System

Abstract

A system is provided.SOLUTION: A system comprising: means for analyzing a message log using a generative model and extracting a negative word and a dangerous word; means for calculating a risk score of a message based on the extracted words; and means for aggregating the risk scores for each employee and determining and notifying an escalation destination when a risk threshold is exceeded.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The technology of the present disclosure relates to a system. [Background technology]

[0002] Patent document 1 discloses a persona chatbot control method performed by at least one processor, the method including the steps of receiving a user utterance, adding the user utterance to a prompt including an instruction sentence related to a description of the chatbot character, encoding the prompt, and inputting the encoded prompt into a language model to generate a chatbot utterance in response to the user utterance. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-180282 Summary of the Invention [Problem to be solved by the invention]

[0004] It is important for companies to monitor the health and mental state of their employees in real time and to detect and respond to risks such as the loss of human resources or information leaks at an early stage. However, due to a lack of methods to effectively analyze unstructured data such as chats between employees and detect risks at an early stage, managing these risks is difficult. Furthermore, by identifying employee complaints at an early stage and taking appropriate measures, companies need to improve the working environment and reduce corporate risks. [Means for solving the problem]

[0005] The present invention provides a system that includes a means for analyzing chat logs using a generative model and extracting negative and dangerous words, a means for calculating a risk score for messages based on the extracted words, and a means for aggregating the risk scores for each employee and determining and notifying an escalation destination when a risk threshold is exceeded. Furthermore, the system includes a means for periodically conducting pulse surveys and integrating the results with the risk scores to perform a final risk assessment. Finally, the system provides a dashboard that includes a means for monitoring and displaying employee mental state and information leakage risk in real time, thereby enhancing risk management across the entire company.

[0006] A "generative model" is an algorithm that uses natural language processing technology to analyze text data and recognize and extract specific patterns and words.

[0007] A "message log" is a record of text messages exchanged between employees on internal company chat platforms, etc.

[0008] "Negative words" are specific words or phrases that may indicate stress, dissatisfaction, or mental strain in employees.

[0009] "Dangerous words" are specific words or phrases that could lead directly to information leaks or corporate risks.

[0010] A "risk score" is a number calculated based on negative or dangerous words contained in a message, and indicates the employee's risk level.

[0011] An "escalation destination" is a specific department or person within the company who will be notified if the risk score exceeds a certain threshold.

[0012] A "pulse survey" is a simple questionnaire survey conducted to periodically check employee satisfaction and mental health.

[0013] A "dashboard" is an interface for monitoring and displaying data on employee mental health and information leakage risk in real time.

[0014] The "risk threshold" is a numerical value that is the standard at which a risk score is deemed to require escalation. [Brief explanation of the drawings]

[0015] [Figure 1] 1 is a conceptual diagram showing an example of the configuration of a data processing system according to a first embodiment. [Figure 2] 1 is a conceptual diagram showing an example of main functions of a data processing device and a smart device according to a first embodiment. [Figure 3] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a second embodiment. [Figure 4] FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and smart glasses according to a second embodiment. [Figure 5] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a third embodiment. [Figure 6] FIG. 11 is a conceptual diagram showing an example of main functions of a data processing device and a headset-type terminal according to a third embodiment. [Figure 7] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a fourth embodiment. [Figure 8] FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and a robot according to a fourth embodiment. [Figure 9] 1 shows an emotion map onto which multiple emotions are mapped. [Figure 10] 1 shows an emotion map onto which multiple emotions are mapped. [Figure 11] FIG. 3 is a sequence diagram showing a processing flow of the data processing system according to the first embodiment. [Figure 12] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 1. [Figure 13]FIG. 10 is a sequence diagram showing the flow of processing in the data processing system according to the second embodiment when an emotion engine is combined. [Figure 14] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 2 when an emotion engine is combined. DETAILED DESCRIPTION OF THE INVENTION

[0016] An example of an embodiment of a system according to the technology of the present disclosure will be described below with reference to the accompanying drawings.

[0017] First, the terms used in the following description will be explained.

[0018] In the following embodiments, a coded processor (hereinafter simply referred to as a "processor") may be a single arithmetic device or a combination of multiple arithmetic devices. Furthermore, a processor may be a single type of arithmetic device or a combination of multiple types of arithmetic devices. Examples of arithmetic devices include a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a GPGPU (General-Purpose computing on Graphics Processing Units), and an APU (Accelerated Processing Unit).

[0019] In the following embodiments, a coded RAM (Random Access Memory) is a memory in which information is temporarily stored and is used as a working memory by a processor.

[0020] In the following embodiments, the coded storage is one or more non-volatile storage devices that store various programs, various parameters, etc. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), magnetic disks (e.g., hard disks), and magnetic tapes.

[0021] In the following embodiments, a communication I / F (Interface) with a symbol is an interface including a communication processor, an antenna, etc. The communication I / F controls communication between multiple computers. Examples of communication standards applied to the communication I / F include wireless communication standards including 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), Bluetooth (registered trademark), etc.

[0022] In the following embodiments, "A and / or B" is synonymous with "at least one of A and B." In other words, "A and / or B" means that it may be only A, only B, or a combination of A and B. Furthermore, in this specification, the same concept as "A and / or B" is also applied when three or more things are expressed connected by "and / or."

[0023] [First embodiment]

[0024] FIG. 1 shows an example of the configuration of a data processing system 10 according to the first embodiment.

[0025] 1, a data processing system 10 includes a data processing device 12 and a smart device 14. An example of the data processing device 12 is a server.

[0026] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0027] The smart device 14 includes a computer 36, a reception device 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The reception device 38, the output device 40, and the camera 42 are also connected to the bus 52.

[0028] The reception device 38 includes a touch panel 38A, a microphone 38B, and the like, and receives user input. The touch panel 38A detects contact with an indicator (for example, a pen or a finger) to receive user input by the touch of the indicator. The microphone 38B detects the user's voice to receive user input by voice. The control unit 46A transmits data indicating the user input received by the touch panel 38A and the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data indicating the user input.

[0029] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting the data in a form of expression that the user 20 can perceive (for example, audio and / or text). The display 40A displays visible information such as text and images in accordance with instructions from the processor 46. The speaker 40B outputs audio in accordance with instructions from the processor 46. The camera 42 is a compact digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0030] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 control the exchange of various information between the processor 46 and the processor 28 via the network 54.

[0031] FIG. 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0032] 2, in the data processing device 12, a specific process is performed by the processor 28. A specific processing program 56 is stored in the storage 32. The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific process is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0033] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0034] In the smart device 14, the processor 46 performs the reception output process. The storage 50 stores a reception output program 60. The reception output program 60 is used in conjunction with the specific processing program 56 by the data processing system 10. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0035] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0036] This invention is a system for understanding the health and mental state of employees in a company in real time, and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze chat logs, extract and score negative and dangerous words, and further integrates the results of pulse surveys to perform a final risk assessment.

[0037] Program processing (natural language explanation)

[0038] server

[0039] 1. Collecting chat logs

[0040] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, MICROSOFT® TEAMS®). Automatically collecting this data using API enables timely monitoring.

[0041] 2. Log Analysis

[0042] The server inputs the collected chat logs into a generative model, which analyzes and extracts negative or dangerous words. The generative model then uses natural language processing technology to analyze the message content and detect specific words and phrases.

[0043] 3. Calculating the risk score

[0044] The server calculates a risk score for each message based on the words detected, for example, assigning a medium risk score to negative words like "painful" and a high risk score to dangerous words like "confidentiality leak."

[0045] 4. Score tallying and escalation

[0046] The server compiles risk scores for each employee and escalates when the cumulative score exceeds a certain threshold. The escalation destination is automatically determined based on the level of risk and notified.

[0047] 5. Conducting a pulse survey

[0048] The server periodically prepares pulse surveys and generates questions for employees, and the survey results are also incorporated into risk assessments to understand the overall risk status of employees.

[0049] Terminal

[0050] 1. Sending chat logs

[0051] The devices (employee PCs and smartphones) send chat logs generated during daily work to the server in real time. This data is used for the analysis process described below.

[0052] 2. Pulse Survey Responses

[0053] The device sends a pulse survey notification to the employee, who then responds. The response results are sent to the server and reflected in the risk assessment.

[0054] User (Administrator)

[0055] 1. Dashboard Monitoring

[0056] Users can use a dashboard provided by the server to monitor the mental state of employees and the risk of information leaks in real time. The dashboard provides visual information, allowing users to immediately check whether there is any risk.

[0057] 2. Escalation

[0058] Users receive escalation notifications from the server and can take action, such as arranging for specialized support for employees with high risk scores or reporting the situation to their immediate supervisor.

[0059] Specific examples

[0060] For example, suppose Employee A frequently mentions in chat that he or she "can't stand the work anymore." In this case, the server collects these messages and uses a generative model to detect the negative word "can't stand it." It then assigns a higher risk score to the message. If the cumulative risk score exceeds a threshold, the server sends an escalation notification to the HR department. The HR staff then schedules a meeting with Employee A and provides the necessary support.

[0061] In this way, the system of the present invention allows companies to detect risks of losing human resources or leaking information at an early stage and take prompt and appropriate measures. Furthermore, by integrating the results of regular pulse surveys, it is possible to more accurately grasp the overall health status of employees.

[0062] The processing flow will be explained below.

[0063] Step 1:

[0064] server

[0065] Get chat logs

[0066] The server periodically collects all chat messages using the API of the company's internal chat platform (e.g., Slack, Microsoft Teams), along with metadata such as the message sender, recipient, and timestamp.

[0067] Step 2:

[0068] server

[0069] Data Preprocessing

[0070] The server cleans up the chat logs by removing unnecessary information, such as removing HTML tags, removing special characters, and standardizing different languages.

[0071] Step 3:

[0072] server

[0073] Input to the generative model

[0074] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3®), which uses natural language processing techniques to analyze each message and extract important features based on the text content.

[0075] Step 4:

[0076] server

[0077] Extracting negative and dangerous words

[0078] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are saved along with the metadata.

[0079] Step 5:

[0080] server

[0081] Calculating the risk score

[0082] The server calculates a risk score for each message based on the negative and dangerous words detected, taking into account the importance and frequency of the words, as well as the message context.

[0083] Step 6:

[0084] server

[0085] Score tally

[0086] The server aggregates the risk scores for each employee to calculate a cumulative risk score for that employee, which is updated over time to keep it up to date.

[0087] Step 7:

[0088] server

[0089] Escalation decision

[0090] The server determines whether escalation is necessary when the cumulative risk score exceeds a pre-defined threshold, and if so, determines the appropriate escalation point (e.g., HR department, direct manager).

[0091] Step 8:

[0092] server

[0093] Escalation Notification

[0094] The server sends an escalation notification, which includes the employee's risk score, any negative or dangerous words detected, and recommended actions to take.

[0095] Step 9:

[0096] Terminal

[0097] Conducting a pulse survey

[0098] The terminal receives notifications of pulse surveys sent from the server and periodically asks employees to answer the pulse surveys. Employees answer the questions through the terminal, and the results are sent to the server.

[0099] Step 10:

[0100] server

[0101] Integration of pulse survey results

[0102] The server combines the pulse survey results with the risk score to create an overall risk assessment for the employee, which also influences escalation decisions.

[0103] Step 11:

[0104] User

[0105] Dashboard Monitoring

[0106] The user (administrator) uses a dashboard provided by the server to monitor the mental state of employees and the risk of information leakage in real time. The dashboard displays visual graphs and numerical values, allowing the user to immediately grasp the risk status.

[0107] Step 12:

[0108] User

[0109] Escalation

[0110] The user receives escalation notifications from the server and can take necessary action, such as arranging a meeting with an employee or arranging for professional support.

[0111] Example 1

[0112] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0113] To prevent employees' mental health from deteriorating and information leaks from occurring, a system that monitors employee behavior in real time and takes appropriate action is required. However, conventional systems lacked sufficient chat log analysis, making it difficult to quickly detect and respond to risks. Furthermore, there was a lack of tools to grasp employees' overall status, making it difficult to provide support in a timely manner.

[0114] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0115] In this invention, the server includes a means for analyzing communication logs using a generative model and extracting negative and dangerous expressions, a means for calculating a risk score for communication based on the extracted expressions, and a means for aggregating risk scores for each employee and determining and notifying notification recipients when a risk threshold is exceeded. This allows for real-time monitoring of employee mental state and information leakage risk, enabling timely escalation and response. Furthermore, by conducting periodic simple surveys and integrating the results with risk scores, more accurate risk assessments are possible.

[0116] A "generative model" is an algorithm that uses natural language processing technology to analyze text data and extract or generate specific information.

[0117] "Communication logs" refer to data that records the communication history of employees on chat platforms, messengers, etc. used during work hours.

[0118] "Negative expressions" are words or phrases that employees use to express feelings of stress, dissatisfaction, anxiety, etc.

[0119] "Dangerous expressions" are words or phrases that suggest risks such as information leaks or compliance violations.

[0120] A "risk score" is an index that quantifies the risk of communication content based on the frequency and importance of negative or dangerous expressions.

[0121] A "simple survey" is a questionnaire survey conducted over a short period of time to understand the mental state and awareness of employees.

[0122] The "information display screen" is an interface that visually displays employees' mental state and the risk of information leakage, and allows for real-time monitoring.

[0123] "Escalation" is the process of notifying the person or department that needs to take action when the risk score exceeds a certain threshold.

[0124] This invention is a system that monitors the mental state of employees and the risk of information leakage in real time within a company, and analyzes communication logs containing negative or dangerous words. This system includes functions such as analyzing communication logs using a generative model, calculating and aggregating risk scores, notifying escalation at appropriate times, conducting periodic simple surveys, and providing an information display screen.

[0125] Hardware and Software

[0126] Server: The server is the main processing component, analyzing communication logs using generative AI models (e.g., GPT-4®), managing risk scores using a database, and sending escalation notifications using an email service.

[0127] Device: The device is an employee's PC or smartphone, and is used to send chat logs and answer simple surveys.

[0128] User (Administrator): Administrators use the information display screen to monitor employees' mental state and the risk of information leakage, and respond to escalations.

[0129] Program processing

[0130] Server Processing

[0131] 1. Collecting chat logs:

[0132] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams) via API, performs API authentication, and collects messages from the required channels and users.

[0133] Examples:

[0134] OAuth2 is used for API authentication, and the / channels.history endpoint is called to retrieve messages for a specific channel. The retrieved data is saved in JSON format.

[0135] 2. Log analysis:

[0136] The server inputs the collected chat logs into a generative AI model to analyze and extract negative and dangerous words. It then tokenizes the log data to generate prompt sentences and requests the generative model to analyze them.

[0137] Examples:

[0138] Generate a prompt such as "Please parse the following message: 'I can no longer stand the work.'"

[0139] 3. Calculating the risk score:

[0140] The server calculates a risk score based on the analysis results, assigning a score based on specific keywords to quantify the risk of each message.

[0141] Examples:

[0142] The keyword "painful" is assigned a medium risk score (5 points), and the keyword "confidentiality leak" is assigned a high risk score (10 points).

[0143] 4. Score tallying and escalation:

[0144] The server compiles risk scores for each employee, and if the cumulative risk score exceeds a certain threshold, it uses an email service to send an escalation notification to the HR department or other department.

[0145] Terminal handling

[0146] 1. Sending chat logs:

[0147] The device sends the employee's chat log in real time to a server, including all messages generated during work hours.

[0148] 2. Quick survey answers:

[0149] The device sends a simple survey notification to the employee, who then answers it. The results are sent to a server and used for risk assessment.

[0150] User Action

[0151] 1. Information display screen monitoring:

[0152] Managers can use the information display screen to monitor employees' mental health and the risk of information leaks, with risk scores and warnings displayed in real time.

[0153] 2. Escalation Response:

[0154] Administrators receive escalation notifications from the server and take appropriate action, such as suggesting interviews or counseling for employees with high risk scores.

[0155] In this way, the system of the present invention allows for real-time monitoring of employees' mental state and information leakage risks, enabling prompt and appropriate responses. Furthermore, comprehensive risk assessments linked to periodic simple surveys allow for a more accurate understanding of employees' health conditions.

[0156] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0157] Step 1: Collect chat logs

[0158] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams) via API. The input is messages from specific channels and users. The output is a JSON-formatted file containing the retrieved chat log data. Specifically, the server performs API authentication using OAuth2. It then calls the / channels.history endpoint to retrieve messages from specific channels and saves them in JSON format.

[0159] Step 2: Analyze the logs

[0160] The server inputs the collected chat logs into a generative AI model to analyze and extract negative and dangerous words. The input is a JSON file of the saved chat logs. The output is a list of negative and dangerous words as the analysis results. Specifically, the server reads the log data and tokenizes each message. It then inputs the generated prompt sentences into the generative AI model and obtains the analysis results from the model.

[0161] Step 3: Calculate the risk score

[0162] The server calculates a risk score based on the analysis results. The input is the analysis results from the generative AI model. The output is a risk score for each message. Specifically, the server lists the detected keywords, references the risk scores corresponding to each keyword, and calculates the total risk score. For example, the keyword "painful" is assigned 5 points, and the keyword "confidentiality leak" is assigned 10 points.

[0163] Step 4: Score aggregation and escalation

[0164] The server aggregates the risk scores for each employee and starts the escalation process if a certain threshold is exceeded. The input is the calculated risk score. The output is the sending of an escalation notification. Specifically, the server aggregates the risk scores for each employee from the database. If the cumulative score exceeds the threshold, it sets an escalation flag, determines the escalation destination, and sends the notification using an email service.

[0165] Step 5: Conduct a pulse survey

[0166] The server periodically prepares pulse surveys and generates questions for employees. The input is a list of employees and the results of past pulse surveys. The output is the generated pulse survey questions and their results. Specifically, the server automatically generates survey questions, creates a survey form, and sends it to each employee's email address. The collected survey responses are analyzed and reflected in the risk assessment.

[0167] Step 6: Monitoring the Dashboard

[0168] Users use a dashboard provided by the server to monitor employees' mental state and information leakage risks. The input is the risk score and pulse survey results from the server. The output is real-time risk information displayed on the dashboard. Specifically, users check the dashboard to visually understand the scores and warnings for each employee.

[0169] (Application example 1)

[0170] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0171] Companies need to understand the mental and physical health of their employees in real time, and quickly detect and respond to risks of losing talent or leaking information. However, current systems often have delays in analyzing chat logs and pulse survey results, making it difficult to detect risks early.

[0172] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0173] In this invention, the server includes means for analyzing message logs using a generative model and extracting negative and dangerous words, means for calculating risk scores for messages based on the extracted words, means for aggregating risk scores for each employee and determining and notifying an escalation destination when a risk threshold is exceeded, means for collecting data using the chat platform's API, means for periodically notifying employees of pulse survey questions and collecting their responses, and means for integrating the collected data with the risk scores to make a final risk assessment. This allows for a real-time understanding of employees' mental and health states, enabling early detection of risks and rapid response.

[0174] A "generative model" is an AI model that uses natural language processing technology to analyze message logs and detect specific words and phrases.

[0175] A "message log" is historical data of messages exchanged on a company's internal chat platform.

[0176] "Negative words" are specific words that may indicate an employee's mental or physical health is deteriorating.

[0177] "Dangerous words" are specific words that are likely to lead to information leaks or disrupt business operations.

[0178] A "risk score" is a numerical representation of the risk level of a message based on the negative or dangerous words contained in the message.

[0179] "Escalation" is a notification process that prompts support personnel or administrators to take action when a detected risk exceeds a certain threshold.

[0180] A "chat platform API" is an application program interface for obtaining data from chat systems used by companies.

[0181] A "pulse survey" is a brief questionnaire survey conducted to periodically check the mental and physical state of employees.

[0182] A "dashboard" is an interface that visually displays employees' mental state and the risk of information leaks in real time.

[0183] The system for implementing the present invention comprises the following steps.

[0184] server

[0185] The server periodically collects message logs using the API of the company's internal chat platform (e.g., Slack or Microsoft Teams). The collected message logs are then analyzed using a generative AI model (e.g., OpenAI® GPT-4 or Hugging Face Transformers). This analysis extracts negative and dangerous words from messages and calculates a risk score for each message.

[0186] The server also aggregates risk scores for each employee, and if the cumulative risk score exceeds a set threshold, it automatically determines the appropriate escalation and notifies the appropriate person. The server also periodically sends pulse surveys to employees and collects their responses. The results of the pulse surveys are integrated with existing risk scores and used in the final risk assessment.

[0187] Furthermore, the server provides managers with a dashboard that visualizes the mental state of employees and the risk of information leakage in real time, allowing them to monitor the risk situation and take necessary measures promptly.

[0188] Terminal

[0189] Employee devices (smartphones and PCs) send chat logs generated during daily work to a server in real time. This data is used for the analysis process described below. The devices also receive survey notifications, and employees respond to pulse surveys via their devices. Responses are sent to the server and reflected in risk assessments.

[0190] User (Administrator)

[0191] Administrators use a dashboard provided by the server to monitor employees' mental health and information leakage risks in real time. The dashboard provides visual information, allowing them to immediately check whether there is any risk. If an escalation occurs, administrators receive a notification from the server and take appropriate countermeasures, such as arranging for professional support for employees with high risk scores.

[0192] Specific examples

[0193] For example, if Employee A frequently says in chat that he or she "can't stand the work anymore," the server collects those messages and uses a generative model to detect the negative word "can't stand it." The server assigns a high risk score to those messages, and if the cumulative risk score exceeds a threshold, the server sends an escalation notification to the administrator. The administrator receives the escalation notification, schedules a meeting with Employee A, and provides the necessary support.

[0194] Prompt Sentence Examples

[0195] "Please list the negative and dangerous words frequently used by employees in chat, calculate a risk score for each, and also perform a sentiment analysis on those words, outputting the results in JSON format."

[0196] In this way, the server, terminals, and administrators work together to grasp the mental and health status of employees in real time, creating a system that can detect risks early and respond quickly.

[0197] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0198] Step 1:

[0199] The server periodically collects message logs using the API of the company's chat platform. The data obtained via the API is in JSON format and includes the date and time, sender information, and message content.

[0200] Input: Message logs obtained from the chat platform's API

[0201] Output: Message log in JSON format, including date, time, sender information, and message content

[0202] Step 2:

[0203] The server inputs the collected message logs into a generative AI model to analyze the message content. The generative model uses natural language processing technology to extract negative and dangerous words from the messages.

