Multi-agent based psychological intervention method and system

Abstract

The application discloses a psychological intervention method and system based on multiple agents, and the psychological intervention method comprises the following steps: a master routing agent acquires initial information input by a patient and calculates information entropy of the initial information; according to the information entropy of the initial information, the master routing agent calculates a cognitive computing power gap when the patient processes the initial information; when the cognitive computing power gap is greater than a preset value, the master routing agent wakes up a corresponding sub-agent according to a gap feature of the cognitive computing power gap, and distributes the initial information and an allocated weight to the sub-agent; the sub-agent performs information dimension reduction processing on the initial information to obtain reduced dimension information and sends the reduced dimension information to the master routing agent; the master routing agent integrates the reduced dimension information according to the allocated weight, and feeds back the integration result to the patient in the form of natural language, so that the computing power is transferred and the information entropy is balanced, the psychological intervention on the patient is realized, and the patient is prevented from being in a state of 'full load crash'.

Description

Technical Field

[0001] This invention relates to the field of psychotherapy technology, and in particular to a psychological intervention method and system based on multiple agents. Background Technology

[0002] From the perspective of information theory and cognitive science, when an individual's cognitive processing power is insufficient to handle the high-entropy (i.e., high-complexity) information in the real environment, psychological symptoms such as anxiety and depression will occur, and it may even cause the brain to "crash from overload." The essence of real psychological counseling is that the counselor, through the acoustic-semantic channel, transfers "cognitive processing power" to the patient to assist the patient in completing a compensatory dimensionality reduction of high-entropy information.

[0003] Most current mainstream AI (Artificial Intelligence) psychological intervention products employ a single, conversational, large-scale language model. This means they merely act as a CBT (Cognitive Behavior Therapy) counselor, engaging in superficial conversations with patients without addressing the essence of psychological counseling and lacking the theoretical legitimacy supporting psychological intervention. Furthermore, this single-agent architecture has weak technological barriers, heavily relying on prompts for role simulation. Competitors can easily circumvent this by modifying the prompts and achieving similar functionality.

[0004] The statements herein provide only background information in relation to this invention and do not necessarily constitute prior art. Summary of the Invention

[0005] The purpose of this invention is to provide a psychological intervention method and system based on multi-agent systems, which can complete the transfer of computing power and balance of information entropy when the patient inputs complex information, thereby realizing psychological intervention for the patient and avoiding the patient being in a "fully loaded and crashed" state.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: A multi-agent-based psychological intervention method, comprising a master routing agent and multiple sub-agents awakened under the control of the master routing agent; the psychological intervention method includes: The master routing agent acquires the initial information input by the patient and calculates the information entropy of the initial information; Based on the information entropy of the initial information, the master control routing agent calculates the cognitive computing power gap when the patient processes the initial information. When the cognitive computing power gap is greater than the preset value, the master control routing agent wakes up the corresponding sub-agent based on the gap characteristics of the cognitive computing power gap and distributes initial information and assigns weights to it; The sub-agent performs dimensionality reduction processing on the initial information to obtain dimensionality-reduced information and sends it to the master routing agent; The master routing agent integrates the dimensionality reduction information according to the assigned weights and feeds back the integration results to the patient in the form of natural language.

[0007] Optionally, the initial information is natural language information, and the information entropy of the initial information is calculated using the following formula: in, Information entropy represents the initial information. This represents the set of semantic units extracted from initial information using natural language processing methods. Represents the first in the set of semantic units i A semantic unit, Indicates the first i The probability of a semantic unit appearing in the context and in the patient's historical input information.

[0008] Optionally, the cognitive computing power gap is calculated using the following formula: in, This indicates a cognitive computing power gap. This represents the patient's baseline cognitive computing ability estimate.

[0009] Optionally, when the cognitive computing power gap is greater than a preset value, before the step of the main control routing agent waking up the corresponding sub-agent based on the gap characteristics of the cognitive computing power gap, the method further includes: the main control routing agent performing semantic analysis on the initial information to determine the gap characteristics of the cognitive computing power gap; The steps to determine the characteristics of the cognitive computing power gap include: If the initial information contains words with emotional polarity but lacks specific event logic, the master routing agent will identify the gap feature of the cognitive computing power gap as an emotional catharsis gap. If the initial information contains cognitive distortions, the master routing agent will identify the gap feature of the cognitive computing power gap as a cognitive reconstruction gap; If the initial information contains keywords that trigger multiple types of alarms and are accompanied by stressor keywords, the master routing agent will identify the cognitive computing power gap as a complex trauma gap.

