System for improving user well being via llm

Abstract

A system for measuring and enhancing user well-being employs large language models (LLMs) to process physiological and neurological parameters from various signals, assess user status, generate prompts for improvement, and maintain continuous interaction for iterative reassessment.

Description

System for Improving User Well Being via LLMFIELD OF THE INVENTION

[0001] The technology pertains to systems and methods for measuring and improving mental, neurological, and physical well-being using a large language model (LLM).BACKGROUND

[0002] The field of this technology pertains to systems for assessing and enhancing well-being, including mental, neurological, and physical states. Traditional methods of evaluation often rely on periodic assessments by medical professionals or self-reported data, which may not capture the individual's continuous or real-time condition. Existing systems may lack the capacity to continuously monitor, evaluate, and respond to a user's holistic state.

[0003] Current approaches in the field often employ either singular modal assessments or basic heuristic algorithms that react to predefined inputs without leveraging adaptive learning capabilities. For example, some systems might solely rely on questionnaires or periodic surveys to gauge mental health status, often leading to delays in recognizing shifts in a user's condition. Similarly, wearables that monitor physiological signals frequently provide raw data without contextual emotional or psychological interpretation, limiting their utility for comprehensive well-being management.

[0004] There is also a notable gap in integrating advanced language models with wellness monitoring. Many existing systems do not integrate sophisticated natural language processing technologies with real-time sensor data to dynamically and iteratively interact with users. As such, they are unable to foster a continuous dialogue that adapts to changing emotional and physiological states of individuals, potentially limiting the efficacy of intervention strategies aimed at improving well-being.

[0005] What is needed is a system that provides continuous assessment and improvement of an individual's well-being by integrating advanced language models with real-time data from various signals, including user interactions and physiological metrics. This system should be capable of dynamically generating prompts whose outcome guides users from one state of wellbeing to another, allowing for iterative reassessment and engagement. By leveraging such acomprehensive approach, the system may support the enhancement of mental, neurological, and physical health in a more effective and timely manner compared to existing technologies.SUMMARY

[0006] In one aspect, a system for measuring and enhancing user well-being through large language models (LLMs) is disclosed. The system comprises a signal acquisition module configured to receive signals from user online interactions. In addition, when available, the signal acquisition module can receive signals from sensors measuring a user's physiological and neurological parameters. Additionally, the system includes an LLM module trained to assess and quantify emotional and physiological status based on the received signals, an assessment module configured to evaluate the output from the LLM module, and an interaction module to maintain dialogue and perform continuous reassessment of user status.

[0007] One object of the technology is to improve user well-being by continuously monitoring and analyzing various input signals that reflect a user's emotional and physiological state. Through the use of LLMs and cognitive computing models, the system aims to guide the user to a more favorable emotional or physiological state by generating targeted prompts.

[0008] In an embodiment, the input signals received by the signal acquisition module may include text input, physiological data such as heart rate and sweating, and engagement metrics including likes, comments, and clicks. The signal acquisition module may also be configured to receive signals from third-party devices providing physiological metrics.

[0009] In another aspect, the LLM module employed in the system may detect sentiment from textual inputs and signals, outputting structured information categorized into various emotional and physiological states, each quantified on a scale. This allows for a detailed and nuanced understanding of the user's current status.

[0010] Yet another object of the system is to maintain sustained user engagement, ensuring that any improvements in emotional or physiological states are maintained over time. The interaction module plays a crucial role in this by frequently reassessing the user's status and adapting prompts as needed.

[0011] In another embodiment, the system may employ a secondary LLM specifically configured to analyze physiological data received from auxiliary devices such as smartwatches.This secondary LLM functions in conjunction with the primary LLM module to provide a comprehensive evaluation of the user's status.

[0012] In yet another aspect, the assessment module generates new prompts based on cognitive computing models aimed at psychological or emotional improvement whenever a user's emotional or physiological status falls below a predetermined value. This iterative process continues with further prompts being generated and user status being reassessed until the desired status is achieved.

[0013] In an additional embodiment, the system further includes a user interface to facilitate real-time communication between the user and the system. This enhances the overall user experience by making interactions more seamless and intuitive.

[0014] In yet another aspect, the method for evaluating and enhancing user well-being comprises receiving signals from a a user's online interactions, evaluating the user's status through an LLM, generating new prompts when necessary, and engaging in ongoing interaction with the user to periodically re-evaluate their status. This continuous cycle aims to enhance the user's well-being in a dynamic and adaptive manner.BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Fig. 1 shows a block diagram of a system for measuring and enhancing user well-being through large language models (LLMs), comprising components including a signal acquisition module, user interactions signals, and an assessment module.

[0016] Fig. 2 is a block diagram illustrating the system for measuring and enhancing user wellbeing through large language models, showing interactions between the user, signal acquisition module, LLM, assessment module, and secondary LLM.

[0017] Fig. 3 is a flowchart illustrating the process of acquiring signals, processing them through an LLM module, assessing user well-being, and generating prompts for user interaction and continuous reassessment.DETAILED DESCRIPTION

[0018] The present subject matter relates to a system configured to measure and enhance user well-being through the application of large language models (LLMs). The system operatesthrough a feedback loop where it continuously receives signals from the user, evaluates the user's emotional and physiological status, generates prompts aimed at improving the user's well-being, and maintains an ongoing dialogue to reassess the user's status. This detailed description outlines various embodiments of the system's components, their functions, and how they interact to achieve the desired outcomes.

