident correlation entries") from the support record document data acquired by the support record acquisition means, and derives an incident occurrence risk value for each user by scoring each extracted incident correlation entry. A risk value transmission means transmits the risk value derived by the risk value derivation means to the user terminal. Incident prediction program for welfare support facilities, characterized by its ability to function in this way. This is solved by providing [a solution].

[0008] By deriving and displaying the risk level of incident occurrence for each user in this way, staff at welfare support facilities can identify users who are highly likely to experience an incident in the near future. Therefore, facility staff providing support to those users can prevent incidents from occurring by providing appropriate support. Consequently, the time previously spent on incident aftermath can be allocated to core duties, allowing for improved service even with limited personnel.

[0009] In the incident prediction program for welfare support facilities of the present invention, the risk value derivation means is not particularly limited as long as it derives an incident risk value using support record document data. In this regard, the applicant found, after interviewing several welfare support facilities, that users who cause incidents have common tendencies in the preceding days (several days to the day before, etc.). Specifically, it was found that these users had previously exhibited impulsive behavior or negative psychological states. Therefore, if it is possible to extract entries that correlate with incident occurrence (incident correlation entries) from past support record document data, an incident risk value can be derived. However, performing this extraction work manually would be time-consuming for the person who fills out the support records (facility staff). Therefore, it is preferable that the risk value derivation means can automatically extract incident correlation entries. However, because facility staff have their own quirks in how they write support records, a method of extracting incident correlation entries using simple keyword searches is prone to omissions and mis-extractions, making it impossible to derive a highly accurate risk value. Therefore, it is preferable that the risk value derivation means derives the risk value using artificial intelligence (AI).

[0010] in particular, The means for deriving the risk value is, Using training data and a neural network that associates each event previously recognized as having a correlation with the occurrence of an incident with the likelihood of an incident occurring when each event occurs, incident correlation descriptions are extracted. It is preferable to make it so. In this way, by utilizing artificial intelligence (AI), it becomes possible to extract incident correlation records with high accuracy.

[0011] Furthermore, while the risk value derivation method may derive the risk value of incident occurrence based on support record document data for only one day (the previous day), it is preferable to derive the risk value of incident occurrence based on support record document data for multiple recent days. This is because signs of an incident may begin to appear not only the day before, but several days prior. However, the probability of an incident occurring is higher when signs of an incident appear one day prior (the previous day) than when signs of an incident appear several days prior (for example, three days prior). For this reason, when the risk value derivation method performs scoring, it is preferable to change the weight depending on the day on which those incident correlation records were made.

[0012] in particular, The risk value calculation means is We will extract incident correlation descriptions from the support record document data for the most recent several days, The scoring system will weight incident correlation descriptions extracted from recent support record document data. It is preferable to make it so. This makes it possible to derive the risk value of incident occurrence with greater accuracy.

[0013] By the way, the incident prediction program for welfare support facilities of the present invention is Computers, A support method proposal tool that suggests support methods tailored to a specific user based on support record document data for that user. It is preferable to also make it function as such. As a result, the facility staff of the welfare support facility can easily grasp more appropriate support methods for users (such as users with a high risk value of incidents occurring). Therefore, it becomes possible to more reliably prevent the occurrence of incidents.

[0014] The above-mentioned support method proposal means is not particularly limited, but using the incident correlation description for a specific user and the disability characteristics and personal characteristics of that user, automatically generate a prompt asking to propose a support method for that user, automatically input that prompt into an artificial intelligence chatbot, and display the content replied by the artificial intelligence chatbot as a support method. It is preferably such. Artificial intelligence chatbots in recent years, such as "ChatGPT" (registered trademark), are known to be able to obtain answers with very high accuracy. By using this type of artificial intelligence chatbot to obtain support methods, it becomes possible to make highly accurate proposals without incorporating complex processing into the incident prediction program for welfare support facilities.

