Answer generation apparatus and method

Abstract

To obtain an answer suitable for a question (inquiry) even when the inquiry is personal.SOLUTION: A recording table is prepared which stores a plurality of records in each of which recording contents and a recording name indicating the recording contents in a simplified manner are recorded in association with each other. A question sentence is received, and a record name matching the received question is selected from among a plurality of record names provided in each of a plurality of records stored in the recording table (step 51). The record content associated with the selected record name is read from the record table (step 38), and the record content is summarized (step 52). The summary of the recorded contents is displayed as an answer to the question (step 26).SELECTED DRAWING: Figure 8

Description

[Technical Field]

[0001] The present invention relates to an answer generation device and method. [Background technology]

[0002] Patent Document 1 describes a chat application that opens a channel (group chat) for discussing the target medical treatment. Multiple members can be registered in a channel, and the registered members create chat messages. Message candidates are identified and displayed based on each member's role information, progress information on the target medical treatment, and the message history of the target channel. When a displayed message candidate is selected, that message is displayed in the message display field. Using the message history can reduce the effort required to input character strings. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2023-157873

[0004] There are also known programs (so-called chatbots) that respond to user questions. However, chatbots are mainly general-purpose automatic responses to general-purpose inquiries. Chatbots are generally not suitable when the content of an inquiry is highly personal. Summary of the Invention [Problem to be solved by the invention]

[0005] The object of the present invention is to enable a person to obtain an appropriate answer to a question (inquiry), even if the question (inquiry) is personal.

[0006] Another object of the present invention is to provide an answer that can be used for handover between staff members, etc.

[0007] The answer generating device according to the first aspect of the invention comprises a record table storing a plurality of records, each of which has detailed record content associated with a record name that abbreviates the detailed record content, a record name selection means for selecting a record name that matches a received question from among a plurality of record names provided for each of the plurality of records stored in the record table, a detailed record content reading means for reading from the record table the detailed record content associated with the record name selected by the record name selection means, a summarization means for summarizing the detailed record content read by the detailed record content reading means, and an output means for outputting the summary of the detailed record content summarized by the summarization means as an answer to the question.

[0008] The record table includes a plurality of records, each of which stores detailed record content and a record name that abbreviates the detailed record content in association with the record content.

[0009] In one embodiment, the detailed record contents are various care (support) contents that are provided or have been provided to users (typically elderly people or people with disabilities) of a care business (nursing facility). At the care business, various records are kept, such as the user's vital data (body temperature, blood pressure, transcutaneous arterial oxygen saturation, etc.), dietary details (e.g., food intake for breakfast, lunch, and dinner), water intake, oral care, height, weight, exercise, visitation records, and recreational details. Each of these records is a detailed record content.

[0010] When the types of detailed record content are diverse, they are generally divided into fewer categories than the types of detailed record content, based on the nature of the detailed record content, ease of recording, etc., and records are kept for each category. For example, a user's vital data (body temperature, blood pressure, transcutaneous arterial oxygen saturation, etc.) are classified as "health management." When health management is performed on a user at a nursing facility, body temperature, blood pressure, transcutaneous arterial oxygen saturation, etc. are measured and recorded together. "Health management" is an example of a record name that abbreviates the above detailed record content. Taking a nursing care facility (nursing care facility) as an example, in addition to "health management," record names such as "hydration," "visitation," "dinner," "recreation," "daytime activities / conditions," "breakfast," "lunch," "nighttime care," and "morning care" are used. The name of the ingested fluid (e.g., "tea") and the amount of fluid (e.g., "200cc") are associated with the record name "hydration" and recorded as detailed record content. If the record title is "Visit," the details of the record will be the visitor (e.g., "Taro"), their relationship (e.g., "eldest son"), and the content (e.g., "They seemed to be having a fun conversation").

[0011] According to the first invention, a record name that matches the question is first selected from among the multiple record names contained in each of the multiple records stored in the record table, and then a summary is created using the detailed record content associated with the selected record name. This summary is then treated as the answer to the question. Because a broad framework, the "record name," is first selected (determined) based on the question content, it becomes easier to obtain an answer that does not deviate significantly from the question. And because the summary that is finally output as the answer is based on the detailed record content associated with the selected record name, it becomes easier to obtain an answer that includes detailed information. Even if the inquiry is personal, an answer that suits the inquiry can be obtained.

[0012] In one embodiment, the record name selection means includes a record name list including a plurality of record names included in a plurality of records stored in the record table and the question, and includes a first prompt generation means for generating a prompt requesting the user to select a record name from the record name list that matches the content of the question, a first sending means for sending the prompt generated by the first prompt generation means to a generation AI device, and a first receiving means for receiving the record name selected by the generation AI device operating in accordance with the prompt generated by the first prompt generation means.Since a record name that matches the question is selected from a plurality of record names included in the record name list, it is possible to prevent the generation AI device from generating a record name that is not in the record name list, i.e., not in a record.

[0013] In another embodiment, the summarizing means includes a second prompt generating means for generating a prompt including the detailed record content read by the detailed record content reading means and requesting that the detailed record content be summarized, a second transmitting means for transmitting the prompt generated by the second prompt generating means to a generation AI device, and a second receiving means for receiving a summary generated by the generation AI device operating in accordance with the prompt generated by the second prompt generating means.Even if the detailed record content merely indicates a numerical value (such as body temperature or percutaneous arterial oxygen saturation), it is possible to obtain an answer in natural language that clearly indicates the meaning of the numerical value (for example, that the body temperature is higher than normal).

