Information processing system and information processing method

Abstract

[Problem] To enable a user to readily participate in a consultation. [Solution] This information processing system is characterized by comprising: an interaction processing unit for generating a response to a user on the basis of an input from the user; an abnormality detection unit for detecting an abnormality of the user on the basis of the input; and a consultation center presentation unit for making an inquiry of the user regarding urgency in cases where an abnormality is detected, and, if there is no urgency, discovering the situation from the user and presenting candidates for a consultation center according to the situation.

Description

Information processing system and information processing method

[0001] The present invention relates to an information processing system and an information processing method.

[0002] Patent Document 1 discloses a system that allows communication between a user and a medical service provider.

[0003] Special Publication No. 2010-503122

[0004] Some users are hesitant to contact medical institutions directly.

[0005] The present invention has been made in view of the above background, and aims to provide a technology that allows users to easily seek advice.

[0006] The main invention of the present invention for solving the above problem is an information processing system comprising: an interaction processing unit that generates a response to a user based on input from the user; an abnormality detection unit that detects an abnormality in the user based on the input; and a consultation center presentation unit that, when the abnormality is detected, asks the user about the urgency of the situation, and, when there is no urgency, asks the user about the situation and presents possible consultation centers depending on the situation.

[0007] Other problems and solutions disclosed in this application will be made clear in the section on preferred embodiments of the invention and the drawings.

[0008] According to the present invention, users can easily seek advice.

[0009] It is a diagram showing an example of the overall configuration of an information processing system. It is a diagram showing an example of the hardware configuration of a management server 2. It is a diagram showing an example of the software configuration of a management server 2. It is a diagram explaining the operation of a management server 2.

[0010] <System Overview> An information processing system according to one embodiment of the present invention will be described below. The information processing system of this embodiment is designed to interact with a user, detect any abnormalities in the user during the interaction, inquire about the level of urgency, and even if there is no level of urgency, inquire about the user's situation and present a consultation window appropriate to the situation.

[0011] 1 is a diagram showing an example of the overall configuration of an information processing system. The information processing system of this embodiment is configured to include a management server 2. The management server 2 is communicably connected to a user terminal 1 via a communication network. The communication network is, for example, the Internet, and is constructed using a public telephone line network, a mobile phone line network, a wireless communication path, Ethernet (registered trademark), or the like.

[0012] The user terminal 1 is a computer operated by a user, and may be, for example, a smartphone, a tablet computer, or a personal computer.

[0013] The management server 2 may be a general-purpose computer such as a workstation or a personal computer, or may be logically realized by cloud computing.

[0014] <Management Server> FIG. 2 is a diagram illustrating an example of the hardware configuration of the management server 2. Note that the illustrated configuration is an example, and other configurations may also be used. The management server 2 includes a CPU 201, memory 202, storage device 203, communication interface 204, input device 205, and output device 206. The storage device 203 stores various data and programs, and is, for example, a hard disk drive, solid state drive, or flash memory. The communication interface 204 is an interface for connecting to a communication network, and is, for example, an adapter for connecting to Ethernet (registered trademark), a modem for connecting to a public telephone network, a wireless communication device for wireless communication, or a USB (Universal Serial Bus) connector or RS232C connector for serial communication. The input device 205 is, for example, a keyboard, mouse, touch panel, button, microphone, or the like for inputting data. The output device 206 is, for example, a display, printer, speaker, or the like for outputting data. Each functional unit of the management server 2 described below is realized by the CPU 201 reading a program stored in the storage device 203 into the memory 202 and executing it, and each storage unit of the management server 2 is realized as part of the storage area provided by the memory 202 and the storage device 203.

[0015] 3 is a diagram showing an example of the software configuration of the management server 2. The management server 2 includes a dialogue processing unit 211, an anomaly detection unit 212, a consultation window presentation unit 213, and a reporting unit 214.

[0016] The dialogue processing unit 211 dialogues with the user. The dialogue processing unit 211 accepts input from the user and generates a response to the user based on the input. In this embodiment, the input from the user is assumed to be text data. The dialogue processing unit 211 can generate a response to the user by providing a prompt including the input from the user and an instruction to create a response to the input to a large-scale language model (LLM). The LLM may be provided in the management server 2, or an API provided by an external server that provides an LLM and generates answers may be called to provide a prompt to the LLM and generate an answer.

[0017] The dialogue processing unit 211 may be adapted to accept a variety of input methods to realize more natural dialogue with the user. For example, the dialogue processing unit 211 may accept multimodal input such as voice, images, and videos in addition to text data.

[0018] In the case of voice input, the dialogue processing unit 211 is equipped with a voice recognition module that converts the user's speech into text. This voice recognition module employs the latest voice recognition technology using deep learning, enabling it to convert voice into text with high accuracy even in noisy environments. It also analyzes the prosodic information of the voice (intonation, accent, pauses, etc.), enabling it to capture the user's emotions and intentions that cannot be fully expressed in text.

[0019] In the case of image input, the dialogue processor 211 includes an image recognition module. This module uses deep learning models such as convolutional neural networks (CNN) to analyze the content of images sent by users. For example, if a user sends a photo of their injury, the module can automatically determine the type and severity of the injury and suggest appropriate responses. It is also possible to use facial recognition technology to estimate the user's emotional state from their facial expression.

[0020] For video input, the dialogue processor 211 includes a video analysis module. This module not only recognizes the images of individual frames but also analyzes changes over time. For example, it can use a recurrent neural network (RNN) or a three-dimensional convolutional neural network (3D-CNN) to capture changes in the user's behavior and the surrounding environment. This makes it possible to understand complex situations and emergency situations that the user cannot fully express in words.

[0021] The dialogue processing unit 211 also has the ability to comprehensively analyze these multimodal inputs. For example, by combining voice input and facial image input, it is possible to detect discrepancies between the user's speech and their emotional state, thereby identifying potential problems early on. It is also possible to more accurately grasp the user's situation by taking into account context information such as location information and surrounding environmental sounds sent in addition to text input.

[0022] The dialogue processing unit 211 can handle both cases where one of these various input methods is used alone and cases where multiple input methods are used in combination. Furthermore, the input method can be dynamically switched depending on the user's settings and the situation.

[0023] The anomaly detection unit 212 detects an anomaly in a user based on an input from the user. The anomaly detection unit 212 can detect an anomaly by combining a plurality of methods.