Effects of the Invention

[0015] As described above, according to the present invention, not only can the work of inputting support records for users of welfare support facilities be performed on a computer, but also from the electronically stored support records (support record text data), it becomes possible to provide an incident prediction program for welfare support facilities that can grasp the risk that the user will have a trouble (incident) soon.

Brief Description of the Drawings

[0016] [Figure 1] It is a diagram showing an example of a management system for welfare support facilities. [Figure 2] It is a diagram showing an example of teacher data listing events of "sign (high)". [Figure 3]This figure shows an example of training data listing "low-level warning signs." [Figure 4] This figure shows an example of a risk value correspondence table for incident occurrence. [Figure 5] This diagram shows an example of the top screen (menu screen) of a management system for welfare support facilities. [Figure 6] This diagram shows an example of a support record input screen in a management system for welfare support facilities. [Figure 7] This figure shows an example of an incident risk analysis screen in a management system for welfare support facilities. [Figure 8] This diagram shows an example of a misidentification reporting screen in a management system for welfare support facilities. [Figure 9] This figure shows an example of a prompt confirmation screen in a management system for welfare support facilities. [Figure 10] This diagram shows an example of a support method confirmation screen in a management system for welfare support facilities. [Modes for carrying out the invention]

[0017] The incident prediction program for welfare support facilities and the system using the same (management system for welfare support facilities) of the present invention will be described in more detail with reference to the drawings. However, the configuration described below is merely a preferred embodiment. For this reason, the technical scope of the present invention is not limited to the configuration described below. The incident prediction program for welfare support facilities of the present invention can be modified as appropriate without impairing the spirit of the invention.

[0018] 1. Overview of the Management System for Welfare Support Facilities Figure 1 shows an example of a system (management system for welfare facilities) using the incident prediction program for welfare support facilities of the present invention. As shown in Figure 1, this management system for welfare support facilities uses a support record management server and an AI server to predict the occurrence of incidents by users of welfare support facilities.

[0019] 1.1 Support Record Management Server The support record management server is used to manage the support records of users of welfare support facilities. The support record management server is installed on the cloud. User terminals are connected to this support record management server via the internet. The user terminals are installed in welfare support facilities that use the welfare support facility management system and are operated by the facility staff. Facility staff use these user terminals to fill out daily support records for the users they support (users of the welfare support facility). The completed support records are sent as electronic data (support record document data) to the database (support record database) on the support record management server and are stored in that database.

[0020] This reduces the workload for facility staff compared to manually recording support information. Furthermore, support records can be stored as electronic data. Therefore, when it becomes necessary to review a user's past support records, the desired records can be easily retrieved. In addition, it facilitates the sharing of user information among facility staff.

[0021] By the way, in the example shown in Figure 1, a personal computer is used as the user terminal. However, the user terminal does not have to be a personal computer; a tablet or smartphone can also be used as a user terminal. The number of user terminals does not have to be one; multiple terminals can be used. It is also possible to provide each facility staff member with a user terminal.

[0022] 1.2 AI Server (Means for acquiring support records, means for deriving risk values, and means for transmitting risk values) The AI ​​server, like the support management server, is installed on the cloud. The incident prediction program for welfare support facilities of the present invention (hereinafter referred to as the "incident prediction program") is implemented on this AI server.

[0023] This incident prediction program enables the AI ​​server to function as a means of acquiring support record document data from the support record database (support record acquisition means), a means of deriving an incident risk value for each user of the welfare support facility (risk value derivation means), and a means of transmitting the risk value derived by the risk value derivation means to the user terminal (risk value transmission means). The risk value derived by the risk value derivation means is stored in the AI ​​server's risk value database for each user. The risk value stored in this risk value database is transmitted to the user terminal.

[0024] Let me explain the method for deriving the risk value in more detail. The means for deriving the risk value is, [1] A step to extract entries that correlate with the occurrence of an incident (incident correlation entries extraction step) from the support record document data for the past three days (the most recent three days) obtained from the support record database, [2] A step to derive an incident risk value for each user by scoring the incident correlation descriptions extracted in the incident correlation description extraction step (risk value derivation step) The risk value is derived through these two steps.

