System and method for ai-guided group psychological therapy
An AI-guided system with a multilayered data structure addresses the limitations of traditional and digital mental health services by providing personalized and adaptive group therapy sessions, enhancing engagement and effectiveness without a human therapist.
Patent Information
- Application Number
- PCT/IL2026/050377
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-05
- Filing Date
- 2026-05-05
- Publication Date
- 2026-11-12
AI Technical Summary
Traditional mental health services face challenges such as high costs, limited access, and the lack of human interaction in digital interventions, which hinder the effectiveness of group therapy sessions and recipient engagement.
An AI-guided system that utilizes a multilayered, interconnected, weighted, and temporal group profile data structure to analyze recipient reactions and adapt session content in real-time, providing personalized and adaptive psychological support without the need for a human therapist.
The system ensures accessible, scalable, and personalized therapeutic experiences by dynamically selecting and adapting interventions based on recipient profiles and group dynamics, enhancing engagement and effectiveness.
Smart Images

Figure IL2026050377_12112026_PF_FP_ABST
Abstract
Description
SYSTEM AND METHOD FOR AI-GUIDED GROUP PSYCHOLOGICAL THERAPYCROSS-REFERENCE TO RELATED APPLICATIONS
[0001] This application claims the benefit of priority to U.S. Application No. 63 / 799,900, titled “SYSTEM AND METHOD FOR AI-GUIDED GROUP PSYCHOLOGICAL THERAPY”, filed 05 May 2025, which is hereby incorporated by reference in its entirety.FIELD OF INVENTION
[0002] The present invention relates to systems and methods for delivering psychological therapy, and more particularly to an Al-guided system for managing group psychological therapeutical experience sessions without a human therapist.BACKGROUND
[0003] The mental health crisis has become increasingly prevalent across all age groups, manifesting as for example as heightened anxiety, loneliness, and diminished emotional resilience. Despite a widespread demand for emotional support, existing mental health services often struggle to meet the needs of individuals seeking help. Traditional therapies can be expensive, difficult to access, and hampered by a shortage of qualified professionals. This situation has led to long waiting times and limited availability of mental health resources for many people.
[0004] In recent years, digital solutions have emerged as a potential way to address some of these challenges. These technologies aim to provide more accessible mental health support through smartphone apps, online platforms, and other digital tools. However, many of these solutions face their own set of limitations. Recipient engagement and retention can be problematic, as individuals may struggle to maintain consistent use of digital mental health tools over time. Additionally, the lack of human connection in some digital interventions may reduce their effectiveness for certain recipients who benefit from interpersonal interaction as part of their mental health support.SUMMARY
[0005] This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the detailed description. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.
[0006] The integration of artificial intelligence (Al) into mental health applications has opened up new possibilities for personalized and adaptive interventions. Al-powered systems canpotentially analyze recipient data to provide tailored recommendations and support. However, the development of Al systems for mental health applications presents its own set of challenges, including ensuring the accuracy and reliability of Al-generated insights, maintaining recipient privacy and data security, and addressing ethical considerations surrounding the use of Al in sensitive healthcare contexts.
[0007] Group therapy approaches have long been recognized as valuable in mental health treatment, offering benefits such as peer support, shared experiences, and cost-effectiveness. However, traditional group therapy sessions typically require the presence of a trained therapist to guide discussions and ensure a therapeutic environment. The limited availability of qualified therapists can restrict access to group therapy options for many individuals.
[0008] As the demand for mental health support continues to grow, there is an increasing need for innovative approaches that can combine the benefits of human interaction, the accessibility of digital platforms, and the potential of Al-driven personalization. Developing systems that can effectively facilitate group-based psychological support while addressing the limitations of both traditional and existing digital approaches remains an ongoing challenge in the field of mental health technology.
[0009] Embodiments of the invention may include a method of controlling a group session of psychological therapeutical experience by at least one computing device or processor. According to some embodiments, the at least one processor may be configured to obtain a group profile data structure characterizing a psychological profile of at least one recipient in a group of recipients engaged in a group session. The processor may be further configured to analyze reaction of the at least one recipient in relation to the group session.
[0010] The term “reaction” may be used in this context to indicate a behaviour of the recipient, or an input of the recipient that may be acquired during a group session, e.g., online, in response to occurrences in the session, or content that has been presented to the recipient during the session. Additionally, or alternatively, the term “reaction” may be used to indicate behaviour or input that may be acquired from the recipient offline, e.g., via a questionnaire to follow up or summarize the events of a session.
[0011] In some embodiments, the processor may also be adapted to enrich the group profile data structure based on said analysis.
[0012] According to some embodiments, the processor may be configured to generate an intervention data element based on the enriched group profile data structure. The intervention data element may define a required selection or adaptation of session content. The at least one processor may be further configured to select, or adapt said session content, based on the intervention data element. In some embodiments, the processor may also be adapted to present the session content to the group of recipients, as part of the psychological therapeutical experience.
[0013] According to some embodiments, the group profile data structure may include a multilayered, interconnected, temporal weighted structure, such as a graph.
[0014] The term "interconnected" in the context of the present invention may refer to the complex relationships and dependencies between different elements within the group profile data structure. This interconnectedness may allow system 10 to represent and analyze the multifaceted nature of psychological profiles and group dynamics. For example, a user's emotional state may be connected to their past experiences, current relationships, and ongoing interactions within the group therapy session. This interconnected structure may enable the system to consider multiple factors when making decisions about interventions or analyzing user progress.
[0015] The "weighted" aspect of the group profde data structure may indicate that different connections or elements within the structure have varying levels of importance or influence. These weights may be dynamically adjusted based on the system's ongoing analysis and learning processes. For instance, recent behavioural patterns may be given more weight than historical data when assessing a user's current psychological state. The weighting system may allow the AI-guided therapy system to prioritize the most relevant information when making decisions about interventions or personalizing content for users.
[0016] The "temporal" nature of the group profile data structure may reflect its ability to capture and represent changes over time. This temporal aspect may allow tracking the progress of users throughout their therapeutic journey, identifying trends in group dynamics, and assessing the effectiveness of interventions over multiple sessions. The temporal dimension may allow the system to consider both short-term fluctuations in users' psychological states and long-term patterns of change. This capability may enable system 10 to provide more nuanced and adaptive therapeutic support, taking into account the evolving needs and progress of individual users and the group as a whole.
[0017] Embodiments of the invention may include a method of controlling a group session of psychological therapeutical experience by at least one computing device. Embodiments of the method may include obtaining a group profile data structure, characterizing a psychological profile of at least one recipient in a group of at least two recipients engaged in the group session, analyzing reaction of the at least one recipient in relation to the group session; enriching the group profile data structure based on said analysis, and generating an intervention data element based on the enriched group profile data structure. The intervention data element may define a required selection or adaptation of session content.
[0018] Embodiments of the method may further include selecting, or adapting said session content, based on the intervention data element; and presenting the session content to the group of recipients, as part of the psychological therapeutical experience.
[0019] According to some embodiments, the group profile data structure may be a multilayered, interconnected, weighted, temporal structure, and may include at least one user layer, representing psychological profiles of one or more recipients of the group of recipients.
[0020] Each user layer of the group profile data structure may include one or more user identifier nodes, identifying specific recipients. Each user identifier node may be connected to one or more characterizing nodes of the user layer, each indicating a specific condition or characteristic of the identified recipient, in a context of a current or historical session.
[0021] Additionally, or alternatively, the group profile data structure may further include a relationship layer, representing relationships between recipients of the group of recipients. The relationship layer may include one or more interaction nodes, defining interactions among the recipients, in a context of a current, or historic group session.
[0022] Additionally, or alternatively, the group profile data structure may further include one or more knowledge layers, associating said psychological profiles and relationships with corresponding knowledge instances in one or more psychological knowledge databases.
[0023] Embodiments of the method may further include analyzing reaction of a specific recipient by capturing audiovisual representation of the specific recipient during the group session, and employing one or more Machine Learning (ML)-based feature extraction models, to extract one or more audiovisual features, representing behaviour of the specific recipient within or beyond the context of the group session.
[0024] Additionally, or alternatively, analyzing reaction of a specific recipient may include prompting the specific recipient to provide a textual data element, representing that recipient’s relation to the session; receiving the textual data element from that specific recipient; and employing one or more ML-based feature extraction models, to extract one or more textual features, describing behaviour of the specific recipient within, or beyond the context of the group session.
[0025] Additionally, or alternatively, analyzing reaction of a specific recipient may include employing one or more first ML-based language models, to generate diagnosis of the specific recipient’s psychological or behavioural condition, based on the one or more audiovisual, and / or textual features.
[0026] Additionally, or alternatively, analyzing reactions of at least one recipient may include employing one or more second ML-based language models, to identify interaction among the recipients, based on their audiovisual or textual features.
[0027] According to some embodiments, enriching the group profile data structure may include, for example producing a new user node pertaining to the specific recipient based on the generated diagnosis and integrating the new user node into the user layer; recalculating a value of at least one user node based on the generated diagnosis; producing a new edge, connecting previously existing user nodes of the user layer; and recalculating a weight of an edge, connecting between user nodes of the user layer.
[0028] Additionally, or alternatively, enriching the group profile data structure may include, for example producing a new interaction node, based on the identified interaction and integrating the new interaction node into the relationship layer; recalculating a value of at least one interaction node based on the generated diagnosis; producing a new edge, connecting a previously existing interaction node of the relationship layer with a user node of the user layer; and recalculating a weight of an edge, connecting between an interaction node of the relationship layer and a user node of the user layer.
[0029] According to some embodiments, at least one knowledge layer may be an ontology layer that may include a plurality of ontology nodes. Each ontology node may represent a knowledge instance that is a psychological term in the one or more psychological knowledge databases.
[0030] Additionally, or alternatively, each ontology node may include a knowledge instance that represents (i) association of the psychological term to a diagnosis, manifested by at least on user node of the user layer, and / or (ii) association of the psychological term to a relationship type, manifested by a relationship node of the relationship layer.
[0031] Additionally, or alternatively, at least one knowledge layer may be a research layer that may include a plurality of research nodes. Each research node may be, or may include a knowledge instance that associates between psychological terms, manifested by ontology nodes of the ontology layer, based on clinical publications.
[0032] According to some embodiments, generating an intervention data element may include based on weights of the group profile data structure, extracting an interconnected substructure of nodes of the enriched group profile data structure, wherein said substructure represents a current context of at least one recipient’s psychological profile; applying a third ML-based language model on the substructure to generate an intervention prompt. Embodiments may further include providing the intervention prompt as input to at least one fourth ML-based language model. The at least one fourth language model may be configured to generate the intervention data element based on said intervention prompt.
[0033] Embodiments of the invention may include a system for controlling a group session of psychological therapeutical experience. Embodiments of the system may include a non-transitory memory device, where modules of instruction code may be stored, and at least one processor associated with the memory device, and configured to execute the modules of instruction code. Upon execution of said modules of instruction code, the at least one processor may be configured to obtain a group profile data structure, characterizing a psychological profile of at least one recipient in a group of at least two recipients engaged in the group session; analyze reaction of the at least one recipient in relation to the group session; enrich the group profile data structure based on said analysis; generate an intervention data element based on the enriched group profile data structure, wherein said intervention data element defines a required selection or adaptation of session content; select, or adapting said session content, based on the intervention data element; and present the session content to the group of recipients, as part of the psychological therapeutical experience.
[0034] Embodiments of the invention may include a method of real-time adaptation of a psychological group therapy session by at least one computing device. Embodiments of the methodmay include presenting, via a session interface implemented on the at least one computing device, session content to a group of recipients participating in the group therapy session, and receiving, by an online machine-learning based session analyzer executing on the at least one computing device, multi-modal input data from the session interface during the group therapy session. The multi-modal input data may include, for example audio data and / or video data captured via input devices.
[0035] According to some embodiments, the online session analyzer may process the audio data using a speech-to-text transcriber to generate textual data representing verbal content from the group therapy session, and / or process the video data using a motion analysis module to extract features representing physical movements, gestures, and facial expressions of at least one recipient.
