Employee loss risk prediction method, device and equipment

Abstract

The invention provides an employee loss risk prediction method, device and equipment, and relates to the technical field of data processing, and the method comprises the steps: obtaining a target department and a target post where a target employee is located, and the work data and potential data of the target employee; evaluating the work data based on a target evaluation model to obtain a first evaluation score of the target employee; inputting the potential data into a trained multi-modal potential recognition model to obtain a potential evaluation score of the target employee; based on the first evaluation score and the potential evaluation score, obtaining a comprehensive evaluation result of the target employee; and inputting the work data and the comprehensive evaluation result into a trained loss risk prediction model to obtain a loss risk prediction result of the target employee. According to the method, the ability evaluation of the employees is comprehensively carried out from multiple aspects of performance, work experience, potential and achievement; and meanwhile, the loss risk of the employee is predicted by combining salary and capability evaluation.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device and equipment for predicting employee turnover risk. Background Art

[0002] Currently, companies use relatively single reference data sources for talent assessment, such as the performance of the year and the ratings of immediate superiors. Such single data sources are overly subjective, resulting in non-objective, incomplete and inaccurate assessment results.

[0003] At the same time, companies lack intuitive tools to view the corresponding talent comparison data, resulting in relatively primitive tools used in talent inventory, such as spreadsheet documents or one-time data tables, which is time-consuming and laborious. Since it is impossible to intuitively compare the work data of employees with different potentials and different ratings, it is impossible to focus on the development of employees, make corresponding adjustments or provide effective guidance and help, resulting in a large number of talent losses and poor quality of talent reserves. Companies often face emergencies such as high staff turnover, insufficient successors to key positions, or excessive staff reserves. These problems seriously restrict the development of companies. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide a method, device and equipment for predicting the risk of employee turnover, so as to solve the above-mentioned problems existing in the prior art, and to comprehensively and objectively evaluate employees, predict the risk of employee turnover, and facilitate timely adjustments.

[0005] In a first aspect, a method for predicting employee turnover risk is provided, which may include:

[0006] Obtain the target department and target position of the target employee, as well as the target employee's work data and potential data; wherein the potential data includes a video of the target employee during the process of answering questions in the potential test and the results of the potential test;

[0007] Match the target evaluation model corresponding to the target department and the target position from the configured comparison table of different departments, different positions and different evaluation models;

[0008] Evaluate the work data based on the target evaluation model to obtain a first evaluation score for the target employee;

[0009] Inputting the potential data into a trained multimodal potential identification model to obtain a potential assessment score of the target employee;

[0010] Obtaining a comprehensive evaluation result of the target employee based on the first evaluation score and the potential evaluation score;

[0011] The work data and the comprehensive evaluation results are input into a trained turnover risk prediction model to obtain a turnover risk prediction result of the target employee.

[0012] In an optional implementation, the work data includes: work experience, historical achievements and performance data;

[0013] The target evaluation model includes: a target performance evaluation model, a target achievement evaluation model and a target work experience evaluation model;

[0014] The work data is evaluated based on the target evaluation model to obtain a first evaluation score of the target employee, including:

[0015] Based on the target performance evaluation model, the target achievement evaluation model and the target work experience evaluation model, the performance data, the historical achievements and the work experience are evaluated respectively to obtain the performance evaluation score, the historical achievement evaluation score and the work experience evaluation score of the target employee;

[0016] Based on the performance evaluation score, the historical achievement evaluation score and the work experience evaluation score, a first evaluation score is obtained.

[0017] In an optional implementation, before inputting the potential data into a trained multimodal potential identification model to obtain a potential assessment score of the target employee, the method further includes:

[0018] For a video frame of a target employee at any time point during the process of answering questions in the potential test, extract the expression feature value, posture feature value, voice feature value and speech speed feature value of the target employee from the video frame;

[0019] Generate a feature matrix based on the expression feature value, posture feature value, sound feature value, and speech rate feature value;

[0020] Performing feature extraction on the potential test answer results to obtain text feature values;

[0021] The expression feature value, the posture feature value, the sound feature value, the speech rate feature value and the text feature value are fused to obtain a fused feature of the target employee at the time point;

[0022] The fusion feature values ​​of the target employees at each time point during the potential test are arranged in chronological order to obtain the fusion feature sequence of the target employees.

[0023] In an optional implementation, the potential data is input into a trained multimodal potential identification model to obtain a potential assessment score of the target employee, including:

[0024] Inputting the fused feature sequence of the target employee into the trained multimodal potential recognition model to obtain the probability that the target employee belongs to different potential categories;

[0025] Get the potential score range corresponding to different potential categories;

[0026] Based on the potential score range and the probability that the target employee belongs to different potential categories, a potential assessment score of the target employee is determined.

[0027] In an optional implementation, after obtaining the comprehensive evaluation result of the target employee, the method further includes:

[0028] Based on the comprehensive evaluation result and the configured talent classification rules, the target talent category corresponding to the target employee is determined; wherein the talent classification rules include: multiple talent categories and classification standards corresponding to each talent category.

[0029] In an optional implementation, the work data and the comprehensive evaluation result are input into a trained turnover risk prediction model to obtain a turnover risk prediction result of the target employee, including:

[0030] If the target talent category meets the preset talent category with the loss risk to be predicted, the work data of the target employee and the comprehensive evaluation result are input into the trained loss risk prediction model to obtain the loss risk prediction result of the target employee.

[0031] In an optional implementation, the work data also includes: the salary of the target employee;

[0032] The work data and the comprehensive evaluation results are input into the trained loss risk prediction model to obtain the loss risk prediction results of the target employees, including:

[0033] The comprehensive evaluation result, the work experience and the salary are input into a trained turnover risk prediction model to obtain a turnover risk prediction result of the target employee.

[0034] In a second aspect, a device for predicting employee turnover risk is provided, which may include:

[0035] An acquisition unit is used to acquire the target department and target position of the target employee, as well as the work data and potential data of the target employee; wherein the potential data includes a video of the target employee in the process of answering questions in the potential test and the answer results of the potential test;

[0036] A matching unit, used to match the target evaluation model corresponding to the target department and the target position from the configured comparison table of different departments, different positions and different evaluation models;

[0037] An evaluation unit, configured to evaluate the work data based on the target evaluation model to obtain a first evaluation score of the target employee; input the potential data into a trained multimodal potential recognition model to obtain a potential evaluation score of the target employee; and obtain a comprehensive evaluation result of the target employee based on the first evaluation score and the potential evaluation score;

[0038] The prediction unit is used to input the work data and the comprehensive evaluation result into the trained loss risk prediction model to obtain the loss risk prediction result of the target employee.

