A psychological test scale recommendation method and system based on multi-agent

Abstract

This invention belongs to the interdisciplinary field of artificial intelligence and psychology, and relates to a method and system for recommending psychological assessment scales based on multi-agent intelligence. The method includes: constructing a contextual state model using multi-dimensional information from the user during the consultation process; filtering scales according to the contextual state model to obtain candidate scales; calculating the fit between the candidate scales and the user's current psychological state, and setting the priority of the candidate scales; obtaining a recommendation score for the candidate scales based on the fit and priority; and determining the optimal scale based on the recommendation score. The system includes a contextual state management module, a dynamic scale recommendation module, a scale execution and scoring module, a risk detection and intervention module, and an auditing and logging module. This invention can intelligently recommend psychological assessment scales based on the user's state and actual needs, while emphasizing security and traceability, and is applicable to psychological health assessment and consultation in various scenarios.

Description

Technical Field

[0001] This invention provides a psychological assessment technology system that can intelligently recommend psychological assessment scales based on user status and actual needs, while emphasizing security and traceability. The technical solution of this invention overcomes the problems of existing technologies, such as fixed assessment scales, difficulty in timely risk detection, incomplete assessment process records, and limited range of selectable scales. It is applicable to mental health assessment and counseling systems in various scenarios. This invention belongs to the interdisciplinary field of artificial intelligence and psychology. Background Technology

[0002] Many psychological counseling platforms and psychological testing tools require users to fill out psychological scales, such as common tests for depression and anxiety. There are many types of psychological scales, and even different versions of the same scale, each with its own proven effectiveness in different scenarios. However, current assessment systems generally use a fixed-scale completion method. Users enter the system and directly fill out a pre-set, specific psychological scale (such as the PHQ-9 depression scale, GAD-7 anxiety scale, ADHD assessment scale, etc.). The system then calculates scores and generates results according to predetermined rules.

[0003] However, existing technologies have the following shortcomings: 1) Inflexible scale recommendations: Existing systems typically cannot automatically select the most suitable test scale based on the user's emotional changes or current needs during the conversation. 2) Difficulty in timely detection of high risks: When users exhibit self-harm tendencies or drastic emotional changes, most online psychological systems lack real-time identification and early warning mechanisms. 3) Existing systems lack necessary records regarding the operation of recommendation logic, the calculation basis for risk judgment, and the detailed trajectory of scale execution. Specifically, the decision-making steps of the recommendation logic are not retained; the norms, thresholds, and characteristic variables used for risk judgment are not recorded; and information such as the user's answering time, question jumps, interruption recovery, and system interactions during scale execution are not fully saved, making it difficult to meet the stringent requirements of privacy protection, compliance auditing, and traceability in the field of mental health. 4) Weak scalability: Existing systems typically only support a few psychological scales, making it difficult to flexibly expand as assessment needs increase.

[0004] Existing publicly available literature, such as CN 114944219 A and CN 117609486 A, mostly adopts a method based on user dialogue text → training model → recommendation scale, which makes it difficult to realize a dynamic evaluation process based on user status and needs, and also lacks identification and early warning mechanisms for high risks. Methods based on clustering algorithms or collaborative filtering recommendation algorithms, such as CN 112530598 B, CN 115994271 B, and CN 117352114 B, can only make recommendations on scales within a limited domain, lack generalization performance, and cannot match the user's actual needs.

[0005] Specifically, through analysis of existing psychological assessment system technical solutions, the following main shortcomings can be identified: 1. The recommendation logic is simplistic and lacks dynamic context adaptation. Existing technologies primarily rely on single user input or limited historical data for scale recommendations. The recommendation process lacks multi-turn dialogue and continuous modeling of user states, specifically manifested in: 1) a lack of contextual state modeling capabilities, making it impossible to adjust recommendation strategies based on historical user input and behavioral trends; 2) a lack of dynamic tracking of changes in user psychological states, leading to a mismatch between recommended scales and current user needs; and 3) a lack of comprehensive consideration of multi-dimensional psychological indicators, resulting in simplistic recommendation results and insufficient matching accuracy. The resulting technical challenge is: how to construct a recommendation mechanism that can continuously update user psychological states and dynamically select the most suitable scale based on the user's dialogue context, thereby improving the relevance and matching accuracy of scale recommendations.

[0006] 2. Insufficient High-Risk Behavior Identification and Emergency Response Capabilities. Most existing psychological assessment systems cannot detect potential psychological crises in users in real time. Specifically: 1) They lack real-time risk identification models, failing to provide timely warnings for extreme behaviors such as self-harm tendencies and severe emotional fluctuations; 2) They lack automatic intervention mechanisms, unable to provide immediate emergency psychological support or guide manual intervention when users exhibit high-risk behaviors; 3) They lack the ability to integrate local psychological assistance resources, unable to provide effective user assistance in high-risk situations. The resulting technical challenge is: how to analyze users' psychological risks in real time during the psychological assessment process and provide automated intervention and emergency response mechanisms to ensure user safety and system compliance.