[0014] Preferably, a plurality of the record tables are provided, each of the record tables having a different identification code, and among the plurality of record tables, a record table having a given identification code is used for the record name selection process by the record name selection means and the detailed record content reading process by the detailed record content reading means. For example, only the record table of a specific user can be processed.

[0015] In another embodiment, a plurality of the record tables are provided, each of the record tables containing data representing a different date, and among the plurality of record tables, a record table containing date data for a given date is used for the record name selection process by the record name selection means and the detailed record content reading process by the detailed record content reading means. For example, only the record table for the current day (the day the question was received) can be processed.

[0016] In one embodiment, the answer generation device includes a personal information determination means for determining whether the detailed record contents read by the detailed record contents reading means contain personal information, and a personal information concealment means for concealing the determined personal information when the personal information determination means determines that the detailed record contents contain personal information. This can prevent personal information from being transmitted to an external party. For example, if a person's name is included in the detailed record contents, the name can be concealed (replaced with a symbol) to prevent the name from being transmitted to an external party.

[0017] The first invention also provides an answer generation method. In the answer generation method according to the first invention, a record name selection means uses a record table storing a plurality of record records, each of which stores detailed record contents associated with a record name that abbreviates the detailed record contents, to select a record name that matches the received question from among the plurality of record names included in each of the plurality of record records stored in the record table, a detailed record content reading means reads the detailed record contents associated with the selected record name from the record table, a summarizing means summarizes the read detailed record contents, and an output means outputs the summary of the detailed record contents as an answer to the question on the display device. An example of the output means is a display device, in which case the answer to the question is displayed on the display device in the form of characters (text).

[0018] The answer generation device according to the second invention comprises a filtering means for extracting recorded content recorded in a record that meets predetermined filter conditions from a record table in which a plurality of records, each of which records recorded content, is stored, a summarizing means for summarizing the recorded content extracted by the filtering means, and an output means for outputting a summary of the recorded content summarized by the summarizing means.

[0019] According to the second aspect of the present invention, a summary is created using the record contents recorded in records that meet a predetermined filter condition from among multiple records stored in a record table. Even if a large number of records are stored in the record table, a summary is created using the records that meet the filter condition, so it is possible to obtain a summary of record contents that require attention, for example.

[0020] If the recorded content includes the user's vital data (body temperature, blood pressure, percutaneous arterial oxygen saturation, etc.), dietary details (e.g., food intake for breakfast, lunch, and dinner), fluid intake, oral care, height, weight, exercise, visitation records, and recreational details, then body temperature, for example, can be used as a filter condition. If a body temperature of 37 degrees or higher is set as a filter condition, a summary will be created using recorded content with a body temperature of 37 degrees or higher. For example, a summary will be created for users who have a measured body temperature of 37 degrees or higher out of multiple users, so users who require attention will be clearly indicated in the summary. A summary suitable for handing over information can be obtained.

[0021] In one embodiment, the summarizing means includes prompt generating means for generating a prompt including the recorded content and a filter condition, requesting that a summary be made using recorded content that conforms to the filter condition from among the recorded content, transmitting means for transmitting the prompt generated by the prompt generating means to a generation AI device, and receiving means for receiving a summary generated by the generation AI device that operates in accordance with the prompt generated by the prompt generating means. The generation AI device can be made to create a summary from the recorded content that conforms to the filter condition.

[0022] The present invention further provides an answer generation method corresponding to the answer generation device according to the second aspect of the present invention. [Brief explanation of the drawings]

[0023] [Figure 1] FIG. 1 is a block diagram showing the hardware configuration of the care content input / chat system. [Figure 2] Shows the user master database. [Figure 3] Shows the user master database. [Figure 4] Shows the keyword master database. [Figure 5] Filter criteria showing master database. [Figure 6] 1 shows a record table. [Figure 7] 10 is a flowchart showing the flow of a login process. [Figure 8] 10 is a flowchart of a chat process when a family member of a facility user logs in. [Figure 9A] 10 is a flowchart of a chat process when a staff member of a nursing care facility logs in. [Figure 9B] 10 is a flowchart of a chat process when a staff member of a nursing care facility logs in. [Figure 9C] 10 is a flowchart of a chat process when a staff member of a nursing care facility logs in. [Figure 10] The chat screen when a family member of a facility user logs in and the process for displaying the chat screen are shown. [Figure 11] The following shows the chat screen when a care facility staff member logs in, and the process for displaying the chat screen. [Figure 12] The following shows the chat screen when a care facility staff member logs in, and the process for displaying the chat screen. [Figure 13] The following shows the chat screen when a care facility staff member logs in, and the process for displaying the chat screen. [Figure 14] The following shows the chat screen when a care facility staff member logs in, and the process for displaying the chat screen. [Figure 15] FIG. 10 is a block diagram showing the hardware configuration of another aspect of a care content input / chat system. [Example]

[0024] Figure 1 is a block diagram showing the hardware configuration of the care content input / chat system.

[0025] Referring to Figure 1, the care content input / chat system includes a record management server 1, a record terminal 2, a generation AI device 3, and a user terminal 4, which are connected to each other via a network (e.g., the Internet) so that data can be sent and received between them.

