Employee demission portrait model construction method based on fusion of multi-source heterogeneous data
By integrating the basic attributes, behavior logs and emotional data of employees, and building multi-dimensional employee behavioral portraits, the problem of insufficient utilization of employee data in the existing technology is solved, and the accuracy of employee resignation risks and the retention of high-value talents is achieved. Improve the efficiency of human resource management and scientific decision-making.
Patent Information
- Application Number
- CN202411131815.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-16
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing technology lacks the integration and utilization of multi-source heterogeneous employee data in human resource management, making it difficult to fully explore the value contained in different types of data, and cannot form a comprehensive portrayal of employees, especially in employee resignation risk warning and high-value talents. There are shortcomings in identification.
By collecting basic attribute data and behavior log data of employees, building multi-dimensional employee behavior portraits, and combining employee text data on various media channels for sentiment analysis, calculating negative emotions density, forming a comprehensive resignation risk score, and then portraying employees A two-dimensional matrix of resignation risks and match personalized retention plans for high-risk, high-value employees.
It realizes accurate identification and early warning of employee resignation risks, improves the efficiency of human resource management, helps enterprises identify and retain high-value employees, reduces the risk of high-value employees, and provides scientific decision-making support.
Smart Images

Figure CN119025919A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of human resource management, and more specifically, to a method for constructing an employee resignation portrait model based on the fusion of multi-source heterogeneous data. Background Art
[0002] Traditional human resource management mainly relies on manual experience and judgment, and has problems such as low efficiency and strong subjectivity. Therefore, it is urgent to use artificial intelligence technology to develop an intelligent human resource management system to achieve accurate employee portraits and resignation risk warnings, thereby providing strong support for corporate talent management decisions.
[0003] At present, domestic and foreign scholars have done some exploratory research on risk prediction and early warning in human resource management. Tang et al. (Tang M, Zhao T, Hu Z, et al. Research on Risk Prediction and Early Warning of Human Resource Management Based on Machine Learning and Ontology Reasoning[J]. vjesnik, 2023, 30(6): 2036-2045.) proposed a human resource management risk prediction and early warning method based on machine learning and ontology reasoning. The method first evaluates the similarity of the source cases in the case library by calculating the similarity of concept names and attributes, and realizes the similarity prediction of knowledge workers and company cases and the prediction of human resource management risks. Then, based on the evaluation results, the most suitable job match is found in specific risk problems and situations, providing solutions for companies to evaluate candidates. Secondly, they developed a machine learning prediction model and optimized the model through cross-validation and regularization methods, and finally achieved a prediction accuracy of 82%. This study has made useful explorations in the field of human resource management risk prediction and early warning, but there are still some shortcomings: the method mainly focuses on the matching problem between candidates and positions, and involves less early warning of the resignation risk of employees in office. It only considers the impact of employees' personal attributes on the resignation risk, and ignores factors such as employees' behavioral performance and emotional state in the organization.
[0004] Zhou et al. (Zhou D. [Retracted] Application of Data Mining Technology in Enterprise Digital Human Resource Management [J]. Security and Communication Networks, 2022, 2022 (1): 7611623.) studied the application of big data mining and analysis technology in enterprise digital human resource management. They used the risk management theory of quantitative enterprise management data characteristics to analyze the management model and business data in the human resource information system, introduced quantitative management theory, and introduced the basic concepts of data feature mining system and six commonly used data analysis and calculation methods. Through case analysis, the data mining theory was applied to the enterprise human resource project management information system. The experimental results show that the application effect of big data mining and analysis technology in talent type analysis, talent loss prevention, performance evaluation, etc. has increased by 25%. This study mainly explores the application of big data technology in the digital transformation of human resource management. It has certain theoretical and practical value, but it also has limitations: lack of analysis at the individual level of employees, it is difficult to accurately profile employees, the data dimension is single, mainly limited to structured data, and the processing of unstructured data such as text, audio and video is insufficient. There is not much involvement in the identification and targeted retention of high-value talents.
[0005] In summary, although existing research has made some attempts in the field of risk prediction and early warning in human resource management, it lacks the integrated utilization of multi-source heterogeneous employee data, making it difficult to fully tap the value contained in different types of data and form a comprehensive characterization of employees. Summary of the invention
[0006] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method for constructing an employee resignation portrait model based on the fusion of multi-source heterogeneous data.
[0007] To achieve the above object, the present invention provides the following technical solutions:
[0008] The method for constructing an employee resignation portrait model based on the integration of multi-source heterogeneous data includes:
[0009] Collect employees' basic attribute data and set their initial risk level for resignation based on the basic attribute data; collect employee behavior log data generated in their daily work and build a multi-dimensional employee behavior profile based on the basic attribute data and behavior log data. Modify the initial risk level for resignation based on the employee behavior profile to form a second risk level for resignation;
[0010] Collect employees' text data on various media channels, build employee sentiment dictionaries based on the text data, judge the sentiment tendency of the text data, and obtain employee sentiment analysis results; calculate negative sentiment density based on employee sentiment analysis results, set hierarchical negative sentiment density thresholds, and trigger sentiment warnings when employees' negative sentiment density exceeds the density threshold of the level for multiple consecutive cycles;
[0011] Calculate the employee value index and form a comprehensive turnover risk score based on the employee's second turnover risk level, negative emotion density and employee value index; characterize a two-dimensional matrix of employee turnover risk based on the comprehensive turnover risk score; match the best personalized retention plan for high-risk and high-value employees.
[0012] Further, the method comprises:
[0013] Obtain the basic attribute data of employees, structure the unstructured basic attribute data, convert it into numerical variables, and form a structured employee attribute table; based on the employee attribute table, construct employee attribute characteristics, divide employees into initial resignation risks according to employee attribute characteristics, and set the initial resignation risk level of employees;
[0014] Collect the behavior log data generated by employees in their daily work, characterize the key behavior characteristics of employees based on the behavior log data, and combine it with the employee attribute characteristics to build a multi-dimensional employee behavior portrait;
[0015] According to the employee behavior profile, risk weights for key behavior characteristics are set, and the initial resignation risk level is weighted and modified according to the risk weights to form a second resignation risk level;
[0016] Set a warning threshold for the second risk level of resignation. When the second risk level of resignation exceeds the warning threshold, an alert is triggered and a list of key employees is generated, prompting the HR department and direct supervisor to pay attention and communicate in a timely manner.