[0039] In a third aspect, an electronic device is provided, the electronic device comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0040] Memory, used to store computer programs;

[0041] The processor is used to implement any method step described in the first aspect when executing the program stored in the memory.

[0042] In a fourth aspect, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, any method step described in the first aspect is implemented.

[0043] This application conducts a comprehensive assessment of the employee's capabilities from multiple aspects such as performance, work experience, potential and achievements. The assessment is more comprehensive and can build a more three-dimensional employee portrait, which helps the company better understand the specific needs and characteristics of each employee; at the same time, combined with salary and capability assessment, it comprehensively and accurately predicts the employee's loss risk, helps to provide targeted intervention measures, thereby improving employee satisfaction, optimizing human resource management, and ensuring the continuity and stability of the company's business.

[0044] This application not only improves the pertinence and accuracy of assessment and turnover risk prediction, but also enables real-time monitoring and prediction of employee turnover risks, allowing companies to respond quickly to potential risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0046] Figure 1A flowchart of a method for predicting employee turnover risk provided in an embodiment of the present application;

[0047] Figure 2 A schematic diagram of the structure of an employee turnover risk prediction device provided in an embodiment of the present application;

[0048] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0050] The employee turnover risk prediction method provided in the embodiment of the present application can be applied in a server or in a terminal with strong computing power. The server can be a physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and basic cloud computing services such as big data and artificial intelligence platforms. The terminal can be a mobile phone, smart phone, laptop, digital broadcast receiver, personal digital assistant (PDA), tablet computer (PAD) and other employee equipment (User Equipment, UE), handheld device, vehicle-mounted device, wearable device, computing device or other processing equipment connected to a wireless modem, mobile station (Mobile Station, MS), mobile terminal (Mobile Terminal), etc. The terminal and the server can be directly or indirectly connected by wired or wireless communication, which is not limited in this application.

[0051] The preferred embodiments of the present application are described below in conjunction with the drawings in the specification. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application. In addition, the embodiments and features in the embodiments of the present application may be combined with each other if there is no conflict.

[0052] Figure 1 A flow chart of a method for predicting employee turnover risk provided in an embodiment of the present application. Figure 1 As shown, the method may include:

[0053] Step S110, obtaining the target department and target position of the target employee, as well as the target employee's work data and potential data; matching the target evaluation model corresponding to the target department and target position from the configured comparison table of different departments, different positions and different evaluation models.

[0054] In an embodiment of the present application, the work data of the target employee includes: work experience, historical achievements, salary and performance data; specifically, work experience includes: training experience, project experience, cross-professional field experience, front-line / front-end business practice and participation in innovative business experience, etc.; historical achievements include: personal historical achievements, the achievements of the team (or department) and / or personal influence; performance data includes job performance data and department performance data; at the same time, job performance data includes the quality, quantity and efficiency of work completed, etc. For employees in different positions, the categories of their job performance data and the corresponding evaluation standards are different. The categories of the job performance data and the corresponding evaluation standards can be preset by the enterprise, dynamically adjusted, or obtained through data mining technology or intelligent analysis model analysis.

[0055] In an embodiment of the present application, the potential data includes a video of the target employee in the process of answering the potential test and the results of the potential test (i.e., the answer text); specifically, the test questions include multiple-choice questions and oral questions; the video includes the employee's facial expressions captured in real time by the camera during the answering process and the audio data of the oral questions collected by the microphone.

[0056] In an embodiment of the present application, a potential test question bank is pre-constructed, in which a plurality of potential test questions are stored, covering potential test questions of different categories and different difficulty levels, and the potential test questions are used to test employees' innovation ability, self-drive and leadership, etc.; the initial potential test questions of the target employees can be obtained according to the target department and target position of the target employees; during the process of the target employees answering the questions, the initial potential test questions will be adjusted in real time according to the performance of the target employees.

[0057] In an embodiment of the present application, initial potential test questions are obtained by matching the target department and target position of the target employee, which may include: determining the first potential category required for the main goal of the target department and the second potential category required for the target position through data mining technology; obtaining the target potential category based on the required first potential category and second potential category; extracting potential test questions containing the target potential category from the potential test question bank; and analyzing the degree of influence of different potential categories on the results to obtain weight coefficients of each potential category, and determining the number of questions of different potential categories contained in the initial potential test questions according to the weight coefficients.

[0058] For example, the main goal of a commercialization department is related to innovation and self-motivation, and the employee's target position is a management position, which requires the employee to have leadership; at the same time, through analysis, it can be seen that leadership occupies a relatively important position, so its weight ratio is determined to be 5, the weight ratio of innovation and self-motivation is 2, and other potential is 1. Then, from the potential test question bank, 50% of leadership potential test questions, 20% of innovation potential test questions, 20% of self-motivation potential test questions and 10% of other potential test questions can be selected.

[0059] In an embodiment of the present application, the target evaluation model includes: a target performance evaluation model, a target achievement evaluation model and a target work experience evaluation model.

[0060] In the embodiment of the present application, the enterprise includes multiple departments, each department includes multiple positions, and the target performance evaluation model corresponding to any position in any department can be obtained by the following method:

[0061] Obtain first data related to the performance of the department and second data related to the performance of the position; preprocess the first data and the second data; use feature engineering to select multiple initial performance evaluation features from the preprocessed first data and the second data based on data such as the performance evaluation goals of the department and the job responsibilities of the position; use the obtained multiple initial performance evaluation features as training data, input them into the constructed random forest model, train the random forest model, and obtain importance scores corresponding to different initial performance evaluation features; based on the configured performance evaluation feature screening threshold, select the second performance evaluation feature from the multiple initial performance evaluation features; perform correlation analysis on the multiple second performance evaluation features and the business focus of the department and the current work focus of the position to obtain the correlation analysis results; based on the correlation analysis results, extract the target performance evaluation features and the weight coefficients of each target performance evaluation feature from the multiple second performance evaluation features to obtain the target performance evaluation model.