[0025] In the incident correlation description extraction step described in [1] above, descriptions such as "grabbed another person," "broke something," or "sounded distressed" are extracted as incident correlation descriptions. However, because facility staff have their own unique ways of writing support records, extracting incident correlation descriptions using a simple keyword search may result in omissions or mis-extractions. For this reason, in this embodiment, artificial intelligence (AI) is used to extract incident correlation descriptions.

[0026] In other words, the system uses a neural network trained on training data to extract incident correlation descriptions. Figures 2 and 3 show examples of training data. Figure 2 is training data listing events with a high correlation to the occurrence of an incident ("high warning" events), while Figure 3 is training data listing events that show a correlation to the occurrence of an incident, but not to the same degree as the "high warning" events in Figure 2 ("low warning" events).

[0027] The training data for "High Warning Signs" shown in Figure 2 lists numerous incidents where impulsivity or negative emotions could not be controlled, resulting in impacts on those around the individual, such as "violence," "property damage," and "lashing out at objects." On the other hand, the training data for "Low Warning Signs" shown in Figure 3 lists numerous incidents that are considered to be highly impulsive, or influenced by negative psychological states, among "subjective expressions" and "objective expressions." By training artificial intelligence (AI) using such training data, it becomes possible to extract incident correlation descriptions corresponding to "High Warning Signs" or "Low Warning Signs" with high accuracy, even if there is variability in the support record document data due to the habits of the person who wrote it (facility staff). If an incident does not fall under either "High Warning Signs" or "Low Warning Signs" (i.e., no incident correlation description is extracted), it is judged as "Normal." It is preferable to use a natural language processing model that can accurately classify documents for extracting incident correlation descriptions. For example, the natural language processing model "BERT," developed by researchers at Google, can understand the context of support record text data, thus improving the accuracy of extracting incident correlation descriptions.

[0028] The incident correlation records extracted in this way are aggregated for each user over a three-day period to derive the risk value for incident occurrence. Figure 4 shows an example of an incident occurrence risk value correspondence table. As shown in Figure 4, the risk value for incident occurrence on the current day is derived from a total of three days, from three days prior to one day prior. In the risk value correspondence table, the unit of the "numerical value" for "risk value on the current day" is "%", but this "%" is used for convenience to make the degree of risk intuitively easy to understand (if it is close to "100%", the risk of incident occurrence is quite high, and if it is close to "0%", the risk of incident occurrence is quite low), and this numerical value does not actually have the meaning of a percentage.

[0029] Incidentally, this risk value correspondence table (Figure 4) was created following steps 1 to 5 below. [Step 1] In the risk value correspondence table, enter either "High Warning Sign," "Low Warning Sign," or "Normal" in the "1 day ago," "2 days ago," and "3 days ago" columns. If you record one of the three levels ("High Warning Sign," "Low Warning Sign," or "Normal") for each of the three days, there will be a total of 27 possible patterns (3 to the power of 3). [Step 2] Assign points to each day. Specifically, 10 points are allocated to "high" signs, 5 points to "low" signs, and 1 point to "average" signs. [Step 3] Apply weighting by day. In other words, the points awarded for recent days (closer to the current day) will be higher than the points awarded for earlier days (further away from the current day). Specifically, the points awarded for one day ago will be multiplied by 5, the points awarded for two days ago will be multiplied by 2.5, while the points awarded for three days ago will remain at 1. [Step 4] Derive the risk value (numerical value) for each pattern on that day. Specifically, the sum of the points for the three days derived in step 3 above is calculated, and this value (sum) is taken as the "risk value (numerical value) for that day" for that pattern. This assigns a risk value of 9 to 85% to each of the 27 patterns shown in the risk value correspondence table. [Step 5] Determine the risk level (high, medium, low) for each pattern on that day. In other words, the risk values ​​(numerical values) derived in step 4 above are classified into three stages: "high," "medium," and "low." Specifically, patterns where the risk value (numerical value) for the day is 60% or higher are classified as "high," patterns where the risk value (numerical value) for the day is between 31% and 59% are classified as "medium," and patterns where the risk value (numerical value) for the day is 30% or lower are classified as "low."