[0024] For example, the anomaly detection unit 212 can use a predefined list of anomaly-related keywords. This list includes a variety of keywords, ranging from direct expressions such as "help me," "it hurts," "I'm scared," and "I want to die," to more ambiguous expressions such as "what should I do" and "I can't do this anymore." Anomalies can be detected by performing a morphological analysis on the user's input text and checking whether these keywords are included.

[0025] For example, the anomaly detection unit 212 may use a machine learning model. Specifically, a natural language processing model (e.g., a modified model such as Bidirectional Encoder Representations from Transformers (BERT) or Generative Pre-trained Transformer (GPT)) previously trained on conversational data may be used. This model may understand the context of input text and quantify and output the likelihood of an anomaly. For example, the anomaly detection unit 212 may analyze text data to determine whether the text data contains keywords suggesting an anomaly. For example, the anomaly detection unit 212 may provide the LLM with a prompt including a user input and a query regarding whether the input is an anomaly, causing the LLM to determine whether an anomaly exists.

[0026] For example, the anomaly detection unit 212 can perform a comparative analysis with the user's past input patterns. For each user, characteristics such as the frequency of word usage in normal conversation, sentence length, and frequency of use of emotional expressions are recorded, and if the current input significantly deviates from these normal patterns, it can be determined that there is a possibility of an anomaly.

[0027] For example, the anomaly detection unit 212 can perform time series analysis. By treating a series of user inputs as time series data, anomalies can be detected by detecting sudden changes in emotions, fixations on specific topics, etc. For this analysis, time series data analysis techniques such as hidden Markov models (HMMs) and long short-term memory (LSTM) networks can be used.

[0028] The anomaly detection unit 212 comprehensively evaluates the detection results obtained by each of the above methods and ultimately determines whether or not an anomaly has occurred. The weighting of each method may be periodically optimized based on system operation data.

[0029] The anomaly detection unit 212 can also classify the type and severity of detected anomalies. For example, it may classify the anomalies into categories such as "health problems," "mental problems," and "external threats," and assign a severity level such as mild, moderate, or severe to each category. The classification results are used in subsequent processes (for example, selecting a help desk or determining the urgency).

[0030] The anomaly detection unit 212 may also include mechanisms to reduce false positives. For example, the anomaly detection unit 212 may reassess the possibility of an anomaly based on the context, taking into account the possibility that the user is using a joke or metaphor. It may also generate a prompt to ask the user to confirm the detected anomaly and present it to the user via the dialogue processing unit 211, thereby reducing false positives.

[0031] The consultation window presentation unit 213 presents a consultation window according to the user's situation. When an abnormality is detected, the unit inquires of the user about the urgency of the situation, and when there is no urgency, the unit asks the user about the situation and presents possible consultation windows according to the situation.

[0032] The consultation window presentation unit 213 asks whether there is an emergency, where the user is, and what the user's situation is, and provides the LLM with a prompt to instruct it to present consultation window candidates according to the user's situation, and also provides input from the user to the LLM to cause the LLM to generate consultation window candidates. In this case, the LLM can determine whether there is an emergency based on the user's response.

[0033] The consultation service presentation unit 213 can employ an advanced recommendation system that takes into consideration multiple factors in order to present the most suitable consultation service according to the user's situation. An example of a method for presenting consultation services will be described below.

[0034] 1. Use of Geographical Location Information The consultation center presentation unit 213 can utilize the user's current location information. This location information can be obtained from the GPS function of the user terminal 1, location estimation based on the IP address, or location information explicitly entered by the user. Specifically, the following processes can be performed: (a) Listing consultation centers within a certain range (e.g., within a 10-kilometer radius) from the user's location; (b) Calculating the distance and travel time to each consultation center to evaluate ease of access; and (c) Presenting closer consultation centers preferentially if the need is urgent. Based on the location information, the consultation center presentation unit 213 can identify the user's current location, ask about the identified location and the user's situation, and provide a prompt to the LLM that includes an instruction to present candidate consultation centers based on the user's situation.

[0035] 2. The user profile-based recommended consultation service presentation unit 213 refers to the user profile, which includes the user's attribute information and past usage history. This profile may include the following information: Basic attributes such as age, gender, and occupation Medical history and health status Previous consultation services used and their evaluations Language and cultural background

[0036] Based on this information, the following recommendations can be made: (a) Presentation of specialized helplines according to age (e.g., counseling for young people, elderly support centers) (b) Prioritization of helplines that can handle sensitive issues taking gender and cultural background into consideration (c) Recommendation of specialists and support groups related to the user's medical history (d) Prioritization of helpline types that the user has given a high rating based on past usage history

[0037] 3. Selection based on the nature and urgency of the problem The consultation center presentation unit 213 considers the nature and urgency of the problem detected by the anomaly detection unit 212. Specifically, based on the following information, it can select a consultation center with appropriate expertise and present 24-hour service centers with priority according to the urgency: (a) Categorization of problems into categories such as health problems, mental health problems, legal problems, and financial problems; and (b) Urgency assessment within each category (e.g., mild, moderate, severe).

[0038] 4. Consideration of Real-Time Service Status The consultation service presentation unit 213 also takes into consideration the real-time status of each consultation service, which may include the following: (a) current congestion status and waiting time, (b) business hours and current availability, and (c) the availability of specific experts. Based on this information, it is possible to preferentially present services that will enable the user to receive appropriate support more quickly.

[0039] 5. Optimization by Machine Learning The consultation service presentation unit 213 continuously optimizes the recommendation algorithm using a machine learning model. (a) Collects user selection results and satisfaction ratings as feedback. (b) Periodically re-learns the ranking algorithm using the collected data. (c) Conducts A / B tests to verify the effectiveness of new recommendation logic.

[0040] By combining the above methods, the consultation service presentation unit 213 generates a list of consultation services optimized for the individual situation and urgency of the user. This list is presented in a prioritized order based on a comprehensive evaluation of the ease of access for the user, the suitability of expertise, and the ability to respond to the urgency.

[0041] The consultation service presentation unit 213 can also add a brief explanation of the reason for selecting each of the consultation services presented.

[0042] The help desk display unit 213 can set multifaceted and detailed criteria to determine the urgency of the user's situation. This determination of the urgency plays an important role in determining an appropriate response and selecting a help desk. Examples of methods and criteria for determining the urgency are described below.

[0043] 1. Definition of Urgency Levels In this system, urgency can be defined in the following four levels: Level 0: No urgency Level 1: Low urgency (warning) Level 2: Medium urgency (prompt response recommended) Level 3: High urgency (immediate response required)

[0044] 2. Keyword-based Urgency Judgment Urgency is judged based on specific keywords and expressions contained in the user's input text. For example: Level 3: "suicide," "murder," "massive bleeding," "difficulty breathing," "chest pain," etc. Level 2: "violence," "panic attack," "high fever," "stalker," etc. Level 1: "stress," "insomnia," "pain," "anxiety," etc. These keyword lists are created based on the advice of medical, psychological, and legal experts and are updated regularly.