[0007] 3. Lack of traceability and compliance. Psychological testing involves sensitive personal information and legal requirements, but existing technologies have the following shortcomings: 1) They cannot record the decision-making basis and algorithm logic of the system's recommended scales; 2) Risk assessment and intervention triggering processes are not traceable; 3) Assessment data and scoring processes lack complete logs, making secure traceability impossible. The resulting technical problem is: how to establish a secure logging mechanism to ensure the traceability and transparency of recommendation decisions, risk assessments, and intervention processes, and meet compliance requirements.

[0008] 4. Insufficient scalability, making it difficult to support multi-scale and multi-dimensional assessments. Existing systems are typically designed with fixed processes, making it difficult to flexibly integrate multiple scales or address different psychological counseling scenarios. Specifically: 1) The system only supports a small number of preset scales, making it difficult to add new scales or multi-dimensional assessments; 2) It has limited support for composite assessment processes driven by changes in user status, failing to achieve personalized assessment paths; 3) The system architecture has high coupling, making it costly to expand new functions or integrate new scenarios. The resulting technical challenge is: how to design a scalable scale library and modular assessment architecture to enable the system to flexibly support multi-scale, multi-dimensional assessments and personalized assessment processes.

[0009] Therefore, there is an urgent need for a psychological assessment system that can understand the context of the conversation, dynamically match assessment scales, detect risks in a timely manner, and be fully recordable and traceable. Summary of the Invention

[0010] To address the aforementioned problems, this invention provides a method and system for recommending psychological assessment scales based on multi-agent systems.

[0011] The technical solution adopted in this invention is as follows: A method for recommending psychological assessment scales based on multi-agent systems includes the following steps: Construct a contextual state model using multi-dimensional information from users during the consultation process; Candidate scales are obtained by filtering the scales based on the context state model; Calculate the fit between candidate scales and users' current psychological state, and set the priority of candidate scales; Recommendation scores for candidate scales are obtained based on fit and priority, and the optimal scale is determined based on the recommendation scores.

[0012] Furthermore, the context state model includes semantic information, behavioral information, assessment information, and scenario information; the step of filtering the scales according to the context state model includes: obtaining the user's current state and historical assessment results according to the context state model, and applying predefined filtering rules to exclude inapplicable or duplicate scales based on the user's current state and historical assessment results.

[0013] Furthermore, the calculation of the fit between the candidate scale and the user's current psychological state includes: Extract semantic features from user-input text using large-scale language models; Assess the user's current psychological risk status; Obtain users' historical assessment results and behavioral patterns from the contextual state model, and analyze historical behavioral trends. The fit is calculated based on semantic features, risk status, and historical behavioral trends.

[0014] Furthermore, the priority of the candidate scales is set according to business rules and the correlation and complementarity between the scales.

[0015] Furthermore, using the aforementioned optimal scale, the following steps are employed to guide users in conducting psychological assessments, thereby enabling the scale to be executed and scored: displaying the questions; receiving the user's submitted answers; calculating the score according to the scale rules; outputting the dimension scores and the total score; and feeding the results back to the context state model.

[0016] Furthermore, risk detection and intervention are conducted during the process of guiding users to conduct psychological assessments using the optimal scale; the risk detection and intervention include: The system analyzes users' current and historical emotional tendencies using natural language processing techniques and calculates changes in emotional tendencies. Detect risky keywords in user input and calculate risky keyword scores; By analyzing changes in the tone and expression patterns of user input, abrupt changes in tone or expression patterns can be obtained. Based on the user's historical evaluation scores, past evaluation scores are obtained; Detect changes in user behavior; The risk index is calculated by taking into account changes in emotional tendency, risk keyword scores, sudden changes in tone or expression patterns, past test scores, and behavioral changes. Multiple risk levels are set based on the risk index, and interventions are made in the psychological assessment process according to the risk level.

[0017] Furthermore, the above method records the basis for each scale recommendation, the trajectory of risk index changes, user input text, assessment answers and scoring rules, as well as emergency process triggering nodes and reasons. The recorded content will be used for compliance audits, risk event reviews, system optimization, or internal supervision scenarios in medical institutions.

[0018] A multi-agent-based psychological assessment scale recommendation system, comprising: The context state management module is used to build a context state model by utilizing multi-dimensional information from users during the consultation process. The dynamic scale recommendation module is used to filter scales based on the context state model to obtain candidate scales, calculate the fit between the candidate scales and the user's current psychological state, set the priority of the candidate scales, obtain the recommendation score of the candidate scales based on the fit and priority, and determine the optimal scale based on the recommendation score.