[0026] The recording terminal 2 is used by staff (employees, etc.) of a care facility (business office) to input various care (support) details that are provided or have been provided to users of the care facility (typically elderly people or people with disabilities), and the input care details are recorded (saved) (care records). The care records are used to consider care plans for facility users, to utilize them in high-quality care services and care management, to improve the qualifications of staff, and to share information. The care content input system / chat system is used in common by multiple care facilities, etc., and is operated by staff from each of the multiple care facility business offices. Multiple recording terminals 2 operated by multiple staff members are used to input care details (care records).

[0027] There are multiple nursing care facilities, each of which generally has multiple users (elderly people and people with disabilities). Nursing care (support) for each of the multiple users at multiple nursing care facilities is provided almost daily, and records are kept almost daily. The recorded nursing care (support) details include many items (care items), such as the user's vital signs (body temperature, blood pressure, percutaneous arterial oxygen saturation, etc.), dietary details (e.g., amount of food consumed for breakfast, lunch, and dinner), fluid intake, oral care, height, weight, exercise, visitation records, and recreational details. The nursing care (support) details recorded in the nursing care content input / chat system include not only details related to personal care for the user, but also the user's own personal records (such as the aforementioned body temperature, height, weight), accident reports, etc.

[0028] A record management server 1 is connected to the record terminal 2 via a network (such as the Internet). The record management server 1 manages the care (support) provided and to be provided for each facility user (record management, schedule management). Various reports are also managed (stored) in the record management server 1.

[0029] Both the record management server 1 and the record terminal 2 are computer devices equipped with a CPU that controls overall operation, RAM that temporarily stores data and programs, an operating system, storage (SSD, HDD, etc.) that stores various application programs, etc., an input device (keyboard, etc.) for inputting letters, numbers, etc., a communication device that sends and receives data over a network, and a display device that displays record forms / document forms, etc. The record management server 1 can be a stationary computer device that can be connected to a network. The record terminal 2 can be a portable computer device that can be connected to a network.

[0030] In this embodiment, as described below, the generation AI device 3 is used in the process of generating summaries (answers) to inquiries (questions) from the record management server 1. The generation AI device 3 can be a device capable of executing a generation program capable of natural language processing, such as ChatGPT, Bard, Bing, or Copilot. For example, information is input to the generation AI device 3 using an API (application programming interface) provided by a business that provides the generation AI device 3, and the information generated (answered) by the generation AI device 3 is obtained.

[0031] The user terminal 4 is used by staff at the nursing care facility and family members of nursing care facility users (e.g., the user's children or grandchildren), and is a mobile terminal such as a smartphone. The user terminal 4 also has a CPU, RAM, storage, an input device, a communication device, and a display device. A chat application program is stored in the storage of the user terminal 4, and when the chat application program is executed on the user terminal 4, a chat screen is displayed on the display device of the user terminal 4. When a user of the user terminal 4 (a staff member at the nursing care facility or a family member of the nursing care facility user) uses this chat screen to input text questions about the facility user's condition, answers to these questions are generated in the care content input / chat system, and the generated answers are displayed on the chat screen. In addition to the above-mentioned care (support) content management process, the record management server 1 also performs chat processing (answer generation processing) by controlling the generation AI device 3. Details of the chat processing performed by the user terminal 4, the record management server 1, and the generation AI device 3 in cooperation with each other, and the chat screen displayed on the display device of the user terminal 4 during the chat processing will be described later.

[0032] Databases and tables (user master database 11, user master database 12, keyword master database 13, filter condition master database 14, and record table database 15) that store various data used in the operation of the care content input / chat system are connected to the record management server 1. These databases and tables used in the operation of the care content input / chat system will be explained below with reference to Figures 2 to 6.

[0033] Referring to Figure 2, the user master database 11 (hereinafter referred to as user master 11) stores data about users who use the care content input / chat system, such as the user ID, the name (full name) of the user identified by the user ID, the classification of "family member" or "staff member," and the user ID of the facility user.

[0034] As described above, the care content input / chat system is used by staff of a care facility, etc., to input the care (support) details that will be provided or have been provided to facility users and to record the input care details. User master 11 is registered with a user ID and name that is individually assigned to each staff member of the care facility, etc., who performs their duties using the care content input / chat system. For staff members of a care facility, etc., classification data (shown as "staff member" in Figure 2) that indicates that the person identified by the user ID is a staff member of the care facility, etc., is registered in the classification column.

[0035] As will be described later, the care content input / chat system is used not only by staff at care facilities, but also, typically, by family members of facility users. Family members of facility users are also assigned user IDs, and their names are registered. In the category field, category data (shown as "family" in Figure 2) is registered, indicating that the person identified by the user ID is a family member of the facility user.

[0036] For the family members of facility users, a user ID for identifying each facility user is also registered in association with the user ID. This identifies them as family members of a specific facility user. Conversely, the user IDs of staff at nursing care facilities, etc., are not associated with user IDs.

[0037] Referring to Figure 3, the user master database 12 (hereinafter referred to as the user master 12) stores a user ID assigned to each facility user who uses a care facility, etc., the facility user's name (given name) (written in kanji), the name written in kana, age, gender, etc. The user ID is an identification code unique to a facility user who uses a care facility, etc., and one user ID is assigned to one facility user. The facility user ID may include a code that identifies the establishment, in which case the establishment to which the facility user belongs can also be identified by the user ID.

