Methods and systems for conversational agent and natural language processing interface for caregivers

Abstract

Embodiments described herein provide systems and methods for caregiver and / or patient support conversational agents based on user-generated content datasets. Embodiments described herein provide systems and methods for conversational agents that use natural language processing (NLP) models and code (e.g. executable code or instructions encoding operations for NLP) to process the user query and provide responses, predictions, and so on. In an aspect, embodiments described herein provide a computer system for training a user model using conversational features. In an aspect, embodiments described herein provide a computer-implemented method for training a user model using conversational features.

Description

METHODS AND SYSTEMS FOR CONVERSATIONAL AGENT AND NATURAL LANGUAGE PROCESSING INTERFACE FOR CAREGIVERSCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] The present application claims priority on United States Patent Application No. 63 / 703,527 filed October 4, 2024, United States Patent Application No. 63 / 706,362 filed October 11 , 2024, United States Patent Application No. 63 / 723,937 filed November 22, 2024 and United States Patent Application No. 63 / 723,943 filed November 22, 2024, the entire contents of which are incorporated herein by reference.FIELD

[0002] The improvements generally relate to the field of conversational agents, artificial intelligence, machine learning and natural language processing.INTRODUCTION

[0003] Embodiments described herein relate to conversational agents. In particular, embodiments described herein relate to conversational agents for performing and providing conversations for patient or dependent care. Further embodiments described herein relate to use of artificial intelligence (Al) to program, configure and / or refine the conversational agent.

[0004] There exists a need for improved conversational agents. For example, there exists a need for improved agents to support a patient or dependent care, and / or a caregiver. Embodiments described herein relate to a system configured for receiving indications from a caregiver and adapt the parameters of the conversational agent accordingly. Embodiments described herein relate to a system that is able to receive input data from a user interface accessible by a caregiver and adjust the parameters of the conversational agent. Embodiments described herein also relate to a conversational agent that can prompt a user with inquiries designed to generate data that can be used for sentiment analysis. Embodiments described herein relate to a computer system or method for reviewing this generated data against Cognitive Behavioral Therapy (CBT) and other therapy modalities to automatically make recommendations for caregivers.

[0005] There also exists a need for improved conversational agents for automatically diagnosing a cognitive function of a user. Conventionally, cognitive assessment is conducted via a series of questions queries by a mental health specialist who determines the cognitive functionof the patient based on the responses obtained from the questionnaire. However, it is generally inconvenient to perform mental health assessment on a regular basis, and the assessment may be inconsistent across different mental health specialists. For example, mental health patients can be confused when consulting in front of a specialist, as they can easily become bored and / or nervous, which can falsify the mental evaluations performed by the specialist. Various factors may cause nervousness, such as the mood of the patient, the relationship they have with the specialist, and so on. In addition, the treatment plan can be performed by more than one specialist, each having their own opinion of the mental state of the patient, making conventional cognitive assessment incongruous.

[0006] Conversational agents can assist users in their daily needs. Conversational agents are based on Large Language Models (LLMs), which is a type of artificial intelligence designed to understand and generate human language. They are generally built using deep learning techniques, which enables processing and generating text based on vast amounts of data.

[0007] Conversational agents based on LLMs carry out a specific function in a single response to a user input. Conventional conversational agent generally cannot have a sustained conversation with a user. Furthermore, the responses produced by the conversational agents may be biased or may contain inaccurate outputs, as the conversational agent learns from large databases that may contain biases.

[0008] Therefore, there is need for improvements or alternatives.SUMMARY

[0009] In an aspect, embodiments described herein provide systems and methods for conversational agents based on user-generated content datasets. Embodiments described herein provide systems and methods for conversational agents that use natural language processing (NLP) models and code (e.g. executable code or instructions encoding operations for NLP) to process the user query and provide responses, predictions, and so on.

[0010] In an aspect, embodiments described herein provide a computer system for training a user model using conversational features.

[0011] In an aspect, embodiments described herein provide a computer-implemented method for training a user model using conversational features.

[0012] In an aspect, embodiments described herein provide a computing system for conversational agents to support patient or dependent care, the system comprising: a device with a caregiver interface providing caregiver conversational content for patient or dependent care using a first conversational agent; a device with a patient interface providing patient conversational content using a second conversational agent, and receiving user feedback; a server system comprising: at least one memory storing thereon a natural language processing (NLP) engine and an interaction software application configured to provide the first conversational agent that generates the caregiver conversational content for patient or dependent care for provision to the caregiver interface, and the second conversational agent that generates the patient conversational content for provision to the patient interface; a processing system that includes one or more processors and one or more memories coupled with the one or more processors, the processing system configured to: receive, from the patient interface, a string data object representative of the user feedback received; identify, using the NLP engine, a first token representative of a user preference in the string data object and label the first token with a respective semantic category; update or train the second conversational agent using the labelled first token; generate the patient conversational content for the second conversational agent using the trained second conversational agent, the conversational content to support patient or dependent care; and generate a notification as part of the caregiver conversational content based on the user preference, the notification to support patient or dependent care.

[0013] In some embodiments, the second conversational agent is configured to refine the patient conversational content based on the user preference.

[0014] In some embodiments, the memory stores a user profile of data fields and corresponding values, a data field and corresponding value storing the user preference, wherein the second conversational agent is configured to refine the patient conversational content based on the user profile.

[0015] In some embodiments, the processing system is further configured to: identify, using the NLP engine, a second token representative of a cognitive pattern in the string data object and label the second token with a respective semantic category; and update or train the second conversational agent using the labelled second token.

[0016] In some embodiments, the processing system is further configured to generate a recommendation data object using the labelled second token and the first conversational agent,the recommendation data object including instructions for a caregiver to address the identified cognitive pattern of the user, wherein the caregiver conversational content comprises the instructions for the caregiver to address the identified cognitive pattern of the user.

[0017] In some embodiments, said updating further comprises updating a model parameter of a model for the first conversational agent associated with the second labelled token using the recommendation data object.

[0018] In some embodiments, the cognitive pattern includes at least one of perceptive functions, memory functions, problem-solving functions, judgment functions, emotional functions and social cognition functions.

[0019] In some embodiments, the processing system is further configured to: receive an indication to start a mental state evaluation procedure; provide the patient interface with the patient conversational content including pre-determined questions of the mental state evaluation procedure; and generate a cognitive function report using the user feedback and the patient conversational content, wherein the user feedback includes a response to one or more of the predetermined questions.

[0020] In some embodiments, the processing system is further configured to record a duration between said providing the conversational content and said receiving the string data object, further wherein said generating the cognitive function report is performed using the recorded duration.

[0021] In some embodiments, the processing system is further configured to stop the mental state evaluation procedure if the recorded duration is greater than a predetermined threshold or if the NLP engine determines that the response does not semantically correspond to a desired response of a corresponding pre-determined question.

[0022] In some embodiments, the string data object includes a conversation log including a plurality of string data objects each representative of corresponding user feedback and a plurality of string data object each representative of patient conversational content.

[0023] In some embodiments, the processing system updates a user profile with the first token representative of the user preference, the user profile storing a data field corresponding to the user preference and a data value corresponding to the first token, wherein the user preference indicates a like or dislike.

[0024] In some embodiments, the string data object includes data representative of instructions received from the caregiver interface, wherein the user preference includes the received instructions, wherein the caregiver conversational content is generated using a recommendation data object including instructions for a caregiver.

[0025] In some embodiments, the first token representative of the user preference defines functionality of the second conversational agent to be adjusted when generating the patient conversational content.

[0026] In some embodiments, the string data object representative of the user feedback received comprises data indicating a health concern or health condition, wherein the processing system updates a user profile with the data indicating the health concern or health condition, and wherein the patient conversational content is generated based on the data indicating the health concern or health condition.

[0027] In an aspect, embodiments described herein provide a computing device for training or updating a conversational agent to support patient or dependent care. The device has: at least one memory storing thereon a natural language processing (NLP) engine and an interaction software application configured to provide one or more models for a conversational agent that generates conversational content for performing and providing conversations for patient or dependent care; a processing system that includes one or more processors and one or more memories coupled with the one or more processors, a user interface and a database, the processing system configured to cause the computing device to: receive, from the user interface, a string data object representative of a user feedback received; receive, from the database, a model of the conversational agent including a plurality of model parameters; identify, using the NLP engine, a first token representative of a user preference in the string data object and label the first token with a respective semantic category; associate the labelled first token with a corresponding model parameter from the plurality of model parameters; update the model parameter associated with the first labelled token using the labelled first token and train the model of the conversational agent using the updated model parameters; and generate conversational content for the conversational agent using the trained model, the conversational content to support patient or dependent care.

[0028] In some embodiments, the associated model parameter is configured to refine the conversational content generated by the conversational agent based on the user preference.

[0029] In some embodiments, at least one model parameter of the plurality of model parameters includes data fields and corresponding values obtained from a user profile of the user.

[0030] In some embodiments, the processing system is further configured to cause the computing device to: identify, using the NLP engine, a second token representative of a cognitive pattern in the string data object and label the second token with a respective semantic category; and associate the labelled second token with a corresponding model parameter from the plurality of model parameters.

[0031] In some embodiments, the processing system is further configured to cause the computing device to generate a recommendation data object using the labelled second token, the recommendation data object including instructions for a caregiver to address the identified cognitive pattern of the user.

[0032] In some embodiments, said updating further comprises updating the model parameter associated with the second labelled token using the recommendation data object.

[0033] In some embodiments, the cognitive pattern includes at least one of perceptive functions, memory functions, problem-solving functions, judgment functions, emotional functions and social cognition functions.

[0034] In some embodiments, the processing system is further configured to cause the computing device to: receive an indication to start a mental state evaluation procedure; provide the user interface with the conversational content including pre-determined questions of the mental state evaluation procedure; and generate a cognitive function report using the user feedback and the conversational content, wherein the user feedback includes a response to one or more of the pre-determined questions.

[0035] In some embodiments, the processing system is further configured to cause the computing device to record a duration between said providing the conversational content and said receiving the string data object, further wherein said generating the cognitive function report is performed using the recorded duration.

[0036] In some embodiments, the processing system is further configured to cause the computing device to stop the mental state evaluation procedure if the recorded duration is greater than a predetermined threshold or if the NLP engine determines that the response does not semantically correspond to a desired response of a corresponding pre-determined question.

[0037] In some embodiments, the string data object includes a conversation log including a plurality of string data objects each representative of corresponding user feedback and a plurality of string data object each representative of conversational content.

[0038] In some embodiments, the processing system updates a user profile with the first token representative of the user preference, the user profile storing a data field corresponding to the user preference and a data value corresponding to the first token, wherein the user preference indicates a like or dislike.

[0039] In some embodiments, the string data object includes data representative of instructions received from a caregiver interface, wherein the user preference includes the received instructions, wherein the conversational content is generated using a recommendation data object including instructions for a caregiver.

[0040] In some embodiments, the first token representative of the user preference defines functionality of the conversational agent to be adjusted when generated the conversational content for assisting the user.

[0041] In some embodiments, the string data object representative of the user feedback received comprises data indicating a health concern or health condition, wherein the processing system updates a user profile with the data indicating the health concern or health condition, and wherein the conversational content is generated based on the data indicating the health concern or health condition.

[0042] In another aspect, there is provided a computer-implemented method for defining parameters for an interaction software application running a conversational agent for assisting a user, the method comprising: receiving an input signal representative of a user command including an indication for the conversational agent, the user command being in the form of human language and the indication defining a functionality of the conversational agent to be adjusted when assisting the user; using natural language processing, process the input signal to identify the indication and associate the indication with a corresponding parameter of the interaction software; generating an updated value of the parameter based on the identified limitation; and providing the updated value to the conversational agent.