[0045] 3. Urgency assessment through context understanding: Rather than simply matching keywords, natural language processing technology is used to understand the context and assess urgency. For example: ・The expression "There's no point in living anymore" is judged as Level 3 because it is highly likely to suggest the risk of suicide. ・The expression "I'm having trouble breathing" is judged as Level 2, considering the possibility of a panic attack.

[0046] 4. Consideration of time factor: Take into account the duration of the problem and the rate of deterioration. For example: ・"I haven't slept for three days" is judged as level 2 ・"A headache that has continued for one month" is judged as level 1

[0047] 5. Assessment of multiple symptoms: When multiple symptoms or problems exist simultaneously, the urgency is adjusted upward based on the combination. For example: If "insomnia" (level 1) and "auditory hallucinations" (level 2) are reported simultaneously, the level is raised to 3.

[0048] 6. Utilizing user background information: Information registered in the user profile is taken into account. For example: If there is a history of suicide attempts, suicide-related keywords will be given more weight. If there are certain chronic illnesses, the urgency of related symptoms will be increased.

[0049] 7. Detecting changes in behavior patterns If a sudden deviation from a user's normal behavior pattern is detected, the urgency is adjusted upward. For example: Multiple accesses at unusual times Sudden use of extreme language that is not normally used

[0050] 8. Consider external factors: External data such as news and weather information will be referenced and reflected in the urgency assessment. For example: In the event of a large-scale disaster, the urgency of related symptom reports will be revised upward. In the event of an infectious disease outbreak, the urgency of related symptoms will be revised upward.

[0051] 9. Comprehensive judgment using machine learning models To comprehensively evaluate the above factors, machine learning models (e.g., gradient boosting decision trees or neural networks) are used. This model takes the following features as input and outputs an urgency level: - Text analysis results (keywords, sentiment analysis) - User profile information - Time factors - External factor data This model is trained using a dataset of expert judgment cases and past response results, and is periodically retrained.

[0052] The results of the urgency assessment are used to determine the order and type of consultation services to be presented, as well as whether to take immediate action (e.g., automatically notify emergency services). To ensure transparency in the assessment, the system can also provide users with a function to briefly explain the basis for the urgency assessment.

[0053] In order to improve the accuracy of emergency assessments, the assessment results are regularly compared with the actual response results, and the assessment criteria and algorithms can be continuously optimized.

[0054] The reporting unit 214 reports an emergency if there is an emergency. The reporting unit 214 can report an emergency to emergency contacts such as an ambulance, the police, or a security company.

[0055] <Operation> FIG. 4 is a diagram illustrating the operation of the management server 2.

[0056] The management server 2 interacts with the user (S301), determines whether an abnormality has occurred based on input from the user (S302), and if an abnormality has occurred (S303: YES), inquires of the user about the urgency (S304), and if there is an emergency (S305: YES), issues a report (S306), and if there is no emergency (S305: NO), inquires of the user about the situation (S307), and presents possible consultation centers according to the situation (S308). Note that if the urgency cannot be determined in step S305, step S304 can be repeated, and if the situation is not sufficiently understood to present a consultation center, step S307 can be repeated.

[0057] As described above, according to the information processing system of this embodiment, if an abnormality in the user is detected during a dialogue with the user, the system can inquire about the user's situation and present a consultation window appropriate to the situation.

[0058] Although the present embodiment has been described above, the above embodiment is intended to facilitate understanding of the present invention and is not intended to limit the present invention. The present invention may be modified or improved without departing from the spirit thereof, and equivalents thereof are also included in the present invention.

[0059] For example, the processing by each of the functional units of the management server 2 described above may be performed by any of the functional units. Also, a different functional unit that performs part of the processing by each of the functional units described above may be added. Also, the functional units of the management server 2 may be distributed across multiple computers.

[0060] Furthermore, the information stored in each storage unit of the management server may be stored in any of the storage units. That is, the information stored in the above-mentioned multiple storage units may be stored in one storage unit, or part of the information stored in one storage unit may be stored in another storage unit.

[0061] <Modification 1> In Modification 1, user privacy protection and data security measures are implemented.

[0062] 1. Data encryption All user data handled within the system is protected using the latest encryption technology. Specifically, the following encryption methods are used: (a) Communication path encryption: TLS 1.3 or higher protocol is used to encrypt communication between client terminals and servers. (b) Stored data encryption: AES-256 bit encryption is used to encrypt data stored in server storage. (c) Key management: Encryption keys are strictly managed using a hardware security module (HSM).

[0063] 2. Access control and user authentication Access to the system is strictly controlled, and user authentication is performed using the following methods: (a) Multi-factor authentication: In addition to passwords, biometric authentication and time-based one-time passwords (TOTP) are used in combination. (b) Risk-based authentication: The access source IP address, terminal information, user behavior patterns, etc. are analyzed, and additional authentication is requested depending on the risk level. (c) Session management: A function is implemented to automatically log out if there is no operation for a certain period of time.

[0064] 3. Data minimization and anonymization: Collect and retain only the minimum amount of data necessary, and anonymize data as much as possible. (a) Data minimization: Collect only information essential to providing services, and do not collect unnecessary data. (b) Pseudonymization: Personally identifiable information is pseudonymized using techniques such as hashing. (c) Anonymization: Data used for purposes such as statistical analysis is anonymized using techniques such as k-anonymization and l-diversification.

[0065] 4. Separate management of data Separate management of different types of data to reduce security risks: (a) Manage personal identification information and other data in separate databases; (b) Manage encryption keys and encrypted data in separate systems.

[0066] 5. Consent management and transparency Manage user consent appropriately and ensure transparency in data usage: (a) Obtain explicit consent: clearly explain the purpose of data collection and use and obtain explicit consent from users. (b) Withdrawal of consent: provide a mechanism that allows users to easily withdraw consent. (c) Data portability: provide a function that allows users to quickly obtain their own data and transfer it to other services.

[0067] 6. Security Monitoring and Response: Continuously monitor the security of the system and establish a system for rapid response. (a) Real-time monitoring: Implement a security monitoring system that utilizes AI functions to detect abnormal access patterns and attacks. (b) Incident response: Establish response procedures in advance in the event of a security incident and conduct regular training. (c) Vulnerability management: Conduct regular vulnerability scans and penetration tests and promptly fix any vulnerabilities discovered.

