Psychological consultation data processing method and system and electronic equipment

Abstract

The invention relates to a psychological consultation data processing method and system and electronic equipment. The method comprises the following steps: acquiring multi-modal sensing data of a target user; constructing a multi-layer target for the target user based on the multi-modal sensing data; based on the intelligent agent types contained in the multi-layer target dynamic configuration parallel thinking collaborative network, configuring parallel computing resources for different types of intelligent agents in the parallel thinking collaborative network; the parallel thinking collaborative network is used for simulating the multi-dimensional parallel thinking ability of a consultant; through a parallel thinking collaborative network, a dynamic collaborative decision-making mechanism is adopted to schedule different types of agents to collaboratively process different psychological tasks, psychological consultation interaction with a target user is realized in combination with a parallel psychological reasoning pipeline, and a personalized psychological support scheme in a complex scene is provided for the target user. According to the method, the psychological consultation process can be highly simulated and personalized, the coping ability and adaptability to complex psychological problems are improved, and the quality and effect of psychological consultation services are improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a method, system, and electronic device for processing psychological counseling data. Background Art

[0002] With the development of artificial intelligence technology, it has begun to be applied in the field of psychological counseling. Currently, it usually relies on pre-trained large models to process user consultation requests and provide psychological therapy services through fixed workflows.

[0003] In related technologies, traditional psychological counseling systems are mostly linear conversational systems, which generally have technical problems such as insufficient psychological planning capabilities, poor adaptability to complex systems, and high response delays. In actual services, the psychological counseling processes and tasks in traditional psychological counseling systems still rely on manual design, without innovation in planning capabilities, and functional jumps are also limited. There is also a lack of personalized customization capabilities for different users, which makes it difficult for traditional psychological counseling systems to simulate the actual counseling process and unable to cope with the complex scenarios and dynamic needs in psychological counseling, resulting in limited actual application effects. Therefore, it is urgent to propose a new technical solution to solve at least one technical problem in related technologies. Summary of the Invention

[0004] This application aims to solve the technical problems existing in the prior art and provides a psychological counseling data processing method, system, and electronic device to solve at least one technical problem existing in the prior art.

[0005] In a first aspect, an embodiment of the present application provides a method for processing psychological consultation data, the method comprising:

[0006] Acquire multimodal perception data of the target user; wherein the multimodal perception data includes at least: first session data, psychological scale data, and perception interaction data;

[0007] Constructing multi-layer goals for target users based on multimodal perception data; the multi-layer goals are used to dynamically plan the target users' interaction goals in the psychological counseling process from top to bottom;

[0008] Based on the multi-layered objectives, the types of agents included in the parallel thinking collaboration network are dynamically configured, and parallel computing resources are allocated to different types of agents in the parallel thinking collaboration network; wherein the parallel thinking collaboration network is used to simulate the multi-dimensional parallel thinking ability of the counselor; different types of agents are responsible for data processing tasks under different psychological counseling dimensions; the types of agents in the parallel thinking collaboration network include one or a combination of the following: emotion recognition agent, cognitive assessment agent, intervention planning agent, empathy expression agent, and meta-supervision agent;

[0009] Through a parallel thinking collaborative network, a dynamic collaborative decision-making mechanism is adopted to schedule different types of intelligent agents to collaborate on different psychological tasks. Combined with a parallel psychological reasoning pipeline, psychological consultation interaction with target users is realized, providing target users with personalized psychological support solutions in complex scenarios.

[0010] In a second aspect, an embodiment of the present application provides a psychological counseling data processing system, which includes at least the following units:

[0011] An acquisition unit is configured to acquire multimodal perception data of a target user; wherein the multimodal perception data includes at least: first session data, psychological scale data, and perception interaction data;

[0012] A construction unit is configured to construct a multi-layer goal for a target user based on the multimodal perception data; the multi-layer goal is used to dynamically plan the target user's interaction goals in the psychological consultation process from top to bottom;

[0013] The collaboration unit is configured to dynamically configure the types of agents included in the parallel thinking collaboration network based on the multi-layered goals, and to allocate parallel computing resources for different types of agents in the parallel thinking collaboration network; wherein the parallel thinking collaboration network is used to simulate the multi-dimensional parallel thinking ability of a counselor; different types of agents are responsible for data processing tasks under different psychological counseling dimensions; the types of agents in the parallel thinking collaboration network include one or a combination of the following: an emotion recognition agent, a cognitive assessment agent, an intervention planning agent, an empathy expression agent, and a meta-supervision agent;

[0014] The interaction unit is configured to use a parallel thinking collaborative network and a dynamic collaborative decision-making mechanism to schedule different types of intelligent agents to collaboratively process different psychological tasks. It combines a parallel psychological reasoning pipeline to achieve psychological consultation interaction with the target user, providing the target user with personalized psychological support solutions in complex scenarios.

[0015] In a third aspect, an embodiment of the present application provides an electronic device, comprising:

[0016] Memory for storing computer software programs;

[0017] The processor is used to read and execute the computer software program, thereby implementing the psychological counseling data processing method of the first aspect.

[0018] In a fourth aspect, a computer-readable storage medium is provided, which includes instructions. When the instructions are run on a computer, the computer executes the psychological counseling data processing method of the first aspect.

[0019] The beneficial effects of the present application are: providing a psychological counseling data processing method, system, and electronic device. In this technical solution, first, the multimodal perception data of the target user is obtained; wherein, the multimodal perception data at least includes: first conversation data, psychological scale data, and perception interaction data. Then, based on the multimodal perception data, a multi-layer goal is constructed for the target user; the multi-layer goal is used to dynamically plan the interaction goals of the target user in the psychological counseling process from top to bottom. Then, based on the multi-layer goal, the types of agents included in the parallel thinking collaborative network are dynamically configured, and parallel computing resources are configured for different types of agents in the parallel thinking collaborative network; wherein, the parallel thinking collaborative network is used to simulate the multi-dimensional parallel thinking ability of the counselor; different types of agents are responsible for data processing tasks under different psychological counseling dimensions; the types of agents in the parallel thinking collaborative network include one of the following or a combination: emotion recognition agent, cognitive assessment agent, intervention planning agent, empathy expression agent, meta-supervision agent. Finally, through a parallel thinking collaborative network, a dynamic collaborative decision-making mechanism was employed to schedule different types of agents to collaborate on different psychological tasks. Combined with a parallel psychological reasoning pipeline, this approach enabled psychological counseling interactions with target users, providing them with personalized psychological support solutions in complex scenarios. This technical solution, through multi-layered goals and a parallel thinking collaborative network, achieved a highly simulated and personalized psychological counseling process, improving the ability to respond to and adapt to complex psychological issues, significantly enhancing the quality and effectiveness of psychological counseling services. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flowchart of a method for processing psychological counseling data according to an embodiment of the present application;

[0021] Figure 2 This is a schematic diagram of the structure of a psychological counseling data processing system according to an embodiment of the present application;

[0022] Figure 3 It is a structural diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0024] In order to solve at least one technical problem in the related art, the embodiments of the present application provide a psychological counseling data processing method, system, and electronic device.

[0025] In the technical solution of this application, a multi-layered goal structure constructed from top to bottom plans the psychological counseling framework for users as a whole. Compared with existing technologies, it can plan the counseling process more systematically and comprehensively, effectively solving the problem of insufficient psychological planning capabilities. By dynamically adjusting the goal priority, the needs of different users can be better matched, the matching degree between the model and the user can be significantly improved, and the accuracy of model interaction can be improved. Multi-layer goals are applied to parallel thinking collaborative networks, and multiple agents promote multiple intervention strategies in parallel. Faced with the complex psychological state of users, the system can output or switch intervention plans in real time, quickly adapt to changes, greatly improve the adaptability of the model to complex systems, and provide users with accurate and personalized psychological support plans. Parallel computing resource optimization scheduling and dynamic collaboration of multiple agents reduce information processing time and shorten interactive response delays. The efficient operation of phased parallel processing pipelines and parallel psychological reasoning pipelines greatly improves the processing efficiency of psychological counseling scenarios. At the same time, precise planning and timely solution adjustments ensure processing quality, provide users with better psychological counseling services, and effectively improve user experience.

[0026] The technical solution of the present application and the psychological counseling data processing solution provided in the embodiments of the present application can also be executed by an electronic device, which can be a server, a server cluster, or a cloud server. The electronic device can also be a terminal device such as a mobile phone, a computer, a tablet computer, a wearable device, or a dedicated device (such as a dedicated terminal device with a psychological counseling data processing method system). These electronic devices can also be equipped with the chips or other hardware processing units introduced in the above embodiments. Alternatively, these electronic devices can also be installed with a service program for executing the psychological counseling data processing solution.

[0027] Figure 1 A flowchart of a psychological consultation data processing method provided in an embodiment of the present application is shown as follows: Figure 1 As shown, the method includes the following steps:

[0028] 101, obtaining multimodal perception data of the target user;

[0029] 102, constructing multi-layer goals for target users based on multimodal perception data;

[0030] 103, dynamically configuring the types of agents included in the parallel thinking collaborative network based on the multi-layer goal, and configuring parallel computing resources for different types of agents in the parallel thinking collaborative network;

[0031] 104. Through a parallel thinking collaborative network, a dynamic collaborative decision-making mechanism is adopted to schedule different types of intelligent agents to collaboratively process different psychological tasks. Combined with a parallel psychological reasoning pipeline, psychological consultation interaction with target users is achieved, providing target users with personalized psychological support solutions in complex scenarios.

[0032] In this embodiment of the present application, multimodal perception data includes at least: first-time conversation data, psychological scale data, and perception interaction data. In the context of psychological counseling, multimodal perception data refers to user psychology-related information collected through multiple channels and in various forms, covering multiple dimensions such as language, behavior, and emotions, and is used to fully understand the user's psychological state, needs, and problems.

[0033] First-time conversation data refers to the interactive data generated when a user first comes into contact with a psychological counseling service. It can be collected through interviews, questionnaires, or system registration, reflecting the user's basic background, motivation for seeking help, and initial demands. For example, information such as "severe insomnia in the past two weeks, high work pressure leading to depression" and "previous experience of seeking treatment for anxiety disorders" actively stated by the user during the first consultation; age, occupation, and marital status provided when filling out the consultation registration form; selection of consultation goals (such as "relieving anxiety" and "improving interpersonal relationships"), etc. This information can be used to build user portraits, identify core issues (such as stressors, past medical history), determine the direction of consultation (such as cognitive behavioral therapy, mindfulness intervention), and assess whether referral to a psychiatric department or other special treatment is needed.

[0034] Psychological scale data is quantitative data collected through standardized psychological assessment tools (scales) and is used to objectively evaluate the user's psychological state, symptom severity or personality traits, reducing subjective judgment bias.

[0035] For example, the frequency with which users select questions such as "I often feel lonely" and "I have no hope for the future" when completing the Self-Rating Depression Scale (SDS); the total score and scores of each dimension in the Self-Rating Anxiety Scale (SAS); and the scores of dimensions such as "neuroticism" and "extraversion" in the Big Five Personality Inventory.