[0204] Input: Message log in JSON format

[0205] Output: Message log with negative and dangerous words extracted

[0206] Step 3:

[0207] The server calculates a risk score for each message based on the extracted negative and dangerous words, for example, assigning a medium risk score to "painful" and a high risk score to "confidentiality leak."

[0208] Input: Message log from which negative and dangerous words were extracted

[0209] Output: Message log with risk score assigned

[0210] Step 4:

[0211] The server aggregates the risk scores for each employee and determines the escalation destination when the cumulative risk score exceeds a set threshold. The server automatically notifies the escalation destination.

[0212] Input: Message logs with risk scores

[0213] Output: Escalation notification

[0214] Step 5:

[0215] The server periodically sends pulse survey questions to employees and collects their responses, which are also included in the risk assessment.

[0216] Input: Pulse Survey Question Notification

[0217] Output: Employee pulse survey responses

[0218] Step 6:

[0219] The server combines the collected pulse survey results with existing risk scores to produce a final risk assessment.

[0220] Input: Pulse survey results and risk score

[0221] Output: Final integrated risk assessment

[0222] Step 7:

[0223] The server provides a dashboard to administrators, which monitors and displays the mental state of employees and the risk of information leakage in real time. Administrators can check the risk status through the dashboard and take appropriate measures.

[0224] Input: Final Integrated Risk Assessment

[0225] Output: Real-time monitoring via dashboard

[0226] Step 8:

[0227] The device sends chat platform message logs and pulse survey responses to the server in real time, ensuring that the latest data is always used for analysis and evaluation.

[0228] Input: Employee chat logs, pulse survey responses

[0229] Output: Real-time updated data

[0230] Step 9:

[0231] Administrators receive escalation notifications from the server and take appropriate action for employees with high risk scores, such as arranging psychological support or scheduling a meeting with the employee.

[0232] Input: Escalation notification

[0233] Output: Implementing countermeasures

[0234] Furthermore, an emotion engine that estimates the user's emotion may be combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59 and perform identification processing using the user's emotion.

[0235] This invention is a system for understanding the health and mental state of employees in a company in real time and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze chat logs, extracting and scoring negative and dangerous words, and combining it with an emotion engine to recognize emotional states and perform comprehensive risk assessments. Furthermore, by integrating pulse survey results and displaying them on a dashboard in real time, it is possible to accurately grasp the overall health status of employees.

[0236] Program processing (natural language explanation)

[0237] server

[0238] 1. Obtaining chat logs

[0239] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams), automatically collecting message data and metadata (sender, recipient, timestamp) using an API.

[0240] 2. Data Preprocessing

[0241] The server cleans up the chat logs by removing unnecessary information, such as removing HTML tags, removing special characters, and standardizing different languages.

[0242] 3. Input to the generative model

[0243] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3), which uses natural language processing techniques to analyze each message and extract important features based on the text content.

[0244] 4. Extraction of negative and dangerous words

[0245] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are saved along with the metadata.

[0246] 5. Calculating the risk score

[0247] The server calculates a risk score for each message based on the words detected, taking into account the importance and frequency of the words, as well as the message context.

[0248] 6. Emotion analysis using an emotion engine

[0249] The server uses an emotion engine to analyze the emotional state of the detected message. For example, the emotion engine reads emotions from the message, recognizes emotions such as "anger," "sadness," and "joy," and calculates an emotion score.

[0250] 7. Score tallying and escalation

[0251] The server aggregates the risk and sentiment scores for each employee and calculates the cumulative risk rating for that employee. If this rating exceeds a set threshold, it determines who should be escalated and notifies them.

[0252] 8. Conducting a pulse survey

[0253] The server periodically prepares pulse surveys and generates questions for employees, and the survey results are also incorporated into risk assessments to understand the overall risk status of employees.

[0254] Terminal

[0255] 1. Sending chat logs

[0256] The devices (employee PCs and smartphones) send chat logs generated during daily work to the server in real time. This data is used for the analysis process described below.

[0257] 2. Pulse Survey Responses

[0258] The device sends a pulse survey notification to the employee, who then responds. The response results are sent to the server and reflected in the risk assessment.

[0259] User (Administrator)

[0260] 1. Dashboard Monitoring

[0261] Users can use a dashboard provided by the server to monitor the mental state of employees and the risk of information leaks in real time. The dashboard displays visual graphs and numerical values, allowing users to immediately grasp the risk status.

[0262] 2. Escalation

[0263] The user receives escalation notifications from the server and can take necessary action, such as arranging a meeting with an employee or arranging for professional support.

[0264] Specific examples

[0265] For example, suppose Employee A frequently mentions in chat that he or she "can't stand the work anymore." In this case, the server collects these messages and uses a generative model to detect the negative word "can't stand it." The emotion engine then recognizes strong emotions such as "anger" or "sadness" from these messages and calculates an emotion score based on that. This gives the message a high risk score.

[0266] If the cumulative risk score exceeds the threshold, the server sends an escalation notification to the HR department. HR personnel then schedule a meeting with Employee A and provide the necessary support. In this way, the system of the present invention enables companies to detect risks of losing human resources or leaking information early and respond quickly and appropriately. Furthermore, by combining it with an emotion engine, it is possible to more accurately grasp the emotional state of employees and achieve more effective risk management.

[0267] The processing flow will be explained below.

[0268] Step 1:

[0269] server

[0270] Get chat logs

[0271] The server periodically collects all chat messages using the API of the company's internal chat platform (e.g., Slack, Microsoft Teams), including metadata such as message sender, recipient, and timestamp.

[0272] Step 2:

[0273] server

[0274] Data Preprocessing

[0275] The server cleans the collected chat logs of unnecessary information, including removing HTML tags, special characters, and language uniformity. The preprocessed data is then prepared for analysis.

[0276] Step 3:

[0277] server

[0278] Input to the generative model

[0279] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3), which uses natural language processing techniques to analyze each message and extract important features based on its content.

[0280] Step 4:

[0281] server

[0282] Extracting negative and dangerous words

[0283] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are then saved along with the metadata.

[0284] Step 5:

[0285] server

[0286] Calculating the risk score

[0287] The server calculates a risk score for each message based on the words detected, for example, assigning a medium risk score to the negative word "painful" and a high risk score to the dangerous word "confidentiality leak."

[0288] Step 6:

[0289] server

[0290] Emotion analysis using an emotion engine

[0291] The server uses an emotion engine to analyze the emotional state of the message. Specifically, the emotion engine recognizes emotions such as "anger," "sadness," and "joy," and calculates an emotion score based on that.

[0292] Step 7:

[0293] server

[0294] Score tally

[0295] The server aggregates the risk and sentiment scores for each employee to calculate a cumulative risk rating, which is updated periodically to keep the rating up to date.

[0296] Step 8:

[0297] server

[0298] Escalation decision

[0299] If the cumulative risk assessment exceeds the set threshold, the server determines that escalation is necessary, determines the escalation destination (e.g., HR department, direct supervisor), and notifies the appropriate person.

[0300] Step 9:

[0301] server

[0302] Escalation Notification

[0303] The server then sends a notification to the determined escalation point, which includes the employee's risk score, any negative or dangerous words detected, and recommended actions to take.

[0304] Step 10:

[0305] Terminal

[0306] Conducting a pulse survey

[0307] The terminal displays the pulse survey notification received from the server to the employee, periodically prompting them to answer the survey. The employee answers the survey on the terminal and sends the answer data to the server.

[0308] Step 11:

[0309] server

[0310] Integration of pulse survey results

[0311] The server integrates the collected pulse survey results into a risk assessment, providing a more accurate picture of an employee's overall risk status based on this integrated data.

[0312] Step 12:

[0313] User

[0314] Dashboard Monitoring

[0315] Users (administrators) use a dashboard provided by the server to monitor the mental state of employees and the risk of information leakage in real time. The dashboard displays intuitive graphs and indicators, allowing users to instantly check the risk status.

[0316] Step 13:

[0317] User

[0318] Escalation

[0319] The user receives escalation notifications from the server and can take necessary measures, such as arranging a meeting with an employee or arranging for professional support.

[0320] Example 2

[0321] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0322] Understanding the mental and physical state of employees within a company is an important issue in managing the risks of losing talent and leaking information. However, conventional methods have made it difficult to comprehensively evaluate this in real time, making it difficult to respond in a timely manner. There is a need for a method to accurately and quickly understand employees' mental state and to detect and respond to the risks of losing talent and leaking information early on.

[0323] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0324] In this invention, the server includes means for analyzing message logs using a generative model and extracting negative and dangerous words, means for calculating a risk score for messages based on the extracted words, means for analyzing the emotional state of messages using an emotion engine and calculating an emotion score, and means for aggregating the risk and emotion scores for each employee and determining and notifying an escalation destination when a risk threshold is exceeded. This makes it possible to accurately evaluate the mental state of employees in real time and to detect and respond to risks of human resource outflow and information leaks at an early stage.

[0325] A "generative model" is an algorithm that uses natural language processing technology to analyze text data and extract important features.

[0326] "Message Log" refers to message data and metadata obtained from an enterprise's internal chat platform.

[0327] "Negative words" refers to a group of words that suggest an employee's mental state is deteriorating.

[0328] "Dangerous words" refers to a group of words that pose a high risk of information leakage or security risk.

[0329] An "emotion engine" refers to technology that analyzes emotional states from text data and recognizes emotions such as "anger," "sadness," and "joy."

[0330] The "risk score" is a numerical value that indicates the degree of risk of a message, calculated based on the extracted negative and dangerous words.

[0331] The "emotion score" is a numerical representation of the emotional state analyzed by the emotion engine.

[0332] A "pulse survey" is a questionnaire survey that is used to easily understand the current status of employees in a short period of time.

[0333] A "dashboard" is an interface that visually displays various data and allows for real-time monitoring of employees' mental state and the risk of information leaks.

[0334] "Escalation" refers to the process of notifying management of employees or situations deemed to be high risk when a set threshold is exceeded.

[0335] The present invention is a system for understanding the health and mental state of employees in a company in real time and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze message logs, extract and score negative and dangerous words, and further combines it with an emotion engine to recognize emotional states and perform comprehensive risk assessment. Specific embodiments for implementing the present invention are described below.

[0336] Server Operation

[0337] 1. Obtaining chat logs

[0338] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams) using an API to collect message data and metadata (sender, recipient, timestamp).

[0339] 2. Data Preprocessing

[0340] The server cleans up the chat logs by removing unnecessary information, using the BeautifulSoup library to remove HTML tags, regular expressions to remove special characters, and the langdetect library to unify the language.

[0341] 3. Input to the generative model

[0342] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3), which analyzes the text and extracts important features through the generative model's API (e.g., OpenAI GPT-3 API).

[0343] 4. Extraction of negative and dangerous words

[0344] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are saved along with the metadata.

[0345] 5. Calculating the risk score

[0346] The server calculates a risk score for each message based on the extracted words, taking into account the importance and frequency of each word, as well as the message context, to calculate an overall score.

[0347] 6. Emotion analysis using an emotion engine

[0348] The server uses an emotion engine (e.g., TextBlob, VADER) to analyze the emotional state of the detected messages, recognizing emotions such as "anger," "sadness," and "joy" and calculating an emotion score.

[0349] 7. Score tallying and escalation

[0350] The server aggregates the risk and emotion scores for each employee and calculates a cumulative risk rating. If the rating exceeds a set threshold, it determines who should be escalated and notifies them.

[0351] 8. Conducting a pulse survey

[0352] The server periodically prepares pulse surveys and generates questions for employees using the SurveyMonkey API, and the employee responses are reflected in the risk assessment.

[0353] Device behavior

[0354] 1. Sending chat logs

[0355] The devices (employee PCs and smartphones) send chat logs generated during daily work to the server in real time, allowing the server to always analyze the latest data.

[0356] 2. Pulse Survey Responses

[0357] The device sends a pulse survey notification to the employee, and when the employee responds, the results are sent to the server and reflected in the risk assessment.

[0358] User (administrator) actions

[0359] 1. Dashboard Monitoring

[0360] Users can use a dashboard provided by the server to monitor the mental state of employees and the risk of information leaks in real time. The dashboard uses data visualization tools such as Grafana and Tableau to display visual graphs and figures, allowing users to immediately grasp risks.

[0361] 2. Escalation

[0362] Users receive escalation notifications from the server and take necessary action. Notifications are sent via email or a dedicated app. For example, HR personnel can schedule a meeting with high-risk employees or arrange for specialized support.

[0363] Specific examples

[0364] For example, if Employee A frequently mentions "I can't stand my job anymore" in chat, the server collects the messages and uses the BERT model to detect the negative word "unbearable." At the same time, the VADER emotion engine recognizes strong emotions such as "anger" or "sadness" in the messages and calculates an emotion score. This gives the message a high risk score.

[0365] If the cumulative risk score exceeds the set threshold, the server sends an escalation notification to the HR department. HR personnel then schedule a meeting with Employee A and provide the necessary support. This system enables companies to detect risks of losing human resources or leaking information early and respond quickly and appropriately.

[0366] Prompt Sentence Examples

[0367] "Employee A frequently says, 'I can't take it anymore.' Extract negative words and sentiment scores from these messages and calculate a risk score."

[0368] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0369] Step 1:

[0370] The server periodically retrieves chat logs from the company's internal chat platforms (e.g., Slack, Microsoft Teams). To retrieve the logs, it uses the API provided by each platform to collect message data and metadata (sender, recipient, timestamp) via an authentication token. Using this, it sends an HTTP request to the API endpoint and receives chat log data in JSON format as a response.

[0371] Input: Authentication token obtained from the chat platform API

[0372] Output: Message data and metadata in JSON format

[0373] Specific operation: Sending an API request and receiving a response

[0374] Step 2:

[0375] The server cleans up the chat logs by removing unnecessary information and message texts. First, it uses the BeautifulSoup library to remove HTML tags, then it uses regular expressions to remove special characters, and it uses the langdetect library to detect the language of the text and unify it if necessary.

[0376] Input: Chat log data in JSON format

[0377] Output: Preprocessed message text

[0378] Specific actions: HTML tag removal, special character removal, language detection and unification

[0379] Step 3:

[0380] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3). Using the generative model's API (e.g., OpenAI GPT-3 API), it sends the message text to the model and receives the parsed results.

[0381] Input: Preprocessed message text

[0382] Output: Analysis results by generative model

[0383] Specific operation: Sending a request to the Generative Model API and receiving a response

[0384] Step 4:

[0385] The generative model extracts negative and dangerous words from the analysis results, matches them with predefined negative and dangerous word lists, and stores the detected words in a database along with metadata.

[0386] Input: Analysis results from the generative model

[0387] Output: Detected negative and dangerous words

[0388] Specific operation: Matching with word list, saving detection results to database

[0389] Step 5:

[0390] The server calculates a risk score for each message based on the extracted words. The server calculates an overall score by taking into account the importance and frequency of each word, as well as the message context. The server then uses a scoring algorithm to calculate the risk score and stores it in a database.

[0391] Input: Detected negative and dangerous words

[0392] Output: Risk score for each message

[0393] Specific operation: Calculate risk score and save it to the database

[0394] Step 6:

[0395] The server analyzes the emotional state of each message using an emotion engine (e.g., TextBlob, VADER). Through the emotion engine API, the server analyzes the message and obtains an emotion score such as "anger," "sadness," or "joy."

[0396] Input: Preprocessed message text

[0397] Output: Sentiment score from the sentiment engine

[0398] Specific operation: Sending a request to the Emotion Engine API and receiving a response

[0399] Step 7:

[0400] The server aggregates the risk and emotion scores for each employee and calculates a cumulative risk rating. If the rating exceeds a set threshold, it determines who should be escalated and notifies them. Ratings and notifications are performed automatically based on the information in the database.

[0401] Input: Risk score and sentiment score for each message

[0402] Output: Cumulative risk assessment and escalation notification

[0403] Specific actions: Calculating risk assessment, sending escalation notifications

[0404] Step 8:

[0405] The server periodically prepares pulse surveys and generates questions for employees. The surveys are created and administered using the SurveyMonkey API, etc. The survey results are also incorporated into risk assessments and are automatically saved in a database.

[0406] Input: Pulse survey execution instruction

[0407] Output: Pulse survey response results

[0408] Specific operations: Generate and send pulse surveys, save response results

[0409] (Application example 2)

[0410] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0411] The difficulty of understanding the health and mental state of employees in a company in real time and quickly detecting and responding to risks of losing human resources or information leaks is a major problem. In particular, extracting negative or dangerous words from employees' chat logs and daily communications to conduct risk assessments is cumbersome, and manual assessments are difficult to monitor efficiently and accurately. Furthermore, a systematic system is needed to comprehensively assess employees' mental and emotional states and escalate or take necessary action in a timely manner.

[0412] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for analyzing message logs using a generative model and extracting negative and dangerous words, means for calculating a risk score for messages based on the extracted words, means for aggregating risk scores for each employee and determining and notifying an escalation destination when a risk threshold is exceeded, means for analyzing the emotional state of the message using an emotion engine and reflecting the results in the risk score, and means for integrating the results of the pulse survey to perform a final risk assessment. This makes it possible to accurately grasp the health and mental state of employees in real time and efficiently perform risk assessment and escalation.

[0413] A "generative model" is a model that uses natural language processing technology to analyze message logs and extract important features and patterns.

[0414] A "message log" is a record of messages sent and received by employees on company chat platforms, etc.

[0415] "Negative words" are words with negative meanings, such as "painful" or "tired."

[0416] "Dangerous words" are words related to information leaks and risks, such as "confidential" and "leak."

[0417] A "risk score" is a risk index calculated based on the importance and frequency of occurrence of extracted negative and dangerous words.

[0418] An "emotion engine" is a technology for analyzing emotions such as "anger," "sadness," and "joy" from the content of a message.

[0419] A "pulse survey" is a survey that periodically asks simple questions to understand employees' health and mental state.

[0420] "Escalation" is the process of notifying higher-level management when a certain risk score is exceeded and urging them to take action.

[0421] A "dashboard" is a tool that visually monitors and displays employees' mental and risk status in real time.

[0422] This invention is a system for understanding the health and mental state of employees in a company in real time and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze chat logs, extracting and scoring negative and dangerous words, and combining it with an emotion engine to recognize emotional states and perform comprehensive risk assessments. Furthermore, by integrating pulse survey results and displaying them on a dashboard in real time, it is possible to accurately grasp the overall health status of employees.

[0423] Server Processing

[0424] The server automatically collects message data and metadata (sender, recipient, timestamp) from the company's internal chat platform (e.g., a general messaging system) using an API to periodically retrieve chat logs. It then removes unnecessary information from the collected chat logs and cleans up the message text. Specifically, it performs processes such as removing HTML tags, deleting special characters, and standardizing different languages.

[0425] The preprocessed chat logs are input into a generative model (e.g., BERT, GPT-3), which uses natural language processing techniques to analyze each message and extract important features based on the text content. Negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") are detected from the analysis results. The detected words are saved along with metadata.

[0426] Next, a risk score is calculated for each message based on the detected words, taking into account the importance and frequency of the words, as well as the message context. Furthermore, an emotion engine is used to analyze the emotional state of the detected message. For example, the emotion engine can read emotions from the message, recognizing emotions such as "anger," "sadness," and "joy," and then calculate an emotion score.

[0427] The server aggregates the risk and emotion scores for each employee and calculates the employee's cumulative risk rating. If this rating exceeds a set threshold, it determines who should be escalated and notifies them.

[0428] Additionally, the server periodically prepares pulse surveys and generates questions for employees. The results of the survey are also incorporated into risk assessments to understand the employee's overall risk status. For example, questions such as, "Have you been feeling stressed at work recently?"

[0429] Terminal handling

[0430] Employee devices (PCs and smartphones) send chat logs generated during daily work to a server in real time. This data is used for the analysis process described below. The data sent from the device goes through the processing steps described above on the server and is used for risk assessment. The device also sends pulse survey notifications to employees, who respond to these. The responses are sent to the server and reflected in the risk assessment.

[0431] User operations

[0432] Administrators use a dashboard provided by the server to monitor employees' mental state and the risk of information leaks in real time. The dashboard displays visual graphs and figures, allowing them to immediately grasp the risk status. Upon receiving an escalation notification, administrators can take the necessary countermeasures, such as arranging a meeting with the employee or arranging for professional support.

[0433] Specific examples

[0434] For example, suppose Employee A frequently mentions in chat that he or she "can't stand the work anymore." In this case, the server collects these messages and uses a generative model to detect the negative word "can't stand it." The emotion engine then recognizes strong emotions such as "anger" or "sadness" from these messages and calculates an emotion score based on that. This gives the message a high risk score.

[0435] If the cumulative risk score exceeds the threshold, the server sends an escalation notification to the management department. The manager then schedules a meeting with Employee A and provides the necessary support. In this way, the system of the present invention enables companies to detect risks of losing human resources or leaking information early and respond quickly and appropriately.

[0436] Prompt Sentence Examples

[0437] "Employee A says: 'I'm at my limit in this job.' Please analyze the emotional state of this message and calculate an appropriate risk score."

[0438] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0439] Step 1:

[0440] The server periodically retrieves chat logs from the company's internal chat platform. For example, it uses an API to automatically collect message data and metadata (sender, recipient, timestamp) from the company's internal messaging system. The input is the chat platform's API endpoint, and the output is chat log data in JSON format.

[0441] Step 2:

[0442] The server removes unnecessary information from the collected chat logs and cleans up the message text. Specifically, it performs processes such as removing HTML tags, deleting special characters, and unifying different languages. The input is the chat log data from the previous step, and the output is the cleaned message text in the target language.

[0443] Step 3:

[0444] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3). The model uses natural language processing techniques to analyze each message and extract important features based on the text content. The input is the cleaned message text from the previous step, and the output is the feature-extracted data.

[0445] Step 4:

[0446] The server detects negative and risky words from the analysis results of the generative model. The detected words are saved along with the associated metadata. The input is the feature extraction data from the previous step, and the output is a list of detected negative and risky words.

[0447] Step 5:

[0448] The server calculates a risk score for each message based on the words detected, taking into account the importance and frequency of the words, as well as the message context. The input is the list of negative and dangerous words from the previous step, and the output is a risk score for each message.

[0449] Step 6:

[0450] The server uses an emotion engine to analyze the emotional state of the message. Specifically, it reads emotions from the message, recognizes emotions such as "anger," "sadness," and "joy," and calculates an emotion score. The input is the message to which the risk score from the previous step has been assigned, and the output is the emotion score.

[0451] Step 7:

[0452] The server aggregates the risk and sentiment scores for each employee and calculates the employee's cumulative risk rating. If this rating exceeds a set threshold, it determines who should be escalated and notifies them. The input is each employee's risk and sentiment score, and the output is an escalation notification.

