A psychological portrait generation and closed-loop intervention system based on multi-source data aggregation

Abstract

The application discloses a psychological portrait generation and closed-loop intervention system based on multi-source data aggregation, and relates to the technical field of human-computer interaction, in particular to a psychological portrait generation and closed-loop intervention system based on multi-source data aggregation.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction technology, and in particular to a psychological profile generation and closed-loop intervention system based on multi-source data aggregation. Background Technology

[0002] Corporate psychological services currently face two major pain points: 1) Data silos, unrealized value: Services are still mainly based on individual consultations and single assessments, with multi-source data (logs, dialogues, scales) displayed in a scattered manner, lacking cross-source fusion analysis. Management cannot gain insight into the stress structure of groups and high-risk groups from surface indicators; 2) The dilemma of compliance and analysis: Group profiling is constrained by privacy, simple aggregation lacks risk identification capabilities, and excessive segmentation is prone to leaking individual information, making it difficult to form a compliant and effective intervention loop.

[0003] There is an urgent need for a technical method that can achieve multi-source data fusion, generate effective group profiles, and provide executable intervention plans and effect tracking under de-identification conditions. Summary of the Invention

[0004] This invention aims to address at least one of the technical problems existing in the prior art. To this end, one objective of this invention is to propose a psychological profiling generation and closed-loop intervention system based on multi-source data aggregation, which balances privacy protection and insightful analysis.

[0005] The psychological profile generation and closed-loop intervention system based on multi-source data aggregation according to an embodiment of the present invention includes: a de-identified data access module, a data processing module, a risk identification module, an intervention closed-loop module, and an output module. The output terminal of the de-identified data access module is electrically connected to the input terminal of the data processing module; the output terminal of the data processing module is electrically connected to the input terminal of the risk identification module; the output terminal of the risk identification module is electrically connected to the input terminal of the intervention closed-loop module; and the output terminal of the intervention closed-loop module is electrically connected to the input terminal of the output module.

[0006] The de-identified data access module is used to acquire and de-identify multi-source mental health data;

[0007] The data processing module is used to perform time alignment, field mapping, and group aggregation on the de-identified data to generate group characteristics;

[0008] The risk identification module is used to calculate the risk index based on group characteristics and identify key groups of concern;

[0009] The intervention closed-loop module is used to generate intervention suggestions based on the identification results and to complete the effect evaluation and data write-back.

[0010] The output module is used to output group profiles and intervention-related information.

[0011] The psychological profile generation and closed-loop intervention system based on multi-source data aggregation according to embodiments of the present invention can achieve the following technical effects by employing this method and system:

[0012] First, this invention uses a de-identified data access module and a data processing module to align individual psychological service data (such as chat summaries, emotion assessment indicators, AI usage logs, etc.) through time alignment, field mapping, and group aggregation to generate group characteristics (such as average stress, help-seeking rate, topic drift indicators, etc.), enabling customers to upgrade from "looking at personal problems" to "looking at group trends," and realizing the transformation of individual-level data into group management insights.

[0013] Second, the present invention performs user identifier replacement, sensitive information deletion and group sample number threshold masking (k≥10) in the de-identified data access module, and realizes time unification and field mapping fusion of multi-source heterogeneous data through the data processing module. Under the premise of strictly protecting individual privacy, it outputs group-level analysis results, taking into account both privacy security and insight value.

[0014] Third, this invention not only uses a weighted linear summation model to calculate the group risk index Rg through the risk identification module, and identifies key groups of concern based on the Rg value, duration and rate of change; it can also identify thematic key groups (with moderate overall risk but a sudden increase in a single thematic indicator), and monitor thematic changes by combining the thematic drift indicator Drift(t), and output early warning signals, making management actions more targeted and effective.

[0015] Fourth, this invention generates intervention suggestions (such as themed activities, work adjustments, content pushes, etc.) through the intervention closed-loop module, and completes the effect evaluation by comparing the changes in Rg before and after the intervention. The evaluation results are written back to the system, forming a continuous service closed loop of "identifying problems - making suggestions - verifying results". At the same time, the output module provides group profile reports and structured data packages, supporting the periodic iteration of BI dashboards and management monthly reports.