[0043] In another aspect, there is provided a computer-implemented method for analyzing a conversation between a user and a conversational agent, the method comprising: receiving aninput signal representative of the conversation between the user and the conversational agent, the conversation including user entries; using natural language processing, processing the input signal to identify patterns in the user entries; and generating a treatment based on the identified patterns.

[0044] In another aspect, there is provided a computer-implemented method for performing a mental state evaluation using a conversational agent, the method comprising: receiving an indication to perform the mental state examination based on a cognitive tool using a conversational agent; generating a conversational content based on a user profile using natural language processing; providing an output signal including the conversational content to a user interface; in response to providing the output signal, receiving an input signal from the user interface, the input signal being representative of user feedback, and adapting the user profile based on the user feedback; in response to said receiving the indication, retrieving the cognitive tool from a database, and generating a series of questions using the cognitive tool, the cognitive tool being configured for assessing the cognitive function of a user; sequentially introducing each question in the conversational content; using the natural language processing, identifying a response for each question in the user feedback, and associating the response with each respective question; and generating a cognitive function report using the series of question and the associated responses.

[0045] In another aspect, there is provided a computer-implemented method for training a user model using conversational features, the method comprising: receiving, from a database, a user model representative of a user profile; receiving, from a user interface, an input signal representative of a user feedback; using natural language processing, processing the input signal to identify features representative of a user preference and / or an emotional behaviour in the user feedback; updating the user profile using the identified features and; using machine learning, training the user model with the updated user profile.

[0046] In another aspect, there is provided a non-transitory machine-readable medium having stored thereon a plurality of instructions that, when executed by at least one computing device, cause the at least one computing device to perform any of the methods herein.DESCRIPTION OF THE FIGURES

[0047] FIG. 1 shows a schematic diagram of a computer system 100 according to embodiments described herein.

[0048] FIGS. 2A-2E show diagrams of methods 200, 220, 240, 260 according to embodiments described herein.

[0049] FIGS. 3A-3C shows block diagrams of systems 300, 330, 360 according to embodiments described herein.

[0050] FIGS. 4A-4E shows example interfaces 400, 410, 420, 430, 440 according to embodiments described herein.

[0051] FIG. 5 shows a diagram 500 for caregiver parameters used to check and influence Al agent responses to an end user according to embodiments described herein.

[0052] FIG. 6 shows exemplary analysis graphs 600, 650 according to embodiments described herein.

[0053] FIG. 7 shows a schematic diagram of an electronic device 700 according to embodiments described herein.DETAILED DESCRIPTION

[0054] Embodiments described herein provide systems and methods for data solicitation, natural language processing (NLP), and conversational agent programming. Embodiments described herein provide methods for processing a caregiver command and updating parameters of the conversational agent accordingly. Further embodiments described herein relate to use of artificial intelligence to process the user input using an NLP engine, and to define parameters of the conversational agent. The parameters, as will be described bellow, may vary, but the commands are suited for adjusting the parameters to the need of the patient of dependent care. Embodiments described herein provide improved interfaces for setting and updating parameters of the conversational agent to avoid the use complex interfaces to set parameters for the agent. A user may not have technical knowledge to use complex interfaces, and the improved interface enables users without such technical knowledge to configure parameters for the agent.

[0055] Methods and systems described herein can significantly streamline and improve caregiver input on parameters that influence and control the function of the system. Furthermore, the methods and systems described can enable any user to specify behavioral parameters for a system used by other users, particularly where there is a dependant / caregiver relationship as common in eldercare, parental systems, or educational systems.

[0056] Further embodiments described herein provide conversational agents for performing and providing sentiment analysis and evaluation for patient or dependent care. Further embodiments described herein relate to use of artificial intelligence to assess conversational records to derive sentiment analysis and recommend behavioral treatments for one or more caregivers of a dependent or patient. Embodiments described herein provide an improved approach for digesting large bodies of conversation with a patient or dependant. This aspect can be followed by summarizing the user status and other details, and making useful recommendations for a caregiver, teacher, provider, or other responsible party.

[0057] Further embodiments described herein relate to implementing a conversational agent for performing a mental state examination. The conversational agents presented herein are configured to follow the mental behavior and mood of a user on a daily basis and to recurrently have discussion with the patient. Such environments have been found to be optimal to perform a mental state examination, as the patient interacting with the conversational agent is generally in a calm state because of the ability of the virtual agent to tailor the conversational content around the personality of the user. Embodiments described herein support recurrent interactions between the user and the conversational agent, the introduction of mental state examinations in those interactions is often uncomplicated and undemanding as opposed to mental state examinations performed by specialists.

[0058] Embodiments described herein provide systems and methods for data solicitation, natural language processing (NLP), and conversational agent programming. Embodiments described herein provide devices and methods for diagnosing a cognitive function of a user. Further embodiments described herein relate to use of artificial intelligence to automatically process the user input using, e.g., NLP processes, and to determine cognitive feature is mental examination results.

[0059] Cognitive tools such as a Mini-Mental State Examination (MMSE) can be used to diagnose cognitive functions. However, the output of such tools need to be evaluated by a specialist to produce a proper diagnostic. The methods and system presented herein use artificial intelligence (Al) to automatically produce an examination report, and does so, in a reproducible manner. While an MMSE is a cognitive tool implementable using the present technology, it will be understood the that other types of examinations and surveys may be used, such as Standardized MMSE (SMMSE), Montreal Cognitive Assessment (MoCA), Mini-Cog, General Practitioner Assessment of Cognition (GPCOG), custom-made surveys, and the like. The cognitive toolsdescribed herein may be provided in the form of a predetermined set of questions configured for assessing the cognitive function of the user, or a set of program instructions for generating the set of questions.

[0060] Testing, evaluations and examinations may not be done consistently which impacts the accuracy and integrity of the results. Embodiments described provide devices and methods for diagnosing a cognitive function using a conversational agent or agent to improve administration of tests and examinations to improve consistency. Testing and examinations may involve human err and bias which impacts the accuracy and integrity of the results. For example, a test may have a timing-based score (e.g. how quickly or how slowly the test taker completes a test activity) and a human administrator may introduce error in the accuracy of the timing in relation to their reaction time or if they do not start or stop of the timing consistently or accurately. For example, there may be a delay in stopping the timing. Further, an evaluation may be repeatedly done for a user. Accurate benchmarking of multiple evaluations may require time or date consistency. This may help track improvements or compare results over time. Embodiments described provide devices and methods for diagnosing a cognitive function that can be consistently repeated with a programmed schedule and prompts for the user to take the evaluation or test. Further, users or test takers may require simple interfaces that are easy to use and not too complex. If the interface is confusing for the user then the test results may not be accurate as the interface may have caused delays or other impacts in the testing.

[0061] Further embodiments described herein provide methods for training a user model using conversational features. Further embodiments described herein relate to the use of artificial intelligence to process the user input using, e.g., NLP, and to update a user profile using features extracted from the user input.

[0062] Provided herein are systems and methods for training a conversational agent based on a user model. The user model is constantly updated with features extracted from conversations held between the conversational agent and the user. A user model is trained, updated and refined for a specific user. That is, user models can be differential across different users. A user model can be trained, updated and refined for a specific user using user feedback. Using bespoke and tailored dataset cultivated through user feedback, the conversational agent is able to create custom, tailored conversational content that promotes user engagement and create connections between the conversational agent and the user. The ability of the conversational agent to recursively extract features in the user feedback and refine the user model using thefeatures enables the conversational agent to be reactive to the emotional response of the user when interacting with the latter. This provides an improved agent configured specifically for patient support.

[0063] In the context of mental healthcare, elder care, teaching, and other types of care and education of the like, it may prove challenging to capture the user’s interest and attention, as users have specific conditions and needs that must be taken into consideration during interactions. For mental health specialists, caregivers, teachers, and the like, it takes years of formation and practice to become efficient in understanding the needs of the patient and providing the right treatment or education. Even so, specialists often rely on physical and mental notes taken during user interaction or after reading through background documentation on the user. When interacting with patients, specialists are thus limited when recollecting information about their patient, which can be difficult when having a large number of patients.

[0064] The present technology uses custom User Generated Content (UGC) databases per user. Using a user interface, the conversational agent communicates with the user to collect data a builds a data structure that stores emotional patterns, behaviors, cognitive patterns, memories, and any other type of information received from the user. The use of this data structure enables the conversational agent to build a user profile that tailors to the user. The data object may be a collection of vectors, matrices, scalars and a combination thereof.

[0065] FIG. 1 shows a schematic diagram of a computer system 100 for training a user model using conversational features. System 100 can have a server 104 that connects, by a network 170, to one or more user devices 180. In some embodiments, a user device 180 can have a conversational agent 102. In some embodiments, a user device 180 can have a caregiver interface 106. In some embodiment, system 100 can be distributed across a plurality of devices or machines or nodes which can be clients, servers, peers, and so on. The system 100 can utilize computing resources across the devices or machines or nodes. The devices or machines can be physically separate computing resources.

[0066] The system 100 connects user devices 180 to exchange data and commands. In some embodiments, user device 180 can be implemented by an electronic device with memory storing instructions and one or more processors that execute the instructions. In some embodiments, user device 180 has at least one processor, memory, and at least one interface (e.g. I / O interface, network interface, user interface).

[0067] In some embodiments, user device 180 can be implemented by an electronic device with memory storing instructions for the conversational agent 102 and one or more processors that execute the instructions. The conversational agent 102 can prompt a user with inquiries designed to generate data that can be used for input data processing. In some embodiments, the user device 180 can store (e.g. in memory) the data received in response to the prompts. In some embodiments, the user device 180 can transmit the data to server 104 for processing and storage. The memory can store instructions for the agent 102 to generate one or more user interfaces to collect data, and exchange data with server 104 and / or other user devices 180 of the system 100. The conversational agent 102 can provide a user interface to provide output and receive input. In some embodiments, the conversational agent 102 has an interface to provide visualizations of data received at a display device.

[0068] In some embodiments, user device 180 can be implemented by an electronic device with memory storing instructions for a caregiver interface 106 and one or more processors that execute the instructions. The caregiver interface 106 can receive the user input data from the caregiver and provide the user input data to the agent 102. Further, system 100 may generate prompts or display information to the caregiver via the caregiver interface 106.

[0069] In some embodiments, server 104 has hardware components such as at least one processor 110, memory 120, at least one I / O interface 130, at least one network interface 140, and an application programming interface (API) 150. Memory 120 can include instructions (e.g. for different computing applications) executable by the processor 110 for generating and managing conversational agent(s) 102. Memory 120 can store data generated by the conversational agent 102 and its prompts.

[0070] The I / O interface 130 enables the server 104 to interconnect with one or more input devices, such as a keyboard, mouse, camera, touch screen and a microphone, or with one or more output devices such as a display screen and a speaker. The network interface 140 enables the server 104 to communicate with other components, to exchange data with other components, to access and connect to network resources, to serve applications, and perform other computing applications by connecting to a network 170 (or multiple networks, or a combination of different networks) capable of carrying data. The hardware components of the server 104 may be connected in various ways including directly coupled, indirectly coupled via a network, and distributed over a wide geographic area and connected via a network (which may be referred to as “cloud computing”). For example, and without limitation, server 104 may be a networkappliance, embedded device, computer expansion module, or other computing device capable of being configured to carry out the methods described herein. In some embodiments, memory 120 can be distributed storage devices.