[0453] Step 8:

[0454] The server periodically prepares pulse surveys and generates questions for employees. The survey results are also incorporated into the risk assessment. The inputs are the survey questions and employee responses, and the output is an updated risk assessment.

[0455] Step 9:

[0456] The terminals send chat logs generated during employees' daily work to the server in real time. This data is used for the analysis process described above. The input is the chat messages generated on the terminals, and the output is the data sent to the server.

[0457] Step 10:

[0458] The terminal sends a pulse survey notification to the employee, who then responds. The response results are sent to the server and reflected in the risk assessment. The input is the response of the employee who received the survey notification on the terminal, and the output is the survey data sent to the server.

[0459] Step 11:

[0460] Users use a dashboard provided by the server to monitor the mental state of employees and the risk of information leakage in real time. The dashboard displays visual graphs and numerical values, allowing users to immediately grasp the risk status. The input is real-time risk data received from the server, and the output is the visualized dashboard screen.

[0461] Step 12:

[0462] The user receives the escalation notification and takes necessary measures, such as setting up a meeting with an employee or arranging for professional support. The input is the escalation notification from the server, and the output is the implemented measures.

[0463] The specific processing unit 290 transmits the result of the specific processing to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[0464] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (registered trademark) (Internet search engine).<URL: https: / / openai.com / blog / chatgpt> ), Gemini (registered trademark) (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0465] In the above embodiment, an example in which the specific process is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific process may be performed by the smart device 14.

[0466] [Second embodiment]

[0467] FIG. 3 shows an example of the configuration of a data processing system 210 according to the second embodiment.

[0468] 3, the data processing system 210 includes the data processing device 12 and smart glasses 214. An example of the data processing device 12 is a server.

[0469] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0470] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, and the camera 42 are also connected to the bus 52.

[0471] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[0472] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[0473] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 control the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[0474] Fig. 4 shows an example of the main functions of the data processing device 12 and the smart glasses 214. As shown in Fig. 4, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[0475] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0476] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0477] In the smart glasses 214, the reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0478] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal."

[0479] This invention is a system for understanding the health and mental state of employees in a company in real time, and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze chat logs, extract and score negative and dangerous words, and further integrates the results of pulse surveys to perform a final risk assessment.

[0480] Program processing (natural language explanation)

[0481] server

[0482] 1. Collecting chat logs

[0483] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams). Automatically collecting this data using an API enables timely monitoring.

[0484] 2. Log Analysis

[0485] The server inputs the collected chat logs into a generative model, which analyzes and extracts negative or dangerous words. The generative model then uses natural language processing technology to analyze the message content and detect specific words and phrases.

[0486] 3. Calculating the risk score

[0487] The server calculates a risk score for each message based on the words detected, for example, assigning a medium risk score to negative words like "painful" and a high risk score to dangerous words like "confidentiality leak."

[0488] 4. Score tallying and escalation

[0489] The server compiles risk scores for each employee and escalates when the cumulative score exceeds a certain threshold. The escalation destination is automatically determined based on the level of risk and notified.

[0490] 5. Conducting a pulse survey

[0491] The server periodically prepares pulse surveys and generates questions for employees, and the survey results are also incorporated into risk assessments to understand the overall risk status of employees.

[0492] Terminal

[0493] 1. Sending chat logs

[0494] The devices (employee PCs and smartphones) send chat logs generated during daily work to the server in real time. This data is used for the analysis process described below.

[0495] 2. Pulse Survey Responses

[0496] The device sends a pulse survey notification to the employee, who then responds. The response results are sent to the server and reflected in the risk assessment.

[0497] User (Administrator)

[0498] 1. Dashboard Monitoring

[0499] Users can use a dashboard provided by the server to monitor the mental state of employees and the risk of information leaks in real time. The dashboard provides visual information, allowing users to immediately check whether there is any risk.

[0500] 2. Escalation

[0501] Users receive escalation notifications from the server and can take action, such as arranging for specialized support for employees with high risk scores or reporting the situation to their immediate supervisor.

[0502] Specific examples

[0503] For example, suppose Employee A frequently mentions in chat that he or she "can't stand the work anymore." In this case, the server collects these messages and uses a generative model to detect the negative word "can't stand it." It then assigns a higher risk score to the message. If the cumulative risk score exceeds a threshold, the server sends an escalation notification to the HR department. The HR staff then schedules a meeting with Employee A and provides the necessary support.

[0504] In this way, the system of the present invention allows companies to detect risks of losing human resources or leaking information at an early stage and take prompt and appropriate measures. Furthermore, by integrating the results of regular pulse surveys, it is possible to more accurately grasp the overall health status of employees.

[0505] The processing flow will be explained below.

[0506] Step 1:

[0507] server

[0508] Get chat logs

[0509] The server periodically collects all chat messages using the API of the company's internal chat platform (e.g., Slack, Microsoft Teams), along with metadata such as the message sender, recipient, and timestamp.

[0510] Step 2:

[0511] server

[0512] Data Preprocessing

[0513] The server cleans up the chat logs by removing unnecessary information, such as removing HTML tags, removing special characters, and standardizing different languages.

[0514] Step 3:

[0515] server

[0516] Input to the generative model

[0517] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3), which uses natural language processing techniques to analyze each message and extract important features based on the text content.

[0518] Step 4:

[0519] server

[0520] Extracting negative and dangerous words

[0521] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are saved along with the metadata.

[0522] Step 5:

[0523] server

[0524] Calculating the risk score

[0525] The server calculates a risk score for each message based on the negative and dangerous words detected, taking into account the importance and frequency of the words, as well as the message context.

[0526] Step 6:

[0527] server

[0528] Score tally

[0529] The server aggregates the risk scores for each employee to calculate a cumulative risk score for that employee, which is updated over time to keep it up to date.

[0530] Step 7:

[0531] server

[0532] Escalation decision

[0533] The server determines whether escalation is necessary when the cumulative risk score exceeds a pre-defined threshold, and if so, determines the appropriate escalation point (e.g., HR department, direct manager).

[0534] Step 8:

[0535] server

[0536] Escalation Notification

[0537] The server sends an escalation notification, which includes the employee's risk score, any negative or dangerous words detected, and recommended actions to take.

[0538] Step 9:

[0539] Terminal

[0540] Conducting a pulse survey

[0541] The terminal receives notifications of pulse surveys sent from the server and periodically asks employees to answer the pulse surveys. Employees answer the questions through the terminal, and the results are sent to the server.

[0542] Step 10:

[0543] server

[0544] Integration of pulse survey results

[0545] The server combines the pulse survey results with the risk score to create an overall risk assessment for the employee, which also influences escalation decisions.

[0546] Step 11:

[0547] User

[0548] Dashboard Monitoring

[0549] The user (administrator) uses a dashboard provided by the server to monitor the mental state of employees and the risk of information leakage in real time. The dashboard displays visual graphs and numerical values, allowing the user to immediately grasp the risk status.

[0550] Step 12:

[0551] User

[0552] Escalation

[0553] The user receives escalation notifications from the server and can take necessary action, such as arranging a meeting with an employee or arranging for professional support.

[0554] Example 1

[0555] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0556] To prevent employees' mental health from deteriorating and information leaks from occurring, a system that monitors employee behavior in real time and takes appropriate action is required. However, conventional systems lacked sufficient chat log analysis, making it difficult to quickly detect and respond to risks. Furthermore, there was a lack of tools to grasp employees' overall status, making it difficult to provide support in a timely manner.

[0557] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0558] In this invention, the server includes a means for analyzing communication logs using a generative model and extracting negative and dangerous expressions, a means for calculating a risk score for communication based on the extracted expressions, and a means for aggregating risk scores for each employee and determining and notifying notification recipients when a risk threshold is exceeded. This allows for real-time monitoring of employee mental state and information leakage risk, enabling timely escalation and response. Furthermore, by conducting periodic simple surveys and integrating the results with risk scores, more accurate risk assessments are possible.

[0559] A "generative model" is an algorithm that uses natural language processing technology to analyze text data and extract or generate specific information.

[0560] "Communication logs" refer to data that records the communication history of employees on chat platforms, messengers, etc. used during work hours.

[0561] "Negative expressions" are words or phrases that employees use to express feelings of stress, dissatisfaction, anxiety, etc.

[0562] "Dangerous expressions" are words or phrases that suggest risks such as information leaks or compliance violations.

[0563] A "risk score" is an index that quantifies the risk of communication content based on the frequency and importance of negative or dangerous expressions.

[0564] A "simple survey" is a questionnaire survey conducted over a short period of time to understand the mental state and awareness of employees.

[0565] The "information display screen" is an interface that visually displays employees' mental state and the risk of information leakage, and allows for real-time monitoring.

[0566] "Escalation" is the process of notifying the person or department that needs to take action when the risk score exceeds a certain threshold.

[0567] This invention is a system that monitors the mental state of employees and the risk of information leakage in real time within a company, and analyzes communication logs containing negative or dangerous words. This system includes functions such as analyzing communication logs using a generative model, calculating and aggregating risk scores, notifying escalation at appropriate times, conducting periodic simple surveys, and providing an information display screen.

[0568] Hardware and Software

[0569] Server: The server is the main processing component, analyzing communication logs using generative AI models (e.g., GPT-4), maintaining risk scores using a database, and sending escalation notifications using an email service.

[0570] Device: The device is an employee's PC or smartphone, and is used to send chat logs and answer simple surveys.

[0571] User (Administrator): Administrators use the information display screen to monitor employees' mental state and the risk of information leakage, and respond to escalations.

[0572] Program processing

[0573] Server Processing

[0574] 1. Collecting chat logs:

[0575] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams) via API, performs API authentication, and collects messages from the required channels and users.

[0576] Examples:

[0577] OAuth2 is used for API authentication, and the / channels.history endpoint is called to retrieve messages for a specific channel. The retrieved data is saved in JSON format.

[0578] 2. Log analysis:

[0579] The server inputs the collected chat logs into a generative AI model to analyze and extract negative and dangerous words. It then tokenizes the log data to generate prompt sentences and requests the generative model to analyze them.

[0580] Examples:

[0581] Generate a prompt such as "Please parse the following message: 'I can no longer stand the work.'"

[0582] 3. Calculating the risk score:

[0583] The server calculates a risk score based on the analysis results, assigning a score based on specific keywords to quantify the risk of each message.

[0584] Examples:

[0585] The keyword "painful" is assigned a medium risk score (5 points), and the keyword "confidentiality leak" is assigned a high risk score (10 points).

[0586] 4. Score tallying and escalation:

[0587] The server compiles risk scores for each employee, and if the cumulative risk score exceeds a certain threshold, it uses an email service to send an escalation notification to the HR department or other department.

[0588] Terminal handling

[0589] 1. Sending chat logs:

[0590] The device sends the employee's chat log in real time to a server, including all messages generated during work hours.

[0591] 2. Quick survey answers:

[0592] The device sends a simple survey notification to the employee, who then answers it. The results are sent to a server and used for risk assessment.

[0593] User Action

[0594] 1. Information display screen monitoring:

[0595] Managers can use the information display screen to monitor employees' mental health and the risk of information leaks, with risk scores and warnings displayed in real time.

[0596] 2. Escalation Response:

[0597] Administrators receive escalation notifications from the server and take appropriate action, such as suggesting interviews or counseling for employees with high risk scores.

[0598] In this way, the system of the present invention allows for real-time monitoring of employees' mental state and information leakage risks, enabling prompt and appropriate responses. Furthermore, comprehensive risk assessments linked to periodic simple surveys allow for a more accurate understanding of employees' health conditions.

[0599] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0600] Step 1: Collect chat logs

[0601] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams) via API. The input is messages from specific channels and users. The output is a JSON-formatted file containing the retrieved chat log data. Specifically, the server performs API authentication using OAuth2. It then calls the / channels.history endpoint to retrieve messages from specific channels and saves them in JSON format.

[0602] Step 2: Analyze the logs

[0603] The server inputs the collected chat logs into a generative AI model to analyze and extract negative and dangerous words. The input is a JSON file of the saved chat logs. The output is a list of negative and dangerous words as the analysis results. Specifically, the server reads the log data and tokenizes each message. It then inputs the generated prompt sentences into the generative AI model and obtains the analysis results from the model.

[0604] Step 3: Calculate the risk score

[0605] The server calculates a risk score based on the analysis results. The input is the analysis results from the generative AI model. The output is a risk score for each message. Specifically, the server lists the detected keywords, references the risk scores corresponding to each keyword, and calculates the total risk score. For example, the keyword "painful" is assigned 5 points, and the keyword "confidentiality leak" is assigned 10 points.

[0606] Step 4: Score aggregation and escalation

[0607] The server aggregates the risk scores for each employee and starts the escalation process if a certain threshold is exceeded. The input is the calculated risk score. The output is the sending of an escalation notification. Specifically, the server aggregates the risk scores for each employee from the database. If the cumulative score exceeds the threshold, it sets an escalation flag, determines the escalation destination, and sends the notification using an email service.

[0608] Step 5: Conduct a pulse survey

[0609] The server periodically prepares pulse surveys and generates questions for employees. The input is a list of employees and the results of past pulse surveys. The output is the generated pulse survey questions and their results. Specifically, the server automatically generates survey questions, creates a survey form, and sends it to each employee's email address. The collected survey responses are analyzed and reflected in the risk assessment.

[0610] Step 6: Monitoring the Dashboard

[0611] Users use a dashboard provided by the server to monitor employees' mental state and information leakage risks. The input is the risk score and pulse survey results from the server. The output is real-time risk information displayed on the dashboard. Specifically, users check the dashboard to visually understand the scores and warnings for each employee.

[0612] (Application example 1)

[0613] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0614] Companies need to understand the mental and physical health of their employees in real time, and quickly detect and respond to risks of losing talent or leaking information. However, current systems often have delays in analyzing chat logs and pulse survey results, making it difficult to detect risks early.

[0615] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0616] In this invention, the server includes means for analyzing message logs using a generative model and extracting negative and dangerous words, means for calculating risk scores for messages based on the extracted words, means for aggregating risk scores for each employee and determining and notifying an escalation destination when a risk threshold is exceeded, means for collecting data using the chat platform's API, means for periodically notifying employees of pulse survey questions and collecting their responses, and means for integrating the collected data with the risk scores to make a final risk assessment. This allows for a real-time understanding of employees' mental and health states, enabling early detection of risks and rapid response.

[0617] A "generative model" is an AI model that uses natural language processing technology to analyze message logs and detect specific words and phrases.

[0618] A "message log" is historical data of messages exchanged on a company's internal chat platform.

[0619] "Negative words" are specific words that may indicate an employee's mental or physical health is deteriorating.

[0620] "Dangerous words" are specific words that are likely to lead to information leaks or disrupt business operations.

[0621] A "risk score" is a numerical representation of the risk level of a message based on the negative or dangerous words contained in the message.

[0622] "Escalation" is a notification process that prompts support personnel or administrators to take action when a detected risk exceeds a certain threshold.

[0623] A "chat platform API" is an application program interface for obtaining data from chat systems used by companies.

[0624] A "pulse survey" is a brief questionnaire survey conducted to periodically check the mental and physical state of employees.

[0625] A "dashboard" is an interface that visually displays employees' mental state and the risk of information leaks in real time.

[0626] The system for implementing the present invention comprises the following steps.

[0627] server

[0628] The server periodically collects message logs using the API of the company's internal chat platform (e.g., Slack or Microsoft Teams). The collected message logs are then analyzed using a generative AI model (e.g., OpenAI GPT-4 or Hugging Face Transformers). During this analysis, negative and dangerous words are extracted from the messages, and a risk score is calculated for each message.

[0629] The server also aggregates risk scores for each employee, and if the cumulative risk score exceeds a set threshold, it automatically determines the appropriate escalation and notifies the appropriate person. The server also periodically sends pulse surveys to employees and collects their responses. The results of the pulse surveys are integrated with existing risk scores and used in the final risk assessment.

[0630] Furthermore, the server provides managers with a dashboard that visualizes the mental state of employees and the risk of information leakage in real time, allowing them to monitor the risk situation and take necessary measures promptly.

[0631] Terminal

[0632] Employee devices (smartphones and PCs) send chat logs generated during daily work to a server in real time. This data is used for the analysis process described below. The devices also receive survey notifications, and employees respond to pulse surveys via their devices. Responses are sent to the server and reflected in risk assessments.

[0633] User (Administrator)

[0634] Administrators use a dashboard provided by the server to monitor employees' mental health and information leakage risks in real time. The dashboard provides visual information, allowing them to immediately check whether there is any risk. If an escalation occurs, administrators receive a notification from the server and take appropriate countermeasures, such as arranging for professional support for employees with high risk scores.

[0635] Specific examples

[0636] For example, if Employee A frequently says in chat that he or she "can't stand the work anymore," the server collects those messages and uses a generative model to detect the negative word "can't stand it." The server assigns a high risk score to those messages, and if the cumulative risk score exceeds a threshold, the server sends an escalation notification to the administrator. The administrator receives the escalation notification, schedules a meeting with Employee A, and provides the necessary support.

[0637] Prompt Sentence Examples

[0638] "Please list the negative and dangerous words frequently used by employees in chat, calculate a risk score for each, and also perform a sentiment analysis on those words, outputting the results in JSON format."

[0639] In this way, the server, terminals, and administrators work together to grasp the mental and health status of employees in real time, creating a system that can detect risks early and respond quickly.

[0640] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0641] Step 1:

[0642] The server periodically collects message logs using the API of the company's chat platform. The data obtained via the API is in JSON format and includes the date and time, sender information, and message content.

[0643] Input: Message logs obtained from the chat platform's API

[0644] Output: Message log in JSON format, including date, time, sender information, and message content

[0645] Step 2:

[0646] The server inputs the collected message logs into a generative AI model to analyze the message content. The generative model uses natural language processing technology to extract negative and dangerous words from the messages.

[0647] Input: Message log in JSON format

[0648] Output: Message log with negative and dangerous words extracted

[0649] Step 3:

[0650] The server calculates a risk score for each message based on the extracted negative and dangerous words, for example, assigning a medium risk score to "painful" and a high risk score to "confidentiality leak."

[0651] Input: Message log from which negative and dangerous words were extracted

[0652] Output: Message log with risk score assigned

[0653] Step 4:

[0654] The server aggregates the risk scores for each employee and determines the escalation destination when the cumulative risk score exceeds a set threshold. The server automatically notifies the escalation destination.

[0655] Input: Message logs with risk scores

[0656] Output: Escalation notification

[0657] Step 5:

[0658] The server periodically sends pulse survey questions to employees and collects their responses, which are also included in the risk assessment.

[0659] Input: Pulse Survey Question Notification

[0660] Output: Employee pulse survey responses

[0661] Step 6:

[0662] The server combines the collected pulse survey results with existing risk scores to produce a final risk assessment.

[0663] Input: Pulse survey results and risk score

[0664] Output: Final integrated risk assessment

[0665] Step 7:

[0666] The server provides a dashboard to administrators, which monitors and displays the mental state of employees and the risk of information leakage in real time. Administrators can check the risk status through the dashboard and take appropriate measures.

[0667] Input: Final Integrated Risk Assessment

[0668] Output: Real-time monitoring via dashboard

[0669] Step 8:

[0670] The device sends chat platform message logs and pulse survey responses to the server in real time, ensuring that the latest data is always used for analysis and evaluation.

[0671] Input: Employee chat logs, pulse survey responses

[0672] Output: Real-time updated data

[0673] Step 9:

[0674] Administrators receive escalation notifications from the server and take appropriate action for employees with high risk scores, such as arranging psychological support or scheduling a meeting with the employee.

[0675] Input: Escalation notification

[0676] Output: Implementing countermeasures

[0677] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[0678] This invention is a system for understanding the health and mental state of employees in a company in real time and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze chat logs, extracting and scoring negative and dangerous words, and combining it with an emotion engine to recognize emotional states and perform comprehensive risk assessments. Furthermore, by integrating pulse survey results and displaying them on a dashboard in real time, it is possible to accurately grasp the overall health status of employees.

[0679] Program processing (natural language explanation)

[0680] server

[0681] 1. Obtaining chat logs

[0682] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams), automatically collecting message data and metadata (sender, recipient, timestamp) using an API.

[0683] 2. Data Preprocessing

[0684] The server cleans up the chat logs by removing unnecessary information, such as removing HTML tags, removing special characters, and standardizing different languages.

[0685] 3. Input to the generative model

[0686] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3), which uses natural language processing techniques to analyze each message and extract important features based on the text content.

[0687] 4. Extraction of negative and dangerous words

[0688] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are saved along with the metadata.

[0689] 5. Calculating the risk score

[0690] The server calculates a risk score for each message based on the words detected, taking into account the importance and frequency of the words, as well as the message context.

[0691] 6. Emotion analysis using an emotion engine

[0692] The server uses an emotion engine to analyze the emotional state of the detected message. For example, the emotion engine reads emotions from the message, recognizes emotions such as "anger," "sadness," and "joy," and calculates an emotion score.

[0693] 7. Score tallying and escalation

[0694] The server aggregates the risk and sentiment scores for each employee and calculates the cumulative risk rating for that employee. If this rating exceeds a set threshold, it determines who should be escalated and notifies them.

[0695] 8. Conducting a pulse survey

[0696] The server periodically prepares pulse surveys and generates questions for employees, and the survey results are also incorporated into risk assessments to understand the overall risk status of employees.

[0697] Terminal

[0698] 1. Sending chat logs

[0699] The devices (employee PCs and smartphones) send chat logs generated during daily work to the server in real time. This data is used for the analysis process described below.

[0700] 2. Pulse Survey Responses

[0701] The device sends a pulse survey notification to the employee, who then responds. The response results are sent to the server and reflected in the risk assessment.

[0702] User (Administrator)

[0703] 1. Dashboard Monitoring

[0704] Users can use a dashboard provided by the server to monitor the mental state of employees and the risk of information leaks in real time. The dashboard displays visual graphs and numerical values, allowing users to immediately grasp the risk status.

[0705] 2. Escalation

[0706] The user receives escalation notifications from the server and can take necessary action, such as arranging a meeting with an employee or arranging for professional support.

[0707] Specific examples

[0708] For example, suppose Employee A frequently mentions in chat that he or she "can't stand the work anymore." In this case, the server collects these messages and uses a generative model to detect the negative word "can't stand it." The emotion engine then recognizes strong emotions such as "anger" or "sadness" from these messages and calculates an emotion score based on that. This gives the message a high risk score.