[0038] Referring to Figure 4, the keyword master database 13 (hereinafter referred to as the keyword master 13) stores a keyword number, a keyword, one or more filter numbers, etc. The keywords stored in the keyword master 13 are registered as multiple words (terms) that are input to the user terminal 4 when the chat process executed by the user terminal 4, the record management server 1, and the generation AI device 3 in cooperation is used for the work of staff at a care facility or the like. When a keyword (word) registered in the keyword master 13 is input to the user terminal 4, the care content input / chat system is used for staff work (for example, "handover" work) (operates in staff mode). Details of the process when the keyword master 13 is used and the care content input / chat system is used for staff work will be described later.

[0039] Referring to FIG. 5, the filter condition master database 14 (hereinafter referred to as the filter condition master 14) stores filter conditions (record data extraction conditions) that are applied when keywords stored in the above-described keyword master 13 are entered into the user terminal 4 and the care content input / chat system is used for staff work. For example, filter conditions can be created (set) for each type of keyword (keyword number) stored in the above-described keyword master 13. Referring to FIGS. 4 and 5, three filter numbers, 1, 2, and 3, are associated with the keyword number 1 in the keyword master 13, "transfer." This means that when the keyword "transfer" is entered into the user terminal 4, the filter conditions corresponding to the three filter numbers, 1, 2, and 3, are applied. Specifically, referring to FIG. 5, the three filter conditions, "body temperature 37 degrees or higher," "record of meals where staple food accounts for 50% or less," and "record of meals where side dishes account for 50% or less," are applied. Details of the processing of the care content input / chat system using the filter condition master 14 will be described later.

[0040] Referring to Figure 6, Figure 6 shows one of a plurality of record tables stored in record table database 15. Record table 15A records various care (support) details that are provided or have been provided to facility users by staff of a care facility or the like. Record table 15A is created, for example, for each facility user and for each date, and includes data that identifies the user ID and the date.

[0041] A plurality of types of care record forms can be displayed on the record terminal 2, and a care record is entered by inputting data into data entry fields provided in the care record form, which is then stored (registered) in the record table 15A. Each time a care record is entered into the care record form and registered in the record table 15A, a record including the record name (care record form name), date, time, and record content is accumulated in the record table 15A.

[0042] Figure 6 shows examples of record names (names of care record forms) such as "Health Management," "Fluids," "Visit," and "Dinner." For example, if we look at the record with the record name "Health Management," 2024 / 7 / 17 is stored in the date column, 13:30 is stored in the time column, and data is stored in the temperature column, SPO2 column, and condition observation column. This means that at 13:30 on July 17, 2024, a care record was created by a staff member of a care facility or other facility using the "Health Management" form, which is one of the care record forms, and that the body temperature (36°C), SPO2 (98%), and condition observation (appears to be in good health) were entered in the health management form, and this was stored in record table 15A. The record name "Health Management" can be said to be an abbreviated name that indicates the detailed record content (in the above case, body temperature, SPO2, and condition observation).

[0043] Figure 7 is a flowchart showing the flow of chat processing executed in the care content input / chat system at the time of login. The processing shown in Figure 7 starts by starting a chat program pre-installed in the user terminal 4. Details of the chat processing will be described later.

[0044] A user ID and password are entered using the user terminal 4 (step 21). The entered user ID and password are sent from the user terminal 4 to the records management server 1 via the network (step 22). The records management server 1 performs authentication processing using the user ID and password (step 31). If authentication is successful, subsequent processing proceeds, but if authentication fails, subsequent processing stops.

[0045] If the authentication is successful, a record including the received user ID is read from the user master 11 (FIG. 2) (step 32).

[0046] As described above, the user master 11 includes classification data of either "family" or "staff." The process branches depending on whether the classification of the user identified by the received user ID is "family" or "staff" (step 33).

[0047] If the user category identified by the received user ID is "family" (family of the facility user), the user ID of the facility user associated with the received user ID and stored in the user master 11 (Figure 2) is sent from the record management server 1 to the user terminal 4 along with the category data "family" ("user family" in step 33, step 34A).

[0048] If the user category identified by the received user ID is "staff" (staff of a nursing facility, etc.), the category data "staff" and the set of user ID and name of the facility user stored in the user master 12 (Figure 3) (generally multiple sets of user IDs and names) are sent from the record management server 1 to the user terminal 4 ("staff" in step 33, step 34B).

[0049] The processing of the user terminal 4 (chat program) branches depending on whether the user identified by the received user ID, i.e., the person who started the chat program using the user terminal 4 (logged in person), is a family member of the facility user or a facility staff member (step 23). The classification data sent from the record management server 1 to the user terminal 4 is used for this branching processing. If the classification data sent from the record management server 1 is "family," the user terminal 4 proceeds to the user family flow ("user family" at step 23). If the classification data is "staff," the user terminal 4 proceeds to the staff flow ("staff" at step 23).

[0050] Instead of the classification data, whether one user ID or multiple user IDs have been sent from the record management server 1 may be used to determine the branching process in step 23. If one user ID has been sent from the record management server 1, the user terminal 4 proceeds to the user family flow (step 34A, step 23: "User family"). If multiple user IDs have been sent from the record management server 1, the user terminal 4 proceeds to the staff flow (step 34B, step 23: "Staff"). In this case, the classification data does not necessarily need to be sent from the record management server 1 to the user terminal 4.

[0051] Figure 8 is a flowchart showing the flow of chat processing (answer generation processing) executed in the care content input / chat system when the person who started the chat program (logged-in person) is a family member of a facility user. Figure 10 shows an example of a chat screen (left side of Figure 10) displayed on the user terminal 4, and an example of the processing of the record management server 1 and the generation AI device 3 for displaying (generating) this chat screen (center and right side of Figure 10).

