Method and device for training credit-guarding evaluation model of key personnel, medium and equipment

Abstract

The invention discloses a training method and device for a credit-guarding evaluation model of key personnel, a medium and equipment, and belongs to the technical field of deep learning. Obtaining personal data of a large number of key persons, wherein direct quantitative data in the personal data comprises demographic statistical data, family and social background data and economic data; the indirect quantitative data comprises social data, specific behavior data, psychological and cognitive data and management performance data; quantizing the directly quantized data into features; quantizing the indirect quantized data into features by using a large language model; calculating the features by using a Gini coefficient calculation formula to obtain a first numerical value; calculating the features by using a calculation formula of chi-square test to obtain a second numerical value; and training model parameters of the trusting evaluation model according to the first value and the second value. According to the method, the data can be quantified by using the large language model, the model parameters of the trusting evaluation model are determined according to the Gini coefficient and the chi-square test, and the accuracy of the model is improved.

Description

Technical Field

[0001] The present application relates to the field of deep learning technology, and in particular to a training method, device, medium and equipment for a trustworthiness evaluation model for key personnel. Background Art

[0002] The management of key drug abusers has always been an important issue in social governance. By evaluating the behavior of key drug abusers and giving them positive incentives, we can encourage them to abide by laws and regulations and actively reform themselves.

[0003] In related technologies, the trustworthiness evaluation of key personnel mainly relies on manual assessment, which has problems such as strong subjectivity, low efficiency, and inconsistent standards, resulting in inaccurate trustworthiness evaluation. Summary of the invention

[0004] This application provides a training method, device, medium and equipment for a trustworthiness evaluation model of key personnel, which is used to solve the problem that the trustworthiness evaluation of key personnel is highly subjective, inefficient and has inconsistent standards, resulting in inaccurate trustworthiness evaluation. The technical solution is as follows:

[0005] According to a first aspect of the present application, a training method for a trustworthiness evaluation model for key personnel is provided, the method comprising:

[0006] Obtaining a large amount of personal data of key personnel, the personal data including direct quantitative data and indirect quantitative data, the direct quantitative data including at least one of demographic data, family and social background data, and economic data; the indirect quantitative data including at least one of social data, specific behavior data, psychological and cognitive data, and management performance data;

[0007] quantizing the directly quantized data into corresponding features according to a preset quantization relationship;

[0008] quantizing the indirect quantization data into corresponding features using a large language model;

[0009] Creating a trustworthiness evaluation model, wherein the trustworthiness evaluation model is a logistic regression model based on the Gini coefficient and the chi-square test;

[0010] The feature is calculated using a calculation formula of the Gini coefficient to obtain a first value; the feature is calculated using a calculation formula of the chi-square test to obtain a second value; and the model parameters of the trustworthiness evaluation model are trained according to the first value and the second value.

[0011] In a possible implementation, the step of using a large language model to quantize the indirect quantization data into corresponding features includes:

[0012] Obtaining the collected data and prompt words of each dimension in the indirect quantitative data;

[0013] Inputting the collected data and the prompt words corresponding to the dimension into a large language model;

[0014] The collected data is processed according to the prompt word using the large language model to obtain features corresponding to the dimension.

[0015] In a possible implementation, the acquiring of the collected data and prompt words of each dimension in the indirect quantitative data includes:

[0016] Obtaining collected data of each dimension in the indirect quantitative data;

[0017] The personnel type information and dimension information of the key personnel are obtained, and the personnel type information and dimension information are added to a preset prompt word template to obtain a prompt word corresponding to the dimension.

[0018] In a possible implementation, the training of the model parameters of the trustworthiness evaluation model according to the first value and the second value includes:

[0019] Performing a logistic regression calculation on the first value and the second value using different model parameters;

[0020] The model parameters corresponding to the maximum calculation result are determined as the model parameters of the trustworthiness evaluation model.

[0021] In a possible implementation manner, quantizing the directly quantized data into corresponding features according to a preset quantization relationship includes:

[0022] For the collected data of each dimension in the directly quantized data, obtaining a quantization relationship corresponding to the dimension, wherein the quantization relationship includes a mapping relationship between different collected data and different features;

[0023] The features corresponding to the collected data are searched in the quantitative relationship.

[0024] In a possible implementation, the method further includes:

[0025] Obtaining personal data of key personnel to be evaluated, the personal data includes direct quantitative data and indirect quantitative data, the direct quantitative data includes demographic data, family and social background data, economic data; the indirect quantitative data includes social data, specific behavior data, psychological and cognitive data, and management performance data;

[0026] quantizing the directly quantized data into corresponding features according to a preset quantization relationship;

[0027] quantizing the indirect quantization data into corresponding features using a large language model;

[0028] The features are processed using the trustworthiness evaluation model to obtain the trustworthiness score of the key personnel.

[0029] In a possible implementation, the using the trustworthiness evaluation model to process the features to obtain the trustworthiness score of the key personnel includes:

[0030] Calculate the feature using the calculation formula of the Gini coefficient in the trustworthiness evaluation model to obtain a first value;

[0031] Calculate the feature using a calculation formula of a chi-square test to obtain a second value;

[0032] A logistic regression calculation is performed on the first value and the second value according to the model parameters to obtain the trustworthiness score of the key personnel.