[0019] Furthermore, the aforementioned system also includes: The scale execution and scoring module is used to guide users to conduct psychological assessments using the optimal scale, thereby realizing the execution and scoring of the scale. The risk detection and intervention module is used to detect and intervene in the process of guiding users to conduct psychological assessments using the optimal scale. The audit and log module is used to record the basis for each scale recommendation, the trajectory of risk index changes, user input text, assessment answers and scoring rules, as well as emergency process triggering nodes and reasons. The recorded content will be used for auditing, risk event review, system optimization, or internal supervision scenarios in medical institutions.

[0020] Furthermore, each module in the system is an intelligent agent.

[0021] Through the aforementioned innovative technical means, this invention significantly improves the intelligence level, security, and usability of psychological assessment systems, with the following specific beneficial effects: 1) Improve the accuracy and personalization of scale recommendations.

[0022] Principles and mechanisms: Based on the context state model and dynamic scale recommendation mechanism, the system can continuously track the user's psychological state, historical behavior and emotional changes, calculate the scale fit and prioritize them.

[0023] Technical benefits: Compared with existing fixed-process systems, it can recommend scales that best meet the user's current psychological needs, avoid duplicate or irrelevant scales, and improve assessment efficiency and user satisfaction.

[0024] 2) Enhance the ability to identify psychological risks and handle emergencies.

[0025] Principle and mechanism: Through multi-factor risk index calculation and multi-level risk assessment mechanism, real-time monitoring of users' self-harm tendencies, extreme emotional fluctuations and other behaviors can be achieved.

[0026] Technical effect: When a user exhibits high-risk behavior, the system can immediately halt the process and trigger intervention measures (such as psychological hotline prompts and human customer service intervention), significantly reducing potential safety hazards during the psychological assessment process.

[0027] 3) Provide full-process traceability and audit capabilities.

[0028] Principles and mechanisms: The system records the scale recommendation logic, risk assessment basis, user input and intervention trigger nodes to form a complete log.

[0029] Technical benefits: It provides reliable data for compliance audits, risk reviews, and system optimization, meets regulatory requirements in the field of mental health services, and enhances system transparency and credibility.

[0030] 4) Enables multi-scale, multi-dimensional, and scalable assessments.

[0031] Principles and mechanisms: The modular scale execution architecture supports key-value pair storage of scale information, enabling decoupling and scalable operations.

[0032] Technical benefits: The system can flexibly integrate new scales, support multi-dimensional psychological assessments and personalized assessment pathways, and improve the coverage and applicability of psychological assessments.

[0033] 5) Enhance the overall intelligence level and application value of the system.

[0034] Principles and mechanisms: Contextual dynamic modeling, intelligent recommendation, risk intervention, and collaborative work of the audit module.

[0035] Technical benefits: Under the premise of ensuring security and compliance, the system realizes personalized psychological assessment and risk management, which is significantly better than existing systems with rigid processes, delayed risk response and poor scalability, thus improving the practical application value of mental health assessment and online counseling. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the steps of a multi-agent-based psychological assessment scale recommendation method according to an embodiment of the present invention.

[0037] Figure 2 This is a module composition diagram of a multi-agent psychological assessment scale recommendation system according to an embodiment of the present invention.

[0038] Figure 3 This is a module composition diagram of a multi-agent psychological assessment scale recommendation system according to another embodiment of the present invention.

[0039] Figure 4 This is a schematic diagram of the architecture of a multi-agent psychological assessment scale recommendation system according to another embodiment of the present invention. Detailed Implementation

[0040] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0041] This invention proposes a multi-agent-based method and system for recommending psychological scales. Its main objectives are: 1) to construct a user context state model to achieve dynamic tracking of psychological states and multi-round scale recommendation; 2) to design a risk detection and intervention module to achieve real-time identification and emergency handling of high-risk psychological behaviors; 3) to provide an audit and log recording mechanism to achieve traceability of recommendation decisions, risk analysis, and intervention processes; and 4) to construct a modular and scalable scale library and assessment architecture to achieve flexible expansion of multi-scale, multi-dimensional, and personalized assessments.

[0042] To achieve the aforementioned technical objectives, large-scale language model technology needs to be introduced. Furthermore, the workflow of this large-scale language model needs to be defined and constrained according to the requirements. Therefore, this invention employs multi-agent technology, where each agent corresponds to the invocation of one or more large-scale language models. This is organized through different prompts, different input / output constraints, and different workflow orchestration methods, thereby achieving efficient, controlled, interpretable, and reasonable scale recommendation functionality. By achieving these technical objectives, this invention overcomes the shortcomings of existing psychological assessment systems in terms of recommendation accuracy, security, compliance, and scalability, significantly improving the intelligence, reliability, and operability of psychological assessment systems in practical applications.