[0062] In the embodiment of the present application, the first data includes: the current business focus of the department, performance evaluation goals, performance data of employees in various positions in the department, the department's income, department's expenditures, department's profits, customer evaluations, and evaluations from other departments or leaders; the second data includes: the job responsibilities, ability requirements, and current work focus of the position; among them, the current business focus of the department and the current work focus of the position will change over time. For example, for a R&D position in a department, the talent required for the stage of building a technical architecture from 0 to 1 is different from that required for the maintenance stage, and the ability requirements for talent are different, so the performance evaluation model should also be different.

[0063] In the embodiment of the present application, preprocessing the first data and the second data includes: cleaning and standardizing the first data and the second data.

[0064] In an embodiment of the present application, the target achievement evaluation model and the target work experience evaluation model are the same as the target performance evaluation model. The final evaluation model is determined by analyzing relevant data of the target department and the target position and using random forest or other regression models and classification models.

[0065] In another embodiment of the present application, the target achievement evaluation model may include three fixed achievement evaluation characteristics, and the weight coefficients of the three achievement evaluation characteristics are adjusted by analyzing the relevant data of the target department and the target position, so as to obtain a target achievement evaluation model that matches the current stage of the target department and the target position; specifically, the three achievement evaluation characteristics include: personal achievement, team achievement and influence.

[0066] In another embodiment of the present application, the target work experience evaluation model may include four fixed experience evaluation characteristics, and the weight coefficients of the four experience evaluation characteristics are adjusted by analyzing the relevant data of the target department and the target position, so as to obtain a target work experience evaluation model that matches the current stage of the target department and the target position; specifically, the four experience evaluation characteristics include: cross-professional field experience, front-line / front-end business practice, participation in innovative business, and the degree of match with the target position.

[0067] In the embodiment of the present application, employees are divided into groups with different technical levels through K-means clustering. The deep learning model is used to explore the potential relationship of improving employees' technical capabilities. The collaborative filtering algorithm is used to select the evaluation tools or evaluation models of similar employees to recommend a suitable evaluation method for each employee.

[0068] Step S120: Evaluate the work data based on the target evaluation model to obtain a first evaluation score of the target employee; input the potential data into the trained multimodal potential recognition model to obtain a potential evaluation score of the target employee.

[0069] In the embodiment of the present application, the work data is evaluated based on the target evaluation model to obtain a first evaluation score of the target employee, including:

[0070] Based on the target performance evaluation model, the target achievement evaluation model and the target work experience evaluation model, the performance data, historical achievements and work experience are evaluated respectively to obtain the performance evaluation score, historical achievement evaluation score and work experience evaluation score of the target employee; based on the performance evaluation score, historical achievement evaluation score and work experience evaluation score, a first evaluation score is obtained.

[0071] In another embodiment of the present application, after obtaining the target evaluation model, corresponding data to be evaluated can be extracted from relevant data of the target employee according to the target evaluation features in the target evaluation model, and the data to be evaluated can be evaluated using the target evaluation model to obtain a first evaluation score for the target employee; the category of the data to be evaluated does not need to be consistent with the evaluation model; for example, the target performance evaluation model can extract performance data, or it can extract work experience and other data as the data to be evaluated, and then the target performance evaluation model can be used to evaluate the data to obtain a corresponding performance evaluation score.

[0072] For example, when it is at the stage of building a technical architecture from 0 to 1, the target performance evaluation characteristics and corresponding weight coefficients involved in the target performance evaluation model may include: technical architecture building ability, technical innovation ability, teamwork ability and work experience; among them, the technical architecture building ability is used to evaluate objective data such as the amount of code built in the technical architecture in past projects, the complexity of the architecture, the cutting-edge nature of the technology used, and its evaluation criteria or scoring criteria can be obtained by analyzing the same type of projects in the same industry; technical innovation ability is used to evaluate objective data such as the number of technical innovation results submitted by employees, the number of technical patents applied for, and contributions to the open source community; teamwork ability is used to evaluate data such as the frequency of communication between team members, the completion of collaborative tasks, and the efficiency of conflict resolution extracted from the project management system; background experience (including relevant project experience, participation in large-scale architecture construction, etc.) is used to evaluate the scale, technical difficulty, and achievements of related projects that employees have participated in in the past and whose matching degree with current business needs exceeds a certain threshold.

[0073] To give another example, assuming that the current business focus of a department is to develop new markets and increase market share, the target performance evaluation characteristics of sales positions may include: new customer development capabilities, which are used to evaluate the growth rate of the number of employees' new customers and the growth rate of sales brought by new customers. The specific scores can be obtained by comparing and analyzing the average level of companies of the same size in the same industry; sales performance (total sales, sales growth rate), which is used to compare the absolute value and growth rate of sales with the goals set by the company and the industry average; customer relationship maintenance capabilities, which are used to evaluate data such as the decline in the employee's customer complaint rate and the growth rate of customer repurchase rate; background resume (sales experience, industry reputation, etc.), which is used to evaluate the size of the company where the employee worked in the past, the degree of match with current business needs, sales performance ranking, sales awards received, etc.

[0074] In an embodiment of the present application, a first evaluation score is obtained based on the performance evaluation score, the historical achievement evaluation score, and the work experience evaluation score, including: directly adding the performance evaluation score, the historical achievement evaluation score, and the work experience evaluation score to obtain the first evaluation score; or, according to the target department and target position of the target employee, determining the weights of the performance evaluation score, the historical achievement evaluation score, and the work experience evaluation score, multiplying the performance evaluation score, the historical achievement evaluation score, and the work experience evaluation score by the corresponding weights, respectively, and adding them up to obtain the first evaluation score.

[0075] In an embodiment of the present application, the training process of the multimodal potential recognition model may include:

[0076] Obtain multiple categories of training data and a facial expression recognition model, a posture recognition model, a sound feature recognition model, a speech rate feature recognition model and a text feature extraction model based on a convolutional neural network to extract the features of the target employee's expression, voice, posture and the features of the answer text when answering questions; use the corresponding categories of training data to train the facial expression recognition model, the posture recognition model, the sound feature recognition model, the speech rate feature recognition model and the text feature extraction model respectively to obtain the trained facial expression recognition model, the posture recognition model, the sound feature recognition model, the speech rate feature recognition model and the text feature extraction model; through the multimodal fusion technology, the features of facial expression, posture, voice emotion and answer text are fused and trained, and a comprehensive multimodal potential recognition model is constructed based on the trained facial expression recognition model, the posture recognition model, the sound feature recognition model, the speech rate feature recognition model and the text feature extraction model, so that it can fully and accurately understand the employee's answering status and emotions, thereby obtaining the corresponding potential scoring results.