[0036] It is understandable that psychological scale data can quantify the severity of depression or anxiety symptoms (such as mild, moderate, severe), assist in determining whether the diagnostic criteria for mental illness are met; analyze personality traits (such as those with high neuroticism are more prone to anxiety), and provide a basis for personalized intervention plans (such as designing emotion regulation exercises for highly sensitive personalities).

[0037] Perceptual interaction data is the user's non-verbal behavior, emotional reactions and interaction details captured in real time during the consultation process. It is collected through technologies such as voice intonation analysis, facial expression recognition, and body movement monitoring, revealing subconscious or unclear psychological states.

[0038] Of particular note are voice intonation data, such as a trembling voice or sudden acceleration of speech when recounting childhood experiences, or a drop in volume or prolonged silence when mentioning a specific person. Facial expression data, such as furrowed brows and averted gaze when discussing family conflicts, or a raised mouth and slightly dilated pupils when recounting successes. Body movements, such as frequent hand rubbing and leg shaking when anxious, or leaning back with arms hanging naturally when relaxed, can also be used to determine the order and placement of specific toys in sandplay therapy.

[0039] Perceptual interaction data can supplement emotional clues beyond verbal expression (such as calm appearance but body language showing tension), identify resistance points (such as physiological reactions when avoiding specific topics), evaluate consultation progress (such as the change from tense posture to relaxed posture reflecting increased sense of security), and provide entry points for non-verbal intervention techniques (such as mindfulness body scan).

[0040] For example, in step 101, first-time session data is collected through the user's initial interaction with the system, such as the basic information the user fills in when registering or first using the system, the preferred tags selected, and the responses to initial introductory questions. This data reflects the user's basic attributes and initial needs. Psychological scale data can be obtained using pre-set psychological scale tools. For example, after completing a specific task or browsing content, the user is guided to complete a brief psychological state assessment, such as a self-emotional self-rating scale or a stress index scale. Each question in the scale corresponds to a different dimension of psychological characteristics, and the user's choices or scores constitute quantitative psychological state data for analyzing their inherent psychological needs. Perceptual interaction data can be generated by capturing the user's real-time interaction behavior with the system interface, such as the user's dwell time, click location, and scroll speed when browsing product pages; response delays and keyword frequency in conversational interactions; touch trajectory and voice intonation changes when using smart devices. This data reflects the user's attention distribution, interests, and emotional fluctuations at a behavioral level. Furthermore, optionally, physical response data during the interaction process can also be collected, such as heart rate, limb stiffness, and frequency of posture changes.

[0041] Further optionally, the multi-layer goals are used to dynamically plan the target user's interaction goals in the psychological consultation process from top to bottom.

[0042] In psychological counseling scenarios, multi-layered goals, through top-down structured design, break down users' complex psychological improvement needs into quantifiable, actionable, and phased interactive objectives, building a dynamic planning system from macro-visions to specific behaviors. Conversational macro goals, as macro-visions of psychological improvement, represent the fundamental changes users desire to achieve. These goals are often abstract and require long-term, multi-stage implementation. For example, User A may wish to alleviate intimacy difficulties caused by childhood trauma, while User B may want to establish a stable sense of self-worth and break free from habitual self-denial. These conversational macro goals anchor the entire counseling process, ensuring that subsequent interventions revolve around these core needs.

[0043] As phased functional goals, these goals break down the macro-goal of a conversation into 3-5 tasks that can be completed within 8-12 weeks. Each goal focuses on improving specific psychological functions and has clear measurement indicators. Taking the alleviation of intimacy disorders as an example, phased goals can be set at the cognitive, emotional, and behavioral levels. For example, identifying the connection between childhood trauma and current relationship patterns, practicing expressing authentic emotions in safe situations, and exploring the establishment of new, low-risk social connections. These phased goals transform abstract visions into actionable medium-term goals, leveraging methods such as cognitive behavioral therapy to help users accumulate specific skills.

[0044] Micro-interaction goals are specific interactive tasks implemented in each consultation session. They provide micro-level, immediate behavioral guidance that directly supports the phased goals. For the phased goal of "identifying trauma connections," a micro-interaction goal for a particular consultation might be guiding the user to describe a "timeline" of key childhood events and corresponding cases of relationship conflict in adulthood, or using mindful breathing exercises to help the user maintain emotional stability while recalling trauma. Through open-ended questions and visual tools, macro-goals can be broken down into actionable conversational processes.

[0045] The dynamic planning implementation logic begins with a top-down goal breakdown. First-session data and psychological scale data are used to confirm the macro-goal for the session. Phased goals are designed based on psychological counseling theory. Micro-interaction goals are then generated based on pre-session perceptual interaction data. Furthermore, a bottom-up feedback correction mechanism is implemented. If users experience strong resistance during micro-interaction goal execution, such as a sudden panic attack during a simulated social scenario exercise, manifested by rapid breathing and pale complexion, the counselor will pause the task, return to the phased goal, and adjust the intervention intensity. If the phased goal is achieved ahead of schedule, the next phase is accelerated or supplementary goals are added.

[0046] This multi-layered goal system has produced significant technical benefits in psychological counseling. It enhances goal visualization and user engagement. By creating a "goal ladder" diagram and other methods, users can clearly see the path from their current problem to their ideal state, reducing ambiguity and enhancing treatment motivation. It also enables refined intervention and efficiency optimization. The dynamic design of micro-interaction goals tailors progressive tasks to different user characteristics, such as designing progressive emotional expression tasks for highly sensitive personalities, improving intervention efficiency. It also provides risk prevention and control and personalized adaptation capabilities, adjusting the priority of phased goals based on user status changes. For example, when a user experiences increased stress due to a job change, they can temporarily shift from building self-worth to coping with acute stress, returning to their original plan once the crisis subsides.

[0047] For example, a 30-year-old female user who developed depression due to workplace bullying had a conversation macro goal of improving depression, with phased goals set sequentially for trauma-related cognitive restructuring, emotional regulation skills development, and workplace boundary reestablishment. In the trauma-related cognitive restructuring phase (weeks 1-4), micro-interaction goals included using a table to identify the connection between the bullying incident and core beliefs and challenging negative automatic thoughts. In the emotional regulation skills development phase (weeks 5-8), micro-interaction goals included practicing body scanning techniques and distinguishing subtle emotional nuances. In the workplace boundary reestablishment phase (weeks 9-12), micro-interaction goals included role-playing to simulate refusing unreasonable work requests and developing a checklist of safe workplace behaviors. Through this layered design, users gradually transitioned from cognitive modification to behavioral reengineering, achieving measurable achievement of the conversation macro goal, such as a significant reduction in depression scale scores, returning to normal levels. This dynamic multi-layered goal programming transforms psychological intervention into a "manageable growth unit," balancing structure and flexibility. This makes the abstract psychological improvement process concrete and actionable, providing counselors with a clear intervention framework and ensuring the scientific and effective nature of the counseling process.

[0048] As an optional embodiment, after step 101, the psychological state information contained in the multimodal perception data can be projected into the psychological task state space to obtain a psychological task state vector. The psychological task state vector includes at least the following dimensions: emotion vector features, cognitive relationship graph features, behavioral sequence features, and narrative structure features. Furthermore, based on the collaborative protocol of the psychological task state space, the psychological task state vector is shared among different types of intelligent agents according to the processing requirements of different data processing tasks.

[0049] In the digital scenario of psychological counseling, projecting the psychological state information in multimodal perception data into the psychological task state space and then generating a psychological task state vector is a key link in achieving precise intervention and intelligent collaboration. Specifically, first, the user's psychological state information is extracted from the first conversation data, psychological scale data, and perception interaction data, and this information is converted into quantifiable and structured features through natural language processing, voice sentiment analysis, behavioral pattern recognition and other technologies. For example, when a user describes recent work pressure, the system captures multimodal signals such as accelerated speech speed and frequent use of negative vocabulary (such as "collapse" and "helpless"), and combines them with the anxiety score data in the scale to integrate them into the original psychological state information. Then, these original information are mapped to different dimensions of the psychological task state space to form a psychological task state vector.

[0050] The psychological task state vector contains multiple key dimensions. The emotion vector feature focuses on the user's emotional level. By analyzing voice intonation, facial expressions, and text emotional vocabulary, it determines the user's current emotion type (such as anger, depression, anxiety) and intensity. For example, it identifies that the user is in a moderately anxious state when talking about family conflicts. The cognitive relationship graph feature is used to deconstruct the user's thinking mode, extract core beliefs, automatic thoughts, and the logical connections between them from the dialogue text, and construct a cognitive network such as "I think I am worthless because of a mistake at work." The behavior sequence feature captures the user's behavioral trajectory during the consultation process, including the frequency of speaking, topic switching patterns, and response speed to specific questions. For example, it records the user's long silence when asked about his childhood experiences. The narrative structure feature analyzes the organization of the user's narrative content, such as whether there are logical jumps, focus shifts, or whether a certain type of event is repeatedly mentioned, such as the user repeatedly emphasizing the experience of betrayal in the past.

[0051] The collaborative protocol based on the psychological task state space can achieve accurate information sharing according to the processing needs of different agents. For example, the emotion recognition agent mainly receives emotion vector features and quickly determines the user's emotional state. If it detects that the user is emotionally excited, it immediately transmits this information to the empathy expression agent, prompting it to adjust its response strategy and use more soothing language. The cognitive assessment agent obtains the characteristics of the cognitive relationship graph and deeply analyzes the user's cognitive bias. If it finds that the user has a "black or white" thinking mode, the result will be transmitted to the intervention planning agent to formulate targeted cognitive reconstruction tasks. The behavior sequence characteristics and narrative structure characteristics assist each agent in understanding the psychological motivations behind the user's behavior and provide a supplementary basis for the overall intervention plan.

[0052] For example, a user seeking counseling for social anxiety exhibited stuttering and evasive eye contact during their first consultation (perceptual interaction data). They also reported being "afraid of interacting with others and feeling they would be laughed at" (first conversation data), and their anxiety scale score indicated high anxiety (psychological scale data). The system projected this information into the psychological task state space. The resulting psychological task state vector features emotion vectors characterized by "high anxiety and low confidence." Cognitive relationship graphs revealed a negative cognitive chain of "interacting with people → being evaluated → being denied." Behavioral sequence features documented brevity and repeated avoidance of key questions. Narrative structure revealed a disorganized narrative logic, with repeated emphasis on past social failures. Subsequently, the emotion recognition agent prioritized the emotion vector features to determine the user's emotional crisis level. The empathy expression agent, based on this information, responded empathetically, "I feel your nervousness right now. This must be very difficult." The cognitive assessment agent, combined with cognitive relationship graph features, identified a cognitive bias toward overgeneralization. The intervention planning agent then integrated all these features to design a phased task, beginning with simple exercises in imagining social scenarios and gradually guiding the user through real-life social situations.

[0053] This approach facilitates the deep integration and structured expression of multimodal data, transforming fragmented user information into a systematic psychological profile, avoiding information redundancy and conflict. Through the precise mapping of psychological task state vectors, standardized input is provided to intelligent agents, improving data processing efficiency and accuracy. An information sharing mechanism based on collaborative protocols breaks down data barriers between agents, promoting collaborative work among them and making intervention plans more tailored to individual user needs. For example, intervention intensity and pacing can be dynamically adjusted based on the user's cognitive bias type and emotional state, ultimately enhancing the effectiveness of psychological counseling and user experience.