[0016] According to some embodiments of the present invention, multi-source mental health data includes at least one of: chat summaries, emotion assessment indicators, AI usage logs, help-seeking behavior records, and organizational structure metadata;

[0017] The desensitization process includes: user identifier replacement, sensitive information deletion, and output masking when the number of group samples is less than a threshold k, where k≥10.

[0018] According to some embodiments of the present invention, the data processing module includes a time alignment and field mapping unit and a group feature construction unit. The time alignment and field mapping unit is used to unify time by week, month or quarter, and map heterogeneous fields into stress, mood, sleep and behavior-related feature clusters.

[0019] According to some embodiments of the present invention, the group feature construction unit is used to perform hierarchical aggregation according to at least one dimension of department, job level, region, project group, and time window to form a multi-level group unit, and only outputs the group-level analysis results.

[0020] According to some embodiments of the present invention, the group characteristics include: mean stress level and percentage of high stress, mean sleep level and percentage of low sleep, mean mood, help-seeking rate, AI activity level, topic frequency, topic drift index, and percentage of sustained high stress. The topic drift index is calculated using the following formula:

[0021] In the formula: Let t be the theme drift index for period t; Let t be the frequency of the theme in the t-th period; The frequency of the theme in the (t-1)th period.

[0022] According to some embodiments of the present invention, the risk identification module uses a weighted linear summation model to calculate the group risk index. :

[0023]

[0024] In the formula: For group risk index; The total number of group characteristics; The preset weights for the i-th feature; The value of the i-th feature is standardized; the risk identification module is based on... Values, duration, and rate of change are used to identify key target groups.

[0025] According to some embodiments of the present invention, the risk identification module is also used to identify thematic key groups, where the overall risk of thematic key groups is moderate, but a single risk theme indicator shows a sudden increase, and a warning signal is output separately after identification.

[0026] According to some embodiments of the present invention, the intervention suggestions generated by the intervention closed-loop module include at least one of the following: themed activities, management training, job adjustments, content delivery, offline workshops, and special care; and by comparing before and after the intervention... Evaluation of the effectiveness of the change and data write-back.

[0027] According to some embodiments of the present invention, the output module outputs a group profile report and a structured data packet. The group profile report includes: risk level, core indicators, and period. Comparison, key themes, key groups, intervention recommendations, and feedback on effects.

[0028] According to some embodiments of the present invention, structured data packages are used to interface with BI dashboards, monthly management reports, and special management reports to achieve data visualization and periodic iteration.

[0029] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0030] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0031] Figure 1 This is an overall flowchart of an embodiment of the present invention;

[0032] Figure 2 This is a flowchart of data processing in an embodiment of the present invention;

[0033] Figure 3 This is the signal transmission direction pattern for generating the psychological profile in this embodiment of the invention;

[0034] Figure 4 This is a signal transmission diagram of the risk identification module in an embodiment of the present invention;

[0035] Figure 5 This is a flowchart of the intervention closed loop and module output in an embodiment of the present invention;

[0036] Figure 6 This is a flowchart illustrating how demand pattern routing, RAG, memory, and security policies jointly determine a round of responses in an embodiment of the present invention.

[0037] Figure 7 This is a flowchart of the speech continuous emotion recognition and intervention route in an embodiment of the present invention;

[0038] Figure 8 This is a flowchart illustrating how a personal psychological knowledge base simultaneously serves the next round of conversation, user reports, and organizational-level aggregation in an embodiment of the present invention.

[0039] Figure 9 This is a flowchart in an embodiment of the present invention showing how the assessment and the thermometer enter a closed loop of AI interpretation and subsequent courses / retests. Detailed Implementation

[0040] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0041] refer to Figure 1-9 According to an embodiment of the present invention, a psychological profile generation and closed-loop intervention system based on multi-source data aggregation includes: a de-identified data access module, a data processing module, a risk identification module, an intervention closed-loop module, and an output module. The output terminal of the de-identified data access module is electrically connected to the input terminal of the data processing module; the output terminal of the data processing module is electrically connected to the input terminal of the risk identification module; the output terminal of the risk identification module is electrically connected to the input terminal of the intervention closed-loop module; and the output terminal of the intervention closed-loop module is electrically connected to the input terminal of the output module.