[0077] In an embodiment of the present application, the training data includes: a large number of facial images of employees showing different expressions (such as hesitation, pain, happiness, relaxation, etc.) when answering questions and expression annotations of each facial image, a large number of audio data of employees answering oral test questions and corresponding emotion annotations, and a large number of answer texts and corresponding text annotations; among which, the emotion annotations are the emotional characteristics reflected by the sound, such as tension, confidence, etc.; the text annotations include: fluency of language expression, confidence, etc.

[0078] In an embodiment of the present application, a facial expression recognition model (based on a convolutional neural network (CNN)) includes an input layer, a convolutional layer, and a fully connected layer; the input layer receives preprocessed facial image data, the convolutional layer uses multiple convolution kernels for feature extraction, and the pooling layer performs downsampling to reduce the data dimension; the fully connected layer maps the features extracted by convolution and pooling to the final expression category, and outputs the probability distribution through the Softmax function.

[0079] In the embodiment of the present application, the sound emotion recognition model uses audio processing technology and deep learning models to extract sound features, such as Mel-frequency cepstral coefficients (MFCC), after audio preprocessing (denoising, framing, etc.). Model training can use a recurrent neural network (RNN) or a long short-term memory network (LSTM).

[0080] In the embodiment of the present application, the answer text feature analysis model (natural language processing) is preprocessed (word segmentation, stop word removal, etc.) and then uses methods such as bag-of-words model, TF-IDF or word embedding to extract features. The model can be a convolutional neural network (CNN) or a Transformer architecture.

[0081] In the embodiment of the present application, before the potential data is input into the trained multimodal potential recognition model, the potential data needs to be processed, including:

[0082] For the video frames of the target employee at any time point during the potential test, the expression feature value, posture feature value, voice feature value and speech speed feature value of the target employee are extracted from the video frames;

[0083] Based on the facial expression feature values, posture feature values, sound feature values, and speech rate feature values, a feature matrix is ​​generated; feature extraction is performed on the potential test answer results to obtain text feature values; the facial expression feature values, posture feature values, sound feature values, speech rate feature values, and text feature values ​​are fused to obtain the fused features of the target employee at that time point; the fused feature values ​​of the target employee at each time point in the process of answering the potential test are arranged in chronological order to obtain the fused feature sequence of the target employee.

[0084] In the embodiment of the present application, feature extraction is performed on the potential test answer results to obtain text feature values, including:

[0085] Split the potential test results into words or phrases and remove stop words; use TF-IDF, Word2Vec or other methods to convert the text into a fixed-dimensional numerical vector.

[0086] In an embodiment of the present application, since there is not one text feature value at each time point, the text feature vector can be repeated multiple times to obtain a text feature matrix for fusion with the expression feature matrix, posture feature matrix, sound feature matrix, and speech rate feature matrix; the size of the text feature matrix is ​​similar or the same as that of the other feature matrices.

[0087] In an embodiment of the present application, when performing feature fusion, feature splicing can be directly performed to obtain fused features; different weight values ​​can also be set for the text feature matrix, expression feature matrix, posture feature matrix, sound feature matrix and speech rate feature matrix to perform weighted fusion.

[0088] In another embodiment of the present application, the expression feature value, posture feature value, sound feature value, speech rate feature value and text feature value are fused to obtain the fusion feature of the target employee at a time point, including:

[0089] For any time point, the expression feature value, posture feature value, sound feature value and speech rate feature value at that time point are fused to obtain the initial fused feature value; the attention layer is used to calculate the correlation between the first fused feature value and the text feature value at that time point to obtain the attention weight; the first fused feature value and the text feature value are weightedly fused according to the attention weight to obtain the fused feature.

[0090] In another embodiment of the present application, the fusion can also be performed according to a pre-set rule. For example, if the expression classification result shows confusion, the voice emotion recognition result shows anxiety, and the text feature analysis result shows uncertainty about the knowledge point, when these three conditions are met at the same time, it is determined that the employee is in a less than ideal state when answering the question, and this state is marked as Y; if these conditions are not met, it is marked as N.

[0091] In another embodiment of the present application, the questions answered by the target employee at different time points during the answering process can also be obtained synchronously, and the employee's answer feature vectors for different questions can be obtained based on the labels corresponding to the different questions and the employee's choice of the question; for any question, the employee's answer feature vector for the question and the facial expression feature value, posture feature value, sound feature value and speech speed feature value when answering the question are fused to obtain the fused features of the employee's answer to the question; based on the fused features of the employee when answering different questions and the labels of the corresponding questions, the target employee's fused feature sequence is obtained.

[0092] In an embodiment of the present application, feature fusion can adopt early fusion, directly splicing the features of facial expressions, voice emotions and answer text into a vector, and input it into a subsequent fusion model; it can also adopt late fusion, respectively obtain the prediction results of the three modalities, and then fuse them by weighted averaging, and the weights can be determined according to experiments or prior knowledge; it can also adopt intermediate fusion, in the middle layer of the model, the features of different modalities are interacted and fused.

[0093] In an embodiment of the present application, for facial images, expression features are extracted and input into a facial expression recognition model to obtain classification results. For sounds, sound features are extracted and input into a sound emotion recognition model to obtain classification results. For answer texts, text features are extracted and input into a natural language processing model to obtain analysis results. The features of the three modalities are integrated to comprehensively judge the employee's answering status and emotions. Based on the results of multimodal fusion, if the employee shows confusion or hesitation, the system selects related similar questions from the question bank for retesting. The employee's selection process and emotional expression are recorded, and the hesitation and pain indexes are calculated. After the test is completed, an answer score report is generated, including a subjective topic selection report and a system-inferred result analysis report.

[0094] In the embodiment of the present application, the potential data is input into the trained multimodal potential identification model to obtain the potential assessment score of the target employee, including:

[0095] The fused feature sequence of the target employee is input into the trained multimodal potential recognition model to obtain the probability that the target employee belongs to different potential categories; the potential score range corresponding to different potential categories is obtained; based on the potential score range and the probability that the target employee belongs to different potential categories, the potential assessment score of the target employee is determined.