[0054] It is worth noting that in the above scheme, the cognitive relationship graph feature can be constructed by extracting cognitive elements (such as events, beliefs, emotions, behaviors, etc.) from the user's expressions and their interrelationships through natural language processing (NLP) and graph neural networks (GNN) technologies based on the multimodal data generated during the consultation process (such as conversation text, voice intonation, and body movements), thereby constructing a dynamic cognitive relationship graph. Specifically, the user input text is first preprocessed through word segmentation and named entity recognition to identify key cognitive nodes (such as "childhood criticism," "self-denial," and "social avoidance"). Then, through dependency parsing and sentiment analysis, causal relationships (such as "childhood criticism → formation of self-denial beliefs"), associative relationships (such as "self-denial → social avoidance behaviors"), or hierarchical relationships (such as the subordination of core beliefs and automatic thoughts") between nodes are discovered, and the nodes are connected by edges. Finally, the graph structure is converted into a low-dimensional dense vector feature to represent the user's cognitive patterns and thought network.

[0055] For example, if a user mentions during a consultation that "when the leader frowns slightly, I feel like I haven't done my job well and I worry about being fired," nodes such as "leader frowns," "work not done well," and "worry about being fired" can be extracted, and the causal chain of "leader's expression → triggering self-denial cognition → triggering anxiety" can be analyzed. A subgraph containing cognitive-emotional-behavioral associations is constructed, and the corresponding cognitive relationship graph feature vector is generated through graph embedding technology (such as GraphSAGE). The technical effect of this feature is to structure the user's fragmented and nonlinear cognitive content, intuitively presenting the logical loopholes (such as overgeneralization and catastrophic thinking) or cognitive biases (such as all-or-nothing thinking) in their thinking patterns, helping counselors or intelligent agents quickly locate cognitive intervention targets. At the same time, the dynamically updated cognitive relationship graph can track the changes in the user's cognitive structure. For example, by comparing the weakening of the connection strength of "negative cognitive nodes" and the addition of "adaptive cognitive nodes" in the graph before and after consultation, the effect of cognitive intervention can be quantitatively evaluated, providing data support for the adjustment of personalized consultation plans.

[0056] In addition, in the embodiment of the present application, the technical implementation logic of the behavior sequence feature is to build a time-series behavior sequence model by capturing the user's explicit behavior data (such as language expression frequency, body movement amplitude, conversation response duration, etc.) during the psychological consultation process. Specifically, the system first uses multimodal sensors (such as voice recognition modules, video cameras) to collect real-time behavior data of users in consultation, such as changes in speech speed, gesture frequency or silence duration when users talk about specific events. After pre-processing, these data are used to extract dynamic features through a time series modeling algorithm to identify patterns in behavior sequences (such as repetitive small action sequences when anxious, response delay patterns when avoiding topics).

[0057] Taking the "deep breath-silence-averted gaze" behavior sequence, a common occurrence when discussing workplace conflicts, as an example, a time series modeling algorithm learns the temporal dependencies of this sequence and generates a corresponding behavioral sequence feature vector, characterizing the user's behavioral response patterns when faced with stressful situations. By dynamically modeling the temporal continuity of behavioral sequences, the system accurately identifies users' subconscious psychological defense mechanisms or emotional triggers. It also transforms fragmented behaviors into structured features, providing data support for intelligent agents to analyze user behavioral motivations. Combined with historical behavioral sequence features, it predicts user behavioral response trends under specific interventions, assisting counselors in optimizing real-time interaction strategies.

[0058] Among them, the core dimensions of behavioral sequence features include: the timestamp order of behavioral elements, the transition probability of behavior types (such as the frequency of behavioral jumps from "expressing anger" to "self-denial"), and the fluctuation curve of behavior intensity (such as the fluctuation pattern of voice volume with changes in emotions). These features together constitute a dynamic psychological mapping at the user behavior level, providing specific behavioral intervention targets for the design of micro-interaction goals in the multi-layer goal system.

[0059] In 102, a multi-layer target is constructed for the target user based on the multimodal perception data.

[0060] As an optional embodiment, in 102, the decision coordinator is used to parse the first conversation data and the psychological scale data to generate a three-layer nested goal structure; the three-layer nested goal structure in the multi-layer goal includes at least: conversation macro goals, phase goals, and micro-interaction goals; the conversation macro goals are obtained for long-term psychological counseling planning, the phase goals are obtained for medium-term psychological counseling planning, and the micro-interaction goals are obtained by dynamic configuration for each conversation process; and each goal in the multi-layer goal is connected based on a directed acyclic graph.

[0061] In the digital scenario of psychological counseling, building multi-layer goals for target users based on multimodal perception data through decision coordinators is a core step in achieving systematic and personalized intervention.

[0062] In 102, the decision coordinator first conducts an in-depth analysis of the initial session data and psychological scale data. The initial session data includes the user's self-reported core concerns and background information, while the psychological scale data quantifies the severity of the user's psychological symptoms and personality traits. The decision coordinator extracts key information from this data. For example, if the user mentions "long-term anxiety, which affects work and social life" (initial session data), combined with the high score on the anxiety scale (psychological scale data), the decision coordinator generates a session macro goal, the core direction of the long-term psychological counseling plan, such as "alleviating anxiety symptoms and restoring normal social functioning." This abstract and broad goal sets the tone for the entire counseling process.

[0063] Next, the decision coordinator, focusing on the macro-goal of the session and combining psychological counseling theory with past experience from similar cases, breaks down phased goals into specific tasks for the mid-term counseling plan. Taking anxiety relief as an example, these phased goals might be divided into three stages: the first stage involves identifying anxiety triggers and internal cognitive patterns; the second stage involves learning and practicing emotion regulation techniques; and the third stage involves consolidating coping strategies in real-world scenarios. Each phased goal has clear expected outcomes and timelines, such as completing the first phase within 8-12 weeks, making the goals actionable and measurable.

[0064] Finally, for each consultation session, the decision coordinator dynamically configures micro-interaction goals based on the perceptual interaction data (such as the user's emotional state, verbal expression details, and behavioral responses). For example, during a consultation, if a user becomes emotional and speaks rapidly while discussing a work report (perceptual interaction data), the micro-interaction goal is to guide the user to calm their emotions through mindful breathing exercises and record this anxiety experience using an "event-thought-feeling" form to provide material for subsequent cognitive analysis. These micro-interaction goals are closely aligned with the context of the conversation and directly serve to advance the phased goals.

[0065] To organize and connect goals, a directed acyclic graph (DAG) connects the session macro goal, phased goals, and micro-interaction goals, creating a clear logical thread. The DAG uses the session macro goal as the root node, with each phased goal branching out from it. Each phased goal, in turn, extends to multiple micro-interaction goals as child nodes. Arrows indicate the dependencies and advancements between goals. For example, the completion of micro-interaction goals drives the progress of phased goals, ultimately contributing to the achievement of the session macro goal.

[0066] For example, a user experiencing depression due to interpersonal conflict in the workplace described feeling "isolated at work and worthless" during the first session (first session data). A depression scale indicated moderate depression (psychological scale data). Based on this, the decision coordinator determined the session's macro goal to be "improving depression and rebuilding positive self-perception and workplace social skills." Mid-term phased goals were broken down into the following: Phase 1: identifying negative core beliefs and interpersonal conflict patterns; Phase 2: learning confident expression and emotional communication skills; and Phase 3: practicing new social strategies in the workplace. In a specific consultation session, the user described experiencing a trembling voice and averted gaze during an argument with a colleague (perceptual interaction data). The micro-interaction goal for this session was to simulate conflict scenarios through role-playing, helping the user practice expressing their opinions firmly and using a mood scoring tool to record emotional changes during the process.

[0067] From the user's perspective, this multi-layered goal construction method makes the abstract psychological improvement process concrete through clear goal stratification. Users can intuitively see the interim results and ultimate direction of the consultation, which enhances their confidence and sense of participation in treatment. From the perspective of the consultation process, dynamically configured micro-interaction goals ensure that each session is targeted, and strategies are adjusted based on real-time data to improve the accuracy and efficiency of intervention. From a system level, the connection method of directed acyclic graphs systematizes the goal management system, making it easier for counselors or intelligent systems to track progress and evaluate results. Once a deviation occurs in a certain link (such as user resistance to a certain micro-interaction goal), the problem can be quickly located and subsequent plans adjusted to achieve personalized and scientific psychological counseling services.

[0068] Further optionally, in 102, each time after receiving the conversation interaction information fed back by the target user, the treatment value function of each intervention path is calculated through real-time path re-evaluation, and the target priority of each target in the multi-layer target is dynamically adjusted based on the recalculated treatment value function.

[0069] It is understandable that real-time path reassessment is a key link in ensuring that intervention strategies meet the dynamic needs of users.

[0070] Specifically, in step 102, each time a target user receives feedback on a conversational interaction, such as conversation content, emotional reactions, and behavioral performance, the calculation of the therapeutic value function is immediately initiated. This function comprehensively considers multiple factors to assess the potential value of different intervention paths for improving the user's psychological well-being. For example, if a user mentions a new traumatic event during a consultation, or expresses strong resistance or positive response to a certain type of intervention, this information will be used as a basis for calculation. Simultaneously, each potential intervention path is analyzed, from deeply exploring the cognitive roots behind the new event to adjusting the intensity of current emotion regulation training. The therapeutic value of each intervention path is quantified by its contribution to achieving the macro-goal, stage-by-stage goals, and micro-interaction goals of the conversation.

[0071] For example, a user who received counseling for social phobia was initially assigned a micro-interaction goal of practicing simple simulated social scenarios. During a consultation, the user experienced a severe panic reaction when attempting to practice, and expressed strong resistance to this approach (conversational interaction information). At this point, the system, through real-time path reassessment, calculated that the therapeutic value of continuing with the original simulation exercise decreased, while the intervention path of exploring the user's early social traumatic experiences and first performing cognitive reconstruction had higher therapeutic value. Based on the recalculated therapeutic value function, the system dynamically adjusts the priority of multiple layers of goals: temporarily lowering the priority of the current micro-interaction goal (simulated practice), setting the exploration of traumatic experiences as a new micro-interaction goal with a higher priority, and adjusting the pace of subsequent phased goals accordingly to ensure that the overall intervention strategy is adapted to the user's current state.

[0072] This real-time dynamic adjustment mechanism first gives psychological counseling greater adaptability, enabling it to quickly respond to and handle sudden emotional changes, exposure of new problems, and other situations that occur during the consultation process, thus avoiding a "one-size-fits-all" fixed intervention process that may lead to user resistance or treatment failure. Secondly, by quantitatively evaluating the therapeutic value of intervention pathways, it ensures that resources and energy are concentrated in the most effective intervention direction for users, thereby improving consultation efficiency and shortening the cycle to achieve the expected treatment effect. In addition, the dynamic adjustment of target priorities reflects the "user-centered" service concept, allowing users to feel the flexibility and pertinence of the consultation strategy, enhancing their trust in and cooperation with the consultation process, and ultimately improving the overall effectiveness of psychological counseling and user satisfaction.