[0033] According to a second aspect of the present application, a training device for a trustworthiness evaluation model of key personnel is provided, the device comprising:

[0034] An acquisition module, used to acquire personal data of a large number of key personnel, wherein the personal data includes direct quantitative data and indirect quantitative data, wherein the direct quantitative data includes at least one of demographic data, family and social background data, and economic data; and the indirect quantitative data includes at least one of social data, specific behavior data, psychological and cognitive data, and management performance data;

[0035] A quantization module, used for quantizing the directly quantized data into corresponding features according to a preset quantization relationship;

[0036] The quantization module is further used to quantize the indirect quantization data into corresponding features using a large language model;

[0037] A creation module is used to create a trustworthiness evaluation model, wherein the trustworthiness evaluation model is a logistic regression model based on the Gini coefficient and the chi-square test;

[0038] The training module is used to calculate the feature using the calculation formula of the Gini coefficient to obtain a first value; calculate the feature using the calculation formula of the chi-square test to obtain a second value; and train the model parameters of the trustworthiness evaluation model according to the first value and the second value.

[0039] According to a third aspect of the present application, a computer-readable storage medium is provided, in which at least one instruction is stored. The at least one instruction is loaded and executed by a processor to implement the training method of the trustworthiness evaluation model for key personnel as described above.

[0040] According to a fourth aspect of the present application, a computer device is provided, which includes a training device for the above-mentioned trustworthiness evaluation model for key personnel.

[0041] The beneficial effects of the technical solution provided by this application include at least:

[0042] For the personal data of key personnel, the large language model can be used to quantify the indirect quantitative data into corresponding features, thereby realizing the quantification of data that cannot be directly quantified. The model parameters of the trustworthiness evaluation model are the model parameters of the logistic regression model. In order to train the trustworthiness evaluation model, the features can be calculated using the calculation formula of the Gini coefficient to obtain the first value; the features can be calculated using the calculation formula of the chi-square test to obtain the second value; the model parameters of the trustworthiness evaluation model are trained based on the first and second values, thereby improving the accuracy of the trustworthiness evaluation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0044] Figure 1 It is a flow chart of a training method for a trustworthiness evaluation model for key personnel provided by an embodiment of the present application;

[0045] Figure 2 It is a flow chart of a training method for a trustworthiness evaluation model for key personnel provided by an embodiment of the present application;

[0046] Figure 3 It is a flow chart of a method for using a trustworthiness evaluation model for key personnel provided by an embodiment of the present application;

[0047] Figure 4 It is a structural block diagram of a training device for a trustworthiness evaluation model for key personnel provided by an embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0049] like Figure 1 As shown, it shows a method flow chart of a training method for a trustworthiness evaluation model of key personnel provided by an embodiment of the present application, and the training method for the trustworthiness evaluation model of key personnel can be applied to a computer device. The training method for the trustworthiness evaluation model of key personnel may include:

[0050] Step 101, obtaining personal data of a large number of key personnel, the personal data includes direct quantitative data and indirect quantitative data, the direct quantitative data includes at least one of demographic data, family and social background data, and economic data; the indirect quantitative data includes at least one of social data, specific behavior data, psychological and cognitive data, and management performance data.

[0051] The personal data of key personnel include direct quantitative data and indirect quantitative data. Direct quantitative data is data that can be directly quantified into corresponding features. For example, direct quantitative data can be different options, and each option corresponds to a different feature value. Indirect quantitative data is data that cannot be directly quantified and needs to be quantified with the help of a large language model. For example, indirect quantitative data can be the content of a psychological counseling conversation or a social chat record. The large language model can be used to score the indirect quantitative data and obtain the feature value.

[0052] First, direct quantitative data of drug abusers include at least one of demographic data, family and social background data, and economic data; indirect quantitative data include at least one of social data, specific behavioral data, psychological and cognitive data.

[0053] (1) Demographic data

[0054] Gender: There may be differences in drug abuse between men and women. The proportion of drug abusers of different genders, reasons for drug abuse, and preferences for different drugs can all be used as statistical data. For example, according to relevant reports, men account for a relatively high proportion of drug abusers in my country. In statistics in some regions, male drug abusers account for more than 80%.

[0055] Marital status: Marital status, such as single, married, divorced or widowed, may affect the psychological state and living environment of drug abusers, and thus be related to drug abuse behavior. For example, divorced or widowed people may be more likely to fall into the predicament of drug abuse due to emotional frustration and other reasons.

[0056] Educational level: People with different educational levels have different understandings of drugs, resistance, and social circles, which will affect their likelihood of drug abuse and their performance after drug abuse. For example, people with low education may be more likely to abuse drugs due to lack of relevant knowledge and correct guidance.

[0057] (2) Family and social background data

[0058] Family background: Family factors such as family income, family structure, parents’ education level, and whether there is a history of drug abuse have an important impact on individual drug abuse behavior. For example, people with financial difficulties, disharmonious parental relationships, or relatives who abuse drugs may be at higher risk of drug abuse.

[0059] Social background: Factors such as the economic development level, social and cultural environment, and drug circulation in different regions can lead to differences in the distribution and characteristics of drug abusers. For example, the proportion of drug abusers and the types of drug abuse may be different between economically developed and relatively backward regions, and the drug abuse situation in urban and rural areas also has its own characteristics.

[0060] (3) Economic data

[0061] Income level: There is a certain correlation between individual or family income and drug abuse. Low-income people may abuse drugs due to high life pressure and low expectations for the future, while high-income people may also use drugs due to social circles and seeking excitement. Drug abusers with different income levels will also have different types and frequencies of drug consumption.