[0096] In an embodiment of the present application, the multimodal potential recognition model is a multi-feature fused LSTM model, which can capture not only the facial image, sound and text features at each time point, but also the transformation features of the time series, thereby comprehensively analyzing the employee status during the entire answering process.

[0097] Assume that at time t1, the multimodal fusion result shows that the employee is in good condition, at time t2 it shows that the employee begins to be confused, and at time t3 the confusion worsens. Through LSTM, the temporal change trend is learned, and finally a comprehensive result based on the entire answering process is output.

[0098] Statistics are collected on the frequency of different states in the entire process of answering questions. For example, statistics are collected on the percentage of the number of times employees show negative states such as confusion and hesitation in the total time points. If this percentage exceeds a certain threshold (such as 50%), the employee's overall state in the entire process of answering questions is judged to be poor; if the percentage is lower than the threshold, the overall state is considered to be good. For example, some thresholds need to be set to divide high potential and low potential. Suppose a threshold of 80 points is set based on the comprehensive score of the employee's answering state and emotions. Scores above 80 points are high potential, and scores below 80 points are low potential.

[0099] In the embodiment of the present application, employees are divided into high-potential talents, medium-potential talents and low-potential talents; the probability of the target employee belonging to different potential categories is the probability of the target employee belonging to high-potential talents, medium-potential talents and low-potential talents.

[0100] In the embodiment of the present application, the potential score ranges corresponding to different potential categories can be customized as needed. For example, the score corresponding to high-potential talents can be defined as 8-10 points.

[0101] In the embodiment of the present application, based on the potential score range and the probability that the target employee belongs to different potential categories, the potential assessment score of the target employee is determined, including:

[0102] The standard values ​​of different potential categories can be determined based on the potential score ranges corresponding to different potential categories; for any potential category, the product of the standard value of the potential category and the probability that the target employee belongs to the potential category is used as the potential score of the target employee under the potential category; based on the potential scores of the target employee under different potential categories, the potential assessment score of the target employee is obtained.

[0103] In the embodiment of the present application, the standard value may be the middle number of the score range, or may be the standard deviation or other values.

[0104] For example, during the employee's answering process, data is collected once every second, and the entire answering process lasts T seconds.

[0105] Face image feature extraction: At each time point t (1≤t≤T), the expression feature value e is extracted from the face image. t (the value range is assumed to be [0,1], 0 represents a very negative expression, 1 represents a very positive expression) and the posture feature value p t The value range is assumed to be [0,1], where 0 represents a very improper posture and 1 represents a very proper posture). For example, at the 5th second, e5 = 0.8, p5 = 0..9.

[0106] Sound feature extraction: Also at each time point t, extract the volume feature value v of the sound t (the value range is assumed to be [0,1], 0 represents a very low volume, 1 represents a very high volume) and the speech rate feature value r t (The value range is assumed to be [0, 1], 0 represents a very slow speaking speed, and 1 represents a very fast speaking speed.) For example, at the 8th second, v8 = 0.6, r8 = 0.7.

[0107] Assign weights w to expression features, posture features, volume features, speech speed features and text features respectively e =0.2, w p =0.1, w v =0.2, w r =0.1, w c =0.4

[0108] At each time point t, the fused feature value x t For: x t=w e ×e t +w p ×p t +w v ×v t +w r × t +w c ×c t .

[0109] Use the LSTM model with multi-feature fusion to process time series: The fused feature sequence {x1, x2, …, x T} is input into the LSTM model. Assume that after being processed by the LSTM model with multi-feature fusion, an intermediate vector h2 (dimension is d) is output; the fully connected layer outputs the probability distribution: the intermediate vector h T After a fully connected layer, the probability distribution of employee potential at different levels is output.

[0110] Assuming that the potential is divided into three levels: high, medium, and low, the weight matrix of the fully connected layer is W = [w ij ](i=1,2,3,j=1,…,d), the bias vector is b=[b1,b2,b3].

[0111] The output probability distribution P = [p1, p2, p3], where

[0112]

[0113] The score is calculated according to the probability distribution: high potential corresponds to 8-10 points, medium potential corresponds to 4-7 points, low potential corresponds to 0-3 points, and the middle values ​​are 9, 5.5, and 1.5 respectively; the final score S = 9×p1+5.5×p2+1.5×p3; for example, if p1=0.6, p2=0.3, p3=0.1, then S=9×0.6+5.5×0.3+1.5×0.1=5.4+1.65+0.15=7.2 points.

[0114] In another embodiment of the present application, during the process of the target employee answering questions, the target employee will also be analyzed in real time, and the features extracted from expressions, sounds and texts will be integrated to comprehensively determine the employee's answering status and emotions. If the result of multimodal fusion shows that the employee is confused or hesitant when answering questions, and his eyes are fixed on option A, but he finally chooses B (obtained by extracting images and eye movements from the video and analyzing them), then the system will select questions related to or similar to question A from the question bank in the database for the answerer to choose, so as to determine whether the answerer deliberately avoids his real choice. If the employee still appears painful and hesitant when choosing, and finally chooses other questions, the system will record the employee's hesitation and pain index when choosing the question, and question the authenticity of this choice. On the contrary, if the employee shows confidence and ease, the system will push more challenging and comprehensive questions based on the tested content to further expand his knowledge and ability.

[0115] In actual applications, the employee's answer results include the comprehensive ability report corresponding to the answer sheet. The report is divided into two parts: one is the employee's subjective topic selection report, and the other is the result analysis report of the employee's real psychological dynamic speculation collected by the system through multimodal recognition technology during the answering process. For example, the original subjective answer score is 80 points, or the comprehensive ability coefficients are all above 5 (the full score is 10 points), but according to the extraction of multimodal data, it is found that the candidate cares about a certain choice and has a clear tendency, but avoids choosing it. Therefore, the system will select the answer to a question that the employee hesitated and looked at many times, and output the result for comparison with the subjective answer. The presented comparison data, employee subjective selection and system objective evaluation data are sent to the company's HR or managers to provide the company with objective and accurate data evaluation and give analysis suggestions.