[0052] When the family member of the facility user successfully logs in, a chat screen 5 is displayed on the display screen of the user terminal 4.

[0053] 10, the chat screen 5 includes a question (inquiry) display field 5A, an answer (response) display field 5B, and an input field 5C. A question (inquiry) is entered into the input field 5C by a user of the user terminal 4 (here, a family member of a facility user). When the enter button 5D (downward triangle button) on the right end of the input field 5C is clicked (tapped), the question entered into the input field 5C is displayed in the question display field 5A on the chat screen 5. The question displayed in the input field 5C is erased.

[0054] A question about a specific facility user is entered in the input field 5C (question display field 5A) of the chat screen 5. The facility user is identified from the user ID entered during the login process described above and is identified by the user ID (see FIG. 7, step 34A) transmitted from the record management server 1. The user ID received from the record management server 1 and the question entered using the chat screen 5 are transmitted from the user terminal 4 to the record management server 1 (steps 24, 25).

[0055] The chat application program is pre-programmed with the address (or telephone number, etc.) of the record management server 1 as the communication destination (contact point). This allows the user ID and questions entered using the chat screen 5 to be sent from the user terminal 4 to the record management server 1.

[0056] Upon receiving the user ID and question, the record management server 1 accesses the record table database 15 and searches for the record table 15A that contains the received user ID and the date of the current day (the day of access from the user terminal 4) (step 35).

[0057] If the record table 15A is found by the search, a record name list (record form name list) is created from the found record table 15A (step 36).

[0058] For example, if record table 15A shown in FIG. 6 is found in the search, a list of record names including "health care," "water," "visit," and "dinner" is created.

[0059] The record management server 1 creates a record name selection (decision) request including the question sent from the user terminal 4 and the created record name list, and sends it to the generation AI device 3 (step 37).

[0060] The record name selection request may include a prompt such as:

[0061] {"messages":[{"role":"user","content":"Select and output record names that match the question based on the constraints.\n#Constraints\nOnly output record names (e.g., moisture)\nDo not include unnecessary plain text in the output results\n#Question\nHow much did you eat for dinner?\n#Record Name\nMoisture\nVisit\nDinner''}],"max_tokens":800,"temperature":0,"top_p":0,"frequency_penalty":0,"presence_penalty":0}

[0062] This record name selection request states, "Please select and output a record name that matches the question based on the constraints." The constraints are "output only the record name (e.g., fluids)" and "no unnecessary plain text should be added to the output result," the question is "How much did you eat for dinner?", and the record name is to be selected from the record name list "health management, fluids, visitation, dinner." The record name selection request does not include the name (full name) of the facility user.

[0063] Upon receiving the record name selection request, the generation AI device 3 executes processing according to the record name selection request, i.e., selects a record name that matches (suits) the content of the query from the record name list (step 51).

[0064] For example, if "Dinner" is selected from the list of record names "Health Management, Hydration, Visitation, Dinner" as the record name that matches the question "How much did you eat for dinner?", the record name "Dinner" is sent from the generation AI device 3 to the record management server 1.

[0065] Upon receiving the record name, the record management server 1 reads the record content corresponding to the received record name from the record table 15A (step 38).

[0066] It is determined whether the read record contents contain personal information (step 39). Specifically, the record management server 1 determines whether the user master 12 contains information that partially matches the read record contents.

[0067] For example, if the read-out dinner record includes "He seemed to be having fun communicating with the people around him (Mr. Shizuoka and Mr. Aichi)," the character strings "Shizuoka" and "Aichi" match part of the facility user's name stored in the name field of user master 12 (Figure 3). The matching character string is treated as personal information, and the personal information is kept secret by replacing that character string with another character string (YES in step 39, step 40). For example, in the case of the dinner record described above, a record is created that keeps the personal information secret, such as "He seemed to be having fun communicating with the people around him (Mr. XX and Mr. △△)."

[0068] If the read record contents do not include personal information, no special processing is performed (NO in step 39).

[0069] A request for summarization is created (step 41). If the record contents read in step 38 contain personal information, the record contents with the personal information concealed as described above are included in the request for summarization. If the record contents read in step 38 do not contain personal information, the read record contents are included in the request for summarization as is.

[0070] The summary request may include prompts such as:

[0071] {“message”:[{“role”:“user”,“content”:“Please summarize the following sentences: [Main dish] 100%, [Side dish] 100%, [Appearance] He enjoyed communicating with the people around him (Mr. / Ms. XX and Mr. / Ms. △△), [Oral care] He performed, [Oral care content] He brushed his teeth"}],"max_tokens":800,"temperature":0,"top_p":0,"frequency_penalty":0,"presence_penalty":0}

[0072] The summary request also does not include the name (first and last name) of the facility user.

[0073] Upon receiving the summary request, the AI ​​generation device 3 executes summarization processing in accordance with the request. If the summary target is "[Main dish] 100%, [Side dish] 100%, [Appearance] Communicated happily with those around (Mr. / Ms. XX and Mr. / Ms. △△), [Oral care] Performed, [Oral care content] Brushed teeth," a summary such as "All of the main dish and side dish were eaten. Communicated happily with those around (Mr. / Ms. XX and Mr. / Ms. △△). For oral care, teeth were brushed" is created (step 52). The created summary is sent from the AI ​​generation device 3 to the record management server 1.