[0071] The systems and methods described herein may involve artificial intelligence (Al), such as machine learning, natural language processing (NLP) application(s) and the like that may be implemented using hardware, software, and a combination thereof. For example, NLP software may be used to break down input data using, e.g., tokenization, parsing, semantic analysis, and machine translation, into processable data. In some cases, NLP software may include Large Language Models (LLMs) trained on large quantities of text data and can be used for generation, translation and summarization of text. Generally, the Al components can be trained using large amount of data that can include various types of user conversations, patterns associated with given text data, theoretical concepts related to the field, and the like.

[0072] The terms “conversational agent”, “virtual agent”, “agent”, “Al agent”, “virtual assistant” or “virtual companion” used herein may refer to a software program or system configured to simulate human conversation and autonomously perform tasks on behalf of a user, such as a caregiver, or another system. For example, FIG. 1 provides an illustration of a system 100 that includes a caregiver interface 106 and an agent 102 which may involve a virtual assistance implementation. Furthermore, the Al agent 102 may be configured for actively generating conversational content for the user. The Al agent 102 is thus configured for transforming user input, such as an input signal representative of a user commend received by the caregiver, into data that can be interpreted for adjusting parameters of the conversational agent. The Al agent 102 may be configured to receive user input via a user interface, process the user input and provide an output signal to the user interface. From the user perspective, the Al agent generally corresponds to a virtual companion or virtual caregiver that can be talked to via the user interface.

[0073] The agent 102 can use NLP, dialogue management, knowledge base, a user interface and machine learning to understand and respond to user inputs in a conversational manner. The NLP component allows the agent to understand and interpret human language. It involves tasks such as tokenization, part-of-speech tagging, named entity recognition, and sentiment analysis. The machine learning component help the agent 102 learn from data and improve its responses over time. Techniques such as supervised learning, unsupervised learning, and reinforcement learning may be used. The dialogue management component manages theflow of conversation, ensuring that the agent responds appropriately to user inputs and maintains context throughout the interaction. The knowledge base is a repository of information that the agent can draw upon to provide accurate and relevant responses. This can include databases, APIs, and other sources of structured and unstructured data. Users interact with the agent via the user interface, which can be text-based (e.g., chat windows) or voice-based (e.g., virtual assistants on smart devices), or a combination, for example.

[0074] As shown in the example of Fig. 1 there may be multiple agents 102 for different user devices 180. An agent 102 can be uniquely configured for a specific user. That is, agent 102 can be differentially configured for a specific user. Memory 120 stores models for conversational agents. There may be one or more models configured or refined for a specific user, for example. Models for NLP are configured to understand, interpret, and generate human language. These models are built using various techniques and can be categorized into several types: rule-based models, statistical models, machine learning models, deep learning models, hybrid models, and so on. These models are use for developing and executing conversational agents 102, language translation systems, sentiment analysis tools, and other applications described herein. They enable system 100 to understand and generate human language, making interactions with technology more natural and intuitive. Rule-based models can rely on predefined linguistic rules to process language. They can be used for tasks like tokenization and part-of-speech tagging. Statistical models can use statistical methods to analyze and predict language patterns. Examples include Hidden Markov Models (HMMs) and Conditional Random Fields (CRFs), which are used for tasks like named entity recognition and sequence labeling. Machine learning models learn from data to improve their performance over time. Examples include Support Vector Machines (SVMs), Decision Trees, and Naive Bayes classifiers. They are used for tasks like text classification and sentiment analysis. Deep learning models use neural networks with multiple layers to process language. Examples include Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), which are used for tasks like text generation and machine translation. More advanced models like Transformers (e.g., BERT, GPT) have revolutionized NLP by providing state-of-the-art performance in various tasks. Hybrid models combine different approaches to leverage the strengths of each. For example, a hybrid model might use rule-based methods for initial processing and deep learning for more complex tasks.

[0075] The term natural language processing (NLP) engine refers to a software engine or set of applications configured for analyzing and interpreting human language input, often in the form of a string data object, to generate semantically labelled tokens. The NLP engine usesmodels, including machine learning models and linguistic rules, to process text data, identify patterns, and extract semantic meaning. The NLP engine integrates components such as tokenization, part-of-speech tagging, named entity recognition, and syntactic parsing to understand the structure and context of the input. The NLP engine is designed to handle various languages and dialects, providing accurate and contextually relevant outputs for applications in the field of cognition assessment.

[0076] The NLP engine can be a special purpose machine such as a physical computer server or rack mounted appliance that resides in a data center or in distributed computing resources and is coupled to a message bus for receiving the user input through an application programming interface, connecting with disparate data tables to generate conversational content and executing specific user inputs. When the conversational content is generated from the recognition of user preferences and cognitive patterns, the content can be encapsulated as a data object to be communicated across the message bus, for example for generating a response to the user’s query.

[0077] The term “interaction software” refers to a collection of instructions, data and program being processed by a processing device and stored on a memory. The interaction software is configured for running or executing the conversational agent and comprises a plurality of parameters that are used to control the conversational agent. One or more interaction software applications are configured to provide one or more models for the various conversational agents 102 to generate conversational content. The content can be personalized for different users, and various conversations with users. The content maintains the context of the conversation. The content can be questions or responses tailored to the individual user, taking into account their preferences, history, and specific needs. The encourages user participation and responds to user inputs in real-time, creating a dynamic exchange of information. The content aims to achieve specific objectives, such as capturing input for treatment and diagnosis, and can also achieve specific objectives such as answering questions, providing information, or guiding users through a cognitive treatment or assessment process.

[0078] The term “user interface” is a system that is able to receive user input and output data from the Al agent. The user interface may be equipped with a keyboard, a microphone and the like to receive the user input, and may be equipped with a screen, a speaker, a headphone and the like to output data to the user. The user interface can also include one or more visual elements or visualization corresponding to output data. In some cases, the software coupled tothe user interface may be equipped with “speech-to-text” and “text-to-speech” modules, which convert an audio signal into a text signal and vice versa. Such modules are preferably implemented in cases where the user interface includes a microphone and speakers or headphones, in which cases the user communicates with the Al agent verbally.

[0079] The user interface can receive string data objects representative of user feedback. The user feedback can be used to train models specific to users, for example. A string data object is a type of data structure used in programming to represent a sequence of characters. Strings can be used to store and manipulate text. Strings can be immutable, meaning that once a string is created, it cannot be changed. Any modification to a string results in the creation of a new string. A string is essentially a sequence of characters, which can include letters, numbers, symbols, and whitespace. Strings can be indexable, meaning that you can access individual characters within the string using their position (index). Strings can be concatenated, or joined together, to form new strings. Strings have a length, which is the number of characters they contain. Many programming languages provide built-in methods and functions for manipulating strings.

[0080] The term “pattern” used herein refers to any type of correlation, trend, structure, regularity or irregularity identifiable in the user input. The patterns may be associated with a cognitive response, a sentimental response, an emotional response, recurrent concepts, abnormal concepts and any other structural form found in the user input. While some features of the present disclosure may be directed towards particular patterns, such as emotional patterns, it will be appreciated that any kind of pattern may be identified and processed in order to create feedback to the user and / or behavioral treatment.

[0081] The terms “treatment”, “behavioral treatment”, “therapeutic techniques” or “Cognitive Behavior Therapy (CBT)" used herein refers to a non-medical therapy that is provided to assess and treat mental health. The treatments proposed herein aim at breaking down mental issues of the user into small and manageable issues. It is thus understood that the treatments proposed herein are not configured for treating diseases or disorders.

[0082] The term “caregiver” used herein relates to someone who provides assistance and support to individuals who may have difficulty managing daily activities on their own. Various types of caregivers may be contemplated, such as family caregivers, professional caregivers, respite caregivers: They temporarily relieve primary caregivers to give them a break, specialized caregivers and so on. In some cases, the caregivers are specialized in assistant elderlyindividuals. In other cases, the caregivers are specialized in assisting children in their daily life an / or in their education.

[0083] The methods described herein may be processed on a variety of consumer grade hardware and software platforms, such as, but not limited to, Apple™, PC™, Google™, Android™, Xaomi™, and so on.

[0084] The computing system 100 for conversational agents provides support patient or dependent care. The system 100 has device 180 with a caregiver interface 106 providing caregiver conversational content for patient or dependent care using a first conversational agent. The system 100 has a device 180 with a patient interface providing patient conversational content using a conversational agent 102 and also for receiving user feedback.

[0085] The server 104 has at least one memory 120 storing thereon a natural language processing (NLP) engine and an interaction software application configured to provide the first conversational agent that generates the caregiver conversational content for patient or dependent care for provision to the caregiver interface 106, and the second conversational agent 102 that generates the patient conversational content for provision to the patient interface. The server 104 has one or more processors 110 configured to: receive, from the patient interface (agent 102), a string data object representative of the user feedback received. The processor 110 can identify, using the NLP engine, a first token representative of a user preference in the string data object and label the first token with a respective semantic category. The processor 110 can update or train the second conversational agent 102 using the labelled first token, and generate the patient conversational content using the trained second conversational agent 102. The conversational content to support patient or dependent care. The processor 110 can generate a notification as part of the caregiver conversational content based on the user preference, the notification to support patient or dependent care. The processor 110 can transmit the notification to the caregiver interface 106.

[0086] In some embodiments, the second conversational agent 102 is configured to refine the patient conversational content based on the user preference.

[0087] In some embodiments, the memory 120 stores a user profile of data fields and corresponding values, a data field and corresponding value storing the user preference. The second conversational agent 102 is configured to refine the patient conversational content based on the user profile.

[0088] In some embodiments, the processor 110 is further configured to: identify, using the NLP engine, a second token representative of a cognitive pattern in the string data object and label the second token with a respective semantic category; and update or train the second conversational agent 102 using the labelled second token.

[0089] In some embodiments, the processor 110 is further configured to generate a recommendation data object using the labelled second token and the first conversational agent (caregiver interface 106), the recommendation data object including instructions for a caregiver to address the identified cognitive pattern of the user. The caregiver conversational content comprises the instructions for the caregiver to address the identified cognitive pattern of the user. In some embodiments, the cognitive pattern includes at least one of perceptive functions, memory functions, problem-solving functions, judgment functions, emotional functions and social cognition functions.

[0090] In some embodiments, the processor 110 is further configured to: receive an indication to start a mental state evaluation procedure; provide the patient interface with the patient conversational content including pre-determined questions of the mental state evaluation procedure; and generate a cognitive function report using the user feedback and the patient conversational content, wherein the user feedback includes a response to one or more of the predetermined questions.

[0091] In some embodiments, the processor 110 is further configured to record a duration between said providing the conversational content and said receiving the string data object, further wherein said generating the cognitive function report is performed using the recorded duration.

[0092] In some embodiments, the processor 110 is further configured to stop the mental state evaluation procedure if the recorded duration is greater than a predetermined threshold or if the NLP engine determines that the response does not semantically correspond to a desired response of a corresponding pre-determined question.

[0093] In some embodiments, the string data object includes a conversation log including a plurality of string data objects each representative of corresponding user feedback and a plurality of string data object each representative of patient conversational content.

[0094] In some embodiments, the processor 110 updates a user profile with the first token representative of the user preference, the user profile storing a data field corresponding to the user preference and a data value corresponding to the first token, wherein the user preference indicates a like or dislike.

[0095] In some embodiments, the string data object includes data representative of instructions received from the caregiver interface 106, wherein the user preference includes the received instructions, wherein the caregiver conversational content is generated using a recommendation data object including instructions for a caregiver.

[0096] In some embodiments, the first token representative of the user preference defines functionality of the second conversational agent 102 to be adjusted when generating the patient conversational content.