[0010] Optionally, the multiple sub-agents include an empathy sub-agent, a cognitive behavioral therapy sub-agent, and a joint consultation sub-agent; the steps for the master-control routing agent to wake up the corresponding sub-agent based on the gap characteristics of the cognitive computing power gap include: When the cognitive computing power gap is characterized by an emotional catharsis gap, the master control routing agent awakens the empathic sub-agent. When the cognitive computing power gap is characterized as a cognitive reconstruction gap, the master control routing agent awakens the cognitive behavioral therapy sub-agent. When the cognitive computing power gap is characterized as a complex trauma gap, the master control routing agent wakes up the joint consultation sub-agent.

[0011] Optionally, the multi-agent system also includes a review agent; prior to performing the step of feeding back the integrated results to the patient in natural language, the following steps are also included: The review agent checks the integration results to determine whether they contain sensitive words; If the integration results contain sensitive words, the review agent will trigger a blocking mechanism. If the integration result does not contain sensitive words, the review agent will inform the main routing agent to send the integration result back to the patient.

[0012] Optionally, sensitive words include one or any combination of the following: words with self-harm or self-inflicted intentions, words with plans or tendencies to threaten others with violence, words with plans or tendencies to incite crime, and words that advise patients to discontinue professional psychiatric prescription medications. The blocking mechanism includes: suspending the process of the current master routing agent and calling staff to speak with the patient.

[0013] Optionally, if the patient responds to the integration results in natural language, the psychological intervention methods may also include: The master routing agent acquires the patient's response information and calculates the information entropy of the response information; The master routing agent calculates the difference between the information entropy of the initial information and the response information to obtain the information entropy reduction. By using reinforcement learning algorithms and taking the maximization of information entropy reduction as the reward function, the distribution strategy of the master routing agent and the weights of each sub-agent are optimized.

[0014] Optionally, the reward function can be expressed as follows: in, Represents the reward function, Indicates the positive return coefficient. Indicates the decrease in information entropy. Indicates the penalty coefficient. This indicates the penalty value for safety violations.

[0015] On the other hand, the present invention also provides a psychological intervention system, including a memory and multiple agents; the memory stores a computer program, and the multiple agents include a master routing agent and multiple sub-agents awakened by the master routing agent; when the multiple agents execute the program, they implement the method described in any of the preceding claims.

[0016] Compared with the prior art, the present invention has at least one of the following advantages: This invention provides a multi-agent-based psychological intervention method and system. By calculating the patient's cognitive computing power gap, the imbalance between the patient's "information entropy" and their own "cognitive computing power" when facing real-world information can be quantified. When the patient's cognitive computing power gap is greater than a preset value, the master control routing agent wakes up the corresponding sub-agent based on the gap characteristics and distributes the patient's input information and assigns corresponding weights to it. The corresponding sub-agent then takes over and performs dimensionality reduction calculations on the information. The master control routing agent integrates the dimensionality reduction information according to the assigned weights and feeds back the integration results to the patient in the form of natural language. This completes the transfer of computing power and balance of information entropy, achieving psychological intervention for the patient and preventing the patient from being in a "fully loaded and crashed" state.

[0017] This invention uses a review agent to detect whether the integrated results to be fed back to the patient contain sensitive words, thereby continuously monitoring the security boundaries of the output results during multi-agent interaction and dimensionality reduction processing. Furthermore, if the integrated results contain sensitive words, the review agent triggers a blocking mechanism to ensure patient safety.

[0018] After completing computational compensation, this invention calculates the information entropy of the patient's response to the integrated results, and calculates the reduction in information entropy between the initial information and the response information to determine whether the patient's immediate information entropy has decreased, thereby evaluating the effectiveness of the psychological intervention. Furthermore, it utilizes a reinforcement learning algorithm, with maximizing the reduction in information entropy as the reward function, to optimize the distribution strategy of the master routing agent and the weights of each sub-agent.

[0019] This invention has scientific and theoretical legitimacy: by proposing and engineering the "computational power compensation rule", it thoroughly explains why psychological intervention methods and systems based on multi-agent systems can replace human counselors to provide treatment for patients, laying the most solid scientific foundation in the industry for subsequent related products.