[0068] 7. Data Disposal Establish procedures for the secure disposal of data that is no longer needed: (a) Automatic deletion: Implement a system to automatically delete data that has exceeded its retention period; (b) Complete erasure: Use techniques to completely erase data from physical storage (e.g., multiple overwriting).

[0069] 8. Employee Education and Access Restrictions Provide security education to system administrators and employees, and strictly manage access privileges: (a) Conduct regular security training; (b) Grant access privileges based on the principle of least privilege; (c) Promptly revoke access privileges when an employee leaves the company.

[0070] 9. Legal Compliance and Security Audits We will comply with relevant laws and guidelines and conduct regular audits: (a) Comply with relevant laws and regulations such as GDPR, CCPA, PIPEDA, and the Personal Information Protection Act. (b) Establish a security management system based on international standards such as ISO 27001 and SOC 2. (c) Conduct regular security audits by a third-party organization.

[0071] <Modification 2> Modification 2 has a multimodal input function that supports voice input and image input in addition to text input.

[0072] 1. The speech-input-compatible dialogue processor 211 is equipped with a speech recognition module and can accept speech input from the user. For speech recognition, the latest speech recognition technology using deep learning is adopted, and the following processes can be performed: (a) Speech-to-text conversion: High-precision speech recognition is achieved using acoustic models and language models using deep neural networks (DNNs). An adaptive learning function is implemented to accommodate dialects and individual speaking habits. (b) Speech prosodic information analysis: Parameters such as pitch, volume, and speaking rate are extracted to estimate emotional states. For example, a higher-than-normal pitch or a faster speaking rate may indicate excitement or anxiety. (c) Non-verbal sound detection: Non-verbal sounds such as coughing, sobbing, and sighing are detected to estimate physical and mental states. For example, frequent coughing may indicate health problems, while sobbing may indicate mental distress.

[0073] 2. The image-input-compatible dialogue processor 211 is equipped with an image recognition module and analyzes images and videos sent by the user, as well as real-time camera footage. Image recognition uses the latest convolutional neural network (CNN) to perform the following processes: (a) Facial Expression Recognition: Using facial landmark detection technology, it extracts feature points such as the eyes, eyebrows, and mouth. Based on the relative positions of these feature points, it classifies basic emotions such as joy, sadness, anger, fear, disgust, and surprise. It also analyzes facial expression changes over time to evaluate changes and intensity of emotions. (b) Posture and Movement Recognition: Using skeletal estimation technology, it analyzes the user's posture and movements. Unnatural postures and sudden movements may indicate physical problems or mental stress. (c) Environmental Recognition: Recognizing objects and backgrounds in images to understand the user's surrounding environment. For example, a messy room or a dark environment may affect the user's mental state.

[0074] 3. The multimodal integration processing anomaly detection unit 212 integrates and analyzes information obtained from the modalities of text, voice, and image. (a) Consistency check between modalities: For example, it checks whether the content of the text matches the tone of the voice and facial expression. If there is a mismatch (e.g., the text is calm but the voice is trembling), a more detailed analysis is performed. (b) Integration of time-series data: It integrates data from each modality in time series to track changes in the user's state. It detects sudden changes and specific patterns, leading to early detection of anomalies. (c) Context-aware anomaly detection: It performs personalized anomaly detection by taking into account the user's past behavioral patterns and personal characteristics. For example, for a user who normally shows little change in facial expression, even a slight change in facial expression can be an important indicator.

[0075] In addition, data processing and integration processing for each modality can be performed by appropriately combining edge computing and cloud computing.

[0076] <Modification 3> In Modification 3, a real-time monitoring function is added. This function enables the system to continuously monitor the user's status and detect any abnormalities without the user having to explicitly input anything.

[0077] 1. Linkage with Wearable Devices In Variation 3, the information processing system links with wearable devices such as smartwatches and fitness trackers. Specifically, the following data is continuously collected and analyzed: (a) Heart Rate: - Analyzes long-term trends in resting heart rate - Detects sudden increases in heart rate and arrhythmias (b) Activity Level: - Tracks diurnal and long-term variations in steps taken, calories burned, and activity intensity - Detects deviations from normal activity patterns (c) Sleep Pattern: - Analyzes sleep duration and quality (deep sleep, REM sleep, etc.) - Detects signs of sleep disorders such as insomnia and hypersomnia (d) Body Temperature: - Tracks body temperature fluctuations and is used for early detection of fever and poor health (e) Blood Oxygen Saturation: - Continuous measurement allows for early detection of respiratory problems

[0078] 2. Utilizing smartphone sensors The following data can be collected and analyzed by utilizing the various sensors installed in the user's smartphone: (a) Location information: - Using GPS data, analyzes the user's movement patterns - Detects long stays in unusual places or sudden changes in movement patterns (b) Acceleration sensor: - Detects changes in the user's movement and posture - Detects falls and abnormal movements (c) Environmental sensor: - Measures ambient brightness, temperature, humidity, etc. - Detects long stays in inappropriate environments (d) Microphone: - Analyzes background sounds and detects abnormal sounds (screams, collisions, etc.) - Analyzes changes in the user's voice characteristics (tone, speed)

[0079] 3. Linking with smart home devices By linking with smart home devices installed in the user's home, the following data can be collected and analyzed: (a) Smart speaker: - Analyzes changes in the user's speech patterns and content - Detects abnormal statements and long periods of silence (b) Smart light: - Analyzes lighting usage patterns - Detects frequent lighting on late at night and lights off for long periods (c) Smart home appliances: - Analyzes daily life patterns such as how often the refrigerator is opened and closed, and how the microwave is used - Detects unusual usage patterns

[0080] 4. Data Integration and Anomaly Detection Algorithms Information collected from the various data sources mentioned above can be integrated and anomalies detected using machine learning algorithms: (a) Personalized Baseline Establishment: Learns each user's normal behavioral patterns and biometric indicators to establish a personalized baseline. (b) Anomaly Detection Algorithms: Detects deviations from normal patterns using unsupervised learning methods (e.g., autoencoders, one-class classifiers). (c) Time Series Analysis: Detects anomalies in time series data using deep learning models such as LSTM (Long Short-Term Memory) networks. (d) Multimodal Data Fusion: Integrates information from different data sources to achieve more reliable anomaly detection.