[0030] When determining the risk of an incident occurring for a particular user, the system obtains three days' worth of support record document data for that user, extracts incident correlation descriptions from that support record document data (if multiple incident correlation descriptions are extracted from the support record document data for the same day, the incident correlation description with the highest score (highest risk) is adopted), and determines which pattern in the risk value correspondence table (Figure 4) the user falls into, and treats the "risk value for the day" corresponding to that pattern as the user's "risk value for the day". The above risk value derivation method derives the user's "risk value for the day" by performing this scoring process.

[0031] The risk value (numerical value and high / medium / low) for incident occurrence is derived for each user of the welfare support facility. The derived risk value for each user is transmitted to the user's terminal via the internet, etc., and can be checked on the user's terminal. This allows welfare support facility staff to know in advance that a particular user is highly likely to experience an incident in the near future. As a result, facility staff who support that user can provide support tailored to that user, thereby preventing incidents from occurring. Consequently, the time previously spent on post-incident handling can be allocated to core duties, enabling service improvement even in welfare support facilities with limited personnel.

[0032] By the way, in the example shown in Figure 1, the cloud where the AI ​​server is installed is shown separately from the cloud where the support record management server is installed. However, there is no particular need for these clouds to be separate; they can be combined into a single common cloud. In fact, there is no particular need for the AI ​​server to be separate from the support record management server; they can be combined into a single common server terminal.

[0033] Furthermore, in the example described above, the AI ​​server calculates the risk value, stores it in a risk value database, and sends it to the user terminal, but this is not the only way to do so. For example, the AI ​​server could perform the incident correlation description extraction step (up to the predictive indicator judgment), send the results to the support record management server and store them in the support record database, and the support record management server could then calculate the risk value from the judgment results (predictive indicator judgments made by the AI ​​server) stored in the support record database.

[0034] 2. How to use the management system for welfare support facilities Next, we will explain how to use the management system for welfare support facilities.

[0035] 2.1 Top screen (menu screen) Figure 5 shows an example of the top screen S1 (menu screen) of the welfare support facility management system. When a user terminal (Figure 1) is started and the welfare support facility management system is accessed from that terminal, a login screen (not shown) is displayed on the user terminal screen. After entering the ID and password on this screen, the top screen S1 (menu screen) shown in Figure 5 is displayed. This top screen S1 is provided with a toolbar 10, a menu selection section 20, a notification display section 30, and a high-risk user display section 40 from the top.

[0036] Menu bar 10 displays, from left to right, the system name, the name of the business using this system, the name of the currently displayed screen, the login date, the name of the logged-in user (the name of the facility staff member), and a "Logout" button.

[0037] Furthermore, the menu selection section 20 has multiple buttons arranged in two rows. In this embodiment, the upper row, from left to right, has the following buttons: "Support Record Input," "Individual Support Plan Input," "User Registration," "Notifications, etc. Input," "Checklist," "Contact Book," and "Incident Risk Analysis." The lower row, from left to right, has the following buttons: "Print," "Inquiry," "Settings," and "Guardian Registration."

[0038] By pressing these buttons, the screen displayed on the user terminal (Figure 1) switches to the screen corresponding to that button. For example, pressing the "Support Record Input" button switches to the screen for inputting support records (Support Record Input Screen S2 shown in Figure 6 below), pressing the "User Registration" button switches to the screen for registering users (User Registration Screen (not shown)), and pressing the "Incident Risk Analysis" button switches to the screen for analyzing incident risks (Incident Risk Analysis Screen S3 shown in Figure 7 below).