[0042] The de-identified data access module is used to acquire and de-identify multi-source mental health data;

[0043] The data processing module is used to perform time alignment, field mapping, and group aggregation on the de-identified data to generate group characteristics;

[0044] The risk identification module is used to calculate the risk index based on group characteristics and identify key groups of concern;

[0045] The intervention closed-loop module is used to generate intervention suggestions based on the identification results and to complete the effect evaluation and data write-back.

[0046] The output module is used to output group profiles and intervention-related information.

[0047] The psychological profile generation and closed-loop intervention system based on multi-source data aggregation according to embodiments of the present invention can achieve the following technical effects by employing this method and system:

[0048] First, this invention uses a de-identified data access module and a data processing module to align individual psychological service data (such as chat summaries, emotion assessment indicators, AI usage logs, etc.) through time alignment, field mapping, and group aggregation to generate group characteristics (such as average stress, help-seeking rate, topic drift indicators, etc.), enabling customers to upgrade from "looking at personal problems" to "looking at group trends," and realizing the transformation of individual-level data into group management insights.

[0049] Second, the present invention performs user identifier replacement, sensitive information deletion and group sample number threshold masking (k≥10) in the de-identified data access module, and realizes time unification and field mapping fusion of multi-source heterogeneous data through the data processing module. Under the premise of strictly protecting individual privacy, it outputs group-level analysis results, taking into account both privacy security and insight value.

[0050] Third, this invention not only uses a weighted linear summation model to calculate the group risk index Rg through the risk identification module, and identifies key groups of concern based on the Rg value, duration and rate of change; it can also identify thematic key groups (with moderate overall risk but a sudden increase in a single thematic indicator), and monitor thematic changes by combining the thematic drift indicator Drift(t), and output early warning signals, making management actions more targeted and effective.

[0051] Fourth, this invention generates intervention suggestions (such as themed activities, work adjustments, content pushes, etc.) through the intervention closed-loop module, and completes the effect evaluation by comparing the changes in Rg before and after the intervention. The evaluation results are written back to the system, forming a continuous service closed loop of "identifying problems - making suggestions - verifying results". At the same time, the output module provides group profile reports and structured data packages, supporting the periodic iteration of BI dashboards and management monthly reports.

[0052] The specific work process is as follows:

[0053] 1) Data access and de-identification processing

[0054] The de-identified data access module first acquires multi-source mental health data (including chat summaries, emotion assessment indicators, AI usage logs, help-seeking behavior records, organizational structure metadata, etc.) and performs de-identification processing on the data: it performs user identifier replacement and deletes sensitive information, while setting a threshold k for the number of group samples (k≥10). When the number of samples in a certain group is less than k, the output is blocked to ensure that individual privacy is irreversible.

[0055] 2) Data processing and population feature generation

[0056] The anonymized data is then sent to the data processing module. This module first performs time alignment (unifying the time granularity by week, month, or quarter) and field mapping (mapping heterogeneous fields into clusters of related features such as stress, mood, sleep, and behavior). Then, it performs hierarchical aggregation by dimensions such as department, job level, region, project team, and time window to generate multi-layered group characteristics, including mean stress and percentage of high-pressure individuals, mean sleep and percentage of low-sleep individuals, mean mood, help-seeking rate, AI activity, topic frequency, and topic drift index. This includes factors such as the proportion of sustained high-pressure situations. This stage only outputs group-level analysis results and does not involve individual information.

[0057] 3) Risk identification and key group determination

[0058] Group characteristic data is fed into the risk identification module. This module uses a weighted linear summation model to calculate the group risk index. Based on the Rg value, duration, and rate of change, it identifies high-risk or continuously deteriorating key groups of concern; it also identifies "theme-based key groups" (overall risk is moderate but a single risk theme indicator shows a sudden increase), and outputs a separate early warning signal.