[0097] In some embodiments, the string data object representative of the user feedback received comprises data indicating a health concern or health condition, wherein the processor 110 updates a user profile with the data indicating the health concern or health condition, and wherein the patient conversational content is generated based on the data indicating the health concern or health condition.

[0098] FIG. 2A shows a diagram of a method 200 for defining parameters for an interaction software running a conversational agent 102 for assisting a user. It will be appreciated that the method 200 may be implemented by a computing device running the agent 102 or being in communication therewith. The method 200 aims at restricting the parameters of the agent 102 based on the commands given by the caregiver. Broadly, the caregiver may review the conversation logs between the conversational agent 102 and the user, or any other pertinent documentation such as cognitive examination, emotional examination, psychologic and / or psychiatric reports and the like. Based on the documentation, the caregiver may use the caregiver interface 106 to write commands in its own words for tuning the agent 102 to better emotionally assist the user. It will be appreciated that the method 200 may be implemented using a plurality of user input signals.

[0099] At step 202, system 100 receives a user input signal or input signal that includes an indication defining a functionality of the conversational agent 102. As mentioned above, the input signal may be received via the user interface 106 from the caregiver. The input signal may be typed in using a keyboard and may be provided in the form of a phrase. Alternatively, the inputsignal may be a voice signal transformed into a text signal using a speech-to-text process. The input signal may include a signal word, a plurality of words or a plurality of phrases. It will be appreciated that the input signal is generally representative of an indication given by the caregiver to set a functionality for the conversational agent 102 in the interaction software, and so, often in a technical jargon.

[0100] The functionality of the conversational agent 102 defined by the caregiver is generally provided in the user input signal as a high-level function or restriction, defined in the words of the caregiver. The functionality may largely vary, and may correspond to suggesting or prohibiting conversation topics, setting trigger alerts when the patient addresses certain topics or demonstrate particular cognitive behavior, defining periods where the conversational agent 102 should engage with the patient, and so on. In some cases where alerts are set, the caregiver may indicate to the conversational agent 102 to approach a particular topic with the user upon the trigger of the alert, and / or to notify the caregiver when the alert is triggered. It will be appreciated that the phrase containing the functionalities defined by the caregiver may generally be understood by another human being.

[0101] In some embodiments, prior to step 202, system 100 may be configured for prompting to the caregiver a request, such as a request to provide instructions for the conversational agent 102, via the caregiver interface 106. In some case, instead of being a user command, the input signal may be provided in the form of a query on the patient status. In such cases, system 100 may be configured to recognize the request from the caregiver and provide the requested information to the caregiver interface instead of performing the method 200.

[0102] At step 204, system 100 processes the input signal to identify the indication and associates the indication with a corresponding parameter of the agent 102. It will be appreciated that the processing may be performed by a program configured to be able to identify the functionalities in the user input signal. In some embodiments, the program codifies a process that includes discretizing the user input signal into a plurality of tokens. The discretization can be performed using techniques such as tokenization, parsing, semantic analysis, language modeling, discourse processing and so on. In some embodiments, the user device 108 is configured to process the input signal. In some embodiments, the server 104 is configured to process the input signal.

[0103] In some cases, at step 204, the user input signal may be pre-processed to remove undesired data, such as punctuation, articles, stop words, etc. Furthermore, the pre-processing may include formatting the user input data into a format processable by the user device 180 and / or the server 104. In some embodiments, the user device 108 is configured to pre-process the input signal e.g. before provision to server 104 for further processing. In some embodiments, the server 104 is configured to pre-process the input signal.

[0104] Once the user input data is processed into tokens, features are extracted from the data using, e.g., a machine learning model. For instance, each token may be weighted, labelled and / or vectorized depending on various factors, such as the content of the token, the frequency of the token in the user input signal, and the like, thereby capturing the semantic relationship amongst the tokens. In some embodiment, the machine learning model may be trained on labelled dataset to improve the extraction of features on the tokens. The features extracted from the tokens are representative of various parameters and values thereof of the conversational agent. For instance, the features may correspond to dialogue management parameters, response generation parameter, context awareness parameters, personalization parameters, emotion recognition parameters and the like. Furthermore, each parameter is associated with a value based on the extracted feature. It will be appreciated that the format of the value may vary, and be, e.g., an increase of the parameter, a decrease of the parameter, a type of topic to be promoted or restricted, a trigger alert and so on. For example, features may include increase the context awareness of the conversational agent 102 on a given topic, trigger an alert when a patient demonstrates a particular emotion and upon trigger of the alert, divert the conversation towards another given topic, prohibit the agent 102 to communicate with the user for given periods of the day and / or the week and the like.

[0105] At step 206, system 100 generates an updated value of the parameter of the agent 102 for each feature extracted from the user input. In order to do so, system 100 processes the value of the parameter obtained from the NLP code. In some embodiments, the previous value of the parameter may be used in the processing of the updated value, e.g., by weighing the value obtained from the NLP code with the previous value. In such cases, system 100 may prevent or compel radical changes to the behaviour of the conversational agent 102. In other cases, the updated value is the value obtained from the NLP code.

[0106] At step 208, system 100 provides the updates value to agent 102 via the interaction software stored on the memory 120. Once step 208 has been processed, the parameters of theconversational agent 102 will be updated based on the input of the caregiver in future conversations with the patient.

[0107] In some embodiments, once the updated value is provided to the agent 102, system 100 may provide to the caregiver via the caregiver interface 106 an indication that the updated value has been updated for the parameter. In such cases, system 100 may be configured to reprocess the updated value and the parameter to obtain a functionality that is introduced in a phrase understandable by the caregiver.

[0108] FIG. 2B shows a diagram of a method 220 for analyzing a conversation between a user and an agent 102. It will be appreciated that the method 220 may be implemented by a computing device running the agent 102 or being in communication therewith. The method 220 aims at generating a behavioral treatment for a user that communicates with the Al agent by generating queries and obtaining user input in response. It will be appreciated that the method 220 may be implemented during one or more conversations between the user and the agent 102. In such cases, multiple queries may be provided to the user and multiple user input signals may be received in response.

[0109] At step 222, queries are prompted to the user via a user interface. That is, user device 180 can provide prompts at a user interface or agent 102. The queries are generally formulated in the form of a sentence and aim at evaluating the mental state of the user. For instance, exemplary queries may be provided as “How are you feeling today?”, “What meal did you have for breakfast?”, “Can you tell me more about your brother?”, “Can you remember the name of your first pet?” and so on. While queries may be formulated in the interrogative, declarative sentences may also be provided to the user, such as “I enjoy having conversations with you.”, “I also like dogs.”, and the like. It will be appreciated that a single query or a plurality of queries may be provided to the user, depending on the embodiment.

[0110] In some cases, step 222 may be omitted, as the computing device may be receiving user input without prompting any query. For instance, the user may voluntarily input a sentence in the user interface, unsolicited by the agent 102.

[0111] At step 224, a user input signal is received at user device 180, the user input signal being representative of a textual or speech response from the user. The user input signal may be representative of the response of the user to the query provided at step 222. The user input signal may correspond to a text signal or an audio signal, depending on the implementation. Dependingon the application, the user input signal may contain a word, a sentence, or multiple sentences. In some embodiments, the user input signal may further include a signal representative of textual or speech from a caregiver via the caregiver interface 106.

[0112] At step 226, system 100 processes the user input signal to identify patterns. It will be appreciated that the processing may be performed by a program configured to be able to identify patterns in the user input signal. In some embodiments, the program codifies a process that includes discretizing the user input signal into a plurality of tokens. The discretization can be performed using techniques such as tokenization, parsing, semantic analysis, language modeling, discourse processing and so on. In some embodiments, the user device 108 is configured to process the input signal. In some embodiments, the server 104 is configured to process the input signal.

[0113] In some cases, at step 226, the user input signal may be pre-processed to remove undesired data, such as punctuation, articles, stop words, etc. Furthermore, the pre-processing may include formatting the user input data into a format processable by the user device 180 and / or the server 104. In some embodiments, the processing of the user input signal may include the queries prompted at step 222, which can be implemented by associating each token with a corresponding query. In some embodiments, the user device 108 is configured to pre-process the input signal e.g. before provision to server 104 for further processing. In some embodiments, the server 104 is configured to pre-process the input signal.

[0114] Once the user input data is processed into tokens, features are extracted from the data using, e.g., a machine learning model. For instance, each token may be weighted, labelled and / or vectorized depending on various factors, such as the content of the token, the frequency of the token in the user input signal, and the like, thereby capturing the semantic relationship amongst the tokens. In some embodiment, the machine learning model may be trained on labeled dataset to improve the extraction of features on the tokens. The features extracted from the tokens may be of various forms and nature, and be representative of the thoughts, the emotions, the physical feelings, the actions and the situations related to the user. The features may also be factual, such as the time of the day, the place where the conversation occurs, the name of the people involved in the conversation should the user be accompanied by other people such caregivers, and the like.

[0115] Using the features extracted from the tokens, the computing device thereafter identifies patterns in the user input signal. The patterns may be established based on types of features, intensity of features and the like. For instance, a pattern relating to a particular emotion may be established should the user demonstrate such emotion in the user input signal. It will be appreciated that multiple patterns may be identified during a conversation between the user and the agent 102.

[0116] At step 228, system 100 generates a treatment using the processed user input signal. In some embodiments, the user device 108 is configured to generate the treatment. In some embodiments, the server 104 is configured to generate the treatment (e.g. by codified instructions stored in memory 120 and executable by the processor 110 to configure a treatment generator application that automatically outputs a treatment based on the input data). The treatment is generally generated by identifying specific problems amongst the patterns and provide specific solutions for each identified problem. It will be appreciated that the concept of “problem” referred to herein corresponds to a type of pattern that affects the mood, thinking, or behavior of the user in a way that makes it difficult to cope with daily life. For each identified problem, the software generates one or more solutions using, e.g., a database including therapeutic techniques and personality traits of the user. For instance, positive traits identified in the patterns of the user, along with a therapeutic technique, may be used to generate a solution for a given problem. In terms of treatment, the solution may be a guideline or software instructions reflecting an approach for the caregiver or the agent 102 to fix the problem of the user. Once a solution is generated for each identified problem, the software generates the treatment based on the solutions and may provide the treatment to a caregiver interface 106 or another conversational agent, depending on the embodiment. As an illustrative example, the user may express health concerns to the agent 102 which may be received user feedback and recorded as a health concern in the user profile. The health concern may be a headache. The health concern may be tagged or labeled and recorded in the user profile and / or daily summary of the user. The processor 110 can identify corresponding treatments or a treatment plan for the health concern, such as laying down, getting rest, drinking fluids, and so on. In some embodiments, the processor 110 can provide a notification to caregiver interface 106. This notification can include the detected health concern and suggested treatments. In some embodiments, the processor 110 can provide a notification to agent 102 with suggested treatments for selection by the user. The agent 102 may receive a selected treatment. The processor 110 can provide a notification to caregiver interface 106 with the selected treatment so that a caregiver can assist in providing the treatmentto the patient or user. For example, the notification can indicate that the patient would like a glass of water, so that a caregiver can provide the glass of water in response to the notification at the caregiver interface 106.

[0117] In some embodiments, at step 228, the system 100 generates recommendations based on the processed user input signal, the recommendations being configured to be provided to the caregiver interface 106. The recommendations are formulated to guide the caregiver to prepare the behavioral treatment based on the identified patterns of the user. In this case, the caregiver may generate the behavioral treatment outside of the system 100 based on the recommendations and input the behavioral treatment into the caregiver interface 106 for the system 100 to implement the behavioral treatment.