[0039] Furthermore, the notification display unit 30 displays information (notifications) that facility staff using this system should be informed of in chronological order. The content displayed on this notification display unit 30 can be updated on the screen that appears when the "Notification Input" button is pressed on the menu selection unit 20 (Notification Input Screen (not shown)).

[0040] Furthermore, the high-risk user display section 40 displays users whose risk value (incident risk value) for that day is classified as "high," in descending order of risk value. The numbers in parentheses in the high-risk user display section 40 (such as "+9" or "+20") indicate the increase or decrease in that user's risk value from the previous day. By displaying users with high risk values ​​on the top screen S1 in this way, all staff members of the welfare support facility can recognize users who are highly likely to cause an incident on that day. As a result, not only the staff member in charge of that user, but also other staff members will pay close attention to the user's words and actions. Consequently, it becomes easier to prevent incidents caused by that user.

[0041] 2.2 Support Record Input Screen Figure 6 shows an example of a support record input screen S2 in a management system for welfare support facilities. This support record input screen S2 is equipped with a toolbar 10, a search condition specification unit 50, a search result display unit 60, a user information display unit 70, and a support record input unit 80.

[0042] The toolbar 10 on the support record input screen S2 is basically the same as the toolbar 10 on the top screen S1, but the rightmost button has been replaced with a "Menu" button instead of a "Logout" button. Pressing this "Menu" button switches the screen back to the top screen S1 (Figure 5). This is also the case for other screens that appear when other buttons are pressed in the menu selection section 20 of the top screen S1 (such as the incident risk analysis screen S3 shown in Figure 7 below).

[0043] The search condition specification unit 50 is a place to specify search conditions in order to search for the target to which support records are entered. In this embodiment, the search condition specification unit 50 is provided with a place to specify the date (date specification unit), a place to specify the business operator (business operator specification unit), a place to specify the service type (service type specification unit), a place to specify the group name (group specification unit), a place to perform partial input (partial input unit), and a place to specify the Japanese syllabary (Japanese syllabary specification unit).

[0044] Support records are entered daily, and by specifying a specific date in the date specification section of the search condition specification section 50, the support record for that day becomes available for input. By default, the current date is specified in the date specification section. Also, since one company may operate multiple welfare support facilities, the business operator can be selected in the business operator specification section of the search condition specification section 50. Furthermore, since users of welfare support facilities receive different services depending on the degree of their disability, the service type can be selected in the service type specification section of the search condition specification section 50. Moreover, even users of the same welfare support facility may belong to different groups, so the group name can be selected in the group specification section of the search condition specification section 50. In addition, the partial input section of the search condition specification section 50 is where you can enter part of the name of a welfare support facility user and add it to the search conditions. Furthermore, the Japanese alphabet specification section of the search condition specification section 50 is where you can search for the names of welfare support facility users alphabetically. By specifying these items, it is possible to quickly find the support record input screen S3 for the target user.

[0045] The search results display unit 60 displays a list of users that meet the conditions specified in the search criteria specification unit 50. A check mark is displayed to the right of the name of a user for whom support records for the relevant day have already been entered. When a specific user is selected from the users displayed in the search results display unit 60, information about that user is displayed in the user information display unit 70, and the support record input unit 80 becomes available to input (new input and editing) support records for that user for the relevant date.

[0046] In this embodiment, the user information display unit 70 displays the name of the selected user, the name of the facility where the user resides, the type of service the user is receiving, the name of the group to which the user belongs, and the user's goals. These goals were previously entered by facility staff or others when registering the user on the user registration screen (not shown) described above.