[0020] This invention avoids the easily replicable "single chatbot" model in existing technologies, and has high innovation and technical sophistication. Attached Figure Description

[0021] Figure 1 A flowchart illustrating a psychological intervention method provided in an embodiment of the present invention; Figure 2 This is a dynamic matching logic diagram of a sub-agent in a psychological intervention method provided by an embodiment of the present invention; Figure 3 This is a logic diagram of a review agent detecting the integration result in a psychological intervention method provided in an embodiment of the present invention. Figure 4This is a logic diagram of optimizing the distribution strategy of the master routing agent and the weights of each sub-agent in a psychological intervention method provided in an embodiment of the present invention. Detailed Implementation

[0022] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The advantages and features of the present invention will become clearer from the following description. It should be noted that the drawings are in a very simplified form and use non-precise proportions, used only to facilitate and clearly illustrate the embodiments of the present invention. Please refer to the drawings to make the objectives, features, and advantages of the present invention more apparent and understandable. It should be understood that the structures, proportions, sizes, etc., depicted in the accompanying drawings are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed in the specification, and are not intended to limit the implementation conditions of the present invention. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in proportions, or adjustments to the size, without affecting the effects and objectives achieved by the present invention, should still fall within the scope of the technical content disclosed in the present invention.

[0023] Combination Figures 1 to 4 As shown, this embodiment provides a psychological intervention method based on multiple agents, which includes a master router agent and multiple sub-agents that are awakened by the master router agent.

[0024] The psychological intervention method includes: Step S1, the master routing agent acquires the initial information input by the patient and calculates the information entropy of the initial information. Step S2, based on the information entropy of the initial information, the master routing agent calculates the cognitive computing power gap when the patient processes the initial information. Step S3, when the cognitive computing power gap is greater than a preset value, the master routing agent wakes up the corresponding sub-agent according to the gap characteristics of the cognitive computing power gap and distributes the initial information and assigns weights to it. Step S4, the sub-agent performs information dimensionality reduction processing on the initial information to obtain dimensionality-reduced information and sends it to the master routing agent. Step S5, the master routing agent integrates the dimensionality-reduced information according to the assigned weights and feeds back the integration result to the patient in the form of natural language, thereby completing the transfer of computing power and the balance of information entropy, realizing psychological intervention for the patient, and preventing the patient from being in a "fully loaded and crashed" state.

[0025] Specifically, in step S1, the initial information is natural language information, and the carrier of natural language information is text and / or sound. In one embodiment, the initial information is voice or text information input by the patient through a terminal (such as a mobile APP, web page dialog box, etc.), for example: "I feel that I am worthless, I can't do anything well, and life is completely meaningless."

[0026] More specifically, after the master control routing agent obtains the initial information input by the patient, it can use the formula (1) shown below to calculate the information entropy of the initial information.

[0027] (1) in, Information entropy represents the initial information. This represents the set of semantic units extracted from the initial information of the current input using natural language processing methods. Represents the first in the set of semantic units i Each semantic unit. Indicates the first i The probability of a semantic unit appearing in the context and in the patient's historical input information; that is, the probability. The calculation relies on the context of the current text (i.e., the initial information of the current input) and the patient's historical input information. Understandably, the context of the current text includes the guidance information (i.e., the integrated result) that the controlling routing agent fed back to the patient in the previous round of dialogue. If a semantic unit of the patient deviates significantly from the aforementioned contextual information or its historical regular expression (i.e., the appearance of rare emotional words or sudden catastrophic expressions), then the probability of that semantic unit appearing... The entropy of the initial information will be extremely low, resulting in a significant increase in the overall information entropy of the initial information. In one embodiment, the master routing agent has a built-in natural language processing (NLP) model. In this case, after the master routing agent obtains the initial information input by the patient, it can directly call the natural language processing model to perform word segmentation and semantic intent extraction on the initial information, thereby obtaining multiple discrete semantic units (e.g., keywords or topic vectors representing anxiety, depression, specific events), and then obtain a set of semantic units composed of multiple semantic units. Subsequently, the natural language processing model can calculate the information entropy of the initial information according to formula (1).

[0028] Understandably, information entropy is a quantitative indicator that measures the degree of confusion, emotional complexity, and semantic uncertainty in a patient's expression. When patients are under high pressure or experiencing psychological symptoms, the information they input often involves logical jumps and emotional vocabulary stacking, leading to a more uniform or unpredictable probability distribution of each semantic unit. This results in a higher information entropy, which is defined as high-entropy information.