[0043] This invention aims to overcome the shortcomings of existing online psychological assessment systems, such as static scale recommendations, weak risk handling capabilities, lack of auditability, and poor scalability. To this end, this invention constructs a method that dynamically updates the user's psychological state based on real-time interactive behavior and automatically matches psychological scales according to the psychological state. It also provides risk monitoring, emergency intervention, and audit tracking capabilities to improve the security, accuracy, and scalability of psychological assessment systems in practical applications. This invention achieves continuous modeling and personalized assessment of the user's psychological state by designing a contextual state model, a dynamic scale recommendation mechanism, a scale execution and scoring mechanism, and a risk index calculation model. Under the premise of ensuring security and compliance, it provides intelligent assessment logic that conforms to the characteristics of psychological counseling scenarios.

[0044] In one embodiment of the present invention, a method for recommending psychological assessment scales based on multi-agent systems is provided, such as... Figure 1 As shown, it includes the following steps: Step S11: Construct a contextual state model using multi-dimensional information from the user during the consultation process; Step S12: Filter the scales according to the context state model to obtain candidate scales; Step S13: Calculate the fit between the candidate scales and the user's current psychological state, and set the priority of the candidate scales; Step S14: Obtain the recommended scores of candidate scales based on fitness and priority, and determine the optimal scale based on the recommended scores.

[0045] In one embodiment, the context state model in step S11 includes semantic information, behavioral information, evaluation information, and scene information.

[0046] In one embodiment, step S12, which involves filtering the scales based on the context state model, includes: obtaining the user's current state and historical assessment results based on the context state model; and applying predefined filtering rules to exclude inapplicable or duplicate scales based on the user's current state and historical assessment results.

[0047] In one embodiment, step S13, calculating the fit between the candidate scale and the user's current psychological state, includes: Extract semantic features from user-input text using large-scale language models; Assess the user's current psychological risk status; Obtain users' historical assessment results and behavioral patterns from the contextual state model, and analyze historical behavioral trends. The fit is calculated based on semantic features, risk status, and historical behavioral trends.

[0048] In one embodiment, setting the priority of candidate scales in step S13 is based on business rules and the correlation and complementarity between scales.

[0049] In one embodiment, the method further includes step S14: using the optimal scale to guide the user to conduct a psychological assessment, thereby implementing the scale's execution and scoring. Step S14 specifically includes the following steps: displaying the questions; receiving the user's submitted answers; calculating the score according to the scale rules; outputting the dimension scores and the total score; and feeding the results back to the context state model.

[0050] In one embodiment, the method further includes step S15: risk detection and intervention during the process of guiding users to conduct psychological assessments using the optimal scale. The risk detection and intervention specifically includes: analyzing the user's current and historical emotional tendencies using natural language processing technology and calculating changes in emotional tendencies; detecting risk keywords contained in the user's input and calculating risk keyword scores; obtaining abrupt changes in tone or expression patterns by analyzing changes in the user's input tone and expression patterns; obtaining past assessment scores based on the user's historical assessment scores; detecting changes in the user's behavior; calculating a risk index by comprehensively considering changes in emotional tendencies, risk keyword scores, abrupt changes in tone or expression patterns, past assessment scores, and behavioral changes; setting multiple risk levels based on the risk index; and intervening in the psychological assessment process according to the risk levels.

[0051] In one embodiment, the above method further includes step S16: recording the basis for each scale recommendation, the trajectory of risk index changes, the user's input text, assessment answers and scoring rules, and the emergency process triggering nodes and reasons, and using the recorded content for compliance audits, risk event reviews, system optimization, or internal supervision scenarios of medical institutions.

[0052] Another embodiment of the present invention provides a recommendation system for psychological assessment scales based on multi-agent systems, such as... Figure 2 As shown, it includes: Context state management module 21 is used to construct a context state model by utilizing multi-dimensional information from users during the consultation process; The dynamic scale recommendation module 22 is used to filter scales according to the context state model to obtain candidate scales, calculate the fit between the candidate scales and the user's current psychological state, set the priority of the candidate scales, obtain the recommendation score of the candidate scales based on the fit and priority, and determine the optimal scale based on the recommendation score.

[0053] In one embodiment, such as Figure 3 As shown, the above system also includes: The scale execution and scoring module 23 is used to guide users to conduct psychological assessments using the optimal scale, thereby realizing the execution and scoring of the scale. The risk detection and intervention module 24 is used to detect and intervene in the process of guiding users to conduct psychological assessments using the optimal scale. The Audit and Log module 25 is used to record the basis for each scale recommendation, the trajectory of risk index changes, user input text, assessment answers and scoring rules, as well as emergency process triggering nodes and reasons. The recorded content will be used for auditing, risk event review, system optimization, or internal supervision scenarios in medical institutions.