[0118] It will be appreciated that sentiment analysis that is turned into recommendations may be subject to various requirements. The first requirement is that the analysis must be derived entirely from actual words turned into token / concepts from the end user. This reduces the risk of “Al Hallucination” or other artifacts of the like distorting the ability to assess intent from the end user. The second requirement is that the guidelines upon which the system is trained apply constraints to the types of recommendations that can be made. This enables, in software deployment, to ensure that the software accomplishes that which is desired and avoids making recommendations which could incur liability, violate user-related constraints, or otherwise extend outside of desired behavior.

[0119] In an exemplary application, a patient inputs the following user input: “My stomach hurts.” In this case, the sentiment analysis identifies the words actually used (e.g., “stomach”) and avoids interpretation (e.g., not “abdomen”). Then, the generated recommendations are limited to the context, for example, to the following suggestion:, “Caregiver should look into this” as opposed to a more specific medical advice, for instance: “Recommend prescribing Pepto Bismol”, which would be restricted by the software.

[0120] In some embodiments, the treatment is provided to the agent 102 at user device 180). In such cases, the treatment may be provided in the form of parameters and instructions that guide the agent 102 to adopt the generated solutions and to fix the problems of the user during conversations. For instance, when the user appears to exhibit an anger pattern in a conversation pertaining to the past, the agent 102 may be programmed to divert the conversation towards a positive memory that may help the user to manage its emotions. In some otherexamples, the agent 102 may start a conversation with the user on subjects that are fit to help reinforce the positive behaviour of the user.

[0121] In some other embodiments, the treatment is provided to a caregiver interface 106 in the form of summaries and recommendations. In such cases, the software may list and grade the identified problems and the created solutions. The software may also generate a contextual background regarding the user’s mental health, which may help the caregiver understanding the treatment to provide to the user. It will be appreciated that the type and form of treatments may vary depending on the embodiments.

[0122] FIG. 2C shows a diagram of a method 240 for performing a mental state evaluation using a conversational agent 102. It will be appreciated that the method 240 may be implemented by a computing device running the conversational agent 102 or being in communication therewith. The method 240 aims at generating a cognitive function report that is consistent, is easy to administer and helps assessing the progression of the patient on a continuous timeline.

[0123] At step 242, the conversational agent 102 receives an indication to perform a mental state examination based on a cognitive tool. It will be appreciated that the conversational agent 102 is generally in communication with the patient via a user interface. The indication may be received while the user is in a conversation with the conversational agent 102, in which case the conversational agent 102 may initiate the mental state evaluation or schedule the latter for another moment. In some cases where the user is not interacting with the conversational agent 102, a notification may be provided to the user interface for notifying the user that the mental state examination should be performed.

[0124] At step 244, the conversational agent 102 retrieves, in response to said receiving the indication, the cognitive tool from a database and generates a series of questions using the cognitive tool. It will be appreciated that the content of the series of questions may vary depending on the type of selected cognitive tool.

[0125] In some embodiments, prior to step 244, the conversational agent notifies the user that the mental state examination is ready to begin, or queries the user by asking if the latter is ready to perform the mental state examination. The notification can be at an interface, for example.

[0126] At step 246, the conversational agent 102 sequentially introduces each question in the conversational content. The questions may be processed by the conversational agent using one or more NLP models or encoded process (e.g. executable code or instructions encoding operations for NLP), and be rephrased in a manner that matches a user profile for the user to better understand the question. Alternatively the questions may be directly inserted in the conversational content. The term “sequentially” used herein implies that the questions are introduced one at the time in the user content.

[0127] At step 248, the conversational agent 102 identifies a response for each question in the user feedback using an NLP model, programme and / or encoded process (e.g. executable code to interpret human language) to enable the conversational agent 102 to recognize, understand and / or generate text. The content of the response may be thereafter compared to a solution contained in the cognitive tool, such that the conversational agent 102 obtains the result of the question by determining if the user has selected the correct response for the respective question. The solution may be Boolean, such as for a “yes or no” question, or be a word or a series of words. In the case of a Boolean solution, the conversational agent 102 may attribute a pass result for a response matching the solution or a fail result if not. In some embodiments, the conversational agent 102 may be configured to determine if the semantic analysis of the response has a degree of confidence that falls within a predetermined threshold of the response, and attribute a pass result if the degree of confidence is in the predetermined threshold or a fail result if not. The conversational agent 102 may also be configured to grade the result of the response based on the semantic analysis thereof.

[0128] In some embodiments, the question (or questions) may be provided as a media including an image, a video or a soundtrack. In this case, the user response may include a description of the media, which is processed by the conversational agent 102 to establish the correctness of the response.

[0129] In some embodiments, the conversational agent 102 may be configured to measure the time that has lapsed between providing the output signal including the question and receiving the input signal including the response. In some embodiments, the conversational agent 102 may be configured to stop the mental state examination or move to the next question should a predefined amount of time have lapsed without having received an input signal from the user. In such a case, an indication that the user was not responsive may be introduced in the response.

[0130] Furthermore, at step 248, the conversational agent 102 associates the response with each respective question of the series of questions. The conversational agent 102 may also include in the response the result of response, the time taken by the user to respond and / or an indication that the user was responsive or not to the question.

[0131] At step 250, the conversational agent 102 generates a cognitive function report using the series of question and the associated responses. The cognitive function report may be provided as a list of questions with respective responses. It will be appreciated that the format of the cognitive function report may vary. The conversational agent 102 may be configured to rephrase the provided questions and / or received responses to facilitate the review of the caregiver or the mental health specialist.

[0132] FIG. 2D shows a diagram of a method 260 for training a user model using conversational features. It will be appreciated that the method 260 may be implemented by a computing device running the conversational agent 102 or being in communication therewith. The method 260 aims at refining a user profile using user feedback and training a user model. By recursively training the user model based on user feedback, treatment profiles, and a user profile updated throughout iterations of back and forth between the user and the conversational agent, the user model becomes tailored to the user’s behaviour. System 100 can develop, train and update different models and agents for different users using user feedback and user profiles.

[0133] At step 262, system 100 receives, from a database, a user model representative of a user profile. The user model may be stored in the memory 120 of the server 104, or in the memory of one or more user devices 180. A user model, as referred to herein, is a machine learning based program configured to find patterns or make decisions from a previously unseen dataset, an updated user profile, and so on. The processing device(s) that receives the user model may vary depending on the implementation, and may be, for instance, the server 104 and / or one or more user devices 180. In the case where the server 104 receives the user model, the latter may be provided to a user device 180 via the network 170. A user profile can refer to a structured representation of information about a user. This profile can include various attributes that help the agent understand and predict the user's preferences, behaviors, and needs. The following are example components of a user profile: Demographic Information: (age, gender, location, and occupation), Behavioral Data (e.g. Information about the user's interactions with the system, such as history, patterns, and usage frequency), Preferences (e.g. Explicitly stated or inferred preferences, such as favorite topics, preferred communication style, and interests), ContextualInformation (e.g. Data about the user's current context, such as time of day, device being used, and location), Historical Data (e.g. Past interactions and transactions that can help in predicting future behavior). User profiles are for personalizing the user experience, improving the accuracy of recommendations, and enhancing the overall effectiveness of NLP engine and intelligent agent 102. By leveraging user profiles, the system 100 can tailor responses and actions to better meet the individual needs and preferences of each user. A user profile may be structured according to a data model or class model that defines multiple data fields and corresponding data values, and links between the data fields. A user profile can have data fields that correspond to a user identifier, a patient legal name, patient nickname or preferred name, user preferences (e.g. likes, dislikes), off limit topics, family history, health concerns, time and data preferences (e.g. wake time, sleep time, activity time), activity calendars, and so on. For example, family history data may indicate that the patient’s child recently passed away so that agent 102 avoids mentioning the child and / or provides consoling comments about children. As another example, an activity calendar may be specific to a facility and indicate different activities and times. The user profile may have a user preference of games such as chess, and the agent 102 may use the user profile to suggest chess as an activity on a specific date / time based on the activity calendar.

[0134] At step 264, system 100 receives, from a user interface, an input signal representative of a user feedback. It will be appreciated that the input signal may vary in form, such as text data, an audio file, an image file or an audio file. In the case where the input signal includes an audio file or a video file, the system 100 may be configured to process the audio file into a text file using a “speech-to-text” operations. In the case where the input file is an image file or a video file, the system 100 may be configured to process the input file using Optical Character Recognition (OCR) to detect text data in the input file or an object detection and recognition process to identify objects in the input file. There are different types of user feedback. For example, user feedback may indicate user preferences (likes, dislikes), nickname, timing data (e.g. wake up at 7am and would like to have a conversation with the agent 102 close to that time), family history, health concerns, and so on.

[0135] At step 266, system 100 processes the input signal to identify features representative of a user preference and / or an emotional behaviour in the user feedback using natural language processing. In some embodiments, the program codifies a process that includes discretizing the input signal into a plurality of tokens. The discretization can be performed using techniques such as tokenization, parsing, semantic analysis, language modeling, discourse processing and so on. In some embodiments, the user device 108 is configured to process theinput signal. In some embodiments, the server 104 is configured to process the input signal. As an example, if the user feedback includes a time that the user wakes up and an extracted feature can be ‘wake time’ and the corresponding value 7am’. The wake time can set a do not disturb parameter for the agent 102 so that the agent 102 is not activated until after the wake time. As another example, if the user feedback includes a time that the user goes to bed and an extracted feature can be ‘bedtime’ and the corresponding value ‘8pm’. The bedtime can set a do not disturb parameter for the agent 102 for after the bedtime. That is, the wake time and the bedtime can be extracted features that can in turn set a user preference relating to activity or inactivity time for the agent 102 (e.g. inactive between the bedtime and the wake time). As another example, the agent 102 may have a default setting to become inactive (e.g. go to sleep) after 5 minutes of inactivity from the user. However, the user may have slowed speech so for that user the default period may adjust to 15 minutes. As another example, the user feedback may include likes or dislikes, and the processor 110 can extract values from the user feedback to populate features for likes and dislikes (e.g. the user feedback may indicate that the user likes a specific poet or poem, and does not like a specific sports team). As a further example, the user feedback may indicate health related concerns, and a feature may relate to a health concern that can be populated with data extracted from the user feedback (e.g. the user has a headache). There may be an associated feature for the agent 102 to check in on the user after a time or event about their health concern to gather more data to reassess the health concern (e.g. check in on the user after a nap event to reassess headache). Accordingly, features can be extracted from user feedback, and different user preferences can map to those features.

[0136] In some cases, at step 266, the input signal may be pre-processed to remove undesired data, such as punctuation, articles, stop words, etc. Furthermore, the pre-processing may include formatting the input data into a format processable by the user device 180 and / or the server 104. In some embodiments, the user device 108 is configured to pre-process the input signal e.g. before provision to server 104 for further processing. In some embodiments, the server 104 is configured to pre-process the input signal.

[0137] In some embodiments, the input signal is received in response of providing conversational data to the user interface. The conversational data may be generated using the user model. When processing the input file, the features used to generate the outputted conversation data may be considered by creating semantic connection with the identified features.

[0138] At step 268, system 100 updates the user profile using the identified features. The system may update the user profile by adding the features thereto. When existing similar features are already present in the user profile, system 100 may adjust the existing features using the identified features. Alternatively, in this case, system 100 may discard the identified features, or replace the already existing features with the identified features. In some embodiments, a semantic evaluation of the identified features may be performed to establish if the identified features fall within a degree of confidence with respect to an expected feature. As an example, the user profile can be updated with features such as wake time, bedtime, inactivity time for the agent 102, likes and dislikes, health concerns, and so on.