[0029] Specifically, in step S2, the cognitive computing power gap is a specific numerical value, which intuitively quantifies the extent to which the current information (e.g., initial information) exceeds the patient's own processing capacity. If the patient's brain is likened to a GPU or CPU, the larger the cognitive computing power gap, the more the patient is in a "fully loaded and crashed" state, and the more psychological intervention is needed. More specifically, the cognitive computing power gap is calculated using the following formula (2): (2) in, This indicates a cognitive computing power gap. Information entropy represents the initial information. This represents the patient's baseline cognitive computing power estimate; whereby the patient's baseline cognitive computing power estimate reflects the patient's basic ability to process complex information under normal conditions.

[0030] In one embodiment, the patient's baseline cognitive computing power estimate is determined by the structural entropy (graph connectivity) of the patient's historical cognitive graph, and the two are inversely proportional (or negatively correlated). That is, the greater the structural entropy of the patient's historical cognitive graph (such as a more chaotic, fragmented, or extremely collapsed, or loop-like state), the lower the patient's baseline cognitive computing power estimate. Furthermore, if the patient's historical cognitive graph exhibits high collapse (such as deep depression or rigid thinking), the patient's baseline cognitive computing power estimate will be extremely low. Understandably, the patient's historical cognitive graph contains structured nodes and connecting edges; nodes include the patient's "core beliefs (such as 'I am unworthy')," "key life events," "significant others," and "high-frequency emotional words"; connecting edges represent the causal relationships or emotional valences (such as positive / negative influence weights) between these nodes.

[0031] In one specific embodiment, the relationship between a patient's baseline cognitive computing power estimate and the structural entropy of their historical cognitive map can be quantitatively mapped using the following inverse proportional smoothing formula: (3) in, This represents the maximum baseline cognitive computing power constant for a healthy person. This represents the structural entropy of the patient's historical cognitive map calculated using a graph neural network algorithm. The modulating factor is denoted as . When the structural entropy of a patient's historical cognitive profile increases sharply (such as due to mental confusion and rigidity caused by deep depression), the corresponding baseline cognitive computing power estimate will exhibit a non-linear decay, approaching an extremely low value.

[0032] In one embodiment, the master routing agent can pre-build a historical cognitive graph for the patient. The specific steps are as follows: Initially, the master routing agent accesses the patient's baseline assessment scale and historical consultation records; then, based on the patient's baseline assessment scale and historical consultation records, it automatically constructs the patient's historical cognitive graph using Named Entity Recognition (NER) and Relation Extraction (RE) techniques and stores it in a cloud database. In subsequent interactions, new entities and relationships can be extracted in real time from each conversation, dynamically updating and storing the historical cognitive graph in the cloud database for the master routing agent to retrieve in real time.

[0033] Specifically, when the cognitive computing power gap is less than or equal to a preset value, for example, when the initial information input by the patient is simply a low-entropy information such as a daily greeting or a routine inquiry, the main control routing agent directly calls a general conversational language model to provide a lightweight natural language response to the initial information input by the patient. That is, it uses the conversational language model to interact with the patient, thereby saving computing power and reducing resource consumption. Optionally, the conversational language model is built into the main control routing agent; optionally, the preset value can be a hyperparameter, or it can be an adjusted value after dynamic baseline adaptation based on the patient's historical average gap value, but this invention is not limited to this.

[0034] Specifically, when the cognitive computing power gap exceeds a preset value, before executing step S3, the process further includes: step S30, where the main control routing agent performs semantic analysis on the initial information to determine the gap characteristics of the cognitive computing power gap. More specifically, step S30 includes: If the initial information contains a large number of words with strong emotional polarity (such as worship, excitement, fanaticism, rage, despair, hatred, disgust, etc.) and lacks specific event logic (such as disordered chronology, reversed or confused cause and effect, sentence jumps without transition, fragmented language, etc.), the master routing agent will identify the gap characteristics of the cognitive computing power gap as an emotional catharsis gap.

[0035] If the initial information contains obvious cognitive distortions (such as absolute requirements like "must be loyal", "must be number one" and "should be absolutely fair and just"; catastrophic thinking like "everything is over" and "what if...?"), the master routing agent will identify the cognitive computing power gap as a cognitive reconstruction gap.

[0036] If the initial information contains keywords that trigger multiple types of alarms (such as self-harm, despair, kidnapping, etc.) and is accompanied by stressor keywords (such as noise, exams, work, disasters, etc.), the master control routing agent will identify the cognitive computing power gap as a complex trauma gap.