[0059] 4) Intervention suggestion generation and closed-loop write-back

[0060] The risk identification results are sent to the intervention closed-loop module. Based on the identified key groups and risk themes, this module generates targeted intervention recommendations, including but not limited to: themed activities, management training, job adjustments, content delivery, offline workshops, and special care. After the intervention is implemented, the module evaluates the effectiveness by comparing the changes in the group's risk index Rg before and after the intervention, and writes the evaluation results and intervention records back to the system's internal data storage for use in subsequent cycles.

[0061] 5) Integration of results output with visualization

[0062] The output module receives the results from the intervention closed-loop module and generates a group profile report and a structured data package. The group profile report includes: risk level, core indicators, periodic Rg comparison, key themes, key groups, intervention recommendations, and effect feedback information. The structured data package can be integrated with BI dashboards, monthly management reports, and special management reports to achieve data visualization and periodic iteration, completing a continuous service closed loop from "identifying problems—making recommendations—verifying results".

[0063] The present invention will now be described in detail.

[0064] 1) Access to and desensitization of multi-source mental health data

[0065] According to some embodiments of the present invention, the system first accesses multi-source heterogeneous mental health data, which includes at least one of the following: chat summaries, emotion assessment indicators, AI usage logs, help-seeking behavior records, and organizational structure metadata. Specifically, chat summaries are unstructured text information generated by the user's interaction with the mental health service system, covering core content such as the user's emotional expression, description of psychological distress, and expression of needs; emotion assessment indicators are quantitative results output by standardized psychological assessment scales (such as the SCL-90 and the Self-Rating Anxiety Scale SAS), accurately reflecting the user's psychological states such as stress, anxiety, and sleep; AI usage logs are used to record in detail the frequency of the user's use of the mental health service system, the duration of each use, and the user's preference for functional modules (such as emotion assessment, AI-assisted expression, and psychological science popularization); help-seeking behavior records include the timing, frequency, and type of help initiated by the user (such as online consultation, emergency intervention, and resource application); and organizational structure metadata includes organizational dimension information for group segmentation, such as the user's department, job level, work location, project team, and employment type, which is only used for group feature aggregation and is not associated with individual identity.

[0066] To strictly meet privacy compliance requirements (in accordance with the relevant provisions of the Personal Information Protection Law and the Data Security Law), the system performs three-level anonymization processing on the aforementioned multi-source data. This anonymization process includes: user identifier replacement, sensitive information deletion, and output masking when the group sample size is less than a threshold k (where k ≥ 10). Specifically, user identifier replacement replaces unique identifiers directly associated with natural persons (such as employee ID, mobile phone number, and national ID number) with irreversible anonymous identifiers generated by the system, achieving complete de-identification of individual identities. Sensitive information deletion uses a semantic recognition model combined with a preset keyword library to filter and remove privacy content that can identify individuals (such as names, specific project names, family information, and descriptions with extreme emotional connotations) from chat summaries and behavior records. The group sample size masking mechanism automatically masks all analysis results for a given dimension when the number of valid samples in a particular aggregation dimension is less than the threshold k, preventing the inference of individual information from small sample groups and further strengthening the security and compliance of privacy protection.

[0067] 2) Specific implementation of the data processing module

[0068] The data processing module includes a time alignment and field mapping unit and a group feature construction unit. The two work together to achieve standardized processing of multi-source heterogeneous data, unified feature mapping, and group-level feature aggregation, providing high-quality data support for subsequent risk identification.

[0069] 2.1) Time Alignment and Field Mapping

[0070] The time alignment and field mapping unit unifies the processing of multi-source heterogeneous data, solving the problem of inconsistent collection frequencies and field formats among different data sources. On the one hand, it aligns all data according to the company's regular management cycle (weekly, monthly, or quarterly), ensuring that data from different sources and with different collection frequencies (such as daily AI usage logs and monthly sentiment assessment data) fall into a preset time window, thus ensuring the consistency of the data's time dimension. On the other hand, through a preset rule engine combined with a lightweight machine learning model, it maps and aggregates multi-source heterogeneous fields into four standardized feature clusters: stress feature cluster, emotion feature cluster, sleep feature cluster, and behavior feature cluster. This enables the fusion and expression of multi-source data under a unified feature system, laying the foundation for subsequent group feature construction. The stress feature cluster includes fields such as stress assessment score, frequency of high-pressure statements, and frequency of seeking help; the emotion feature cluster includes fields such as emotion assessment score, percentage of positive and negative emotions, and frequency of emotion fluctuations; the sleep feature cluster includes fields such as sleep quality assessment score, frequency of insomnia-related statements, and records of abnormal sleep patterns; and the behavior feature cluster includes fields such as AI usage activity, help-seeking rate, interaction duration, and preference for using functional modules.