[0017] Further, the method comprises:
[0018] The employee attribute table contains categorical variables and numerical variables;
[0019] Generate employee attribute categorical features from categorical variables;
[0020] Generate employee attribute numerical features from numerical variables;
[0021] Generate new characteristics of employee career based on employee attribute table;
[0022] Combine employee attribute category characteristics, employee attribute numerical characteristics and employee career new characteristics to form employee attribute characteristics;
[0023] Filter out a subset of employee attribute characteristics whose correlation with the risk of resignation is greater than a relevant threshold from the employee attribute characteristics;
[0024] Construct an initial resignation risk warning model, predict the resignation risk of all employees based on the employee attribute feature subset and the initial resignation risk warning model, output the resignation risk prediction probability, and divide the resignation risk prediction probability into three levels: high, medium, and low to form the initial resignation risk level of employees;
[0025] High risk: predicted probability of resignation risk c≥0.6;
[0026] Medium risk: predicted probability of leaving the company is 0.3 <c<0.6;
[0027] Low risk: The predicted probability of resignation risk c≤0.3.
[0028] Further, the method comprises:
[0029] Collect the behavior log data generated by employees in their daily work from multiple data sources, associate and integrate the behavior log data from different sources according to employee ID and timestamp to form a structured employee behavior data set;
[0030] Identify key behavioral factors that influence employee performance and turnover from employee behavior datasets;
[0031] The employee behavior dataset is divided into behavior time series data according to the time window. Based on the behavior time series data, the frequent subsequences of employee behavior evolution are found as the sequence pattern of employee behavior.
[0032] Extract the key influencing factors of turnover risk from the key behavioral factors and quantify them into key behavioral characteristics;
[0033] Extract the evolutionary trajectory of employee behavior from the sequence pattern and summarize it into the characteristics of behavior change trend;
[0034] Combine employee attribute characteristics, key behavior characteristics and behavior change trend characteristics to construct a multi-dimensional employee behavior portrait.
[0035] Furthermore, the method of weighting and modifying the initial resignation risk level according to the risk weight to form a second resignation risk level includes:
[0036]
[0037] in, For the The second risk level of employee turnover, For the The initial turnover risk level of an employee, n is the number of key behavioral characteristics, j is the index of the key behavioral characteristics, and w is thej is the risk weight of the jth key behavioral characteristic, μ j is the mean value of the jth key behavior feature, For the The performance value of an employee on the jth key behavioral characteristic, σ j is the standard deviation of the jth key behavior feature, K is the adjustment coefficient, and round(·) is the rounding function.
[0038] Furthermore, the method for calculating the negative emotion density based on the employee emotion analysis result includes:
[0039]
[0040] Among them, T i is the timestamp of the ith sentence, T max is the end time of the time window, S i is the sentiment score of the i-th sentence, N is the total number of sentences in the time window, and w i is the time weight of the i-th sentence, λ is the decay coefficient, D neg is the density of negative emotions; δ(S i ) is the negative sentiment indicator function of the i-th sentence, if S i <0, then δ(S i )=1, otherwise δ(S i )=0.
[0041] Further, the method comprises:
[0042] Quantify the value contribution of employees to the organization from multiple dimensions, generate multi-dimensional contribution indicators, perform weighted summation of the contribution indicators of each dimension to form an employee comprehensive value score, normalize the comprehensive value scores of all employees, generate an employee value percentage ranking as the employee value index;
[0043] The second resignation risk level, negative emotion density and employee value index are standardized and unified to form three risk sub-factors;
[0044] Fit the relationship between the second turnover risk level, negative emotion density, employee value index and actual employee turnover rate, and estimate the regression coefficients of the three risk sub-factors;
[0045] The regression coefficients of the three risk sub-factors are used as weights, and the risk sub-factors are weighted and summed to generate a comprehensive turnover risk score;
[0046] The comprehensive turnover risk score is divided into three turnover risk levels: high, medium, and low, which serve as the horizontal axis of the matrix;
[0047] The employee value index is divided into three employee value levels: high, medium, and low, which serve as the vertical axis of the matrix;
[0048] The employee value levels and turnover risk levels are combined in pairs to form a 3×3 turnover risk matrix. Employees in the upper right corner of the matrix are high-risk and high-value employees.
[0049] The system for constructing an employee resignation portrait model based on the fusion of multi-source heterogeneous data is used to implement the above-mentioned method for constructing an employee resignation portrait model based on the fusion of multi-source heterogeneous data. The system includes:
[0050] Data collection module: used to collect employees’ basic attribute data and behavioral log data generated in their daily work from multiple data sources, as well as employees’ text data on various media channels;
[0051] Data processing module: used to structure the collected unstructured basic attribute data and associate and integrate the behavior log data from different sources according to employee ID and timestamp;
[0052] Employee behavior profile building module: used to characterize key employee behavior characteristics based on behavior log data, and combine it with employee attribute characteristics to build a multi-dimensional employee behavior profile;
[0053] Employee resignation risk level classification module: used to set the initial resignation risk level of employees based on their basic attribute data; set the risk weights of key behavioral characteristics based on employee behavior profiles, and perform weighted corrections on the initial resignation risk level based on the risk weights to form a second resignation risk level;
[0054] Sentiment analysis module: used to build an employee sentiment dictionary based on the text data of employees on various media channels, judge the sentiment tendency of the text data, obtain employee sentiment analysis results, and calculate the negative sentiment density based on the employee sentiment analysis results;
[0055] Employee value assessment module: used to quantify the value contribution of employees to the organization from multiple dimensions, generate multi-dimensional contribution indicators, perform weighted summation of the contribution indicators of each dimension to form an employee comprehensive value score, normalize the comprehensive value scores of all employees, and generate an employee value percentage ranking as an employee value index;
[0056] Resignation risk assessment module: used to form a comprehensive resignation risk score based on the employee's second resignation risk level, negative emotion density and employee value index, and to characterize the resignation risk matrix;
[0057] Personalized retention plan generation module: used to match the best personalized retention plan for high-risk and high-value employees.
[0058] An electronic device includes a memory, a central processing unit, and a computer program stored in the memory and executable on the central processing unit. When the central processing unit executes the computer program, the method for constructing an employee resignation portrait model based on the fusion of multi-source heterogeneous data is implemented.
[0059] A computer-readable storage medium having a computer program stored thereon, which, when executed, implements the above-mentioned method for constructing an employee resignation portrait model based on the fusion of multi-source heterogeneous data.