[0074] The record management server 1 receives the summary and transmits it to the user terminal 4 (step 42). The received summary is displayed in the reply display field 5B on the chat screen 5 of the user terminal 4 (step 26, FIG. 10).

[0075] As mentioned above, the "record name" is a simplified name that indicates the detailed record content stored in the record content field. The record name is selected first according to the content of the question; in other words, the broad framework known as the "record name" is selected first according to the content of the question, making it easier to obtain an answer that does not deviate too far from the question. The summary that is ultimately output as an answer is created based on the detailed record content associated with the selected record name, making it easier to obtain an answer (summary) that includes detailed information. Even if the inquiry (question) is personal (a question directed at a specific user), it is possible to obtain an answer that appropriately addresses the inquiry.

[0076] Furthermore, in the above-described process, the personal information of facility users (personal information of other people) is not transmitted to the generation AI device 3. This makes it possible to protect the information of facility users.

[0077] The family members of facility users will be provided with information about the user's condition based on the care records via a chat screen. This allows the family members of facility users to easily find out about the user's condition at the care facility without having to visit the facility.

[0078] If the generation AI device 3 is unable to select a record name from the record name list that matches the content of the question entered by the user's family on the chat screen 5, it may respond to that effect to the record management server 1, and the record management server 1 may prompt the user's family to include a record name included in the record name list in the question sent to the user terminal 4. For example, a message such as "Please include one of the following in your question: 'health care, fluids, visitation, or dinner'" may be displayed on the chat screen 5. This allows the user's family to know, via the chat screen 5, the content of the care provided to the user at a nursing facility or the like.

[0079] Figures 9A to 9C are flowcharts showing the flow of chat processing (answer generation processing) executed in the care content input / chat system when the person who started the chat program (logged-in person) is a facility staff member. Figures 11 to 14 show an example of a chat screen (left side of Figures 11 to 14) displayed on the user terminal 4, and an example of the processing of the record management server 1 and the generation AI device 3 for displaying (generating) this chat screen (center and right side of Figures 11 to 14).

[0080] As mentioned above, the care content input / chat system can determine whether the person logging in is a user family member or a staff member during the login process (Figure 7).

[0081] 11 to 14, the chat screen 6 displayed on the display screen of the user terminal 4 when a staff member of a nursing facility or the like logs in differs from the chat screen 5 (FIG. 10) when a family member of the user logs in in that a selection field 6E for facility users is displayed. When a staff member of a nursing facility or the like logs in, all user IDs and names in the user master 12 are sent from the record management server 1 to the user terminal 4 during the login process (FIG. 7, step 34B), and this data is used to display the selection field 6E for facility users.

[0082] The facility user selection field 6E displays all selectable names of facility users stored in the user master 12 received from the record management server 1. The staff member can use the selection field 6E to select the name of a specific facility user about whom they wish to make an inquiry (question) from among the multiple facility users displayed there (step 61). Figures 11 and 12 show how "Shizuoka Taro" has been selected from among the multiple facility users.

[0083] In the facility user selection field 6E, in addition to selecting a specific facility user, it is also possible to select "no user selection," which does not select a specific facility user. Figures 13 and 14 show a state in which no specific facility user is selected ("no user selection" is selected).

[0084] First, we explain the processing of the care content input / chat system when a specific facility user is selected (Figures 11 and 12).

[0085] A user of the user terminal 4, in this case a staff member of a care facility or the like, inputs a question (inquiry) into the input field 6C (step 62). When the enter button 6D is clicked (tapped), the question input into the input field 6C is displayed in the question display field 6A. The user ID of the facility user selected using the selection field 6E and the question are sent from the user terminal 4 to the record management server 1 (YES in step 63, step 64).

[0086] When a specific user is selected in the selection field 6E and the user ID is received by the record management server 1 (YES in step 81), the record table 15A for the day of login with the received user ID is searched, and a record name list is created from the record table 15A found (steps 82, 84). A record name selection request including a question and a record name list is sent from the record management server 1 to the generation AI device 3 (step 85). For example, the following prompt for a record name selection request is created in the record management server 1 and sent to the generation AI device 3:

[0087] {"messages":[{"role":"user","content":"Select and output record names that match the question based on the constraints.\n#Constraints\nOnly output record names (e.g., water content)\nDo not include unnecessary plain text in the output results\nIf there are no record names that match the question, return "No record name"\n#Question\nHow much did you eat for dinner?\n#Record name\nWater content\nVisiting\nDinner\nRecreation\nDaytime activities / conditions\nBreakfast\nLunch\nNighttime care\nMorning care''}],"max_tokens":800,"temperature":0,"top_p":0,"frequency_penalty":0,"presence_penalty":0}

[0088] The generation AI device 3 determines whether the received record name list contains a record name that matches (fits) the received question (step 101).

[0089] If the record name list contains a record name that matches the question (YES in step 101), a string representing that record name (for example, "dinner") is sent from the generation AI device 3 to the record management server 1 (step 102).

[0090] The record management server 1 determines whether the received character string is in the record name list (step 86). For example, if the received character string is "dinner" and the record name list is "health care, fluids, visit, dinner," the received character string is included in the record name list, so the record content of that record name is read from the record table 15A (step 87), and a summary request including the record content is created and sent to the generation AI device 3 (step 90). As in the case where the logged-in user is a family member of the user described above, the record content is summarized in the generation AI device 3 and sent to the user terminal 4 via the record management server 1 (steps 104, 95). The received summary is displayed in the reply display field 6B on the chat screen 6 of the user terminal 4 (step 66, Figure 11).