[0073] In 103, the types of agents included in the parallel thinking collaborative network are dynamically configured based on the multi-layer goals, and parallel computing resources are configured for different types of agents in the parallel thinking collaborative network.

[0074] In the embodiment of the present application, the parallel thinking collaborative network is used to simulate the multi-dimensional parallel thinking ability of the counselor. Further optionally, different types of agents are responsible for data processing tasks under different psychological counseling dimensions.

[0075] In psychological counseling scenarios, the parallel thinking collaborative network aims to simulate the multi-dimensional parallel thinking capabilities of counselors, enabling efficient and precise psychological intervention through the division of labor and collaboration among different types of agents. The emotion recognition agent captures the user's emotional state in real time, using non-verbal signals such as changes in voice intonation, emotional tendencies in text, and micro-expressions in video footage to determine the user's current emotion type and intensity. It continuously tracks emotional dynamics, constructs a change curve, and identifies emotional triggers, helping counselors or the system quickly identify users' emotional distress. For example, it can accurately identify hidden shame in users describing workplace conflicts, preventing intervention from straying from the core emotional core.

[0076] Exemplarily, the types of agents in the parallel thinking collaborative network include one or a combination of the following: emotion recognition agent, cognitive assessment agent, intervention planning agent, empathy expression agent, and meta-supervision agent.

[0077] The cognitive assessment agent focuses on deconstructing users' thinking patterns and cognitive biases. By analyzing conversation texts, it extracts core beliefs, intermediate beliefs, and automatic thoughts, and constructs a complete cognitive hierarchy map. At the same time, based on cognitive psychology theory, it identifies common types of cognitive distortions and quantifies the degree of bias, providing a theoretical basis for subsequent cognitive behavioral interventions. For example, it designs probability estimation tasks for users' "catastrophic" thinking and gradually corrects irrational cognition.

[0078] The intervention planning agent dynamically generates personalized intervention plans based on the analysis results of the first two. It matches intervention techniques to the user's problem type, breaking down macro-goals into multi-level, actionable tasks. It also features a dynamic adjustment mechanism that optimizes intervention strategies and task difficulty based on real-time user feedback. For example, it designs a gradual exposure path for highly sensitive users to avoid user resistance caused by direct exposure to high-stress scenarios.

[0079] The empathy expression agent simulates the empathy ability of human counselors, generates ethical empathy feedback based on the content and emotions expressed by users, and adjusts the response method according to the emotion recognition results. It conveys understanding and support through language and non-verbal signals, reduces the user's defensive psychology, and enhances their sense of being understood. Studies have shown that high-quality empathy can reduce the consultation dropout rate by 30% and improve users' cooperation with intervention measures.

[0080] The meta-supervisory agent acts as the "guardian" of the entire network, providing global system monitoring and optimization, coordinating data flows between agents to avoid information silos, reviewing intervention plans based on ethical guidelines for psychological counseling, alerting to potential ethical risks, and optimizing system performance by analyzing historical data. For example, if the emotion recognition agent detects that a user's anxiety level exceeds the limit, the meta-supervisory agent can suspend the exposure task and initiate crisis intervention. If a user mentions suicidal thoughts, the crisis intervention plan is immediately triggered and external resources are mobilized.

[0081] Taking the example of a user who suffers from insomnia and self-doubt due to criticism from his leader, the emotion recognition agent detects words such as "despair" and "fear" and voice changes, and determines that the user is in a state of moderate anxiety and low self-esteem; the cognitive assessment agent identifies the core belief "I must be perfect to be recognized" and the "catastrophizing" and "over-generalization" biases in automatic thinking; the intervention planning agent uses CBT technology to design an underlying task to refute the conclusion of "poor ability" by citing successful cases; the empathy expression agent responds by expressing its understanding of the user's panic and guiding him to describe the incident in detail; the meta-supervision agent monitors the intervention process in real time. If the user's anxiety continues to rise, it will prompt the user to adjust the intervention rhythm and add relaxation training sessions.

[0082] The parallel thinking collaborative network provided by this application achieves multi-dimensional coverage from emotions, cognition, behavior to relationship building and system management, avoiding the limitations of a single perspective; the linkage between the meta-supervision agent and the intervention planning agent gives the system strong dynamic adaptability, which can take into account both the structured framework and the personalized needs of users; automated data processing and solution generation greatly improve efficiency, shortening the pre-consultation preparation time from 30 minutes to 5 minutes, while reducing human errors; strict ethical monitoring and risk warning mechanisms effectively ensure the safety of the consultation process, so that it complies with industry standards and requirements, and ultimately transforms complex psychological interventions into computable and collaborative intelligent processes, providing solid technical support for the large-scale application of digital psychological counseling.

[0083] As an optional embodiment, in step 103, dynamically configuring the agent types included in the parallel thinking collaborative network based on the multi-layered goals includes: first, dynamically activating multiple agents matching the target user based on the goal level and task complexity via a central scheduler to form the parallel thinking collaborative network. Then, dynamically allocating and adjusting the functional specialization, processing priority, resource allocation rate, and task response threshold of the multiple agents using a parameterized configuration equation for agent capabilities, thereby completing the intelligent configuration of the multiple agents.

[0084] Exemplarily, the multiple agents include at least: an emotion recognition agent used to adopt an attention-enhanced dual-channel network to support multi-dimensional emotion space recognition; a cognitive assessment agent used to build a core belief network to identify thinking patterns and cognitive distortions; an intervention planning agent used to generate and evaluate multiple intervention paths based on the multi-layer goals, multiple intervention paths are used to simulate user reactions, and multiple intervention paths support parallel advancement; an empathy expression agent is used to optimize the professionalism and gentleness of language expression; and a meta-supervision agent is used to monitor ethical boundaries and professional standards.

[0085] In the digital process of psychological counseling, step 103 realizes the dynamic construction and optimization of the parallel thinking collaborative network through the parameterized configuration of the central scheduler and agent capabilities. Specifically, the central scheduler first analyzes the target user's multi-layer goals (including conversation macro goals, stage goals, and micro-interaction goals), and selects the adapted agents based on the goal level and task complexity. For example, when the user's conversation macro goal is to alleviate severe social anxiety, the central scheduler determines that it is necessary to prioritize the activation of the emotion recognition agent to capture the user's anxiety reaction, the cognitive assessment agent to analyze the cognitive bias behind their social phobia, and at the same time call the intervention planning agent to design a gradually exposed intervention plan, thereby forming the initial parallel thinking collaborative network.

[0086] After the network is established, it undergoes in-depth optimization through parameterized configuration equations based on the agent's capabilities. For example, the Emotion Recognition Agent can adjust the sensitivity parameters of its multidimensional emotion space recognition based on the complexity of the user's emotional fluctuations. If the user's cognitive biases are stubborn, the core belief network analysis accuracy parameters of the Cognitive Assessment Agent will be enhanced. For the Intervention Planning Agent, if rapid verification of multiple intervention paths is required, the resource allocation rate for its parallel tasks will be increased accordingly. The Empathy Expression Agent dynamically adjusts the gentleness weight of its language based on the user's defensiveness. A meta-supervision agent monitors the behavior of each agent throughout the process. When the exposure exercises designed by the Intervention Planning Agent threaten to cross ethical boundaries, their processing priority is immediately lowered and corrections are triggered.

[0087] For example, a user with intimacy difficulties due to childhood trauma will be activated by the central dispatcher to detect their avoidance when discussing emotional topics. The cognitive assessment agent will construct a core belief network regarding the self's unworthiness of love. The intervention planning agent will generate multiple parallel intervention pathways based on multi-layered objectives, including "trauma narrative-cognitive reconstruction-relationship simulation." The empathy expression agent will gently guide the user's expression, while the meta-supervision agent will ensure that the entire process adheres to the ethical principles of psychological counseling. During a consultation, if the emotion recognition agent detects strong emotional fluctuations in the user, a parameterized configuration equation will slow down the intervention planning agent's progress while increasing the empathy expression agent's response frequency. The meta-supervision agent will simultaneously review subsequent intervention pathways for risks.

[0088] This dynamic configuration mechanism provides a clear division of labor among agents at a functional level. The emotion recognition agent accurately identifies complex user emotions through an attention-enhanced dual-channel network. The cognitive assessment agent locates thinking vulnerabilities using a core belief network. The intervention planning agent simulates multiple intervention paths in parallel to efficiently select the optimal solution. The empathy expression agent balances professionalism and affinity to establish trust. The meta-supervision agent strengthens ethical defenses. Regarding collaboration, parameterized configuration enables seamless coordination among agents, while the central dispatcher acts as a central command center, adjusting the priorities and resources of each agent in real time based on user status to avoid information silos and duplication of effort. Ultimately, the system is able to quickly respond to individual differences and provide personalized psychological counseling services tailored to user needs, improving intervention efficiency while ensuring the safety and professionalism of the consultation process.

[0089] Furthermore, in practical applications, ensuring the accuracy of the parameterized configuration equations for agent capabilities requires collaborative optimization across multiple dimensions. First, a dynamic calibration mechanism should be established. A closed-loop verification system should be constructed using feedback from historical conversational data on intervention effects (such as user mood improvement indicators and cognitive bias correction rates). Regular correlation analysis should be conducted between the parameter configuration results output by the equations and actual treatment outcomes. For example, the frequency of positive feedback from users in conversations under certain parameter combinations should be compared to adjust the weight coefficients and threshold ranges in the equations. Secondly, an expert knowledge injection mechanism should be introduced, with experts in the field of psychological counseling conducting empirical verification of the parameter configuration logic. For example, in the setting of the language gentleness parameter for the empathy expression agent, a reasonable range of values ​​should be defined in accordance with clinical communication guidelines to avoid deviations from professional ethical standards. Furthermore, a small-sample pilot testing strategy should be implemented, piloting specific parameter configurations with a local user population. A / B testing should be conducted to compare the intervention effects of different parameter combinations. For example, the accuracy of the cognitive assessment agent in identifying users' core beliefs under different processing priorities should be tested, and the optimal parameter combination should be selected based on statistically significant results. Furthermore, an anomaly detection module is designed to monitor parameter configuration results in real time to ensure they trigger pre-defined safety boundaries (e.g., whether the resource allocation rate of the intervention planning agent causes a particular intervention path to overconsume computing resources). Upon detecting a parameter anomaly, an automatic rollback or manual intervention adjustment is initiated. The functional roles of various agents create a synergistic effect within the parameterized configuration: the emotion recognition agent uses the multidimensional emotion data (e.g., anxiety and pleasure) output by the attention enhancement network to provide basic input to the cognitive assessment agent. The cognitive assessment agent adjusts the path generation parameters of the intervention planning agent based on the core belief network analysis results (e.g., prioritizing cognitive reconstruction intervention paths). The empathy expression agent optimizes the gentleness and expertise weights of the language generation model based on the parameter outputs of the first two agents. The meta-supervision agent continuously monitors the compliance of each agent's parameters with ethical standards, for example, ensuring that the cognitive assessment agent's thought pattern recognition does not violate the user's privacy boundaries. Through this mechanism, the parameterized configuration equation can continuously optimize accuracy under the multiple constraints of data-driven, expert experience, empirical testing, and ethical monitoring, enabling the multi-agent collaborative network to dynamically adapt to individual user needs while maintaining the professionalism and safety of the intervention strategy.