[0081] 5. Privacy and Security Considerations Protecting user privacy is particularly important in continuous monitoring. To this end, the following measures can be implemented: (a) Data encryption: All collected data is end-to-end encrypted. (b) Local processing preference: Data processing is performed on the user's device as much as possible, and transmission to the cloud is minimized. (c) Anonymization: Data that needs to be sent to the cloud is anonymized as much as possible. (d) Opt-in method: Each monitoring function operates only if explicitly enabled by the user. (e) Data retention period limitations: Collected data is retained only for the minimum necessary period and deleted periodically.

[0082] This real-time monitoring function makes it possible to detect abnormalities in users earlier and more comprehensively.

[0083] <Modification 4> In Modification 4, a personalized anomaly detection function is added. This function learns the user's past behavioral patterns and health status, making it possible to achieve highly accurate anomaly detection customized for each individual.

[0084] 1. User profile construction The system constructs and continuously updates a detailed profile for each user. This profile may include the following information: (a) Basic information - age, gender, occupation, living environment, etc. (b) Medical information - medical history, current health status, medication information, allergies, chronic illnesses, and other individual health risks (c) Behavioral patterns - daily activity schedule, sleep cycle, eating habits, exercise habits, and hobbies (d) Psychological profile: - personality traits (e.g., assessment based on the Big Five model) - stress tolerance level - history of past mental health conditions

[0085] 2. Personalization using machine learning models The following machine learning models can be trained using each user's data to achieve personalized anomaly detection: (a) Behavior prediction model: - Uses a recurrent neural network (RNN) to learn the user's daily behavioral patterns, detects deviations from predicted patterns, and evaluates possible anomalies. (b) Health status estimation model: - Uses ensemble learning methods such as random forest to estimate the user's health status, and evaluates health risks using biometric data, activity data, and environmental data as input. (c) Emotional state classifier: - Uses a support vector machine (SVM) or similar to learn the user's normal emotional state, and classifies emotional states from text, voice, and facial expression data to detect abnormal states. (d) Anomaly detection model: - Uses unsupervised learning methods such as one-class SVM and isolation forest to learn the user's unique "normal" state, and detects data points in a multidimensional data space that deviate from the learned normal state as anomalies.

[0086] 3. Context-aware anomaly assessment Taking into account the user's current context, more accurate anomaly assessment can be performed: (a) Temporal context: - Considers differences in behavioral patterns depending on the day of the week and time of day - Assessment takes into account seasonal fluctuations and long-term trends (b) Location context: - Uses GPS data to identify the user's current location - Assessment is based on different criteria depending on location, such as home, work, or travel destination (c) Social context: - Considers the user's schedule information and social events (days off, public holidays, etc.) - Reflects changes in family structure and social relationships (marriage, divorce, job change, etc.) (d) Environmental context: - Considers weather information (temperature, humidity, air pressure, etc.) - Takes into account special situations such as large-scale events and disasters

[0087] 4. Adaptive Threshold Settings The threshold for detecting abnormalities can be dynamically adjusted according to the user's situation: (a) Sensitivity adjustment: - Based on the user's past abnormality detection history, the balance between false positives and false negatives is optimized. - Sensitivity is automatically adjusted according to the user's health risk. (b) Graded assessment: - Multiple threshold levels are set, and assessments are made in stages from mild to severe abnormalities. - Countermeasures are prepared according to each level. (c) Feedback loop: - Threshold settings are continuously optimized based on feedback from users and medical professionals.

[0088] 5. Multimodal data integration Information from different data sources can be integrated to perform comprehensive anomaly assessments: (a) Sensor fusion: - Integrate multiple data sources such as wearable devices, smartphones, and environmental sensors - Use methods such as Kalman filters to make reliable estimates from noisy data (b) Cross-modal analysis: - Analyze correlations between different modalities (e.g., text, voice, biometric data) - Capture subtle changes that are difficult to detect with a single modality (c) Time-scale integration: - Comprehensive assessment of short-term fluctuations (such as momentary changes in emotions) and long-term trends (such as chronic stress accumulation)

[0089] 6. Introduction of Explainable AI (XAI) The basis for anomaly detection can be presented in an understandable form: (a) Visualization of feature importance: - Visualize the contribution of each feature using SHAP (Shapley Additive explanations) values, etc. - Allows users and medical professionals to understand the basis for anomaly detection and take appropriate action. (b) Visualization of decision trees: - Visualize individual decision trees of a random forest model - Allows intuitive understanding of the decision-making process leading to anomaly detection. (c) Counterfactual explanations: - Generate explanations in the form of "If the data for XX had been within the normal range, it would not have been judged to be an anomaly" - Allows users to obtain specific guidelines for action to improve their condition.

[0090] This personalized anomaly detection function enables highly accurate anomaly detection that takes into account the individuality of each user, thereby reducing unnecessary alerts and minimizing the risk of overlooking them, enabling the provision of more valuable support to users.

[0091] <Modification 5> In Modification 5, a function for evaluating the urgency level in stages is added. This function allows the urgency to be evaluated in detail at multiple levels, and appropriate measures to be taken according to each level.

[0092] 1. Definition of urgency levels In this system, urgency levels can be defined in the following five levels: Level 0: Normal state Level 1: Mild concern (caution required) Level 2: Moderate concern (early intervention recommended) Level 3: High concern (prompt response required) Level 4: Emergency (immediate intervention required)

[0093] 2. Urgency Assessment Algorithm A multi-tiered algorithm that takes into account the following factors can be used to assess the urgency: (a) Severity of symptoms: - Degree of physical symptoms (pain, difficulty breathing, bleeding, etc.) - Strength of mental symptoms (anxiety, depression, suicidal thoughts, etc.) (b) Acuteness of symptoms: - Time from onset of symptoms to the present - Rate of progression of symptoms (c) Potential risk: - Risk assessment based on the user's medical history and current health condition - Risk due to environmental factors (weather, geographical location, etc.) (d) Changes in behavioral patterns: - Degree of deviation from normal behavioral patterns - Sudden changes in activity level or abnormal behavior (e) Social context: - User's social situation (living alone, living with family, etc.) - Presence or absence of a support network

[0094] 3. Response measures for each emergency level The following tiered responses can be taken depending on the emergency level: Level 0 (normal): ・Continuing regular well-being checks ・Providing general information on health promotion and prevention Level 1 (mild concern): ・Sending warning messages to the user ・Suggesting self-care methods and ways to deal with mild symptoms ・Increasing the frequency of monitoring Level 2 (moderate concern): ・Providing the user with specific recommendations for action (e.g., visiting a medical institution or booking a counseling appointment) ・Providing links to reliable sources of information and ways to consult with a specialist ・Starting regular follow-up checks Level 3 (high concern): ・Immediately presenting high-urgency consultation hotlines (e.g., 24-hour hotlines) ・Notifying pre-registered emergency contacts with the user's consent ・Providing a real-time dialogue function with medical professionals and psychological counselors Level 4 (emergency): ・Activating an automatic call function to emergency services (e.g., ambulance, police) ・Sharing the user's location information with emergency services ・Providing the user with emergency response procedures (e.g., first aid methods)

[0095] 4. Dynamic Urgency Assessment Urgency assessment is not static, but can be dynamically updated as the situation changes: (a) Real-time monitoring: - Continuously analyze sensor data and input information and update the urgency accordingly. (b) Automatic escalation over time: - Automatically escalate the urgency if the situation does not improve within a certain time. (c) User feedback integration: - Adjust the urgency based on the user's responses and actions. (d) External information integration: - Adjust the urgency taking into account external data such as weather warnings and disaster information.