[0709] If the cumulative risk score exceeds the threshold, the server sends an escalation notification to the HR department. HR personnel then schedule a meeting with Employee A and provide the necessary support. In this way, the system of the present invention enables companies to detect risks of losing human resources or leaking information early and respond quickly and appropriately. Furthermore, by combining it with an emotion engine, it is possible to more accurately grasp the emotional state of employees and achieve more effective risk management.

[0710] The processing flow will be explained below.

[0711] Step 1:

[0712] server

[0713] Get chat logs

[0714] The server periodically collects all chat messages using the API of the company's internal chat platform (e.g., Slack, Microsoft Teams), including metadata such as message sender, recipient, and timestamp.

[0715] Step 2:

[0716] server

[0717] Data Preprocessing

[0718] The server cleans the collected chat logs of unnecessary information, including removing HTML tags, special characters, and language uniformity. The preprocessed data is then prepared for analysis.

[0719] Step 3:

[0720] server

[0721] Input to the generative model

[0722] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3), which uses natural language processing techniques to analyze each message and extract important features based on its content.

[0723] Step 4:

[0724] server

[0725] Extracting negative and dangerous words

[0726] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are then saved along with the metadata.

[0727] Step 5:

[0728] server

[0729] Calculating the risk score

[0730] The server calculates a risk score for each message based on the words detected, for example, assigning a medium risk score to the negative word "painful" and a high risk score to the dangerous word "confidentiality leak."

[0731] Step 6:

[0732] server

[0733] Emotion analysis using an emotion engine

[0734] The server uses an emotion engine to analyze the emotional state of the message. Specifically, the emotion engine recognizes emotions such as "anger," "sadness," and "joy," and calculates an emotion score based on that.

[0735] Step 7:

[0736] server

[0737] Score tally

[0738] The server aggregates the risk and sentiment scores for each employee to calculate a cumulative risk rating, which is updated periodically to keep the rating up to date.

[0739] Step 8:

[0740] server

[0741] Escalation decision

[0742] If the cumulative risk assessment exceeds the set threshold, the server determines that escalation is necessary, determines the escalation destination (e.g., HR department, direct supervisor), and notifies the appropriate person.

[0743] Step 9:

[0744] server

[0745] Escalation Notification

[0746] The server then sends a notification to the determined escalation point, which includes the employee's risk score, any negative or dangerous words detected, and recommended actions to take.

[0747] Step 10:

[0748] Terminal

[0749] Conducting a pulse survey

[0750] The terminal displays the pulse survey notification received from the server to the employee, periodically prompting them to answer the survey. The employee answers the survey on the terminal and sends the answer data to the server.

[0751] Step 11:

[0752] server

[0753] Integration of pulse survey results

[0754] The server integrates the collected pulse survey results into a risk assessment, providing a more accurate picture of an employee's overall risk status based on this integrated data.

[0755] Step 12:

[0756] User

[0757] Dashboard Monitoring

[0758] Users (administrators) use a dashboard provided by the server to monitor the mental state of employees and the risk of information leakage in real time. The dashboard displays intuitive graphs and indicators, allowing users to instantly check the risk status.

[0759] Step 13:

[0760] User

[0761] Escalation

[0762] The user receives escalation notifications from the server and can take necessary measures, such as arranging a meeting with an employee or arranging for professional support.

[0763] Example 2

[0764] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0765] Understanding the mental and physical state of employees within a company is an important issue in managing the risks of losing talent and leaking information. However, conventional methods have made it difficult to comprehensively evaluate this in real time, making it difficult to respond in a timely manner. There is a need for a method to accurately and quickly understand employees' mental state and to detect and respond to the risks of losing talent and leaking information early on.

[0766] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[0767] In this invention, the server includes means for analyzing message logs using a generative model and extracting negative and dangerous words, means for calculating a risk score for messages based on the extracted words, means for analyzing the emotional state of messages using an emotion engine and calculating an emotion score, and means for aggregating the risk and emotion scores for each employee and determining and notifying an escalation destination when a risk threshold is exceeded. This makes it possible to accurately evaluate the mental state of employees in real time and to detect and respond to risks of human resource outflow and information leaks at an early stage.

[0768] A "generative model" is an algorithm that uses natural language processing technology to analyze text data and extract important features.

[0769] "Message Log" refers to message data and metadata obtained from an enterprise's internal chat platform.

[0770] "Negative words" refers to a group of words that suggest an employee's mental state is deteriorating.

[0771] "Dangerous words" refers to a group of words that pose a high risk of information leakage or security risk.

[0772] An "emotion engine" refers to technology that analyzes emotional states from text data and recognizes emotions such as "anger," "sadness," and "joy."

[0773] The "risk score" is a numerical value that indicates the degree of risk of a message, calculated based on the extracted negative and dangerous words.

[0774] The "emotion score" is a numerical representation of the emotional state analyzed by the emotion engine.

[0775] A "pulse survey" is a questionnaire survey that is used to easily understand the current status of employees in a short period of time.

[0776] A "dashboard" is an interface that visually displays various data and allows for real-time monitoring of employees' mental state and the risk of information leaks.

[0777] "Escalation" refers to the process of notifying management of employees or situations deemed to be high risk when a set threshold is exceeded.

[0778] The present invention is a system for understanding the health and mental state of employees in a company in real time and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze message logs, extract and score negative and dangerous words, and further combines it with an emotion engine to recognize emotional states and perform comprehensive risk assessment. Specific embodiments for implementing the present invention are described below.

[0779] Server Operation

[0780] 1. Obtaining chat logs

[0781] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams) using an API to collect message data and metadata (sender, recipient, timestamp).

[0782] 2. Data Preprocessing

[0783] The server cleans up the chat logs by removing unnecessary information, using the BeautifulSoup library to remove HTML tags, regular expressions to remove special characters, and the langdetect library to unify the language.

[0784] 3. Input to the generative model

[0785] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3), which analyzes the text and extracts important features through the generative model's API (e.g., OpenAI GPT-3 API).

[0786] 4. Extraction of negative and dangerous words

[0787] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are saved along with the metadata.

[0788] 5. Calculating the risk score

[0789] The server calculates a risk score for each message based on the extracted words, taking into account the importance and frequency of each word, as well as the message context, to calculate an overall score.

[0790] 6. Emotion analysis using an emotion engine

[0791] The server uses an emotion engine (e.g., TextBlob, VADER) to analyze the emotional state of the detected messages, recognizing emotions such as "anger," "sadness," and "joy" and calculating an emotion score.

[0792] 7. Score tallying and escalation

[0793] The server aggregates the risk and emotion scores for each employee and calculates a cumulative risk rating. If the rating exceeds a set threshold, it determines who should be escalated and notifies them.

[0794] 8. Conducting a pulse survey

[0795] The server periodically prepares pulse surveys and generates questions for employees using the SurveyMonkey API, and the employee responses are reflected in the risk assessment.

[0796] Device behavior

[0797] 1. Sending chat logs

[0798] The devices (employee PCs and smartphones) send chat logs generated during daily work to the server in real time, allowing the server to always analyze the latest data.

[0799] 2. Pulse Survey Responses

[0800] The device sends a pulse survey notification to the employee, and when the employee responds, the results are sent to the server and reflected in the risk assessment.

[0801] User (administrator) actions

[0802] 1. Dashboard Monitoring

[0803] Users can use a dashboard provided by the server to monitor the mental state of employees and the risk of information leaks in real time. The dashboard uses data visualization tools such as Grafana and Tableau to display visual graphs and figures, allowing users to immediately grasp risks.

[0804] 2. Escalation

[0805] Users receive escalation notifications from the server and take necessary action. Notifications are sent via email or a dedicated app. For example, HR personnel can schedule a meeting with high-risk employees or arrange for specialized support.

[0806] Specific examples

[0807] For example, if Employee A frequently mentions "I can't stand my job anymore" in chat, the server collects the messages and uses the BERT model to detect the negative word "unbearable." At the same time, the VADER emotion engine recognizes strong emotions such as "anger" or "sadness" in the messages and calculates an emotion score. This gives the message a high risk score.

[0808] If the cumulative risk score exceeds the set threshold, the server sends an escalation notification to the HR department. HR personnel then schedule a meeting with Employee A and provide the necessary support. This system enables companies to detect risks of losing human resources or leaking information early and respond quickly and appropriately.

[0809] Prompt Sentence Examples

[0810] "Employee A frequently says, 'I can't take it anymore.' Extract negative words and sentiment scores from these messages and calculate a risk score."

[0811] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0812] Step 1:

[0813] The server periodically retrieves chat logs from the company's internal chat platforms (e.g., Slack, Microsoft Teams). To retrieve the logs, it uses the API provided by each platform to collect message data and metadata (sender, recipient, timestamp) via an authentication token. Using this, it sends an HTTP request to the API endpoint and receives chat log data in JSON format as a response.

[0814] Input: Authentication token obtained from the chat platform API

[0815] Output: Message data and metadata in JSON format

[0816] Specific operation: Sending an API request and receiving a response

[0817] Step 2:

[0818] The server cleans up the chat logs by removing unnecessary information and message texts. First, it uses the BeautifulSoup library to remove HTML tags, then it uses regular expressions to remove special characters, and it uses the langdetect library to detect the language of the text and unify it if necessary.

[0819] Input: Chat log data in JSON format

[0820] Output: Preprocessed message text

[0821] Specific actions: HTML tag removal, special character removal, language detection and unification

[0822] Step 3:

[0823] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3). Using the generative model's API (e.g., OpenAI GPT-3 API), it sends the message text to the model and receives the parsed results.

[0824] Input: Preprocessed message text

[0825] Output: Analysis results by generative model

[0826] Specific operation: Sending a request to the Generative Model API and receiving a response

[0827] Step 4:

[0828] The generative model extracts negative and dangerous words from the analysis results, matches them with predefined negative and dangerous word lists, and stores the detected words in a database along with metadata.

[0829] Input: Analysis results from the generative model

[0830] Output: Detected negative and dangerous words

[0831] Specific operation: Matching with word list, saving detection results to database

[0832] Step 5:

[0833] The server calculates a risk score for each message based on the extracted words. The server calculates an overall score by taking into account the importance and frequency of each word, as well as the message context. The server then uses a scoring algorithm to calculate the risk score and stores it in a database.

[0834] Input: Detected negative and dangerous words

[0835] Output: Risk score for each message

[0836] Specific operation: Calculate risk score and save it to the database

[0837] Step 6:

[0838] The server analyzes the emotional state of each message using an emotion engine (e.g., TextBlob, VADER). Through the emotion engine API, the server analyzes the message and obtains an emotion score such as "anger," "sadness," or "joy."

[0839] Input: Preprocessed message text

[0840] Output: Sentiment score from the sentiment engine

[0841] Specific operation: Sending a request to the Emotion Engine API and receiving a response

[0842] Step 7:

[0843] The server aggregates the risk and emotion scores for each employee and calculates a cumulative risk rating. If the rating exceeds a set threshold, it determines who should be escalated and notifies them. Ratings and notifications are performed automatically based on the information in the database.

[0844] Input: Risk score and sentiment score for each message

[0845] Output: Cumulative risk assessment and escalation notification

[0846] Specific actions: Calculating risk assessment, sending escalation notifications

[0847] Step 8:

[0848] The server periodically prepares pulse surveys and generates questions for employees. The surveys are created and administered using the SurveyMonkey API, etc. The survey results are also incorporated into risk assessments and are automatically saved in a database.

[0849] Input: Pulse survey execution instruction

[0850] Output: Pulse survey response results

[0851] Specific operations: Generate and send pulse surveys, save response results

[0852] (Application example 2)

[0853] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0854] The difficulty of understanding the health and mental state of employees in a company in real time and quickly detecting and responding to risks of losing human resources or information leaks is a major problem. In particular, extracting negative or dangerous words from employees' chat logs and daily communications to conduct risk assessments is cumbersome, and manual assessments are difficult to monitor efficiently and accurately. Furthermore, a systematic system is needed to comprehensively assess employees' mental and emotional states and escalate or take necessary action in a timely manner.

[0855] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for analyzing message logs using a generative model and extracting negative and dangerous words, means for calculating a risk score for messages based on the extracted words, means for aggregating risk scores for each employee and determining and notifying an escalation destination when a risk threshold is exceeded, means for analyzing the emotional state of the message using an emotion engine and reflecting the results in the risk score, and means for integrating the results of the pulse survey to perform a final risk assessment. This makes it possible to accurately grasp the health and mental state of employees in real time and efficiently perform risk assessment and escalation.

[0856] A "generative model" is a model that uses natural language processing technology to analyze message logs and extract important features and patterns.

[0857] A "message log" is a record of messages sent and received by employees on company chat platforms, etc.

[0858] "Negative words" are words with negative meanings, such as "painful" or "tired."

[0859] "Dangerous words" are words related to information leaks and risks, such as "confidential" and "leak."

[0860] A "risk score" is a risk index calculated based on the importance and frequency of occurrence of extracted negative and dangerous words.

[0861] An "emotion engine" is a technology for analyzing emotions such as "anger," "sadness," and "joy" from the content of a message.

[0862] A "pulse survey" is a survey that periodically asks simple questions to understand employees' health and mental state.

[0863] "Escalation" is the process of notifying higher-level management when a certain risk score is exceeded and urging them to take action.

[0864] A "dashboard" is a tool that visually monitors and displays employees' mental and risk status in real time.

[0865] This invention is a system for understanding the health and mental state of employees in a company in real time and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze chat logs, extracting and scoring negative and dangerous words, and combining it with an emotion engine to recognize emotional states and perform comprehensive risk assessments. Furthermore, by integrating pulse survey results and displaying them on a dashboard in real time, it is possible to accurately grasp the overall health status of employees.

[0866] Server Processing

[0867] The server automatically collects message data and metadata (sender, recipient, timestamp) from the company's internal chat platform (e.g., a general messaging system) using an API to periodically retrieve chat logs. It then removes unnecessary information from the collected chat logs and cleans up the message text. Specifically, it performs processes such as removing HTML tags, deleting special characters, and standardizing different languages.

[0868] The preprocessed chat logs are input into a generative model (e.g., BERT, GPT-3), which uses natural language processing techniques to analyze each message and extract important features based on the text content. Negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") are detected from the analysis results. The detected words are saved along with metadata.

[0869] Next, a risk score is calculated for each message based on the detected words, taking into account the importance and frequency of the words, as well as the message context. Furthermore, an emotion engine is used to analyze the emotional state of the detected message. For example, the emotion engine can read emotions from the message, recognizing emotions such as "anger," "sadness," and "joy," and then calculate an emotion score.

[0870] The server aggregates the risk and emotion scores for each employee and calculates the employee's cumulative risk rating. If this rating exceeds a set threshold, it determines who should be escalated and notifies them.

[0871] Additionally, the server periodically prepares pulse surveys and generates questions for employees. The results of the survey are also incorporated into risk assessments to understand the employee's overall risk status. For example, questions such as, "Have you been feeling stressed at work recently?"

[0872] Terminal handling

[0873] Employee devices (PCs and smartphones) send chat logs generated during daily work to a server in real time. This data is used for the analysis process described below. The data sent from the device goes through the processing steps described above on the server and is used for risk assessment. The device also sends pulse survey notifications to employees, who respond to these. The responses are sent to the server and reflected in the risk assessment.

[0874] User operations

[0875] Administrators use a dashboard provided by the server to monitor employees' mental state and the risk of information leaks in real time. The dashboard displays visual graphs and figures, allowing them to immediately grasp the risk status. Upon receiving an escalation notification, administrators can take the necessary countermeasures, such as arranging a meeting with the employee or arranging for professional support.

[0876] Specific examples

[0877] For example, suppose Employee A frequently mentions in chat that he or she "can't stand the work anymore." In this case, the server collects these messages and uses a generative model to detect the negative word "can't stand it." The emotion engine then recognizes strong emotions such as "anger" or "sadness" from these messages and calculates an emotion score based on that. This gives the message a high risk score.

[0878] If the cumulative risk score exceeds the threshold, the server sends an escalation notification to the management department. The manager then schedules a meeting with Employee A and provides the necessary support. In this way, the system of the present invention enables companies to detect risks of losing human resources or leaking information early and respond quickly and appropriately.

[0879] Prompt Sentence Examples

[0880] "Employee A says: 'I'm at my limit in this job.' Please analyze the emotional state of this message and calculate an appropriate risk score."

[0881] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0882] Step 1:

[0883] The server periodically retrieves chat logs from the company's internal chat platform. For example, it uses an API to automatically collect message data and metadata (sender, recipient, timestamp) from the company's internal messaging system. The input is the chat platform's API endpoint, and the output is chat log data in JSON format.

[0884] Step 2:

[0885] The server removes unnecessary information from the collected chat logs and cleans up the message text. Specifically, it performs processes such as removing HTML tags, deleting special characters, and unifying different languages. The input is the chat log data from the previous step, and the output is the cleaned message text in the target language.

[0886] Step 3:

[0887] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3). The model uses natural language processing techniques to analyze each message and extract important features based on the text content. The input is the cleaned message text from the previous step, and the output is the feature-extracted data.

[0888] Step 4:

[0889] The server detects negative and risky words from the analysis results of the generative model. The detected words are saved along with the associated metadata. The input is the feature extraction data from the previous step, and the output is a list of detected negative and risky words.

[0890] Step 5:

[0891] The server calculates a risk score for each message based on the words detected, taking into account the importance and frequency of the words, as well as the message context. The input is the list of negative and dangerous words from the previous step, and the output is a risk score for each message.

[0892] Step 6:

[0893] The server uses an emotion engine to analyze the emotional state of the message. Specifically, it reads emotions from the message, recognizes emotions such as "anger," "sadness," and "joy," and calculates an emotion score. The input is the message to which the risk score from the previous step has been assigned, and the output is the emotion score.

[0894] Step 7:

[0895] The server aggregates the risk and sentiment scores for each employee and calculates the employee's cumulative risk rating. If this rating exceeds a set threshold, it determines who should be escalated and notifies them. The input is each employee's risk and sentiment score, and the output is an escalation notification.

[0896] Step 8:

[0897] The server periodically prepares pulse surveys and generates questions for employees. The survey results are also incorporated into the risk assessment. The inputs are the survey questions and employee responses, and the output is an updated risk assessment.

[0898] Step 9:

[0899] The terminals send chat logs generated during employees' daily work to the server in real time. This data is used for the analysis process described above. The input is the chat messages generated on the terminals, and the output is the data sent to the server.

[0900] Step 10:

[0901] The terminal sends a pulse survey notification to the employee, who then responds. The response results are sent to the server and reflected in the risk assessment. The input is the response of the employee who received the survey notification on the terminal, and the output is the survey data sent to the server.

[0902] Step 11:

[0903] Users use a dashboard provided by the server to monitor the mental state of employees and the risk of information leakage in real time. The dashboard displays visual graphs and numerical values, allowing users to immediately grasp the risk status. The input is real-time risk data received from the server, and the output is the visualized dashboard screen.

[0904] Step 12:

[0905] The user receives the escalation notification and takes necessary measures, such as setting up a meeting with an employee or arranging for professional support. The input is the escalation notification from the server, and the output is the implemented measures.

[0906] The specific processing unit 290 transmits the result of the specific processing to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[0907] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0908] In the above embodiment, an example in which the specific processing is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the smart glasses 214.

[0909] [Third embodiment]

[0910] FIG. 5 shows an example of the configuration of a data processing system 310 according to the third embodiment.

[0911] 5, the data processing system 310 includes the data processing device 12 and a headset type terminal 314. An example of the data processing device 12 is a server.

[0912] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0913] The headset type terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, the camera 42, and the display 343 are also connected to the bus 52.

[0914] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[0915] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[0916] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 control the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[0917] Fig. 6 shows an example of the main functions of the data processing device 12 and the headset type terminal 314. As shown in Fig. 6, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[0918] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0919] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0920] In the headset type terminal 314, a reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0921] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as the "server" and the headset type terminal 314 will be referred to as the "terminal."

[0922] This invention is a system for understanding the health and mental state of employees in a company in real time, and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze chat logs, extract and score negative and dangerous words, and further integrates the results of pulse surveys to perform a final risk assessment.

[0923] Program processing (natural language explanation)

[0924] server

[0925] 1. Collecting chat logs

[0926] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams). Automatically collecting this data using an API enables timely monitoring.

[0927] 2. Log Analysis

[0928] The server inputs the collected chat logs into a generative model, which analyzes and extracts negative or dangerous words. The generative model then uses natural language processing technology to analyze the message content and detect specific words and phrases.

[0929] 3. Calculating the risk score

[0930] The server calculates a risk score for each message based on the words detected, for example, assigning a medium risk score to negative words like "painful" and a high risk score to dangerous words like "confidentiality leak."

[0931] 4. Score tallying and escalation

[0932] The server compiles risk scores for each employee and escalates when the cumulative score exceeds a certain threshold. The escalation destination is automatically determined based on the level of risk and notified.

[0933] 5. Conducting a pulse survey

[0934] The server periodically prepares pulse surveys and generates questions for employees, and the survey results are also incorporated into risk assessments to understand the overall risk status of employees.

[0935] Terminal

[0936] 1. Sending chat logs

[0937] The devices (employee PCs and smartphones) send chat logs generated during daily work to the server in real time. This data is used for the analysis process described below.

[0938] 2. Pulse Survey Responses

[0939] The device sends a pulse survey notification to the employee, who then responds. The response results are sent to the server and reflected in the risk assessment.

[0940] User (Administrator)

[0941] 1. Dashboard Monitoring

[0942] Users can use a dashboard provided by the server to monitor the mental state of employees and the risk of information leaks in real time. The dashboard provides visual information, allowing users to immediately check whether there is any risk.

[0943] 2. Escalation

[0944] Users receive escalation notifications from the server and can take action, such as arranging for specialized support for employees with high risk scores or reporting the situation to their immediate supervisor.

[0945] Specific examples

[0946] For example, suppose Employee A frequently mentions in chat that he or she "can't stand the work anymore." In this case, the server collects these messages and uses a generative model to detect the negative word "can't stand it." It then assigns a higher risk score to the message. If the cumulative risk score exceeds a threshold, the server sends an escalation notification to the HR department. The HR staff then schedules a meeting with Employee A and provides the necessary support.

[0947] In this way, the system of the present invention allows companies to detect risks of losing human resources or leaking information at an early stage and take prompt and appropriate measures. Furthermore, by integrating the results of periodic pulse surveys, it is possible to more accurately grasp the overall health status of employees.