[0090] As an optional embodiment, in step 103, parallel computing resources are configured for different types of agents in the parallel thinking collaborative network, including: constructing a response gradient architecture for the target user based on the multi-layered goals; wherein the response gradient architecture is used to set the interactive response time with the target user at different conversation stages; and wherein the interactive response processing layers at different conversation stages include at least: an immediate emotional response layer, a standard cognitive layer, and a deep reflection generation layer. The current conversation stage of the target user is determined, and computing resources are allocated to the target user using the matching interactive response processing layer in the response gradient architecture. The computing resources allocated to the target user are weighted using a differential resource allocation function, taking into account the psychological urgency, cognitive complexity, and interactive needs of the data processing tasks of the different types of agents, to obtain the computing resources corresponding to each type of agent. The weight configuration parameters in the differential resource allocation function vary with the conversation stage to adapt to the specific computing resource requirements of different stages.

[0091] In the psychological counseling scenario, step 103 revolves around the precise configuration of computing resources for intelligent agents in the parallel thinking collaborative network. In 103, a response gradient architecture is first constructed based on the user's multi-layer goals (session macro goals, stage goals, micro-interaction goals). This architecture aims to set interactive response time standards that match the user at different stages of the consultation process. The response gradient architecture includes three interactive response processing layers: immediate emotional response layer, standard cognitive layer, and deep reflection generation layer. Each layer corresponds to different response requirements and computing resource priorities. For example, the response time corresponding to the immediate emotional response layer is 15-100ms for the first packet response, the response time corresponding to the standard cognitive processing layer (SCP) is 200-500ms for the first packet response, and the response time corresponding to the deep reflection generation layer is 1-3 seconds.

[0092] Determining the user's current conversational stage is a key prerequisite for resource allocation. In 103, by analyzing multimodal perception data from real-time user feedback, such as the amplitude of emotional fluctuations, the depth of language expression, and the degree of exposure to cognitive biases, the user's current conversational stage is determined. For example, when a user suddenly becomes emotionally agitated by mentioning a sensitive topic, with symptoms such as rapid speech and trembling voice, the system determines that the user is in a stage of intense emotional fluctuations. If the user begins to rationally analyze their own problems and systematically elaborate their thinking patterns, they enter the cognitive exploration stage. When the user begins to review past experiences and deeply analyze the motivations behind their behavior, they are in a deep reflection stage.

[0093] During resource allocation, basic computing resources are first allocated using the interactive response processing layer, which matches the user's current session phase. The immediate emotional response layer focuses on quickly capturing and responding to emotional signals. Therefore, computing resources are prioritized for the emotion recognition agent and the empathy expression agent, ensuring that the system can quickly respond to changes in user emotions, such as quickly outputting a soothing response when the user is emotionally agitated. The standard cognitive layer is primarily used to analyze user cognitive patterns and design intervention strategies. At this stage, the cognitive assessment agent and the intervention planning agent receive more computing resources to support in-depth analysis of user thinking logic and intervention path planning. The deep reflection generation layer focuses on advancing long-term goals and optimizing strategies. The meta-supervision agent and the intervention planning agent play a leading role in this phase, obtaining sufficient computing resources to evaluate intervention effectiveness and adjust the overall plan.

[0094] After allocating basic resources, 103 will also combine the psychological urgency of different intelligent agent data processing tasks (such as the urgency of emotion recognition tasks when users show suicidal tendencies), cognitive complexity (such as the amount of computing required to analyze complex cognitive biases) and interactive needs (such as the frequency of empathetic expression in real-time dialogue scenarios), and perform weighted calculations through differential resource allocation functions to further refine the resource allocation of each intelligent agent.

[0095] For example, when a user is experiencing an emotional breakdown, the emotional recognition agent's data processing task has a high level of psychological urgency, and the differential resource allocation function significantly increases its resource weight to ensure real-time and accurate analysis of emotional data. During the deep reflection phase, the intervention planning agent, which has a higher cognitive complexity due to the need to simulate multi-path intervention effects, will have its computational resource share increased accordingly. Furthermore, the weight configuration parameters in the differential resource allocation function dynamically change with the conversational stage. For example, when transitioning from the immediate emotional response layer to the standard cognitive layer, the emotional recognition agent's weight decreases, while the cognitive assessment agent's weight increases, thereby flexibly adapting to the specific computational resource requirements of different stages.

[0096] Take a user who is anxious due to workplace pressure as an example. At the beginning of the consultation, the user is anxious and expresses panic. The system determines that he is in the immediate emotional response layer and quickly allocates most of the computing resources to the emotion recognition agent and the empathy expression agent. The emotion recognition agent quickly analyzes the user's anxiety level, and the empathy expression agent gives a soothing response in time; as the consultation progresses, the user begins to talk about specific events and his own thoughts at work, entering the standard cognition layer. At this time, the cognitive assessment agent and the intervention planning agent obtain more resources. The former analyzes the user's irrational cognition of "must complete the work perfectly", and the latter designs an intervention path for cognitive reconstruction and stress management based on this; when the user begins to review his career development process and reflect on his own behavior patterns, he enters the deep reflection generation layer, and the meta-supervision agent and the intervention planning agent become the objects of resource allocation. The former ensures that the intervention process complies with ethical standards, and the latter optimizes long-term intervention strategies to promote users to achieve core goals.

[0097] This resource allocation approach enables precise delivery of computing resources, avoiding waste or under-allocation, and improving overall system efficiency. By dynamically adjusting resource allocation strategies, the intelligent collaborative network closely adapts to user needs at different stages of a conversation, enhancing the real-time and targeted nature of consulting services. Different interactive response processing layers each perform their respective functions. The immediate emotional response layer ensures timely attention and comfort to users' emotions, the standard cognitive layer supports in-depth cognitive analysis and intervention design, and the deep reflection generation layer facilitates the achievement of long-term goals. These three layers work synergistically to provide users with a consistent and efficient psychological consulting experience, promoting a gradual improvement in their mental state.

[0098] Further, optionally, in step 103, after obtaining the computing resources allocated to different types of agents through a weighted calculation of psychological urgency, cognitive complexity, and interaction needs using a differential resource allocation function, a hybrid architecture can be employed to optimize the computing resources allocated to each type of agent. Specifically, a critical path synchronization mechanism is employed to allocate synchronous computing resources to the critical intervention path that executes the core data processing task, thereby ensuring parallel advancement of the critical intervention path. A non-critical path asynchronous processing mechanism is employed to allocate asynchronous computing resources to the non-critical intervention path. Furthermore, a computational dependency graph is dynamically constructed to identify the logical dependencies between data processing tasks in real time, optimize the task execution order between different types of agents, and thereby optimize the computational resource allocation for multiple intervention paths.

[0099] In 103, a hybrid architecture was used to perform secondary optimization on the resources calculated by the differential resource allocation function, aiming to further improve resource utilization efficiency and task execution synergy. The differential resource allocation function performs a weighted calculation based on psychological urgency, cognitive complexity, and interaction needs. After initially allocating computing resources to different agents, the hybrid architecture achieves precise resource allocation through a synchronous processing mechanism for critical paths and an asynchronous processing mechanism for non-critical paths.

[0100] Here, the critical path synchronization mechanism focuses on core data processing tasks, which constitute the key intervention path for users. These tasks are related to consultation effectiveness and user safety, and must ensure strong consistency and immediacy. For example, when the emotion recognition agent detects that the user is experiencing an emergency psychological state such as severe anxiety or even suicidal tendencies, the mechanism quickly allocates synchronous computing resources to the emotion recognition agent, crisis assessment agent, and empathy expression agent, ensuring that these agents can process data simultaneously and quickly. The emotion recognition agent accurately analyzes the intensity and type of emotions, the crisis assessment agent immediately determines the risk level, and the empathy expression agent synchronously generates a soothing response. This avoids intervention disconnects caused by delayed resource allocation or asynchronous execution, ensures the parallel advancement of key intervention paths, and provides timely and effective support to users.

[0101] The non-critical path asynchronous processing mechanism targets non-critical intervention path tasks that don't require high real-time performance, such as background information integration and historical case comparative analysis. As users recount their experiences, the non-critical path asynchronous processing mechanism assigns relevant data processing tasks to the corresponding agents and configures asynchronous computing resources, allowing these tasks to execute in parallel in the background without occupying synchronous resources. For example, while the cognitive assessment agent analyzes the user's current cognitive patterns, the historical data comparison task, supported by asynchronous resources, can analyze the user's past response patterns in similar situations, providing a reference for subsequent intervention without competing with core tasks for computing resources, thereby improving overall resource utilization.

[0102] Dynamically constructing a computational dependency graph is a key means of optimizing the order in which tasks are executed. The system scans the data processing tasks of different types of agents in real time and identifies the logical dependencies between tasks. For example, the "identifying user cognitive biases" task can only be completed after the "designing a targeted cognitive reconstruction plan," thus clarifying the sequential dependency between the two. Using the computational dependency graph, the system employs methods such as topological sorting to rationally arrange the order in which tasks are executed, ensuring that while maximizing parallelism, errors in results caused by mismatched task execution are avoided. For example, when formulating an intervention plan, the user's emotional and cognitive state analysis task is completed first, followed by the simulation tasks of different intervention paths. This ensures that each task proceeds in a logical and orderly manner while fully utilizing computing resources to accelerate the overall process.

[0103] For example, a user experiencing depression due to work pressure exhibits low mood and negative expressions at the beginning of the consultation. This indicates a high level of psychological urgency, leading to a differential resource allocation function that allocates more resources to both the Emotion Recognition Agent and the Empathy Expression Agent. A hybrid architecture further enhances this with a critical path synchronous processing mechanism that ensures the Emotion Recognition Agent rapidly interprets the user's emotional signals while the Empathy Expression Agent simultaneously generates empathic responses, providing emotional support. A non-critical path asynchronous processing mechanism delegates the task of integrating the user's past work experience to the relevant agents for asynchronous processing. As the consultation progresses, when it becomes necessary to analyze the user's cognitive bias of "perfection is a failure" and design an intervention plan, a dependency graph is computed to clarify task dependencies. The Cognitive Assessment Agent first identifies the cognitive bias, and the Intervention Planning Agent then designs a plan based on this bias. This optimizes the task execution sequence and resource allocation, ensuring an efficient and orderly intervention process.

[0104] This hybrid architecture optimization method improves resource utilization efficiency, avoids resource waste caused by synchronous processing of all tasks, or intervention failure caused by asynchronous processing of key tasks; ensures timely response and accuracy of key intervention paths, and enhances users' sense of security and trust in emergency psychological states; through reasonable planning of task execution sequence, reduces task waiting time, accelerates the advancement of multiple intervention paths, and improves the overall service quality and efficiency of the psychological counseling system, enabling the system to more flexibly and efficiently respond to the complex psychological needs of different users.

[0105] For example, in 103, in the digital system of psychological counseling, the source optimization scheduling mechanism achieves a dynamic balance between response efficiency and resource consumption by integrating differential computing resource allocation, intelligent agent parallelization framework and semantic compression technology.