[0091] If the question does not match a record name included in the record name list (NO in step 101), a string indicating that the record name is not in the question (in the case of the above prompt, "No record name") is sent from the generation AI device 3 to the record management server 1 (step 103). In this case, the record management server 1 reads the keyword master 13 (Fig. 4) and determines whether the question contains a keyword in the keyword master 13 (NO in step 86, steps 88 and 89).

[0092] The keyword master 13 contains multiple words used by staff at care facilities in their work, such as "transfer," "special change" (meaning "anything particularly unusual"), and "contact information" (Figure 4). If a question contains a word used by staff at care facilities in their work, the care content input / chat system will operate to support the staff's work.

[0093] Specifically, if the question contains a keyword registered in the keyword master 13 (YES in step 89), the record management server 1 reads the filter condition master 14 (FIG. 5) (step 92) and reads all record contents from the record name list (step 93). Then, a summary request including the filter condition and record contents is sent to the generation AI device 3 (step 94). For example, the following prompt is created and sent to the generation AI device 3:

[0094] {“messages”:[{“role”:“user”,“content”:“Please extract only records that meet the filter conditions and summarize the record contents. Return in sentences, not bullet points.\n#Filter conditions\nTemperature above 37°C\nRecord of meals with staple food amount of 5 or less\nRecord of meals with side dish amount of 5 or less\n#Record contents\n[8:00 Breakfast] Swallowing rehabilitation: Performed, Staple food: 2, Side dish: 0, Fluid: 200cc, Medication: Performed, Oral care: Performed, Condition (meal): Severe cough, meal discontinued\n[10:00 Bath] Bath: Canceled, Reason for cancellation: Poor health\n[7:00 Health management]Temperature: 38.0°C, Blood pressure_systolic: 120mmHg, Blood pressure_diastolic: 88mmHg, SPO2: 97%, Condition observation: Nurse instructed to perform regular re-examination\n[9:00 Excretion] Toilet Urine: Little, Stool: None, Situation (Excretion): Defecation -4\n[9:00 Medications] Medication Name: Laxoberon, Reason / Content: Defecation -4\n[13:45 Health Management] Body Temperature: 36.6℃, Blood Pressure_systolic: 125mmHg, Blood Pressure_diastolic: 80mmHg, Pulse: 55 beats / min"}],"max_tokens":800,"temperature":0,"top_p":0,"frequency_penalty":0,"presence_penalty":0}

[0095] Upon receiving the summary request, the generation AI device 3 executes the summarization process according to the request and sends the created summary to the records management server 1 (steps 104 and 95). The summary is sent to the user terminal 4 and displayed on the chat screen 6.

[0096] As shown in Figure 12, when the user name "Shizuoka Taro" is selected and a question containing the keyword "handover," such as "Tell me about today's handover," is entered, all records for user "Shizuoka Taro" for that day are read, and the AI ​​generation device 3 generates a summary using records that meet the filter condition corresponding to the keyword "handover." As described above, the keyword "handover" is associated with three filter conditions: "body temperature above 37 degrees," "records of meals with less than 50% staple food," and "records of meals with less than 50% side dishes." Therefore, a summary is created using records that include "body temperature above 37 degrees," "records of meals with less than 5% staple food," and "records of meals with less than 5% side dishes." The care content input / chat system can be made to create summaries of content that is considered to require attention for a specific user, i.e., summaries suitable for handover in this case.

[0097] If the question does not include the record name and does not include any keywords stored in the keyword master 13 (NO in step 86, NO in step 89), error processing is performed (step 91). For example, a message urging the facility staff member to use one of the terms registered in the keyword master 13 is displayed on the chat screen 6 of the user terminal 4 used by the facility staff member.

[0098] Next, we will explain the processing of the care content input / chat system when no specific facility user is selected ("No user selected") (Figures 13 and 14).

[0099] If no specific facility user is selected, only the question is sent from the user terminal 4 to the record management server 1, and the user ID is not sent (NO in step 63, NO in step 81). In this case, the record table to be processed is the record table of all users (step 83).

[0100] For example, if the question contains the record name "health management" included in the record name list, such as "Please tell me about today's health management handover," the record tables of all users are used to generate a summary using the record contents of the record name "health management" for all users, and it is displayed on the chat screen 6 (YES in step 101, step 102, YES in step 86, step 87, steps 90, 104, 95, 66) (see Figure 13).

[0101] If the question does not include a record name, but does include a keyword registered in the keyword master 13, such as "handover," all record contents for that day are read out, and the AI ​​generation device 3 uses the record contents that meet the filter conditions corresponding to the keyword "handover" to generate a summary (NO in step 101, NO in step 103, NO in step 86, YES in step 88, step 89, steps 92-94, 104, 95, 66) (see Figure 14).

[0102] If the person who started the chat program (the logged-in person) is a facility staff member, the confidentiality processing of personal information that is performed when the logged-in person is a family member of the user is not performed. Even if personal information is included in the record contents read from record table 15A, it will be summarized as is (see Figure 11).

[0103] In the above-described embodiment, a record table 15A having date data of the day when access was made from the user terminal 4 is used, and the record name selection process and summary generation process are performed in cooperation with the record management server 1 and the generation AI device 3. Instead of the record table 15A of the day when access was made, date data may be sent from the user terminal 4 to the record management server 1, and the record table 15A having that date may be used to perform the record name selection process and summary generation process.