[0060] Compared with the prior art, the present invention has the following beneficial effects:
[0061] By comprehensively analyzing employees' basic attributes, behavior logs, and emotional data, the present invention can construct employee behavior portraits from multiple dimensions and accurately depict employees' risk factors for leaving the company. The initial risk level of leaving the company is corrected using an AI algorithm to form a more accurate second risk level of leaving the company, thereby improving the accuracy of leaving risk identification.
[0062] The present invention collects text data of employees on various media channels, constructs an employee sentiment dictionary, determines the sentiment tendency of the text data, and obtains employee sentiment analysis results. By calculating the negative sentiment density and setting hierarchical negative sentiment density thresholds, it is possible to timely detect and warn employees of negative emotions, which helps companies take measures in advance to avoid employee resignation due to emotional problems.
[0063] This method quantifies the value contribution of employees to the organization from multiple dimensions, generates multi-dimensional contribution indicators, forms a comprehensive employee value score, and performs normalization to generate an employee value index. The standardized employee value index, combined with the risk level of resignation, provides a more comprehensive employee value assessment, which helps companies better identify and retain high-value employees.
[0064] According to the comprehensive turnover risk score and value index of employees, the present invention can depict a two-dimensional matrix of employee turnover risk and match the best personalized retention plan for high-risk and high-value employees. This personalized retention strategy can effectively improve employee satisfaction and loyalty and reduce the risk of loss of high-value employees.
[0065] Through automated data collection, processing and analysis, the present invention greatly reduces the workload of manual intervention and improves the efficiency of human resource management. The HR department can focus more on strategy formulation and employee relationship management to improve the overall management effect. Through AI algorithms and data analysis, the present invention provides a scientific basis and data support for the human resource management decision-making of enterprises, reduces the subjectivity and blindness of decision-making, and improves the scientificity and effectiveness of management decisions. BRIEF DESCRIPTION OF THE DRAWINGS
[0066] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0067] Figure 1 It is a flow chart of the method for constructing an employee resignation portrait model based on the fusion of multi-source heterogeneous data in the present invention;
[0068] Figure 2 A flowchart of a method for generating association rules between behaviors in a method for constructing an employee resignation portrait model based on integration of multi-source heterogeneous data of the present invention;
[0069] Figure 3 A flowchart of a method for constructing a multi-dimensional employee behavior portrait in a method for constructing an employee resignation portrait model based on integration of multi-source heterogeneous data of the present invention;
[0070] Figure 4 This is a functional module diagram of the system for building an employee resignation portrait model based on the fusion of multi-source heterogeneous data in the present invention. DETAILED DESCRIPTION
[0071] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0072] Example 1
[0073] See also Figure 1 As shown, this embodiment provides a method for constructing an employee resignation portrait model based on integrating multi-source heterogeneous data, including:
[0074] Step S1000, collecting basic attribute data of employees, setting the initial risk level of employees according to the basic attribute data; collecting behavioral log data generated by employees in their daily work, constructing a multi-dimensional employee behavior profile according to the basic attribute data and the behavior log data, and revising the initial risk level of employees according to the employee behavior profile to form a second risk level of employees;
[0075] Furthermore, step S1000 includes:
[0076] Step S1100, collecting basic attribute data of employees, using a decision tree algorithm to classify employees into initial resignation risk levels based on the basic attribute data, and setting initial resignation risk levels for employees;
[0077] Furthermore, step S1100 includes:
[0078] Step S1110, obtaining basic attribute data of employees, cleaning the basic attribute data, removing abnormal values, and filling missing values; structuring the unstructured basic attribute data, converting them into numerical variables, and forming a structured employee attribute table;
[0079] Specifically, extract the basic attribute data of employees from the Human Resources Information Management System (HRIS), including age, gender, education, position, length of service, marital status, family situation, etc. Clean the data, remove outliers, and fill in missing values reasonably (such as using mean and median to fill continuous variables, and using mode to fill discrete variables). Structuralize unstructured data (such as education and position) and convert them into numerical variables. Finally, form a structured and standardized employee attribute table.
[0080] Step S1120, constructing employee attribute features based on the employee attribute table, and filtering out attribute feature subsets with high relevance to resignation risk from the employee attribute features; the employee attribute table includes categorical variables and numerical variables;
[0081] Further, step S1120 includes:
[0082] Step S1121, one-hot encoding or serial number encoding is performed on the categorical variable to generate employee attribute categorical features;
[0083] Step S1122, normalizing the numerical variables to generate numerical features of employee attributes;
[0084] One-hot encoding or serial number encoding is performed on categorical variables (such as gender, education, and position), and discrete values are mapped to numerical values to generate categorical features of employee attributes. Numerical variables are normalized to eliminate dimensional differences and generate numerical features of employee attributes.
[0085] Step S1123, generating new features of the employee's career based on the employee attribute table;
[0086] Specifically, the purpose of step S1123 is to use the existing employee attribute table and construct some new and more business-significant combination features based on the professional knowledge and experience of human resource management. Taking the ratio of company length to age as an example, it can be calculated according to the following formula:
[0087] Age ratio = employee's years of service in the company / employee's age;
[0088] This ratio can reflect the career stage and career stability of employees. Generally speaking, the higher the age-to-tenure ratio, the greater the proportion of the employee's working time in the company to his or her age. That is, the younger the age when joining the company, or the longer the time working in the company, often means that the closer the career is to the company, and the lower the risk of leaving. According to the distribution of the age-to-tenure ratio, some thresholds can be set to divide employees into different career stages, such as:
[0089] Age-to-year ratio < 0.2: Early career stage, high risk of leaving;
[0090] 0.2<=age-to-tenure ratio<0.5: mid-career stage, medium risk of leaving;
[0091] Age-to-tenure ratio >=0.5: Late career stage, low risk of leaving;
[0092] Similarly, we can also derive some other combined features, such as the average years of work in each company (total years of work ÷ number of companies worked for), years since the last promotion, etc., to enrich the description of employees' career development status.
[0093] Step S1124, combining the employee attribute category feature, the employee attribute numerical feature and the employee career new feature to form the employee attribute feature;
[0094] By combining employee attribute category characteristics, employee attribute numerical characteristics and employee career new characteristics, a comprehensive employee attribute characteristic is formed. This combination process makes employee attribute data richer and more multi-dimensional, and can more accurately reflect the personal characteristics and career development status of employees.