[0071] 2.2) Group Feature Construction

[0072] The group feature construction unit is hierarchically aggregated according to at least one dimension of department, job level, region, project team, and time window to form multi-layer nested group units (such as "East China Region - R&D Department - Core Project Team - Quarterly" and "Headquarters - Human Resources Department - Supervisor Job Level - Monthly"). The system only outputs group-level statistical results and does not output any individual detailed data, thus completely avoiding the leakage of individual privacy from the mechanism.

[0073] The specific characteristics of the constructed group include: mean stress level and percentage of people under high stress (defined as those whose stress assessment scores exceed a preset threshold), mean sleep level and percentage of people with low sleep quality (defined as those whose sleep assessment scores are below a preset threshold), mean mood, help-seeking rate (the proportion of people seeking help to the total number of people in the group), AI activity (the average frequency of users using AI psychological services within the group), topic frequency (the total number of times risk topics appear within the group), topic drift index, and percentage of people under continuous high stress (the proportion of people who are under high stress for two or more consecutive periods).

[0074] Among them, the theme drift index is used to quantify the intensity of change of risk themes over two consecutive periods, and its expression is shown in formula (1):

[0075]

[0076] In the formula:

[0077] Drift(t) is the theme drift index of a certain risk theme in period t;

[0078] freq(t) represents the frequency of occurrence of this risk theme within period t;

[0079] freq(t−1) represents the frequency of occurrence of the same risk theme within the t−1 period.

[0080] When Drift(t) > 1.5, it indicates that the attention to this risk topic is showing a significant upward trend; when 0.5 < Drift(t) ≤ 1.5, it indicates that the attention to this risk topic is basically stable; when Drift(t) ≤ 0.5, it indicates that the attention to this risk topic is showing a significant downward trend. This indicator can effectively identify risk topics that are rising rapidly and clustering locally, providing accurate quantitative basis for early warning.

[0081] 3) Specific implementation of the risk identification module

[0082] 3.1) Calculation of Group Risk Index

[0083] The risk identification module uses a weighted linear summation model to calculate the group risk index Rg. This model can integrate multi-dimensional group characteristics and accurately quantify the group's mental health risk level. The calculation formula is shown in formula (2):

[0084]

[0085] In the formula:

[0086] This is the comprehensive mental health risk index for the current group unit, with a value range of [0, 100]. The higher the value, the higher the mental health risk of the group.

[0087] n is the total number of group characteristics participating in the risk calculation. In this embodiment, n=8, which means all the above-mentioned group characteristics.

[0088] The preset weight coefficient corresponding to the i-th feature represents the degree of contribution of this feature to the overall risk. The total weight is 1, where the weight of the proportion of the high-pressure population is 1. Continuous high pressure ratio weight Request for help rate weight Theme drift index weight Sentiment mean weight Weighting of low sleep quality AI activity weight Topic frequency weight ;

[0089] is the dimensionless value of the i-th feature after Z-score standardization. The standardized value ranges from [0,1], which is used to eliminate the influence of different feature dimensions and ensure the rationality and accuracy of the risk index calculation.

[0090] 3.2) Focus on group identification

[0091] The system constructs a three-tiered identification rule based on the magnitude, duration, and rate of change of the group risk index Rg to identify high-risk groups of concern: Level 1 risk (Rg≥80) classifies a group as high-risk, requiring immediate emergency intervention; Level 2 risk (70≤Rg<80) classifies a group as medium-to-high-risk, requiring targeted intervention; and Level 3 risk (60≤Rg<70) classifies a group of concern, requiring continuous monitoring. Specifically, the identification rule is as follows: when the Rg of a group unit is ≥70, and the risk duration exceeds two cycles, or the Rg increase in a single cycle exceeds 30%, it is identified as a high-risk group of concern.