[0948] The processing flow will be explained below.

[0949] Step 1:

[0950] server

[0951] Get chat logs

[0952] The server periodically collects all chat messages using the API of the company's internal chat platform (e.g., Slack, Microsoft Teams), along with metadata such as the message sender, recipient, and timestamp.

[0953] Step 2:

[0954] server

[0955] Data Preprocessing

[0956] The server cleans up the chat logs by removing unnecessary information, such as removing HTML tags, removing special characters, and standardizing different languages.

[0957] Step 3:

[0958] server

[0959] Input to the generative model

[0960] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3), which uses natural language processing techniques to analyze each message and extract important features based on the text content.

[0961] Step 4:

[0962] server

[0963] Extracting negative and dangerous words

[0964] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are saved along with the metadata.

[0965] Step 5:

[0966] server

[0967] Calculating the risk score

[0968] The server calculates a risk score for each message based on the negative and dangerous words detected, taking into account the importance and frequency of the words, as well as the message context.

[0969] Step 6:

[0970] server

[0971] Score tally

[0972] The server aggregates the risk scores for each employee to calculate a cumulative risk score for that employee, which is updated over time to keep it up to date.

[0973] Step 7:

[0974] server

[0975] Escalation decision

[0976] The server determines whether escalation is necessary when the cumulative risk score exceeds a pre-defined threshold, and if so, determines the appropriate escalation point (e.g., HR department, direct manager).

[0977] Step 8:

[0978] server

[0979] Escalation Notification

[0980] The server sends an escalation notification, which includes the employee's risk score, any negative or dangerous words detected, and recommended actions to take.

[0981] Step 9:

[0982] Terminal

[0983] Conducting a pulse survey

[0984] The terminal receives notifications of pulse surveys sent from the server and periodically asks employees to answer the pulse surveys. Employees answer the questions through the terminal, and the results are sent to the server.

[0985] Step 10:

[0986] server

[0987] Integration of pulse survey results

[0988] The server combines the pulse survey results with the risk score to create an overall risk assessment for the employee, which also influences escalation decisions.

[0989] Step 11:

[0990] User

[0991] Dashboard Monitoring

[0992] The user (administrator) uses a dashboard provided by the server to monitor the mental state of employees and the risk of information leakage in real time. The dashboard displays visual graphs and numerical values, allowing the user to immediately grasp the risk status.

[0993] Step 12:

[0994] User

[0995] Escalation

[0996] The user receives escalation notifications from the server and can take necessary action, such as arranging a meeting with an employee or arranging for professional support.

[0997] Example 1

[0998] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[0999] To prevent employees' mental health from deteriorating and information leaks from occurring, a system that monitors employee behavior in real time and takes appropriate action is required. However, conventional systems lacked sufficient chat log analysis, making it difficult to quickly detect and respond to risks. Furthermore, there was a lack of tools to grasp employees' overall status, making it difficult to provide support in a timely manner.

[1000] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[1001] In this invention, the server includes a means for analyzing communication logs using a generative model and extracting negative and dangerous expressions, a means for calculating a risk score for communication based on the extracted expressions, and a means for aggregating risk scores for each employee and determining and notifying notification recipients when a risk threshold is exceeded. This allows for real-time monitoring of employee mental state and information leakage risk, enabling timely escalation and response. Furthermore, by conducting periodic simple surveys and integrating the results with risk scores, more accurate risk assessments are possible.

[1002] A "generative model" is an algorithm that uses natural language processing technology to analyze text data and extract or generate specific information.

[1003] "Communication logs" refer to data that records the communication history of employees on chat platforms, messengers, etc. used during work hours.

[1004] "Negative expressions" are words or phrases that employees use to express feelings of stress, dissatisfaction, anxiety, etc.

[1005] "Dangerous expressions" are words or phrases that suggest risks such as information leaks or compliance violations.

[1006] A "risk score" is an index that quantifies the risk of communication content based on the frequency and importance of negative or dangerous expressions.

[1007] A "simple survey" is a questionnaire survey conducted over a short period of time to understand the mental state and awareness of employees.

[1008] The "information display screen" is an interface that visually displays employees' mental state and the risk of information leakage, and allows for real-time monitoring.

[1009] "Escalation" is the process of notifying the person or department that needs to take action when the risk score exceeds a certain threshold.

[1010] This invention is a system that monitors the mental state of employees and the risk of information leakage in real time within a company, and analyzes communication logs containing negative or dangerous words. This system includes functions such as analyzing communication logs using a generative model, calculating and aggregating risk scores, notifying escalation at appropriate times, conducting periodic simple surveys, and providing an information display screen.

[1011] Hardware and Software

[1012] Server: The server is the main processing component, analyzing communication logs using generative AI models (e.g., GPT-4), maintaining risk scores using a database, and sending escalation notifications using an email service.

[1013] Device: The device is an employee's PC or smartphone, and is used to send chat logs and answer simple surveys.

[1014] User (Administrator): Administrators use the information display screen to monitor employees' mental state and the risk of information leakage, and respond to escalations.

[1015] Program processing

[1016] Server Processing

[1017] 1. Collecting chat logs:

[1018] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams) via API, performs API authentication, and collects messages from the required channels and users.

[1019] Examples:

[1020] OAuth2 is used for API authentication, and the / channels.history endpoint is called to retrieve messages for a specific channel. The retrieved data is saved in JSON format.

[1021] 2. Log analysis:

[1022] The server inputs the collected chat logs into a generative AI model to analyze and extract negative and dangerous words. It then tokenizes the log data to generate prompt sentences and requests the generative model to analyze them.

[1023] Examples:

[1024] Generate a prompt such as "Please parse the following message: 'I can no longer stand the work.'"

[1025] 3. Calculating the risk score:

[1026] The server calculates a risk score based on the analysis results, assigning a score based on specific keywords to quantify the risk of each message.

[1027] Examples:

[1028] The keyword "painful" is assigned a medium risk score (5 points), and the keyword "confidentiality leak" is assigned a high risk score (10 points).

[1029] 4. Score tallying and escalation:

[1030] The server compiles risk scores for each employee, and if the cumulative risk score exceeds a certain threshold, it uses an email service to send an escalation notification to the HR department or other department.

[1031] Terminal handling

[1032] 1. Sending chat logs:

[1033] The device sends the employee's chat log in real time to a server, including all messages generated during work hours.

[1034] 2. Quick survey answers:

[1035] The device sends a simple survey notification to the employee, who then answers it. The results are sent to a server and used for risk assessment.

[1036] User Action

[1037] 1. Information display screen monitoring:

[1038] Managers can use the information display screen to monitor employees' mental health and the risk of information leaks, with risk scores and warnings displayed in real time.

[1039] 2. Escalation Response:

[1040] Administrators receive escalation notifications from the server and take appropriate action, such as suggesting interviews or counseling for employees with high risk scores.

[1041] In this way, the system of the present invention allows for real-time monitoring of employees' mental state and information leakage risks, enabling prompt and appropriate responses. Furthermore, comprehensive risk assessments linked to periodic simple surveys allow for a more accurate understanding of employees' health conditions.

[1042] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1043] Step 1: Collect chat logs

[1044] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams) via API. The input is messages from specific channels and users. The output is a JSON-formatted file containing the retrieved chat log data. Specifically, the server performs API authentication using OAuth2. It then calls the / channels.history endpoint to retrieve messages from specific channels and saves them in JSON format.

[1045] Step 2: Analyze the logs

[1046] The server inputs the collected chat logs into a generative AI model to analyze and extract negative and dangerous words. The input is a JSON file of the saved chat logs. The output is a list of negative and dangerous words as the analysis results. Specifically, the server reads the log data and tokenizes each message. It then inputs the generated prompt sentences into the generative AI model and obtains the analysis results from the model.

[1047] Step 3: Calculate the risk score

[1048] The server calculates a risk score based on the analysis results. The input is the analysis results from the generative AI model. The output is a risk score for each message. Specifically, the server lists the detected keywords, references the risk scores corresponding to each keyword, and calculates the total risk score. For example, the keyword "painful" is assigned 5 points, and the keyword "confidentiality leak" is assigned 10 points.

[1049] Step 4: Score aggregation and escalation

[1050] The server aggregates the risk scores for each employee and starts the escalation process if a certain threshold is exceeded. The input is the calculated risk score. The output is the sending of an escalation notification. Specifically, the server aggregates the risk scores for each employee from the database. If the cumulative score exceeds the threshold, it sets an escalation flag, determines the escalation destination, and sends the notification using an email service.

[1051] Step 5: Conduct a pulse survey

[1052] The server periodically prepares pulse surveys and generates questions for employees. The input is a list of employees and the results of past pulse surveys. The output is the generated pulse survey questions and their results. Specifically, the server automatically generates survey questions, creates a survey form, and sends it to each employee's email address. The collected survey responses are analyzed and reflected in the risk assessment.

[1053] Step 6: Monitoring the Dashboard

[1054] Users use a dashboard provided by the server to monitor employees' mental state and information leakage risks. The input is the risk score and pulse survey results from the server. The output is real-time risk information displayed on the dashboard. Specifically, users check the dashboard to visually understand the scores and warnings for each employee.

[1055] (Application example 1)

[1056] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1057] Companies need to understand the mental and physical health of their employees in real time, and quickly detect and respond to risks of losing talent or leaking information. However, current systems often have delays in analyzing chat logs and pulse survey results, making it difficult to detect risks early.

[1058] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[1059] In this invention, the server includes means for analyzing message logs using a generative model and extracting negative and dangerous words, means for calculating risk scores for messages based on the extracted words, means for aggregating risk scores for each employee and determining and notifying an escalation destination when a risk threshold is exceeded, means for collecting data using the chat platform's API, means for periodically notifying employees of pulse survey questions and collecting their responses, and means for integrating the collected data with the risk scores to make a final risk assessment. This allows for a real-time understanding of employees' mental and health states, enabling early detection of risks and rapid response.

[1060] A "generative model" is an AI model that uses natural language processing technology to analyze message logs and detect specific words and phrases.

[1061] A "message log" is historical data of messages exchanged on a company's internal chat platform.

[1062] "Negative words" are specific words that may indicate an employee's mental or physical health is deteriorating.

[1063] "Dangerous words" are specific words that are likely to lead to information leaks or disrupt business operations.

[1064] A "risk score" is a numerical representation of the risk level of a message based on the negative or dangerous words contained in the message.

[1065] "Escalation" is a notification process that prompts support personnel or administrators to take action when a detected risk exceeds a certain threshold.

[1066] A "chat platform API" is an application program interface for obtaining data from chat systems used by companies.

[1067] A "pulse survey" is a brief questionnaire survey conducted to periodically check the mental and physical state of employees.

[1068] A "dashboard" is an interface that visually displays employees' mental state and the risk of information leaks in real time.

[1069] The system for implementing the present invention comprises the following steps.

[1070] server

[1071] The server periodically collects message logs using the API of the company's internal chat platform (e.g., Slack or Microsoft Teams). The collected message logs are then analyzed using a generative AI model (e.g., OpenAI GPT-4 or Hugging Face Transformers). During this analysis, negative and dangerous words are extracted from the messages, and a risk score is calculated for each message.

[1072] The server also aggregates risk scores for each employee, and if the cumulative risk score exceeds a set threshold, it automatically determines the appropriate escalation and notifies the appropriate person. The server also periodically sends pulse surveys to employees and collects their responses. The results of the pulse surveys are integrated with existing risk scores and used in the final risk assessment.

[1073] Furthermore, the server provides managers with a dashboard that visualizes the mental state of employees and the risk of information leakage in real time, allowing them to monitor the risk situation and take necessary measures promptly.

[1074] Terminal

[1075] Employee devices (smartphones and PCs) send chat logs generated during daily work to a server in real time. This data is used for the analysis process described below. The devices also receive survey notifications, and employees respond to pulse surveys via their devices. Responses are sent to the server and reflected in risk assessments.

[1076] User (Administrator)

[1077] Administrators use a dashboard provided by the server to monitor employees' mental health and information leakage risks in real time. The dashboard provides visual information, allowing them to immediately check whether there is any risk. If an escalation occurs, administrators receive a notification from the server and take appropriate countermeasures, such as arranging for professional support for employees with high risk scores.

[1078] Specific examples

[1079] For example, if Employee A frequently says in chat that he or she "can't stand the work anymore," the server collects those messages and uses a generative model to detect the negative word "can't stand it." The server assigns a high risk score to those messages, and if the cumulative risk score exceeds a threshold, the server sends an escalation notification to the administrator. The administrator receives the escalation notification, schedules a meeting with Employee A, and provides the necessary support.

[1080] Prompt Sentence Examples

[1081] "Please list the negative and dangerous words frequently used by employees in chat, calculate a risk score for each, and also perform a sentiment analysis on those words, outputting the results in JSON format."

[1082] In this way, the server, terminals, and administrators work together to grasp the mental and health status of employees in real time, creating a system that can detect risks early and respond quickly.

[1083] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1084] Step 1:

[1085] The server periodically collects message logs using the API of the company's chat platform. The data obtained via the API is in JSON format and includes the date and time, sender information, and message content.

[1086] Input: Message logs obtained from the chat platform's API

[1087] Output: Message log in JSON format, including date, time, sender information, and message content

[1088] Step 2:

[1089] The server inputs the collected message logs into a generative AI model to analyze the message content. The generative model uses natural language processing technology to extract negative and dangerous words from the messages.

[1090] Input: Message log in JSON format

[1091] Output: Message log with negative and dangerous words extracted

[1092] Step 3:

[1093] The server calculates a risk score for each message based on the extracted negative and dangerous words, for example, assigning a medium risk score to "painful" and a high risk score to "confidentiality leak."

[1094] Input: Message log from which negative and dangerous words were extracted

[1095] Output: Message log with risk score assigned

[1096] Step 4:

[1097] The server aggregates the risk scores for each employee and determines the escalation destination when the cumulative risk score exceeds a set threshold. The server automatically notifies the escalation destination.

[1098] Input: Message logs with risk scores

[1099] Output: Escalation notification

[1100] Step 5:

[1101] The server periodically sends pulse survey questions to employees and collects their responses, which are also included in the risk assessment.

[1102] Input: Pulse Survey Question Notification

[1103] Output: Employee pulse survey responses

[1104] Step 6:

[1105] The server combines the collected pulse survey results with existing risk scores to produce a final risk assessment.

[1106] Input: Pulse survey results and risk score

[1107] Output: Final integrated risk assessment

[1108] Step 7:

[1109] The server provides a dashboard to administrators, which monitors and displays the mental state of employees and the risk of information leakage in real time. Administrators can check the risk status through the dashboard and take appropriate measures.

[1110] Input: Final Integrated Risk Assessment

[1111] Output: Real-time monitoring via dashboard

[1112] Step 8:

[1113] The device sends chat platform message logs and pulse survey responses to the server in real time, ensuring that the latest data is always used for analysis and evaluation.

[1114] Input: Employee chat logs, pulse survey responses

[1115] Output: Real-time updated data

[1116] Step 9:

[1117] Administrators receive escalation notifications from the server and take appropriate action for employees with high risk scores, such as arranging psychological support or scheduling a meeting with the employee.

[1118] Input: Escalation notification

[1119] Output: Implementing countermeasures

[1120] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[1121] This invention is a system for understanding the health and mental state of employees in a company in real time and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze chat logs, extracting and scoring negative and dangerous words, and combining it with an emotion engine to recognize emotional states and perform comprehensive risk assessments. Furthermore, by integrating pulse survey results and displaying them on a dashboard in real time, it is possible to accurately grasp the overall health status of employees.

[1122] Program processing (natural language explanation)

[1123] server

[1124] 1. Obtaining chat logs

[1125] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams), automatically collecting message data and metadata (sender, recipient, timestamp) using an API.

[1126] 2. Data Preprocessing

[1127] The server cleans up the chat logs by removing unnecessary information, such as removing HTML tags, removing special characters, and standardizing different languages.

[1128] 3. Input to the generative model

[1129] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3), which uses natural language processing techniques to analyze each message and extract important features based on the text content.

[1130] 4. Extraction of negative and dangerous words

[1131] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are saved along with the metadata.

[1132] 5. Calculating the risk score

[1133] The server calculates a risk score for each message based on the words detected, taking into account the importance and frequency of the words, as well as the message context.

[1134] 6. Emotion analysis using an emotion engine

[1135] The server uses an emotion engine to analyze the emotional state of the detected message. For example, the emotion engine reads emotions from the message, recognizes emotions such as "anger," "sadness," and "joy," and calculates an emotion score.

[1136] 7. Score tallying and escalation

[1137] The server aggregates the risk and sentiment scores for each employee and calculates the cumulative risk rating for that employee. If this rating exceeds a set threshold, it determines who should be escalated and notifies them.

[1138] 8. Conducting a pulse survey

[1139] The server periodically prepares pulse surveys and generates questions for employees, and the survey results are also incorporated into risk assessments to understand the overall risk status of employees.

[1140] Terminal

[1141] 1. Sending chat logs

[1142] The devices (employee PCs and smartphones) send chat logs generated during daily work to the server in real time. This data is used for the analysis process described below.

[1143] 2. Pulse Survey Responses

[1144] The device sends a pulse survey notification to the employee, who then responds. The response results are sent to the server and reflected in the risk assessment.

[1145] User (Administrator)

[1146] 1. Dashboard Monitoring

[1147] Users can use a dashboard provided by the server to monitor the mental state of employees and the risk of information leaks in real time. The dashboard displays visual graphs and numerical values, allowing users to immediately grasp the risk status.

[1148] 2. Escalation

[1149] The user receives escalation notifications from the server and can take necessary action, such as arranging a meeting with an employee or arranging for professional support.

[1150] Specific examples

[1151] For example, suppose Employee A frequently mentions in chat that he or she "can't stand the work anymore." In this case, the server collects these messages and uses a generative model to detect the negative word "can't stand it." The emotion engine then recognizes strong emotions such as "anger" or "sadness" from these messages and calculates an emotion score based on that. This gives the message a high risk score.

[1152] If the cumulative risk score exceeds the threshold, the server sends an escalation notification to the HR department. HR personnel then schedule a meeting with Employee A and provide the necessary support. In this way, the system of the present invention enables companies to detect risks of losing human resources or leaking information early and respond quickly and appropriately. Furthermore, by combining it with an emotion engine, it is possible to more accurately grasp the emotional state of employees and achieve more effective risk management.

[1153] The processing flow will be explained below.

[1154] Step 1:

[1155] server

[1156] Get chat logs

[1157] The server periodically collects all chat messages using the API of the company's internal chat platform (e.g., Slack, Microsoft Teams), including metadata such as message sender, recipient, and timestamp.

[1158] Step 2:

[1159] server

[1160] Data Preprocessing

[1161] The server cleans the collected chat logs of unnecessary information, including removing HTML tags, special characters, and language uniformity. The preprocessed data is then prepared for analysis.

[1162] Step 3:

[1163] server

[1164] Input to the generative model

[1165] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3), which uses natural language processing techniques to analyze each message and extract important features based on its content.

[1166] Step 4:

[1167] server

[1168] Extracting negative and dangerous words

[1169] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are then saved along with the metadata.

[1170] Step 5:

[1171] server

[1172] Calculating the risk score

[1173] The server calculates a risk score for each message based on the words detected, for example, assigning a medium risk score to the negative word "painful" and a high risk score to the dangerous word "confidentiality leak."

[1174] Step 6:

[1175] server

[1176] Emotion analysis using an emotion engine

[1177] The server uses an emotion engine to analyze the emotional state of the message. Specifically, the emotion engine recognizes emotions such as "anger," "sadness," and "joy," and calculates an emotion score based on that.

[1178] Step 7:

[1179] server

[1180] Score tally

[1181] The server aggregates the risk and sentiment scores for each employee to calculate a cumulative risk rating, which is updated periodically to keep the rating up to date.

[1182] Step 8:

[1183] server

[1184] Escalation decision

[1185] If the cumulative risk assessment exceeds the set threshold, the server determines that escalation is necessary, determines the escalation destination (e.g., HR department, direct supervisor), and notifies the appropriate person.

[1186] Step 9:

[1187] server

[1188] Escalation Notification

[1189] The server then sends a notification to the determined escalation point, which includes the employee's risk score, any negative or dangerous words detected, and recommended actions to take.

[1190] Step 10:

[1191] Terminal

[1192] Conducting a pulse survey

[1193] The terminal displays the pulse survey notification received from the server to the employee, periodically prompting them to answer the survey. The employee answers the survey on the terminal and sends the answer data to the server.

[1194] Step 11:

[1195] server

[1196] Integration of pulse survey results

[1197] The server integrates the collected pulse survey results into a risk assessment, providing a more accurate picture of an employee's overall risk status based on this integrated data.

[1198] Step 12:

[1199] User

[1200] Dashboard Monitoring

[1201] Users (administrators) use a dashboard provided by the server to monitor the mental state of employees and the risk of information leakage in real time. The dashboard displays intuitive graphs and indicators, allowing users to instantly check the risk status.

[1202] Step 13:

[1203] User

[1204] Escalation

[1205] The user receives escalation notifications from the server and can take necessary measures, such as arranging a meeting with an employee or arranging for professional support.

[1206] Example 2

[1207] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1208] Understanding the mental and physical state of employees within a company is an important issue in managing the risks of losing talent and leaking information. However, conventional methods have made it difficult to comprehensively evaluate this in real time, making it difficult to respond in a timely manner. There is a need for a method to accurately and quickly understand employees' mental state and to detect and respond to the risks of losing talent and leaking information early on.

[1209] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[1210] In this invention, the server includes means for analyzing message logs using a generative model and extracting negative and dangerous words, means for calculating a risk score for messages based on the extracted words, means for analyzing the emotional state of messages using an emotion engine and calculating an emotion score, and means for aggregating the risk and emotion scores for each employee and determining and notifying an escalation destination when a risk threshold is exceeded. This makes it possible to accurately evaluate the mental state of employees in real time and to detect and respond to risks of human resource outflow and information leaks at an early stage.

[1211] A "generative model" is an algorithm that uses natural language processing technology to analyze text data and extract important features.

[1212] "Message Log" refers to message data and metadata obtained from an enterprise's internal chat platform.

[1213] "Negative words" refers to a group of words that suggest an employee's mental state is deteriorating.

[1214] "Dangerous words" refers to a group of words that pose a high risk of information leakage or security risk.

[1215] An "emotion engine" refers to technology that analyzes emotional states from text data and recognizes emotions such as "anger," "sadness," and "joy."