[0139] At step 270, system 100 trains the user model with the updated user profile using machine learning. It will be appreciated that the type of machine learning may vary, depending on the implementation. In some cases, the machine-learning is based on deep-learning or on generative Al.

[0140] In some embodiments, method 260 includes step 272, in which system generates a summary using the trained user model. At a high level, the summary is a tool used by a caregiver to assess the mental condition of a patient conversing with the conversational agent. The summary may include sentiments, trends, and concerns based on the latest conversation held between the conversational agent and the user. The summary may be provided to the caregiver via the caregiver interface 106.

[0141] FIG. 2E shows a diagram of a method 280 for updating parameters for an interaction software running a conversational agent 102 for assisting a user. It will be appreciated that the method 200 may be implemented by a computing device running the agent 102 or being in communication therewith.

[0142] At step 282, system 100 receives, from the user interface of the user device 180, a string data object representative of a user feedback received. The string data object may contain words, sentences, conversation logs, pictures files, numbers and so on.

[0143] At step 284, system 100 receives, from a database, a model of the conversational agent including a plurality of model parameters. A model parameter can be a variable within a model that is adjusted during the training (or tuning) process to e.g. minimize the difference between the predicted output and the actual output. These parameters can determine the model's behavior and performance. In machine learning, model parameters can be weights and biases inneural networks, which are updated through optimization processes. A goal may be to find the optimal set of parameters that allow the model to make accurate predictions or outputs on new, unseen data. An example model parameter is a safety parameter. For example, a patient may get paranoid and request to alert an emergency service. The parameter may configure the model and agent 102 to detect a safety concern and then collect more data from the user about the concern and calm the user down. Another example parameter may relate to notifications, including triggers and recipients. The parameter may configure the model to send a notification about a safety concern to a caregiver interface 106, along with information gathered from the patient. Another example parameter may relate to fall detection such as for a patient that may have restricted mobility or movement, or history of falls, and so on. The parameter may configure the model to send a notification about a fall event to a caregiver interface 106.

[0144] At step 286, system 100 identifies, using the NLP engine, a first token representative of a user preference in the string data object and labels the first token with a respective semantic category. A semantic category can refer to a group of words or phrases that share a common meaning or concept. These categories help in understanding and organizing the meaning of words within a text. Semantic categories can enable the model to grasp the context and relationships between words more effectively. A semantic category may relate to a data value for a feature, for example. As an example, a user preference may relate to activities that the patient likes, such as painting. For this example, the semantic category may be ‘paint’ which can be used to build out a knowledge graph for the NLP engine. Another example semantic category can relate to a story and likes / dislikes. The agent 102 can generate story content for the user based on the likes / dislikes which may map to different semantic categories. Another example semantic category can relate personality assessments such as an extrovert or introvert.

[0145] At step 288, system 100 associates the labelled first token with a corresponding model parameter from the plurality of model parameters. For example, the token ‘paint’ may be labelled with ‘likes’ for a corresponding model parameter about conversation topics for the user so that the agent 102 includes content about painting. As another example, the token ‘baseball’ may be labelled with ‘dislikes’ for the corresponding model parameter about conversation topics for the user so that the agent 102 does not include content about baseball.

[0146] At step 290, system 100 updates the model parameter associated with the first labelled token using the labelled first token and train the model of the conversational agent using the updated model parameters.

[0147] Now referring to Fig. 3A, there is shown a block diagram of a system 300 for analyzing a conversation between a user and an agent. It will be appreciated that the system 300 may be implemented using components of the system 100 for performing the method 220. The system 300 includes a conversational log database 302, from which a conversation log between the user and the agent can be obtained. The conversational log is thereafter provided to a word to token converter 304, which uses an application, such as NLP code, to convert the words into token. It will be appreciated that the token converter 304 creates the token according to the meaning of the words.

[0148] In some embodiments, the converted tokens are provided to a token handler 306, which groups the tokens and establishes among the tokens. The token handler 306 is configured for determining a contextual intent in the conversation log using the tokens.

[0149] The tokens are thereafter provided to an assessment and digestion module 308, which includes a summary database 310 and a feature extractor 312. In operation, the assessment and digestion module 308 generates a summary of the conversation between the user and the agent using the determine contextual intent and the tokens, and stores the summary in the summary database 310. It will be appreciated that the summary may be accessed by the caregiver via the caregiver interface 106 to obtain a brief overview of the conversation. The feature extractor 312 processes the tokens by weighting, labelling and / or vectorizing the tokens, thereby capturing the semantic relationship amongst the tokens.

[0150] The extracted features are thereafter provided to a recommendation generator 314, which generates, based on the extracted features, a recommendation for the caregiver using, e.g., a machine learning model.

[0151] In some embodiments, the recommendation generator 314 may be in communication with a restriction database 316, which contains a list of restrictions that restrain the recommendation generator 314 from generating unwanted recommendations. The restrictions may be based on laws and regulations, on user-related criteria, on population-related criteria, on caregiver preferences and other categories of the like. The restrictions may be obtained from various sources, such as predefined restriction lists, from the caregiver via the caregiver interface 106, and the like.

[0152] Once the system 300 establishes that the recommendation generated by the recommendation generator 314 meets the restrictions contained in the restriction database 316,the recommendation is then provided to a recommendation provider 318, which is configured to output the recommendation in a memory 120 and / or to the caregiver interface 106.

[0153] FIG. 3B shows a block diagram of a system 330 for performing a mental state evaluation. The system 330 may be used to implement the method 240 of Fig. 2C. The system 330 includes a conversational agent 332 having an indication reception unit 340, a question generating unit 336, a response identifying unit 338 and a cognitive function report generating unit 340. The question generating unit 336 is coupled to a cognitive tool database 342, and the question generating unit 336 and the response identifying unit 338 are coupled to a user interface 344, which can be accessed by the user. It will be appreciated that the system 330 may be implemented in the system 100 described above.

[0154] At a high level, the conversational agent 332 is configured for generating conversational content based on a user profile using one or more NLP models or encoded processes. The conversational content may also be limited to specific parameters defined by the caregiver and / or the mental health specialist. It will be appreciated that the user profile may be a machine learning model trained on a database of data related to the user or to a plurality of other users, depending on the implementation.

[0155] Using the created conversational content, the conversational agent is configured for providing an output signal including the conversational content to the user interface 344, and to receive therefrom an input signal being representative of user feedback. It will be appreciated that the input signal is representative of a user command provided to the user interface 344 by the user. In a general manner, i.e., when not performing the mental state evaluation, the user feedback may or may not be predicted, and the conversational agent may be configured to process the signal using an NLP process to identify user-related properties, such as a behavior, a mental state, general knowledge and the like. Using the processed user feedback, the conversational agent 332 is configured to adapt the user profile to the user behavior, to tailor the conversational content to the user.

[0156] The indication reception unit 334 is configured to receive an indication that the mental state evaluation is to be performed on the user. The indication may be received from the server 10 via the network 170, from a caregiver interface 106 or from any other source of data such as a database. In some cases, the indication may be stored on the memory of the processor of the user device 180 and provided to the conversational agent 102 upon detection of a triggercondition, such as a predetermined amount of time that has lapsed since the last mental state evaluation or if the current date and time match a trigger date and time. In some embodiments, the trigger condition may be configured on detecting a particular cognitive pattern in the user feedback processed by the conversational agent 102.

[0157] Once the indication is received by the indication reception unit 334, the latter prompts the question generating unit 336 to proceed into generating a series a question for cognitive evaluation of the patient. To do so, the question generating unit 336 requests the cognitive tool database 342 a given cognitive tool. It will be appreciated that the cognitive tool database 342 is a storage medium that may be based locally or over the network 170, depending on the implementation. In response, the cognitive tool database 342 may return the question generating unit 336 a software containing instructions to generate the series of questions. Alternatively, the cognitive tool database may provide the question generating unit 336 with a text format or the like of the series of questions, which is transformed into conversational content by the conversational agent 102. Once the conversational content is provided to the user interface 344 via the output signal, the question generating unit 336 is put on standby for further instructions.

[0158] After processing the input signal provided by the user interface 344, the response identifying unit 338 is configured to run a semantic analysis on the user feedback in order to identify a response therein. Once the response is identified by the response identifying unit 338, the latter prompts the question generating unit 336 to output a signal including the next question in the series of questions. If the series of questions have been completed, or if a trigger condition has been identified such as the determination of the user being non-responsive to the conversation content, the question generating unit 336 returns the response identifying unit 338 an indication that the cognitive function report is ready to be generated.

[0159] Once the series of questions has been responded by the user, or if the user has been established to be non-responsive, the response identifying unit 338 provides the cognitive function report generating unit 340 an indication to generate the cognitive function report.

[0160] In some embodiments, the cognitive function report generating unit 340 may be configured to provide the cognitive function report to the caregiver interface 106 or to the server 104 via the network 170.

[0161] FIG. 30 shows a block diagram of a system 360 for training a user model using conversational features. The system 360 may be used to implement the method 260 of Fig. 2D. The system 360 includes a conversational agent 362 having a conversational content generating unit 364, a user input receiving unit 366, a user profile updating unit 368 and a user model training unit 670. The system 360 also includes and a user model database 372 and a user interface 374, which are both connected to the conversational agent 362.

[0162] In operation, the conversational content generating unit 364 is configured for generating conversational content and providing the latter to the user interface 374. The conversational content may be generated using the user model to tailor the conversational content around the user profile. Alternatively, the conversational content may be arbitrarily generated using predefined templates, or by generated using a command received from the caregiver interface, for instance. The conversational content generating unit 364 may also receive a user input from the user input receiving unit 366. In this case, the conversational content generating unit 364 may use the user model to process the user input to generate the conversational content.

[0163] The user input receiving unit 366 is configured to the input signal representative of user feedback received from the user interface 374 and process the input signal to identify features. The processing of the signal may be performed using the methods described above. The features may be representative of emotional patterns, behaviors, cognitive patterns and memories of the user. The identified features are fed to the user profile updating unit 368, which is configured for updating the user profile with the identified features. The user profile may be stored on a user device 180 and / or in the server 104, depending on the implementation. In some embodiments, the identified features may be labelled and categorized by the user profile updating unit 368. For instance, the features may be regrouped by emotions, memories, cognitive patterns, and behavior, and a relationship between the features across different categories may be established. For example, a particular emotion (e.g., happiness) may be associated with a memory (e.g., memory of a child) and a behavior (e.g., feeding the child), and the conversational content generating unit 364 may be configured for creating conversational content based on this combination of features should the user model or other requests queries such content. In one other example, a cognitive pattern (e.g., not remembering the name of a spouse) may be associated with a memory (e.g., walking with the spouse). In such a case, the conversational agent 362 may be configured to generate an item in a summary to be provided to a caregiver based on this combination of features.

[0164] In some embodiments, the user profile updating unit 368 includes restrictions obtained from the caregiver interface or from predefined templates. For instance, the user profile updating unit 368 may be prevented to include certain aspects of the user profile, such as the inclusion of some types of memories or to modify the behavior of a user to a certain extent, based on the restrictions. This effect may limit the conversational content generating unit 364 from creating content that may be harmful to the user in terms of mental health treatment.

[0165] Once updated, the user profile is provided to the user model training unit 370, which also receives the user model from the user model database 372. In some embodiments, the user model may be a trained LLM obtained from a provider such as OpenAI™, Google™, Meta™, and the like, which implies that the user model has been previously or is continuously trained using large data structures. The user model database 372 may further store anonymized patient and / or therapist transcripts with markup data for training the user model. Additionally, the user model database may store caregiver instructions and resident information, which can also be used for training the user model.