[0054] In one embodiment, this invention proposes a multi-agent-based psychological assessment scale recommendation system, whose core consists of modules such as context state management, dynamic scale recommendation, scale execution and scoring, risk detection and intervention, and auditing and logging. These modules, acting as agents, each have large-scale language models and programming implementations. Through collaboration among multiple agent modules, the system achieves modeling of the user's psychological state, dynamic arrangement of assessment paths, and security control of the assessment process. The overall technical framework of the solution is described below, and the combination and coordination of the various modules are as follows. Figure 4 As shown below, the specific implementation methods of each module in this embodiment are explained.

[0055] 1. Context state management module.

[0056] The context state management module plays a crucial role in this system, its function being to uniformly model the multi-dimensional information of users during the consultation process. This module constructs the context state model using the following information: 1) Semantic information: including the classification of user input text, emotional features, and intent inference.

[0057] 2) Behavioral information: including response speed, changes in expression, and keyword usage patterns.

[0058] 3) Assessment information: including completed scales, score data, and assessment trends.

[0059] 4) Scene information: including the evaluation stage, current task, page status, etc.

[0060] The contextual state model is continuously updated with each user input, allowing the entire assessment process to make decisions based on evolving psychological states. For example, when a user repeatedly expresses emotional characteristics such as "fatigue" or "lack of motivation" in the conversation, the system will automatically increase the risk weight associated with depression and adjust subsequent scale recommendations accordingly.

[0061] This module provides the system with highly structured user psychological profiles and serves as the foundation for subsequent recommendation and risk algorithms.

[0062] 2. Dynamic scale recommendation module.

[0063] The dynamic scale recommendation module is the core module in this invention. Compared with existing technologies, this module does not rely solely on model prediction, but combines a contextual state model to execute complex rules and matching logic. Its recommendation process includes: 1) Candidate scale screening: Filter out inapplicable or duplicate scales based on the user's current status and historical assessment results.

[0064] The specific steps for implementing filtering include: a) Extracting the user's current state: The system obtains the user's current state, including their emotional characteristics, psychological needs, and historical behavioral trends, using a contextual state model based on a large-scale language model. This state is stored in a structured attribute table. For example, if a user repeatedly expresses emotional characteristics such as "tired" or "lacking motivation" in a conversation, the system will automatically increase the risk weight associated with depression.

[0065] b) Extract historical assessment results: Obtain historical assessment results from the context state model, including completed scales and their scores.

[0066] c) Apply filtering rules: Based on the user's current status and historical assessment results, apply predefined filtering rules to exclude inapplicable or duplicate scales. For example, if a user has already completed the PHQ-9 depression scale and scored high, the system may exclude other depression-related scales and recommend scales for anxiety or other dimensions.

[0067] 2) Scale fit calculation: Calculate the degree of match between each candidate scale and the user's current psychological state.

[0068] 3) Risk level filtering: Based on real-time risk detection results, if a user's risk index reaches a preset high-risk threshold, self-administered questionnaires that may induce discomfort or require professional guidance will be avoided, and the user will be prioritized for manual intervention or recommended low-risk alternatives.

[0069] 4) Prioritization Strategy Application: Based on the correlation and complementarity between scales (e.g., covering different psychological dimensions, forming joint assessment groups), and the priority set by business rules, candidate scales are sorted or grouped. This module supports joint recommendation, meaning that multiple complementary scales can be output at once to form an assessment plan, reducing the burden of multiple assessments and improving coverage. Business rules refer to the rules formulated during the scale recommendation process based on the actual needs and professional knowledge of mental health assessment. These rules guide the scale fit calculation and priority ranking. For example, in one embodiment, business rules include: a) Covering different psychological dimensions: Recommended scales should cover as many psychological dimensions of users as possible (such as mood, anxiety, depression, attention, etc.) to provide a comprehensive mental health assessment; b) Forming a joint assessment group: The recommended scales should be able to form a joint assessment group, providing a more comprehensive assessment of mental state through the complementarity of multiple scales.

[0070] Priority refers to determining the recommendation order of multiple candidate rating scales based on business rules and the user's current state. For example, in one embodiment, priority can be set based on the following factors: a) Relationship and complementarity between scales: For example, some scales are complementary in assessing the emotion dimension and can be given priority; b) Priority settings for business rules: For example, based on the actual needs of mental health assessment, certain scales may be set as high priority and recommended to users first, such as the ADHD scale.

[0071] 5) Determine the optimal scale: The recommendation result is derived based on a combination of fit and priority.

[0072] The fit algorithm is a core method for evaluating the degree of match between candidate scales and a user's current psychological state. The fit algorithm can comprehensively consider factors such as semantic features, risk status, and historical behavioral trends. For example, when the system identifies significant emotional fluctuations in a user but not yet reaching a dangerous level, it will recommend stress or anxiety-related scales instead of directly proceeding to a high-risk intervention process. The fit algorithm comprehensively considers the following factors: a) Semantic features: Analyzing user input text through large-scale language models to extract information such as emotional features, psychological needs, and intent inferences; b) Risk status: Assessing the user's current psychological risk level based on real-time risk detection results; c) Historical behavioral trends: Analyzing the changing trends of the user's psychological state based on historical assessment results and behavioral patterns.