[0019] The system comprises several key modules:

[0020] Signal Acquisition Module: This module is responsible for collecting input signals from various sources starting with user online interactions (such as text inputs, application usage signals, engagement metrics such as "likes", comments, time spent, content viewed and clicks), and optionally including signals from sensors that measure the user's physiological parameters (such as heart rate and sweating) .Large Language Model (LLM) Module: The LLM module utilizes a pre-trained LLM designed for assessing user status based on the received signals. The LLM evaluates the user's emotional and physiological states, quantifying them into structured categories such as emotional state, fear, and anxiety, each within a specified range, for instance, from 1 to 5. Assessment Module: This module processes the outputs from the LLM, evaluating whether the user's emotional or physiological statuses fall below a predetermined threshold. If so, the assessment module generates a new prompt aimed at transitioning the user's state to a more desirable level. The generated prompts leverage cognitive computing models tailored to psychological and emotional improvement. Interaction Module: This module maintains an ongoing dialogue with the user, facilitating continuous interaction and reassessment of the user's status. It uses the feedback from the LLM module to adjust its engagement strategy and ensures the user's states are frequently monitored and managed. User Interface: A user-friendly interface is provided to facilitate real-time communication between the user and the system. This interface supports text-based interactions and other engagement forms such as clicks and emojis.

[0021] In one embodiment, the system's signal acquisition module receives textual inputs directly from the user and evaluates these inputs using the LLM module to detect overall sentiment and specific emotional states. The LLM quantifies these states, and if the sentiment indicates a negative state below a predetermined threshold, the assessment module produces tailored prompts to improve the user's well-being. For example, if a user indicates feeling anxious, the system may respond with a prompt whose outcome will provide the user with relaxation techniques or reassuring messages.

[0022] Another embodiment includes the use of external physiological sensors, available on devices such as smartwatches, wristbands or patches, integrated with the signal acquisition module. These sensors can measure parameters like heart rate, sweating, blood pressure, blood oxygen, atrial fibrillation, activity records, distance walked, calories burned, or sleep quality. The LLM module uses these physiological data points to provide a comprehensive evaluation of the user's physical and emotional states. The system can generate prompts based on both the detected emotional state from textual inputs and physiological data from the sensors, providing a holistic approach to improving user well-being.

[0023] In a further embodiment, the system incorporates engagement metrics such as user clicks, likes, and comments collected through social media interaction or interaction with specific applications such as health-related applications. The signal acquisition module gathers these metrics, and the LLM module evaluates the user's engagement level and emotional sentiment. The assessment module uses this data to create engagement strategies and prompts that encourage positive interaction and emotional uplift.

[0024] Another embodiment utilizes a dual-LLM setup, where one LLM is specialized for generating natural language dialogue, while the second LLM is dedicated to analyzing physiological and emotional data. The interaction module coordinates input from both LLMs to create prompts that are contextually appropriate and accurately targeted toward improving the user's emotional and physiological states. This approach enables sophisticated and nuanced user interactions.

[0025] In yet another embodiment, the system includes a validation step before continuing with further prompts. After receiving initial signals and generating a prompt, the interaction module validates the user's current status through direct feedback or additional signal measurements. This preliminary validation ensures that the generated prompt accurately reflects the user's needs and that subsequent interactions are effectively tailored to maintain or improve the user's well-being.

[0026] The above embodiments demonstrate the versatility and adaptability of the system in enhancing user well-being through meticulous assessment and targeted interventions facilitated by LLMs. The system continuously refines its strategies based on real-time data, making it an effective tool for managing emotional and physiological health.

[0027] Fig. 1 shows a system for measuring and enhancing user well-being through large language models (LLMs). LLM 100 is interfaced with the signal acquisition module 110, user 120,and the assessment module 130. User 120 interacts with the system, generating user interaction signals 121 and user physiological signals 122, both of which are captured by signal acquisition module 110.

[0028] The signal acquisition module 110 processes these inputs and establishes a communication pathway whereby user interaction signals 121 and user physiological signals 122 are transmitted to LLM 100. LLM 100 is designed to facilitate a user-LLM dialogue 125 with user 120. This dialogue is intended to assess and quantify the emotional and physiological status of user 120 based on the gathered signals.

[0029] The assessment module 130 receives outputs from LLM 100 through a digital interface and evaluates these outputs to form a structured response. If user's 120 emotional or physiological status requires improvement, assessment module 130 prompts LLM 100 to generate a new interaction aimed at enhancing user well-being. This cycle of continuous interaction and re-evaluation persists to ensure that the user's well-being is monitored and improved in real time.

[0030] In one embodiment, the system receives input signals that include textual entries and engagement metrics from the user. For instance, when a user communicates through a chat interface expressing feelings of anxiety by typing "I feel very anxious today," the system's signal acquisition module captures this textual input. Concurrently, the system collects engagement metrics such as the number of clicks, time spent on specific areas within the application, and interactive engagements like comments and likes on social media platforms.

[0031] The LLM module evaluates the emotional state derived from the textual input by processing the sentiment of the message. The message "I feel very anxious today" might be quantified, outputting a structured category such as 'anxiety' assigned a value of 4 out of 5. The engagement metrics, combined with the textual analysis, provide a comprehensive picture of the user's emotional condition. For example, reduced interaction and fewer clicks might correlate with negative emotional states.

[0032] Based on these evaluations, the assessment module determines if the user's emotional status is below a set threshold. For example, if the anxiety level is above 3 (on a 1-5 scale), the module may generate a prompt aimed at mitigating anxiety. A possible prompt might be, "The user has shown an anxiety level of 4 out of 5, for the 2ndtime this week. Please start a conversation to reduce the anxiety level to 2 or less. Continue to evaluate the anxiety level between 1 and 5 and stop when reaching 2 or less". An outcome of such a prompt might be,"Consider practicing deep breathing exercises; let's do it together for five minutes." This prompt is directed to the interaction module, which then engages the user through the interface. The system re-evaluates the user's emotional state periodically and adjusts the interaction if needed.