[0062] Employment: In addition to occupation, employment status such as full-time, part-time, unemployed, and job stability are also important factors. People who are unemployed or have unstable jobs may be more likely to fall into the predicament of drug abuse, and the work pressure and environment of certain specific occupations may also be related to drug abuse behavior.

[0063] (4) Social Data

[0064] Social circle: Whether there are drug abusers among the drug abuser’s friends or relatives, as well as the location and frequency of their social activities, can reflect the impact of their social environment on drug abuse behavior. Drug abuse culture and peer pressure in the social circle may promote or hinder a person from abusing drugs.

[0065] Drug-related social relationships: such as whether the individual is involved in drug transactions, whether he or she has connections with drug dealers, etc. These relationships can indirectly reflect the depth and breadth of his or her drug abuse and are of great value in preventing drug abuse.

[0066] (5) Specific behavior data

[0067] Age of first drug abuse: reflects the time when drug abusers begin to come into contact with drugs. It is of great significance for analyzing drug abuse trends, key age groups for drug abuse prevention work, etc. It can also be used as a reference factor for predicting the future development of drug abusers and the difficulty of quitting drug addiction.

[0068] Frequency of drug abuse: including the number of times of drug abuse in different time dimensions such as daily, weekly, and monthly, can reflect the degree of drug addiction of drug abusers, which in turn affects their physical condition, social function, and the difficulty and need of quitting drug addiction.

[0069] Methods of drug abuse: Different methods of drug abuse, such as oral, injection, and smoking, are not only related to the type of drug, but may also reflect the preferences, economic strength, and awareness of their own health of the drug abuser. The health risks and social hazards brought about by different methods of drug abuse are also different.

[0070] Drug addiction treatment experience: Data such as whether there has been drug addiction treatment experience, the number of times the drug addiction has been treated, the method of drug addiction treatment, and the relapse after drug addiction treatment are very important for evaluating the drug abuser's willingness to treat drug addiction, the effectiveness of drug addiction treatment, and formulating personalized drug addiction treatment plans.

[0071] (6) Psychological and cognitive data

[0072] Mental health status: Drug abusers often have psychological problems of varying degrees, such as anxiety, depression, and low self-esteem. Using professional psychological assessment tools to assess the mental health status of drug abusers and understand their psychological characteristics and problems will help analyze the psychological motivations for drug abuse and provide a basis for psychological intervention and drug addiction treatment.

[0073] Drug awareness: This includes the understanding of the harm of drugs, the knowledge of laws and regulations, etc. People with lower awareness may be more likely to try to abuse drugs, and improving their awareness of drugs is also an important part of the process of quitting drug addiction.

[0074] Second, the direct quantitative data of community management objects include at least one of demographic statistics and family and social background data, and the indirect quantitative data include at least one of specific behavioral data, psychological and cognitive data, and management performance data.

[0075] (1) Demographic data

[0076] Gender: Community management objects of different genders may differ in specific behavior types, management needs, and psychological characteristics. The ratio of men to women and the distribution of different genders in various specific behaviors can all be used as statistical data.

[0077] Marital status: Marital status such as unmarried, married, divorced or widowed will affect the life stability, family support system and psychological state of the community management subjects, and thus affect the management effect. The risk of recidivism of community management subjects with different marital status can be analyzed.

[0078] Educational level: The level of cultural literacy is closely related to legal awareness, cognitive ability and employment opportunities. Community management subjects with lower educational levels may need more educational assistance and skills training to improve their ability to return to society. It is possible to study the characteristics of people with different educational levels in terms of specific behavior types, performance during management, etc.

[0079] (2) Family and social background data

[0080] Family income: Family economic conditions will affect the quality of life and psychological state of community management subjects. Management subjects from low-income families may face greater life pressure, which will affect their management results and recidivism risks. The relationship between family income and performance during management can be analyzed.

[0081] Family structure: such as the number of family members, whether there are children, relationship with family members, etc. A stable family structure and good family relationships can provide emotional support and supervision for the subjects of community management, help reduce the risk of recidivism, otherwise it may increase the risk.

[0082] Social support network: including the number of friends, frequency of social activities, participation in community organizations or volunteer activities, etc. A rich social support network can help community management objects better integrate into society, reduce loneliness and social exclusion, and improve management effectiveness.

[0083] (3) Specific behavior data

[0084] Specific behavior types: such as drunk driving, drug abuse, intentional injury, etc. The community management objects of different specific behavior types have great differences in psychological characteristics, behavior patterns and management needs. Classification, statistics and analysis of them will help to formulate more targeted management plans.

[0085] Number of specific behaviors: Data such as the ratio of first-time offenders to repeat offenders and the frequency of specific behaviors of repeat offenders can reflect the specific behavioral tendencies and management difficulties of community management subjects. Repeat offenders may require stricter supervision and more in-depth psychological correction.

[0086] Judgment situation: including the type of punishment, length of punishment, etc. These factors directly determine the duration and supervision requirements of community management, and will also affect the psychological pressure and management attitude of the community management subjects.

[0087] (4) Psychological and cognitive data

[0088] Mental health level: Through professional psychological assessment tools, we evaluate the anxiety, depression, inferiority, aggressiveness and other psychological characteristics of the community management subjects, understand their mental health status, and provide them with corresponding psychological counseling and intervention measures. Those with more serious mental health problems may need closer attention and more professional psychotherapy.