[0036] The online session analyzer may use a machine-learning based sentiment analysis module in real-time to generate one or more insights regarding the at least one recipient's emotional tone and reactions to the session content.
[0037] The at least one processor may subsequently provide the extracted insights to an online session moderator executing on the at least one computing device to determine, based on the extracted insights, that an adaptation to the session content is required. The online session moderator may subsequently transmit, to the session interface, adapted session content for presentation to the group of recipients.
[0038] The adapted session content may include, for example, clarified content generated in response to detected recipient confusion, extended discussion content generated in response to detected engagement levels exceeding a threshold, supplementary exercises generated in response to detected topic resonance, and the like.
[0039] Embodiments of the invention may include a method of ensuring safety in an AI-guided psychological therapy system by at least one computing device. Embodiments of the method may include receiving, by a first guardrail module executing on the at least one computing device, an input associated with a group therapy session. This input may include, for example, an intervention prompt generated by a decision model or a participant prompt received via input devices. The first guardrail module may analyze the input using at least one of keyword detection, theme analysis, or content similarity search to detect potential issues, including, for example personally identifiable information, off-topic content, inappropriate themes, and the like.
[0040] Responsive to the input being approved by the first guardrail module, embodiments of the invention may include processing, by a large language model (LLM) executing on the at least one computing device, the input to generate an output corresponding to a next intervention for the group therapy session.
[0041] According to some embodiments, a second guardrail module executing on the at least one computing device may analyze the generated output, to detect potential issues therein. Such potential issues may include, for example, hallucinations, content inconsistent with clinical guidelines, harmful content, and the like.
[0042] According to some embodiments, the second guardrail module may classify at least one detected issue by severity and urgency based on a comparison of the detected issue against predefined classification criteria stored in a rules and policies database. In some embodiments, the second guardrail module may enrich the analysis process, e.g., by inject context data comprising user psychological profile information and group dynamics data extracted from a group profde data structure into the analysis. The second guardrail module may thereby adjust sensitivity thresholds based on participant vulnerability indicators.
[0043] Responsive to the output being approved by the second guardrail module, the at least one processor may propagate, or transmit the next intervention to a session interface for presentation to one or more recipients participating in the group therapy session.
[0044] Responsive to the output not being approved by the second guardrail module, the at least one processor may trigger an escalation policy stored in a non-transitory memory. The escalation policy may include, for example, regenerating the output by the language model, consulting one or more (e.g., a plurality) of ML based models to generate an alternative output, or transmitting an escalation request to a human operator interface.BRIEF DESCRIPTION OF FIGURES
[0045] Non-limiting and non-exhaustive examples are described with reference to the following figures.
[0046] Fig. 1 is a block diagram depicting a computing device that may be used within a system for managing group psychological therapy, according to some embodiments of the invention;
[0047] Fig. 2 is a block diagram illustrating a workflow of a psychological therapeutical experience system, which can be implemented according to some embodiments of the invention;
[0048] Fig. 3 is a schematic block diagram showing an architecture of the psychological therapeutical experience system, according to some embodiments of the invention;
[0049] Fig. 3A is a block diagram depicting interrelation between a session analyzer, a diagnostic module, and a group profile data structure, according to some embodiments of the invention.
[0050] Fig. 3B is a block diagram depicting internal architecture of the journey management module, according to some embodiments of the invention.
[0051] Fig. 4A is a block diagram depicting a workflow of session analysis and diagnosis, which can be implemented in the group psychological therapy management system 10, according to some embodiments of the invention;
[0052] Fig. 4B is a block diagram depicting an example of a workflow of a session analyzer module, which can be included in the group psychological therapy management system 10, according to some embodiments of the invention;
[0053] Fig. 5 is a schematic diagram illustrating an example of a group profile data structure, which may be implemented as multi-layered, interconnected graph structure, which can be used in managing group psychological therapy, according to some embodiments of the invention;
[0054] Fig. 6A is a block diagram showing a process of automatically generating an intervention in a psychological group session according to some embodiments of the invention;
[0055] Fig. 6B is a block diagram depicting an example of a workflow of a guardrail model which can be implemented in the group psychological therapy management system, according to some embodiments of the invention; and
[0056] Fig. 7 is a flowchart illustrating a method for automatically controlling a group session of psychological therapeutical experience according to some embodiments of the invention.DETAILED DESCRIPTION
[0057] The following description sets forth exemplary aspects of the present disclosure. It should be recognized, however, that such description is not intended as a limitation on the scope of the present disclosure. Rather, the description also encompasses combinations and modifications to those exemplary aspects described herein.
[0058] The present disclosure relates to a psychological therapeutical experience system that provides automated therapeutic support without direct involvement of (e.g., requiring the presenceof) a professional therapist. The system may utilize artificial intelligence to manage emotional journeys and select appropriate, online or offline interventions for recipients.
[0059] As used herein, the term "intervention" (denoted 300INT) may refer to a purposeful action, strategy, or content implemented within the psychological therapeutical experience system 10, to influence, guide, or support the emotional, cognitive, or behavioral state of one or more recipients participating in a group therapy session. An intervention may, for example, include presenting discussion topics, introducing therapeutic exercises, providing targeted feedback, suggesting coping strategies, or initiating role-playing scenarios. As explained herein, interventions may be dynamically selected, adapted, or generated by the system of the present invention based on recipient profiles, session analysis, and predicted efficacy, to address the current psychological needs and goals of the group or individual participants in the psychological group session.
[0060] In some cases, the psychological therapeutical experience system may be designed to facilitate interactions among groups, or between pairs of non-professional recipients. The system may collect and analyze diverse data from the recipients to inform the therapeutic process. This data collection may include video, audio, text, and responses to psychological questionnaires.
[0061] The system may incorporate one or more Al-powered components that may function as a psychological agent. This agent may autonomously oversee the emotional journey of the recipients and make decisions regarding suitable interventions based on collected data and defined parameters.
[0062] In some implementations, the system may provide therapeutic support through a combination of Al-guided video meetings between recipients and ongoing messaging-based interactions. This multi-modal approach may allow for both structured sessions and continuous engagement.
[0063] By leveraging artificial intelligence and data analysis, embodiments of the invention may provide accessible psychological support that adapts to recipients' needs without requiring direct therapist involvement for each interaction. The automated nature of the system may enable scalable delivery of personalized therapeutic experiences.
[0064] Reference is now made to Fig. 1, which is a block diagram depicting a computing device, which may be included within an embodiment of a system for managing a psychological group therapy session, according to some embodiments.
[0065] Computing device 1 may include a processor or controller 2 that may be, for example, a central processing unit (CPU) processor, a chip or any suitable computing or computational device, an operating system 3, a memory 4, executable code 5, a storage system 6, input devices 7 and output devices 8. Processor 2 (or one or more controllers or processors, possibly across multiple units or devices) may be configured to carry out methods described herein, and / or to execute or act as the various modules, units, etc. More than one computing device 1 may be included in, and one or more computing devices 1 may act as the components of, a system according to embodiments of the invention.
[0066] Operating system 3 may be or may include any code segment (e.g., one similar to executable code 5 described herein) designed and / or configured to perform tasks involving coordination, scheduling, arbitration, supervising, controlling or otherwise managing operation of computing device 1, for example, scheduling execution of software programs or tasks or enabling software programs or other modules or units to communicate. Operating system 3 may be a commercial operating system. It will be noted that an operating system 3 may be an optional component, e.g., in some embodiments, a system may include a computing device that does not require or include an operating system 3.
[0067] Memory 4 may be or may include, for example, a Random- Access Memory (RAM), a read only memory (ROM), a Dynamic RAM (DRAM), a Synchronous DRAM (SD-RAM), a double data rate (DDR) memory chip, a Flash memory, a volatile memory, a non-volatile memory, a cache memory, a buffer, a short term memory unit, a long term memory unit, or other suitable memory units or storage units. Memory 4 may be or may include a plurality of possibly different memory units. Memory 4 may be a computer or processor non-transitory readable medium, or a computer non-transitory storage medium, e.g., a RAM. In one embodiment, a non-transitory storage medium such as memory 4, a hard disk drive, another storage device, etc. may store instructions or code which when executed by a processor may cause the processor to carry out methods as described herein.
[0068] Executable code 5 may be any executable code, e.g., an application, a program, a process, task, or script. Executable code 5 may be executed by processor or controller 2 possibly under control of operating system 3. For example, executable code 5 may be an application that may manage a psychological group therapy session, as further described herein. Although, for the sake of clarity, a single item of executable code 5 is shown in Fig. 1, a system according to someembodiments of the invention may include a plurality of executable code segments similar to executable code 5 that may be loaded into memory 4 and cause processor 2 to carry out methods described herein.
[0069] Storage system 6 may be or may include, for example, a flash memory as known in the art, a memory that is internal to, or embedded in, a micro controller or chip as known in the art, a hard disk drive, a CD-Recordable (CD-R) drive, a Blu-ray disk (BD), a universal serial bus (USB) device or other suitable removable and / or fixed storage unit. Data pertaining to a psychological group therapy session may be stored in storage system 6 and may be loaded from storage system 6 into memory 4 where it may be processed by processor or controller 2. In some embodiments, some of the components shown in Fig. 1 may be omitted. For example, memory 4 may be a non-volatile memory having the storage capacity of storage system 6. Accordingly, although shown as a separate component, storage system 6 may be embedded or included in memory 4.
[0070] Input devices 7 may be or may include any suitable input devices, components, or systems, e.g., a detachable keyboard or keypad, a mouse and the like. Output devices 8 may include one or more (possibly detachable) displays or monitors, speakers and / or any other suitable output devices. Any applicable input / output (I / O) devices may be connected to Computing device 1 as shown by blocks 7 and 8. For example, a wired or wireless network interface card (NIC), a universal serial bus (USB) device or external hard drive may be included in input devices 7 and / or output devices 8. It will be recognized that any suitable number of input devices 7 and output device 8 may be operatively connected to Computing device 1 as shown by blocks 7 and 8.
[0071] A system according to some embodiments of the invention may include components such as, but not limited to, a plurality of central processing units (CPU) or any other suitable multipurpose or specific processors or controllers (e.g., similar to element 2), a plurality of input units, a plurality of output units, a plurality of memory units, and a plurality of storage units.
[0072] The term neural network (NN) or artificial neural network (ANN), e.g., a neural network implementing a machine learning (ML) or artificial intelligence (Al) function, may be used herein to refer to an information processing paradigm that may include nodes, referred to as neurons, organized into layers, with links between the neurons. The links may transfer signals between neurons and may be associated with weights. A NN may be configured or trained for a specific task, e.g., pattern recognition or classification. Training a NN for the specific task mayinvolve adjusting these weights based on examples. Each neuron of an intermediate or last layer may receive an input signal, e.g., a weighted sum of output signals from other neurons, and may process the input signal using a linear or nonlinear function (e.g., an activation function). The results of the input and intermediate layers may be transferred to other neurons, and the results of the output layer may be provided as the output of the NN. Typically, the neurons and links within a NN are represented by mathematical constructs, such as activation functions and matrices of data elements and weights. At least one processor (e.g., processor 2 of Fig. 1) such as one or more CPUs or graphics processing units (GPUs), or a dedicated hardware device may perform the relevant calculations.
[0073] As explained herein, a psychological therapeutical experience system 10 may be implemented to provide automated therapeutic support through guided, psychological group sessions. According to some embodiments, psychological group sessions 400s may be designed to address specific themes or therapeutic goals, providing focused support and guidance for participants. These themed sessions can target various aspects of mental health and personal growth, such as improving interpersonal relationships, overcoming adversity or trauma, managing anxiety or depression, developing emotional resilience, or enhancing self-esteem.
[0074] For example, a session 400S focused on interpersonal relations might guide recipients through exercises in active listening, empathy building, and conflict resolution. Similarly, a session 400S addressing trauma recovery could lead participants through grounding techniques, sharing experiences in a safe environment, and developing coping strategies. These themed sessions are often scripted to ensure a structured and purposeful progression, guiding recipients through carefully designed activities, discussions, and reflections that align with the session's therapeutic objectives. The scripted nature helps maintain focus, ensures all participants receive consistent guidance, and facilitates a cohesive group experience while addressing individual needs within the chosen theme.