[0116] Step S130: Based on the first evaluation score and the potential evaluation score, a comprehensive evaluation result of the target employee is obtained; the work data and the comprehensive evaluation result are input into the trained turnover risk prediction model to obtain the turnover risk prediction result of the target employee.

[0117] In the embodiment of the present application, the comprehensive evaluation result of the target employee includes: a potential evaluation score, a performance evaluation score, a historical achievement evaluation score, and a work experience evaluation score.

[0118] In the embodiment of the present application, after obtaining the comprehensive evaluation result of the target employee, the method further includes:

[0119] Based on the comprehensive evaluation results and the configured talent classification rules, the target talent category corresponding to the target employee is determined; if the target talent category meets the preset talent category with the loss risk to be predicted, the target employee's work data and comprehensive evaluation results are input into the trained loss risk prediction model to obtain the loss risk prediction result of the target employee.

[0120] In an embodiment of the present application, the performance scoring rules include: a first performance scoring threshold and a second performance scoring threshold; wherein the first performance scoring threshold is greater than the second performance scoring threshold; the achievement scoring rules include: a first achievement scoring threshold and a second achievement scoring threshold; wherein the first achievement scoring threshold is greater than the second achievement scoring threshold; the experience scoring rules include: a first experience scoring threshold and a second experience scoring threshold; wherein the first experience scoring threshold is greater than the second experience scoring threshold; the potential scoring rules include: a first potential scoring threshold and a second potential scoring threshold; wherein the first potential scoring threshold is greater than the second potential scoring threshold.

[0121] In an embodiment of the present application, if the performance evaluation score is greater than the first performance score threshold, the target employee is a high-performance employee; if the performance evaluation score is greater than the second performance score threshold and less than the first performance score threshold, the target employee is a medium-performance employee; if the performance evaluation score is less than the second performance score threshold, the target employee is a low-performance employee; similarly, based on the corresponding historical achievement evaluation score, work experience evaluation score and potential evaluation score, the category matching results of the target employee in terms of historical achievements, work experience and potential can be obtained.

[0122] In an embodiment of the present application, the talent classification rules include: multiple talent categories and classification standards corresponding to each talent category; the talent categories include: regular talents, potential stars, star talents, self-improvement talents, reliable contribution talents, outstanding talents, performance improvement talents, stable talents and attention talents.

[0123] In an embodiment of the present application, if the category matching results of the target employee in terms of performance evaluation, historical achievements, work experience and potential meet the classification criteria of a certain talent category, then the talent category is used as the target talent category of the target employee and is mapped in a pre-built enterprise talent inventory nine-square grid map.

[0124] In the embodiment of the present application, the categories of talents whose loss risks are to be predicted include: outstanding talents and key talents.

[0125] In an embodiment of the present application, after mapping the target talent category of the target employee to a pre-built nine-square grid map of enterprise talent inventory, a secondary calculation can be performed based on the data of the inventory results to infer the health of the organization's talent pool and generate an enterprise talent health summary report; at the same time, data mining is performed based on the data of outstanding talents and key talents to predict growth paths and staff loss risks and summarize them.

[0126] In the embodiment of the present application, the health calculation includes: calculating the number of talents in any area of ​​the nine-square grid. Assuming that the company has a total of N employees, and there are n1 talents in the high-performance-high-potential area, the talent ratio in this area is Similarly, calculate the talent ratio of other regions; for example, there are n2 talents in the medium performance-medium potential region, and their ratio is According to the enterprise strategy and business needs, different weights are set for different areas of the nine-square grid. For example, the weight of the high-performance-high-potential area is set to w1=0.3, and the weight of the medium-performance-high-potential area is set to w2=0.25, etc.; the health index H is calculated using the formula H=w1×p1+w2×p2+…;

[0127] When the total proportion of talents in high-potential areas (high performance-high potential, medium performance-high potential, and low performance-high potential areas) is between 30% and 40%, and the distribution of talents in each performance-potential combination area is relatively balanced (the proportion of talents in each area fluctuates roughly between 5% and 20%), the health level is considered normal;

[0128] If the total proportion of talents in high-potential areas is less than 30% (such as falling to 20% or even lower), or although there are no serious problems with the overall talent structure, there is a shortage or surplus of talents in certain key areas (such as the proportion of talents in high-performance-high-potential areas is less than 5%, while the proportion of talents in low-performance-low-potential areas exceeds 25%), then the health level begins to decline;

[0129] When the total proportion of talents in high-potential areas is less than 20% and continues to decline, or there is a serious imbalance in the distribution of talents (for example, most talents are concentrated in low-performance-low-potential areas, more than 40%, while there are almost no talents in high-performance-high-potential and other medium- and high-potential areas; or talents are overly concentrated in high-performance-high-potential areas, more than 50%, but there is an extreme shortage of talents in other areas), it is judged to be unhealthy.

[0130] Generate a summary report on the health of corporate talent based on the judgment results.

[0131] In an embodiment of the present application, the churn risk prediction model may adopt an XGBoost model, and train the churn risk prediction model using the historical work data and historical comprehensive evaluation results of multiple labeled historical employees in the company's historical database; wherein the labels are churned and not churned.

[0132] In an embodiment of the present application, the training method includes: preprocessing historical work data, historical comprehensive evaluation results and historical salaries, and selecting the features that have the greatest impact on the prediction of turnover risk through correlation analysis, feature importance evaluation and other methods; creating a new feature vector based on the features that have the greatest impact on the prediction of turnover risk, including combined features, time series features and interactive features, and performing feature transformations, such as logarithmic transformations and polynomial features; and then training the model using the training data to obtain a model that can predict employee turnover risk.

[0133] In an embodiment of the present application, the churn risk prediction model outputs not only the churn risk of the target employee, but also includes the importance score of the churn risk influencing features and local explanations, such as using SHAP values ​​to explain the prediction results of the target employee, thereby helping the company to formulate corresponding intervention measures, such as training, promotion opportunities, welfare improvements, etc.

[0134] In an embodiment of the present application, the method further includes: generating a report including a summary, analysis and suggestions, and pushing it to HR and managers.