[0096] 5. Urgency assessment through multimodal data analysis Urgency can be assessed by integrating data from multiple modalities such as text, voice, and images: (a) Text analysis: - Analyze the content and emotions of text messages using natural language processing technology - Detect specific keywords and expression patterns (b) Voice analysis: - Analyze the prosodic information of voice (pitch, speed, volume, etc.) - Estimate the emotional state from the voice using an emotion recognition algorithm (c) Image / video analysis: - Estimate the emotional state from the user's facial expression using facial expression recognition technology - Detect abnormalities in posture and movement (d) Vital data analysis: - Analyze fluctuations in vital data such as heart rate, blood pressure, and body temperature - Detect deviations from the normal range

[0097] 6. Privacy and security considerations Handling personal information involved in emergency assessments requires particular care: (a) Data encryption: Protect all data used in assessments with strong encryption methods. (b) Access control: Strictly manage accessible information and permissions according to the emergency level. (c) Anonymization processing: When sharing information with external agencies, such as emergency response agencies, provide only the minimum amount of information necessary. (d) Consent management: Obtain prior consent from users regarding the response details for each emergency level. Consent can be withdrawn or changed at any time.

[0098] 7. Ensuring explainability and transparency The basis for urgency assessment can be presented in an understandable way: (a) Visualization of assessment factors: - Visually displays the main factors that contributed to the determination of each urgency level. (b) Stepwise information disclosure: - Detailed assessment information is disclosed step by step according to the urgency level. (c) Interactive explanation function: - Provides a function to explain the assessment basis in more detail in response to questions from users.

[0099] This gradual urgency assessment function allows for a more detailed understanding of the user's situation and makes it possible to provide the appropriate level of support.

[0100] <Modification 6> In Modification 6, an AI assistant collaboration function is added. This function enables collaboration with an AI assistant that utilizes a large-scale language model (LLM) to effectively grasp the user's situation through more natural dialogue.

[0101] 1. Basic structure of the AI ​​assistant The AI ​​assistant of this system can be composed of the following elements: (a) Language model: - Uses a large-scale language model (e.g., GPT series) based on the latest transformer architecture - Provides knowledge specialized for medical and health consultations through fine tuning (b) Dialogue management system: - A module that tracks the context of the dialogue with the user and maintains a consistent conversation - Generates responses taking into account the dialogue history (c) Sentiment analysis engine: - A module that estimates emotions and psychological state from text - Understands the emotions behind the user's comments and selects appropriate responses (d) Knowledge base: - A database that systematizes medical and health information, psychological knowledge, social resource information, etc. - Regularly updated with the latest information to ensure the accuracy of the AI ​​assistant's responses

[0102] 2. Understanding the situation through natural dialogue An AI assistant can understand the user's situation through natural dialogue using the following methods: (a) Open-ended questions: - Start the dialogue with a question that the user can speak freely, such as "How was your day today?" - Extract important information from the user's answer and use it to ask further questions (b) Active listening: - Appropriately rephrase and confirm what the user has said - Generate empathetic responses and create an atmosphere where the user can speak at ease (c) Context-sensitive questions: - Dynamically generate related questions based on the user's answers - Collect necessary information while maintaining a natural flow of dialogue (d) Multi-turn dialogue: - Gradually understand the situation through a multi-turn dialogue rather than a single question and answer - Adjust the depth and specificity of the questions as the dialogue progresses

[0103] 3. Situation Grasping Algorithms The AI ​​assistant can grasp the user's situation comprehensively using the following algorithms: (a) Keyword Extraction: - Extracts important keywords from the user's comments - Identifies medical terms, symptom expressions, emotional expressions, etc. (b) Topic Modeling: - Analyzes the topics of the entire conversation and identifies the user's main concerns - Uses methods such as Latent Dirichlet Allocation (LDA) (c) Time Series Analysis: - Analyzes information across multiple conversations in chronological order - Identifies changes and trends in the user's condition (d) Network Analysis: - Analyzes the relationships between concepts and events that appear in the user's comments using a graph structure - Identifies the root cause of the problem and related factors

[0104] 4. Personalized dialogue strategies AI assistants can adopt personalized dialogue strategies for each user: (a) Adapting communication style: - Adjusting language and dialogue tone according to the user's age, cultural background, and preferences - Appropriately selecting styles such as formal / casual, direct / indirect (b) Optimizing information provision: - Adjusting the level of information detail according to the user's knowledge level and understanding - Optimizing the balance between visual information (diagrams, illustrations, etc.) and verbal explanations (c) Setting pace: - Adjusting the dialogue pace according to the user's reaction speed and concentration level - Suggesting breaks as needed to reduce the user's burden (d) Identifying areas of interest: - Identifying topics and concerns that the user is particularly interested in and exploring them in a focused manner

[0105] 5. Integration of multimodal inputs AI assistants can integrate input modalities other than text to understand the situation: (a) Speech analysis: - Converts user speech into text using speech recognition technology - Estimates emotional state from speech prosody information (pitch, speed, volume, etc.) (b) Image analysis: - Analyzes images and videos shared by users - Estimates emotional state using facial expression recognition technology (c) Biometric data analysis: - Analyzes data such as heart rate and activity level obtained from wearable devices - Detects stress levels and physical changes

[0106] 6. Dealing with Edge Cases AI assistants can be designed to deal appropriately with edge cases such as: (a) Immediate detection of emergency situations: - Immediately identify urgent situations such as suicidal thoughts or acute medical problems - Activate pre-defined emergency protocols (b) Interpret ambiguous expressions: - Appropriately interpret metaphors and euphemisms - Understand while taking into account cultural background and personal idiosyncrasies (c) Handling contradictory information: - When a user's statement contains a contradiction, carefully confirm and understand the true meaning - If necessary, explore the cause of the contradiction (disordered memory, change in situation, etc.) (d) Dealing with silence and evasive responses: - Appropriate response strategies when the user does not answer a question or is evasive - Apply techniques to extract the necessary information without damaging trust

[0107] 7. Ensuring privacy and security The following measures can be taken to ensure privacy and security in interactions with AI assistants: (a) End-to-end encryption: Encrypt all communications between the user and the AI ​​assistant. (b) Data minimization: Collect and store only the minimum amount of data necessary to understand the situation. (c) Anonymization processing: Remove personally identifiable information from conversation logs. (d) Access control: Strictly manage access to the AI ​​assistant's functions and data. (e) Ethical considerations: Establish a system whereby the AI ​​assistant's decisions are overseen and verified by human experts.