[0075] Fig. 2 is a block diagram illustrating a workflow which may be implemented by a psychological therapeutical experience system 10, according to some embodiments of the invention.
[0076] According to some embodiments of the invention, system 10 may be implemented as a software module, a hardware module, or any combination thereof. For example, system 10 may be, or may include one or more computing devices such as element 1 of Fig. 1, and may be adaptedto execute one or more modules of executable code (e.g., element 5 of Fig. 1) to manage guided psychological group treatment sessions, as further described herein.
[0077] As shown in Fig. 2, arrows may represent flow of one or more data elements to and from system 10 and / or among modules or elements of system 10. Some arrows have been omitted on Fig. 2 for the purpose of clarity.
[0078] Psychological therapeutical experience system 10 may include a group profiles data structure 100. Group profiles data structure 100 may be implemented as a multi-layered graph, representing a profde (also referred to herein as a state) of one or more recipients participating in the group session, and their relationships. As explained herein, this structure may serve as a foundation for providing accurate session context, and for personalizing the therapeutic experience.
[0079] Psychological therapeutical experience system 10 may further include a journey management module 200. Journey management module 200 may comprise a multi-agent system component that functions to make decisions regarding the next psychological intervention. These decisions may be based on the recipients' state and context as represented in the group profiles data structure 100.
[0080] The psychological therapeutical experience system 10 may also include a session creator module 300. Session creator 300 may be, or may include a psychological Large Language Model (LLM) assistant. Session creator 300 may function to generate personalized dyadic, or group psychological sessions, building a library of cooperative interventions and adapting them to specific recipient needs.
[0081] Additionally, or alternatively, session creator 300 may be adapted to produce various types of personalized, ad-hoc session content 300SC to be presented to participating users during the psychological group session 400S.
[0082] For example, session content 300SC may include audiovisual content such as customized video clips or music selections tailored to the therapeutic goals of the session. For instance, a short, animated video explaining cognitive behavioural therapy techniques may be presented to help users understand a specific concept. Alternatively, a calming music track may be played during a guided relaxation exercise.
[0083] In another example, session content 300SC may include task instructions. Session content 300SC may comprise textual or audible instructions for performing therapeutic tasks orexercises. For example, the system may present step-by-step instructions for a mindfulness meditation exercise, guiding users through the process of focusing on their breath and observing their thoughts without judgment.
[0084] In another example, session content 300SC may include interactive questionnaires, session content 300SC may include prompted questionnaires or surveys presented as text on the screen. These may be designed to gather real-time feedback from participants about their emotional states, reactions to specific topics, or progress in applying therapeutic techniques discussed in the session.
[0085] In yet another example, session content 300SC may include role-playing scenarios. Session content 300SC may present interactive role-playing scenarios where users are given specific roles and situations to act out. For instance, the system may provide a scenario describing a conflict situation and assign roles to different participants, prompting them to practice conflict resolution skills in a safe, controlled environment. The scenario details and role descriptions may be presented as text, while instructions and prompts for the exercise may be delivered through audio or video guidance. Other examples are also possible.
[0086] Psychological therapeutical experience system 10 may further include, or be communicatively connected to a session interface 400. Session interface 400 may, for example, be implemented as a web application via web browser, or as a mobile application. Session interface 400 may be adapted to facilitate the psychological group session as a guided video call, among the participating recipients. Session interface 400 may further enable presentation of audiovisual session content 300SC, and facilitate direct interaction and engagement with, and among participating recipients, as part of the therapeutic process.
[0087] Psychological therapeutical experience system 10 may further include a diagnostic module 500. As elaborated herein, Diagnostic module 500 may implement LLM-based diagnostics to identify and classify recipients' emotional and behavioral states and needs.
[0088] For example, the diagnostic module 500 may analyze behavioral and psychoemotional states, assess vulnerabilities and resilience of participants, and evaluate relationship dynamics of recipients within the group, and / or between pairs of recipients.
[0089] As explained herein, diagnostic module 500 may employ one or more (e.g., a plurality of) ML-based models (e.g., 510, 520, 530, 540, 550) to provide a feedback 500F to group profiles data structure 100 to adapt, or modify structure 100 so as to reflect recipients' emotional andbehavioral states, in accordance with its analysis of the session. For example, feedback 500F may include updated classifications or scores for emotional profile parameters, such as revised anxiety or depression levels. Additionally, or alternatively, feedback 500F may include new or updated indications of personality trait assessments, identified coping mechanisms, behavioral pattern changes observed during the session, and the like. In another example, feedback 500F may include relationship dynamic indicators, such as changes in engagement symmetry or conflict patterns between recipients. In some cases, feedback 500F may include vulnerability assessments or abuse indicators requiring attention. Feedback 5 OOF may further include confidence scores associated with each diagnostic output, enabling downstream components to assess the reliability of the diagnostic information.
[0090] This process may be implemented online (e.g., in near real time), to intervene in an ongoing session, so as to match an evolving profile or state on specific participating recipients, as reflected by group profiles data structure 100. Additionally, or alternatively, this process may be performed offline, to prepare personalized future sessions so as to accommodate recipient profiles.
[0091] According to some embodiments, j ourney management module 200 may subsequently utilize an intervention efficacy prediction model to determine the most suitable next step in the therapeutic process. As explained herein, journey management module 200 may take into account the analysis provided by the diagnostic module 500 and the current state represented in the group profdes data structure 100.
[0092] Reference is now made to Fig. 3, which is a schematic block diagram showing an example of an implemented architecture of the psychological therapeutical experience system 10, according to some embodiments of the invention. System 10 of Fig. 3 may be the same as system 10 of Fig. 2.
[0093] As shown in Fig. 3, psychological therapeutical experience system 10 may include multiple interconnected modules and components arranged in a hierarchical structure to facilitate automated psychological therapeutical experience. Arrows may represent flow of one or more data elements to and from system 10 and / or among modules or elements of system 10. Some arrows have been omitted on Fig. 3 for the purpose of clarity.
[0094] In the examples provided herein, psychological therapeutical experience system 10 may utilize a group profiles data structure 100, which may be implemented as a multi-layered, interconnected, weighted, temporal graph. However, it may be appreciated that other appropriatemultilayered structures may also be applicable. In these examples, multilayered data structure 100 may include one or more user layers 110, at least one relationship layer 120 and one or more knowledge layers, such as an ontology layer 130, and / or a research layer 140. A more elaborate implementation example of multilayered graph of data structure 100 is discussed herein, e.g., in relation to Fig. 5.
[0095] Each user layer 110 may represent a psychological profile of one or more unique, respective recipients, who may be engaged in a current, or historical psychological therapeutical experience group session. Additionally, or alternatively, at least one user layer may include, or represent psychological profiles of one or more (e.g., a plurality of) recipients, within, or beyond the group of participating recipient users.
[0096] As used herein, the term "psychological profile" may refer to a comprehensive representation of an individual's mental and emotional characteristics, behaviors, and patterns. This profile may include information about the user's personality traits, cognitive tendencies, emotional states, behavioral patterns, mental health history, and current psychological needs or challenges. The psychological profile may be dynamically updated based on ongoing interactions and analyses within the group therapy sessions. It may incorporate various psychological assessments, diagnostic evaluations, and observed behaviors to create a multifaceted understanding of the individual's psychological state. This profile may serve as a foundation for personalizing therapeutic interventions, predicting potential responses to different treatment approaches, and tracking progress over time within the psychological therapeutical experience system 10.
[0097] For example, to characterize a psychological variable such as anxiety for a specific recipient, system 10 may implement a majority voting approach with differential weighting based on the dominance or relevance of each question in a psychological assessment. This weighted majority voting may reduce the probability of classification errors. System 10 may further employ a confidence score to assess the reliability of the characterization. In some embodiments, this confidence score may be derived from a combination of model-based evaluation and scoring provided by a human expert. In some embodiments, diagnostic module 500, and specifically recipient psychological model 530, may be configured to implement this majority voting approach and generate classifications with associated confidence scores.
[0098] Based on the confidence score, system 10 may determine whether data augmentation is required to improve the reliability of the psychological profile. Such data augmentation may include, for example theme analysis, wherein system 10 may take textual input from a recipient, generate multiple rephrasings of the text using a large language model (LLM), and query the LLM to classify each rephrased version to provide consistent labeling (e.g., an anxiety classification).
[0099] Additionally, or alternatively, data augmentation may accommodate deepening of reasoning. For example, given a particular classification, system 10 may generate an opinion article for review by an expert psychologist, explaining why a specific classification was selected based on the available data. In some embodiments, session analyzer 700, and specifically LLM 740, may be configured to perform these data augmentation processes.
[0100] In some embodiments, once system 10 has accumulated substantial psychological information, including both real data and synthetic data with high reliability, system 10 may perform fine-tuning on an existing model to calibrate it for providing the required psychological assessments. Such a fine-tuned model may not require repeating the data augmentation processes, as the augmented knowledge may already be incorporated as part of its training. In cases where degradation in model capabilities is observed, retraining may be performed. In some embodiments, diagnostic module 500 may be configured to perform this fine-tuning process on accumulated data.
[0101] At the conclusion of this process, each node in the psychological profile may receive both a classification and a confidence score. In cases where the confidence score is insufficient, messaging agent 250 may proactively perform data enrichment by prompting additional questions during sessions or adding screening questions, while session interface 400 may introduce specific exercises during meetings. This approach may be analogous to how a human psychologist would seek additional information when lacking certain details about an emotional or personality aspect of a patient. Meanwhile, decision-making based on nodes with low confidence scores may be performed with appropriate caution by journey management module 200 and decision model 220, taking into account the reduced reliability of those particular profile elements.
[0102] User layer 110 may include one or more interconnected user nodes 11 ON. Each user node 11 ON may define a recipient's condition, or characteristic, in the context of that current or historical session.
[0103] For example, each user node 110N may be implemented as a data (e.g., a table), that may include (i) identifiers (e.g., name, ID number, etc.) of a respective recipient, and / or (ii)characteristics of the respective recipient, such as age, gender, nationality, demography, and the like.
[0104] Additionally, or alternatively, at least one user node 1 ION may represent (iii) clinical parameters representing the respective recipient, such as descriptors of medical history, prescribed medication, hospitalizations, and the like.
[0105] As elaborated herein (e.g., in relation to Fig. 4A), session analyzer module 700 may include one or more machine-learning (ML) based models, adapted to extract a diverse range of features 700F (also referred to as “behavioural features”) from the presented group session 400S. Session analyzer module 700 may thereby provide a comprehensive dataset for further analysis.
[0106] Features 700F may include, for example linguistic patterns, derived from speech -to-text transcription (STT), such as word choice, sentence structure, and topic coherence. Features 700F may further include non-verbal cues extracted from video analysis may encompass facial expressions, body language, and gestures.
[0107] Session analyzer 700 may also identify features 700F as emotional indicators through sentiment analysis of both verbal and non-verbal communication.
[0108] Session analyzer 700 may identify features 700F as quantification of interaction dynamics within the group, such as turn-taking patterns, response times, and engagement levels.
[0109] Additionally, session analyzer 700 may identify features 700F as extraction of contextual information like discussion topics, shared experiences, and collective emotional states.
[0110] The multifaceted features 700F of session analyzer 700 may provide a rich dataset that enables subsequent diagnostic models 500 to perform in-depth psychological assessments and allow tailored interventions for the group session.
[0111] According to some embodiments, recipient psychological model 530, vulnerability model 540, and abuse model 550 may receive session features 700F extracted by the session analyzer 700. Models 530, 540 and 550 may process session features 700F to generate various indications of the recipient's psychological state.
[0112] According to some embodiments, recipient psychological model 530 may produce indicators related to the recipient's emotional state, cognitive patterns, and behavioral tendencies. Vulnerability model 540 may generate assessments of the recipient's susceptibility to specific psychological challenges or risks. Abuse model 550 may identify potential signs of abuse or harmful dynamics within the recipient's environment or relationships. By analyzing the sessionfeatures, these models may collaborate to provide a synergistic effect, generating comprehensive evaluation of the recipient's psychological well-being, potential vulnerabilities, and any concerning patterns that may require attention or intervention.