[0092] Meanwhile, the system can accurately identify key groups of concern based on specific themes: the overall risk index of this group is in the medium range (40≤Rg<70), but a single risk theme indicator shows a significant surge (theme drift index Drift(t)>2.0), indicating a localized clustering risk. The system separately marks this group and outputs a special early warning signal, clearly defining the warning theme and the scope of risk clustering, achieving refined and differentiated risk identification, and avoiding overlooking localized risk hazards.

[0093] 4) Specific implementation of the intervention closed-loop module

[0094] The intervention closed-loop module generates intervention recommendations that are precisely matched with the risk characteristics of the group based on the risk identification results (risk level, risk type, core risk theme). The intervention recommendations include at least one or a combination of the following: thematic activities, management training, work adjustments, content delivery, offline workshops, and special care, to ensure that the intervention measures are targeted and feasible.

[0095] The system performs a full-process quantitative evaluation of the intervention measures, forming a complete closed loop of "risk identification - intervention implementation - effect evaluation - strategy optimization": First, it records the time nodes, intervention measures, and coverage of the intervention; second, it compares the group risk index Rg of the same group unit before and after the intervention, calculates the change (ΔRg = Rg before intervention - Rg after intervention). If ΔRg ≥ 10%, the intervention is considered effective; if 5% ≤ ΔRg < 10%, the intervention is considered basically effective and the intervention measures need to be optimized; if ΔRg < 5% or ΔRg is negative, the intervention is considered ineffective and the intervention plan needs to be reformulated; finally, the intervention effect evaluation results and optimization suggestions are written back to the system data layer for group characteristic weight optimization and intervention strategy iteration, continuously improving the intervention effect.

[0096] 5) Specific implementation of the output module

[0097] The output module outputs two types of standardized results, catering to both management decision-making needs and system integration requirements, as detailed below:

[0098] (1) Group Profile Report: Generated by group unit, using a standardized template, with complete content and clear logic, specifically including: basic information of the group (department, time window, effective sample size), group risk level and Rg value, statistics and analysis of core characteristic indicators, comparison of Rg changes within the period (month-on-month, year-on-year), core risk themes and changing trends, list of key groups to pay attention to (including risk level and aggregation range), list of targeted intervention suggestions (including implementation priority and execution path), and historical intervention effect feedback information. The report only contains group-level data and does not involve any individual information.

[0099] (2) Structured data package: It adopts common formats such as CSV / JSON / Excel and includes structured fields such as group unit identifier, statistical quantities of group characteristics in various dimensions, group risk index Rg, risk level, intervention suggestion code, and effect evaluation data. It can be directly connected to the company's existing BI dashboard, management monthly report, special management report and other systems to realize the visualization of group mental health data and periodic management iteration, and provide accurate decision support for the company's management.

[0100] 6) Specific Application Examples

[0101] 6.1 Example 1: Application in the R&D Department of a Large Internet Company

[0102] The R&D department of a large internet company has 800 employees and 12 project teams. Due to high project delivery pressure and frequent overtime, they urgently need to use the system of this invention to achieve precise management of group mental health risks. The specific application and implementation process is as follows:

[0103] The first step was data access and anonymization: Data from 800 employees in the department was accessed from multiple sources, including chat summaries between employees and the AI ​​psychological assistant (over 5000 messages), monthly emotion assessment data (92% coverage), AI usage logs (over 300 daily uses), help-seeking behavior records (an average of 28 monthly requests), and organizational structure metadata (departments, project teams, job levels). Three levels of anonymization were performed, replacing employee IDs with anonymous identifiers and removing sensitive information such as specific project names and personal names mentioned in the chat summaries. Since the sample size for all project teams was ≥20 (greater than the threshold k=10), all analysis results were allowed to be output.

[0104] The second step is data processing: Time alignment is performed on a monthly basis, and the daily AI usage logs and monthly sentiment assessment data are uniformly mapped to the monthly time window; all heterogeneous data are mapped into four feature clusters, and a multi-layer group unit of "department-project group-month" is constructed. Eight group characteristics such as the average pressure, high pressure ratio, topic frequency, and topic drift index of each group unit are calculated. Among them, the topic frequency of "project delivery pressure" of the core project group "XX system R&D group" is 120 times in August and 75 times in July. The topic drift index Drift(8) = 120 / 75 = 1.6 is calculated, indicating that the attention to this topic has increased significantly.