[0033] In another embodiment, the system utilizes engagement metrics such as likes, time spent, comments, and clicks to gauge the user's emotional state. When a user engages with content by providing multiple 'likes' or 'comments' in a short time span, the signal acquisition module captures these engagements. For instance, a user 'likes' several posts about mental health within a few minutes.

[0034] The LLM module processes these engagement metrics. A high level of interaction with mental health-related content may indicate an elevated interest or need for mental health support. For instance, the LLM might classify the engagement with mental health content and assign 'interest in mental health' a value of 4 out of 5.

[0035] The assessment module evaluates if this engagement metric falls below the predetermined threshold for emotional wellness. If it does, a prompt is generated to foster positive interaction and offer support. For example, the prompt could be, "The user has exhibited a heightened interest in stress management resources over the past week. Please initiate a conversation to offer relaxation techniques customized to their user history". An outcome of such a prompt might be, "It appears you've been looking into stress management frequently. Would you like some personalized strategies for managing stress based on your past interactions?" The interaction module ensures this prompt is shown to the user, thereby providing meaningful engagement and support tailored to the user's observed behavior.

[0036] In a further embodiment, the system gathers and evaluates both textual inputs and engagement metrics from user interactions. For instance, a user types "I feel sad" into the chat interface and simultaneously clicks on self-help articles repeatedly within the application environment. The system's signal acquisition module records both types of input.

[0037] The LLM module assesses the combined textual and engagement signals to quantify the user's emotional state. The message "I feel sad" is analyzed, potentially assigning 'sadness' a value of 3 out of 5, while the repeated interaction with self-help content might signal a heightened need for support. Together, these metrics provide a nuanced understanding of the user's emotional needs.

[0038] If the user's emotional state is assessed to be below an acceptable threshold, the assessment module generates a prompt whose outcome is tailored to improve their well-being. For instance, the prompt might read, "The user has shown a depression level of 3 out of 5. Please start a conversation to reduce the depression level to 2 or less. Constantly evaluate the depression level until it reaches a value of 2 or less out of 5 and then stop when the goal is reached". The outcome of such a prompt can be a system message such as "It seems like you are feeling down. How about we explore some activities that might uplift your mood, like watching a funny video or listening to a favorite song?" The interaction module then delivers this prompt outcome to the user and continues to monitor their response and engagement for further adjustments.

[0039] These described embodiments leverage the combined capabilities of text analysis and engagement metrics to create a dynamic and responsive system focused on enhancing user well-being through large language models (LLMs).

[0040] The present invention relates to a system wherein the input signals comprise engagement metrics, specifically likes, comments, emojis, clicks, or time spent. This system uses large language models (LLMs) to measure and enhance user well-being through continuous assessment and interactive system messages generated by said prompts. Five embodiments of this system are described below to illustrate its functionality and application in various scenarios.

[0041] In some embodiments, the system is designed to collect engagement metrics such as likes, comments, and emojis from a social media platform. The signal acquisition module captures these metrics as the user interacts with the platform. For instance, a high frequency of 'likes' and 'smile' emojis on uplifting content may indicate a positive emotional state, while a predominance of 'sad' emojis and negative comments may suggest a distressed emotional condition.

[0042] The LLM module processes these engagement metrics to quantify the user's emotional status. For example, a combination of multiple 'sad' emojis and historical data indicating a user history of significant sad feelings, might point to a high level of sadness, quantified as 4 out of 5. The assessment module evaluates this emotional state and, if it falls below a predefined threshold, it generates a new prompt aimed at improving the user's emotional state. The interaction module then engages the user, suggesting positive interaction activities like sharing a positive post or engaging in a supportive group discussion.

[0043] Some embodiments include a feature for tracking the user's click behavior and the time spent on specific content. The signal acquisition module registers these engagement metrics, noting patterns such as extended time spent on articles about anxiety or frequent clicks on topics related to stress management.

[0044] The LLM module evaluates these patterns to derive insights about the user's mental condition. For instance, prolonged interaction with stress-related articles may indicate heightened stress levels, quantified as 3 out of 5 on the anxiety scale. The assessment module identifies this as an area needing intervention and generates prompts like, "The user shows increased interaction with stress-related posts and his stress level has been evaluated as 3 out of 5. Please start a conversation aiming to reduce the stress level to 2 or less out of 5. Continue monitoring the stress level and stop when stress level is measured as 2 or less". Such prompt might then generate a message to the user such as "Would you like some tips to deal with stress?" The interaction module displays this message, encouraging the user to engage with stress reduction strategies.

[0045] In some embodiments, the system integrates engagement metrics from both likes and comments to assess user sentiment. The signal acquisition module aggregates these metrics, noting the context of user interactions. For example, a user's frequent 'likes' on fitness posts and positive comments on health-related articles suggest a proactive approach to well-being.

[0046] The LLM module processes these indicators and assigns appropriate values to emotional parameters. Positive comments might be quantified as a high level of well-being, for instance 4 out of 5 on the happiness scale. If the user's positive engagement decreases, the assessment module generates prompts that produce messages like, "You seem less active recently. How about a new fitness challenge?" The interaction module then presents this to the user to reignite positive behaviors.