[0073] The specific calculation of fit can be performed using the following steps: 1) Semantic feature extraction: Use large-scale language models to extract semantic features from user input text, including emotional features (such as fatigue, lack of motivation), psychological needs (such as seeking comfort, hoping to improve mood), etc., and transform these features into a structured attribute table.

[0074] 2) Risk Status Assessment: Based on the risk index provided by the real-time risk detection module, assess the user's current psychological risk level. For example, when the risk index is low, the weight of risk-related factors in the compatibility score is low; when the risk index is high, the weight of risk-related factors in the compatibility score is high.

[0075] 3) Historical Behavioral Trend Analysis: Obtain users' historical assessment results and behavioral patterns from the contextual state model, and analyze their performance and trends in different psychological dimensions.

[0076] 4) Fit Calculation: Based on the above characteristics, a weighted scoring model is used to calculate the fit. For example, the fit score A can be expressed as: Where S represents the semantic feature score, R represents the risk status score, H represents the historical behavior trend score, and w1, w2 and w3 are the corresponding weights.

[0077] Priority scores are also calculated using a weighted summation method. Let C represent the relevance and complementarity score, and P represent the business rule score. The priority score Pr can be expressed as: Here, w4 and w5 are the corresponding weights.

[0078] Finally, by combining fit and priority through a weighted summation, the recommendation score S for each candidate scale is obtained: Where A represents the fit score, Pr represents the priority score, and w6 and w7 are the corresponding weights.

[0079] By using the above methods, the system can comprehensively consider suitability and priority, dynamically adjust the scale recommendation results, ensure that the scale recommendation can change dynamically with the user's status, avoid the limitation of fixed processes on users, and improve the relevance and effectiveness of the assessment.

[0080] 3. Scale implementation and scoring module.

[0081] The scale execution and scoring module is responsible for guiding users to complete psychological assessments in an orderly and structured manner. This module is based on a scalable scale library design, which uses a key-value pair structure to store scale information, including: item number, item text, options and corresponding scores, scoring method, and the corresponding dimension (e.g., attention dimension, emotion dimension). During the assessment process, the module automatically scores according to the scale rules, including: 1) displaying the items; 2) receiving the user's submitted answers; 3) calculating the score according to the scale rules; 4) outputting the dimension scores and total score; and 5) feeding the results back to the context state model. This module is designed to be decoupled from the recommendation and risk modules, allowing new scales to be added at any time without affecting the overall system structure.

[0082] 4. Risk detection and intervention module.

[0083] To ensure the safety of the psychological assessment process, this invention designs a risk detection and intervention module to analyze potential psychological crisis information in user input in real time. The risk index R(t) is updated by incorporating the following factors: 1) Sentiment Change (E(t)): This analyzes a user's current and historical sentiment tendencies using natural language processing techniques, quantifying the degree of emotional fluctuation. Sentiment change can be represented by the change in sentiment score. For example, sentiment analysis models (such as BERT-based sentiment classification models) or the sentiment discrimination capabilities of large-scale language models can be used to score the sentiment of user input text, with scores ranging from -1 (extremely negative) to 1 (extremely positive). Sentiment change E(t) can be expressed as the difference between the current sentiment score and the historical sentiment score: E(t) = Emotional Score t - Mood Score t-1 Where t and t-1 represent the emotion scores at the current moment and the previous moment, respectively.

[0084] 2) Specific Risk Keywords (K(t)): Detects whether user input contains specific high-risk keywords, such as "suicide" or "unbearable pain." Each occurrence of a keyword increases the risk index. The score for specific risk keywords is calculated using the keyword's weight and frequency of occurrence. in, This represents the weight of the i-th keyword. This represents the frequency of the i-th keyword at the current time t, where n is the total number of keywords.

[0085] 3) Changes in tone or expression pattern (T(t)): This involves analyzing changes in the tone and expression patterns of user input, such as speech rate, sentence length, and punctuation usage. Changes in tone or expression pattern T(t) can be quantified by the amount of change in linguistic features.

[0086] 4) Past Assessment Scores (S(t)): This combines the user's historical assessment scores, especially scores from scales related to mental health risk. Past assessment scores can be represented by a weighted average of historical scale scores. in, Let represent the score of the j-th scale at the current time t, and m be the total number of related scales.

[0087] 5) Behavioral Changes (B(t)): Detects changes in user behavior, such as sudden abnormal response times or frequent interruptions. Behavioral changes B(t) can be quantified by the degree of abnormality of the behavioral characteristics.