[0089] Self-cognition: including awareness of one's own specific behaviors, understanding of one's own personality and abilities, expectations for the future, etc. Community management subjects with clear self-cognition are more likely to actively participate in management, formulate reasonable life plans, and thus better return to society.

[0090] Legal awareness: the degree of understanding of laws and regulations and the awareness of compliance. People with insufficient legal awareness may be more likely to break the law again, so legal education and publicity need to be strengthened in the management process.

[0091] (5) Manage performance data

[0092] Compliance with regulations: such as reporting on time, attendance at education and community services, and whether the various regulations of community management have been violated. These data can directly reflect the management attitude and self-discipline of the community management objects, and are of great significance for evaluating management effectiveness and deciding whether the management plan needs to be adjusted.

[0093] Educational learning outcomes: Measure learning outcomes in legal knowledge, professional skills, mental health, etc. through assessment scores, learning experiences, behavioral changes, etc., so as to timely adjust educational content and methods and improve the quality of education management.

[0094] Community service performance: including participation in community service, work attitude, interaction with community residents, etc. Community management subjects who actively participate in community service and perform well are often more likely to gain recognition and acceptance from the community, enhance their sense of social responsibility and belonging, and promote their smooth return to society.

[0095] Third, the direct quantitative data of released personnel include at least one of demographic data, family and social background data, and economic data; the indirect quantitative data include at least one of specific behavioral data, psychological and cognitive data, and management performance data.

[0096] (1) Demographic data

[0097] Gender: Released persons of different genders may differ in specific behavior types, recidivism rates, etc. For example, men account for a higher proportion of certain specific behavior types, while women also account for a certain proportion of certain specific economic behaviors. Analyzing gender factors helps to understand the behavioral characteristics and management needs of released persons of different genders.

[0098] Marital status: Single, married, divorced or widowed marital status has an impact on the stability of life and psychological state of released persons. Married persons may have a family support system, which helps them to better return to society; while divorced or widowed persons may face greater life pressure and psychological burden, which affects their risk of recidivism.

[0099] Educational level: Educational level is related to the cognitive level, employment opportunities and social adaptability of released persons. People with lower educational levels may face more difficulties in finding employment and integrating into society, and have a relatively higher risk of recidivism. At the same time, the need for educational assistance and skills training is more urgent.

[0100] (2) Family and social background data

[0101] Family income: Family economic conditions will affect the quality of life and psychological state of released persons, and thus affect their risk of recidivism. Released persons from low-income families may face greater life pressure and may be more likely to recidivate if they lack effective economic support and employment opportunities.

[0102] Family structure: Family structure factors such as the number of family members, whether there are children, and the relationship with family members have an important impact on the reintegration of released persons into society. A stable family structure and good family relationships can provide emotional support and supervision for them, reducing the risk of recidivism; on the contrary, tense or broken family relationships may increase the possibility of recidivism.

[0103] Living environment: The economic development level, public security situation, community environment and other factors of the residential area will also have an impact on the released persons. People living in economically backward and poorly public security areas may face more bad temptations and social pressures, increasing the risk of recidivism; while a good community environment and community support system will help them integrate into society and reduce recidivism.

[0104] (3) Economic data

[0105] Employment status: The employment status of released prisoners, such as whether they are employed, the type of employment, and job stability, is an important factor affecting their reintegration into society and the risk of recidivism. Stable employment can provide economic income and social recognition, reducing the risk of recidivism; while difficulty in finding a job or frequent unemployment may lead to a lack of livelihood and increase the possibility of recidivism.

[0106] Income level: An individual’s income level is closely related to their quality of life and psychological state. Released prisoners with lower incomes may face financial pressure and be unable to meet their basic living needs, which increases their motivation to reoffend. Improving their income level and economic stability will help reduce the risk of reoffending.

[0107] Vocational skills: The level of vocational skills of released personnel affects their employment opportunities and career development. People who lack vocational skills may only be able to work in low-paying, unstable jobs, or even find it difficult to find a job; improving their employability and competitiveness through vocational skills training is of great significance for their smooth return to society.

[0108] (4) Specific behavior data

[0109] Specific types: such as drunk driving, intentional injury, and specific behavior types involving drug abuse. Different types of specific behaviors reflect the behavioral patterns and psychological characteristics of released personnel, and their recidivism risks and management needs are also different. For example, people involved in drug abuse have a higher risk of relapse and need more targeted drug addiction treatment and psychological counseling measures.

[0110] Number of specific behaviors: Data such as the ratio of first-time offenders to repeat offenders and the frequency of specific behaviors of repeat offenders can reflect the specific behavioral tendencies and management difficulties of released persons. Repeat offenders often have a higher risk of reoffending and may require stricter supervision and more in-depth psychological correction and social assistance.

[0111] Length of punishment: The length of punishment is related to the reform effect and recidivism risk of the released person. Generally speaking, the risk of recidivism of released persons with a short punishment period may be relatively high due to the relatively short time of reform, but this is not absolute and needs to be analyzed in combination with other factors.

[0112] (5) Psychological and cognitive data

[0113] Mental health status: People who have been released from prison after a long period of punishment often have psychological problems to varying degrees, such as anxiety, depression, inferiority, aggression, etc. Assessing their mental health status through professional psychological assessment tools helps to understand their psychological characteristics and problems, and provides a basis for the formulation of psychological intervention and management measures. People with more serious mental health problems may need closer attention and more professional psychological treatment.