[0047] Furthermore, the support record input section 80 displays input fields for "Support Name" (Support Name Input Section), "Time" (Time Input Section), and "User Status" (User Status Input Section) in chronological order. When the support record input section 80 is first displayed, the support name input section, time input section, and user status input section are blank. Facility staff can input text into these blanks to enter the support record. If a pre-defined phrase is to be entered into the user status input section, pressing the "Pre-defined Phrases" button at the bottom displays a list of pre-defined phrases (not shown in the diagram), and selecting one of the pre-defined phrases copies it to the specified blank. If the pre-prepared blanks are insufficient, pressing the "Add" button at the bottom allows for an additional row of input fields in the support name input section, time input section, and user status input section. After completing the necessary inputs, pressing the "Register" button updates the support record for that user for the corresponding date. Updated support records are sent as support record text data to the support record management server (Figure 1) and reflected in the support record database (Figure 1). By performing this process daily for all users, the support record database accumulates daily support record text data for a large number of users.

[0048] 2.3 Incident Risk Analysis Screen Figure 7 shows an example of the incident risk analysis screen S3 in a management system for welfare support facilities. This incident risk analysis screen S3 includes a toolbar 10, a search condition specification unit 50, a search result display unit 60, a name display unit 90, a risk value display unit for the day 100, a warning sign trend display unit 110, a support record display unit for the previous day 120, a detailed personal information display unit 130, and a suggestion request unit 140. Of these, the toolbar 10 is substantially the same as the toolbar 10 in the support record input screen S2 (Figure 6), so its explanation is omitted.

[0049] The search condition specification section 50 on the incident risk analysis screen S3 (Figure 7) is provided for the same purpose as the search condition specification section 50 on the support record input screen S2 (Figure 6). However, the search condition specification section 50 on the incident risk analysis screen S3 (Figure 7) is simpler than the search condition specification section 50 on the support record input screen S2 (Figure 6), as it only has sections for specifying the date (date specification section), specifying the service provider (service provider specification section), specifying the service type (service type specification section), and specifying the group name (group specification section).

[0050] Furthermore, the search results display unit 60 in the incident risk analysis screen S3 (Figure 7) is provided for the same purpose as the search results display unit 60 in the support record input screen S2 (Figure 6). However, the search results display unit 60 in the incident risk analysis screen S3 (Figure 7) differs from the search results display unit 60 in the support record input screen S2 (Figure 6) in that, to the right of each user's name, the user's risk value (numerical value) for that day and the increase or decrease in the risk value (numerical value) from the previous day are displayed. When a specific user is selected from the users displayed in this search results display unit 60, the user's name is displayed in the name display unit 90, and the displays in the current day risk value display unit 100, the warning sign trend display unit 110, the previous day's support record display unit 120, and the user's detailed information display unit 130 are switched to those for that user.

[0051] The daily risk value display unit 100 displays the date, the risk value (numerical value) for that day, and the increase or decrease in the risk value (numerical value) from the previous day. The risk value (numerical value) displayed on the daily risk value display unit 100 is derived by the risk value derivation means described above in accordance with the risk value correspondence table in Figure 4. If the risk value (numerical value) corresponds to "high" (60 or higher), it is displayed in red; if the risk value (numerical value) corresponds to "medium" (31 to 59 or higher), it is displayed in yellow; and if the risk value (numerical value) corresponds to "low" (30 or lower), it is displayed in blue. This makes it easy for facility staff to grasp at a glance the level of incident risk for each user.

[0052] The indicator trend display unit 110 displays the user's indicators (displayed as three values: high, low, and normal, as shown above) for the past week. By displaying the trend of indicators in this way, it is possible to understand whether the user's physical and mental condition is improving or worsening. However, the indicators are not displayed numerically, but in three stages: high, low, and normal. If the indicator is high, it is displayed with a red light; if it is low, it is displayed with a yellow light; and if it is normal, it is displayed with a blue light (the light is not displayed on days when support records have not been entered (the current day)). By representing indicators using a light in this way, it becomes easier to intuitively grasp the indicators for that day.

[0053] The previous day's support record display unit 120 displays items from the previous day's support records in which incident correlation descriptions have been extracted. These support records were entered using the support record input screen S2 (Figure 6) described above and stored in the support record database (Figure 1). By displaying the previous day's support records on the incident risk analysis screen S3 (Figure 7) in this way, it is possible to understand the user's situation on the previous day. To the right of each item, the signs of incident correlation descriptions extracted from the description of that item are displayed as traffic lights. However, the possibility that these signs may be misjudged cannot be ruled out. For this reason, a "Report Misjudgment" button is provided next to the traffic lights, allowing users to report misjudgments.