[0088] Considering the above five factors, the formula for calculating the risk index R(t) can be expressed as: Among them, α, β, γ, δ and These are the weighting coefficients of each factor. These weighting coefficients can be adjusted according to the actual application scenario and expert experience to reflect the relative importance of each factor to the risk index.

[0089] The system sets multiple levels based on the risk index: 1) Low risk (R(t)≤R low When the risk index is lower than or equal to the preset low-risk threshold R low At this point, the system considers the user to be in a low-risk state and continues the normal assessment process. For example, it can be set to 0.3; 2) Medium risk (R) low < R(t)≤R high When the risk index is between the low-risk threshold R low and high-risk threshold R high During this period, the system considers the user to be in a medium-risk state and recommends enhanced monitoring of relevant scales, such as increasing the frequency of recommendations for mental health-related scales or reminding the user to pay attention to their own mental health. For example, R high It can be set to 0.8; 3) High risk (R) high < R(t) :When the risk index is higher than the preset high-risk threshold R high If the system determines that the user is in a high-risk state, it will automatically terminate the current assessment process and trigger emergency measures, such as prompting the user to call the local psychological assistance hotline or guiding human customer service to intervene, to ensure the user's safety.

[0090] This module ensures that the system has critical security mechanisms in psychologically sensitive environments.

[0091] 5. Audit and Log Module.

[0092] To meet legal requirements and industry standards for psychological service systems, this invention provides an audit and log module that records the following: 1) the basis for each scale recommendation; 2) the trajectory of risk index changes; 3) user input text; 4) assessment answers and scoring rules; and 5) emergency process trigger points and reasons. These records can be used for compliance audits, risk event reviews, system optimization, and internal supervision within medical institutions, ensuring the system's traceability and transparency.

[0093] The main innovations of this invention compared to existing technologies are reflected in the following aspects: 1) Dynamic modeling of context state.

[0094] Innovation: A user context state model is constructed by using multi-dimensional information (semantics, behavior, assessment, and scenario) and updated in real time with user interaction.

[0095] Technological innovation: It enables continuous tracking of psychological states in multi-turn dialogues, providing dynamic references for scale recommendations and solving the problems of existing systems having simplistic recommendation logic and lacking context adaptation.

[0096] 2) Dynamic scale adaptive recommendation mechanism.

[0097] Innovation: By combining contextual state models with scale fit calculations, dynamic and personalized scale recommendations can be achieved.

[0098] Technological innovation: The introduction of a comprehensive algorithm that combines candidate scale screening, fit calculation, risk filtering, and priority strategy enables the recommendation results to be adjusted according to changes in user status, overcoming the shortcomings of existing technologies such as fixed processes and low recommendation accuracy.

[0099] 3) Real-time risk detection and intelligent intervention.

[0100] Innovation: The design of a multi-level real-time risk monitoring module based on a risk index can identify high-risk behaviors such as self-harm tendencies and emotional abnormalities, and automatically trigger intervention measures.

[0101] Technological innovation: By combining emotional changes, keywords, tone patterns, behavioral characteristics, and historical test scores to calculate a risk index, the system can provide immediate response and security for high-risk behaviors, a capability lacking in existing technologies.

[0102] 4) Auditable records of decisions and interventions.

[0103] Innovation: Establishing a complete audit log mechanism to record the basis for scale recommendations, changes in risk index, user input, scoring process, and intervention trigger points.

[0104] Technological innovation: It provides full-process traceability and transparency, meets the regulatory and compliance requirements of mental health services, and solves the problem that existing systems cannot audit.

[0105] 5) A modular and scalable scale execution architecture based on multi-agent systems.

[0106] Innovation: The design incorporates an scalable scale library and decoupled modules, including scale execution, scoring, and feedback mechanisms, supporting multi-scale, multi-dimensional, and personalized assessments.

[0107] Technological innovation: The system architecture achieves low coupling and high scalability, allowing for flexible addition of new scales or expansion of psychological assessment scenarios, overcoming the shortcomings of weak scalability in existing technologies.

[0108] The modules of the technical solution provided by this invention have multiple alternatives, including but not limited to: 1) Alternative models for risk index: Sentiment models, Bayesian models, neural networks, etc. can be used as alternatives.

[0109] 2) Scale fit function alternative: Decision trees, rule sets, or deep learning models can be used.

[0110] 3) Semantic parsing alternatives: Keyword matching, traditional NLP, vector models, etc. can be used.

[0111] 4) Emergency response alternatives: can be connected to psychological counseling platforms, notification of users' family members, telephone call centers, etc.

[0112] It should be understood that the methods and systems disclosed in the above embodiments of this invention can be implemented in other ways. For example, the above module division can be implemented in other ways, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. The various steps and modules in this invention can be implemented in the form of software functional units and can be stored in a computer-readable storage medium, including several instructions to cause a computer device to execute some or all of the steps of the method described in this invention.