[0135] In an embodiment of the present application, generating a report containing a summary, analysis, and suggestions includes: selecting a department, for example, selecting a commercialization department, and analyzing through data mining technology to find that the main goals of the department are related to data growth, personal innovation, and self-drive, and its departmental positions are relatively comprehensive, covering R&D, sales, products, testing, design, managers of various positions, etc. Therefore, it is recommended to use multi-dimensional evaluation tools according to different positions. For example, sales positions focus on performance and service, and it is recommended to use 360 ​​evaluation + performance + past performance + background resume for comprehensive consideration; at the same time, the model finds that product positions focus on innovation, self-drive, responsibility, leadership and other requirements, so the evaluation tool + past performance + background resume is used for comprehensive consideration.

[0136] Furthermore, based on the characteristics of the talents in this department, it is recommended to use the talent standard type of "personal contribution", which contains multiple statistical dimensions, and HR can also modify this type.

[0137] In terms of evaluation dimensions, it is recommended to use tools such as [capability, potential, experience, and performance] to identify, cultivate, and evaluate high-potential management talents. Finally, preliminary data on talent placement is obtained based on the evaluation results and multi-dimensional evaluation, forming a nine-square grid view, such as distinguishing performance as high, medium, and low, and management ability as high, medium, and low. Then, based on the multi-dimensional data evaluation, the talent results are allocated to each grid, and the proportion of personnel is calculated, such as regular talents, self-improvement talents, performance enhancers, stable talents, reliable contributors, potential stars, star talents, outstanding talents, and attention talents.

[0138] Based on its characteristics, among the various tools provided by the visualization view, such as multi-dimensional custom filtering, such as management ability-performance, potential-performance, and opening the third dimension view, managers can intuitively view / focus on the content they are concerned about.

[0139] Among them, in terms of calibration function, managers can make subjective evaluations based on multi-dimensional data, and comprehensively evaluate the talents they are concerned about during calibration meetings. For example, at present, "Han Xiaoxiao" is a stable talent with low performance and moderate management ability. During the performance appraisal calibration, the manager will increase the performance of "Han Xiaoxiao", so the position of "Han Xiaoxiao" in the nine-square grid will change. This change can be human intervention or a change made by the system's intelligent calculation. When personnel change, the personnel position will change accordingly. These changes will be marked by the system through the superscript after the personnel name, such as rising two grids and distinguishing the content of the rise, such as left, right, up and down. The arrow markings are different according to the rising position, and the specific rising numbers are displayed accordingly, such as rising "2" grids, rising from stable talents to potential stars, and the arrow is upward. By marking in this way, managers can comprehensively evaluate the potential of personnel based on multi-dimensional data such as talent placement and personnel calibration adjustments (such as rising, falling, and staying the same). Through this tool, managers can intuitively view the corresponding data and pay attention to the status of the corresponding personnel in a timely manner. For example, a person who was a key potential star last quarter becomes a performance improver in the next quarter and is in the observation area. Finally, the relevant calibration data can also be used to display the magnitude of the movement and calibration ratio compared with the previous inventory.

[0140] At the same time, there are 52 star talents, but this position or department does not need so many successors. At this time, we provide VS talent comparison PK function, select the corresponding talents for multi-dimensional comparison again, and set the corresponding ratio. The system will compare the multi-dimensional data according to the set ratio, and recommend the corresponding personnel to retain the current position according to the data comparison, and give the corresponding personnel who need to be demoted and removed from this position. For example, in the multi-dimensional re-comparison, it was found that "Ping Jianzheng", "He Chenlu" and "Yan Sanmu" were insufficient in enthusiasm, teamwork and innovation compared with others, and they were promoted slowly or not at all in the past three years. Therefore, it is judged that they have obvious shortcomings compared with other candidates, and it is recommended that they be adjusted to "excellent talents".

[0141] Finally, a complete talent inventory map is formed, and secondary calculations are performed based on the data of the inventory results to infer the health of the organization's talent pool. Data mining is performed based on the data of outstanding talents and key talents to predict growth paths and staff loss risks and summarize them. For example, if calculations find that talent A has recently frequently browsed recruitment-related websites, paid attention to competitor recruitment information, and has a lower work attitude than before, managers and HR are recommended to pay special attention to prevent staff loss; furthermore, if it is found that there are too many star talents, insufficient room for talent advancement, or too few reserve talents, and there is a risk of a gap, artificial intelligence technology and data analysis models are used to form a report analysis, and risk prevention suggestions are given, and the report is finally pushed to the corresponding HR and managers.

[0142] Corresponding to the above method, the embodiment of the present application also provides an employee turnover risk prediction device, such as Figure 2 As shown, the employee turnover risk prediction device includes:

[0143] The acquisition unit 210 is used to acquire the target department and target position of the target employee, as well as the work data and potential data of the target employee; wherein the potential data includes the video of the target employee in the process of answering questions in the potential test and the answer results of the potential test;

[0144] A matching unit 220 is used to match the target department and the target position with the target evaluation model from the configured comparison table of different departments, different positions and different evaluation models;

[0145] The evaluation unit 230 is used to evaluate the work data based on the target evaluation model to obtain a first evaluation score of the target employee; input the potential data into the trained multimodal potential recognition model to obtain a potential evaluation score of the target employee; and obtain a comprehensive evaluation result of the target employee based on the first evaluation score and the potential evaluation score;

[0146] The prediction unit 240 is used to input the work data and the comprehensive evaluation results into the trained turnover risk prediction model to obtain the turnover risk prediction results of the target employees.

[0147] The functions of each functional unit of the employee turnover risk prediction device provided in the above-mentioned embodiments of the present application can be realized through the above-mentioned method steps. Therefore, the specific working process and beneficial effects of each unit in the employee turnover risk prediction device provided in the embodiments of the present application will not be repeated here.

[0148] The present application also provides an electronic device, such as Figure 3 As shown, it includes a processor 310 , a communication interface 320 , a memory 330 and a communication bus 340 , wherein the processor 310 , the communication interface 320 , and the memory 330 communicate with each other via the communication bus 340 .

[0149] Memory 330, for storing computer programs;

[0150] The processor 310 is used to execute the program stored in the memory 330 to implement the following steps:

[0151] Obtain the target department and target position of the target employee, as well as the target employee's work data and potential data; the potential data includes the video of the target employee answering questions in the potential test and the results of the potential test;

[0152] From the comparison table of different departments, different positions and different evaluation models, match the target evaluation model corresponding to the target department and target position;

[0153] Evaluate the work data based on the target evaluation model to obtain the first evaluation score of the target employee; input the potential data into the trained multimodal potential recognition model to obtain the potential evaluation score of the target employee; obtain the comprehensive evaluation result of the target employee based on the first evaluation score and the potential evaluation score;

[0154] The work data and comprehensive evaluation results are input into the trained turnover risk prediction model to obtain the turnover risk prediction results of the target employees.