[0037] In one embodiment, the gap characteristics of the cognitive computing power gap can be identified as one or any combination of emotional catharsis gap, cognitive reconstruction gap, and complex trauma gap. Optionally, step S30 is performed by a multidimensional intent classifier within the master routing agent, but the present invention is not limited thereto.

[0038] Specifically, such as Figure 2 As shown, multiple sub-agents include an empathy sub-agent (i.e., the empathy agent), a cognitive behavioral therapy sub-agent (i.e., the CBT agent), and a joint consultation sub-agent (i.e., the joint consultation agent), and these sub-agents can operate collaboratively. More specifically, step S3 includes: when the cognitive computing power gap is characterized by an emotional catharsis gap, the master control routing agent awakens the empathy sub-agent and distributes initial information to it. When the cognitive computing power gap is characterized by a cognitive reconstruction gap, the master control routing agent awakens the cognitive behavioral therapy sub-agent and distributes initial information to it. When the cognitive computing power gap is characterized by a complex trauma gap, the master control routing agent awakens the joint consultation agent and distributes initial information to it.

[0039] In one embodiment, when the cognitive computing power gap is determined to be a combination of emotional catharsis gap, cognitive reconstruction gap, and complex trauma gap, the master routing agent will activate the empathy sub-agent, the cognitive behavioral therapy sub-agent, and the joint consultation sub-agent and distribute initial information to these three sub-agents. Simultaneously, the master routing agent will assign different weights to these three sub-agents. Similarly, when the cognitive computing power gap is determined to be a combination of any two of the emotional catharsis gap, cognitive reconstruction gap, and complex trauma gap, the master routing agent will activate the sub-agent corresponding to that combination and distribute initial information. Simultaneously, the master routing agent will assign different weights to the sub-agent corresponding to that combination, i.e., dynamically allocate routing weights.

[0040] Specifically, in step S4, after the sub-agents receive the initial information sent by the master routing agent, each sub-agent performs distributed "computing power compensation"; where "computing power compensation" at the computer system level refers to the information dimensionality reduction and structured rewriting process performed using a large language model.

[0041] More specifically, after receiving initial information (complex high-dimensional emotions or real-world problems), the sub-agent uses its built-in psychological knowledge base (such as the knowledge base storing CBT theory) and large language model to perform information dimensionality reduction processing on the initial information, thereby breaking down the initial information into low-dimensional logical elements and obtaining dimensionality-reduced information to replace the patient's brain in bearing the burden of information processing.

[0042] In one embodiment, the expression for information dimensionality reduction processing is as follows: (4) in, This represents dimensionality reduction information. This represents the processing function of the corresponding sub-agent. Indicates initial information. The table represents the vector of the built-in psychological knowledge base of the corresponding sub-agent.

[0043] Specifically, in step S5, after receiving the dimensionality reduction information (e.g., structured text) output by each sub-agent, the master routing agent strictly applies the weights assigned to each sub-agent in step S3. More specifically, the master routing agent utilizes the attention mechanism of a large language model or weighted prompts to perform weighted fusion of the dimensionality reduction information output by each sub-agent based on their weight coefficients, thereby making the generated integrated result more targeted. For example, if the empathy sub-agent has the highest weight, the final integrated result will focus more on emotional support and reassurance in natural language expression; if the cognitive behavioral therapy sub-agent has the highest weight, the integrated result will focus more on logical guidance and cognitive reconstruction.

[0044] Furthermore, the master routing agent can integrate the weighted fusion results using semantic fusion and semantic smoothing methods, thereby generating a coherent and natural integrated result and transmitting it back to the patient in natural language, thus completing a "computing power transfer" and information entropy balance. For example, as a "summarizing node," the master routing agent can use the "emotional holding strategy" output by the empathy sub-agent and the "cognitive reconstruction guidance" output by the cognitive behavioral therapy sub-agent as context. It can then use semantic fusion and semantic smoothing methods in the large language model to regenerate a coherent, natural, and empathetic natural language message (i.e., the integrated result), ensuring that the patient feels like they are conversing with a real, experienced expert, rather than receiving a piecemeal response from a robot.

[0045] Specifically, such as Figure 3 As shown, the multi-agent system also includes a reviewer agent, whose permissions are higher than those of the master routing agent and each sub-agent. Before the step of feeding back the integrated results to the patient in natural language, the reviewer agent checks the integrated results to determine if they contain sensitive words. Sensitive words refer to expressions that cross medical red lines or violate medical ethics; optionally, sensitive words include one or any combination of the following: words with self-harm or self-mutilation intent, words indicating a plan or tendency to threaten others with violence, words inciting a plan or tendency to commit a crime, and words advising the patient to discontinue professional psychiatric prescription medications. Optionally, sensitive words are pre-stored in a hard-coded or strongly validated rule base (e.g., a medical-grade rule base).