[0113] Models 530, 540 and 550 may subsequently integrate these indications as user nodes 11 ON of user layer 110.
[0114] For example, user nodes 11 ON of user layer 110 may further include or represent conditions such as (iv) current (e.g., pertaining to a present session) mental state diagnoses from the diagnostics model(s) 500, such as emotional needs, a state of depression, a state of anxiety, and the like, (v) previous (e.g., pertaining to a historical session) mental state diagnoses from the diagnostics model(s) 500 (e.g., 530), (vi) a vulnerability indication of the recipient to a specific subject as determined by diagnostics model(s) 500 (e.g., 540), (vii) a state of abuse of the recipient as determined by diagnostics model(s) 500 (e.g., 550), and the like.
[0115] Additionally, or alternatively, user nodes 11 ON may include (viii) one or more indicators of behavioural patterns such as patterns of movement, engagement to current, or previous group therapy sessions, responsiveness, engagement in a discussion, a level of concentration, and the like.
[0116] Relationship layer 120 may represent relationships within subgroups or pairs of participating recipients. Relationship layer 120 may be implemented as an interconnected network of interaction nodes 120N. Each interaction node 120N may include indications of different types of relations between participating recipients in the respective subgroup or pair, in the context of a current or historic group session. These indications may be predicted or generated by respective models (group relationship assessment model 510, group usage pattern model 520) of diagnosis models 500.
[0117] For example, group relationship assessment model 510 may receive from session analyzer 700 session features 700F, such as verbal and non-verbal cues, sentiment analysis of recipient communications, and patterns of agreement or disagreement. Group relationship assessment model 510 may analyze these session features 700F to predict and / or update a first group of relationship indicators. This analysis may be performed online (e.g., in real-time), to affect a currently presented session 400S, or offline, to generate a future session as part of a planned course of therapy. Group relationship assessment model 510 may subsequently integrate these indications as interaction nodes 120N of relationship layer(s) 120.
[0118] For example, interaction nodes 120N pertaining to this first group of relationship indicators may include (i) relationship type indicators, such as "supportive”, "conflicting”, "neutral”, or "evolving", (ii) communication style descriptors, such as "assertive”, "passive”, "collaborative”, or "competitive”., (lii) group role classifications, such as "leader”, "mediator”, or "supporter", and / or (iv) engagement symmetry measures, showing whether interactions are balanced or dominated by certain recipients.
[0119] Additionally, or alternatively, interaction nodes 120N pertaining to the first group of relationship indicators may include (v) interaction frequency metrics, which may quantify how often recipients engage with other, specific recipients during sessions, (vi) emotional resonance scores, indicating the degree of emotional alignment between recipients, (vii) trust level assessments, which may reflect the perceived reliability and openness between recipients, (viii) influence dynamics, representing a relative impact that recipients may have on each other's opinions or behaviors, (ix) conflict resolution patterns, describing how recipients typically navigate disagreements, and / or (x) shared experience indicators, highlighting common backgrounds or challenges among recipients.
[0120] Group usage pattern model 520 may focus on broader trends in how recipients interact over time, to generate interaction nodes 120N pertaining to a second group of relationship indicators. Such relationship indicators may include, for example (i) session attendance patterns for each recipient pair or subgroup, (ii) topic preferences shared among specific recipients, (iii) progress trajectories in relationship development, (iv) consistency of engagement across multiple sessions, and the like.
[0121] By combining outputs from models 510 and 520, the system may dynamically update the interaction nodes 120N to reflect the evolving nature of relationships within the group. This information may be used by other components of the psychological therapeutical experience system 10 to tailor interventions and session content to the specific relational dynamics of the participating recipients.
[0122] As elaborated herein group profdes data structure 100 may include one or more global knowledge layers, that may reflect current psychological knowledge, as reflected by a concurrent, scientific knowledgebase of psychological literature, it may be appreciated that such knowledge layers may provide relevant scientific context to the diagnoses and relations as manifested by respective user layers 110 and relationship layers 120.
[0123] For example, one such knowledge layer may be an ontology layer 130, which may include a network of one or more interconnected ontology nodes 130. Ontology nodes 130 may be implemented as data elements (e.g., tables) that may associate between (a) various types of user conditions (e.g., psychological diagnoses, behavioural and emotional patterns), relationships (e.g., interaction with other recipients), and / or session content with (b) psychological terms, in one or more psychological knowledge databases.
[0124] For example, an ontology node 130 may associate a "conflicting" relationship type indicator from an interaction node 120N with psychological terms such as "interpersonal conflict”, "communication breakdown”, or "cognitive dissonance" in the psychological knowledge database. This association may provide context for interpreting the relationship dynamics observed in the group session.
[0125] Another example may involve linking a "supportive" relationship indicator to terms like "social support”, "empathy”, or "therapeutic alliance" in the knowledge database. This connection may help system 10 understand the positive aspects of recipient interactions and their potential therapeutic benefits.
[0126] In some cases, an ontology node 130 may associate communication style descriptors, such as "assertive" or "passive”, with psychological concepts like "communication patterns”, "selfexpression”, or "interpersonal effectiveness”. This association may aid in understanding the implications of different communication styles on group dynamics and individual therapeutic progress.
[0127] Ontology nodes 130 may also connect group role classifications to relevant psychological theories. For instance, a "leader" role may be associated with concepts like "group dynamics”, "social influence”, or "emergent leadership" in the knowledge database, providing a theoretical framework for interpreting recipient behaviors and interactions within the group context.
[0128] In another example, ontology nodes 130 may associate specific diagnoses (e.g., diagnosed psychological conditions, as represented in nodes 11 ON) with relevant behavioral characteristics described in psychological literature.
[0129] For example, an ontology node 130 may link a diagnosis of Post-Traumatic Stress Disorder (PTSD) with various behavioral and psychological characteristics commonly associated with the condition.
[0130] Such association may include, for example, connections to terms such as "hypervigilance”, "avoidance behaviors”, "intrusive thoughts”, "emotional numbing”, and "exaggerated startle response”. By establishing these links, system 10 may gain a more comprehensive understanding of the potential manifestations and impacts of PTSD within the group therapy context.
[0131] Additionally, the ontology node 130 may connect the PTSD diagnosis to related concepts like "trauma-informed care", "cognitive-behavioral therapy for PTSD", or "exposure therapy". These associations may provide the system with relevant therapeutic approaches and interventions specifically tailored to addressing PTSD symptoms within the group setting.
[0132] Furthermore, ontology nodes 130N may link PTSD to potential comorbid conditions often described in psychological literature, such as depression, anxiety disorders, or substance abuse. This interconnected knowledge may allow the system to consider the complex interplay of multiple psychological factors when analyzing recipient behavior and designing interventions.
[0133] Another example of a knowledge layer may include a research layer 140, which may include one or more interconnected research nodes 140N. Research nodes 140N may associate knowledge instances from clinical publications to the psychological terms of ontology nodes 130N.
[0134] Pertaining to the example of PTSD, a research node 140N might associate the psychological term "PTSD" with specific clinical findings from recent publications, demonstrating the efficacy of group therapy for veterans with combat-related trauma. This association creates a direct link between abstract psychological concepts in the ontology layer and evidence-based research findings that support specific therapeutic approaches.
[0135] Such association between clinical research and psychological terms may ground the system's 10 therapeutic decisions in empirical evidence rather than theoretical constructs alone. It may also enable the Al-guided system 10 to access the most current and relevant research findings when generating interventions for specific psychological conditions or relationship dynamics.
[0136] This association may further create a continuously updatable knowledge foundation that can evolve as new research emerges, ensuring that system 10 remains aligned with best practices in psychological therapeutical experience.
[0137] Additionally, by connecting abstract concepts to concrete research findings, system 10 can provide more transparent and justifiable rationales for its therapeutic recommendations, potentially increasing recipient trust and engagement with the treatment process.
[0138] As shown in Fig. 3, psychological therapeutical experience system 10 may further include a journey management module 200. Journey management module 200 may include an orchestration agent 210, adapted to coordinate with various subcomponents, as elaborated herein.
[0139] According to some embodiments, journey management module 200 may include a decision model 220 that may, in turn, include a rules module 224 and an efficacy prediction model 226. As elaborated herein (e.g., in relation to Fig. 6A), decision model 220 may employ rules module 224 and efficacy prediction model 226 to determine required psychological interventions, based on the participating recipients' profiles, as reflected by group profdes data structure 100.
[0140] Reference is also made to Fig. 3A, which is a block diagram depicting data flow between session analyzer 700, diagnostic module 500, and group profile data structure 100, according to some embodiments of the invention.
[0141] As shown in Fig. 3A, session analyzer 700 outputs extracted features 700F, which are provided to diagnostic module 500. Diagnostic module 500 may include multiple specialized submodels including relationship assessment model 510, group usage pattern model 520, recipient psychological model 530, vulnerability model 540, and abuse model 550. Each sub-model may process the extracted features 700F and generates corresponding diagnostic features: relationship assessment features 51 OF, group usage features 520F, recipient psychological features 530F, vulnerability features 540F, and abuse model output 550F, respectively.
[0142] The collective output of diagnostic module 500 may be provided as feedback 500F, comprising the combined features 510F, 520F, 530F, 540F, and 550F generated by the respective sub-models. Feedback 500F may be directed to group profile data structure 100, which may be implemented as a multi-layered structure comprising user layer 110 containing user nodes 110N, relationship layer 120 containing interaction nodes 120N, knowledge layer (ontology) 130 containing ontology nodes 13 ON, knowledge layer (research) 140 containing research nodes 140N, and context subgraph 150.
[0143] Context subgraph 150 may represent a dynamically extracted interconnected substructure of nodes relevant to the current session context. This extraction may be performed via a Retrieval- Augmented Generation (RAG) query as described herein. For example, contextsubgraph 150 may include user nodes 11 ON representing the current participants, their relevant emotional profile parameters, interaction nodes 120N representing their relationship dynamics, connected ontology nodes 13 ON representing relevant psychological terms, and research nodes 140N representing applicable therapeutic interventions.
[0144] As depicted in Fig. 3 A, feedback 500F specifically updates user nodes 11 ON in user layer 110 and interaction nodes 120N in relationship layer 120. This feedback mechanism may enable diagnostic module 500 to enrich and refine the psychological profiles and relationship data stored in group profile data structure 100 based on session analysis performed by session analyzer 700. By continuously and iteratively updating user nodes 11 ON and interaction nodes 120N, the system may maintain current and accurate representations of recipients’ psychological states and inter-recipient dynamics throughout the course of group therapeutic sessions.
[0145] Reference is also made to Fig. 3B, which is a block diagram depicting internal architecture of journey management module 200, according to some embodiments of the invention.
[0146] As shown in Fig. 3B, journey management module 200 may include an orchestration agent 210, which may coordinate operations among the various subcomponents within journey management module 200. Orchestration agent 210 may manage the flow of data and control signals between subcomponents to facilitate cohesive decision-making and intervention generation.
[0147] Journey management module 200 may further include a decision model 220. As depicted in Fig. 3B, decision model 220 may include several interconnected components such as a rules and policies module 224, an efficacy prediction model 226, and a success metrics module 227. Rules and policies module 224 may store and apply predefined therapeutic guidelines and ethical constraints. Efficacy prediction model 226 may predict the effectiveness of potential interventions based on recipient profiles and session context. Success metrics module 227 may calculate success metrics 227SM and provide these metrics as feedback via efficacy score 226S to efficacy prediction model 226, thereby creating a feedback loop for continuous improvement of intervention effectiveness.
[0148] As shown in Fig. 3B, journey management module 200 may also include an online session analyzer 232 and an online session moderator 234. Online session analyzer 232 may provide real-time analysis capabilities during ongoing group therapy sessions, extracting insightsregarding recipients' reactions to session content as it unfolds. Online session moderator 234 may make rapid adjustments to session content based on real-time insights provided by online session analyzer 232.
[0149] Journey management module 200 may further include a personalization module 240, which may enhance and personalize intervention prompts 200PR by adapting interventions to match specific diagnoses or mental states of participating recipients, as manifested in group profile data structure 100.