[0114] Self-awareness: including awareness of one’s own specific behaviors, understanding of one’s own personality and abilities, and expectations for the future. Released personnel with clear self-awareness are more likely to actively participate in management, develop reasonable life plans, and thus better return to society.

[0115] Social cognition: The cognition and attitude of released prisoners towards society will also affect their behavior. For example, negative cognition such as dissatisfaction with society and distrust of others may increase their risk of recidivism; while positive social cognition and yearning for a normal life will help them integrate into society and reduce the possibility of recidivism.

[0116] (6) Management performance data

[0117] Performance during management: including reform performance, participation in education and labor reform, etc. Released persons who perform well during management have relatively good reform results and relatively low recidivism risks; while persons with poor performance may need further strengthening of supervision and management measures.

[0118] Acceptance of support measures: whether relevant support measures have been accepted after release, such as employment guidance, psychological counseling, social security, etc., as well as the degree of acceptance and effectiveness of these support measures. Effective support measures can help released persons better adapt to society and reduce the risk of recidivism, but if the support measures are not in place or are not accepted, it may affect the effectiveness of their return to society.

[0119] Step 102: quantize the direct quantization data into corresponding features according to a preset quantization relationship.

[0120] The quantitative relationship can be set empirically.

[0121] For example, demographic data includes gender, marital status, and education level; male and female in gender correspond to different characteristic values; unmarried, married, divorced, and widowed in marital status correspond to different characteristic values; primary education, secondary education, and higher education in education level correspond to different characteristic values. After obtaining the demographic data of a key person, the corresponding characteristic value can be determined based on the gender, marital status, and education level of the key person.

[0122] Step 103: quantize the indirect quantization data into corresponding features using a large language model.

[0123] The large language model is a pre-trained model, and the training method of the large language model is not limited in this embodiment.

[0124] The computer device can input each indirect quantization data into the large language model, and the large language model can process the indirect quantization data and score it to obtain a corresponding feature value.

[0125] Step 104: Create a trustworthiness evaluation model, which is a logistic regression model based on the Gini coefficient and chi-square test.

[0126] The Gini coefficient was originally one of the commonly used indicators internationally to measure the income gap among residents of a country or region.

[0127] The chi-square test is a test of the degree of deviation between the actual observed values ​​and the theoretical inference values ​​of a statistical sample. The degree of deviation between the actual observed values ​​and the theoretical inference values ​​determines the size of the chi-square value. The larger the chi-square value, the greater the deviation between the two; conversely, the smaller the deviation between the two; if the two values ​​are completely equal, the chi-square value is 0, indicating that the theoretical value is completely consistent.

[0128] The logistic regression model is a model that estimates the probability of an event occurring based on a given set of independent variable data.

[0129] In this embodiment, the calculation results of the Gini coefficient and the chi-square test on the features are used as independent variables, so that the logistic regression model trains model parameters according to the independent variables.

[0130] Step 105, calculate the feature using the calculation formula of the Gini coefficient to obtain a first value; calculate the feature using the calculation formula of the chi-square test to obtain a second value; and train the model parameters of the trustworthiness evaluation model according to the first value and the second value.

[0131] The calculation formula of the Gini coefficient is: Where t represents time, i represents the feature of the i-th dimension, c represents the total dimension of the feature, and p(i|t) represents the probability that the feature value is i at a given time t. Substituting all features into the calculation formula of the Gini coefficient, the first value can be obtained.

[0132] The calculation formula for the chi-square test is: Where i represents the feature of the i-th dimension, j represents the j-th key personnel in a feature, c represents the total dimension of the feature (i.e., the number of rows), r represents the total number of key personnel (i.e., the number of columns), and O ij Indicates the characteristics of the i-th key person in the i-th characteristic, the total of the rows i Represents the sum of the i-th feature of all key personnel, column sum j represents the sum of all characteristics of the jth key personnel, and the total sample size represents the sum of all characteristic values ​​of all key personnel. ij -E ij ) 2 / E ij It represents the square of the deviation of each cell divided by its expected frequency. Substituting all the features into the calculation formula of the chi-square test, the second value can be obtained.

[0133] On the premise that both the first value and the second value are known, the first value and the second value can be used to train the model parameters of the trustworthiness evaluation model, and finally a trained trustworthiness evaluation model can be obtained.

[0134] In summary, the training method of the trustworthiness evaluation model for key personnel provided in the embodiment of the present application can use a large language model to quantify indirect quantitative data into corresponding features for the personal data of key personnel, thereby realizing the quantification of data that cannot be directly quantified. The model parameters of the trustworthiness evaluation model are the model parameters of the logistic regression model. In order to train the trustworthiness evaluation model, the features can be calculated using the calculation formula of the Gini coefficient to obtain a first value; the features can be calculated using the calculation formula of the chi-square test to obtain a second value; the model parameters of the trustworthiness evaluation model are trained according to the first value and the second value, thereby improving the accuracy of the trustworthiness evaluation model.

[0135] like Figure 2 As shown, it shows a flow chart of a training method for a trustworthiness evaluation model of key personnel provided by an embodiment of the present application, and the training method for the trustworthiness evaluation model of key personnel can be applied to a computer device. The training method for the trustworthiness evaluation model of key personnel may include:

[0136] Step 201, obtaining personal data of a large number of key personnel, the personal data including direct quantitative data and indirect quantitative data, the direct quantitative data including at least one of demographic data, family and social background data, and economic data; the indirect quantitative data including at least one of social data, specific behavior data, psychological and cognitive data, and management performance data.