[0095] Step S1125, selecting a subset of employee attribute characteristics whose relevance to the risk of resignation is greater than a relevant threshold from the employee attribute characteristics;
[0096] The calculation method of the relevant threshold includes:
[0097]
[0098] in:
[0099] θ: correlation threshold, used to filter out the subset of employee attribute features that are more correlated with turnover risk;
[0100] m: the number of employee attribute characteristics;
[0101] w k: The weight of the k-th employee attribute feature, which indicates the importance of this feature in predicting turnover risk, and can be determined by feature selection algorithms (such as recursive feature elimination) or expert knowledge;
[0102] r k : The correlation coefficient between the k-th employee attribute feature and the resignation label, indicating the linear or monotonic correlation between the feature and the resignation label.
[0103] ε: adjustment coefficient, used to balance the effects of feature correlation and feature variability;
[0104] σ k : The standard deviation of the k-th employee attribute feature, indicating the variability of this feature in the data set;
[0105] μ k : The mean of the k-th employee attribute feature, which represents the average value of the feature in the data set;
[0106] k: The index of employee attribute feature.
[0107] The weight of employee attribute characteristics w k Indicates the importance of each feature in predicting turnover risk. The larger the weight, the more important the feature. k Indicates the correlation between the feature and the resignation label. The larger the absolute value, the stronger the correlation. ε is used to adjust the impact of feature correlation and feature variability, and balance the contribution of the two to the threshold. The formula comprehensively considers the importance, correlation, variability and standardization factors of the feature, and dynamically adjusts the relevant threshold to make the selected feature subset more accurate and representative. By dynamically adjusting the relevant threshold, it can more flexibly adapt to the characteristics of different data sets, improve the accuracy and reliability of resignation risk prediction, reduce redundant features, and improve the efficiency and explanatory power of the model.
[0108] The resignation label is a mark used to identify whether an employee has resigned. The resignation label is usually represented by 0 and 1, where 0 means that the employee has not resigned (is still employed) and 1 means that the employee has resigned. The Pearson correlation coefficient or Spearman rank correlation coefficient is used to calculate the correlation between each attribute feature and the resignation label, and the features with an absolute value of the correlation coefficient greater than θ are selected. These features have a high correlation with the resignation label.
[0109] Step S1130, constructing an initial resignation risk warning model, predicting the resignation risk of all employees based on the employee attribute feature subset and the initial resignation risk warning model, outputting the predicted resignation risk probability, and dividing the predicted resignation risk probability into three levels: high, medium, and low, to form the initial resignation risk level of the employees;
[0110] High risk: predicted probability of resignation risk c≥0.6;
[0111] Medium risk: predicted probability of leaving the company is 0.3 <c<0.6;
[0112] Low risk: predicted probability of resignation risk c≤0.3;
[0113] Specifically, the selected subset of employee attribute features is used as input, and whether the employee has resigned is the target variable (0-not resigned, 1-already resigned). The decision tree classification algorithm (such as CART, C4.5, random forest) is used to train the resignation risk prediction model. The historical resignation employee data is used as positive samples, and the current employee data is used as negative samples. The training set and test set are divided into 8:2 ratios. During the training process, the grid search combined with cross-validation is used to optimize the model hyperparameters (such as the maximum depth of the tree and the minimum number of samples of leaf nodes) to improve the generalization ability of the model. The model's classification accuracy, precision, recall rate, F1 value, ROC curve and other performance indicators are evaluated on the test set, and the model with the best overall performance is selected as the initial resignation risk warning model, and the model is used to predict the resignation risk probability of all current employees.
[0114] The basic attributes of employees are important factors that affect their turnover tendency. Generally speaking, the turnover rate of employees who are younger, have a shorter company tenure, and have a lower education level is relatively high, while the turnover rate of employees who are older, have a longer company tenure, and have a higher position is relatively low. This is because young employees have more employment options, while senior employees have a higher opportunity cost. At the same time, married employees are usually less likely to leave than unmarried employees because they have heavier family responsibilities. The initial turnover risk warning model trained with employee attribute data can quickly and roughly identify and classify the employee's turnover risk based on the "portrait" of the employee's personal characteristics, thereby helping the HR department to initially identify high-risk groups and formulate targeted warning and intervention measures. Of course, employee attributes can only reflect static turnover tendencies. To gain a more comprehensive and dynamic insight into the employee's turnover risk, it is necessary to further consider the employee's behavioral data and explore the employee's actual performance and psychological state in the organization. The initial turnover risk level provides a basis and direction for subsequent behavioral data analysis. The two complement each other to form a multi-dimensional assessment of employee turnover risk.
[0115] Step S1200, collecting behavior log data generated by employees in their daily work, characterizing key behavior characteristics of employees based on the behavior log data, and combining it with employee attribute characteristics to construct a multi-dimensional employee behavior portrait;
[0116] Further, step S1200 includes:
[0117] Step S1210, collecting behavior log data generated by employees in their daily work from multiple data sources, and associating and integrating the behavior log data from different sources according to employee ID and timestamp to form a structured employee behavior data set;
[0118] Specifically, employee behavior log data is scattered in different business systems, such as attendance system punch-in records, access control system entry and exit records, mail system sending and receiving records, collaborative office system operation records, etc. It is necessary to connect data with various business systems, collect key elements such as employee ID, timestamp, behavior type, and behavior object, and aggregate and associate the scattered behavior data according to employee ID and timestamp to form a structured employee behavior data set. Employee ID is used as the primary key to associate different behavior data of the same employee. The timestamp records the time when the behavior occurs, provides time series attributes for the behavior data, and is the basis for mining behavior time series patterns.