[0047] Some embodiments utilize engagement metrics from user interaction with emotive icons, such as emojis. The signal acquisition module captures the frequency and type of emojis used across different scenarios. For instance, an increased use of 'angry' emojis in a short time frame may signal escalating frustration levels.

[0048] The LLM module parses these emoji patterns, quantifying the emotional state they represent. A surge in 'angry' emojis might be interpreted as 4 out of 5 on the frustration scale. The assessment module responds by generating prompts that can generate messages such as, "It seems like you're feeling frustrated. Would you like to talk about it or engage in somecalming activities?" The interaction module uses these messages to guide the user towards managing frustration effectively.

[0049] In some embodiments, the system places a focus on the user's interactive patterns over time, analyzing engagement metrics such as time spent on specific applications, and the depth of interaction characterized by clicks and engagement rates. The signal acquisition module logs these metrics, noting any shifts in user engagement.

[0050] The LLM module evaluates these longitudinal engagement patterns to assess trends in the user's well-being. For example, an increase in time spent on wellness apps combined with a steady rate of positive feedback from diverse sources such as user history, past user engagements, or additional present or past signals could signal improved stress management capabilities of the user, quantified as 3 out of 5 on a relaxation scale. If a downward trend is detected, the assessment module may generate a prompt that would start a conversation with the user using messages to encourage continued engagement with wellness activities, such as, "You've been making great progress. How about trying a new relaxation technique today?"

[0051] The system leverages engagement metrics to assess and enhance user well-being through a sophisticated interplay of signal acquisition, LLM evaluation, assessment, and interactive prompting modules. This adaptive approach provides tailored interventions aimed at promoting sustained emotional and physiological health.

[0052] In some embodiments, the system incorporates a signal acquisition module configured to gather biometric data that includes heart rate and electrodermal activity, which are indicative of the user's stress levels. The biometric data is collected through wearable sensors such as smartwatches or fitness bands connected to the user's body. When the user's heart rate and electrodermal activity indicate an increased stress level, this data is transmitted to the LLM module. The LLM module evaluates the user's physiological status and quantifies stress on a scale from 1 to 5. If the stress level is above a set threshold, the assessment module generates prompts aimed at generating calming messages, such as suggesting a breathing exercise or playing relaxing music, to help the user reduce their stress levels. The interaction module may deliver these messages through a mobile application or a wearable device interface and continues to monitor the user's physiological data for further adjustments.

[0053] In another embodiment, the system's signal acquisition module is designed to capture biometric data related to sleep patterns, using sensors embedded in a sleep tracking device such as a smart bed or wearable sleep tracker. The captured data includes metrics such as total sleeptime, sleep stages, and interruptions during sleep. This biometric data is then analyzed by the LLM module to determine the quality of the user's sleep. The assessment module evaluates this data and, if it identifies poor sleep quality indicated by frequent awakenings or insufficient REM sleep, it generates prompts that produce messages aimed at improving the user's sleep habits. These messages may include advice on maintaining a regular sleep schedule, reducing screen time before bed, or suggesting relaxation techniques. The interaction module communicates these messages to the user, for example, through a smartphone application, and the system continues to monitor and assess sleep data to achieve improved sleep quality.

[0054] In a further embodiment, the signal acquisition module is configured to collect biometric data reflecting the user's physical activity levels, such as steps taken, calories burned, and active minutes, using a fitness tracker or smartwatch. This data is analyzed by the LLM module to gauge the user's activity level and overall physical health. The assessment module compares the user's activity data against recommended physical activity guidelines. If the user's activity level falls below the recommended guidelines, the module generates prompts producing motivational messages to encourage increased physical activity, such as suggesting the user take a walk, engage in a short workout session, or set daily step goals. The interaction module presents these messages through the user's fitness app or wearable device, providing ongoing encouragement and challenges to help the user meet their physical activity targets.

[0055] In yet another embodiment, the system collects biometric data on the user's hydration levels through a smart water bottle equipped with sensors that measure the quantity of water consumed throughout the day. This data is sent to the signal acquisition module, which then processes and transmits it to the LLM module. The LLM module assesses the user's hydration status and quantifies it on a scale from 1 to 5. If the user's hydration status is below the optimal level, signifying potential dehydration, the assessment module generates prompts that produce messages to remind the user to drink water at regular intervals. These messages might include reminders to take sips of water every hour or to track daily water intake against hydration goals. The interaction module communicates these messages via a mobile application linked to the smart water bottle, ensuring that the user receives timely reminders and maintains adequate hydration levels throughout the day.

[0056] Fig. 2 shows a system for measuring and enhancing user well-being through two large language models (LLMs). The system involves multiple components interacting with each otherfor seamless operation. At the core is the LLM element 100, which communicates with various modules to process and assess user data.

[0057] The signal acquisition module 110 interfaces with the user 120 to gather both user interaction signals 121 and user physiological signals 122. These signals, which might include data from online activities and physiological measurements like heart rate, are relayed to the LLM element 100 for processing.

[0058] The user 120 engages in a user-LLM dialogue 125 facilitated by the system. During this interaction, user data is continuously collected and fed into the LLM element 100 through the signal acquisition module 110. The system maintains an active dialogue, ensuring the user's emotional and physiological status is monitored and evaluated in real-time.

[0059] Next, the LLM element 100 transmits the processed data to the assessment module 130, which evaluates the user's status. If any emotional or physiological metrics are below the desired thresholds, the assessment module 130 generates appropriate prompts that create messages to be sent back to the user. These messages aim at improving the user's well-being and are transmitted back to the LLM element 100.