[0166] In operation, the conversational agent 362 may be configured to recursively outputting conversational content from the conversational content generating unit 364 and receiving user input from the user input receiving unit, which results in sustaining a conversation with the user via the user interface 374. The conversational content generating unit 364 may be configured for processing new conversational content using the user model and the user input. The user model may be retrained by the user model training unit 370 at every iteration of the conversation, or after a predetermined number of iterations.

[0167] In some embodiments, the conversational agent 362 may be configured for processing the identified features and establishing if the identified features fall within a degree of confidence with respect to an expected feature. In this case, if the identified features fall outside of the degree of confidence, which may correspond on a user model diverging from the user characteristics received by the user input receiving unit, the user model training unit 370 may be configured to retrain the user model using a user profile received from the user profile updating unit 368 and updated using the newly identified features.

[0168] In some embodiments, the user profile updating unit 368 may be configured for identifying gaps in the user profile, such as features that have not been updated for a predetermined amount of time, categories of feature that are not or lowly populated, features thatno longer match the newly identified features and other types of deficiencies of the like. Upon detection of a gap in the user profile, the user profile updating unit 368 may be configured for requesting the conversational content generating unit 364 to generate conversational content formulated to receive features that would fill the gap in the user profile.

[0169] FIG. 4A shows an example interface 400 for defining parameters for an interaction software (e.g. conversational agent 102) for assisting a user. The interface 400 can be implemented using the method 200 of Fig. 2A. The interface provides an improved way to allow one user to apply specific and fuzzy constraints for how the interaction software will engage with another user.

[0170] In this example, the interface receives information or feedback from a user (e.g. a caregiver) that marks certain topics as off limits for communications with another user (e.g. a patient). That is, interface 400 can capture or receive user feedback indicating one or more topics that should not be part of the conversational content for the specific user. The interface also prompts and encourages certain other conversational parameters. The interaction software correctly understood and saved these parameters into a custom end user database, user profile, and so on. A user may not have technical knowledge to use complex interfaces, and the improved interface enables users without such technical knowledge to configure parameters for the agent by receiving this user feedback. As another example, user feedback can also indicate topics that should form part of the conversational content for a specific user. The topics may help the user feel comfortable which may improve the accuracy of cognitive assessments. The topic may also be useful as part of the treatment, for example.

[0171] Eldercare providers, parents, teachers, and the guardians of other individuals or populations who lack significant autonomy may rely on conversational Agents 102 to provide companionship, instruction, and other types of interactions. The constraints that caregivers wish to put upon these Agents 102 can be customized - differing from end user to user, and evolving over a user’s life. Caretakers may not have the patience or technical sophistication to use complex interfaces to set parameters for the Agent 102. Moreover, conversational engagement with an end user is inherently complex and constantly evolving. Caretakers will benefit from the ability to set constraints for the Agent 102, label certain topics off limits, and encourage certain types of interactions by the Agent 102 while discouraging other types of interactions by the Agent 102. Embodiments described herein provide improved systems, methods and interfaces that approach allow Caretakers to explain preferences in natural language free from technical jargon; the system100 then uses logic and applies data from LLMs to reinforce these constraints during engagements with the end user and others.

[0172] Historically, providing data on how an end user will engage with software or the allowable parameters of a dataset relies on often complex and tedious data entry. These approaches lack the fluidity and adaptability required for conversational systems that are designed to engage with users in real-world conditions (like schools, nursing homes, etc.) They can be burdensome where only a small percentage of the population being able to use them.

[0173] FIG. 4B shows an example interface 410 for performing a mental state evaluation using the system 100. The interface 410 shows an exemplary graphical user interface (GUI) including selection box provided to the caregiver interface 106. The selection box includes a plurality of cognitive tools to be selected by the caregiver. In this embodiment, the subsequent(s) mental state evaluation will be performed using the selected cognitive tool.

[0174] FIG. 4G shows an example interface 420 for performing a mental state evaluation using the system 100. The interface 420 shows a GUI provided to the patient, on which recorded responses are displayed. In this case, the interface 420 outputs the signal using a speaker device or the like, and the user responds to the questions using a microphone.

[0175] FIG. 4D shows an example interface 430 for performing a mental state evaluation using the system 100. The interface 430 shows a GUI provided to the patient, on which recorded responses are displayed. In this case, the interface 430 outputs the signal, including a phrase and an image, on a display device, and the user responds to the questions using a microphone.

[0176] FIG. 4E shows an example interface 440 for training a user model using conversational features. The interface 400 can be obtained using the method 260 of Fig. 2D. The interface 440 provides an improved way to tailor conversational content around user engagement. The conversation provided in the interface 440 is exemplary and the current technology is not bound to the content of the interface 440.

[0177] Now referring to FIG. 5, there is shown a diagram 500 for caregiver parameters that can be used to check and influence Al agent responses to an end user. The diagram 500 includes an Al agent module 502, an end user module 504, a software parameters module 506 and a caregiver interface module 508. In some embodiments, the memory 120 stores code instructions for the software parameters module 506 and processor 110 executes the codeinstructions for the software parameters module 506 to implement functionality and operations described herein. In some embodiments, the end user module 504 is executed by a processor at user device 180. In some embodiments, the Al agent module 502 is executed by Al agent 102 or forms part of or integrates with the Al agent 102. In some embodiments, the Al agent module 502 is executed by processor to generate output data for the Al agent 102 and receive input data from the Al agent 102. In some embodiments, the caregiver interface module 508 is executed by caregiver interface 106 or forms part of or integrates with the caregiver interface 106. In some embodiments, the caregiver interface module 508 is executed by processor to generate output data for the caregiver interface 106 and receive input data from the caregiver interface 106.

[0178] The diagram 500 represents the high-level process of the software controlling the Al agent 102 conversing with the end user. The Al agent 102 receives input data from the end user and generates output data for provision to the end user (e.g. for display at user device 180 or for generating speech or audio output at user device 180). In other words, the diagram 500 shows a conversation loop (e.g. data exchange) between the Al agent module 502 and the end user module 504, in which the software parameters module 506 is interposed to apply the parameters set by the caregiver interface module 508.

[0179] In operation, the Al agent module 502 provides signals representative of a conversation generated with a machine learning process to the end user module 504 via the software parameter module 506. Upon reception of the signal provided by the Al agent module 502, the software parameter module 506 analyzes the signal and may apply modifications to the signal based on the parameters. It will be appreciated that the parameters (or values associated therewith) can be obtained from the caregiver interface 508, in accordance with the embodiments presented above. In some embodiments, the software parameter module 506 may be configured for modifying the parameters of the machine learning process of the Al agent by providing a signal including those modifications to the Al agent module 502.

[0180] Once the signal of the Al agent module 502 is received and processed by the software parameter module 506, the software parameter module 506 is configured to provide the processed signal to the end user module 504, which is configured to display the signal to the end user via a suited user interface. The end user module 504 is also configured to receive a signal indicative of a user input received from the user interface, and to provide this signal to the Al agent module 502, which will use this signal in the machine learning code to continue or terminate the conversation.

[0181] Now referring to Fig. 6, there is shown a first graph 600 representing the sentiment analysis of the user over time and a second graph 650 representing the engagement of the user over time, as processed by the agent 102. The first graph 600 and the second graph 650 can be obtained as a result of performing the method 220 of Fig. 2B.

[0182] In the first graph 600, there is shown a first sentiment curve 602, and second sentiment curve 604 and a third sentiment curve 606. It will be appreciated that the sentiment curves 602, 604, 606 are exemplary, and the form and shape thereof may vary. In the embodiment depicted in the first graph 600, the first sentiment curve 602 represents the intensity of happiness of the user as a function of time, the second sentiment curve 604 represents the intensity of surprise of the user as a function of time and the third sentiment curve 606 represents other nonevaluated sentiments. The intensity of the sentiment curves 602, 604, 606 may be established via various types of metrics, such as the semantic analysis of the conversation between the agent 102 and the user centered on a given sentiment or being directly established by, e.g., a therapist analyzing the behaviour of the user.

[0183] In the second graph 650, there is shown an engagement curve 652 representative of the level of involvement of the user in the conversation with the agent 102. The engagement curve 652 may be established semantically by analyzing the user input signal, or be representative of the amount of time per day the user is actively in communication with the agent 102.

[0184] In some implementations, the first graph 600 and the second graph 650 may be displayed on a graphical user interface (GUI), along with other pertinent information such as summaries and recommendations, to a caregiver via a display device coupled to the caregiver interface 106.

[0185] FIG. 7 is a schematic diagram of an electronic device 700 for implementing a natural language processing interface for caregivers. The electronic device 700 may be used to perform the steps of the methods 200, 220, 240, 260. As depicted, electronic device 700 includes at least one processor 702, memory 704, at least one I / O interface 706, and at least one network interface 708.

[0186] Each processor 702 may be, for example, any type of general-purpose microprocessor or microcontroller, a digital signal processing (DSP) processor, an integratedcircuit, a field programmable gate array (FPGA), a reconfigurable processor, a programmable read-only memory (PROM), or any combination thereof.

[0187] Memory 704 may include a suitable combination of any type of computer memory that is located either internally or externally such as, for example, random-access memory (RAM), read-only memory (ROM), compact disc read-only memory (CDROM), electro-optical memory, magneto-optical memory, erasable programmable read-only memory (EPROM), and electrically- erasable programmable read-only memory (EEPROM), Ferroelectric RAM (FRAM) or the like.

[0188] Each I / O interface 706 enables electronic device 700 to interconnect with one or more input devices, such as a keyboard, mouse, camera, touch screen and a microphone, or with one or more output devices such as a display screen and a speaker.

[0189] Each network interface 708 enables electronic device 700 to communicate with other components, to exchange data with other components, to access and connect to network resources, to serve applications, and perform other computing applications by connecting to a network (or multiple networks) capable of carrying data.

[0190] The embodiments of the devices, systems, and methods described herein may be implemented in a combination of both hardware and software. These embodiments may be implemented on programmable computers, each computer including at least one processor, a data storage system (including volatile memory or non-volatile memory or other data storage elements or a combination thereof), and at least one communication interface.

[0191] Program code is applied to input data to perform the functions described herein and to generate output information. The output information is applied to one or more output devices. In some embodiments, the communication interface may be a network communication interface. In embodiments in which elements may be combined, the communication interface may be a software communication interface, such as those for inter-process communication. In still other embodiments, there may be a combination of communication interfaces implemented as hardware, software, and combination thereof.

[0192] Throughout the foregoing discussion, numerous references will be made regarding servers, services, interfaces, portals, platforms, or other systems formed from computing devices. It should be appreciated that the use of such terms is deemed to represent one or more computing devices having at least one processor configured to execute software instructions stored on acomputer readable tangible, non-transitory medium. For example, a server can include one or more computers operating as a web server, database server, or other type of computer server in a manner to fulfill described roles, responsibilities, or functions.

[0193] One should appreciate that the systems and methods described herein may provide better memory usage, improved processing, and / or improved bandwidth usage.

[0194] The following discussion provides many example embodiments. Although each embodiment represents a single combination of inventive elements, other examples may include all possible combinations of the disclosed elements. Thus, if one embodiment comprises elements A, B, and C, and a second embodiment comprises elements B and D, other remaining combinations of A, B, C, or D, may also be used.

[0195] The term “connected” or “coupled to” may include both direct coupling (in which two elements that are coupled to each other contact each other) and indirect coupling (in which at least one additional element is located between the two elements).