[0054] Figure 8 shows an example of the misjudgment report screen S4 that appears when the "Report Misjudgment" button is pressed. In this embodiment, the misjudgment report screen S4 (Figure 8) opens in a separate window from the incident risk analysis screen S3 (Figure 7) described above. As shown in Figure 8, the misjudgment report screen S4 is provided with a section for copying the text reporting the misjudgment (report target display section 150), a section for correcting the judgment (judgment correction section 160), and a section for entering optional comments (optional comment input section 170). In the judgment correction section 160, the judgment can be corrected using radio buttons. After correcting the judgment in the judgment correction section 160 and pressing the "Send Report" button in the lower right corner of the screen, the judgment of the corresponding item is updated to the correct one, and a report is also sent to the system administrator. The misjudgment report screen S4 can be closed by pressing the "Close" button in the lower left corner of the screen.

[0055] Let's return to the Incident Risk Analysis screen S3 (Figure 7). In the Personal Information Display Section 130 of the Incident Risk Analysis screen S3, it is possible to view and edit detailed information such as the user's (individual's) disability characteristics, personal characteristics, and case studies. Clicking the "+" mark in the heading of each item will display detailed information below that item, and the "+" mark will change to a "-" mark. Clicking the "-" mark will hide the displayed detailed information. The detailed information displayed in the Personal Information Display Section 130 is information that facility staff or others have entered in advance when registering the user, but it is also possible to modify (edit) it from this Personal Information Display Section 130 by clicking the "Edit" button.

[0056] The suggestion request unit 140 is equipped with a "Receive AI's suggested support methods (standard version)" button (hereinafter referred to as the "standard suggestion button"), a "Receive AI's suggested support methods (simplified version)" button (hereinafter referred to as the "simplified suggestion button"), and a "prompt" button. When the standard suggestion button or the simplified suggestion button is pressed, the AI ​​will suggest a support method tailored to the user. In other words, the standard suggestion button and the simplified suggestion button are designed to make the AI ​​server (Figure 1) function as a means (support method suggestion means) for suggesting support methods tailored to the user. When suggesting a support method, the AI ​​server refers to the user's support record document data, etc.

[0057] Specifically, the support methods are proposed according to steps 1-3 below. [Step 1] When the standard suggestion button or the simplified suggestion button is pressed, the AI ​​server (Figure 1) automatically generates a prompt stating, "Please suggest a support method for that user." The prompt may include incident correlation information extracted from the user's support record text data from the previous day, as well as detailed information about the user. [Step 2] The prompt automatically generated in step 1 is automatically entered into the artificial intelligence chatbot. In this embodiment, the "ChatGPT" (registered trademark) API is used. [Step 3] The AI ​​chatbot's responses are displayed as support methods tailored to the user.

[0058] Figure 9 shows an example of a prompt that is automatically generated in step 1 above. The screen that displays this prompt (prompt display screen S5) opens in a separate window when the "Prompt" button in the proposal request section 140 on the incident risk analysis screen S3 (Figure 7) is clicked.

[0059] As can be seen in Figure 9, the prompt includes information about the user (disability characteristics and personal characteristics from the detailed personal information mentioned above) in bullet points. This allows for suggestions from multiple perspectives. In addition, support record data from the previous day that showed an incident correlation are extracted and included in the report section. This allows for suggestions that are more appropriate for that user.

[0060] Furthermore, the initial command assigns the AI ​​the role of a staff member at a welfare support facility. This allows for suggestions from the perspective of a facility staff member. Moreover, if the suggestions are too general, it will be difficult to translate them into actionable steps in the field. Therefore, the initial command specifies that the AI ​​should respond "with concrete examples." Additionally, since suggestions that include perspectives that facility staff members often overlook are desirable, the initial command specifies that the AI ​​should make "creative and innovative" suggestions.