[0150] Additionally, journey management module 200 may include a messaging agent 250, which may provide an auxiliary messaging communication channel with session interface 400. Messaging agent 250 may prompt recipients to respond to ongoing sessions, present questionnaires, and facilitate other forms of interaction outside the primary video-based therapy sessions.
[0151] As depicted in Fig. 3B, decision model 220 may output an intervention prompt 200PR. Intervention prompt 200PR may define a required psychological intervention to be generated by downstream components such as session creator 300.
[0152] Reference is now made to Fig. 4A, which is a block diagram depicting workflow of session analysis, which can be implemented in the group psychological therapy management system 10, according to some embodiments of the invention. This workflow may be designed to analyze reactions of recipients participating in group sessions and generate diagnostic features based on various inputs.
[0153] According to some embodiments, session interface 400 may receive, e.g., via input devices 7 of Fig. 1, various types of input data 400IN during a group session. The input devices 7 may include, for example a microphone, a camera, a keyboard, a touchscreen and the like. The respective input data 400IN may include, for example audio, video, text, and responses to questionnaires. The diverse input data 400IN may provide a comprehensive representation of recipient behavior and interactions during the session.
[0154] Session analyzer 700 may process the input data 400IN received from the session interface 400. The session analyzer 700 may comprise multiple analysis modules designed to extract various features 700F from the input data, as elaborated in the following examples:
[0155] Session analyzer 700 may employ a speech-to-text transcriber (STT) 710, that may be configured to convert audio input 400IN into text, enabling further analysis of verbal content from the session.
[0156] In yet another example, session analyzer 700 may also a large language model (LLM) 740, configured to process textual data from STT 710, to extract features 700F representing deeper semantic meaning and context from recipient interactions.
[0157] Additionally, or alternatively, session analyzer 700 may include motion analysis modules 720. Motion analysis modules 720 may process video input to extract features 700F representing information about recipients' physical movements, gestures, and facial expressions during the session.
[0158] In another example, session analyzer 700 may also include sentiment analysis modules 730, configured to analyze text and audio inputs 400IN to determine features 700F representing the emotional tone and sentiment expressed by recipients during the session. Other types of features 700F may also be possible.
[0159] Reference is further made to Fig. 4B, which is a block diagram depicting an example workflow of session analyzer 700, which can be implemented in the group psychological therapy management system 10, according to some embodiments of the invention. In this example, the workflow may be designed to answer diagnostic questions about participants, such as determining the stress level of a participant, understanding how a participant reacts when angry, or identifying what input triggers anger with a participant.
[0160] As shown in Fig. 4B, the workflow may receive multi-modal data as input, which may include participant audio tracks and participant video tracks. Session analyzer 700 may process this input through multiple parallel analysis paths, each utilizing LLM 740 in various configurations including generic, trained, or mini model variants.
[0161] In the example of Fig. 4B, in a first analysis path, the multi-modal data may be processed by querying LLM 740, which may then search history data for conversational content. This path may produce a first answer, with an associated confidence level.
[0162] In a second analysis path, the multi-modal data may be processed by querying LLM 740 in its generic, trained, or mini-model configuration. This query may search the interconnected, multi-layered data structure of group profile data structure 100 for the stress level of similar people. This path may produce a second answer with an associated confidence level.
[0163] In a third analysis path, participant audio tracks may be processed by asking LLM 740 for tone analysis. This analysis may generate a third answer with an associated confidence level.
[0164] In a fourth analysis path, participant video tracks may be processed by asking ML or LLM 740 to perform sentiment analysis via facial expression analysis. This path may similarly produce a forth answer with an associated confidence level.
[0165] The outputs from all four analysis paths, each comprising an answer and confidence level optionally obtained via a different ML or LLM modality, may then be combined to produce a final output, e.g., through majority voting. The final output may include a majority vote result along with an overall confidence level. This multi-path approach with majority voting may reduce the probability of classification errors by leveraging multiple analysis modalities and data sources to arrive at more reliable diagnostic conclusions about participant psychological states.
[0166] According to some embodiments, session analyzer 700 and diagnostic module 500 may perform operations on group profile data structure 100 using graph database frameworks as known in the art. Such frameworks may natively support weighted, interconnected graph structures of the type described herein, enabling efficient traversal, querying, and updating of the multilayered structure comprising user layers 110, relationship layers 120, and knowledge layers 130, 140.
[0167] Session analyzer 700 and diagnostic module 500 may combine LLM reasoning with graph retrieval to perform LLM-based feature extraction over graph structures. This approach may be directly applicable to the RAG-based querying described herein for extracting context subgraphs 150 from group profile data structure 100.
[0168] According to some embodiments, session analyzer modules (e.g., 710, 720, 730, and 740) may utilize pre-trained models for audiovisual feature extraction, sentiment analysis, and behavioral pattern recognition. In some cases, post-training fine-tuning may be applied to proprietary session data to improve extraction quality for the specific therapeutic domain.
[0169] According to some embodiments, session analyzer 700 may extract features 700F such as emotional profile parameters 700F-EP representing psycho-emotional state indicators for each recipient. These indicators may include, for example, stress, depression, anxiety, emotional wellbeing, self-esteem, emotional valence, emotional regulation capacity, loneliness, and resilience. Each indicator may be characterized with a baseline value, a current state value, and a trend over time.
[0170] Session analyzer 700 may further extract a dominant mood parameter, representing the prevailing affective tone characterizing the recipient across sessions. Examples of dominant mood may include irritable, anxious, flat, or elevated. Additionally, session analyzer 700 may identify dominant emotions, representing the most frequently expressed or detected discrete emotions. Examples of dominant emotions may include sadness, anger, fear, guilt, shame, hope, gratitude, frustration, jealousy, contempt, curiosity, joy, disgust, surprise, and embarrassment. These emotional profile parameters 700F-EP may be used to populate user nodes 110N of user layer 110 in group profile data structure 100. For example, a user node 11 ON may be populated with an anxiety classification and associated confidence score characterizing the recipient's current emotional state.
[0171] Extraction of emotional profile parameters may be performed using various techniques. These techniques may include thematic analysis of the content of meetings, sentiment analysis using video and voice data, facial micro-expression classification using proprietary or third-party ML-based models, vocal prosody analysis, and self-report questionnaires presented via session interface 400 or messaging agent 250.
[0172] According to some embodiments, session analyzer 700 may extract features 700F such as personality parameters 700F-PP characterizing enduring psychological traits of recipients. These parameters may include, for example, openness, interpersonal sensitivity, assertiveness, empathy capacity, conflict style, need for approval, tolerance for ambiguity, and need for control. Conflict style may be further categorized as accommodating, competing, avoiding, collaborating, or compromising.
[0173] Extraction of personality parameters may involve analyzing various behavioral indicators. For example, openness may be assessed based on depth of self-disclosure (distinguishing surface content from vulnerable content), body language openness (such as posture and arm positioning), and topic shifting patterns when emotional content is introduced. These parameters may be calculated for each session and aggregated to establish a baseline for each participant. These personality parameters 700F-PP may be used to populate user nodes 11 ON of user layer 110 in group profile data structure 100. For example, a user node 11 ON may be populated with an openness score and a conflict style classification characterizing the recipient's enduring psychological traits.
[0174] According to some embodiments, session analyzer 700 may extract features 700F such as coping mechanism indicators 700F-CM employed by recipients during group sessions. These coping mechanisms may include, for example, avoidance, intellectualization, denial, humor as deflection, cognitive reframing, social support seeking, passive aggression, and splitting.
[0175] Extraction of coping mechanism indicators may involve detecting specific behavioral patterns. For example, avoidance may be identified through topic avoidance detection, such as recurring redirection from specific themes, and non-verbal withdrawal cues, such as gaze aversion and physical distancing. These coping mechanism indicators 700F-CM may be used to populate user nodes 110N of user layer 110 in group profile data structure 100. For example, a user node 11 ON may be populated with an avoidance indicator and associated behavioral pattern data characterizing the recipient's coping strategies.
[0176] According to some embodiments, session analyzer 700 may extract features 700F such as behavioral pattern features 700F-BP representing observable behaviors during group sessions. These behavioral patterns may include linguistic patterns such as word choice, frequency of stress-related terms, speech rate, and sentence fragmentation. Vocal tone analysis may capture features such as pitch elevation, speech tempo, and vocal tension. Facial expression and body language analysis may identify indicators such as jaw clenching, fidgeting, and reduced eye contact.
[0177] Additional behavioral patterns may include response latency, verbal output volume, self-disclosure depth and frequency, listening versus talking ratio, conversation initiation patterns, and gaze patterns such as gaze directed at speaker, averted gaze, or scanning of the group. These parameters may be measured to establish a baseline for each recipient, and subsequent deviation from the baseline may be used to assess changes in emotional parameters. For example, changes in tone of voice and body language may indicate that a current topic has greater emotional influence on the recipient. These behavioral pattern features 700F-BP may be used to populate user nodes 110N of user layer 110 in group profile data structure 100. For example, a user node 11 ON may be populated with a baseline speech rate value and a current engagement level characterizing the recipient's observable behaviors during sessions.
[0178] According to some embodiments, session analyzer 700 may extract features 700F such as relationship profde parameters 700F-RP characterizing interpersonal dynamics between recipients. These parameters may be used to populate interaction nodes 120N of relationship layer 120 in group profile data structure 100. For example, an interaction node 120N may be populatedwith a turn-taking frequency value and an engagement symmetry measure characterizing the conversational dynamics between a specific pair of recipients.
[0179] Conversational dynamics parameters may include turn-taking frequency, synchronization in body language, smile ratio, silence ratio, deep conversations ratio, and interruption patterns including frequency and type such as supportive versus competitive interruptions. Additional conversational dynamics parameters may include verbal mirroring representing adoption of a partner's language, terms, or phrasing, response latency between partners, topic initiation balance indicating who introduces new topics, laughter co-occurrence distinguishing shared from unilateral laughter, emotional contagion indicators representing mood alignment over session duration, engagement symmetry indicating balanced versus one-sided participation, and repair attempts following misunderstandings or ruptures.
[0180] Quality of empathetic dialogue parameters may include response to partner's disclosures, which may be categorized as validation, dismissal, redirection to self-narrative, advice-giving, or silence. Additional parameters may include change in engagement level when sensitive topics arise, reflective listening indicators such as paraphrasing and summarizing partner's content, emotional attunement representing appropriate matching of emotional tone, follow-up references to partner's previously shared content across sessions, ratio of questions asked about partner versus statements about self, and tolerance for partner's emotional expression indicating whether the recipient holds space versus rushing to fix or resolve.
[0181] In some cases, the session features 700F generated by the session analyzer 700 may be further processed by one or more diagnostic models within the diagnostic module 500. These diagnostic models may include the relationship assessment model 510, the group usage pattern model 520, the recipient psychological model 530, the vulnerability model 540, and the abuse model 550.
[0182] Relationship assessment model 510 may analyze the session features 700F to generate relationship assessment features 51 OF. As shown in Fig. 3 A, relationship assessment model 510 may be one of the specialized sub-models within diagnostic module 500. Features 51 OF may provide insights into the dynamics and quality of relationships between recipients in the group session, and may be provided as feedback 500F for updating groups profde data structure 100.
[0183] Group usage pattern model 520 may process the session features 700F to produce group usage features 520F. These features may reflect patterns reflecting how recipients engagewith system 10 and interact within the group over time, and may be provided as feedback 5 OOF for updating group profile data structure 100.
[0184] Recipient psychological model 530 may examine the session features 700F to create recipient psychological features 530F. As shown in Fig. 3A, recipient psychological model 530 may process extracted features 700F and generate corresponding diagnostic features 530F. These features may offer diagnoses, and insights into individual recipients' psychological conditions and behaviors during the session, and may be provided as feedback 500F for updating group profile data structure 100.
[0185] Vulnerability model 540 may analyze the session features 700F to generate vulnerability features 540F. These features may indicate potential vulnerabilities or risk factors for individual recipients based on their behavior and interactions, and may be provided as feedback 500F for updating group profile data structure 100.