[0113] For example, one embodiment of the present invention provides a computer device (computer, server, etc.) including a memory and a processor. The memory stores a computer program configured to be executed by the processor. The computer program includes instructions for performing the steps of the method of the present invention. Another embodiment of the present invention provides a computer-readable storage medium (such as ROM / RAM, disk, optical disk, etc.) storing a computer program that, when executed by a computer, implements the steps of the method of the present invention. Yet another embodiment of the present invention provides a computer program product including a computer program that, when executed by a computer, implements the steps of the method of the present invention.

[0114] The specific embodiments of the present invention disclosed above are intended to help understand the content of the present invention and to implement it accordingly. Those skilled in the art will understand that various substitutions, changes, and modifications are possible without departing from the spirit and scope of the present invention. The present invention should not be limited to the content disclosed in the embodiments of this specification; the scope of protection of the present invention is defined by the claims.

Claims

1. A method for recommending psychological assessment scales based on multi-agent systems, characterized in that, Includes the following steps: Construct a contextual state model using multi-dimensional information from users during the consultation process; Candidate scales are obtained by filtering the scales based on the context state model; Calculate the fit between candidate scales and users' current psychological state, and set the priority of candidate scales; Recommendation scores for candidate scales are obtained based on fit and priority, and the optimal scale is determined based on the recommendation scores.

2. The method according to claim 1, characterized in that, The context state model includes semantic information, behavioral information, evaluation information, and scene information; The step of filtering scales based on the context state model includes: obtaining the user's current state and historical assessment results based on the context state model, and applying predefined filtering rules to exclude inapplicable or duplicate scales based on the user's current state and historical assessment results.

3. The method according to claim 1, characterized in that, The calculation of the fit between the candidate scale and the user's current psychological state includes: Extract semantic features from user-input text using large-scale language models; Assess the user's current psychological risk status; Obtain users' historical assessment results and behavioral patterns from the contextual state model, and analyze historical behavioral trends. The fit is calculated based on semantic features, risk status, and historical behavioral trends.

4. The method according to claim 1, characterized in that, The priority of the candidate scales is determined based on business rules and the relationships and complementarities between the scales.

5. The method according to claim 1, characterized in that, Using the aforementioned optimal scale, the following steps are employed to guide users in conducting psychological assessments, thereby enabling the scale to be executed and scored: displaying the questions; receiving the user's submitted answers; calculating the score according to the scale rules; outputting the dimension scores and the total score; and feeding the results back to the context state model.

6. The method according to claim 5, characterized in that, Risk detection and intervention are conducted during the process of guiding users to conduct psychological assessments using the optimal scale; the risk detection and intervention include: The system analyzes users' current and historical emotional tendencies using natural language processing techniques and calculates changes in emotional tendencies. Detect risky keywords in user input and calculate risky keyword scores; By analyzing changes in the tone and expression patterns of user input, abrupt changes in tone or expression patterns can be obtained. Based on the user's historical evaluation scores, past evaluation scores are obtained; Detect changes in user behavior; The risk index is calculated by taking into account changes in emotional tendency, risk keyword scores, sudden changes in tone or expression patterns, past test scores, and behavioral changes. Multiple risk levels are set based on the risk index, and interventions are made in the psychological assessment process according to the risk level.

7. The method according to claim 6, characterized in that, Record the basis for each scale recommendation, the trajectory of risk index changes, user input text, assessment answers and scoring rules, as well as emergency process trigger points and reasons. The recorded content will be used for compliance audits, risk event reviews, system optimization, or internal supervision scenarios in medical institutions.

8. A recommendation system for psychological assessment scales based on multi-agent systems, characterized in that, include: The context state management module is used to build a context state model by utilizing multi-dimensional information from users during the consultation process. The dynamic scale recommendation module is used to filter scales based on the context state model to obtain candidate scales, calculate the fit between the candidate scales and the user's current psychological state, set the priority of the candidate scales, obtain the recommendation score of the candidate scales based on the fit and priority, and determine the optimal scale based on the recommendation score.

9. The system according to claim 8, characterized in that, Also includes: The scale execution and scoring module is used to guide users to conduct psychological assessments using the optimal scale, thereby realizing the execution and scoring of the scale. The risk detection and intervention module is used to detect and intervene in the process of guiding users to conduct psychological assessments using the optimal scale. The audit and log module is used to record the basis for each scale recommendation, the trajectory of risk index changes, user input text, assessment answers and scoring rules, as well as emergency process triggering nodes and reasons. The recorded content will be used for auditing, risk event review, system optimization, or internal supervision scenarios in medical institutions.

10. The system according to claim 8 or 9, characterized in that, Each module in the system is an intelligent agent.

Images