[0105] The third step is risk identification: a weighted linear summation model is used to calculate the group risk index Rg for each project group. Among them, the Rg of the XX system R&D group was 76 in August (level 2 risk), and Rg was 62 in July, with a single-cycle increase of 22.6% for one cycle, which was determined to be a comprehensive high-risk concern group. At the same time, the overall Rg of the "test group" was identified as 58 (medium risk), but the "interpersonal conflict" theme drift index was 2.1, which was determined to be a theme-based key concern group, and a special early warning signal was issued.

[0106] The fourth step is the intervention loop: For the XX system R&D team, a combined intervention suggestion of "adjusting project delivery schedule + offline stress management workshops + customized stress reduction content delivery" is generated; for the test team, an intervention suggestion of "interpersonal communication training + one-on-one online consultation" is generated; one month after the intervention is implemented, the Rg of the XX system R&D team drops to 65, ΔRg=11%, and the intervention is judged to be effective; the topic drift index of "interpersonal conflict" in the test team drops to 1.2, and Rg drops to 52, with significant intervention effect. The evaluation results are written back to the system to optimize the relevant feature weights.

[0107] The fifth step is to output the following results: generate a group profile report for the R&D department in August, which clarifies the risk level of each project team, key groups of concern, and intervention recommendations; output a structured data package, connect it to the enterprise BI dashboard, and allow management to view the risk trends and intervention effects of each project team in real time, providing support for project management decisions.

[0108] 6.2 Example 2: Company-wide application in a medium-sized manufacturing enterprise

[0109] A medium-sized manufacturing enterprise has a total of 500 employees, covering six departments including production, sales, and administration, distributed across two regions. The enterprise emphasizes employee mental health management but lacks precise methods for group risk identification and intervention. The specific implementation process of the system of this invention is as follows:

[0110] The first step is data access and anonymization: Access chat summaries, quarterly sentiment assessment data, AI usage logs, help-seeking behavior records, and organizational structure metadata of 500 employees across the company; perform anonymization processing, replace employee ID numbers, mobile phone numbers, and other identifiers, and delete sensitive information. Among them, the sample size of the "Regional Sales Team" under a certain sales department is 8 (less than the threshold k=10), so the system automatically hides all analysis results of this team and only outputs the analysis results of the department-level and above groups.

[0111] The second step is data processing: time alignment is performed by quarter, multi-source data is mapped into four feature clusters, and group units of "region-department-quarter" are constructed. The group characteristics of each group unit are calculated. Among them, the average "sleep quality" of the production department in the third quarter was 62 (out of 100), and the proportion of low sleep quality reached 45%, which was significantly higher than that of other departments.

[0112] The third step is risk identification: calculate the risk index Rg for each department. The production department's Rg in the third quarter was 78 (level 2 risk), and its Rg in the second quarter was 65. The duration was one cycle, and the increase reached 20%, so it was identified as a comprehensive high-risk attention group; there was no specific theme-based key attention group.

[0113] The fourth step is the intervention loop: For the production department, an intervention suggestion of "sleep health themed activities + night shift work schedule adjustment suggestions + sleep improvement content push" is generated and implemented in conjunction with the human resources department and production management department; after one quarter of intervention implementation, the average sleep value of the production department increased to 75, the proportion of low sleep quality decreased to 28%, Rg decreased to 66, ΔRg=12%, the intervention was effective, and the effect data was written back to the system to iterate the intervention strategy.

[0114] The fifth step is to deliver results: generate a company-wide Q3 group profile report, compare the risk levels of different regions and departments, identify the production department as a key department to focus on and the effectiveness of interventions; output structured data packages, connect with the company's monthly management reports, realize the periodic management of employee mental health data, and help the company optimize its human resource management strategies.

[0115] The two embodiments above demonstrate that the system of the present invention can effectively solve the problems of data silos, privacy compliance and group analysis in existing enterprise psychological services, realize the compliant integration of multi-source mental health data, accurate identification of group risks, accurate implementation of intervention measures and closed-loop management of effects, adapt to the application needs of enterprises of different sizes and types, and has strong practicality and scalability.