[1216] The "risk score" is a numerical value that indicates the degree of risk of a message, calculated based on the extracted negative and dangerous words.

[1217] The "emotion score" is a numerical representation of the emotional state analyzed by the emotion engine.

[1218] A "pulse survey" is a questionnaire survey that is used to easily understand the current status of employees in a short period of time.

[1219] A "dashboard" is an interface that visually displays various data and allows for real-time monitoring of employees' mental state and the risk of information leaks.

[1220] "Escalation" refers to the process of notifying management of employees or situations deemed to be high risk when a set threshold is exceeded.

[1221] The present invention is a system for understanding the health and mental state of employees in a company in real time and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze message logs, extract and score negative and dangerous words, and further combines it with an emotion engine to recognize emotional states and perform comprehensive risk assessment. Specific embodiments for implementing the present invention are described below.

[1222] Server Operation

[1223] 1. Obtaining chat logs

[1224] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams) using an API to collect message data and metadata (sender, recipient, timestamp).

[1225] 2. Data Preprocessing

[1226] The server cleans up the chat logs by removing unnecessary information, using the BeautifulSoup library to remove HTML tags, regular expressions to remove special characters, and the langdetect library to unify the language.

[1227] 3. Input to the generative model

[1228] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3), which analyzes the text and extracts important features through the generative model's API (e.g., OpenAI GPT-3 API).

[1229] 4. Extraction of negative and dangerous words

[1230] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are saved along with the metadata.

[1231] 5. Calculating the risk score

[1232] The server calculates a risk score for each message based on the extracted words, taking into account the importance and frequency of each word, as well as the message context, to calculate an overall score.

[1233] 6. Emotion analysis using an emotion engine

[1234] The server uses an emotion engine (e.g., TextBlob, VADER) to analyze the emotional state of the detected messages, recognizing emotions such as "anger," "sadness," and "joy" and calculating an emotion score.

[1235] 7. Score tallying and escalation

[1236] The server aggregates the risk and emotion scores for each employee and calculates a cumulative risk rating. If the rating exceeds a set threshold, it determines who should be escalated and notifies them.

[1237] 8. Conducting a pulse survey

[1238] The server periodically prepares pulse surveys and generates questions for employees using the SurveyMonkey API, and the employee responses are reflected in the risk assessment.

[1239] Device behavior

[1240] 1. Sending chat logs

[1241] The devices (employee PCs and smartphones) send chat logs generated during daily work to the server in real time, allowing the server to always analyze the latest data.

[1242] 2. Pulse Survey Responses

[1243] The device sends a pulse survey notification to the employee, and when the employee responds, the results are sent to the server and reflected in the risk assessment.

[1244] User (administrator) actions

[1245] 1. Dashboard Monitoring

[1246] Users can use a dashboard provided by the server to monitor the mental state of employees and the risk of information leaks in real time. The dashboard uses data visualization tools such as Grafana and Tableau to display visual graphs and figures, allowing users to immediately grasp risks.

[1247] 2. Escalation

[1248] Users receive escalation notifications from the server and take necessary action. Notifications are sent via email or a dedicated app. For example, HR personnel can schedule a meeting with high-risk employees or arrange for specialized support.

[1249] Specific examples

[1250] For example, if Employee A frequently mentions "I can't stand my job anymore" in chat, the server collects the messages and uses the BERT model to detect the negative word "unbearable." At the same time, the VADER emotion engine recognizes strong emotions such as "anger" or "sadness" in the messages and calculates an emotion score. This gives the message a high risk score.

[1251] If the cumulative risk score exceeds the set threshold, the server sends an escalation notification to the HR department. HR personnel then schedule a meeting with Employee A and provide the necessary support. This system enables companies to detect risks of losing human resources or leaking information early and respond quickly and appropriately.

[1252] Prompt Sentence Examples

[1253] "Employee A frequently says, 'I can't take it anymore.' Extract negative words and sentiment scores from these messages and calculate a risk score."

[1254] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1255] Step 1:

[1256] The server periodically retrieves chat logs from the company's internal chat platforms (e.g., Slack, Microsoft Teams). To retrieve the logs, it uses the API provided by each platform to collect message data and metadata (sender, recipient, timestamp) via an authentication token. Using this, it sends an HTTP request to the API endpoint and receives chat log data in JSON format as a response.

[1257] Input: Authentication token obtained from the chat platform API

[1258] Output: Message data and metadata in JSON format

[1259] Specific operation: Sending an API request and receiving a response

[1260] Step 2:

[1261] The server cleans up the chat logs by removing unnecessary information and message texts. First, it uses the BeautifulSoup library to remove HTML tags, then it uses regular expressions to remove special characters, and it uses the langdetect library to detect the language of the text and unify it if necessary.

[1262] Input: Chat log data in JSON format

[1263] Output: Preprocessed message text

[1264] Specific actions: HTML tag removal, special character removal, language detection and unification

[1265] Step 3:

[1266] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3). Using the generative model's API (e.g., OpenAI GPT-3 API), it sends the message text to the model and receives the parsed results.

[1267] Input: Preprocessed message text

[1268] Output: Analysis results by generative model

[1269] Specific operation: Sending a request to the Generative Model API and receiving a response

[1270] Step 4:

[1271] The generative model extracts negative and dangerous words from the analysis results, matches them with predefined negative and dangerous word lists, and stores the detected words in a database along with metadata.

[1272] Input: Analysis results from the generative model

[1273] Output: Detected negative and dangerous words

[1274] Specific operation: Matching with word list, saving detection results to database

[1275] Step 5:

[1276] The server calculates a risk score for each message based on the extracted words. The server calculates an overall score by taking into account the importance and frequency of each word, as well as the message context. The server then uses a scoring algorithm to calculate the risk score and stores it in a database.

[1277] Input: Detected negative and dangerous words

[1278] Output: Risk score for each message

[1279] Specific operation: Calculate risk score and save it to the database

[1280] Step 6:

[1281] The server analyzes the emotional state of each message using an emotion engine (e.g., TextBlob, VADER). Through the emotion engine API, the server analyzes the message and obtains an emotion score such as "anger," "sadness," or "joy."

[1282] Input: Preprocessed message text

[1283] Output: Sentiment score from the sentiment engine

[1284] Specific operation: Sending a request to the Emotion Engine API and receiving a response

[1285] Step 7:

[1286] The server aggregates the risk and emotion scores for each employee and calculates a cumulative risk rating. If the rating exceeds a set threshold, it determines who should be escalated and notifies them. Ratings and notifications are performed automatically based on the information in the database.

[1287] Input: Risk score and sentiment score for each message

[1288] Output: Cumulative risk assessment and escalation notification

[1289] Specific actions: Calculating risk assessment, sending escalation notifications

[1290] Step 8:

[1291] The server periodically prepares pulse surveys and generates questions for employees. The surveys are created and administered using the SurveyMonkey API, etc. The survey results are also incorporated into risk assessments and are automatically saved in a database.

[1292] Input: Pulse survey execution instruction

[1293] Output: Pulse survey response results

[1294] Specific operations: Generate and send pulse surveys, save response results

[1295] (Application example 2)

[1296] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1297] The difficulty of understanding the health and mental state of employees in a company in real time and quickly detecting and responding to risks of losing human resources or information leaks is a major problem. In particular, extracting negative or dangerous words from employees' chat logs and daily communications to conduct risk assessments is cumbersome, and manual assessments are difficult to monitor efficiently and accurately. Furthermore, a systematic system is needed to comprehensively assess employees' mental and emotional states and escalate or take necessary action in a timely manner.

[1298] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for analyzing message logs using a generative model and extracting negative and dangerous words, means for calculating a risk score for messages based on the extracted words, means for aggregating risk scores for each employee and determining and notifying an escalation destination when a risk threshold is exceeded, means for analyzing the emotional state of the message using an emotion engine and reflecting the results in the risk score, and means for integrating the results of the pulse survey to perform a final risk assessment. This makes it possible to accurately grasp the health and mental state of employees in real time and efficiently perform risk assessment and escalation.

[1299] A "generative model" is a model that uses natural language processing technology to analyze message logs and extract important features and patterns.

[1300] A "message log" is a record of messages sent and received by employees on company chat platforms, etc.

[1301] "Negative words" are words with negative meanings, such as "painful" or "tired."

[1302] "Dangerous words" are words related to information leaks and risks, such as "confidential" and "leak."

[1303] A "risk score" is a risk index calculated based on the importance and frequency of occurrence of extracted negative and dangerous words.

[1304] An "emotion engine" is a technology for analyzing emotions such as "anger," "sadness," and "joy" from the content of a message.

[1305] A "pulse survey" is a survey that periodically asks simple questions to understand employees' health and mental state.

[1306] "Escalation" is the process of notifying higher-level management when a certain risk score is exceeded and urging them to take action.

[1307] A "dashboard" is a tool that visually monitors and displays employees' mental and risk status in real time.

[1308] This invention is a system for understanding the health and mental state of employees in a company in real time and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze chat logs, extracting and scoring negative and dangerous words, and combining it with an emotion engine to recognize emotional states and perform comprehensive risk assessments. Furthermore, by integrating pulse survey results and displaying them on a dashboard in real time, it is possible to accurately grasp the overall health status of employees.

[1309] Server Processing

[1310] The server automatically collects message data and metadata (sender, recipient, timestamp) from the company's internal chat platform (e.g., a general messaging system) using an API to periodically retrieve chat logs. It then removes unnecessary information from the collected chat logs and cleans up the message text. Specifically, it performs processes such as removing HTML tags, deleting special characters, and standardizing different languages.

[1311] The preprocessed chat logs are input into a generative model (e.g., BERT, GPT-3), which uses natural language processing techniques to analyze each message and extract important features based on the text content. Negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") are detected from the analysis results. The detected words are saved along with metadata.

[1312] Next, a risk score is calculated for each message based on the detected words, taking into account the importance and frequency of the words, as well as the message context. Furthermore, an emotion engine is used to analyze the emotional state of the detected message. For example, the emotion engine can read emotions from the message, recognizing emotions such as "anger," "sadness," and "joy," and then calculate an emotion score.

[1313] The server aggregates the risk and emotion scores for each employee and calculates the employee's cumulative risk rating. If this rating exceeds a set threshold, it determines who should be escalated and notifies them.

[1314] Additionally, the server periodically prepares pulse surveys and generates questions for employees. The results of the survey are also incorporated into risk assessments to understand the employee's overall risk status. For example, questions such as, "Have you been feeling stressed at work recently?"

[1315] Terminal handling

[1316] Employee devices (PCs and smartphones) send chat logs generated during daily work to a server in real time. This data is used for the analysis process described below. The data sent from the device goes through the processing steps described above on the server and is used for risk assessment. The device also sends pulse survey notifications to employees, who respond to these. The responses are sent to the server and reflected in the risk assessment.

[1317] User operations

[1318] Administrators use a dashboard provided by the server to monitor employees' mental state and the risk of information leaks in real time. The dashboard displays visual graphs and figures, allowing them to immediately grasp the risk status. Upon receiving an escalation notification, administrators can take the necessary countermeasures, such as arranging a meeting with the employee or arranging for professional support.

[1319] Specific examples

[1320] For example, suppose Employee A frequently mentions in chat that he or she "can't stand the work anymore." In this case, the server collects these messages and uses a generative model to detect the negative word "can't stand it." The emotion engine then recognizes strong emotions such as "anger" or "sadness" from these messages and calculates an emotion score based on that. This gives the message a high risk score.

[1321] If the cumulative risk score exceeds the threshold, the server sends an escalation notification to the management department. The manager then schedules a meeting with Employee A and provides the necessary support. In this way, the system of the present invention enables companies to detect risks of losing human resources or leaking information early and respond quickly and appropriately.

[1322] Prompt Sentence Examples

[1323] "Employee A says: 'I'm at my limit in this job.' Please analyze the emotional state of this message and calculate an appropriate risk score."

[1324] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1325] Step 1:

[1326] The server periodically retrieves chat logs from the company's internal chat platform. For example, it uses an API to automatically collect message data and metadata (sender, recipient, timestamp) from the company's internal messaging system. The input is the chat platform's API endpoint, and the output is chat log data in JSON format.

[1327] Step 2:

[1328] The server removes unnecessary information from the collected chat logs and cleans up the message text. Specifically, it performs processes such as removing HTML tags, deleting special characters, and unifying different languages. The input is the chat log data from the previous step, and the output is the cleaned message text in the target language.

[1329] Step 3:

[1330] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3). The model uses natural language processing techniques to analyze each message and extract important features based on the text content. The input is the cleaned message text from the previous step, and the output is the feature-extracted data.

[1331] Step 4:

[1332] The server detects negative and risky words from the analysis results of the generative model. The detected words are saved along with the associated metadata. The input is the feature extraction data from the previous step, and the output is a list of detected negative and risky words.

[1333] Step 5:

[1334] The server calculates a risk score for each message based on the words detected, taking into account the importance and frequency of the words, as well as the message context. The input is the list of negative and dangerous words from the previous step, and the output is a risk score for each message.

[1335] Step 6:

[1336] The server uses an emotion engine to analyze the emotional state of the message. Specifically, it reads emotions from the message, recognizes emotions such as "anger," "sadness," and "joy," and calculates an emotion score. The input is the message to which the risk score from the previous step has been assigned, and the output is the emotion score.

[1337] Step 7:

[1338] The server aggregates the risk and sentiment scores for each employee and calculates the employee's cumulative risk rating. If this rating exceeds a set threshold, it determines who should be escalated and notifies them. The input is each employee's risk and sentiment score, and the output is an escalation notification.

[1339] Step 8:

[1340] The server periodically prepares pulse surveys and generates questions for employees. The survey results are also incorporated into the risk assessment. The inputs are the survey questions and employee responses, and the output is an updated risk assessment.

[1341] Step 9:

[1342] The terminals send chat logs generated during employees' daily work to the server in real time. This data is used for the analysis process described above. The input is the chat messages generated on the terminals, and the output is the data sent to the server.

[1343] Step 10:

[1344] The terminal sends a pulse survey notification to the employee, who then responds. The response results are sent to the server and reflected in the risk assessment. The input is the response of the employee who received the survey notification on the terminal, and the output is the survey data sent to the server.

[1345] Step 11:

[1346] Users use a dashboard provided by the server to monitor the mental state of employees and the risk of information leakage in real time. The dashboard displays visual graphs and numerical values, allowing users to immediately grasp the risk status. The input is real-time risk data received from the server, and the output is the visualized dashboard screen.

[1347] Step 12:

[1348] The user receives the escalation notification and takes necessary measures, such as setting up a meeting with an employee or arranging for professional support. The input is the escalation notification from the server, and the output is the implemented measures.

[1349] The specific processing unit 290 transmits the result of the specific processing to the headset type terminal 314. In the headset type terminal 314, the control unit 46A causes the speaker 240 and the display 343 to output the result of the specific processing. The microphone 238 acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[1350] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1351] In the above embodiment, an example was given in which the specific processing is performed by the data processing device 12, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the headset type terminal 314.

[1352] [Fourth embodiment]

[1353] FIG. 7 shows an example of the configuration of a data processing system 410 according to the fourth embodiment.

[1354] 7, a data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[1355] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[1356] The robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a control target 443. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, the camera 42, and the control target 443 are also connected to the bus 52.

[1357] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[1358] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[1359] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 control the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[1360] The control object 443 includes a display device, LEDs in the eyes, and motors for driving the arms, hands, and feet. The posture and gestures of the robot 414 are controlled by controlling the motors of the arms, hands, and feet. Some of the emotions of the robot 414 can be expressed by controlling these motors. In addition, the facial expressions of the robot 414 can also be expressed by controlling the light emission state of the LEDs in the eyes of the robot 414.

[1361] Fig. 8 shows an example of the main functions of the data processing device 12 and the robot 414. As shown in Fig. 8, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[1362] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[1363] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[1364] In the robot 414, the processor 46 performs the reception output process. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[1365] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1366] This invention is a system for understanding the health and mental state of employees in a company in real time, and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze chat logs, extract and score negative and dangerous words, and further integrates the results of pulse surveys to perform a final risk assessment.

[1367] Program processing (natural language explanation)

[1368] server

[1369] 1. Collecting chat logs

[1370] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams). Automatically collecting this data using an API enables timely monitoring.

[1371] 2. Log Analysis

[1372] The server inputs the collected chat logs into a generative model, which analyzes and extracts negative or dangerous words. The generative model then uses natural language processing technology to analyze the message content and detect specific words and phrases.

[1373] 3. Calculating the risk score

[1374] The server calculates a risk score for each message based on the words detected, for example, assigning a medium risk score to negative words like "painful" and a high risk score to dangerous words like "confidentiality leak."

[1375] 4. Score tallying and escalation

[1376] The server compiles risk scores for each employee and escalates when the cumulative score exceeds a certain threshold. The escalation destination is automatically determined based on the level of risk and notified.

[1377] 5. Conducting a pulse survey

[1378] The server periodically prepares pulse surveys and generates questions for employees, and the survey results are also incorporated into risk assessments to understand the overall risk status of employees.

[1379] Terminal

[1380] 1. Sending chat logs

[1381] The devices (employee PCs and smartphones) send chat logs generated during daily work to the server in real time. This data is used for the analysis process described below.

[1382] 2. Pulse Survey Responses

[1383] The device sends a pulse survey notification to the employee, who then responds. The response results are sent to the server and reflected in the risk assessment.

[1384] User (Administrator)

[1385] 1. Dashboard Monitoring

[1386] Users can use a dashboard provided by the server to monitor the mental state of employees and the risk of information leaks in real time. The dashboard provides visual information, allowing users to immediately check whether there is any risk.

[1387] 2. Escalation

[1388] Users receive escalation notifications from the server and can take action, such as arranging for specialized support for employees with high risk scores or reporting the situation to their immediate supervisor.

[1389] Specific examples

[1390] For example, suppose Employee A frequently mentions in chat that he or she "can't stand the work anymore." In this case, the server collects these messages and uses a generative model to detect the negative word "can't stand it." It then assigns a higher risk score to the message. If the cumulative risk score exceeds a threshold, the server sends an escalation notification to the HR department. The HR staff then schedules a meeting with Employee A and provides the necessary support.

[1391] In this way, the system of the present invention allows companies to detect risks of losing human resources or leaking information at an early stage and take prompt and appropriate measures. Furthermore, by integrating the results of regular pulse surveys, it is possible to more accurately grasp the overall health status of employees.

[1392] The processing flow will be explained below.

[1393] Step 1:

[1394] server

[1395] Get chat logs

[1396] The server periodically collects all chat messages using the API of the company's internal chat platform (e.g., Slack, Microsoft Teams), along with metadata such as the message sender, recipient, and timestamp.

[1397] Step 2:

[1398] server

[1399] Data Preprocessing

[1400] The server cleans up the chat logs by removing unnecessary information, such as removing HTML tags, removing special characters, and standardizing different languages.

[1401] Step 3:

[1402] server

[1403] Input to the generative model

[1404] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3), which uses natural language processing techniques to analyze each message and extract important features based on the text content.

[1405] Step 4:

[1406] server

[1407] Extracting negative and dangerous words

[1408] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are saved along with the metadata.

[1409] Step 5:

[1410] server

[1411] Calculating the risk score

[1412] The server calculates a risk score for each message based on the negative and dangerous words detected, taking into account the importance and frequency of the words, as well as the message context.

[1413] Step 6:

[1414] server

[1415] Score tally

[1416] The server aggregates the risk scores for each employee to calculate a cumulative risk score for that employee, which is updated over time to keep it up to date.

[1417] Step 7:

[1418] server

[1419] Escalation decision

[1420] The server determines whether escalation is necessary when the cumulative risk score exceeds a pre-defined threshold, and if so, determines the appropriate escalation point (e.g., HR department, direct manager).

[1421] Step 8:

[1422] server

[1423] Escalation Notification

[1424] The server sends an escalation notification, which includes the employee's risk score, any negative or dangerous words detected, and recommended actions to take.

[1425] Step 9:

[1426] Terminal

[1427] Conducting a pulse survey

[1428] The terminal receives notifications of pulse surveys sent from the server and periodically asks employees to answer the pulse surveys. Employees answer the questions through the terminal, and the results are sent to the server.

[1429] Step 10:

[1430] server

[1431] Integration of pulse survey results

[1432] The server combines the pulse survey results with the risk score to create an overall risk assessment for the employee, which also influences escalation decisions.

[1433] Step 11:

[1434] User

[1435] Dashboard Monitoring

[1436] The user (administrator) uses a dashboard provided by the server to monitor the mental state of employees and the risk of information leakage in real time. The dashboard displays visual graphs and numerical values, allowing the user to immediately grasp the risk status.

[1437] Step 12:

[1438] User

[1439] Escalation

[1440] The user receives escalation notifications from the server and can take necessary action, such as arranging a meeting with an employee or arranging for professional support.

[1441] Example 1

[1442] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1443] To prevent employees' mental health from deteriorating and information leaks from occurring, a system that monitors employee behavior in real time and takes appropriate action is required. However, conventional systems lacked sufficient chat log analysis, making it difficult to quickly detect and respond to risks. Furthermore, there was a lack of tools to grasp employees' overall status, making it difficult to provide support in a timely manner.

[1444] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[1445] In this invention, the server includes a means for analyzing communication logs using a generative model and extracting negative and dangerous expressions, a means for calculating a risk score for communication based on the extracted expressions, and a means for aggregating risk scores for each employee and determining and notifying notification recipients when a risk threshold is exceeded. This allows for real-time monitoring of employee mental state and information leakage risk, enabling timely escalation and response. Furthermore, by conducting periodic simple surveys and integrating the results with risk scores, more accurate risk assessments are possible.

[1446] A "generative model" is an algorithm that uses natural language processing technology to analyze text data and extract or generate specific information.

[1447] "Communication logs" refer to data that records the communication history of employees on chat platforms, messengers, etc. used during work hours.

[1448] "Negative expressions" are words or phrases that employees use to express feelings of stress, dissatisfaction, anxiety, etc.

[1449] "Dangerous expressions" are words or phrases that suggest risks such as information leaks or compliance violations.

[1450] A "risk score" is an index that quantifies the risk of communication content based on the frequency and importance of negative or dangerous expressions.

[1451] A "simple survey" is a questionnaire survey conducted over a short period of time to understand the mental state and awareness of employees.

[1452] The "information display screen" is an interface that visually displays employees' mental state and the risk of information leakage, and allows for real-time monitoring.

[1453] "Escalation" is the process of notifying the person or department that needs to take action when the risk score exceeds a certain threshold.