[0108] This AI assistant collaboration function allows users to express their situation through more natural and flexible dialogue.

[0109] <Modification 7> In Modification 7, a community support function is added. This function makes it possible to present specialized consultation services in cases where the emergency is low, as well as to perform community matching with users who have had similar experiences.

[0110] 1. Overview of the Community Support Feature This feature consists of the following elements: (a) User profiling: - Analyzes users' experiences, concerns, interests, etc. to create profiles. (b) Matching algorithm: - Evaluates similarities and compatibility between users to make optimal matches. (c) Community management system: - Promotes interaction between users and maintains a safe and constructive environment. (d) Feedback system: - Evaluates the effectiveness of community support and makes continuous improvements.

[0111] 2. The user profiling community support function can create a user profile using the following methods: (a) Initial questionnaire: Collects basic information, experience, and interests of the user when they first start using the system. (b) Behavioral analysis: Analyzes the user's interests based on system usage history, posted content, search keywords, etc. (c) Natural language processing: Analyzes the text of users' posts and inquiries to extract key topics and emotions. (d) Dynamic updates: Continuously updates the profile in response to changes in the user's activities and situation.

[0112] 3. Matching Algorithms The following algorithms can be used to achieve optimal community matching: (a) Cosine Similarity: - User profiles are converted into vectors and the similarity between profiles is calculated. (b) Collaborative Filtering: - Recommends communities that similar users have previously rated as useful. (c) Topic Modeling: - Extracts topics of interest to users using Latent Dirichlet Allocation (LDA) and recommends users and communities with similar topics. (d) Graph-Based Algorithms: - Represents the relationships between users in a graph structure and identifies influential users and communities using algorithms such as PageRank.

[0113] 4. Types of communities This system can provide a variety of communities, such as the following: (a) Experience sharing group: A group where users who have experienced a particular health problem or life challenge gather together (b) Support forum: A place where users can post questions or inquiries about a particular topic and receive answers from other users (c) Peer mentoring: One-on-one matching where experienced users support new users (d) Challenge group: A group where users encourage each other toward a common goal (e.g., quitting smoking, losing weight) (e) Information exchange hub: A community where users can share the latest health information and useful resources

[0114] 5. Community Management System The following functions can be implemented to maintain a safe and constructive community environment: (a) Content moderation: - Using natural language processing and machine learning, automatically detect and filter inappropriate posts, and have human moderators monitor and intervene. (b) User rating system: - Rating and reward system for users who make useful contributions, and warning and penalty system for users with problematic behavior. (c) Privacy settings: - A function that allows users to fine-tune their own information disclosure level, and providing an option for anonymous participation. (d) Promotion of compliance with guidelines: - Clearly indicating community guidelines and regular awareness-raising, and providing a system for reporting violations of the guidelines.

[0115] 6. Involvement of experts To improve the quality and reliability of community support, the involvement of experts can be promoted: (a) Certified expert system: - Certify qualified experts such as medical professionals, counselors, and nutritionists - Prioritize the display of replies and posts by certified experts (b) Expert moderation: - Experts regularly review and correct community content - Correct incorrect information and dangerous advice (c) Online seminars: - Experts regularly hold webinars and Q&A sessions - Answer questions from participants in real time

[0116] 7. Feedback and AI To continuously improve the effectiveness of learning community support, the following mechanisms can be introduced: (a) User feedback: - Surveys to evaluate the appropriateness of matching, the usefulness of support, etc. - Adjustment of matching algorithms based on feedback results (b) Outcome tracking: - Tracking changes in users' health status and quality of life over the long term - Correlation analysis between community participation and outcomes (c) AI-assisted content recommendation: - An AI system that recommends optimal communities and content based on the user's situation and needs - Continuous improvement of recommendation accuracy

[0117] 8. Privacy and Security The following measures may be implemented to ensure the safety of users' personal information and communications: (a) Data encryption: - End-to-end encryption of all communications and storage data (b) Anonymization options: - Ability for users to participate in the community anonymously if they wish (c) Data access control: - Ability for users to fine-tune the level of access and sharing of their data (d) Security audits: - Periodic third-party security audits

[0118] 9. Ethical Considerations The following ethical considerations can be taken into account when operating the community support function: (a) Informed consent: - Clearly explain the risks and benefits and obtain consent before participating in the community. (b) Recommendation for combined use with professional support: - Caution should be exercised to ensure that community support does not substitute for professional medical care or counseling. (c) Diversity and inclusion: - Create an environment where users from all backgrounds can easily participate - Strictly address discrimination and prejudice.

[0119] This community support feature allows users to receive practical advice and emotional support from peers with similar experiences in addition to expert support. This reduces users' sense of isolation and provides a multifaceted approach to problem-solving, enabling more effective support. Mutual support within the community is also expected to strengthen users' self-efficacy and social connections, leading to improved long-term health and well-being.

[0120] <Modification 8> In Modification 8, a follow-up function is added. With this function, after a consultation window is presented, the user's situation can be reconfirmed after a certain period of time has passed, and continuous support can be provided.

[0121] 1. Overview of the follow-up function This function consists of the following elements: (a) Follow-up scheduler: - Sets and manages the appropriate follow-up timing for each user (b) Situation reassessment system: - Reassesses the user's current situation from multiple angles (c) Adaptive support function: - Provides optimal additional support based on the reassessment results (d) Progress tracking system: - Tracks and analyzes changes in the user's long-term situation

[0122] 2. Follow-up Scheduler To properly manage the timing of follow-ups, the following features can be implemented: (a) Dynamic Scheduling: Dynamically set the frequency and interval of follow-ups depending on the nature and urgency of the issue. For example, 24 hours for high urgency, one week for medium urgency, one month for low urgency, etc. (b) User Settings: A feature that allows users to select the timing and frequency of follow-ups they prefer. (c) Adaptive Reminder: Learns and adjusts the optimal reminder method and frequency based on the user's response patterns. (d) Multimodal Notification: Use a combination of multiple notification methods, such as in-app notifications, email, SMS, and push notifications.