[0061] When you press the Standard Proposal button, the prompt shown in Figure 9 is automatically generated. When you press the Simplified Proposal button, a simplified version of the prompt shown in Figure 9 is automatically generated. You can switch prompts as needed; for example, press the Standard Proposal button when you want a thorough proposal, and the Simplified Proposal button when you want a summarized proposal.

[0062] Figure 10 shows an example of the screen (support method confirmation screen S6) displayed when an AI chatbot responds in step 3 above. As can be seen in Figure 10, quite specific and useful suggestions are provided. Implementing a function to return such useful suggestions independently would require writing a complex program and incurring high development costs. However, by utilizing existing AI chatbot APIs such as the "ChatGPT" (registered trademark) API, it becomes possible to obtain highly accurate suggestions while keeping development costs down.

[0063] The suggestions received can be used as a reference when providing support to users at high risk of incidents. This will allow for more effective prevention of incidents and free up time previously spent on incident aftermath to be allocated to core duties. As a result, welfare support facilities can improve services even with limited personnel.

[0064] By the way, it can sometimes take some time for all of the responses from the AI ​​chatbot to be displayed. In this embodiment, a robot-like icon is displayed on the support method confirmation screen S6 (Figure 10), and while receiving responses from the AI ​​chatbot, an animation process is performed that makes it appear as if the robot is speaking. The rendering speed of the string of characters is adjusted so that the string of characters that will be the response is displayed one character at a time in accordance with the movement of the robot's mouth. This makes the user feel as if the robot is talking to them, which helps to alleviate the stress of waiting for a response. [Explanation of Symbols]

[0065] 10 Toolbar 20 Menu Selection Section 30 Notice display section 40. High-risk user display section 50 Search Criteria Section 60 Search Results Display Section 70 User information display section 80 Support Record Input Unit 90 Name Display Section 100 Daily Risk Value Display Section 110 Predictive Trend Display Section 120 Previous Day Support Record Display Section 130 Personal details information display section 140 Proposal request section 150 Reportable Display Section 160 Judgment correction section 170 Optional comment input section S1 Top Screen (Menu Screen) S2 Support Record Input Screen S3 Incident Risk Analysis Screen S4 false judgment report screen S5 prompt confirmation screen S6 support method confirmation screen

Claims

1. In order to predict the occurrence of incidents involving users of welfare support facilities, Computers, A means for obtaining support record text data from a support record database containing support records of users of welfare support facilities, A risk value derivation means extracts entries that correlate with incident occurrence (hereinafter referred to as "incident correlation entries") from the support record document data acquired by the support record acquisition means, and derives an incident occurrence risk value for each user by scoring each extracted incident correlation entry. A risk value transmission means transmits the risk value derived by the risk value derivation means to the user terminal. An incident prediction program for welfare support facilities, characterized by its ability to function in this manner.

2. The means for deriving the risk value is, Using training data and a neural network that associates each event previously recognized as having a correlation with the occurrence of an incident with the likelihood of an incident occurring when each event occurs, incident correlation descriptions are extracted. The incident prediction program for welfare support facilities according to claim 1.

3. The risk value calculation means is We will extract incident correlation descriptions from the support record document data for the most recent several days, The scoring system will weight incident correlation descriptions extracted from recent support record document data. The incident prediction program for welfare support facilities according to claim 2.

4. Computers, A support method proposal tool that suggests support methods tailored to a specific user based on support record document data for that user. The incident prediction program for welfare support facilities according to claim 1, which also functions as such.

5. The means of proposing support methods are The system automatically generates a prompt requesting suggestions for support methods for a specific user, using incident correlation descriptions, the user's disability characteristics, and personal characteristics. The prompt is automatically entered into the AI ​​chatbot, The responses received from the AI ​​chatbot will be displayed as support methods. The incident prediction program for welfare support facilities described in claim 4.

Images