[0137] Key personnel include at least drug abusers, community management subjects and released persons. The personal data of each type of personnel is detailed in the description of step 101 and will not be repeated here.

[0138] Step 202, for the collected data of each dimension in the direct quantization data, obtain the quantization relationship corresponding to the dimension, wherein the quantization relationship includes a mapping relationship between different collected data and different features; and search for features corresponding to the collected data in the quantization relationship.

[0139] Quantitative relationships can be set based on experience. For example, if the proportion of males among drug abusers is relatively high, the quantitative relationship set can be: male ~1 point, female ~2 points. For another example, divorced or widowed people may be more likely to fall into the dilemma of drug abuse due to emotional frustration, etc., so the quantitative relationship set can be: unmarried ~3 points, married ~2 points, divorced ~1 point, widowed ~1 point. For another example, people with low education may be more likely to abuse drugs due to lack of relevant knowledge and correct guidance, so the quantitative relationship set can be: higher education ~3 points, secondary education ~2 points, lower education ~1 point.

[0140] Assuming that the collected data of a key person is male, divorced, and low education, then his characteristics = 1+1+1=3 points.

[0141] Step 203, obtaining the collected data and prompt words of each dimension in the indirect quantitative data; inputting the collected data and the prompt words corresponding to the dimension into the large language model; using the large language model to process the collected data according to the prompt words to obtain features corresponding to the dimension.

[0142] For each type of key personnel, the collected data of each dimension needs to be quantified into corresponding features through the large language model. Each dimension can have multiple collected data. For example, in the dimension of psychological and cognitive data, one collected data is the mental health status, and the other collected data is the degree of drug cognition.

[0143] Specifically, obtaining the collected data and prompt words for each dimension in the indirect quantitative data may include: obtaining the collected data for each dimension in the indirect quantitative data; obtaining the personnel type information and dimension information of key personnel, adding the personnel type information and dimension information to a preset prompt word template, and obtaining the prompt word corresponding to the dimension.

[0144] Among them, the prompt words can be: "Please give the mental health level of the drug abuser based on the chat record between the drug abuser and the psychologist below. 5 points are very healthy, 4 points are relatively healthy, 3 points are average, 2 points are relatively unhealthy, and 1 point is very unhealthy. The chat record is as follows: xxxxxx. What is the score given?".

[0145] Step 204: Create a trustworthiness evaluation model, which is a logistic regression model based on the Gini coefficient and chi-square test.

[0146] For details about the Gini coefficient, chi-square test, and logistic regression model, please refer to the description in step 104.

[0147] In this embodiment, the calculation results of the Gini coefficient and the chi-square test on the features are used as independent variables, so that the logistic regression model trains model parameters according to the independent variables.

[0148] Step 205, calculate the feature using the calculation formula of the Gini coefficient to obtain a first value; calculate the feature using the calculation formula of the chi-square test to obtain a second value; and train the model parameters of the trustworthiness evaluation model according to the first value and the second value.

[0149] The calculation formulas for the Gini coefficient and the chi-square test are detailed in the description of step 105.

[0150] Under the premise that the first value and the second value are both known, the first value and the second value can be used to train the model parameters of the trustworthiness evaluation model. Specifically, different model parameters can be used to perform logistic regression calculation on the first value and the second value; and the model parameter corresponding to the maximum calculation result is determined as the model parameter of the trustworthiness evaluation model.

[0151] In summary, the training method of the trustworthiness evaluation model for key personnel provided in the embodiment of the present application can use a large language model to quantify indirect quantitative data into corresponding features for the personal data of key personnel, thereby realizing the quantification of data that cannot be directly quantified. The model parameters of the trustworthiness evaluation model are the model parameters of the logistic regression model. In order to train the trustworthiness evaluation model, the features can be calculated using the calculation formula of the Gini coefficient to obtain a first value; the features can be calculated using the calculation formula of the chi-square test to obtain a second value; the model parameters of the trustworthiness evaluation model are trained according to the first value and the second value, thereby improving the accuracy of the trustworthiness evaluation model.

[0152] like Figure 3 As shown, it shows a flow chart of a method for using a trustworthiness evaluation model for key personnel provided by an embodiment of the present application, and the method for using the trustworthiness evaluation model for key personnel can be applied to a computer device. The method for using the trustworthiness evaluation model for key personnel may include:

[0153] Step 301, obtaining personal data of key personnel to be evaluated, the personal data includes direct quantitative data and indirect quantitative data, the direct quantitative data includes at least one of demographic data, family and social background data, and economic data; the indirect quantitative data includes at least one of social data, specific behavior data, psychological and cognitive data, and management performance data.

[0154] When the key person to be evaluated is a drug abuser, the direct quantitative data includes at least one of demographic data, family and social background data, and economic data, and the indirect quantitative data includes at least one of social data, specific behavioral data, and psychological and cognitive data.

[0155] When the key personnel to be evaluated are the subjects of community management, the direct quantitative data include at least one of demographic statistics and family and social background data, and the indirect quantitative data include at least one of specific behavioral data, psychological and cognitive data, and management performance data.

[0156] When the key personnel to be evaluated are released personnel, the direct quantitative data include at least one of demographic statistics, family and social background data, and economic data; the indirect quantitative data include at least one of specific behavioral data, psychological and cognitive data, and management performance data.