[0123] 3. Situation Reassessment System The following methods can be used to comprehensively reassess the user's current situation: (a) Structured questionnaire: ・Quantitative assessment using standardized rating scales (e.g., PHQ-9, GAD-7, etc.) (b) Open-ended questions: ・Hearing about the situation in natural language using an AI chatbot (c) Behavioral data analysis: ・Analysis of objective data such as app usage, activity level, and sleep patterns (d) Biometric data monitoring: ・Analysis of heart rate variability, stress levels, etc. in conjunction with wearable devices (e) Social media analysis: ・Analysis of SNS postings and activity patterns with the user's consent

[0124] 4. Adaptive support function Based on the results of the reassessment, it is possible to provide personalized additional support: (a) Stepped intervention: - Adjust the intensity of support according to the degree of improvement in the situation - Example: Providing self-help resources → Online counseling → Recommending face-to-face medical treatment (b) Resource recommendation engine: - Recommends information, tools, and content optimized for the user's current needs (c) (Goal setting support:) - Set next steps and short-term goals through dialogue with an AI assistant (d) Peer support matching: - Match with users with similar experiences and recovery cases (e) Expert re-linkage: - If necessary, refer the user back to an appropriate expert or medical institution

[0125] 5. Progress Tracking System Examples of features for tracking and analyzing long-term changes in users include: (a) Data visualization: - Graphing mood, symptoms, behavioral changes, etc. over time (b) Milestone recording: - Recording important events and achievements and visualizing progress (c) Predictive analysis: - Predicting future changes in condition using machine learning models (d) Comparative analysis: - Evaluating relative progress by comparing with anonymized data from other users in similar situations

[0126] 6. Optimization using AI To maximize the effectiveness of follow-ups, optimization using AI can be performed: (a) Personalization: - Individually optimize the content and timing of follow-ups based on the user's characteristics, past response patterns, and progress. (b) Sentiment analysis: - Analyze the emotional state of the user from their answers and reactions and select an appropriate response. (c) Natural language generation: - Generate personalized messages tailored to the user's situation and personality. (d) Anomaly detection: - Detect reactions or behaviors that deviate from normal patterns and determine the need for early intervention.

[0127] 7. Privacy and Security The following measures can be implemented to protect user data in follow-up functions: (a) Data minimization: Collect and store only the minimum amount of data necessary for follow-up. (b) Data anonymization: Anonymize data used for analysis and comparison as much as possible. (c) Consent management: Obtain explicit consent for data use at each stage of follow-up. (d) Access control: Strictly control access to follow-up data and grant only the minimum necessary permissions.

[0128] 8. Ethical Considerations The following ethical considerations can be taken into account when operating the follow-up function: (a) Respect for self-determination: - Guarantee the user's right to stop follow-up or change the frequency at any time. (b) Prevention of overreliance: - Promote the improvement of self-management skills to prevent overreliance on the system. (c) Expert supervision: - Judgements and recommendations made by AI are periodically reviewed by experts. (d) Cultural sensitivity: - Provide follow-up that takes into account the user's cultural background and values.

[0129] 9. Effectiveness measurement and continuous improvement Efforts to evaluate the effectiveness of the follow-up function and to continuously improve it include the following: (a) Setting outcome indicators: - Setting objective indicators such as the degree of improvement in symptoms, improvement in quality of life, and reduction in recurrence rate. (b) User feedback: - Conducting regular surveys on the usefulness and satisfaction of follow-up. (c) A / B testing: - Comparative verification of multiple approaches regarding follow-up methods and frequency. (d) Updating machine learning models: - Regularly updating the prediction model and recommendation engine based on accumulated data and results.

[0130] This follow-up function allows users to receive ongoing support even after the initial consultation. The system will accurately identify changes in the user's condition and provide additional support and intervention as needed, promoting long-term problem resolution and maintaining health. Furthermore, the accumulation and analysis of data is expected to improve the overall effectiveness of the system's support and enable more precise preventative intervention.

[0131] <Disclosures> The present disclosure also includes the following configurations. [Item 1] An information processing system comprising: a dialogue processing unit that generates a response to a user based on input from the user; an anomaly detection unit that detects an abnormality in the user based on the input; and a consultation center presentation unit that, when the abnormality is detected, inquires of the user about an urgency, and, when there is no urgency, asks the user about the situation and presents candidate consultation centers according to the situation. [Item 2] The information processing system according to Item 1, further comprising: a reporting unit that reports the abnormality when there is an urgency. [Item 3] The information processing system according to Item 1, wherein the consultation center presentation unit asks whether there is an urgency, where the user is, and what situation the user is in, provides a prompt to a large-scale language model instructing it to present the candidate consultation centers according to the situation of the user, and provides the input from the user to the large-scale language model to cause the large-scale language model to generate the candidate consultation centers. [Item 4] An information processing method comprising: a computer generating a response to a user based on input from the user; detecting an abnormality in the user based on the input; inquiring of the user about an urgency when the abnormality is detected; and, when the abnormality is not an urgency, asking the user about the situation and presenting possible consultation centers according to the situation.

[0132] 1 User terminal 2 Management server

Claims

1. An information processing system comprising: a dialogue processing unit that generates a response to a user based on input from the user; an anomaly detection unit that detects an abnormality in the user based on the input; and a consultation center presentation unit that, when the abnormality is detected, asks the user about the urgency of the situation, and, when there is no urgency, asks the user about the situation and presents possible consultation centers depending on the situation.

2. An information processing system according to claim 1, characterized in that it comprises a reporting unit that reports the abnormality when the emergency occurs.

3. An information processing system as described in claim 1, wherein the consultation window presentation unit asks whether there is an urgency, where the user is, and what situation the user is in, provides a prompt to a large-scale language model instructing it to present the consultation window candidates according to the user's situation, and provides input from the user to the large-scale language model to cause the large-scale language model to generate the consultation window candidates.

4. An information processing method comprising: a computer generating a response to a user based on input from the user; detecting an abnormality in the user based on the input; inquiring about the urgency of the situation from the user when the abnormality is detected; and, when there is no urgency, asking the user about the situation and presenting possible consultation centers according to the situation.

Images