[0104] In the above embodiment, both the user's family and the facility staff use the same chat program, but the program used by the user's family and the program used by the facility staff may be different programs. In this case, it is not necessary to determine whether the person logging in to the user terminal 4 is a user's family member or a facility staff member (step 23).

[0105] Fig. 15 is a block diagram showing the hardware configuration of a care content input / chat system according to another embodiment. This system differs from the care content input / chat system shown in Fig. 1 in that it is equipped with a text-to-speech device 7 and an intercom 8. The intercom 8 is typically worn by staff at a care facility or the like.

[0106] The text-to-speech device 7 has a function of converting sentences and words represented by given text data into sound (audio) data.

[0107] The intercom 8 includes a microphone 8A and an earphone 8B, and has the function of outputting sound data representing the sound picked up by the microphone 8A and reproducing the input sound data through the earphone 8B.

[0108] The user terminal 4 is connected to the text-to-speech device 7 and the intercom 8 by wired or wireless connection. The text-to-speech device 7 is also connected to the intercom 8 by wired or wireless connection. The voice data of the staff member picked up by the microphone 8A of the intercom 8 is converted into text data in the user terminal 4. Questions (inquiries) can be entered by voice into the input field 6C of the chat screen 6. Furthermore, the text data representing the summary displayed in the answer display field 6B on the chat screen 6 is given to the text-to-speech device 7 and output as voice. The voice data output from the text-to-speech device 7 is given to the earphone 8B of the intercom 8, allowing the staff member wearing the intercom 8 to hear the summary by voice. [Explanation of symbols]

[0109] 1 Records Management Server 2 Recording terminal 3 Generation AI device 4 User terminal 5,6 Chat screen 11 User Master Database 12 User Master Database 13 Keyword Master Database 14 Filter Condition Master Database 15 Record Table Database 15A Recording Table

Claims

1. a record table storing a plurality of records, each of which is associated with a detailed record content and a record name that briefly indicates the detailed record content; a record name selection means for selecting a record name that matches the received query from among a plurality of record names provided in each of a plurality of records stored in the record table; detailed record content reading means for reading detailed record content associated with the record name selected by the record name selecting means from the record table; a summarizing means for summarizing the detailed record contents read by the detailed record contents reading means; and an output means for outputting a summary of the detailed record contents summarized by the summarizing means as an answer to the question; Answer generator.

2. The record name selection means a first prompt generating means for generating a prompt including a record name list including a plurality of record names included in a plurality of records stored in the record table and the question, the prompt requesting the user to select a record name from the record name list that matches the content of the question; a first transmitting means for transmitting the prompt generated by the first prompt generating means to the generating AI device; and a first receiving means for receiving a record name selected by a generating AI device operating in accordance with the prompt generated by the first prompt generating means; The answer generating device according to claim 1 .

3. The summarizing means: second prompt generating means for generating a prompt including the detailed record content read by the detailed record content reading means, the prompt requesting a summary of the detailed record content; a second transmitting means for transmitting the prompt generated by the second prompt generating means to the generating AI device; and and second receiving means for receiving a summary generated by the generating AI device operating in accordance with the prompt generated by the second prompt generating means; The answer generating device according to claim 1 or 2.

4. a plurality of the recording tables, each of the recording tables including a different identification code; Among the plurality of record tables, a record table having a given identification code is used in a record name selection process by the record name selection means and a detailed record content reading process by the detailed record content reading means. The answer generating device according to claim 1 .

5. a plurality of the record tables, each of the record tables containing data representing a different date; Among the plurality of record tables, a record table having date data of a given date is used in the record name selection process by the record name selection means and the detailed record content reading process by the detailed record content reading means. The answer generating device according to claim 1 .

6. personal information determination means for determining whether personal information is included in the detailed record contents read by the detailed record contents reading means; and and a personal information concealment means for concealing the determined personal information when the personal information determination means determines that the detailed record contents include personal information. The answer generating device according to claim 1 .

7. a record name selection means, using a record table in which a plurality of record records are stored, each of which is recorded in association with a detailed record content and a record name that abbreviates the detailed record content, to select a record name that matches the received question from a plurality of record names included in each of the plurality of record records stored in the record table; A detailed record content reading means reads the detailed record content associated with the selected record name from the record table; A summarizing means summarizes the read detailed record contents, The output means outputs a summary of the detailed record contents as an answer to the question. Answer generation method.

8. a filtering means for extracting record contents recorded in records that meet predetermined filter conditions from a record table in which a plurality of records, each of which has record contents recorded therein, are stored; a summarizing means for summarizing the record contents extracted by the filtering means; and an output means for outputting a summary of the recorded content summarized by the summarizing means; Answer generator.

9. The summarizing means: prompt generation means for generating a prompt including the record contents and a filter condition, the prompt requesting a summary using record contents that conform to the filter condition among the record contents; a transmitting means for transmitting the prompt generated by the prompt generating means to the generating AI device; and receiving means for receiving a summary generated by a generation AI device that operates in accordance with the prompt generated by the prompt generation means; The answer generating device according to claim 8 .

10. a plurality of the recording tables, each of the recording tables including a different identification code; Among the plurality of recording tables, a recording table having a given identification code is used for filtering processing by the filtering means. The answer generating device according to claim 8 .

11. A filtering means extracts record contents recorded in records that meet predetermined filter conditions from a record table in which a plurality of records, each of which has record contents recorded therein, A summarization means summarizes the extracted record content; An output means outputs a summary of the summarized record content. Answer generation method.

Images