[0157] Step 302: quantize the directly quantized data into corresponding features according to a preset quantization relationship.

[0158] Step 303: quantize the indirect quantization data into corresponding features using a large language model.

[0159] Step 304, use the trustworthiness evaluation model to process the features to obtain the trustworthiness score of the key personnel.

[0160] Specifically, using the trustworthiness evaluation model to process the characteristics to obtain the trustworthiness score of the key personnel can include: using the calculation formula of the Gini coefficient in the trustworthiness evaluation model to calculate the characteristics to obtain a first value; using the calculation formula of the chi-square test to calculate the characteristics to obtain a second value; performing a logistic regression calculation on the first value and the second value according to the model parameters to obtain the trustworthiness score of the key personnel.

[0161] like Figure 4 As shown, it shows a structural block diagram of a training device for a trustworthiness evaluation model of key personnel provided by an embodiment of the present application, and the training device for the trustworthiness evaluation model of key personnel can be applied to a computer device. The training device for the trustworthiness evaluation model of key personnel may include:

[0162] An acquisition module 410 is used to acquire personal data of a large number of key personnel, wherein the personal data includes direct quantitative data and indirect quantitative data, wherein the direct quantitative data includes at least one of demographic data, family and social background data, and economic data; and the indirect quantitative data includes at least one of social data, specific behavior data, psychological and cognitive data, and management performance data;

[0163] A quantization module 420, configured to quantize the directly quantized data into corresponding features according to a preset quantization relationship;

[0164] The quantization module 420 is further used to quantize the indirect quantization data into corresponding features using the large language model;

[0165] A creation module 430 is used to create a trustworthiness evaluation model, where the trustworthiness evaluation model is a logistic regression model based on the Gini coefficient and the chi-square test;

[0166] The training module 440 is used to calculate the feature using the calculation formula of the Gini coefficient to obtain a first value; calculate the feature using the calculation formula of the chi-square test to obtain a second value; and train the model parameters of the trustworthiness evaluation model according to the first value and the second value.

[0167] In an optional embodiment, the quantization module 420 is further configured to:

[0168] Obtain the collected data and prompt words for each dimension of the indirect quantitative data;

[0169] Input the collected data and the prompt words corresponding to the dimensions into the large language model;

[0170] The collected data is processed according to the prompt words using a large language model to obtain features corresponding to the dimensions.

[0171] In an optional embodiment, the quantization module 420 is further configured to:

[0172] Obtain the collected data for each dimension in the indirect quantitative data;

[0173] The personnel type information and dimension information of the key personnel are obtained, and the personnel type information and dimension information are added to the preset prompt word template to obtain the prompt word corresponding to the dimension.

[0174] In an optional embodiment, the training module 440 is further configured to:

[0175] Performing a logistic regression calculation on the first value and the second value using different model parameters;

[0176] The model parameters corresponding to the maximum calculation results are determined as the model parameters of the trustworthiness evaluation model.

[0177] In an optional embodiment, the quantization module 420 is further configured to:

[0178] For the collected data of each dimension in the direct quantitative data, a quantitative relationship corresponding to the dimension is obtained, where the quantitative relationship includes a mapping relationship between different collected data and different features;

[0179] Find the features corresponding to the collected data in the quantitative relationship.

[0180] In an optional embodiment, the acquisition module 410 is further used to acquire personal data of the key personnel to be evaluated, the personal data including direct quantitative data and indirect quantitative data, the direct quantitative data including demographic data, family and social background data, economic data; the indirect quantitative data including social data, specific behavior data, psychological and cognitive data, management performance data;

[0181] The quantization module 420 is further used to quantize the direct quantization data into corresponding features according to a preset quantization relationship;

[0182] The quantization module 420 is further used to quantize the indirect quantization data into corresponding features using the large language model;

[0183] The evaluation module is used to process the features using the trustworthiness evaluation model to obtain the trustworthiness score of key personnel.

[0184] In an optional embodiment, the evaluation module is further used to:

[0185] The calculation formula of the Gini coefficient in the trustworthiness evaluation model is used to calculate the characteristics to obtain a first value;

[0186] The characteristic is calculated using the calculation formula of the chi-square test to obtain a second value;

[0187] A logistic regression calculation is performed on the first value and the second value according to the model parameters to obtain the trustworthiness score of the key personnel.

[0188] In summary, the training device of the trustworthiness evaluation model for key personnel provided in the embodiment of the present application can use a large language model to quantify indirect quantitative data into corresponding features for the personal data of key personnel, thereby realizing the quantification of data that cannot be directly quantified. The model parameters of the trustworthiness evaluation model are the model parameters of the logistic regression model. In order to train the trustworthiness evaluation model, the features can be calculated using the calculation formula of the Gini coefficient to obtain a first value; the features can be calculated using the calculation formula of the chi-square test to obtain a second value; the model parameters of the trustworthiness evaluation model are trained according to the first value and the second value, thereby improving the accuracy of the trustworthiness evaluation model.

[0189] An embodiment of the present application provides a computer-readable storage medium, in which at least one instruction is stored. The at least one instruction is loaded and executed by a processor to implement the training method of the trustworthiness evaluation model for key personnel as described above.

[0190] An embodiment of the present application provides a computer device, which includes a training device for the trustworthiness evaluation model of any key personnel mentioned above.