[0106] Specifically, the differential computing resource allocation mechanism achieves refined scheduling based on a three-level response gradient architecture. The immediate emotional response layer rapidly activates the emotion recognition agent and the empathy expression agent in response to sudden user emotional crises (such as panic attacks) with a first-packet response of 15-100ms. Computing resources are prioritized to capture immediate physiological signals such as micro-expressions and voice tremors and generate soothing responses. The standard cognitive processing layer, with a response time of 200-500ms, prioritizes resources for the cognitive assessment agent and the intervention planning agent. These agents support the construction of core belief networks and the initial generation of intervention paths, such as addressing logical vulnerabilities in users' "black-and-white" thinking. The deep reflection generation layer, with a response time of 1-3 seconds, provides sufficient resources for the meta-supervision agent and the intervention planning agent to simulate the effectiveness of multi-path interventions and examine ethical risks, such as assessing the potential impact of "exposure therapy" on traumatized users. The resource allocation function achieves stage-by-stage adaptation by dynamically adjusting the weights of psychological urgency, cognitive complexity, and interaction needs. In the emotional crisis stage, the weight of psychological urgency can be increased to 0.6 to ensure immediate response; while in the cognitive reconstruction stage, the weight of cognitive complexity is increased to 0.5 to support complex belief network analysis.

[0107] Optionally, the agent computational parallelization framework adopts a hybrid synchronous-asynchronous processing architecture. The critical path synchronous processing (CPSP) mechanism ensures strong consistency of core tasks such as emotion recognition and crisis assessment. For example, when the emotion recognition agent detects that the user's anxiety value exceeds the standard, the empathy expression agent is synchronously triggered to adjust the response strategy to avoid intervention disconnection due to asynchronous delays. Non-critical path asynchronous processing (NPAP) allows tasks such as background information integration and historical data comparison to be executed in parallel. For example, when a user describes a workplace conflict, the cognitive assessment agent is asynchronously started to analyze their past social patterns. Dependency-aware task scheduling dynamically identifies the logical relationship between tasks through the computational dependency graph (CDG). For example, there is a causal dependency between "identifying trauma trigger points" and "designing exposure scenarios". The system optimizes the execution order through topological sorting, maximizes parallelism and reduces waiting time while ensuring result consistency.

[0108] Furthermore, semantic-preserving compression technology can be used to reduce token consumption in large models through lightweight state management. Lightweight Session State Snapshot (LSS) technology creates incremental checkpoints at task switching points (such as transitioning from the emotional layer to the cognitive layer), recording only the changes in the user's state (such as new negative automatic thoughts) rather than the entire session history. This ensures 99.9% state consistency while reducing duplicate computations. The Core-Periphery Semantic Structure (CPSS) model divides user information into core semantics (such as core beliefs and crisis signals) that must be retained and peripheral content (such as details of repeated instances) that can be compressed, compressing redundant information while ensuring intervention effectiveness. The context-aware reference mechanism uses intelligent indexing technology to replace repeated full content in a conversation with short tags (such as "R3: Anxiety Triggering Event"). For example, if a user repeatedly mentions "being criticized by a leader," the system automatically generates a reference tag, reducing token duplication and improving computational efficiency.

[0109] In an example where these mechanisms work together, a first-packet response time of 87ms was achieved in scenarios involving user emotional outbursts, while token consumption was reduced by 42%. During complex cognitive interventions, perceptual scheduling increased task parallelism by 65%, keeping inference latency below 410ms. Semantic compression technology reduced state synchronization loss in long-distance conversations to less than 0.3%. This optimization not only meets the stringent real-time and professional requirements of psychological counseling, but also reduces computing costs through efficient resource utilization, driving intelligent counseling systems towards practical and scalable application.

[0110] In 104, through a parallel thinking collaborative network, a dynamic collaborative decision-making mechanism is adopted to schedule different types of intelligent agents to collaboratively process different psychological tasks, and a parallel psychological reasoning pipeline is combined to realize psychological consultation interaction with target users.

[0111] As an optional embodiment, in step 104, first, through a dynamic collaborative decision-making mechanism, a parallel psychological reasoning pipeline is employed to implement psychological consultation interaction with the target user in the form of a staged parallel processing pipeline. The decision weights of different types of agents in different conversation stages are dynamically adjusted based on historical intervention effects. Through the parallel thinking collaborative network and dynamic collaborative decision-making mechanism, the system achieves deep interaction with the user in the form of a staged pipeline. The data processing flow for each conversation stage is described in detail below:

[0112] In the first conversation stage, the emotion recognition agent, cognitive assessment agent, behavior analysis-related agent and the agent responsible for narrative processing are scheduled to perform a multi-dimensional psychological state assessment of the target user; the emotion vector features, cognitive relationship graph features, behavior sequence features and narrative structure features obtained by each currently scheduled agent are integrated into a psychological task state vector and passed to the next stage through a dedicated psychological state representation format.

[0113] Among them, the emotion recognition agent identifies the target user's emotional state in the multi-dimensional emotional space by analyzing the target user's language expression, tone, textual emotional vocabulary, etc., and generates emotion vector features; the cognitive evaluation agent builds a core belief network based on the target user's expression, sorts out the target user's thinking mode, finds cognitive distortion points, and forms cognitive relationship graph features; the behavior analysis agent sorts out the behavior sequence features based on the target user's past behavior data and behavioral performance in the current interaction; the narrative processing agent analyzes the content logic of the target user's narration to obtain narrative structure features.

[0114] Specifically, the first conversation phase involves multidimensional psychological state assessment. The system deploys emotion recognition, cognitive assessment, behavioral analysis, and narrative processing agents to analyze the user's current state using multimodal data. For example, the emotion recognition agent analyzes the user's accelerated speech rate and frequent use of words like "desperate" when discussing unemployment, generating emotion vector features that indicate anxiety and depressive tendencies. The cognitive assessment agent extracts the core belief "self-denial" from the user's statement "I will never find a job" and constructs a cognitive relationship graph feature that includes catastrophizing. The behavioral analysis agent records behavioral sequence features such as increased silences and stiff body movements when answering sensitive questions. The narrative processing agent identifies narrative structural features such as logical jumps and repeated emphasis on failure experiences in the user's narrative. These features are integrated into a psychological task state vector and passed to the next stage in the psychological state representation format (PSRF), providing a foundation for targeted intervention. The PSRF uses a standardized structure (e.g., a nested model encompassing emotion, cognition, behavior, and narrative dimensions) to ensure consistency and interpretability of multidimensional data.

[0115] In the second session stage, the intervention planning agent, cognitive evaluation agent, and meta-supervision agent are scheduled; the intervention planning agent generates multiple candidate psychological hypotheses based on the psychological task state vector, combined with the general psychological principles and evidence-based intervention methods in the first-level knowledge base; the cognitive evaluation agent logically verifies the multiple candidate psychological hypotheses from a professional perspective to check whether the multiple candidate psychological hypotheses are consistent with psychological theories and the actual conditions of the target users; the meta-supervision agent examines whether the multiple candidate psychological hypotheses comply with ethical norms and professional standards; the verification results of each currently scheduled agent are fed back to the intervention planning agent, and the intervention planning agent adjusts and optimizes the multiple candidate psychological hypotheses based on the verification results, screens out the target psychological hypothesis, and passes it to the next stage in a dedicated psychological state representation format.

[0116] Specifically, the second session involves the generation and verification of psychological hypotheses. The intervention planning agent, based on the state vector transmitted via the PSRF in the first session and incorporating theories from the primary knowledge base, such as cognitive behavioral therapy (CBT) and psychodynamics, generates multiple candidate hypotheses. For example, for users experiencing unemployment anxiety, hypotheses might include "anxiety stems from catastrophizing about financial pressure" or "the core belief 'I am a loser' drives avoidance behavior." The cognitive assessment agent professionally verifies the logical consistency of the hypotheses, such as whether "catastrophizing" matches the narrative structure features in the user's PSRF. The meta-supervision agent reviews the hypotheses for ethical risks (e.g., whether they over-exploit traumatic details). After feedback from the PSRF, the intervention planning agent adjusts the hypotheses, selecting the target hypothesis that best aligns with the user's reality (e.g., "unemployment anxiety is dually related to low self-worth and catastrophizing") and passing it on to the third session using the PSRF to ensure the hypothesis's scientific and ethical compliance.

[0117] In the third conversation stage, the intervention planning agent generates multiple intervention paths based on the target psychological hypotheses obtained through screening, simulates the target user's possible reactions to different intervention measures, and evaluates the expected effects of each path; the empathy expression agent optimizes the language expression in the intervention strategy; based on the treatment value function, combined with the expected effect coefficient, time efficiency coefficient, customer acceptance and risk factors, a comprehensive evaluation of multiple intervention strategies is conducted, the program priority is dynamically adjusted, and the best intervention strategy is screened out; the best intervention strategy is output through a dedicated psychological state representation format, and real-time interaction information with the target user is generated to provide personalized psychological support for the target user.

[0118] Specifically, the third session involves intervention strategy generation and optimization. The intervention planning agent generates multiple intervention pathways based on the target psychological hypothesis, such as "cognitive restructuring + relaxation training" and "trauma narrative + building a social support network." The agent simulates user responses to different pathways (e.g., the probability of resistance to exposure therapy) using the PSRF. The empathy expression agent optimizes intervention language, changing "You're overgeneralizing" to "Does this pattern of thinking make you feel trapped?" to enhance acceptance. The optimized language parameters are synchronized in real time to the intervention pathway model via the PSRF. The treatment value function integrates expected outcomes (e.g., the likelihood of reducing anxiety scale scores), time efficiency (e.g., the feasibility of completing the intervention within 8 weeks), user acceptance (e.g., willingness to work from home), and risk factors (e.g., the probability of triggering a panic attack). Path priority is dynamically adjusted based on dynamic data weights in the PSRF, ultimately selecting the optimal strategy: "First, alleviate physical symptoms of anxiety through mindfulness training, then gradually challenge negative beliefs." The PSRF then outputs interactive information (e.g., guiding users to keep an anxiety diary) to ensure personalized and implementable strategies.

[0119] It is understandable that the aforementioned stages can be divided by automatically identifying the current process through the stage characteristics of the multimodal data in the PSRF (such as emotional stability, cognitive complexity, and narrative coherence indicators). For example, the user's transition from emotional breakdown (anxiety vector value > 0.8 in the PSRF in the first stage) to rational discussion of the problem (a 30% reduction in the complexity of the cognitive relationship diagram in the second stage) and then to active experimentation with new behaviors (an increase in the frequency of positive responses in the behavioral sequence in the third stage) marks the entry of the consultation into different levels. The core role of the stage division is to achieve a closed-loop advancement of "assessment-hypothesis-intervention", with each stage focusing on a specific task to avoid information overload. The first stage uses the PSRF to comprehensively collect data to ensure the correct direction of the intervention. The second stage verifies the hypothesis based on the PSRF to improve the scientific nature of the plan. The third stage uses the PSRF to simulate dynamic feedback to achieve personalized adaptation. The consistency between stages is guaranteed by the Mind State Representation Format (PSRF), which adopts a hierarchical storage structure (such as a nested dictionary indexed by timestamps) to update user interaction information (such as feedback scores on a certain intervention strategy) to the data nodes of each stage of the PSRF in real time. For example, the user's resistance to "exposure exercises" (identified by the behavioral sequence feature of silence duration greater than 120 seconds in the PSRF) will trigger the system to go back to the second stage to re-verify the hypothesis, or adjust the path priority of the third stage, forming a dynamic cycle of "perception-decision-action-feedback" to ensure that the consultation strategy evolves flexibly with the user's state.