[0116] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0117] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A psychological profile generation and closed-loop intervention system based on multi-source data aggregation, characterized in that, include: The system includes a de-identified data access module, a data processing module, a risk identification module, an intervention closed-loop module, and an output module. The output of the de-identified data access module is electrically connected to the input of the data processing module. The output of the data processing module is electrically connected to the input of the risk identification module. The output of the risk identification module is electrically connected to the input of the intervention closed-loop module. The output of the intervention closed-loop module is electrically connected to the input of the output module. The de-identified data access module is used to acquire and de-identify multi-source mental health data; The data processing module is used to perform time alignment, field mapping, and group aggregation on the de-identified data to generate group characteristics; The risk identification module is used to calculate the risk index based on group characteristics and identify key groups of concern; The intervention closed-loop module is used to generate intervention suggestions based on the identification results and to complete the effect evaluation and data write-back. The output module is used to output group profiles and intervention-related information.

2. The psychological profile generation and closed-loop intervention system based on multi-source data aggregation according to claim 1, characterized in that, Multi-source mental health data includes at least one of the following: chat summaries, emotion assessment indicators, AI usage logs, help-seeking behavior records, and organizational structure metadata; The desensitization process includes: user identifier replacement, sensitive information deletion, and output masking when the number of group samples is less than a threshold k, where k≥10.

3. The psychological profile generation and closed-loop intervention system based on multi-source data aggregation according to claim 1, characterized in that, The data processing module includes a time alignment and field mapping unit and a group feature construction unit. The time alignment and field mapping unit is used to unify time by week, month or quarter and map heterogeneous fields into stress, mood, sleep and behavior-related feature clusters.

4. The psychological profile generation and closed-loop intervention system based on multi-source data aggregation according to claim 3, characterized in that, The group feature building unit is used to hierarchically aggregate data according to at least one dimension, such as department, job level, region, project team, or time window, to form multi-level group units, and only outputs group-level analysis results.

5. The psychological profile generation and closed-loop intervention system based on multi-source data aggregation according to claim 4, characterized in that, Group characteristics include: mean stress level and percentage of high-stress individuals, mean sleep level and percentage of low-sleep individuals, mean mood, help-seeking rate, AI activity level, topic frequency, topic drift index, and percentage of individuals experiencing continuous high stress. The formula for calculating the topic drift index is as follows: In the formula: Let t be the theme drift index for period t; Let t be the frequency of the theme in the t-th period; The frequency of the theme in the (t-1)th period.

6. The psychological profile generation and closed-loop intervention system based on multi-source data aggregation according to claim 1, characterized in that, The risk identification module uses a weighted linear summation model to calculate the group risk index. : In the formula: For group risk index; The total number of group characteristics; The preset weights for the i-th feature; The value of the i-th feature after standardization; Risk identification module based on Values, duration, and rate of change are used to identify key target groups.

7. The psychological profile generation and closed-loop intervention system based on multi-source data aggregation according to claim 6, characterized in that, The risk identification module is also used to identify key thematic groups. These key thematic groups have a moderate overall risk, but a single risk theme indicator shows a sudden increase. After identification, a separate warning signal is output.

8. The psychological profile generation and closed-loop intervention system based on multi-source data aggregation according to claim 1, characterized in that, The intervention recommendations generated by the closed-loop intervention module include at least one of the following: themed activities, management training, job adjustments, content delivery, offline workshops, and special care; and the recommendations are compared before and after the intervention. Evaluation of the effectiveness of the change and data write-back.

9. The psychological profile generation and closed-loop intervention system based on multi-source data aggregation according to claim 1, characterized in that, The output module outputs a group profile report and a structured data package. The group profile report includes: risk level, core indicators, and period. Comparison, key themes, key groups, intervention recommendations, and feedback on effects.

10. The psychological profile generation and closed-loop intervention system based on multi-source data aggregation according to claim 9, characterized in that, Structured data packages are used to connect to BI dashboards, monthly management reports, and special management reports to achieve data visualization and periodic iteration.

Images