[0046] In one embodiment, if the integration result does not contain sensitive words, the review agent performs a release operation, that is, instructs the master routing agent to feed back the integration result to the patient.

[0047] In one embodiment, if the integration result contains sensitive words, the review agent immediately triggers a blocking mechanism within milliseconds. The blocking mechanism includes: suspending the process of the current master routing agent and forcing a seamless call to a staff member (e.g., a human crisis interventionist) to speak with the patient, thus enabling full takeover by a human counselor. Optionally, the blocking mechanism also includes switching to a structured secure response channel, where a pre-defined, manually reviewed text template (e.g., "We understand you are currently in great pain. For your safety, please immediately call the 24-hour crisis intervention hotline XXX") is provided to the patient, thereby buying more time for full takeover by a human counselor.

[0048] Specifically, such as Figure 4 As shown, if the integration result does not contain sensitive words and the patient responds to the integration result in natural language, the psychological intervention method further includes: Step S6, the master routing agent obtains the patient's response information and calculates the information entropy of the response information; Step S7, the master routing agent calculates the difference between the information entropy of the response information and the initial information to obtain the information entropy reduction; Step S8, using a reinforcement learning algorithm and taking the maximization of the information entropy reduction as the reward function, the distribution strategy of the master routing agent and the weights of each sub-agent are optimized, so that the effect of psychological intervention becomes better and better. It can be understood that the distribution strategy not only includes the discrete awakening rule of sub-agents based on gap features, but also includes the continuous dynamic allocation (Soft Weighting) mapping function of the collaborative weights among multiple awakened sub-agents, as well as adjusting the boundary threshold for classifying gap feature categories; in this way, when facing complex and ambiguous gap features, the gap features can be reasonably determined and the weights among the awakened sub-agents can be accurately allocated.

[0049] In one embodiment, the calculation process of the information entropy of the response information in step S6 is similar to the calculation process of the information entropy of the initial information, and will not be described again here.

[0050] In one embodiment, in step S7, the information entropy reduction can be used to determine whether the patient's immediate information entropy has decreased after the computational compensation is completed, so as to evaluate the effect of the psychological intervention. Specifically, if the information entropy reduction is greater than 0, that is, the information entropy of the initial information is greater than the information entropy of the response information, it indicates that the patient's emotions are stabilizing and logic is recovering. At this time, the patient's information entropy has decreased, and the effect of the psychological intervention is good. If the information entropy reduction is less than 0, that is, the information entropy of the initial information is less than the information entropy of the response information, it indicates that the patient's emotions are more agitated and logic is more confused. At this time, the patient's information entropy has not decreased, and the effect of the psychological intervention is poor.

[0051] In one embodiment, in step S8, the expression for the reward function is as follows: (5) in, Represents the reward function, Indicates the positive return coefficient. Indicates the decrease in information entropy. Indicates the penalty coefficient. This represents the penalty value for safety violations (this value is extremely large if it causes the review agent to trigger the blocking mechanism). It should be noted that safety violation penalties include not only situations that trigger the review agent's blocking mechanism, but also various other violations of ethical red lines in psychological counseling, such as: the inclusion of unethical clinical medical diagnoses in the integrated results, the appearance of repetitive invalid responses in a vicious cycle, or the appearance of negative expressions that deprive the patient of emotional nuances (such as statements like "Just try to be more positive," which deprive the patient of their feelings). The safety violation penalty value is a pre-set set of stepped numerical constants based on the severity of the potential adverse consequences (e.g., violations involving life safety are assigned a very large negative constant, while general invalid responses are assigned a smaller negative constant). In the reinforcement learning environment, these values ​​serve as negative penalty signals to constrain the behavioral boundaries of the master routing agent.

[0052] Optionally, the reinforcement learning algorithm is the PPO (Proximal Policy Optimization) algorithm; the value of the reward function calculated according to formula (4) will be fed back to the policy network (see Figure 4 PPO, through a gradient ascent algorithm, enables the master routing agent, when faced with similar high-entropy input information in the future, to increase the probability of selecting sub-agent combination strategies that produce higher reward function values ​​and decrease the probability of selecting sub-agent combination strategies that lead to negative penalties, thus achieving self-iterative collaborative weight allocation among sub-agents. This can be understood as... Figure 4 As shown, after each optimization of the distribution strategy of the master routing agent and the weights of each sub-agent, the current iteration ends, and a computational power compensation or transfer is completed. The sub-agent combination strategy and participation weights can be recorded. Furthermore, in each iteration, the response information can be used as initial information and the process returns to steps S1-S8.