[0186] Abuse model 550 may process session features 700F to produce a fifth model output 550F. This output may identify potential signs of abuse or harmful dynamics within the group or between specific recipients, and may be provided as feedback 5 OOF for updating group profile data structure 100.
[0187] By employing these multiple diagnostic models, the psychological therapeutical experience system 10 may generate a comprehensive set of diagnostic features based on the analyzed reactions of recipients during the group session, providing valuable insights for personalizing interventions and guiding the therapeutic process.
[0188] The psychological therapeutical experience system 10 may include a messaging agent 250. This module may provide an auxiliary messaging communication channel with the session interface 400, which may be implemented as a mobile application on a participant's device. The messaging agent 250 may be configured to prompt users to respond to ongoing sessions, present questionnaires, and facilitate other forms of interaction outside the primary video-based therapy sessions. This auxiliary channel may allow the system to receive online feedback for improving future sessions, as well as enabling interrupts or interventions during a current session.
[0189] For example, the messaging agent 250 may send a quick survey about a participant's emotional state after a particularly intense discussion, or provide additional resources related to topics covered in the session. By offering this supplementary mode of communication, themessaging agent 250 may enhance the overall therapeutic experience and provide valuable data for the continuous improvement of the psychological therapeutical experience system.
[0190] For example, psychological therapeutical experience system 10 may employ messaging agent 250 to prompt a specific recipient to provide a textual data element (e.g., a response to a question), representing that recipient’s relation to the session. A session analyzer (e.g., 232 / 700) may receive the textual data element from that specific recipient, and employ one or more ML-based feature extraction models (e.g., 710, 720, 730, 740), to extract one or more textual features 700F, describing the specific recipient in reaction to the group session. As elaborated herein, decision model may collaborate with session creator LM to utilize the one or more textual features 700F to create therefrom an appropriate intervention prompt 300INT. As explained herein, intervention prompt 300INT will in turn be converted to an appropriate session content 300SC, to be presented or included in session 400S.
[0191] Reference is now made to Fig. 5. According to some embodiments, the group profiles data structure 100 may be implemented as a multi-layered, interconnected, weighted, temporal structure (e.g., an interconnected graph). Fig. 5 illustrates a non-limiting example of such an implementation of the group profiles data structure 100.
[0192] As shown in Fig. 5, the group profiles data structure 100 may include four interconnected layers. The basic layer may be labeled as " 110 user layer", where each node 11 ON may pertain to, or represent a user, or recipient of group therapy sessions. The relations between these users may be manifested by nodes 120N of a relationship layer 120. Above the user layer, two knowledge layers may be present - one labeled as "140 Knowledge layer (Research)" and another labeled as "130 Knowledge layer (Ontology)".
[0193] In the example of Fig. 5, user layer 110 is presented as a weighted, interconnected graph, consisting of a plurality of user layer nodes, also referred to herein as "user nodes" 11 ON. User layer node(s) 11 ON may include various elements representing different aspects of specific recipient's profiles.
[0194] For example, Fig. 5 shows user layer nodes 110N that may be used to identify specific recipients. These are labeled in Fig. 5 as "User j5dk", "User 77d2" and "User 33b3", and may be referred to herein as identifier nodes. One or more (e.g., each) of the identifier user nodes 11 ON may be connected to other nodes 110N of user layer 110, that may indicate specific characteristics of the identified recipients in a context of a current or historical session. These are labeled in Fig.5 as "Veteran", "Session xw3e", "Gen Z", and "Engagement", and may be referred to herein as user characterizing nodes 11 ON. Each user identifier node 11 ON and corresponding and user characterizing nodes 1 ION may be viewed as representing a minimal psychological profile of the respective recipient user. This psychological profile being minimal in a sense that it only pertains to the properties of the specific recipient. These user layer nodes 11 ON may be interconnected within the user layer, or via relationship layer 120 to represent relationships or associations between different recipient attributes or sessions.
[0195] Research layer 140, may contain research nodes 140N. In Fig. 5, these research nodes 140N may include elements labeled as "Intervention 4c77", "Article e71c", "Intervention 3d37", "Article xd4g", "Intervention be6f ', and "Article 6982". These research nodes 140N may represent specific research findings, interventions, or articles that may be relevant to the psychological therapeutical experience process.
[0196] Ontology layer 130 may contain ontology nodes 130N. In the example shown in Fig.5, ontology nodes 13 ON may include elements labeled as "Physiological Stimulation", "PTSD", "Grief1, "Sadness", and "Stress". These ontology nodes 130N may represent broader psychological concepts or conditions that may be relevant to the treatment process.
[0197] As shown in the example of Fig. 5, Group profiles data structure 100 may feature interconnections between nodes across all layers. These interconnections may be represented by weighted edges, with numerical values indicating the strength of connections between different nodes. For example, Fig. 5 shows connections between user layer nodes 110N and research nodes 140N, as well as connections between research nodes 140N and ontology nodes 13 ON.
[0198] In some cases, the group profiles data structure 100 may be dynamically updated based on analysis of recipient interactions and behaviors during psychological therapeutical experience sessions. For example, the session analyzer 700 may analyze recipient reactions during a group session, and the diagnostic module 500 may generate new diagnostic information based on this analysis. This new diagnostic information may be used to produce a new user node 110N, which may then be integrated into (e.g., connected to other nodes of) group profiles data structure 100, thereby enriching the psychological profde of the specific recipient.
[0199] For example, 10 may create a new user node 11 ON when a newly detected characteristic differs from existing nodes by more than a predefined similarity threshold, or when the confidence score of the new classification exceeds a minimum threshold value. Conversely,system 10 may update an existing node when the new information pertains to a characteristic already represented in the group profile data structure 100.
[0200] Similarly, the relationship assessment model 510 and the group usage pattern model 520 may identify new interactions or patterns among recipients. Based on this identification, a new interaction node may be produced and integrated into the relationship layer 120 of the group profdes data structure 100, further enriching the representation of relationships within the recipient group.
[0201] The interconnected nature of the group profiles data structure 100 may allow the psychological therapeutical experience system 10 to leverage information across different layers when making decisions about interventions or personalizing treatment approaches. For example, the journey management module 200 may use the connections between user layer node(s) 11 ON, research nodes 140N, and ontology nodes 13 ON to inform the selection of appropriate interventions for specific recipients or groups, as explained herein.
[0202] Reference is now made to Fig. 6A, which is a block diagram showing a process of automatically generating an intervention 300INT in a psychological group session according to some embodiments of the invention.
[0203] According to some embodiments, psychological therapeutical experience system 10 (e.g., the same as system 10 of Figs. 2 and 3) may implement a psychological intervention process, as illustrated in Fig. 6A. This process may involve selecting, evaluating, and optionally adapting interventions 300INT based on specific recipient's profdes, as manifested in group profiles data structure 100.
[0204] In some embodiments, decision model 220 may employ a Retrieval-Augmented Generation (RAG) approach to query the group profiles data structure 100. The psychological therapeutical experience system 10 may extract an interconnected substructure 150 of nodes from the group profiles data structure 100 in response to the RAG query. This extraction may be based on weights of the group profiles data structure 100.
[0205] Additionally, decision model 220 may further query an intervention library 900 that may contain additional, cooperative psychological interventions.
[0206] In other words, decision model 220 may employ the RAG approach to retrieve relevant information from the group profiles data structure 100, which may include recipient psychological profdes, relationship dynamics, and historical session data, and further extend to the interventionlibrary 900, to access information about previously successful interventions or established therapeutic techniques.
[0207] The retrieved information may be used to augment the current context of the group session. This augmentation process may combine real-time session data with historical and contextual information from the group profiles data structure 100 and, in some cases, the intervention library 900.
[0208] In some implementations, the efficacy prediction model 226 may be a machinelearning based model, adapted to predict the most effective intervention based on the query response, which may include substructure 150 and optionally psychological interventions from the intervention library 900. This process may produce an optimal intervention prompt 200PR.
[0209] The creation of prompt 200PR may involve a dual-stage process. In the first stage, efficacy prediction model 226 may select the most appropriate intervention from a library of interventions known to address specific diagnoses or mental states of participants. For instance, efficacy prediction model 226 may select a cognitive restructuring intervention for a participant exhibiting symptoms of depression. Based on this selection, the decision module 220 may generate an initial prompt 200PR. In the second stage, a personalization module 240, which may be implemented as a large language model, may enhance, and personalize the initial prompt 200PR. For example, personalization module 240 may adapt the required intervention to match the specific diagnoses or mental states of the participating users, as manifested in the group profiles data structure 100. Pertaining to the above example, personalization module 240 may tailor the cognitive restructuring intervention by incorporating specific themes or experiences relevant to the participant's personal history or current life situation, as reflected in their user profile within the group profiles data structure 100. This dual-stage process may enable system 10 to generate prompts 200PR for interventions 300INT that are both evidence-based and personalized to the unique needs of each participant in the group therapy session.
[0210] For example, in embodiments where decision model 220 is implemented as a rulebased model 224, the selection of interventions may depend on both the discussion topics from previous sessions and the current psychological state indicators (e.g., group profile 100) of the recipients. In a first scenario, two students who discussed or experienced exam-related stress in a previous session and have medium to high stress levels, as indicated by their user nodes 110N in group profile data structure 100, may receive stress regulation exercises in the next session. In asecond scenario, two students who discussed the same topic but have low stress level indicators may receive a more reflective session dealing with the meaning of success and failure and their implications on self-worth. This rule-based approach may enable decision model 220 to match intervention types (and subsequent prompts 200PR) to recipient states based on predefined therapeutic guidelines stored in rules and policies module 224.
[0211] According to some embodiments, efficacy prediction model 226 may (e.g., during a training stage) undergo supervised or semi-supervised training. This training process may involve using historical data from past group therapy sessions, including the interventions applied and their measured outcomes. Efficacy prediction model 226 may be trained on pairs of input data, which may include (a) substructures 150, extracted from the group profiles data structure 100 and intervention descriptions 900 and (b) corresponding output data (representing the efficacy of those interventions).
[0212] In semi-supervised scenarios, efficacy prediction model 226 may further leverage unlabeled data to improve its performance. The training process may involve iterative optimization of the model's parameters to minimize prediction errors, potentially using techniques such as gradient descent or other appropriate optimization algorithms.
[0213] During inference on new data instances, efficacy prediction model 226 may take as input substructure 150 extracted from the group profiles data structure 100, which may represent the current state and context of the group therapy session. Efficacy prediction model 226 may also consider potential interventions, either from the intervention library 900 or generated dynamically.
[0214] For each potential intervention, efficacy prediction model 226 may produce a predicted efficacy score 226S. These predictions may be based on patterns learned during training, allowing the model to generalize from past experiences to new, unseen situations. The intervention with the highest predicted efficacy score 226S may be selected as basis for generating a next intervention 300INT, as explained herein. Additionally, or alternatively, efficacy scores 226S may be used to rank multiple interventions for consideration by other components of system 10.
[0215] According to some embodiments, session creator 300 may be, or may include one or more ML-based large language models (LLMs). Session creator model 300 may be configured to receive intervention prompt 200PR as input, and generate, based on that input, an intervention data element 300INT. In some cases, generated intervention data element 300INT may be anaudiovisual element, such as a new session content 300SC, which may be presented via the session interface 400 to participating recipients.
[0216] As shown in Fig. 6A, psychological therapeutical experience system 10 may incorporate a success metrics module 227. The success metrics module 227 may calculate success metrics 227SM, and provide these metrics as feedback to the efficacy prediction model 226. Psychological therapeutical experience system 10 may thus creating a feedback loop for evaluating, retraining and improving intervention effectiveness.
[0217] Success metrics module 227 may calculate various success metrics 227SM to evaluate the effectiveness of interventions and overall session performance. Metrics 227SM may provide feedback to the efficacy prediction model 226, creating a feedback loop for continuous improvement of the psychological therapeutical experience system 10.
[0218] For example, success metrics 227SM may include measures of user session engagement, which could be derived from analyzing participants' interaction patterns, duration of active participation, and frequency of contributions during group sessions.