[0060] Additionally, the system incorporates a secondary LLM element 200, which communicates with both the user 120 and the assessment module 130. The secondary LLM element 200 may provide further analysis or support, ensuring comprehensive coverage of the user's well-being metrics. Communication between the user and the secondary LLM 200 also supports the completion of the evaluation and dialogue process, as illustrated by the bidirectional flow with label 225.

[0061] This interconnected system facilitates continuous interaction, assessment, and enhancement of user well-being by leveraging the processing capabilities of multiple LLM elements and various input signals from the user.

[0062] The present subject matter relates to a system and method configured for measuring and enhancing user well-being through the application of large language models (LLMs), specifically focusing on the detection of sentiment from textual inputs and signals. The system operates through a continuous feedback loop where it receives signals from the user, evaluates these signals to assess the user's emotional and physiological status, generates prompts to produce messages aimed at improving the user's well-being, and maintains an ongoing dialogue to reassess the user's status.

[0063] In one embodiment, the system receives text inputs directly from the user through a chat-based interface. The received text is then processed by the LLM module, which detects the sentiment expressed in the text. For example, if a user types "I feel very sad today," the LLM module identifies the expressed sentiment and quantifies it as 'sadness' at a level of, for instance, 4 out of 5. This quantification is then evaluated by the assessment module, which may also consider other parameters such as past user conversations, data from devices measuring physiological signals such as heart rate and sleep patterns and more. If the sadness level exceeds a predetermined threshold, the assessment module generates a prompt such as "The sadness level of the user has been measured as 4 out of 5. Please start a conversation with the user to reduce the sadness level to 3 or less out of 5. Continue to monitor the user sadness level until it reaches 3 or less out of 5 and then stop". This prompt can produce a message to the user such as, "Let's discuss what's troubling you and find a way to improve your mood." The interaction module delivers this message through the user interface and continues to monitor further interactions to reassess the user's emotional state.

[0064] In another embodiment, the system integrates user engagement metrics with textual sentiment analysis. When a user engages in an application by providing feedback in the form of text comments, likes, or clicks, the signal acquisition module captures both the textual content and the engagement metrics. For instance, a user commenting, "I am worried about my work," and frequently clicking on stress-related articles would be indicative of anxiety. The LLM module processes these inputs and assigns a quantitative value to the detected sentiment, such as 'anxiety' at a level of 3 out of 5. The assessment module evaluates this sentiment and generates prompts like, "The anxiety level of the user has been measured as 3 out of 5. Please start a conversation with the user to reduce the anxiety level to 2 or less out of 5. Continue to monitor the user anxiety level until it reaches 2 or less out of 5 and then stop". This prompt can produce a message to the user such as "Would you like to explore some stress management techniques?" The interaction module presents this suggestion through the user interface, enabling the user to access relevant resources and support.

[0065] In a further embodiment, the system utilizes physiological signal input in conjunction with detected textual sentiment from user interactions. The signal acquisition module gathers data from wearable devices, such as a smartwatch or wristband measuring heart rate, alongside textual inputs like, "I feel nervous." The LLM module evaluates both the textual sentiment and physiological data, assigning values such as 'nervousness' at a level of 3 out of 5 and elevatedheart rate. The assessment module then integrates these values to generate a prompt like, "The nervousness level of the user has been measured as 3 out of 5. Please start a conversation with the user to reduce the nervousness level to 2 or less out of 5. Continue to monitor the user nervousness level until it reaches 2 or less out of 5 and then stop". This prompt can produce a message to the user such as "Your heart rate is elevated, and you've mentioned feeling nervous. Shall we do a quick relaxation exercise?" The interaction module communicates this message to the user and monitors the effectiveness of the relaxation exercise through continued interaction and signal acquisition.

[0066] Another embodiment focuses on engaging with users through social media interactions that include text and engagement metrics. The signal acquisition module collects data such as user posts, comments, and interactions like 'likes' and 'shares.' The LLM module processes the textual content to detect sentiment, for instance, identifying a post stating, "I am feeling overwhelmed," and assigns an 'overwhelmed' value of 4 out of 5. The module also considers engagement metrics, such as minimal 'likes' on positive content, to support this evaluation. The assessment module then generates an appropriate prompt like, "The user has shown an overwhelmed level as 3 out of 5. Please start a conversation with the user to reduce the overwhelmingness level to 2 or less out of 5. Continue to monitor the user overwhelmingness level until it reaches 2 or less out of 5 and then stop". This prompt can produce a message to the user such as "I noticed you've mentioned feeling overwhelmed. How about we set some manageable goals for today?" The interaction module delivers this message and engages with the user through the social media platform's messaging service, ensuring continuous interaction and emotional support.

[0067] These described embodiments illustrate the system's capability to detect and quantify sentiment from various user inputs, including textual content and engagement signals, using advanced LLMs. The system continuously adapts its responses based on real-time data to enhance user well-being through targeted interventions and ongoing dialogue.

[0068] The present subject matter relates to a system and method configured to assess and enhance user well-being through the application of large language models (LLMs). The system involves receiving signals from the user, evaluating these signals to categorize the user's emotional and physiological states into structured categories, and generating prompts that produce messages aimed at improving well-being based on these categorizations.

[0069] LLM Module Output Functionality

[0070] The LLM module is designed to process input signals and output structured information that categorizes the user's emotional and physiological states. This module operates by receiving raw signals from various sources, interpreting these signals to derive meaningful metrics, and quantifying these metrics into categories such as emotional state, fear, and anxiety. Each category is assigned a value on a predefined scale, typically ranging from 1 to 5. The values in these categories are used by other components of the system to generate prompts and facilitate continuous user interaction.