[0155] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0156] The communication interface is used for communication between the above electronic device and other devices.

[0157] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0158] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0159] The implementation methods and beneficial effects of the components of the electronic device in the above embodiments to solve the problems can be seen in Figure 1 The various steps in the illustrated embodiment are implemented, therefore, the specific working process and beneficial effects of the electronic device provided by the embodiment of the present application are not repeated here.

[0160] In another embodiment provided in the present application, a computer-readable storage medium is provided, in which instructions are stored. When the computer-readable storage medium is executed on a computer, the computer executes any one of the employee turnover risk prediction methods in the above embodiments.

[0161] In another embodiment provided by the present application, a computer program product including instructions is also provided, which, when executed on a computer, enables the computer to execute any of the employee turnover risk prediction methods in the above embodiments.

[0162] Those skilled in the art will appreciate that the embodiments in the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt a complete hardware embodiment, a complete software embodiment, or a form of an embodiment combining software and hardware. Moreover, the present application may adopt a form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0163] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0164] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0165] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0166] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present application.

[0167] Obviously, those skilled in the art can make various changes and modifications to the embodiments in the present application without departing from the spirit and scope of the embodiments in the present application. Thus, if these modifications and variations of the embodiments in the present application fall within the scope of the claims and their equivalents in the embodiments of the present application, the embodiments of the present application are also intended to include these modifications and variations.

Claims

1. A method for predicting employee turnover risk, characterized in that: The method comprises: Obtain the target department and target position of the target employee, as well as the target employee's work data and potential data; wherein the potential data includes a video of the target employee during the process of answering questions in the potential test and the results of the potential test; Match the target evaluation model corresponding to the target department and the target position from the configured comparison table of different departments, different positions and different evaluation models; Evaluate the work data based on the target evaluation model to obtain a first evaluation score for the target employee; Inputting the potential data into a trained multimodal potential identification model to obtain a potential assessment score of the target employee; Obtaining a comprehensive evaluation result of the target employee based on the first evaluation score and the potential evaluation score; The work data and the comprehensive evaluation results are input into a trained turnover risk prediction model to obtain a turnover risk prediction result of the target employee.

2. The method according to claim 1, characterized in that The work data includes: work experience, historical achievements and performance data; The target evaluation model includes: a target performance evaluation model, a target achievement evaluation model and a target work experience evaluation model; The work data is evaluated based on the target evaluation model to obtain a first evaluation score of the target employee, including: Based on the target performance evaluation model, the target achievement evaluation model and the target work experience evaluation model, the performance data, the historical achievements and the work experience are evaluated respectively to obtain the performance evaluation score, the historical achievement evaluation score and the work experience evaluation score of the target employee; Based on the performance evaluation score, the historical achievement evaluation score and the work experience evaluation score, a first evaluation score is obtained.

3. The method according to claim 1, characterized in that Before inputting the potential data into the trained multimodal potential identification model to obtain the potential assessment score of the target employee, the method further includes: For a video frame of a target employee at any time point during the process of answering questions in the potential test, extract the expression feature value, posture feature value, voice feature value and speech speed feature value of the target employee from the video frame; Generate a feature matrix based on the expression feature value, posture feature value, sound feature value, and speech rate feature value; Performing feature extraction on the potential test answer results to obtain text feature values; The expression feature value, the posture feature value, the sound feature value, the speech rate feature value and the text feature value are fused to obtain a fused feature of the target employee at the time point; The fusion feature values ​​of the target employees at each time point during the potential test are arranged in chronological order to obtain the fusion feature sequence of the target employees.

4. The method according to claim 3, characterized in that The potential data is input into the trained multimodal potential identification model to obtain the potential assessment score of the target employee, including: Inputting the fused feature sequence of the target employee into the trained multimodal potential recognition model to obtain the probability that the target employee belongs to different potential categories; Get the potential score range corresponding to different potential categories; Based on the potential score range and the probability that the target employee belongs to different potential categories, the potential assessment score of the target employee is determined.

5. The method according to claim 1, characterized in that After obtaining the comprehensive evaluation result of the target employee, the method further includes: Based on the comprehensive evaluation result and the configured talent classification rules, the target talent category corresponding to the target employee is determined; wherein the talent classification rules include: multiple talent categories and classification standards corresponding to each talent category.

6. The method according to claim 5, characterized in that The work data and the comprehensive evaluation results are input into the trained loss risk prediction model to obtain the loss risk prediction results of the target employees, including: If the target talent category meets the preset talent category with the loss risk to be predicted, the work data of the target employee and the comprehensive evaluation result are input into the trained loss risk prediction model to obtain the loss risk prediction result of the target employee.

7. The method according to claim 2, characterized in that The job data also includes: the target employee’s salary; The work data and the comprehensive evaluation results are input into the trained loss risk prediction model to obtain the loss risk prediction results of the target employees, including: The comprehensive evaluation result, the work experience and the salary are input into a trained turnover risk prediction model to obtain a turnover risk prediction result of the target employee.

8. An employee turnover risk prediction device, characterized in that: The device comprises: An acquisition unit is used to acquire the target department and target position of the target employee, as well as the work data and potential data of the target employee; wherein the potential data includes a video of the target employee in the process of answering questions in the potential test and the answer results of the potential test; A matching unit, used to match the target evaluation model corresponding to the target department and the target position from the configured comparison table of different departments, different positions and different evaluation models; An evaluation unit, configured to evaluate the work data based on the target evaluation model to obtain a first evaluation score of the target employee; input the potential data into a trained multimodal potential recognition model to obtain a potential evaluation score of the target employee; and obtain a comprehensive evaluation result of the target employee based on the first evaluation score and the potential evaluation score; The prediction unit is used to input the work data and the comprehensive evaluation result into the trained loss risk prediction model to obtain the loss risk prediction result of the target employee.

9. An electronic device, characterized in that: The electronic device comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory, used to store computer programs; A processor, for implementing any of the methods described in claims 1-7 when executing a program stored in a memory.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images