[0219] System 10 may also incorporate metrics 227SM of behavior, potentially based on behavioral features 700F extracted by the session analyzer 700. These behavioral metrics 227SM may include changes in emotional expression, improvements in communication patterns, or reductions in indicators of stress or anxiety as observed through video and audio analysis.
[0220] Content-based success metrics 227SM may be derived from responses to questionnaires or surveys presented during or after sessions, providing direct feedback on participants' perceived progress and satisfaction with the therapeutic process.
[0221] Additionally, or alternatively, success metrics 227SM may encompass measures of temporal improvement in recipients' psychological conditions, as manifested in the group profile data structure 100. These could include tracking changes in user nodes 11 ON over time, improvements in relationship dynamics as reflected in interaction nodes 120N, or positive shifts in psychological states as identified by the various diagnostic models (510, 520, 530, 540, 550). By analyzing these multifaceted success metrics 227SM, system 10 may continuously refine its intervention strategies and adapt to the evolving needs of participants, potentially enhancing the overall efficacy of the Al-guided group therapy approach.
[0222] The psychological therapeutical experience system 10 may incorporate a guardrail model 800, which may be implemented as a large language model (LLM) adapted to identifyprecarious conditions based on session features 700F. The guardrail model 800 may collaborate with the personalization module 240 to ensure the safety and ethical conduct of group sessions. By analyzing the extracted features 700F, the guardrail model 800 may detect potential risks or sensitive situations that could arise during the therapy session. This model may then work in conjunction with the personalization module 240 to avoid producing intervention prompts 200PR (and subsequent interventions 300INT) that could jeopardize the well-being of participants or violate ethical guidelines. The guardrail model 800 may reference the rules and policies database 224 to align its decision-making process with established ethical standards and safety protocols for psychological therapeutical experience. This additional layer of oversight may help maintain the integrity of the Al-guided therapy system, ensuring that all interventions are not only effective but also safe and ethically sound.
[0223] The online session analyzer module 232 may be implemented as a lightweight version of the session analyzer 700, adapted to provide real-time analysis capabilities during ongoing group therapy sessions. This module may include a speech-to-text analysis component, which may be configured to extract insights regarding users' reactions or relations to the content of the session as it unfolds. The online session analyzer 232 may process audio input in near real-time, identifying key phrases, emotional tones, or patterns of interaction that indicate how participants are engaging with the session content.
[0224] The online session moderator 234 may function as a streamlined version of the decision model 220, designed to make rapid adjustments to the session content based on the realtime insights provided by the online session analyzer 232. This module may be capable of adapting the presented content in real-time or near real-time to accommodate participants' reactions and ensure optimal engagement and therapeutic benefit. For example, if the online session analyzer 232 detects that a participant is expressing confusion or difficulty understanding certain content, the online session moderator 234 may promptly intervene to clarify or rephrase the information. In another scenario, if the analyzer module 232 identifies that a particular topic is resonating strongly with the group, the session moderator 234 may extend the discussion on that topic or introduce related exercises to capitalize on the engagement. These real-time adaptations may help maintain the relevance and effectiveness of the therapy session, ensuring that it remains responsive to the immediate needs and reactions of the participants.
[0225] Reference is now made to Fig. 6B which is a block diagram depicting an example for a workflow of guardrail model 800, which can be implemented in the group psychological therapy management system 10, according to some embodiments of the invention. The workflow may be designed to detect and manage unsafe or inappropriate behaviors from both participants and the system agent 200.
[0226] As shown in Fig. 6B, the system may include an LLM Application component that interfaces with online session analyzer 232.
[0227] In some embodiments, the online session analyzer 232 may receive an intervention prompt 200PR as input and process it through an LLM to generate an output, which may correspond to the next intervention 300INT. Additionally, or alternatively, online session analyzer 232 may receive a user prompt from a participant via input devices (e.g., input devices 7 of Fig.1) and process the user prompt through the LLM to generate an appropriate intervention 300INT based on the user's input. This may enable real-time responsiveness to participant queries or requests during the session.
[0228] In the example of Fig. 6B, guardrail model(s) 800 may be implemented as two separate modules positioned at the input and output stages of the LLM processing pipeline. A first guardrail module 800, which may function as an Input Guard (denoted input guardrail module 800IN), may be positioned before the LLM processing stage. This input-side guardrail module 800 may be configured to analyze incoming prompts and detect potential issues including, for example, whether the input contains personally identifiable information (PII), whether the input is off topic, whether the input pertains to a predefined inappropriate theme (e.g., a jailbreak attempt), and the like. These checks may help ensure that inappropriate or potentially harmful inputs are identified before being processed by the LLM.
[0229] A second guardrail module 800, which may function as an output guard (denoted output guardrail module 8000UT), may be positioned after the LLM processing stage. This output-side guardrail module 800 may be configured to analyze the generated output and detect potential issues including hallucinations, previous content adjustment, and clinical justification. These checks may help ensure that the next intervention 300INT generated by the system is appropriate and safe for presentation to participating recipients.
[0230] According to some embodiments, guardrail model 800 may implement detection mechanisms for participant input. These mechanisms may include real-time and offlineconversation monitoring for keywords such as curses or sexual content, as well as theme analysis for detecting abusive messages, sexual seduction, or suicidal intent. Additionally, guardrail model 800 may perform video analysis to detect inappropriate visual content. The multi-modal data may be chunked into smaller pieces based on content similarity and used as input for detection. In some embodiments, the detection process may employ third-party models trained specifically for this task, as well as a proprietary guardrail layer with proprietary algorithms and similarity search capabilities.
[0231] According to some embodiments, guardrail model 800 may classify harmful events by severity and urgency to respond. This classification may inform the selection of appropriate controlled actions. In some cases, system 10 may employ different model configurations or flavors to accommodate different audiences and sensitivity levels. For example, a first configuration may be appropriate for participants with aggressive communication styles, while a second configuration may be more suitable for vulnerable participants who require heightened sensitivity.
[0232] Guardrail model 800 may further implement anomaly detection by analyzing past conversations and behavioral patterns of participants. In some embodiments, guardrail model 800 may employ context injection via prompt engineering, incorporating user context, psychological profde information, and group dynamics into the LLM processing. This context injection may include vulnerability indicators and specific sensitive topics based on a participant's personal history. For example, the system may be configured to recognize sensitivity around certain topics given a participant's background or experiences. For instance, if a participant's profde indicates that a family member died from cancer, guardrail model 800 may be configured to apply heightened sensitivity when topics related to cancer or terminal illness arise in the session.
[0233] Based on the classification of a harmful event, guardrail model 800 may select a controlled action from a set of predefined actions. These controlled actions may include stopping the session, stopping the conversation, intervening to change the content discussed, escalating to a human operator, or implementing any other text, content, or intervention defined in the system’s set of controlled actions.
[0234] According to some embodiments, the overall flow of the guardrail process may proceed as follows: input from a participant action may be received and analyzed, followed by classification of any detected issues, selection of a controlled action with customization according to the participants’ state, implementation of an agent guardrail process before delivering aresponse, and finally generation of the agent response. This dual-guardrail architecture may provide oversight at both the input and output stages of the intervention generation process, helping to maintain the integrity and safety of the Al-guided therapy system.
[0235] According to some embodiments, the output-side guardrail module 800 may verify that the agent's suggested response is appropriate and does not contain harmful content before delivery to participants. If the response is not approved by the output-side guardrail module 800, an escalation policy may be triggered. Escalation options may include, for example, retry (regenerating the response), mixture of experts (consulting multiple models to generate an alternative response), or human in the loop (escalating to a human operator for review and intervention).
[0236] Psychological therapeutical experience system 10 may include a matching module 600, which may be implemented as a rule-based or machine learning-based model. This module may be trained to identify groups or pairs of individuals based on their psychological profiles, as manifested in the group profile data structure 100, particularly within the user layer 120UL. The matching module 600 may analyze various aspects of users' profiles, such as their psychological states, therapeutic goals, communication styles, and past session experiences, to suggest optimal groupings for future sessions. These suggestions may be based on predicted success metrics, such as potential for positive interactions, complementary therapeutic needs, or likelihood of achieving shared therapeutic objectives. For example, the matching module 600 may recommend pairing individuals with similar challenges but different coping strategies, potentially fostering mutual support and learning. Alternatively, it may suggest grouping users with complementary personality traits to encourage balanced and productive group dynamics. By leveraging the rich data available in the group profde data structure 100, the matching module 600 may enhance the effectiveness of group sessions by creating synergistic participant combinations tailored to maximize therapeutic benefits.
[0237] According to some embodiments, the psychological therapeutical experience system 10 may implement a method for controlling a group session of psychological therapeutical experience by at least one processor (e.g., processor 2 of Fig. 1). Reference is now made to Fig. 7, which is a flow diagram, illustrating an implementation example of such a method, according to some embodiments of the invention.
[0238] As shown in step SI 005, the at least one processor 2 may obtain a group profile data structure (e.g., element 100 of Fig. 3). The group profile data structure may characterize a psychological profile of at least one recipient in a group of recipients engaged in the group session. In some cases, group profile data structure 100 may include user layers 110, relationship layers 120 and knowledge layers such as ontology layers 130, and research layers 140, as described previously.
[0239] Following the obtaining of the group profile data structure, the at least one processor 2 may proceed to a step SI 010 of analyzing the reaction of the at least one recipient in relation to the group session. This analysis may be performed by components such as the session analyzer 700 and the diagnostic module 500 of Fig. 3.
[0240] As explained herein, session analyzer 700 may extract various features 700F from the session, while diagnostic module 500 may employ one or more (e.g., a plurality of) models such as the relationship assessment model (e.g., 510 of Fig. 3), group usage pattern model (e.g., 520 of Fig. 3), recipient psychological model (e.g., 530 of Fig. 3), vulnerability model (e.g., 540 of Fig.3) and abuse model (e.g., 550 of Fig. 3) to interpret these features.
[0241] Based on the analysis performed in step SI 010, the at least one processor 2 may proceed to a step SI 015 of enriching the group profile data structure. This enrichment process may involve updating or adding new nodes (110N, 120N, 130N, 140N of Fig. 3) to the user layers, relationship layers, or other components of the group profile data structure. The enriched group profde data structure 100 may thus reflect the latest insights gained from the session analysis.
[0242] Additionally, or alternatively, the enrichment process of the group profile data structure 100 may involve recalculating weights of the nodes and / or edges within the structure. As new information is gathered from session analysis and user interactions, diagnostic models 500 may dynamically adjust the weights associated with various nodes and connections in the multilayered structure (e.g., graph).
[0243] For instance, the weights of edges connecting diagnostic nodes 110N in the user layer 110 may be updated to reflect recent changes in a user's psychological state. Similarly, the weights of connections between interaction nodes 120N in the relationship layer 120 may be recalculated based on new observations of group dynamics. This weight recalibration may also extend to the knowledge layers, where the relevance of certain psychological concepts or research findings to the current therapeutic context may be reevaluated. By continuously refining these weights, system10 may maintain an up-to-date and nuanced representation of the users' psychological profiles and group dynamics, potentially enabling more accurate and effective interventions.
[0244] For example, edge weights may be recalculated using a temporal decay function, wherein more recent observations are assigned higher weights than older observations. In one example, a weight update may be calculated as w_new = a * w_observed + (1 - a) * w_previous, where a represents a learning rate parameter.
[0245] The at least one processor 2 may then proceed to a step SI 020 of generating an intervention data element (300INT of Figs. 3, 6A) based on the enriched group profde data structure 100. In some cases, the at least one processor 2 may employ a journey management module (e.g., element 200 of Fig. 3) to implement this step.
[0246] As explained herein, an orchestration agent 210 within journey management module 200 may coordinate with a decision model (e.g., element 220 of Fig. 3) to employ a rules module 224 and the efficacy prediction model 226 to determine the most appropriate intervention. Decision module 220 may then collaborate with a session creator language model (e.g., element 300 of Fig. 3, also referred to as a session creation studio) to select or generate an optimal intervention data element (e.g., element 300INT of Fig. 3) based on these inputs.