[0120] Phased parallel processing enables the system to simultaneously complete multi-dimensional evaluation (the first phase takes about 5 minutes), multi-hypothesis verification (the second phase takes about 8 minutes), and multi-path simulation (the third phase takes about 7 minutes) in a single session, improving overall efficiency compared to traditional linear processes. The standardized data transmission mechanism of PSRF enables the dynamic weight adjustment mechanism to optimize the decision priority of the intelligent agent based on historical intervention data (such as the response pattern in the PSRF of similar users), thereby improving the accuracy of the hypothesis. The treatment value function obtains comprehensive evaluation parameters in real time through PSRF, which improves user acceptance of the intervention plan and reduces the incidence of risk events. Through the full process of PSRF, the system realizes the seamless connection of multi-stage data and the precise scheduling of intelligent agent collaboration, providing users with clear-cut, scientific and safe personalized psychological support, and promoting the development of digital psychological counseling towards precision and intelligence.

[0121] Further optionally, during the interaction process, the received user interaction information is updated to the parallel thinking collaboration network through a dedicated mental state representation format to ensure that the user interaction information is fed back to each conversation stage and flows coherently between each conversation stage.

[0122] In terms of data layering design, PSRF can be divided into a three-level structure of "base layer-feature layer-decision layer". The base layer stores original multimodal data (such as speech waveforms and text logs), the feature layer integrates structured features processed by the intelligent agent (such as emotion vectors, nodes and edge data of cognitive relationship graphs), and the decision layer records the processing results of each stage (such as candidate hypotheses and intervention path priorities). This hierarchical structure allows specific-level data to be retrieved on demand during parsing, avoiding the redundancy of loading the full amount of data. For example, when verifying psychological hypotheses in the second conversation stage, the system only needs to parse the cognitive relationship graph features of the feature layer and the historical hypothesis data of the decision layer. There is no need to repeatedly process the original speech data of the base layer, which can improve the parsing speed.

[0123] To establish a dynamic indexing mechanism, we add index tags to frequently accessed data in the PSRF (such as core features of the current session and real-time user feedback). For example, by adding the "Current Session - Emotion - Real-Time" tag to sentiment vector features, the system can directly locate the latest sentiment data through the index when adjusting intervention strategies in real time, eliminating the need to traverse the entire data structure. Combining caching technologies (such as Redis) to store hot data can further reduce the response time of high-frequency parsing operations to milliseconds, meeting the low-latency requirements of real-time interaction scenarios.

[0124] In terms of lightweight and on-demand loading, a scalable modular structure is designed, allowing the system to dynamically load required data modules based on the current session phase. For example, in the first session phase, only the emotion, cognition, behavior, and narrative modules related to multidimensional assessment are loaded, while the intervention path simulation data for the third phase is temporarily not loaded. When entering the third phase, the assessment module is dynamically unloaded and the strategy optimization data is loaded. This "on-demand loading" mechanism reduces memory usage, improves parsing efficiency, and reduces parsing lag in complex session scenarios.

[0125] In terms of compatibility and cross-stage linkage optimization, ensure that the PSRF structure supports data backtracking and updating in different conversation stages. For example, when a user shows resistance to a certain intervention strategy in the third stage, the system can quickly locate the psychological hypothesis data of the second stage and the original behavioral characteristics of the first stage through the hierarchical structure of the PSRF to analyze the root cause of the resistance. To this end, a "stage association pointer" field can be added to the PSRF to record the logical dependency between the data (such as "current intervention path → target psychological hypothesis → original cognitive characteristics"), so that the parser can quickly concatenate cross-stage data along the pointer to avoid repeated queries. Experiments have shown that the PSRF with association pointers improves parsing efficiency when retrieving cross-stage data.

[0126] Through these optimizations, PSRF parsing efficiency is improved, memory usage is reduced, and latency in real-time interaction scenarios is stabilized within the user's required response time, meeting the needs of high-frequency data exchange and dynamic decision-making in psychological counseling. The structured layered design and indexing mechanism also enhance data traceability, facilitating the system's review of intervention effects and model iteration. For example, by parsing historical PSRF data, the characteristic distribution of different user groups at each stage can be analyzed, providing data support for the adjustment of decision weights by the agent, further improving the accuracy and efficiency of psychological counseling.

[0127] In another optional embodiment, in step 104, when a parallel psychological reasoning pipeline is used to implement psychological counseling interactions with target users, if several pre-strategies conflict, a multi-agent consensus mechanism based on game theory can be used to arbitrate the intervention strategies output by different agents. For example, in low-conflict situations, different intervention strategies are weighted and fused; in moderate-conflict situations, a dual-path testing approach is used to select the target intervention strategy with the best test results from among different intervention strategies; and in high-conflict situations, a meta-supervisory agent arbitration decision is used to select the target intervention strategy from among different intervention strategies.

[0128] In psychological counseling interactions, when conflicting intervention strategies generated by parallel psychological reasoning pipelines arise, a multi-agent consensus mechanism based on game theory dynamically assesses the degree of conflict and implements hierarchical arbitration to ensure the scientific and consistent nature of the intervention strategies. The degree of conflict is determined based on the dimensions and scope of strategic differences: low conflict manifests as differences in strategic details (such as the duration of relaxation training), moderate conflict involves divergent theoretical frameworks (such as the choice of pathways between cognitive behavioral therapy and mindfulness therapy), and high conflict touches on ethical boundaries or safety risks (such as the conflict between forced trauma exposure and the principle of avoiding secondary harm). By calculating the parameter differences between strategies in dimensions such as psychological urgency, cognitive complexity, and ethical compliance (e.g., the difference in risk factors between two strategies is greater than 0.4), and combining this with a database of historical conflict cases, the system automatically categorizes the conflict into different levels.

[0129] Specifically, in low-conflict scenarios (e.g., differences in strategy details), when the emotion recognition agent suggests extending mindfulness breathing exercises to 8 minutes per session, while the intervention planning agent recommends 5 minutes per session based on time efficiency considerations, the system initiates a weighted fusion mechanism. Based on the weight configuration of the current session (e.g., the psychological urgency weight of 0.6 in the immediate emotional response layer), the emotion recognition agent's suggestion weight is set to 0.6, and the intervention planning agent's suggestion weight is set to 0.4, generating a compromise strategy of "first 5 minutes of practice, then extending to 8 minutes based on real-time user feedback." This fusion preserves the core demands of the different agents and avoids delays in intervention due to disagreements over details.

[0130] In moderate conflict scenarios (e.g., theoretical framework disagreement), if the cognitive assessment agent believes the user's social anxiety stems from a cognitive bias toward self-denial and recommends a cognitive reconstruction strategy, while the behavioral analysis agent, based on the user's history of avoidance behavior, prioritizes systematic desensitization, creating a moderate conflict between the two. In this scenario, the system initiates a dual-path test, validating both strategies in parallel within a safe and controllable simulated environment (e.g., a virtual social setting): the cognitive reconstruction group guides users to record positive self-statements, while the systematic desensitization group gradually exposes them to virtual social situations. The two groups are compared for the magnitude of their anxiety reduction (e.g., heart rate change and subjective distress score over 20 minutes), and the strategy with the best test results (e.g., a 25% reduction in anxiety in the systematic desensitization group vs. an 18% reduction in the cognitive reconstruction group) is selected as the target intervention strategy.

[0131] In high-conflict scenarios (such as ethical risk aversion), when the intervention planning agent generates a strategy of "forced recall of traumatic details to promote exposure," the meta-supervision agent detects that this strategy may trigger the risk of post-traumatic stress disorder (PTSD) relapse (risk factor > 0.7), creating a high level of conflict. At this point, the meta-supervision agent activates an arbitration mechanism, directly rejecting the high-risk strategy based on ethical principles of psychological counseling (such as the "minimum harm principle") and invoking the crisis intervention plan, recommending the use of safe base technology to gradually build trust and avoid psychological harm to the user.

[0132] This mechanism improves the efficiency of resolving strategic conflicts, increases the acceptance of fusion strategies in low-conflict scenarios, enhances the effectiveness of strategies through dual-path testing in moderate-conflict scenarios, and achieves zero risk events through ethical arbitration in high-conflict scenarios. By quantifying the interests of intelligent agents through game theory models (e.g., emotion recognition agents pursue emotional soothing efficiency, and meta-supervision agents pursue ethical safety), the system can maintain the consistency and professionalism of intervention strategies in conflict scenarios. For example, in experiments simulating social anxiety intervention, the consensus mechanism improved user compliance with intervention plans and shortened treatment durations. This hierarchical arbitration mechanism respects the professional division of labor among multiple intelligent agents while avoiding the limitations of a single perspective through scientific decision-making, providing a reliable conflict resolution solution for complex psychological intervention scenarios.

[0133] In the technical solution of this application, through multi-layer goals and parallel thinking collaborative networks, it is possible to achieve a high degree of simulation and personalization of the psychological counseling process, improve the ability and adaptability to cope with complex psychological problems, and significantly improve the quality and effectiveness of psychological counseling services.

[0134] In another embodiment of the present application, a psychological consultation data processing system is also provided. Figure 2 Said system comprises the following units:

[0135] An acquisition unit is configured to acquire multimodal perception data of a target user; wherein the multimodal perception data includes at least: first session data, psychological scale data, and perception interaction data;

[0136] A construction unit is configured to construct a multi-layer goal for a target user based on the multimodal perception data; the multi-layer goal is used to dynamically plan the target user's interaction goals in the psychological consultation process from top to bottom;

[0137] The collaboration unit is configured to dynamically configure the types of agents included in the parallel thinking collaboration network based on the multi-layered goals, and to allocate parallel computing resources for different types of agents in the parallel thinking collaboration network; wherein the parallel thinking collaboration network is used to simulate the multi-dimensional parallel thinking ability of a counselor; different types of agents are responsible for data processing tasks under different psychological counseling dimensions; the types of agents in the parallel thinking collaboration network include one or a combination of the following: an emotion recognition agent, a cognitive assessment agent, an intervention planning agent, an empathy expression agent, and a meta-supervision agent;

[0138] The interaction unit is configured to use a parallel thinking collaborative network and a dynamic collaborative decision-making mechanism to schedule different types of intelligent agents to collaboratively process different psychological tasks. It combines a parallel psychological reasoning pipeline to achieve psychological consultation interaction with the target user, providing the target user with personalized psychological support solutions in complex scenarios.

[0139] The system can implement various steps in the above method embodiments, which will not be expanded here.

[0140] In the embodiment of the present application, a psychological counseling data processing system is adopted, which can achieve a high degree of simulation and personalization of the psychological counseling process through multi-layer goals and parallel thinking collaborative networks, improve the ability to cope with and adapt to complex psychological problems, and significantly improve the quality and effectiveness of psychological counseling services.