[0119] Step S1220, using an association rule mining algorithm to discover frequently co-occurring behavior patterns in the employee behavior data set, using association rules to characterize the correlation between different behaviors, and identifying key behavior factors that affect employee performance and resignation;
[0120] For further information, see Figure 2 , step S1220 includes:
[0121] Step S1221, preprocessing the employee behavior data set and encoding the behavior elements by 0-1;
[0122] Step S1222, each employee forms a behavior transaction, and the behavior transactions of all employees constitute a behavior transaction database;
[0123] Step S1223, finding frequently co-occurring behavior combinations based on the behavior transaction database, and generating association rules between the behaviors, where the association rules consist of antecedents and consequents;
[0124] Specifically, behavioral elements refer to the various components of specific behavioral records generated by employees in their daily work. These behavioral elements can include various specific actions or events related to employee work activities. For example, clock-in records: employee clock-in and clock-out times. Access control records: employee access control system records for entering and leaving the company. Email records: employee emails sent and received. Collaborative office operation records: employee operation records in the collaborative office system, such as document editing, task updating, meeting participation, etc. Based on the behavioral transaction database, frequent item set mining algorithms such as Apriori and FP-Growth are used to find frequently co-occurring behavioral combinations, and thus generate association rules between behaviors. Association rules are composed of antecedents and consequents. The antecedents are certain behavioral combinations of employees, and the consequents are highly correlated results, such as excellent performance or resignation. Through association rule mining, key behavioral factors that affect management results such as employee performance and resignation are identified, revealing the inherent relationship between certain employee behavioral combinations and management results. These key behavioral factors are the core elements for building employee behavioral profiles. For example, if {lateness, early departure} is identified as the key behavior combination that leads to poor performance, then indicators such as "frequency of lateness" and "frequency of early departure" should be used as key features of employee behavior profiles. For another example, if {overtime, high work intensity} is a key factor affecting employee resignation, then indicators such as "overtime hours" and "work intensity index" are the focus of describing employee behavior.
[0125] Step S1230, using a sequence pattern mining algorithm to discover the changing pattern of employee behavior over time, and using a sequence pattern to describe the evolution trajectory of employee behavior in the time dimension;
[0126] Specifically, step S1230 includes:
[0127] Step S1231, dividing the employee behavior data set into behavior time series data according to the time window;
[0128] Step S1232, based on the behavior time series data, find out the frequent subsequences of employee behavior evolution as the sequence pattern of employee behavior;
[0129] Specifically, taking a time window of fixed length (such as 1 day, 1 week) as a unit, the behavior of employees in the window is abstracted as a composite event, and the employee behavior thus forms time series data. On the time series data, sequential pattern mining algorithms such as GSP, SPADE, and PrefixSpan are used to find frequent subsequences of employee behavior evolution, which are the sequential patterns of employee behavior. The sequential pattern is composed of multiple composite events arranged in chronological order, which depicts the change process of employee behavior over time and reveals the dynamic evolution law of behavior. For example, the sequence pattern of <{late}, {absenteeism}, {resignation}> shows that from lateness to absenteeism to resignation, it presents the decline trajectory of the employee life cycle.
[0130] Step S1240, based on the key behavior factors and the sequence pattern of employee behavior, characterize the key behavior characteristics of employees, and combine them with the employee attribute characteristics to construct a multi-dimensional employee behavior portrait;
[0131] For further information, see Figure 3 , step S1240 includes:
[0132] Step S1241, extracting key influencing factors of turnover risk from key behavioral factors, and quantifying them into specific key behavioral features;
[0133] Step S1242, extracting the evolution trajectory of employee behavior from the sequence pattern and summarizing it as a behavior change trend feature;
[0134] Step S1243, combining employee attribute characteristics, key behavior characteristics, and behavior change trend characteristics to construct a multi-dimensional employee behavior profile;
[0135] Specifically, we extract the key factors affecting the risk of leaving the company from the results of association rule mining, such as "frequent lateness", "long overtime hours", "few email interactions", etc., and quantify them into specific key behavioral characteristics. These indicators should be able to reflect the frequency, intensity or change trend of behavioral factors, and calculate the specific value of each quantitative indicator based on the employee's behavioral data. These values can be frequency, duration, number of times, etc.
[0136] Examples: Frequent lateness;
[0137] Key behavioral factors: Frequent tardiness;
[0138] Quantitative indicators: frequency of lateness;
[0139] Calculation method:
[0140] Define the time window: for example, in months.
[0141] Count the number of lateness: Count the number of employees' lateness in a month.
[0142] Example calculation: Employee A was late 5 times in June, so his lateness frequency is 5.
[0143] The evolutionary trajectory of employee behavior is extracted from the results of sequential pattern mining, such as "the number of latenesses is gradually increasing" and "the degree of completion of work tasks is gradually decreasing", which are summarized as the characteristics of behavioral change trends as a dynamic supplement to the behavioral portrait. Finally, the employee attribute characteristics (such as age, gender, length of service, etc.), key behavioral characteristics (such as frequency of lateness, overtime hours, communication frequency, etc.), and behavioral change trend characteristics (such as attendance trends, performance trends, etc.) are combined to construct a multi-dimensional employee behavioral portrait. For example, the behavioral portrait of an employee can be summarized as: "A male born in the 1990s, who has been employed for 1 year, is frequently late and on an increasing trend, has an average overtime of 3 hours / day, exchanges less than 5 emails with the supervisor / week, and has a decreasing performance completion rate month by month." This portrait comprehensively portrays the employee from the perspectives of personal attributes, behavioral performance, and behavioral evolution, and intuitively presents the status characteristics of the employee in the organization. Based on the employee behavioral portrait, the initial resignation risk can be dynamically corrected to achieve a more accurate resignation warning. The HR department can also use behavioral portraits to provide targeted coaching and incentives to high-risk employees and retain talents in a personalized manner.
[0144] Step S1300, setting risk weights for key behavioral characteristics based on employee behavioral profiles, and weightedly modifying the initial resignation risk level based on the risk weights to form a second resignation risk level;
[0145] Specifically, the system matches and compares the actual behavior profile of each employee with the constructed key behavior characteristics (such as frequency of lateness, overtime hours, communication frequency, etc.), and calculates the employee's score on each key behavior characteristic, such as the matching degree between the employee's actual frequency of lateness and the key characteristic of "frequent lateness". At the same time, the similarity of the employee's actual behavior sequence and the mined abnormal behavior sequence pattern (such as from lateness to absenteeism to resignation) is calculated to determine the degree of consistency between the employee's behavior evolution trajectory and the typical negative pattern. Different risk weights are assigned to each key behavior characteristic according to the degree of correlation with the risk of resignation. The weight can be determined by mathematical statistics methods (such as logistic regression coefficients) or expert experience. The more key behavior characteristics an employee matches, the higher the weight of each characteristic, and the closer the behavior sequence pattern is to the negative decline trajectory, the greater the comprehensive risk weight value. If the employee's behavior is good, the key features are positive, the behavior sequence is rising steadily, and the risk weighting value is low, the initial level will be lowered from high risk to medium risk, or from medium risk to low risk; on the contrary, if the employee's behavior is abnormal, the key features are negative, the behavior sequence is decreasing, and the risk weighting value is high, the initial level will be raised. The adjustment range is positively correlated with the risk weighting value. The higher the weighting value, the more levels will be raised. After behavioral portrait matching and risk weighting correction, the employee's resignation risk level is more dynamic and three-dimensional, and the initial judgment based on static attributes is corrected to make it more in line with the employee's actual situation.