[0191] It should be noted that: the training device for the trustworthiness evaluation model of key personnel provided in the above embodiment only uses the division of the above functional modules as an example when conducting the trustworthiness evaluation of key personnel. In actual applications, the above functional allocation can be completed by different functional modules as needed, that is, the internal structure of the training device for the trustworthiness evaluation model of key personnel is divided into different functional modules to complete all or part of the functions described above. In addition, the training device for the trustworthiness evaluation model of key personnel provided in the above embodiment and the training method embodiment of the trustworthiness evaluation model of key personnel belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0192] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0193] The above description is not intended to limit the embodiments of the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the protection scope of the embodiments of the present application.

Claims

1. A training method for a trustworthiness evaluation model for key personnel, characterized in that: The method comprises: Obtaining a large amount of personal data of key personnel, the personal data including direct quantitative data and indirect quantitative data, the direct quantitative data including at least one of demographic data, family and social background data, and economic data; the indirect quantitative data including at least one of social data, specific behavior data, psychological and cognitive data, and management performance data; quantizing the directly quantized data into corresponding features according to a preset quantization relationship; quantizing the indirect quantization data into corresponding features using a large language model; Creating a trustworthiness evaluation model, wherein the trustworthiness evaluation model is a logistic regression model based on the Gini coefficient and the chi-square test; The feature is calculated using a calculation formula of the Gini coefficient to obtain a first value; the feature is calculated using a calculation formula of the chi-square test to obtain a second value; and the model parameters of the trustworthiness evaluation model are trained according to the first value and the second value.

2. The training method of the trustworthiness evaluation model for key personnel according to claim 1 is characterized in that: The using of a large language model to quantize the indirect quantization data into corresponding features includes: Obtaining the collected data and prompt words of each dimension in the indirect quantitative data; Inputting the collected data and the prompt words corresponding to the dimension into a large language model; The collected data is processed according to the prompt word using the large language model to obtain features corresponding to the dimension.

3. The training method of the trustworthiness evaluation model for key personnel according to claim 2 is characterized in that: The step of obtaining the collected data and prompt words of each dimension in the indirect quantitative data includes: Obtaining collected data of each dimension in the indirect quantitative data; The personnel type information and dimension information of the key personnel are obtained, and the personnel type information and dimension information are added to a preset prompt word template to obtain a prompt word corresponding to the dimension.

4. The training method of the trustworthiness evaluation model for key personnel according to claim 1 is characterized in that: The step of training the model parameters of the trustworthiness evaluation model according to the first value and the second value includes: Performing a logistic regression calculation on the first value and the second value using different model parameters; The model parameters corresponding to the maximum calculation result are determined as the model parameters of the trustworthiness evaluation model.

5. The training method of the trustworthiness evaluation model for key personnel according to claim 1 is characterized in that: The step of quantizing the directly quantized data into corresponding features according to a preset quantization relationship includes: For the collected data of each dimension in the directly quantized data, obtaining a quantization relationship corresponding to the dimension, wherein the quantization relationship includes a mapping relationship between different collected data and different features; The features corresponding to the collected data are searched in the quantitative relationship.

6. The training method of the trustworthiness evaluation model for key personnel according to any one of claims 1 to 5, characterized in that: The method further comprises: Obtaining personal data of key personnel to be evaluated, the personal data includes direct quantitative data and indirect quantitative data, the direct quantitative data includes demographic data, family and social background data, economic data; the indirect quantitative data includes social data, specific behavior data, psychological and cognitive data, and management performance data; quantizing the directly quantized data into corresponding features according to a preset quantization relationship; quantizing the indirect quantization data into corresponding features using a large language model; The features are processed using the trustworthiness evaluation model to obtain the trustworthiness score of the key personnel.

7. The training method of the trustworthiness evaluation model for key personnel according to claim 6 is characterized in that: The using of the trustworthiness evaluation model to process the features to obtain the trustworthiness score of the key personnel includes: Calculate the feature using the calculation formula of the Gini coefficient in the trustworthiness evaluation model to obtain a first value; Calculate the feature using a calculation formula of a chi-square test to obtain a second value; A logistic regression calculation is performed on the first value and the second value according to the model parameters to obtain the trustworthiness score of the key personnel.

8. A training device for a trustworthiness evaluation model for key personnel, characterized in that: The device comprises: An acquisition module, used to acquire personal data of a large number of key personnel, wherein the personal data includes direct quantitative data and indirect quantitative data, wherein the direct quantitative data includes at least one of demographic data, family and social background data, and economic data; and the indirect quantitative data includes at least one of social data, specific behavior data, psychological and cognitive data, and management performance data; A quantization module, used for quantizing the directly quantized data into corresponding features according to a preset quantization relationship; The quantization module is further used to quantize the indirect quantization data into corresponding features using a large language model; A creation module is used to create a trustworthiness evaluation model, wherein the trustworthiness evaluation model is a logistic regression model based on the Gini coefficient and the chi-square test; The training module is used to calculate the feature using the calculation formula of the Gini coefficient to obtain a first value; calculate the feature using the calculation formula of the chi-square test to obtain a second value; and train the model parameters of the trustworthiness evaluation model according to the first value and the second value.

9. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the training method of the trustworthiness evaluation model for key personnel as described in any one of claims 1 to 7.

10. A computer device, characterized in that: The computer equipment includes: a training device for the trustworthiness evaluation model of key personnel as described in claim 8.

Images