[0141] See also Figure 3 , Figure 3 This is a schematic diagram of an embodiment of an electronic device provided in an embodiment of the present application. Figure 3 As shown, an embodiment of the present application provides an electronic device 500, including a memory 510, a processor 520, and a computer program 511 stored in the memory 510 and executable on the processor 520. When the processor 520 executes the computer program 511, a psychological counseling data processing method is implemented.

[0142] The embodiment of the present application provides a computer-readable storage medium. The embodiment provides a computer-readable storage medium having a computer program stored thereon, which implements a psychological consultation data processing method when executed by a processor.

[0143] It should be noted that, in the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0144] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0145] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0146] Obviously, those skilled in the art may make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if such changes and modifications of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include such changes and modifications.

Claims

1. A psychological consultation data processing method, characterized in that: The method comprises: Acquire multimodal perception data of the target user; wherein the multimodal perception data includes at least: first session data, psychological scale data, and perception interaction data; Constructing multi-layer goals for target users based on multimodal perception data; the multi-layer goals are used to dynamically plan the target users' interaction goals in the psychological counseling process from top to bottom; Based on the multi-layered objectives, the types of agents included in the parallel thinking collaboration network are dynamically configured, and parallel computing resources are allocated to different types of agents in the parallel thinking collaboration network; wherein the parallel thinking collaboration network is used to simulate the multi-dimensional parallel thinking ability of the counselor; different types of agents are responsible for data processing tasks under different psychological counseling dimensions; the types of agents in the parallel thinking collaboration network include one or a combination of the following: emotion recognition agent, cognitive assessment agent, intervention planning agent, empathy expression agent, and meta-supervision agent; Through a parallel thinking collaborative network, a dynamic collaborative decision-making mechanism is adopted to schedule different types of intelligent agents to collaborate on different psychological tasks. Combined with a parallel psychological reasoning pipeline, psychological consultation interaction with target users is realized, providing target users with personalized psychological support solutions in complex scenarios.

2. The psychological consultation data processing method according to claim 1, characterized in that: After obtaining the multimodal perception data of the target user, the method further includes: Projecting the psychological state information contained in the multimodal perception data into the psychological task state space to obtain a psychological task state vector; wherein the psychological task state vector includes at least the following dimensions: emotion vector features, cognitive relationship graph features, behavior sequence features, and narrative structure features; Based on the collaborative protocol of the psychological task state space, the psychological task state vector is shared among different types of intelligent agents according to the processing requirements of different data processing tasks.

3. The psychological consultation data processing method according to claim 1, characterized in that: The step of constructing a multi-layer target for a target user based on multimodal perception data includes: The decision coordinator analyzes the first conversation data and the psychological scale data to generate a three-layer nested goal structure; the three-layer nested goal structure in the multi-layer goal includes at least: conversation macro goals, stage goals, and micro-interaction goals; the conversation macro goals are obtained based on long-term psychological counseling planning, the stage goals are obtained based on medium-term psychological counseling planning, and the micro-interaction goals are dynamically configured for each conversation process; Connecting each target in the multi-layer target based on a directed acyclic graph; The method also includes: each time after receiving conversation interaction information fed back by the target user, calculating the treatment value function of each intervention path through real-time path re-evaluation, and dynamically adjusting the target priority of each target in the multi-layer target based on the recalculated treatment value function.

4. The psychological consultation data processing method according to claim 1, characterized in that: The method of dynamically configuring the types of agents included in the parallel thinking collaborative network based on the multi-layer goals includes: According to the target level and task complexity, multiple agents matching the target user are dynamically started through a central scheduler to form the parallel thinking collaborative network; The multiple agents include at least: an emotion recognition agent that uses an attention-enhanced dual-channel network to support multi-dimensional emotion space recognition; a cognitive assessment agent that builds a core belief network and identifies thinking patterns and cognitive distortions; an intervention planning agent that generates and evaluates multiple intervention paths based on the multi-layered goals, wherein the multiple intervention paths are used to simulate user reactions and support parallel advancement; an empathy expression agent that optimizes the professionalism and gentleness of language expression; and a meta-supervision agent that monitors ethical boundaries and professional standards. Through the parameterized configuration equation of the agent's capabilities, the functional specialization, processing priority, resource allocation rate and task response threshold of multiple agents are dynamically allocated and set and dynamically adjusted to complete the intelligent setting of multiple agents.

5. The psychological consultation data processing method according to claim 1, characterized in that: The configuring of parallel computing resources for different types of agents in the parallel thinking collaborative network includes: Based on the multi-layered goals, a response gradient architecture for the target user is constructed; wherein the response gradient architecture is used to set the interactive response time with the target user at different conversation stages; the interactive response processing layer at different conversation stages includes at least: an immediate emotional response layer, a standard cognitive layer, and a deep reflection generation layer; Determine the current session stage of the target user and allocate computing resources to the target user using the matching interactive response processing layer in the response gradient architecture; For the computing resources allocated to target users, the psychological urgency, cognitive complexity, and interaction requirements of the data processing tasks of different types of agents are combined, and weighted calculations are performed through the differential resource allocation function to obtain the computing resources corresponding to different types of agents. Among them, the weight configuration parameters in the differential resource allocation function change with the change of session stage to adapt to the specific requirements of computing resources in different stages.

6. The psychological consultation data processing method according to claim 5, characterized in that: After performing weighted calculations on psychological urgency, cognitive complexity, and interaction requirements using a differential resource allocation function to obtain computing resources allocated to different types of agents, the method further includes: A hybrid architecture is used to optimize the computing resources allocated to different types of agents. A critical path synchronization mechanism is used to allocate synchronous computing resources to the critical intervention paths that perform core data processing tasks, ensuring that the critical intervention paths advance in parallel. A non-critical path asynchronous processing mechanism is used to allocate asynchronous computing resources to non-critical intervention paths. Dynamically construct a computational dependency graph, identify the logical dependencies between data processing tasks in real time, and optimize the task execution order between different types of intelligent agents to optimize the computing resource configuration of multiple intervention paths.

7. The psychological consultation data processing method according to claim 1, characterized in that: The parallel thinking collaborative network adopts a dynamic collaborative decision-making mechanism to schedule different types of intelligent agents to collaboratively process different psychological tasks, and combines a parallel psychological reasoning pipeline to achieve psychological consultation interaction with the target user, including: Through a dynamic collaborative decision-making mechanism, a parallel psychological reasoning pipeline is adopted to achieve psychological consultation interaction with the target user in the form of a staged parallel processing pipeline. The decision weights of different types of agents in different conversation stages are dynamically adjusted based on historical intervention effects. In the first session, the emotion recognition agent, cognitive assessment agent, behavior analysis agent, and narrative processing agent are scheduled to perform a multi-dimensional psychological state assessment of the target user. The emotion vector features, cognitive relationship graph features, behavior sequence features, and narrative structure features obtained by each of the currently scheduled agents are integrated into a psychological task state vector and passed to the next stage using a dedicated psychological state representation format. The emotion recognition agent identifies the target user's emotional state in a multidimensional emotional space by analyzing the target user's language expression, tone of voice, and textual emotional vocabulary, and generates emotion vector features. The cognitive evaluation agent constructs a core belief network based on the target user's expression, sorts out the target user's thinking pattern, identifies cognitive distortion points, and forms cognitive relationship graph features. The behavior analysis agent compiles behavior sequence features based on the target user's past behavior data and current behavior performance in the interaction. The narrative processing agent analyzes the content logic of the target user's narration to obtain narrative structure features. In the second session, the intervention planning agent, cognitive evaluation agent, and meta-supervision agent are scheduled. The intervention planning agent generates multiple candidate psychological hypotheses based on the psychological task state vector and combines the general psychological principles and evidence-based intervention methods in the first-level knowledge base. The cognitive evaluation agent logically verifies the multiple candidate psychological hypotheses from a professional perspective to check whether the multiple candidate psychological hypotheses are consistent with psychological theories and the actual conditions of the target users. The meta-supervision agent examines whether the multiple candidate psychological hypotheses comply with ethical norms and professional standards. The verification results of each currently scheduled agent are fed back to the intervention planning agent, which adjusts and optimizes the multiple candidate psychological hypotheses based on the verification results, selects the target psychological hypothesis, and passes it to the next stage in a dedicated psychological state representation format. In the third conversation phase, the intervention planning agent generates multiple intervention paths based on the screened target psychological hypotheses, simulates the target user's possible reactions to different intervention measures, and evaluates the expected effects of each path. The empathy expression agent optimizes the language expression of the intervention strategy. Based on the treatment value function, combined with the expected effect coefficient, time efficiency coefficient, customer acceptance and risk factors, a comprehensive evaluation of multiple intervention strategies is conducted, and the priority of the solutions is dynamically adjusted to screen the optimal intervention strategy. The optimal intervention strategy is output through a dedicated psychological state representation format, and real-time interaction information is generated with the target user to provide personalized psychological support for the target user. During the interaction process, the received user interaction information is updated to the parallel thinking collaboration network through a dedicated mental state representation format to ensure that the user interaction information is fed back to each conversation stage and flows coherently between each conversation stage.

8. The psychological consultation data processing method according to claim 7, characterized in that: The process of implementing psychological consultation interaction with the target user by combining the parallel psychological reasoning pipeline also includes: If there is a conflict in intervention strategies, a multi-agent consensus mechanism based on game theory is used to arbitrate the intervention strategies output by different agents; Among them, in low-conflict situations, different intervention strategies are weightedly fused; in moderate-conflict situations, a dual-path testing method is used to select the target intervention strategy with the best test effect from different intervention strategies; in high-conflict situations, a meta-supervisory intelligent agent arbitration decision is used to select the target intervention strategy from different intervention strategies.

9. A psychological consultation data processing system, characterized in that: The system includes the following units, wherein: An acquisition unit is configured to acquire multimodal perception data of a target user; wherein the multimodal perception data includes at least: first session data, psychological scale data, and perception interaction data; A construction unit is configured to construct a multi-layer goal for a target user based on the multimodal perception data; the multi-layer goal is used to dynamically plan the target user's interaction goals in the psychological consultation process from top to bottom; The collaboration unit is configured to dynamically configure the types of agents included in the parallel thinking collaboration network based on the multi-layered goals, and to allocate parallel computing resources for different types of agents in the parallel thinking collaboration network; wherein the parallel thinking collaboration network is used to simulate the multi-dimensional parallel thinking ability of a counselor; different types of agents are responsible for data processing tasks under different psychological counseling dimensions; the types of agents in the parallel thinking collaboration network include one or a combination of the following: an emotion recognition agent, a cognitive assessment agent, an intervention planning agent, an empathy expression agent, and a meta-supervision agent; The interaction unit is configured to use a parallel thinking collaborative network and a dynamic collaborative decision-making mechanism to schedule different types of intelligent agents to collaboratively process different psychological tasks. It combines a parallel psychological reasoning pipeline to achieve psychological consultation interaction with the target user, providing the target user with personalized psychological support solutions in complex scenarios.

10. An electronic device, characterized in that: including a memory for storing a computer software program; A processor is used to read and execute the computer software program, thereby implementing the psychological counseling data processing method described in any one of claims 1-8.

Images