[0071] Signal Processing: The LLM module is configured to process a variety of input signals, including textual inputs from direct user communication, biometric data from sensors, and engagement metrics from user interactions. The signals are parsed to extract relevant features that contribute to the assessment of the user's state. Categorization: The processed signals are categorized into predefined categories that represent the user's emotional and physiological states. Key categories include emotional state, fear, and anxiety. Each category has sub-levels defined in a quantifiable range, typically from 1 to 5, where 1 represents a minimal state and 5 represents an extreme state. Quantification: After categorizing the signals, the LLM module assigns a numerical value to each category. For example, if a textual input indicates high anxiety, the LLM module may quantify this as 'anxiety: 4 out of 5.' This structured output enables the system to systematically interpret and respond to the user's condition.

[0072] In a textual Input analysis embodiment, the LLM module primarily processes textual inputs received from the user interface. When the user inputs text such as "I am feeling very anxious today," the LLM module analyzes the sentiment and context of the statement. The module categorizes the emotional state as 'anxiety' and quantifies it within a range. In this case, the output may be 'anxiety: 4 out of 5.' This structured data is then sent to the assessment module, which determines if the quantified anxiety level requires intervention and generates appropriate prompts that create appropriate messages to the user to mitigate it.

[0073] In a biometric and physiological data integration embodiment, the LLM module receives biometric data such as heart rate, sweating, and other physiological metrics from wearable sensors. For instance, if a user has an elevated heart rate and increased sweating, the LLM module categorizes these signals into physiological and emotional stress factors. Each category is quantified, e.g., 'heart rate: 5 out of 5' and 'sweating: 4 out of 5.' This comprehensive evaluation allows the system to generate prompts that produce holistic messages that address both emotional and physiological aspects of the user's well-being.

[0074] In an engagement metrics evaluation embodiment, the LLM module analyzes engagement metrics such as likes, comments, clicks, and time spent on specific content. For instance, if a user frequently engages with stress-related content and leaves comments indicating worry, the LLM module processes these metrics to categorize the emotional state as 'anxiety' or 'stress.' The quantified output may be 'stress: 3 out of 5' and 'anxiety: 2 out of 5.' This data is utilized by the assessment module to develop prompts whose outcome (messages to the user) is aimed at redirecting user engagement towards more positive interactions and content.

[0075] These embodiments illustrate the capability of the LLM module to translate diverse user signals into structured and quantifiable categories, facilitating targeted interventions to enhance user well-being. The system's ability to integrate textual inputs, biometric data, and engagement metrics ensures comprehensive assessment and effective user support.

[0076] The assessment module is designed to create prompts based on cognitive computing models aimed at psychological or emotional improvement by leveraging structured information from the LLM module. The assessment module processes the categorized and quantified data output by the LLM module, such as emotional state, anxiety, or fear, each represented on a predefined scale.

[0077] Upon determining that the user's emotional or physiological status falls below a predetermined threshold, the assessment module engages cognitive computing techniques to generate prompts whose outcome are personalized messages. These techniques include natural language processing, machine learning algorithms, and psychological frameworks tailored to address specific user needs.

[0078] The generated messages are designed to facilitate cognitive-behavioral interactions that encourage positive mental and emotional states. For instance, if the user's emotional state is identified as anxious and quantified at a high level, the assessment module may generate a prompt that produces a message suggesting breathing exercises, mindfulness practices, or motivational messages aimed at alleviating anxiety.

[0079] Additionally, the prompts' outcome (generated messages) may incorporate evidencebased strategies known to improve well-being, such as encouraging physical activities for stress relief, recommending social interactions for emotional support, or guiding users through relaxation techniques. These messages are communicated through the interaction module to ensure the user receives timely and contextually relevant support.

[0080] The assessment module continuously monitors the effectiveness of the prompts and generated messages and adjusts the prompts them if necessary based on real-time feedback and ongoing data analysis, ensuring a tailored and adaptive approach to enhancing user wellbeing.

[0081] In some embodiments, the communication module is configured to use a generic LLM for dialogue generation and a domain-specific LLM for data analysis as follows:

[0082] In some embodiments, the communication module leverages the capabilities of two distinct types of large language models (LLMs) to achieve its functions. The generic LLM is tasked with generating dialogue for interactions with the user. This model is pre-trained on a broad dataset and is adept at natural language generation, facilitating the creation of coherent, contextually appropriate responses during user engagement. The generic LLM ensures that the system can maintain an ongoing, natural conversation with the user, enhancing the overall user experience.

[0083] Conversely, the domain-specific LLM is specialized in processing and analyzing data pertinent to the user's physiological and emotional states. This model is specifically trained on datasets that include biometric and physiological data, user engagement metrics, and textual sentiment inputs related to well-being. It excels in interpreting and categorizing these inputs into structured outputs, providing precise evaluations necessary for accurate user status assessments.

[0084] The communication module orchestrates the collaboration between the two LLMs. Upon receiving input signals from the user, it routes these signals to the domain-specific LLM for detailed analysis. The domain-specific LLM processes the data to quantify emotional and physiological states, outputting structured categories such as stress, anxiety, or overall wellbeing on a predefined scale.

[0085] Subsequently, the communication module utilizes the generic LLM to generate dialogue based on the structured information provided by the domain-specific LLM. If the assessment indicates the need for an intervention, the communication module instructs the generic LLM to create prompts whose outcome addresses the user's current state, employing natural language that is both engaging and supportive. This dual-LLM configuration allows the system to deliver refined and contextually relevant interactions aimed at enhancing the user's well-being.