[0146] The adjusted resignation risk level fully combines the individual endowments of employees and organizational behavior, and can more comprehensively and dynamically predict the possibility of employee turnover. On the one hand, individual attributes such as age, length of service, and position determine the tendency of employee turnover and are internal factors that affect their resignation choices. On the other hand, behavioral performance such as attendance, performance, and interaction reflect the actual status of employees and are external windows for observing their willingness to resign. The integration of the two types of information forms a more robust risk judgment than a single dimension, avoiding the problem of biased predictions based solely on individual attributes or isolated behavioral fragments.
[0147] Methods for weighting and modifying the initial resignation risk level according to the risk weight to form a second resignation risk level include:
[0148]
[0149] in:
[0150] No. The second risk level of employee turnover;
[0151] No. The employee's initial risk level for leaving the company (integer, 1 = low risk, 2 = medium risk, 3 = high risk);
[0152] n: the number of key behavioral features;
[0153] j: index of key behavioral features;
[0154] w j : Risk weight of the jth key behavioral characteristic;
[0155] μ j : The mean of the jth key behavior feature;
[0156] No. The performance value of an employee on the jth key behavioral characteristic (can be a numerical indicator such as frequency, duration, etc.);
[0157] σ j : The standard deviation of the jth key behavior feature;
[0158] K: adjustment coefficient, used to control the impact range of the adjustment factor and ensure that the revised resignation risk level changes within a reasonable range;
[0159] round(·): rounding function, used to round the calculation result to the nearest integer to ensure the correct resignation risk level is an integer;
[0160] The employee's performance value on the jth key behavioral characteristic Standardization eliminates the impact of different feature dimensions so that the performance values of different features can be directly compared;
[0161] If an employee's performance on a key behavioral characteristic is significantly higher than the average level (abnormal performance), the risk level of resignation will increase; conversely, if an employee's performance on a key behavioral characteristic is significantly lower than the average level (good performance), the risk level of resignation will decrease. This formula takes into account the impact of multiple key behavioral characteristics through weighted summation. Through standardization, the impact of different feature dimensions is eliminated, so that the performance values of different features can be directly weighted summed.
[0162] Step S1400, setting a second resignation risk level warning threshold, when the second resignation risk level exceeds the warning threshold, triggering a warning, generating a list of key employees, and prompting the HR department and direct supervisor to pay attention and communicate in a timely manner;
[0163] Resignation warnings should not be issued to every employee whose resignation risk increases. This may lead to too frequent warnings, which will make HR and supervisors tired of responding, and the pertinence and effectiveness of warnings will decrease. A looser warning threshold should be set, and the warning will be triggered only when the employee's resignation risk level reaches a higher level (such as high risk) or the number of high-risk employees exceeds a certain proportion. Once the employee's second resignation risk level reaches the warning standard, the system will automatically include him / her in the key focus list and notify the employee's direct superior and the human resources department specialist as soon as possible. The list clearly lists the employee's basic information, current resignation risk level, risk level change trend, key behavioral abnormalities, etc., so that supervisors and HR can quickly understand the situation and carry out follow-up communication in a targeted manner. The list is presented in various clear and intuitive forms (such as lists and cards), highlighting key information and simplifying the information acquisition process. For employees who appear in the key focus list for many consecutive times and whose risk level continues to rise, the system will automatically issue warning prompts and set high-priority labels to guide supervisors to pay more attention and take more proactive retention measures to curb the further deterioration of risks. Graded warnings help to reasonably allocate time and energy for human resource management and focus on high-risk personnel.
[0164] Step S2000, collecting text data of employees on various media channels, constructing an employee sentiment dictionary based on the text data, judging the sentiment tendency of the text data, and obtaining employee sentiment analysis results; based on the employee sentiment analysis results, calculating the negative sentiment density, setting the negative sentiment density threshold of each level, and triggering a sentiment warning when the negative sentiment density of the employee exceeds the density threshold of the level for multiple consecutive cycles;
[0165] Specifically, through the data interface with the enterprise OA system, internal forums, and the personal information filled in by employees on recruitment websites and industry communities, unstructured text data such as employee user names, posting content, comment replies, and personal profiles on various media channels are collected. Using web crawler technology, the latest employee text data is regularly captured to ensure the real-time and continuity of the data. Using natural language processing toolkits such as Jieba and NLTK, the crawled text data is processed for Chinese word segmentation to split the text into the smallest semantic units. Meaningless words in the text, such as punctuation marks and modal particles, are removed to reduce data noise. Part-of-speech tagging is performed on the word segmentation results to provide a grammatical basis for sentiment analysis. Text preprocessing can improve the pertinence and accuracy of subsequent analysis. Sentiment analysis can be based on existing Chinese sentiment dictionaries, such as Hownet sentiment dictionary, or the enterprise can build its own professional sentiment dictionary suitable for specific industries and groups. Using methods such as dictionary matching and dependency syntax analysis, the sentiment words and their polarity contained in the text, such as "happy" and "unhappy", are identified. At the same time, emotional semantic rules are designed, such as "XX, but YY" often expresses negative emotions, while "not only XX, but also YY" tends to be positive emotions. Combining word polarity and semantic rules, the overall emotional tendency of the text is judged, such as positive, negative or neutral. Continuously track the changing trends of employees' emotional tendencies and gain insight into their psychological dynamics.
[0166] The employee sentiment analysis results include the sentiment score of each sentence, which is the sentiment score obtained after sentiment analysis of each sentence in the employee's emails, chat records and other texts. The sentiment score is usually a numerical value, indicating the sentiment tendency of the sentence, which can be a continuous value from negative to positive; the timestamp of the sentiment score, which records the time of the sentiment score of each sentence, so as to analyze the change of sentiment over time; the sentiment classification label, in addition to the numerical score, can also classify the sentiment of the sentence, such as positive, neutral, negative, etc.; the average sentiment score: the average sentiment score within a period of time (such as daily, weekly), indicating the overall sentiment tendency within the time period; the fluctuation of the sentiment score, the fluctuation range of the sentiment score within a period of time, reflecting the severity of the sentiment change.