[0053] Furthermore, this embodiment also provides a psychological intervention system, including a memory and multiple agents; the memory stores a computer program, and the multiple agents include a master routing agent and multiple sub-agents awakened by the master routing agent; when the multiple agents execute the program, they implement the method described above.

[0054] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0055] The artificial intelligence-based psychological assessment and medical auxiliary diagnosis solutions involved in the embodiments of the present invention strictly comply with national regulations on data security, personal information protection and medical industry management in terms of data collection, processing, model training and application, and meet the requirements of medical ethics review.

[0056] The diagnostic data involved in the embodiments of this invention (such as, but not limited to, behavioral observation data, clinical consultation records, physiological and biochemical indicators, etc.) are all derived from legal and compliant publicly available de-identified datasets, or legally collected with the subject's (or their legal guardian's) explicit consent after full disclosure. During the data collection process, this invention does not involve concealment, deception, or unauthorized acquisition of sensitive personal information.

[0057] In the embodiments of this invention, when collecting personal biometric information such as facial micro-expressions, eye gaze trajectories, voice acoustic features, or body movements of subjects for auxiliary assessment in psychological-related diagnoses, the collection is conducted in a specific environment where the subjects are fully informed and voluntarily cooperate. The subjects have been clearly informed of the specific purpose and scope of use of the data collection and have obtained their specific consent. This invention does not contain any covert, non-intrusive, or unauthorized image and sound capture methods.

[0058] If the embodiments of this invention involve multimodal medical data fusion (e.g., simultaneously inputting text questionnaire data, facial expression videos, and voice recordings of subjects), this system adopts a strict permission isolation and encrypted transmission mechanism when processing such data. The original data of different modalities are physically or logically isolated during the feature extraction stage. The artificial intelligence algorithm model only performs fusion calculations on the extracted abstract feature vectors. The entire calculation process and the final output diagnostic results cannot be reversed to reconstruct the original multimodal joint information of the subject, thus eliminating the privacy leakage risk that may be caused by cross-modal data association.

[0059] Before using the aforementioned data for model training or actual diagnosis, all data underwent rigorous de-identification and anonymization processes. This eliminated identity-related biases while retaining the disease characteristic dimensions required for psychological assessment. Furthermore, the algorithmic decision-making mechanism of this invention is built upon objective psychopathological characteristics. During model training, it fully considers the diversity, representativeness, and balance of the samples, and avoids setting biased, stereotypical, or discriminatory decision rules based on non-disease factors such as the subject's gender, age, occupation, or social status.

[0060] The systems, methods, and devices claimed in this invention for psychological assessment and medical auxiliary diagnosis, etc., whose output analysis results, risk probabilities, or classification labels are intended solely for clinical reference by legally qualified psychiatrists, psychotherapists, and other medical professionals. This invention serves only as a clinical auxiliary assessment tool and cannot be used directly as the final basis for clinical diagnosis, nor can it automatically generate drug treatment prescriptions or psychological intervention instructions. The final psychological state assessment and treatment plan must be made by a professional physician through interviews and comprehensive clinical examinations. The implementation of this invention does not replace the diagnostic and treatment actions of professional medical personnel.

[0061] Although the present invention has been described in detail through the preferred embodiments above, it should be understood that the above description should not be considered as a limitation of the present invention. Various modifications and substitutions to the present invention will be apparent to those skilled in the art after reading the above description. Therefore, the scope of protection of the present invention should be defined by the appended claims.

Claims

1. A psychological intervention method based on multi-agent systems, characterized in that, The multi-agent system includes a master routing agent and multiple sub-agents that are controlled and awakened by the master routing agent. The psychological intervention methods include: The master control routing agent acquires the initial information input by the patient and calculates the information entropy of the initial information; Based on the information entropy of the initial information, the master control routing agent calculates the cognitive computing power gap when the patient processes the initial information; When the cognitive computing power gap is greater than a preset value, the master control routing agent wakes up the corresponding sub-agent according to the gap characteristics of the cognitive computing power gap and distributes the initial information and assigns weights to it; The sub-agent performs dimensionality reduction processing on the initial information to obtain dimensionality-reduced information and sends it to the master control routing agent; The master control routing agent integrates the dimensionality reduction information according to the assigned weights and feeds back the integration result to the patient in the form of natural language.