[0086] The interaction module is configured to maintain user engagement to sustain the achieved emotional and physiological states as follows:

[0087] The interaction module employs several strategies to ensure continuous user engagement. Initially, it utilizes personalized prompts generated based on the user's quantifiable emotional and physiological statuses. These prompts are crafted so their outcome encourages active interaction, such as suggesting specific activities, providing motivational messages, or guiding the user through relaxation exercises.

[0088] Additionally, the interaction module continuously monitors the user's engagement levels by tracking responses, adherence to recommendations, and overall interaction frequency. This information is processed to adjust the engagement strategy dynamically. For example, if a user shows decreased interaction, the module may increase the frequency of messages or modify the content to better capture the user's interest.

[0089] The interaction module leverages a combination of push notifications, reminders, and interactive sessions through the user interface to keep the user involved. Push notifications and reminders are timed to coincide with periods of inactivity or at optimal times based on the user's routine, encouraging them to re-engage with the system.

[0090] Through the user interface, the interaction module facilitates real-time dialogues, responding promptly to user inputs and providing instant feedback. This real-time interaction ensures that the user feels supported and heard, promoting consistent engagement.

[0091] Furthermore, in some embodiments, the interaction module incorporates gamification elements where applicable, such as setting achievable goals, providing rewards for maintaining certain wellness activities, and tracking progress. These elements create a sense of accomplishment and motivate the user to continue participating.

[0092] The module also supports personalized content delivery, such as articles, videos, or exercises tailored to the user's interests and needs, which is essential in maintaining prolonged engagement.

[0093] By employing these multifaceted strategies, the interaction module ensures that the user remains engaged, thereby sustaining the achieved emotional and physiological states.

[0094] In some embodiments, the interaction module is configured to perform validation of the user's status before continuing with further prompts as follows:

[0095] Upon generating an initial prompt based on the assessment of the user's emotional and physiological data, the interaction module initiates a validation process to confirm the user's current status. This process involves soliciting direct feedback from the user or obtaining additional signal measurements to verify the accuracy of the previously assessed state. Themodule may prompt the user with questions such as, "How are you feeling right now?" or request the user to perform specific actions that can be monitored, such as checking their heart rate again.

[0096] The validation process integrates real-time feedback and additional physiological data into the system for analysis by the LLM module, ensuring the LLM's assessments remain up-to- date. If the validated status aligns with the initial assessment, the system proceeds with the next steps as planned. Otherwise, the interaction module re-evaluates the situation, generating new prompts based on the updated user data. This ensures personalized and accurate responses tailored to the user's actual condition, maintaining the effectiveness of the interaction strategy aimed at enhancing user well-being.

[0097] Fig. 3 shows a flow chart illustrating the process for measuring and enhancing user wellbeing through large language models (LLMs). At step 300, signals are acquired from sensors measuring the user's physiological and neurological parameters and from the user's online interactions. These signals may include biometric data such as heart rate, signals from smartwatches, and user engagement metrics from online activities.

[0098] At step 310, the acquired signals are input to the LLM module. The LLM module, which is trained to assess and quantify emotional and physiological status, processes these inputs. Subsequently, at step 320, the emotional and physiological status of the user is assessed and quantified using the LLM module.

[0099] The next step involves decision-making at step 330, where it is determined whether the emotional or physiological status is below a predetermined value. If the status meets the required level, the process may end at step 370. However, if the status is below the predetermined value, the system proceeds to step 340.

[0100] At step 340, a new prompt aimed at improving the user's well-being is generated. This new prompt generates a message that is structured to transition the user from their current state to a more desirable state. Following this, step 350 maintains dialogue with the user based on the generated prompt.

[0101] Finally, step 360 executes continuous reassessment of the user's status using the LLM module. This step ensures that the user's well-being is continuously monitored and adjusted as necessary, maintaining a feedback loop to achieve and sustain the required physiological and emotional states.

[0102] In another aspect, the present invention relates to a method for evaluating and enhancing user well-being through large language models (LLMs), which comprises several key steps: receiving signals from sensors measuring a user's physiological and neurological parameters, as well as user online interactions; evaluating and quantifying the user's emotional and physiological status through an LLM based on the received signals; generating a new message if an emotional or physiological status is below a predetermined value to transition the user to another state; and engaging in ongoing interaction with the user, periodically reevaluating the user's status.

[0103] The method begins with the acquisition of signals from various sources. These signals include user online interactions, and optionally physiological and neurological parameters typically measured by sensors , . Physiological signals may be collected from wearable devices such as smartwatches, fitness bands, other devices such as a smartphone or smart water bottle, or other bio-sensing devices that monitor metrics like heart rate, sweating, and electrodermal activity. Neurological parameters can be inferred from certain physiological data or collected via specialized neuro-sensors. Additionally, the method incorporates signals from user online interactions, which can consist of textual inputs, application usage data, engagement metrics like likes, comments, clicks, and time spent on specific content.

[0104] Once the signals are received, the LLM evaluates and quantifies the user's emotional and physiological status. The LLM is pre-trained on datasets that include various forms of input signals related to user well-being. In this phase, the LLM processes the incoming data by parsing textual inputs, analyzing biometric measurements, and evaluating engagement metrics. The LLM quantifies these inputs and categorizes the user's status into structured categories such as emotional state, fear, anxiety, etc., each assessed on a predefined scale typically ranging from 1 to 5. For instance, a textual input expressing sadness might be assigned 'sadness: 4 out of 5,' and an elevated heart rate might be quantified as 'heart rate: 5 out of 5.'