[0167] Automating the process of collecting, processing, and analyzing text data forms a real-time computing pipeline for text sentiment analysis. Employees' emotional expressions inside and outside the organization often more truly reflect their inner state. By continuously monitoring the dynamic changes in employees' emotional tendencies, we can understand their workplace experience and psychological fluctuations. Employees with large emotional fluctuations and continuously rising negative emotions will also have a sudden increase in potential resignation risks. Therefore, text sentiment analysis provides an important psychological warning signal for the resignation warning system and is a key link in achieving real-time monitoring of human risks. Fully tapping the value of unstructured employee text data can make up for the blind spots of structured behavioral data and form an important supplement to employee portraits.
[0168] Based on the results of employee sentiment analysis, methods for calculating negative sentiment density include:
[0169]
[0170] in:
[0171] T i : The timestamp of the ith sentence. Used to calculate the time weight, reflecting the time when the sentence appeared;
[0172] T max : The end time of the time window, ensuring that recent sentiment changes have a greater impact on density calculations;
[0173] S i : The sentiment score of the ith sentence, indicating the sentiment tendency of each sentence, with negative sentiment scores as negative values and positive sentiment scores as positive values;
[0174] N: the total number of sentences in the time window;
[0175] w i : The time weight of the ith sentence is calculated based on the timestamp of the sentence, reflecting the importance of the sentence in the entire time window. Generally speaking, the most recent sentence has a large weight and the older sentence has a small weight;
[0176] λ: attenuation coefficient;
[0177] D neg : Negative sentiment density is the ratio of the total negative sentiment score to the total number of text characters;
[0178] δ(S i ): Negative sentiment indicator function of the ith sentence. Used to indicate whether the sentence is a negative sentiment sentence. If S i Negative emotions (i.e. S i <0), then δ(S i )=1; otherwise δ(S i )=0.
[0179] A time window is a specific time range selected during sentiment analysis to collect and analyze text data from employees during this time. By limiting the time range of analysis, a time window can focus on analyzing sentiment changes within a specific time period. This is particularly useful for identifying short-term sentiment fluctuations. i Helps identify the temporal characteristics of emotional changes, S i directly reflects the sentiment intensity of the sentence, N represents the number of all sentences in the selected time period, δ(S i ) helps filter out negative sentiment sentences.
[0180] Negative emotion density can be defined as the number or frequency of negative emotion expressions generated by employees per unit time, such as "employees post more than 3 negative emotion posts per week". The threshold of negative emotions is set in layers based on the attributes of employee rank, department, project, etc. For example, grassroots employees can be allowed a higher tolerance for negative emotions, while the negative emotion standards for management employees are stricter. When the negative emotion density of an employee exceeds the threshold of the level for N consecutive cycles (such as 3 weeks), the system automatically triggers an emotional warning. That is, the second risk level of employee resignation is assessed and appropriately raised to the third level, prompting the supervisor to appease the employee's emotions in time. Continuity is the key basis for judging the severity of negative emotions, not individual negative expressions. This formula not only takes into account the number of negative emotion sentences, but also the emotional intensity and time distribution of each sentence, providing a comprehensive and accurate sentiment analysis result that can be applied to texts of different lengths and structures, and has good generalization ability.
[0181] Step S3000, calculating the employee value index, and forming a comprehensive resignation risk score according to the employee's second resignation risk level, negative emotion density and employee value index; characterizing the employee resignation risk two-dimensional matrix according to the comprehensive resignation risk score; matching the best personalized retention plan for high-risk and high-value employees;
[0182] Furthermore, step S3000 includes:
[0183] Step S3100, quantifying the value contribution of employees to the organization from multiple dimensions, generating multi-dimensional contribution indicators, performing weighted summation on the contribution indicators of each dimension to form an employee comprehensive value score, normalizing the comprehensive value scores of all employees, generating an employee value percentage ranking as an employee value index;
[0184] Specifically, employee value assessment should be a multi-dimensional concept, and it is not possible to judge a person by a single standard. It is necessary to comprehensively consider the employee's job performance, professional skills, development potential, social capital, etc. Job performance mainly refers to the employee's historical performance evaluation results, including performance achievement rate, work quality, efficiency and other indicators. Professional skills can be quantified through employee training, skill assessment, professional certification, etc., reflecting the professional knowledge and practical experience mastered by employees. Development potential can be evaluated from the aspects of employees' learning ability, innovation ability, leadership, etc., reflecting the future growth space of employees. Social capital reflects the employee's interpersonal network and influence inside and outside the organization, such as whether he holds a key position and whether he has rich industry connections. By assigning weights to the contribution indicators of each dimension and taking weighted sum, the comprehensive value score of the employee can be obtained.
[0185] After calculating the comprehensive value score of the employee, it is necessary to compare and normalize the scores of all employees horizontally. Map the employee value score to a range of 0-100 points to obtain the employee value percentage, which intuitively reflects the relative value ranking of each employee in the entire company. The higher the ranking, the higher the value percentage, which means that the employee contributes more value to the company and is a key talent for the organization. The employee value percentage ranking is the employee value index, which reveals the horizontal status of employee value and facilitates targeted development of talent retention strategies.
[0186] The generation of the employee value index provides a powerful tool for quantitatively identifying key talents in an organization. Traditional talent evaluation often relies on impressions and subjective judgments, while the employee value index is based on objective data and displays employee value in multiple aspects, which can identify the core backbone of an organization more scientifically and accurately. For employees with a higher risk of leaving, if their employee value index is also relatively high, they should be the focus of retention and given targeted care and incentives in a timely manner to avoid the damage of talent loss to the competitiveness of the organization. The employee value index allows organizations to identify and retain talents more focused and more accurately.
[0187] Step S3200, forming a comprehensive resignation risk score according to the employee's second resignation risk level, negative emotion density and employee value index;
[0188] Further, step S3200 includes:
[0189] Step S3210, standardizing the second resignation risk level, negative emotion density, and employee value index, unifying the dimensions, and forming three risk sub-factors, namely, the second resignation risk level sub-factor, the negative emotion density sub-factor, and the employee value index sub-factor;
[0190] Step S3220, fitting the relationship between the second resignation risk level, negative emotion density, employee value index and actual employee turnover rate, and estimating the regression coefficients of the three risk sub-factors;
[0191] Step S3230, taking the regression coefficients of the three risk sub-factors as weights, performing weighted summation on the risk sub-factors to generate a comprehensive resignation risk score;
[0192] Specifically, when forming a comprehensive resignation risk assessment, it is necessary to take into account information from multiple risk dimensions, including the second resignation risk level, negative emotion density, and employee value index. Since the three indicators are calculated in different ways, the numerical differences are large and cannot be directly arithmetic operations. They need to be standardized first, and the dimensions must be unified so that they fall within a similar numerical range. Common methods such as Z-score and Min-Max can be used for standardization.