[1454] This invention is a system that monitors the mental state of employees and the risk of information leakage in real time within a company, and analyzes communication logs containing negative or dangerous words. This system includes functions such as analyzing communication logs using a generative model, calculating and aggregating risk scores, notifying escalation at appropriate times, conducting periodic simple surveys, and providing an information display screen.

[1455] Hardware and Software

[1456] Server: The server is the main processing component, analyzing communication logs using generative AI models (e.g., GPT-4), maintaining risk scores using a database, and sending escalation notifications using an email service.

[1457] Device: The device is an employee's PC or smartphone, and is used to send chat logs and answer simple surveys.

[1458] User (Administrator): Administrators use the information display screen to monitor employees' mental state and the risk of information leakage, and respond to escalations.

[1459] Program processing

[1460] Server Processing

[1461] 1. Collecting chat logs:

[1462] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams) via API, performs API authentication, and collects messages from the required channels and users.

[1463] Examples:

[1464] OAuth2 is used for API authentication, and the / channels.history endpoint is called to retrieve messages for a specific channel. The retrieved data is saved in JSON format.

[1465] 2. Log analysis:

[1466] The server inputs the collected chat logs into a generative AI model to analyze and extract negative and dangerous words. It then tokenizes the log data to generate prompt sentences and requests the generative model to analyze them.

[1467] Examples:

[1468] Generate a prompt such as "Please parse the following message: 'I can no longer stand the work.'"

[1469] 3. Calculating the risk score:

[1470] The server calculates a risk score based on the analysis results, assigning a score based on specific keywords to quantify the risk of each message.

[1471] Examples:

[1472] The keyword "painful" is assigned a medium risk score (5 points), and the keyword "confidentiality leak" is assigned a high risk score (10 points).

[1473] 4. Score tallying and escalation:

[1474] The server compiles risk scores for each employee, and if the cumulative risk score exceeds a certain threshold, it uses an email service to send an escalation notification to the HR department or other department.

[1475] Terminal handling

[1476] 1. Sending chat logs:

[1477] The device sends the employee's chat log in real time to a server, including all messages generated during work hours.

[1478] 2. Quick survey answers:

[1479] The device sends a simple survey notification to the employee, who then answers it. The results are sent to a server and used for risk assessment.

[1480] User Action

[1481] 1. Information display screen monitoring:

[1482] Managers can use the information display screen to monitor employees' mental health and the risk of information leaks, with risk scores and warnings displayed in real time.

[1483] 2. Escalation Response:

[1484] Administrators receive escalation notifications from the server and take appropriate action, such as suggesting interviews or counseling for employees with high risk scores.

[1485] In this way, the system of the present invention allows for real-time monitoring of employees' mental state and information leakage risks, enabling prompt and appropriate responses. Furthermore, comprehensive risk assessments linked to periodic simple surveys allow for a more accurate understanding of employees' health conditions.

[1486] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1487] Step 1: Collect chat logs

[1488] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams) via API. The input is messages from specific channels and users. The output is a JSON-formatted file containing the retrieved chat log data. Specifically, the server performs API authentication using OAuth2. It then calls the / channels.history endpoint to retrieve messages from specific channels and saves them in JSON format.

[1489] Step 2: Analyze the logs

[1490] The server inputs the collected chat logs into a generative AI model to analyze and extract negative and dangerous words. The input is a JSON file of the saved chat logs. The output is a list of negative and dangerous words as the analysis results. Specifically, the server reads the log data and tokenizes each message. It then inputs the generated prompt sentences into the generative AI model and obtains the analysis results from the model.

[1491] Step 3: Calculate the risk score

[1492] The server calculates a risk score based on the analysis results. The input is the analysis results from the generative AI model. The output is a risk score for each message. Specifically, the server lists the detected keywords, references the risk scores corresponding to each keyword, and calculates the total risk score. For example, the keyword "painful" is assigned 5 points, and the keyword "confidentiality leak" is assigned 10 points.

[1493] Step 4: Score aggregation and escalation

[1494] The server aggregates the risk scores for each employee and starts the escalation process if a certain threshold is exceeded. The input is the calculated risk score. The output is the sending of an escalation notification. Specifically, the server aggregates the risk scores for each employee from the database. If the cumulative score exceeds the threshold, it sets an escalation flag, determines the escalation destination, and sends the notification using an email service.

[1495] Step 5: Conduct a pulse survey

[1496] The server periodically prepares pulse surveys and generates questions for employees. The input is a list of employees and the results of past pulse surveys. The output is the generated pulse survey questions and their results. Specifically, the server automatically generates survey questions, creates a survey form, and sends it to each employee's email address. The collected survey responses are analyzed and reflected in the risk assessment.

[1497] Step 6: Monitoring the Dashboard

[1498] Users use a dashboard provided by the server to monitor employees' mental state and information leakage risks. The input is the risk score and pulse survey results from the server. The output is real-time risk information displayed on the dashboard. Specifically, users check the dashboard to visually understand the scores and warnings for each employee.

[1499] (Application example 1)

[1500] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1501] Companies need to understand the mental and physical health of their employees in real time, and quickly detect and respond to risks of losing talent or leaking information. However, current systems often have delays in analyzing chat logs and pulse survey results, making it difficult to detect risks early.

[1502] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[1503] In this invention, the server includes means for analyzing message logs using a generative model and extracting negative and dangerous words, means for calculating risk scores for messages based on the extracted words, means for aggregating risk scores for each employee and determining and notifying an escalation destination when a risk threshold is exceeded, means for collecting data using the chat platform's API, means for periodically notifying employees of pulse survey questions and collecting their responses, and means for integrating the collected data with the risk scores to make a final risk assessment. This allows for a real-time understanding of employees' mental and health states, enabling early detection of risks and rapid response.

[1504] A "generative model" is an AI model that uses natural language processing technology to analyze message logs and detect specific words and phrases.

[1505] A "message log" is historical data of messages exchanged on a company's internal chat platform.

[1506] "Negative words" are specific words that may indicate an employee's mental or physical health is deteriorating.

[1507] "Dangerous words" are specific words that are likely to lead to information leaks or disrupt business operations.

[1508] A "risk score" is a numerical representation of the risk level of a message based on the negative or dangerous words contained in the message.

[1509] "Escalation" is a notification process that prompts support personnel or administrators to take action when a detected risk exceeds a certain threshold.

[1510] A "chat platform API" is an application program interface for obtaining data from chat systems used by companies.

[1511] A "pulse survey" is a brief questionnaire survey conducted to periodically check the mental and physical state of employees.

[1512] A "dashboard" is an interface that visually displays employees' mental state and the risk of information leaks in real time.

[1513] The system for implementing the present invention comprises the following steps.

[1514] server

[1515] The server periodically collects message logs using the API of the company's internal chat platform (e.g., Slack or Microsoft Teams). The collected message logs are then analyzed using a generative AI model (e.g., OpenAI GPT-4 or Hugging Face Transformers). During this analysis, negative and dangerous words are extracted from the messages, and a risk score is calculated for each message.

[1516] The server also aggregates risk scores for each employee, and if the cumulative risk score exceeds a set threshold, it automatically determines the appropriate escalation and notifies the appropriate person. The server also periodically sends pulse surveys to employees and collects their responses. The results of the pulse surveys are integrated with existing risk scores and used in the final risk assessment.

[1517] Furthermore, the server provides managers with a dashboard that visualizes the mental state of employees and the risk of information leakage in real time, allowing them to monitor the risk situation and take necessary measures promptly.

[1518] Terminal

[1519] Employee devices (smartphones and PCs) send chat logs generated during daily work to a server in real time. This data is used for the analysis process described below. The devices also receive survey notifications, and employees respond to pulse surveys via their devices. Responses are sent to the server and reflected in risk assessments.

[1520] User (Administrator)

[1521] Administrators use a dashboard provided by the server to monitor employees' mental health and information leakage risks in real time. The dashboard provides visual information, allowing them to immediately check whether there is any risk. If an escalation occurs, administrators receive a notification from the server and take appropriate countermeasures, such as arranging for professional support for employees with high risk scores.

[1522] Specific examples

[1523] For example, if Employee A frequently says in chat that he or she "can't stand the work anymore," the server collects those messages and uses a generative model to detect the negative word "can't stand it." The server assigns a high risk score to those messages, and if the cumulative risk score exceeds a threshold, the server sends an escalation notification to the administrator. The administrator receives the escalation notification, schedules a meeting with Employee A, and provides the necessary support.

[1524] Prompt Sentence Examples

[1525] "Please list the negative and dangerous words frequently used by employees in chat, calculate a risk score for each, and also perform a sentiment analysis on those words, outputting the results in JSON format."

[1526] In this way, the server, terminals, and administrators work together to grasp the mental and health status of employees in real time, creating a system that can detect risks early and respond quickly.

[1527] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1528] Step 1:

[1529] The server periodically collects message logs using the API of the company's chat platform. The data obtained via the API is in JSON format and includes the date and time, sender information, and message content.

[1530] Input: Message logs obtained from the chat platform's API

[1531] Output: Message log in JSON format, including date, time, sender information, and message content

[1532] Step 2:

[1533] The server inputs the collected message logs into a generative AI model to analyze the message content. The generative model uses natural language processing technology to extract negative and dangerous words from the messages.

[1534] Input: Message log in JSON format

[1535] Output: Message log with negative and dangerous words extracted

[1536] Step 3:

[1537] The server calculates a risk score for each message based on the extracted negative and dangerous words, for example, assigning a medium risk score to "painful" and a high risk score to "confidentiality leak."

[1538] Input: Message log from which negative and dangerous words were extracted

[1539] Output: Message log with risk score assigned

[1540] Step 4:

[1541] The server aggregates the risk scores for each employee and determines the escalation destination when the cumulative risk score exceeds a set threshold. The server automatically notifies the escalation destination.

[1542] Input: Message logs with risk scores

[1543] Output: Escalation notification

[1544] Step 5:

[1545] The server periodically sends pulse survey questions to employees and collects their responses, which are also included in the risk assessment.

[1546] Input: Pulse Survey Question Notification

[1547] Output: Employee pulse survey responses

[1548] Step 6:

[1549] The server combines the collected pulse survey results with existing risk scores to produce a final risk assessment.

[1550] Input: Pulse survey results and risk score

[1551] Output: Final integrated risk assessment

[1552] Step 7:

[1553] The server provides a dashboard to administrators, which monitors and displays the mental state of employees and the risk of information leakage in real time. Administrators can check the risk status through the dashboard and take appropriate measures.

[1554] Input: Final Integrated Risk Assessment

[1555] Output: Real-time monitoring via dashboard

[1556] Step 8:

[1557] The device sends chat platform message logs and pulse survey responses to the server in real time, ensuring that the latest data is always used for analysis and evaluation.

[1558] Input: Employee chat logs, pulse survey responses

[1559] Output: Real-time updated data

[1560] Step 9:

[1561] Administrators receive escalation notifications from the server and take appropriate action for employees with high risk scores, such as arranging psychological support or scheduling a meeting with the employee.

[1562] Input: Escalation notification

[1563] Output: Implementing countermeasures

[1564] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[1565] This invention is a system for understanding the health and mental state of employees in a company in real time and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze chat logs, extracting and scoring negative and dangerous words, and combining it with an emotion engine to recognize emotional states and perform comprehensive risk assessments. Furthermore, by integrating pulse survey results and displaying them on a dashboard in real time, it is possible to accurately grasp the overall health status of employees.

[1566] Program processing (natural language explanation)

[1567] server

[1568] 1. Obtaining chat logs

[1569] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams), automatically collecting message data and metadata (sender, recipient, timestamp) using an API.

[1570] 2. Data Preprocessing

[1571] The server cleans up the chat logs by removing unnecessary information, such as removing HTML tags, removing special characters, and standardizing different languages.

[1572] 3. Input to the generative model

[1573] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3), which uses natural language processing techniques to analyze each message and extract important features based on the text content.

[1574] 4. Extraction of negative and dangerous words

[1575] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are saved along with the metadata.

[1576] 5. Calculating the risk score

[1577] The server calculates a risk score for each message based on the words detected, taking into account the importance and frequency of the words, as well as the message context.

[1578] 6. Emotion analysis using an emotion engine

[1579] The server uses an emotion engine to analyze the emotional state of the detected message. For example, the emotion engine reads emotions from the message, recognizes emotions such as "anger," "sadness," and "joy," and calculates an emotion score.

[1580] 7. Score tallying and escalation

[1581] The server aggregates the risk and sentiment scores for each employee and calculates the cumulative risk rating for that employee. If this rating exceeds a set threshold, it determines who should be escalated and notifies them.

[1582] 8. Conducting a pulse survey

[1583] The server periodically prepares pulse surveys and generates questions for employees, and the survey results are also incorporated into risk assessments to understand the overall risk status of employees.

[1584] Terminal

[1585] 1. Sending chat logs

[1586] The devices (employee PCs and smartphones) send chat logs generated during daily work to the server in real time. This data is used for the analysis process described below.

[1587] 2. Pulse Survey Responses

[1588] The device sends a pulse survey notification to the employee, who then responds. The response results are sent to the server and reflected in the risk assessment.

[1589] User (Administrator)

[1590] 1. Dashboard Monitoring

[1591] Users can use a dashboard provided by the server to monitor the mental state of employees and the risk of information leaks in real time. The dashboard displays visual graphs and numerical values, allowing users to immediately grasp the risk status.

[1592] 2. Escalation

[1593] The user receives escalation notifications from the server and can take necessary action, such as arranging a meeting with an employee or arranging for professional support.

[1594] Specific examples

[1595] For example, suppose Employee A frequently mentions in chat that he or she "can't stand the work anymore." In this case, the server collects these messages and uses a generative model to detect the negative word "can't stand it." The emotion engine then recognizes strong emotions such as "anger" or "sadness" from these messages and calculates an emotion score based on that. This gives the message a high risk score.

[1596] If the cumulative risk score exceeds the threshold, the server sends an escalation notification to the HR department. HR personnel then schedule a meeting with Employee A and provide the necessary support. In this way, the system of the present invention enables companies to detect risks of losing human resources or leaking information early and respond quickly and appropriately. Furthermore, by combining it with an emotion engine, it is possible to more accurately grasp the emotional state of employees and achieve more effective risk management.

[1597] The processing flow will be explained below.

[1598] Step 1:

[1599] server

[1600] Get chat logs

[1601] The server periodically collects all chat messages using the API of the company's internal chat platform (e.g., Slack, Microsoft Teams), including metadata such as message sender, recipient, and timestamp.

[1602] Step 2:

[1603] server

[1604] Data Preprocessing

[1605] The server cleans the collected chat logs of unnecessary information, including removing HTML tags, special characters, and language uniformity. The preprocessed data is then prepared for analysis.

[1606] Step 3:

[1607] server

[1608] Input to the generative model

[1609] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3), which uses natural language processing techniques to analyze each message and extract important features based on its content.

[1610] Step 4:

[1611] server

[1612] Extracting negative and dangerous words

[1613] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are then saved along with the metadata.

[1614] Step 5:

[1615] server

[1616] Calculating the risk score

[1617] The server calculates a risk score for each message based on the words detected, for example, assigning a medium risk score to the negative word "painful" and a high risk score to the dangerous word "confidentiality leak."

[1618] Step 6:

[1619] server

[1620] Emotion analysis using an emotion engine

[1621] The server uses an emotion engine to analyze the emotional state of the message. Specifically, the emotion engine recognizes emotions such as "anger," "sadness," and "joy," and calculates an emotion score based on that.

[1622] Step 7:

[1623] server

[1624] Score tally

[1625] The server aggregates the risk and sentiment scores for each employee to calculate a cumulative risk rating, which is updated periodically to keep the rating up to date.

[1626] Step 8:

[1627] server

[1628] Escalation decision

[1629] If the cumulative risk assessment exceeds the set threshold, the server determines that escalation is necessary, determines the escalation destination (e.g., HR department, direct supervisor), and notifies the appropriate person.

[1630] Step 9:

[1631] server

[1632] Escalation Notification

[1633] The server then sends a notification to the determined escalation point, which includes the employee's risk score, any negative or dangerous words detected, and recommended actions to take.

[1634] Step 10:

[1635] Terminal

[1636] Conducting a pulse survey

[1637] The terminal displays the pulse survey notification received from the server to the employee, periodically prompting them to answer the survey. The employee answers the survey on the terminal and sends the answer data to the server.

[1638] Step 11:

[1639] server

[1640] Integration of pulse survey results

[1641] The server integrates the collected pulse survey results into a risk assessment, providing a more accurate picture of an employee's overall risk status based on this integrated data.

[1642] Step 12:

[1643] User

[1644] Dashboard Monitoring

[1645] Users (administrators) use a dashboard provided by the server to monitor the mental state of employees and the risk of information leakage in real time. The dashboard displays intuitive graphs and indicators, allowing users to instantly check the risk status.

[1646] Step 13:

[1647] User

[1648] Escalation

[1649] The user receives escalation notifications from the server and can take necessary measures, such as arranging a meeting with an employee or arranging for professional support.

[1650] Example 2

[1651] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1652] Understanding the mental and physical state of employees within a company is an important issue in managing the risks of losing talent and leaking information. However, conventional methods have made it difficult to comprehensively evaluate this in real time, making it difficult to respond in a timely manner. There is a need for a method to accurately and quickly understand employees' mental state and to detect and respond to the risks of losing talent and leaking information early on.

[1653] The specific processing by the specific processing unit 290 of the data processing device 12 in the second embodiment is realized by the following means.

[1654] In this invention, the server includes means for analyzing message logs using a generative model and extracting negative and dangerous words, means for calculating a risk score for messages based on the extracted words, means for analyzing the emotional state of messages using an emotion engine and calculating an emotion score, and means for aggregating the risk and emotion scores for each employee and determining and notifying an escalation destination when a risk threshold is exceeded. This makes it possible to accurately evaluate the mental state of employees in real time and to detect and respond to risks of human resource outflow and information leaks at an early stage.

[1655] A "generative model" is an algorithm that uses natural language processing technology to analyze text data and extract important features.

[1656] "Message Log" refers to message data and metadata obtained from an enterprise's internal chat platform.

[1657] "Negative words" refers to a group of words that suggest an employee's mental state is deteriorating.

[1658] "Dangerous words" refers to a group of words that pose a high risk of information leakage or security risk.

[1659] An "emotion engine" refers to technology that analyzes emotional states from text data and recognizes emotions such as "anger," "sadness," and "joy."

[1660] The "risk score" is a numerical value that indicates the degree of risk of a message, calculated based on the extracted negative and dangerous words.

[1661] The "emotion score" is a numerical representation of the emotional state analyzed by the emotion engine.

[1662] A "pulse survey" is a questionnaire survey that is used to easily understand the current status of employees in a short period of time.

[1663] A "dashboard" is an interface that visually displays various data and allows for real-time monitoring of employees' mental state and the risk of information leaks.

[1664] "Escalation" refers to the process of notifying management of employees or situations deemed to be high risk when a set threshold is exceeded.

[1665] The present invention is a system for understanding the health and mental state of employees in a company in real time and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze message logs, extract and score negative and dangerous words, and further combines it with an emotion engine to recognize emotional states and perform comprehensive risk assessment. Specific embodiments for implementing the present invention are described below.

[1666] Server Operation

[1667] 1. Obtaining chat logs

[1668] The server periodically retrieves chat logs from the company's internal chat platform (e.g., Slack, Microsoft Teams) using an API to collect message data and metadata (sender, recipient, timestamp).

[1669] 2. Data Preprocessing

[1670] The server cleans up the chat logs by removing unnecessary information, using the BeautifulSoup library to remove HTML tags, regular expressions to remove special characters, and the langdetect library to unify the language.

[1671] 3. Input to the generative model

[1672] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3), which analyzes the text and extracts important features through the generative model's API (e.g., OpenAI GPT-3 API).

[1673] 4. Extraction of negative and dangerous words

[1674] The generative model detects negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") from the analysis results. The detected words are saved along with the metadata.

[1675] 5. Calculating the risk score

[1676] The server calculates a risk score for each message based on the extracted words, taking into account the importance and frequency of each word, as well as the message context, to calculate an overall score.

[1677] 6. Emotion analysis using an emotion engine

[1678] The server uses an emotion engine (e.g., TextBlob, VADER) to analyze the emotional state of the detected messages, recognizing emotions such as "anger," "sadness," and "joy" and calculating an emotion score.

[1679] 7. Score tallying and escalation

[1680] The server aggregates the risk and emotion scores for each employee and calculates a cumulative risk rating. If the rating exceeds a set threshold, it determines who should be escalated and notifies them.

[1681] 8. Conducting a pulse survey

[1682] The server periodically prepares pulse surveys and generates questions for employees using the SurveyMonkey API, and the employee responses are reflected in the risk assessment.

[1683] Device behavior

[1684] 1. Sending chat logs

[1685] The devices (employee PCs and smartphones) send chat logs generated during daily work to the server in real time, allowing the server to always analyze the latest data.

[1686] 2. Pulse Survey Responses

[1687] The device sends a pulse survey notification to the employee, and when the employee responds, the results are sent to the server and reflected in the risk assessment.

[1688] User (administrator) actions

[1689] 1. Dashboard Monitoring

[1690] Users can use a dashboard provided by the server to monitor the mental state of employees and the risk of information leaks in real time. The dashboard uses data visualization tools such as Grafana and Tableau to display visual graphs and figures, allowing users to immediately grasp risks.

[1691] 2. Escalation

[1692] Users receive escalation notifications from the server and take necessary action. Notifications are sent via email or a dedicated app. For example, HR personnel can schedule a meeting with high-risk employees or arrange for specialized support.

[1693] Specific examples

[1694] For example, if Employee A frequently mentions "I can't stand my job anymore" in chat, the server collects the messages and uses the BERT model to detect the negative word "unbearable." At the same time, the VADER emotion engine recognizes strong emotions such as "anger" or "sadness" in the messages and calculates an emotion score. This gives the message a high risk score.

[1695] If the cumulative risk score exceeds the set threshold, the server sends an escalation notification to the HR department. HR personnel then schedule a meeting with Employee A and provide the necessary support. This system enables companies to detect risks of losing human resources or leaking information early and respond quickly and appropriately.

[1696] Prompt Sentence Examples

[1697] "Employee A frequently says, 'I can't take it anymore.' Extract negative words and sentiment scores from these messages and calculate a risk score."