2. The multi-agent-based psychological intervention method as described in claim 1, characterized in that, The initial information is natural language information, and the information entropy of the initial information is calculated using the following formula: ; in, The information entropy represents the initial information. This represents the set of semantic units extracted from the initial information using natural language processing methods. Represents the first in the set of semantic units i A semantic unit, Indicates the first i The probability of a semantic unit appearing in the context and in the patient's historical input information.

3. The multi-agent-based psychological intervention method as described in claim 2, characterized in that, The cognitive computing power gap is calculated using the following formula: ; in, This indicates a cognitive computing power gap. This represents the patient's baseline cognitive computing ability estimate.

4. The multi-agent-based psychological intervention method as described in claim 1, characterized in that, When the cognitive computing power gap is greater than a preset value, before the step of the master control routing agent waking up the corresponding sub-agent according to the gap characteristics of the cognitive computing power gap is executed, the method further includes: the master control routing agent performing semantic analysis on the initial information to determine the gap characteristics of the cognitive computing power gap; The steps for determining the gap characteristics of the cognitive computing power gap include: If the initial information contains words with emotional polarity but lacks specific event logic, the master control routing agent will determine the gap feature of the cognitive computing power gap as an emotional catharsis gap. If the initial information contains cognitive distortion, the master control routing agent will determine the gap feature of the cognitive computing power gap as a cognitive reconstruction gap; If the initial information contains keywords that trigger multiple types of alarms and are accompanied by stressor keywords, the master control routing agent will determine the gap characteristics of the cognitive computing power gap as a complex trauma gap.

5. The multi-agent-based psychological intervention method as described in claim 4, characterized in that, The plurality of sub-agents includes an empathy sub-agent, a cognitive behavioral therapy sub-agent, and a joint consultation sub-agent; the step by which the master control routing agent wakes up the corresponding sub-agent based on the gap characteristics of the cognitive computing power gap includes: When the cognitive computing power gap is characterized by the emotional catharsis gap, the main control routing agent wakes up the empathy sub-agent. When the cognitive computing power gap is characterized by the cognitive reconstruction gap, the master control routing agent wakes up the cognitive behavior therapy sub-agent; When the cognitive computing power gap is characterized by the complex trauma gap, the master control routing agent wakes up the joint consultation sub-agent.

6. The multi-agent-based psychological intervention method as described in claim 1, characterized in that, The multi-agent system also includes a review agent; Before the step of providing feedback to the patient in natural language, the following steps are also included: The review agent detects the integration result to determine whether the integration result contains sensitive words; If the integration result contains the sensitive words, the review agent triggers a blocking mechanism. If the integration result does not contain the sensitive words, the review agent instructs the master routing agent to feed back the integration result to the patient.

7. The multi-agent-based psychological intervention method as described in claim 6, characterized in that, The sensitive words include: words with self-harm or self-inflicted injury intentions, words with plans or tendencies to threaten others with violence, words with plans or tendencies to incite crime, and words that advise patients to discontinue professional psychiatric prescription drugs, or any combination thereof. The blocking mechanism includes: suspending the current process of the main control routing agent and calling staff to speak with the patient.

8. The multi-agent-based psychological intervention method as described in claim 1, characterized in that, If the patient responds to the integration results in natural language, the psychological intervention method further includes: The master control routing agent acquires the patient's response information and calculates the information entropy of the response information; The master control routing agent calculates the difference between the information entropy of the initial information and the response information to obtain the information entropy reduction. By using reinforcement learning algorithms and taking the maximization of information entropy reduction as the reward function, the distribution strategy of the master routing agent and the weights of each sub-agent are optimized.

9. The multi-agent-based psychological intervention method as described in claim 8, characterized in that, The expression for the reward function is as follows: ; in, Represents the reward function, Indicates the positive return coefficient. Indicates the decrease in information entropy. Indicates the penalty coefficient. This indicates the penalty value for safety violations.

10. A psychological intervention system, characterized in that, It includes a memory and multiple agents; the memory stores a computer program, and the multiple agents include a master routing agent and multiple sub-agents that are controlled and awakened by the master routing agent; when the multiple agents execute the program, they implement the method as described in any one of claims 1 to 9.

Images