[0105] If the user's emotional or physiological status is determined to be below a predetermined threshold value, the system proceeds to generate a new prompt aimed at transitioning the user from their current state to a more desirable one. This prompt and the message output by the prompt are crafted using cognitive computing models that leverage natural language processing and machine learning algorithms tailored to psychological or emotional improvement. For example, if the user's anxiety level is high, the generated message might suggest relaxation techniques, mindfulness practices, or engaging in a soothing activity.The messages are personalized based on the structured information obtained from the LLM's evaluation and are intended to guide the user towards better emotional and physiological health.

[0106] The method includes maintaining ongoing interaction with the user via a user interface that supports real-time communication. The interaction involves the system presenting the generated prompts to produce messages to the user and facilitating activities, exercises, or recommendations designed to improve their well-being. The interaction may employ various communication channels, such as text messages, in-app notifications, or voice commands, depending on the user's preferences and the system's capabilities.

[0107] To ensure continuous improvement and accurate monitoring, the system periodically re-evaluates the user's status using the LLM. This involves regularly collecting new signals and re-assessing the user's emotional and physiological states. The periodic assessments enable the system to adapt its strategies and prompts based on the user's real-time data and responses, ensuring that the interactions remain relevant and effective in enhancing the user's well-being.

[0108] The method of the invention facilitates a comprehensive approach to enhancing user well-being by integrating continuous signal acquisition, advanced LLM evaluation, personalized prompt generation, and ongoing user interaction. By leveraging the processing capabilities of LLMs, the method ensures an adaptive, real-time response to user needs, promoting sustained emotional and physiological health.

Claims

Claims1. A computer-implemented interactive system for processing, through large language models (LLMs), user data representing user well-being, and interacting with the user to increase the measured user data representing well-being, comprising:• a signal acquisition module configured to receive signals from user online interactions;• an LLM module trained to assess and quantify emotional and physiological status based on said received signals;• an assessment module configured to evaluate the output from the LLM module, and if an emotional or physiological status is below a predetermined value, then the assessment module is configured to generate a new prompt whose outcome is one or more messages configured to improve the measured user data representing user's well-being until said status is at the required level; and• an interaction module configured to maintain dialogue with the user and perform continuous reassessment of user status using the LLM module.

2. The system of claim 1, wherein the received signals comprise at least one of text input, application usage signals, engagement metrics such as likes, comments, or clicks.

3. The system of claim 2, wherein said engagement metrics comprise likes, comments, choice of emojis, clicks or time spent.

4. The system of claim 1, wherein the received signals comprise signals from sensors measuring a user's physiological and neurological parameters.

5. The system of claim 4, wherein the input signals comprise biometric data.

6. The system of claim 1, wherein the signal acquisition module is configured to receive signals from third-party devices providing physiological metrics.

7. The system of claim 1, further comprising: a secondary LLM configured to analyze physiological data such as heart rate and sweating received from auxiliary devices like smartwatches.

8. The system of claim 1, wherein the LLM module's evaluation includes detecting sentiment from textual inputs and signals.

9. The system of claim 1, wherein the LLM module outputs structured information including categories such as emotional state, fear, anxiety, each quantified on a scale from 1 to 5.

10. The system of claim 1, wherein the assessment module is configured to create prompts based on cognitive computing models aimed at psychological or emotional improvement.

11. The system of claim 1, further comprising a user interface to facilitate real-time communication between the user and the system.

12. The system of claim 1, wherein the communication module is configured to use a generic LLM for dialogue generation and a domain-specific LLM for data analysis.

13. The system of claim 1, wherein the interaction module is configured to maintain user engagement to sustain the achieved emotional and physiological states.

14. The system of claim 1, wherein the interaction module is configured to perform validation of the user's status before continuing with further prompts and / or messages.

15. A computer-implemented interactive method for processing, through large language models (LLMs), user data representing user well-being and interacting with the user to increase the measured user data representing well-being, comprising:• receiving signals from user online interactions;• evaluating and quantifying the user's emotional and physiological status through an LLM based on the received signals;• if an emotional or physiological quantified status is below a predetermined value, generating a new prompt whose outcome are one or more messages configured to transition the user from one emotional or physiological state to another so that said measured user data representing the user's emotional or physiological status reaches said predetermined value; and• engaging in ongoing interaction with the user and periodically re-evaluating the user's status.

16. The method of claim 15, wherein receiving signals comprises collecting textual inputs and engagement metrics such as likes, comments, or clicks from the user.

17. The method of claim 15, wherein evaluating the user's status includes categorizing emotional status, fear, and anxiety on a scale of 1 to 5 through the LLM.

18. The method of claim 15, wherein generating the new prompt includes using cognitive computing models targeted at psychological or emotional improvement.

19. The method of claim 18, further comprising validating the user's status before continuing with further prompts or messages.

20. The method of claim 15, wherein receiving signals comprises signals from sensors measuring a user's physiological and neurological parameters.

21. The method of claim 20, wherein receiving signals includes collecting physiological data such as heart rate and sweating from auxiliary devices like smartwatches.

22. The method of claim 15, wherein evaluating the user's status through the LLM includes detecting sentiment from textual inputs and signals.

23. The method of claim 15 further comprising: generating the new prompt based on iterative user feedback.

24. The method of claim 15, further comprising: maintaining sustained user engagement to achieve and hold targeted emotional and physiological states.

25. The method of claim 15, wherein re-evaluating the user's status includes periodic assessments based on ongoing user interactions and new signals received.

26. The method of claim 15, wherein generating a new prompt involves using a secondary LLM configured for analyzing distinct types of user input, such as emotional and physiological signals.

Images