[1698] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1699] Step 1:

[1700] The server periodically retrieves chat logs from the company's internal chat platforms (e.g., Slack, Microsoft Teams). To retrieve the logs, it uses the API provided by each platform to collect message data and metadata (sender, recipient, timestamp) via an authentication token. Using this, it sends an HTTP request to the API endpoint and receives chat log data in JSON format as a response.

[1701] Input: Authentication token obtained from the chat platform API

[1702] Output: Message data and metadata in JSON format

[1703] Specific operation: Sending an API request and receiving a response

[1704] Step 2:

[1705] The server cleans up the chat logs by removing unnecessary information and message texts. First, it uses the BeautifulSoup library to remove HTML tags, then it uses regular expressions to remove special characters, and it uses the langdetect library to detect the language of the text and unify it if necessary.

[1706] Input: Chat log data in JSON format

[1707] Output: Preprocessed message text

[1708] Specific actions: HTML tag removal, special character removal, language detection and unification

[1709] Step 3:

[1710] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3). Using the generative model's API (e.g., OpenAI GPT-3 API), it sends the message text to the model and receives the parsed results.

[1711] Input: Preprocessed message text

[1712] Output: Analysis results by generative model

[1713] Specific operation: Sending a request to the Generative Model API and receiving a response

[1714] Step 4:

[1715] The generative model extracts negative and dangerous words from the analysis results, matches them with predefined negative and dangerous word lists, and stores the detected words in a database along with metadata.

[1716] Input: Analysis results from the generative model

[1717] Output: Detected negative and dangerous words

[1718] Specific operation: Matching with word list, saving detection results to database

[1719] Step 5:

[1720] The server calculates a risk score for each message based on the extracted words. The server calculates an overall score by taking into account the importance and frequency of each word, as well as the message context. The server then uses a scoring algorithm to calculate the risk score and stores it in a database.

[1721] Input: Detected negative and dangerous words

[1722] Output: Risk score for each message

[1723] Specific operation: Calculate risk score and save it to the database

[1724] Step 6:

[1725] The server analyzes the emotional state of each message using an emotion engine (e.g., TextBlob, VADER). Through the emotion engine API, the server analyzes the message and obtains an emotion score such as "anger," "sadness," or "joy."

[1726] Input: Preprocessed message text

[1727] Output: Sentiment score from the sentiment engine

[1728] Specific operation: Sending a request to the Emotion Engine API and receiving a response

[1729] Step 7:

[1730] The server aggregates the risk and emotion scores for each employee and calculates a cumulative risk rating. If the rating exceeds a set threshold, it determines who should be escalated and notifies them. Ratings and notifications are performed automatically based on the information in the database.

[1731] Input: Risk score and sentiment score for each message

[1732] Output: Cumulative risk assessment and escalation notification

[1733] Specific actions: Calculating risk assessment, sending escalation notifications

[1734] Step 8:

[1735] The server periodically prepares pulse surveys and generates questions for employees. The surveys are created and administered using the SurveyMonkey API, etc. The survey results are also incorporated into risk assessments and are automatically saved in a database.

[1736] Input: Pulse survey execution instruction

[1737] Output: Pulse survey response results

[1738] Specific operations: Generate and send pulse surveys, save response results

[1739] (Application example 2)

[1740] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1741] The difficulty of understanding the health and mental state of employees in a company in real time and quickly detecting and responding to risks of losing human resources or information leaks is a major problem. In particular, extracting negative or dangerous words from employees' chat logs and daily communications to conduct risk assessments is cumbersome, and manual assessments are difficult to monitor efficiently and accurately. Furthermore, a systematic system is needed to comprehensively assess employees' mental and emotional states and escalate or take necessary action in a timely manner.

[1742] The identification process by the identification processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for analyzing message logs using a generative model and extracting negative and dangerous words, means for calculating a risk score for messages based on the extracted words, means for aggregating risk scores for each employee and determining and notifying an escalation destination when a risk threshold is exceeded, means for analyzing the emotional state of the message using an emotion engine and reflecting the results in the risk score, and means for integrating the results of the pulse survey to perform a final risk assessment. This makes it possible to accurately grasp the health and mental state of employees in real time and efficiently perform risk assessment and escalation.

[1743] A "generative model" is a model that uses natural language processing technology to analyze message logs and extract important features and patterns.

[1744] A "message log" is a record of messages sent and received by employees on company chat platforms, etc.

[1745] "Negative words" are words with negative meanings, such as "painful" or "tired."

[1746] "Dangerous words" are words related to information leaks and risks, such as "confidential" and "leak."

[1747] A "risk score" is a risk index calculated based on the importance and frequency of occurrence of extracted negative and dangerous words.

[1748] An "emotion engine" is a technology for analyzing emotions such as "anger," "sadness," and "joy" from the content of a message.

[1749] A "pulse survey" is a survey that periodically asks simple questions to understand employees' health and mental state.

[1750] "Escalation" is the process of notifying higher-level management when a certain risk score is exceeded and urging them to take action.

[1751] A "dashboard" is a tool that visually monitors and displays employees' mental and risk status in real time.

[1752] This invention is a system for understanding the health and mental state of employees in a company in real time and for early detection and response to risks of human resource outflow and information leaks. This system uses a generative model to analyze chat logs, extracting and scoring negative and dangerous words, and combining it with an emotion engine to recognize emotional states and perform comprehensive risk assessments. Furthermore, by integrating pulse survey results and displaying them on a dashboard in real time, it is possible to accurately grasp the overall health status of employees.

[1753] Server Processing

[1754] The server automatically collects message data and metadata (sender, recipient, timestamp) from the company's internal chat platform (e.g., a general messaging system) using an API to periodically retrieve chat logs. It then removes unnecessary information from the collected chat logs and cleans up the message text. Specifically, it performs processes such as removing HTML tags, deleting special characters, and standardizing different languages.

[1755] The preprocessed chat logs are input into a generative model (e.g., BERT, GPT-3), which uses natural language processing techniques to analyze each message and extract important features based on the text content. Negative words (e.g., "painful," "tired") and dangerous words (e.g., "confidential," "leak") are detected from the analysis results. The detected words are saved along with metadata.

[1756] Next, a risk score is calculated for each message based on the detected words, taking into account the importance and frequency of the words, as well as the message context. Furthermore, an emotion engine is used to analyze the emotional state of the detected message. For example, the emotion engine can read emotions from the message, recognizing emotions such as "anger," "sadness," and "joy," and then calculate an emotion score.

[1757] The server aggregates the risk and emotion scores for each employee and calculates the employee's cumulative risk rating. If this rating exceeds a set threshold, it determines who should be escalated and notifies them.

[1758] Additionally, the server periodically prepares pulse surveys and generates questions for employees. The results of the survey are also incorporated into risk assessments to understand the employee's overall risk status. For example, questions such as, "Have you been feeling stressed at work recently?"

[1759] Terminal handling

[1760] Employee devices (PCs and smartphones) send chat logs generated during daily work to a server in real time. This data is used for the analysis process described below. The data sent from the device goes through the processing steps described above on the server and is used for risk assessment. The device also sends pulse survey notifications to employees, who respond to these. The responses are sent to the server and reflected in the risk assessment.

[1761] User operations

[1762] Administrators use a dashboard provided by the server to monitor employees' mental state and the risk of information leaks in real time. The dashboard displays visual graphs and figures, allowing them to immediately grasp the risk status. Upon receiving an escalation notification, administrators can take the necessary countermeasures, such as arranging a meeting with the employee or arranging for professional support.

[1763] Specific examples

[1764] For example, suppose Employee A frequently mentions in chat that he or she "can't stand the work anymore." In this case, the server collects these messages and uses a generative model to detect the negative word "can't stand it." The emotion engine then recognizes strong emotions such as "anger" or "sadness" from these messages and calculates an emotion score based on that. This gives the message a high risk score.

[1765] If the cumulative risk score exceeds the threshold, the server sends an escalation notification to the management department. The manager then schedules a meeting with Employee A and provides the necessary support. In this way, the system of the present invention enables companies to detect risks of losing human resources or leaking information early and respond quickly and appropriately.

[1766] Prompt Sentence Examples

[1767] "Employee A says: 'I'm at my limit in this job.' Please analyze the emotional state of this message and calculate an appropriate risk score."

[1768] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1769] Step 1:

[1770] The server periodically retrieves chat logs from the company's internal chat platform. For example, it uses an API to automatically collect message data and metadata (sender, recipient, timestamp) from the company's internal messaging system. The input is the chat platform's API endpoint, and the output is chat log data in JSON format.

[1771] Step 2:

[1772] The server removes unnecessary information from the collected chat logs and cleans up the message text. Specifically, it performs processes such as removing HTML tags, deleting special characters, and unifying different languages. The input is the chat log data from the previous step, and the output is the cleaned message text in the target language.

[1773] Step 3:

[1774] The server inputs the preprocessed chat logs into a generative model (e.g., BERT, GPT-3). The model uses natural language processing techniques to analyze each message and extract important features based on the text content. The input is the cleaned message text from the previous step, and the output is the feature-extracted data.

[1775] Step 4:

[1776] The server detects negative and risky words from the analysis results of the generative model. The detected words are saved along with the associated metadata. The input is the feature extraction data from the previous step, and the output is a list of detected negative and risky words.

[1777] Step 5:

[1778] The server calculates a risk score for each message based on the words detected, taking into account the importance and frequency of the words, as well as the message context. The input is the list of negative and dangerous words from the previous step, and the output is a risk score for each message.

[1779] Step 6:

[1780] The server uses an emotion engine to analyze the emotional state of the message. Specifically, it reads emotions from the message, recognizes emotions such as "anger," "sadness," and "joy," and calculates an emotion score. The input is the message to which the risk score from the previous step has been assigned, and the output is the emotion score.

[1781] Step 7:

[1782] The server aggregates the risk and sentiment scores for each employee and calculates the employee's cumulative risk rating. If this rating exceeds a set threshold, it determines who should be escalated and notifies them. The input is each employee's risk and sentiment score, and the output is an escalation notification.

[1783] Step 8:

[1784] The server periodically prepares pulse surveys and generates questions for employees. The survey results are also incorporated into the risk assessment. The inputs are the survey questions and employee responses, and the output is an updated risk assessment.

[1785] Step 9:

[1786] The terminals send chat logs generated during employees' daily work to the server in real time. This data is used for the analysis process described above. The input is the chat messages generated on the terminals, and the output is the data sent to the server.

[1787] Step 10:

[1788] The terminal sends a pulse survey notification to the employee, who then responds. The response results are sent to the server and reflected in the risk assessment. The input is the response of the employee who received the survey notification on the terminal, and the output is the survey data sent to the server.

[1789] Step 11:

[1790] Users use a dashboard provided by the server to monitor the mental state of employees and the risk of information leakage in real time. The dashboard displays visual graphs and numerical values, allowing users to immediately grasp the risk status. The input is real-time risk data received from the server, and the output is the visualized dashboard screen.

[1791] Step 12:

[1792] The user receives the escalation notification and takes necessary measures, such as setting up a meeting with an employee or arranging for professional support. The input is the escalation notification from the server, and the output is the implemented measures.

[1793] The specific processing unit 290 transmits the result of the specific processing to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the control target 443 to output the result of the specific processing. The microphone 238 acquires voice indicating a user input regarding the result of the specific processing. The control unit 46A transmits voice data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[1794] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1795] In the above embodiment, an example was given in which the specific processing is performed by the data processing device 12, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the robot 414.

[1796] The emotion identification model 59 as an emotion engine may determine the user's emotion according to a specific mapping. Specifically, the emotion identification model 59 may determine the user's emotion according to an emotion map (see FIG. 9), which is a specific mapping. Similarly, the emotion identification model 59 may determine the robot's emotion, and the identification processing unit 290 may perform identification processing using the robot's emotion.

[1797] FIG. 9 is a diagram illustrating an emotion map 400 on which multiple emotions are mapped. In the emotion map 400, emotions are arranged in concentric circles radiating from the center. Emotions closer to the center of the concentric circles are more primitive. Emotions representing states and actions arising from a state of mind are arranged on the outer edges of the concentric circles. The concept of emotion includes both affect and mental states. Emotions generally generated from reactions occurring in the brain are arranged on the left side of the concentric circles. Emotions generally induced by situational judgment are arranged on the right side of the concentric circles. Emotions generally generated from reactions occurring in the brain and induced by situational judgment are arranged on the upper and lower sides of the concentric circles. Furthermore, the emotion of "pleasure" is arranged on the upper side of the concentric circles, and the emotion of "discomfort" is arranged on the lower side. In this way, in the emotion map 400, multiple emotions are mapped based on the structure by which emotions are generated, and emotions that tend to occur simultaneously are mapped close to each other.

[1798] These emotions are distributed in the 3 o'clock direction on emotion map 400, and typically fluctuate between relief and anxiety. In the right half of emotion map 400, situational awareness dominates over internal sensations, resulting in a sense of calm.

[1799] The inside of emotion map 400 represents what is going on in the mind, and the outside of emotion map 400 represents behavior, so the further you go outside emotion map 400, the more visible the emotions become (the more they are expressed in behavior).

[1800] Human emotions are based on various balances, such as posture and blood sugar levels. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. Emotions can also be created for robots, automobiles, and motorcycles, based on various balances, such as posture and remaining battery life. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. An emotion map can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on Voice Emotion Recognition and Emotional Brain Physiological Signal Analysis Systems, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). The left half of the emotion map lists emotions belonging to the "reaction" domain, where sensation is dominant. The right half of the emotion map lists emotions belonging to the "situation" domain, where situational awareness is dominant.

[1801] The emotion map defines two emotions that promote learning. One is a negative emotion on the situation side, around the middle of "repentance" or "reflection." In other words, this occurs when the robot experiences negative emotions such as "I never want to feel this way again" or "I don't want to be scolded again." The other is a positive emotion on the response side, around "desire." In other words, this occurs when the robot experiences positive feelings such as "I want more" or "I want to know more."

[1802] The emotion identification model 59 inputs user input into a pre-trained neural network, obtains emotion values ​​indicating each emotion shown in the emotion map 400, and determines the user's emotion. This neural network is pre-trained based on multiple pieces of training data that are combinations of user input and emotion values ​​indicating each emotion shown in the emotion map 400. Furthermore, this neural network is trained so that emotions that are located close to each other have similar values, as in the emotion map 900 shown in FIG. 10. FIG. 10 shows an example in which multiple emotions, "relieved," "calm," and "reassuring," have similar emotion values.

[1803] The system according to the present disclosure has been described above mainly with respect to the functions of the data processing device 12, but the system according to the present disclosure is not necessarily implemented on a server. The system according to the present disclosure may be implemented as a general information processing system. The present disclosure may be implemented, for example, as a software program running on a personal computer or an application running on a smartphone, etc. The method according to the present disclosure may be provided to users in the form of SaaS (Software as a Service).

[1804] In the above embodiment, an example was given in which the specific processing is performed by one computer 22, but the technology of the present disclosure is not limited to this, and the specific processing may be distributed and performed by a plurality of computers including the computer 22. For example, the data generation model 58 may be provided in an external device of the data processing device 12, and data may be generated in the external device in accordance with input data.

[1805] In the above embodiment, an example in which the specific processing program 56 is stored in the storage 32 has been described, but the technology of the present disclosure is not limited to this. For example, the specific processing program 56 may be stored in a portable, computer-readable, non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed in the computer 22 of the data processing device 12. The processor 28 executes the specific processing in accordance with the specific processing program 56.

[1806] Alternatively, the specific processing program 56 may be stored in a storage device such as a server connected to the data processing device 12 via the network 54, and the specific processing program 56 may be downloaded and installed on the computer 22 in response to a request from the data processing device 12.

[1807] It is not necessary to store all of the specific processing program 56 in a storage device such as a server connected to the data processing device 12 via the network 54, or to store all of the specific processing program 56 in the storage 32; only a portion of the specific processing program 56 may be stored.

[1808] The hardware resource for executing a specific process can be any of the following processors: An example of a processor is a CPU, which is a general-purpose processor that functions as a hardware resource for executing a specific process by executing software, i.e., a program. Another example of a processor is a dedicated electrical circuit, such as an FPGA (Field-Programmable Gate Array), a PLD (Programmable Logic Device), or an ASIC (Application Specific Integrated Circuit), which is a processor with a circuit configuration designed specifically for executing a specific process. Each processor has built-in or connected memory, and each processor uses the memory to execute the specific process.

[1809] The hardware resource that executes the specific processing may be configured with one of these various processors, or may be configured with a combination of two or more processors of the same or different types (for example, a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Also, the hardware resource that executes the specific processing may be a single processor.

[1810] As an example of a system configured with a single processor, first, one processor is configured by combining one or more CPUs and software, and this processor functions as a hardware resource that executes a specific process. Second, there is a system that uses a processor that realizes the functions of an entire system including multiple hardware resources that execute a specific process on a single IC chip, as typified by SoC (System-on-a-chip). In this way, a specific process is realized using one or more of the above-mentioned various processors as hardware resources.

[1811] Furthermore, the hardware structure of these various processors can be, more specifically, an electric circuit that combines circuit elements such as semiconductor devices. The specific processing described above is merely an example. Therefore, it goes without saying that unnecessary steps may be deleted, new steps may be added, or the processing order may be rearranged, without departing from the spirit of the invention.

[1812] The above-described description and illustrations are a detailed explanation of the parts related to the technology of the present disclosure and are merely an example of the technology of the present disclosure. For example, the above description of the configuration, functions, actions, and effects is an explanation of an example of the configuration, functions, actions, and effects of the parts related to the technology of the present disclosure. Therefore, it goes without saying that unnecessary parts may be deleted, new elements may be added, or replacements may be made to the above-described description and illustrations within the scope of the gist of the technology of the present disclosure. Furthermore, to avoid confusion and facilitate understanding of the parts related to the technology of the present disclosure, the above-described description and illustrations omit explanations of common technical knowledge that do not require particular explanation to enable the implementation of the technology of the present disclosure.

[1813] All publications, patent applications, and technical standards mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent application, or technical standard was specifically and individually indicated to be incorporated by reference.

[1814] The following is further disclosed regarding the above embodiment.

[1815] (Claim 1)

[1816] A means for analyzing message logs using a generative model and extracting negative words and dangerous words;

[1817] means for calculating a risk score for the message based on the extracted words;

[1818] A method for aggregating risk scores for each employee and determining and notifying the appropriate escalation point when a risk threshold is exceeded;

[1819] A system including:

[1820] (Claim 2)

[1821] 10. The system of claim 1, further comprising means for periodically conducting pulse surveys and integrating the results of the pulse surveys with the risk score to provide a final risk assessment.

[1822] (Claim 3)

[1823] 10. The system of claim 1, further comprising means for providing a dashboard to monitor and display employee mental states and information leakage risks in real time.

[1824] "Example 1"

[1825] (Claim 1)

[1826] A means for analyzing a communication log using a generative model and extracting negative expressions and dangerous expressions;

[1827] means for calculating a risk score for the communication based on the extracted expressions;

[1828] A method for aggregating risk scores for each employee and determining who to notify when a risk threshold is exceeded;

[1829] A system including:

[1830] (Claim 2)

[1831] 10. The system of claim 1, further comprising means for periodically administering a short survey and integrating the results with the risk score to provide a final risk assessment.

[1832] (Claim 3)

[1833] 10. The system of claim 1, further comprising means for providing an information display screen to monitor and display the mental state of employees and information leakage risks in real time.

[1834] "Application Example 1"

[1835] (Claim 1)

[1836] A means for analyzing message logs using a generative model and extracting negative words and dangerous words;

[1837] means for calculating a risk score for the message based on the extracted words;

[1838] A method for aggregating risk scores for each employee and determining and notifying the appropriate escalation point when a risk threshold is exceeded;

[1839] A means of collecting data using the chat platform's API;

[1840] A method for periodically sending pulse survey questions to employees and collecting their responses;

[1841] A means of integrating the collected data with risk scores to produce a final risk assessment;

[1842] A system including:

[1843] (Claim 2)

[1844] 10. The system of claim 1, further comprising means for providing a dashboard to monitor and display employee mental states and information leakage risks in real time.

[1845] (Claim 3)

[1846] The system of claim 1 , further comprising: means for receiving an escalation notification.

[1847] "Example 2: Combining Emotion Engines"

[1848] (Claim 1)

[1849] A means for analyzing message logs using a generative model and extracting negative words and dangerous words;

[1850] means for calculating a risk score for the message based on the extracted words;

[1851] means for analyzing the emotional state of the message using an emotion engine and calculating an emotion score;

[1852] A method for aggregating risk scores and sentiment scores for each employee, and determining and notifying the appropriate escalation point when a risk threshold is exceeded;

[1853] A system including:

[1854] (Claim 2)

[1855] 10. The system of claim 1, further comprising means for periodically conducting pulse surveys and integrating the results of the pulse surveys with the risk score to provide a final risk assessment.

[1856] (Claim 3)

[1857] 10. The system of claim 1, further comprising means for providing a dashboard to monitor and display employee mental states and information leakage risks in real time.

[1858] "Application example 2 when combining emotion engines"

[1859] (Claim 1)

[1860] A means for analyzing message logs using a generative model and extracting negative words and dangerous words;

[1861] means for calculating a risk score for the message based on the extracted words;

[1862] A method for aggregating risk scores for each employee and determining and notifying the appropriate escalation point when a risk threshold is exceeded;

[1863] A means for analyzing the emotional state of a message using an emotion engine and reflecting the result in a risk score;

[1864] A means of integrating the results of pulse surveys into a final risk assessment;

[1865] A system including:

[1866] (Claim 2)

[1867] 10. The system of claim 1, further comprising means for periodically conducting pulse surveys and integrating the results of the pulse surveys with the risk score to provide a final risk assessment.

[1868] (Claim 3)

[1869] 10. The system of claim 1, further comprising means for providing a dashboard to monitor and display employee mental states and information leakage risks in real time. [Explanation of symbols]

[1870] 10, 210, 310, 410 Data Processing Systems 12 Data Processing Device 14 Smart Devices 214 Smart Glasses 314 Headset-type terminal 414 Robot< / url:> < / url:> < / url:> < / url:>

Claims

1. A means for analyzing message logs using a generative model and extracting negative words and dangerous words; means for calculating a risk score for the message based on the extracted words; A method for aggregating risk scores for each employee and determining and notifying the appropriate escalation point when a risk threshold is exceeded; A system including:

2. The system of claim 1 , further comprising means for periodically conducting pulse surveys and integrating the results of the pulse surveys with the risk score to provide a final risk assessment.

3. The system of claim 1 , further comprising means for providing a dashboard to monitor and display the mental state of employees and the risk of information leakage in real time.

Images