[0247] Once intervention data element 300INT has been generated, the at least one processor 2 may move to a step S 1025 of selecting or adapting session content 400S based on the intervention data element.
[0248] The at least one processor 2 may then proceed to step SI 030 of presenting the session content to the group of participating recipients. This presentation may be facilitated through a session interface (e.g., element 400 of Fig. 3, such as a mobile application), which may display the session content 400S to the participating recipients.
[0249] The present invention may provide practical application in improving technology in the field of automated psychological therapy. The streamlined integration of the highly elaborate group profde data structure 100 and online management of group sessions in real time may require a unique configuration of highly specialized, complex Al systems. Such integration may not be implemented mentally or through conventional means.
[0250] According to some embodiments, the technical improvements achieved by system 10 may include real-time processing of multi-modal data, including audio, video, and text, through specialized ML modules such as speech-to-text transcriber 710, motion analysis module 720,sentiment analysis module 730, and LLM 740. System 10 may further provide dynamic updating of binary or weighted graph structures within group profile data structure 100, e.g., utilizing temporal decay functions for edge weight recalculation as described herein.
[0251] Additionally, system 10 may implement parallel multi-path analysis with majority voting, as described in relation to Fig. 4B, to arrive at reliable psychological assessments with associated confidence levels. The dual-guardrail architecture of guardrail model 800, positioned at both input and output stages of the LLM processing pipeline, may provide safety verification that could not be performed through manual oversight alone given the real-time nature of the therapeutic sessions.
[0252] System 10 may further employ RAG-based querying of the interconnected, multilayered group profile data structure 100 to extract context subgraphs 150 relevant to current session contexts. The feedback loops between success metrics module 227 and efficacy prediction model 226 may enable continuous refinement of intervention strategies based on measured outcomes, representing a technical improvement over static therapeutic approaches.
[0253] The combination of these technical elements may enable system 10 to provide automated psychological therapeutic support at a scale and responsiveness that would not be achievable through conventional therapeutic methods or mental processes alone.
Claims
CLAIMS1. A method of controlling a group session of psychological therapeutical experience by at least one computing device, the method comprising:obtaining a group profile data structure, characterizing a psychological profile of at least one recipient in a group of at least two recipients engaged in the group session;analyzing reaction of the at least one recipient in relation to the group session; enriching the group profile data structure based on said analysis;generating an intervention data element based on the enriched group profile data structure, wherein said intervention data element defines a required selection or adaptation of session content;selecting, or adapting said session content, based on the intervention data element; and presenting the session content to the group of recipients, as part of the psychological therapeutical experience.
2. The method of claim 1, wherein the group profile data structure is a multi-layered, interconnected, weighted, temporal structure, comprising at least one user layer, representing psychological profiles of one or more recipients of the group of recipients.
3. The method according to any one of claims 1-2, wherein each user layer comprises one or more user identifier nodes, identifying specific recipients, and wherein each user identifier node is connected to one or more characterizing nodes of the user layer, each indicating a specific condition or characteristic of the identified recipient, in a context of a current or historical session.
4. The method according to any one of claims 1-3, wherein the group profde data structure further comprises:a relationship layer, representing relationships between recipients of the group of recipients, wherein the relationship layer comprises one or more interaction nodes, defining interactions among the recipients, in a context of a current, or historic group session; andone or more knowledge layers, associating said psychological profiles and relationships with corresponding knowledge instances in one or more psychological knowledge databases.
5. The method according to any one of claims 1-4, wherein analyzing reaction of a specific recipient comprises:capturing audiovisual representation of the specific recipient during the group session; and employing one or more Machine Learning (ML)-based feature extraction models, to extract one or more audiovisual features, representing behaviour of the specific recipient within or beyond the context of the group session.
6. The method according to any one of claims 1-5, wherein analyzing reaction of a specific recipient further comprises:prompting the specific recipient to provide a textual data element, representing that recipient’s relation to the session;receiving, the textual data element from that specific recipient; andemploying one or more ML-based feature extraction models, to extract one or more textual features, describing behaviour of the specific recipient within, or beyond the context of the group session.
7. The method according to any one of claims 1-6, wherein analyzing reaction of a specific recipient further comprises employing one or more first ML-based language models, to generate diagnosis of the specific recipient’s psychological or behavioural condition, based on the one or more audiovisual, and / or textual features.
8. The method according to any one of claims 1-7, wherein enriching the group profile data structure is selected from a list consisting of:producing a new user node pertaining to the specific recipient based on the generated diagnosis and integrating the new user node into the user layer;recalculating a value of at least one user node based on the generated diagnosis; producing a new edge, connecting previously existing user nodes of the user layer; and recalculating a weight of an edge, connecting between user nodes of the user layer.
9. The method according to any one of claims 6-8 wherein analyzing reaction of at least one recipient further comprises employing one or more second ML-based language models, to identify interaction among the recipients, based on their audiovisual or textual features.
10. The method of claim 9, wherein enriching the group profile data structure is selected from:producing a new interaction node, based on the identified interaction and integrating the new interaction node into the relationship layer;recalculating a value of at least one interaction node based on the generated diagnosis; producing a new edge, connecting a previously existing interaction node of the relationship layer with a user node of the user layer; andrecalculating a weight of an edge, connecting between an interaction node of the relationship layer and a user node of the user layer.
11. The method according to any one of claims 8-10 wherein at least one knowledge layer is an ontology layer, comprising a plurality of ontology nodes, wherein each ontology node represents a psychological term in the one or more psychological knowledge databases.
12. The method of claim 11, wherein each ontology node further represents (i) association of the psychological term to a diagnosis, manifested by at least on user node of the user layer, and / or (ii) association of the psychological term to a relationship type, manifested by a relationship node of the relationship layer.
13. The method according to any one of claims 11-12 wherein at least one knowledge layer is a research layer, comprising a plurality of research nodes, each associating between psychological terms, manifested by ontology nodes of the ontology layer, based on clinical publications.
14. The method according to any one of claims 2-13, wherein generating an intervention data element comprises:based on weights of the group profile data structure, extracting an interconnected substructure of nodes of the enriched group profile data structure, wherein said substructure represents a current context of at least one recipient’s psychological profile;applying a third ML-based language model on the substructure to generate an intervention prompt; andproviding said intervention prompt as input to at least one fourth ML-based language model, wherein the at least one fourth language model is configured to generate the intervention data element based on said intervention prompt.
15. A system for controlling a group session of psychological therapeutical experience, the system comprising: a non-transitory memory device, wherein modules of instruction code are stored, and at least one processor associated with the memory device, and configured to execute the modules of instruction code, whereupon execution of said modules of instruction code, the at least one processor is configured to:obtain a group profile data structure, characterizing a psychological profile of at least one recipient in a group of at least two recipients engaged in the group session;analyze reaction of the at least one recipient in relation to the group session;enrich the group profile data structure based on said analysis;generate an intervention data element based on the enriched group profile data structure, wherein said intervention data element defines a required selection or adaptation of session content;select, or adapting said session content, based on the intervention data element; and present the session content to the group of recipients, as part of the psychological therapeutical experience.
16. The system of claim 15, wherein the group profile data structure is a multi-layered, interconnected, weighted, temporal structure, comprising at least one user layer, representing psychological profiles of one or more recipients of the group of recipients.
17. The system according to any one of claims 15-16, wherein each user layer comprises one or more user identifier nodes, identifying specific recipients, and wherein each user identifier node is connected to one or more characterizing nodes of the user layer, each indicating a specific condition or characteristic of the identified recipient, in a context of a current or historical session.
18. The system according to any one of claims 15-17, wherein the group profile data structure further comprises:a relationship layer, representing relationships between recipients of the group of recipients, wherein the relationship layer comprises one or more interaction nodes, defining interactions among the recipients, in a context of a current, or historic group session; andone or more knowledge layers, associating said psychological profiles and relationships with corresponding knowledge instances in one or more psychological knowledge databases.
19. The system according to any one of claims 15-18, wherein the at least one processor is further configured to analyze reaction of a specific recipient by:capturing audiovisual representation of the specific recipient during the group session; and employing one or more Machine Learning (ML)-based feature extraction models, to extract one or more audiovisual features, representing behaviour of the specific recipient within or beyond the context of the group session.
20. The system according to any one of claims 15-19, wherein the at least one processor is configured to further analyze reaction of a specific recipient by:prompting the specific recipient to provide a textual data element, representing that recipient’s relation to the session;receiving, the textual data element from that specific recipient; andemploying one or more ML-based feature extraction models, to extract one or more textual features, describing behaviour of the specific recipient within, or beyond the context of the group session.
21. The system according to any one of claims 15-20, wherein the at least one processor is configured to further analyze reaction of a specific recipient by employing one or more first ML-based language models, to generate diagnosis of the specific recipient’s psychological or behavioural condition, based on the one or more audiovisual, and / or textual features.
22. The system according to any one of claims 15-21, wherein the at least one processor is configured to enrich the group profile data structure by performing one or more of:producing a new user node pertaining to the specific recipient based on the generated diagnosis and integrating the new user node into the user layer;recalculating a value of at least one user node based on the generated diagnosis; producing a new edge, connecting previously existing user nodes of the user layer; and recalculating a weight of an edge, connecting between user nodes of the user layer.
23. The system according to any one of claims 20-22 wherein the at least one processor is configured to further analyze reaction of at least one recipient by employing one or more second ML-based language models, to identify interaction among the recipients, based on their audiovisual or textual features.
24. The system according to any one of claims 15-23, wherein the at least one processor is configured to further enrich the group profile data structure by performing one or more of :producing a new interaction node, based on the identified interaction and integrating the new interaction node into the relationship layer;recalculating a value of at least one interaction node based on the generated diagnosis; producing a new edge, connecting a previously existing interaction node of the relationship layer with a user node of the user layer; andrecalculating a weight of an edge, connecting between an interaction node of the relationship layer and a user node of the user layer.
25. The system according to any one of claims 22-24 wherein at least one knowledge layer is an ontology layer, comprising a plurality of ontology nodes, wherein each ontology node represents a psychological term in the one or more psychological knowledge databases.
26. The system of claim 25, wherein each ontology node further represents (i) association of the psychological term to a diagnosis, manifested by at least on user node of the user layer, and / or (ii) association of the psychological term to a relationship type, manifested by a relationship node of the relationship layer.
27. The system according to any one of claims 25-26 wherein at least one knowledge layer is a research layer, comprising a plurality of research nodes, each associating between psychological terms, manifested by ontology nodes of the ontology layer, based on clinical publications.
28. The system according to any one of claims 16-27, wherein the at least one processor is configured to generate an intervention data element by:based on weights of the group profile data structure, extracting an interconnected substructure of nodes of the enriched group profile data structure, wherein said substructure represents a current context of at least one recipient’s psychological profile;applying a third ML-based language model on the substructure to generate an intervention prompt; andproviding said intervention prompt as input to at least one fourth ML-based language model, wherein the at least one fourth language model is configured to generate the intervention data element based on said intervention prompt.
29. A method of real-time adaptation of a psychological group therapy session by at least one computing device, the method comprising:presenting, via a session interface implemented on the at least one computing device, session content to a group of recipients participating in the group therapy session;receiving, by an online session analyzer executing on the at least one computing device, multi-modal input data from the session interface during the group therapy session, wherein the multi-modal input data comprises audio data and video data captured via input devices;processing, by the online session analyzer, the audio data using a speech-to-text transcriber to generate textual data representing verbal content from the group therapy session;processing, by the online session analyzer, the video data using a motion analysis module to extract features representing physical movements, gestures, and facial expressions of at least one recipient;extracting, by the online session analyzer in real-time using a sentiment analysis module, insights regarding the at least one recipient's emotional tone and reactions to the session content;providing the extracted insights to an online session moderator executing on the at least one computing device;determining, by the online session moderator based on the extracted insights, that an adaptation to the session content is required; andtransmitting, by the online session moderator to the session interface, adapted session content for presentation to the group of recipients,wherein the adapted session content comprises at least one of clarified content generated in response to detected recipient confusion, extended discussion content generated in response to detected engagement levels exceeding a threshold, and supplementary exercises generated in response to detected topic resonance.