[0196] The technical solution of embodiments may be in the form of a software product. The software product may be stored in a non-volatile or non-transitory storage medium, which can be a compact disk read-only memory (CD-ROM), a USB flash disk, or a removable hard disk. The software product includes a number of instructions that enable a computer device (personal computer, server, or network device) to execute the methods provided by the embodiments.

[0197] The embodiments described herein are implemented by physical computer hardware, including computing devices, servers, receivers, transmitters, processors, memory, displays, and networks. The embodiments described herein provide useful physical machines and particularly configured computer hardware arrangements. The embodiments described herein are directed to electronic machines and methods implemented by electronic machines adapted for processing and transforming electromagnetic signals which represent various types of information. The embodiments described herein pervasively and integrally relate to machines, and their uses; and the embodiments described herein have no meaning or practical applicability outside their use with computer hardware, machines, and various hardware components. Substituting the physical hardware particularly configured to implement various acts for nonphysical hardware, using mental steps for example, may substantially affect the way the embodiments work. Such computer hardware limitations are clearly essential elements of the embodiments described herein, and they cannot be omitted or substituted for mental meanswithout having a material effect on the operation and structure of the embodiments described herein. The computer hardware is essential to implement the various embodiments described herein and is not merely used to perform steps expeditiously and in an efficient manner.

[0198] Although the embodiments have been described in detail, it should be understood that various changes, substitutions and alterations can be made herein without departing from the scope as defined by the appended claims.

[0199] Moreover, the scope of the present application is not intended to be limited to the particular embodiments of the process, machine, manufacture, composition of matter, means, methods and steps described in the specification. As one of ordinary skill in the art will readily appreciate from the disclosure of the present invention, processes, machines, manufacture, compositions of matter, means, methods, or steps, presently existing or later to be developed, that perform substantially the same function or achieve substantially the same result as the corresponding embodiments described herein may be utilized. Accordingly, the appended claims are intended to include within their scope such processes, machines, manufacture, compositions of matter, means, methods, or steps.

[0200] As can be understood, the examples described above and illustrated are intended to be exemplary only.

Claims

WHAT IS CLAIMED IS:1 . A computing system for conversational agents to support patient or dependent care, the system comprising: a device with a caregiver interface providing caregiver conversational content for patient or dependent care using a first conversational agent; a device with a patient interface providing patient conversational content using a second conversational agent, and receiving user feedback; a server system comprising: at least one memory storing thereon a natural language processing (NLP) engine and an interaction software application configured to provide the first conversational agent that generates the caregiver conversational content for patient or dependent care for provision to the caregiver interface, and the second conversational agent that generates the patient conversational content for provision to the patient interface; a processing system that includes one or more processors and one or more memories coupled with the one or more processors, the processing system configured to: receive, from the patient interface, a string data object representative of the user feedback received; identify, using the NLP engine, a first token representative of a user preference in the string data object and label the first token with a respective semantic category; update or train the second conversational agent using the labelled first token; generate the patient conversational content using the trained second conversational agent, the conversational content to support patient or dependent care; and generate a notification as part of the caregiver conversational content based on the user preference, the notification to support patient or dependent care.

2. The computing system of claim 1 , wherein the second conversational agent is configured to refine the patient conversational content based on the user preference.

3. The computing system of claim 1 , wherein the memory stores a user profile of data fields and corresponding values, a data field and corresponding value storing the user preference, wherein the second conversational agent is configured to refine the patient conversational content based on the user profile.

4. The computing system of claim 1 , wherein the processing system is further configured to: identify, using the NLP engine, a second token representative of a cognitive pattern in the string data object and label the second token with a respective semantic category; and update or train the second conversational agent using the labelled second token.

5. The computing system of claim 4, wherein the processing system is further configured to generate a recommendation data object using the labelled second token and the first conversational agent, the recommendation data object including instructions for a caregiver to address the identified cognitive pattern of the user, wherein the caregiver conversational content comprises the instructions for the caregiver to address the identified cognitive pattern of the user.

6. The computing system of claim 5, wherein said updating further comprises updating a model parameter of a model for the first conversational agent associated with the second labelled token using the recommendation data object.

7. The computing system of claim 4, wherein the cognitive pattern includes at least one of perceptive functions, memory functions, problem-solving functions, judgment functions, emotional functions and social cognition functions.

8. The computing system of claim 1 , wherein the processing system is further configured to: receive an indication to start a mental state evaluation procedure; provide the patient interface with the patient conversational content including predetermined questions of the mental state evaluation procedure; and generate a cognitive function report using the user feedback and the patient conversational content,wherein the user feedback includes a response to one or more of the pre-determined questions.

9. The computing system of claim 8, wherein the processing system is further configured to record a duration between said providing the conversational content and said receiving the string data object, further wherein said generating the cognitive function report is performed using the recorded duration.

10. The computing system of claim 9, wherein the processing system is further configured to stop the mental state evaluation procedure if the recorded duration is greater than a predetermined threshold or if the NLP engine determines that the response does not semantically correspond to a desired response of a corresponding pre-determined question.

11. The computing system of claim 1 , wherein the string data object includes a conversation log including a plurality of string data objects each representative of corresponding user feedback and a plurality of string data object each representative of patient conversational content.

12. The computing system of claim 1 , wherein the processing system updates a user profile with the first token representative of the user preference, the user profile storing a data field corresponding to the user preference and a data value corresponding to the first token, wherein the user preference indicates a like or dislike.

13. The computing system of claim 1 , wherein the string data object includes data representative of instructions received from the caregiver interface, wherein the user preference includes the received instructions, wherein the caregiver conversational content is generated using a recommendation data object including instructions for a caregiver.

14. The computing system of claim 1 , wherein the first token representative of the user preference defines functionality of the second conversational agent to be adjusted when generating the patient conversational content.

15. The computing system of claim 1 , wherein the string data object representative of the user feedback received comprises data indicating a health concern or health condition, wherein the processing system updates a user profile with the data indicating the health concern or health condition, and wherein the patient conversational content is generated based on the data indicating the health concern or health condition.

16. A computing device for training or updating a conversational agent to support patient or dependent care, the device comprising: at least one memory storing thereon a natural language processing (NLP) engine and an interaction software application configured to provide one or more models for a conversational agent that generates conversational content for performing and providing conversations for patient or dependent care; a processing system that includes one or more processors and one or more memories coupled with the one or more processors, a user interface and a database, the processing system configured to cause the computing device to: receive, from the user interface, a string data object representative of a user feedback received; receive, from the database, a model of the conversational agent including a plurality of model parameters; identify, using the NLP engine, a first token representative of a user preference in the string data object and label the first token with a respective semantic category; associate the labelled first token with a corresponding model parameter from the plurality of model parameters; update the model parameter associated with the first labelled token using the labelled first token and train the model of the conversational agent using the updated model parameters; and generate conversational content for the conversational agent using the trained model, the conversational content to support patient or dependent care.

17. The computing device of claim 16, wherein the associated model parameter is configured to refine the conversational content generated by the conversational agent based on the user preference.

18. The computing device of claim 16, wherein at least one model parameter of the plurality of model parameters includes data fields and corresponding values obtained from a user profile of the user.

19. The computing device of claim 16, wherein the processing system is further configured to cause the computing device to: identify, using the NLP engine, a second token representative of a cognitive pattern in the string data object and label the second token with a respective semantic category; and associate the labelled second token with a corresponding model parameter from the plurality of model parameters.

20. The computing device of claim 19, wherein the processing system is further configured to cause the computing device to generate a recommendation data object using the labelled second token, the recommendation data object including instructions for a caregiver to address the identified cognitive pattern of the user.

21. The computing device of claim 20, wherein said updating further comprises updating the model parameter associated with the second labelled token using the recommendation data object.

22. The computing device of claim 19, wherein the cognitive pattern includes at least one of perceptive functions, memory functions, problem-solving functions, judgment functions, emotional functions and social cognition functions.

23. The computing device of claim 16, wherein the processing system is further configured to cause the computing device to: receive an indication to start a mental state evaluation procedure; provide the user interface with the conversational content including pre-determined questions of the mental state evaluation procedure; and generate a cognitive function report using the user feedback and the conversational content, wherein the user feedback includes a response to one or more of the pre-determined questions.

24. The computing device of claim 23, wherein the processing system is further configured to cause the computing device to record a duration between said providing the conversationalcontent and said receiving the string data object, further wherein said generating the cognitive function report is performed using the recorded duration.

25. The computing device of claim 23, wherein the processing system is further configured to cause the computing device to stop the mental state evaluation procedure if the recorded duration is greater than a predetermined threshold or if the NLP engine determines that the response does not semantically correspond to a desired response of a corresponding pre-determined question.

26. The computing device of claim 16, wherein the string data object includes a conversation log including a plurality of string data objects each representative of corresponding user feedback and a plurality of string data object each representative of conversational content.

27. The computing device of claim 16, wherein the processing system updates a user profile with the first token representative of the user preference, the user profile storing a data field corresponding to the user preference and a data value corresponding to the first token, wherein the user preference indicates a like or dislike.

28. The computing device of claim 16, wherein the string data object includes data representative of instructions received from a caregiver interface, wherein the user preference includes the received instructions, wherein the conversational content is generated using a recommendation data object including instructions for a caregiver.

29. The computing device of claim 16, wherein the first token representative of the user preference defines functionality of the conversational agent to be adjusted when generated the conversational content for assisting the user.

30. The computing device of claim 16, wherein the string data object representative of the user feedback received comprises data indicating a health concern or health condition, wherein the processing system updates a user profile with the data indicating the health concern or health condition, and wherein the conversational content is generated based on the data indicating the health concern or health condition.

31. A computer-implemented method for defining parameters for an interaction software application running a conversational agent for assisting a user, the method comprising:receiving an input signal representative of a user command including an indication for the conversational agent, the user command being in the form of human language and the indication defining a functionality of the conversational agent to be adjusted when assisting the user; using natural language processing, process the input signal to identify the indication and associate the indication with a corresponding parameter of the interaction software; generating an updated value of the parameter based on the identified limitation; and providing the updated value to the conversational agent.

32. A computer-implemented method for analyzing a conversation between a user and a conversational agent, the method comprising: receiving an input signal representative of the conversation between the user and the conversational agent, the conversation including user entries; using natural language processing, processing the input signal to identify patterns in the user entries; and generating a treatment based on the identified patterns.

33. A computer-implemented method for performing a mental state evaluation using a conversational agent, the method comprising: receiving an indication to perform the mental state examination based on a cognitive tool using a conversational agent; generating a conversational content based on a user profile using natural language processing, providing an output signal including the conversational content to a user interface, in response to providing the output signal, receiving an input signal from the user interface, the input signal being representative of user feedback, and adapting the user profile based on the user feedback;in response to said receiving the indication, retrieving the cognitive tool from a database, and generating a series of questions using the cognitive tool, the cognitive tool being configured for assessing the cognitive function of a user; sequentially introducing each question in the conversational content; using the natural language processing, identifying a response for each question in the user feedback, and associating the response with each respective question; and generating a cognitive function report using the series of question and the associated responses.

34. A computer-implemented method for training a user model using conversational features, the method comprising: receiving, from a database, a user model representative of a user profile; receiving, from a user interface, an input signal representative of a user feedback; using natural language processing, processing the input signal to identify features representative of a user preference and / or an emotional behaviour in the user feedback; updating the user profile using the identified features and; using machine learning, training the user model with the updated user profile.

35. A non-transitory machine-readable medium having stored thereon a plurality of instructions that, when executed by at least one computing device, cause the at least one computing device to perform the method of any one of claims 31 to 34.

Images