[0193] After standardizing the turnover risk level, emotion density, and value index, three risk sub-factors are formed. However, the contribution of each sub-factor to the final turnover risk is different. The relationship between the risk sub-factors and the actual employee turnover rate can be fitted based on the Logistic regression model using historical data. Logistic regression can obtain the regression coefficient of each risk sub-factor through maximum likelihood estimation. The absolute value of the regression coefficient represents the weight of the sub-factor's impact on the turnover risk. A positive coefficient indicates a turnover-promoting factor, such as risk level and emotion density; a negative coefficient indicates a turnover-inhibiting factor, such as the employee value index.
[0194] The comprehensive risk score reflects the overall assessment of employee turnover risk, which goes a step further than the initial risk judgment based on employee attributes and behavioral portraits. It not only considers the static characteristics and dynamic performance of employees, but also integrates the emotional state and value contribution of employees. The three risk factors are cross-corroborated, avoiding the bias of making arbitrary predictions based on a single piece of information, making the portrayal of turnover risk more three-dimensional and comprehensive. When an employee's performance in the three aspects is not ideal, such as high risk level, negative emotions, and low value contribution, his comprehensive risk score will be much higher than that of ordinary people, indicating a very high probability of turnover, which requires the organization to attach great importance to and intervene in time. On the contrary, if the employee's data in each dimension is relatively optimistic, the comprehensive risk score is low, and the possibility of turnover in the short term is not high, so the organization can be relatively assured. The advantage of the comprehensive turnover risk scoring model is that it can take into account multiple factors such as individual employees, organizational environment, and interpersonal interaction, and make a more comprehensive and prudent prediction of talent loss, thereby guiding the organization to formulate targeted human resource policies and minimize the probability of core talent loss.
[0195] Step S3300, based on the comprehensive resignation risk score, a two-dimensional matrix of employee resignation risk is drawn, where the horizontal axis is the risk level and the vertical axis is the employee value;
[0196] Further, step S3300 includes:
[0197] Step S3310, dividing the comprehensive resignation risk score into three resignation risk levels: high, medium, and low, as the horizontal axis of the matrix;
[0198] Step S3320, dividing the employee value index into three employee value levels of high, medium, and low, which serve as the vertical axis of the matrix;
[0199] Step S3330, combining employee value levels and resignation risk levels in pairs to form a 3×3 resignation risk matrix, where employees in the upper right corner of the matrix are high-risk and high-value employees;
[0200] Specifically, the horizontal axis of the matrix represents the level of risk of resignation, and the vertical axis represents the level of employee value. First, the comprehensive risk score of employees needs to be graded. The risk score can be divided into three intervals from low to high by equal interval segmentation or equal frequency segmentation, corresponding to the three levels of low risk, medium risk, and high risk. Similarly, the employee value index is also treated similarly and divided into three levels of high value, medium value, and low value. After establishing the framework of the risk-value matrix, each employee is classified into one of the nine quadrants according to his or her risk level and value level of resignation. The "high-risk-high-value" employees in the upper right corner of the matrix are the top priority of the organization's retention work, with the highest loss cost, and need to invest core resources to focus on retention. The "low-risk-low-value" employees in the lower left corner of the matrix have limited contributions to the organization and are less likely to lose. They belong to the regular talent reserve and can be relatively left alone. Employees in other quadrants should also match differentiated incentive retention strategies based on the combination of risk and value levels. For example, for high-value-medium-risk employees, their value contribution is outstanding but the signs of leaving are gradually showing. They can be moderately incentivized to strengthen their sense of belonging. For medium-value-low-risk employees, their risk of leaving is not obvious but their value needs to be improved. They can be strengthened to promote their migration to the high-value quadrant.
[0201] The advantage of the resignation risk matrix is that it can clearly present the overall picture of the organization's talent loss, identify the high-risk groups that need the most attention, and avoid evenly distributing efforts in retaining employees. Using the resignation risk matrix, organizations can take measures according to their talents and prescribe the right medicine for the right symptoms, so as to achieve targeted results and achieve twice the result with half the effort in human capital management. The matrix also reveals the dynamic distribution of employees in the risk and value dimensions, making it easier for organizations to monitor talent status in real time and adjust human resource policies in a timely manner. The risk matrix is a "war room" for resignation warnings, which has important decision-making support value for managers to gain insight into the trend of talent loss and make scientific decisions.
[0202] Step S3400, matching the best personalized retention plan for high-risk and high-value employees;
[0203] For high-risk and high-value employees, organizations must respond quickly and provide attractive retention plans to retain them sincerely. Traditional retention often adopts a "one-size-fits-all" strategy, with unified salary increases, job increases, stock options, etc. However, for top talents, the marginal utility of such material incentives decreases and may not work. The reasons for the resignation of talents vary greatly, which may be salary, work environment, development space, leadership style, team atmosphere, work location, etc., and need to be analyzed according to individual differences. This requires the design of personalized retention plans for high-risk and high-value employees.
[0204] First, based on employee profile features, such as age, profession, rank, historical performance, and risk level of resignation, machine learning algorithms are used to mine the best practices in past successful retention cases. Through association rule learning, the correspondence between employee characteristics and retention measures can be discovered, and the most targeted retention combination can be identified. Through collaborative filtering, historical cases that are most similar to the target employee can be found, and their successful retention experience can be directly learned. Machine learning can quickly find employees' "portrait twins" from massive historical data and automatically recommend tailored retention menus.
[0205] On this basis, it is also necessary to fully consider the personalized demands of employees. Through daily communication and heart-to-heart talks with employees, managers should find out the deep-seated reasons for employees' resignation, understand their real thoughts, and make targeted efforts to fill the gaps. For example, if employees value job flexibility more, they can be allowed to work flexibly; if employees want more systematic training, they can be provided with opportunities for overseas study; if employees lose interest in existing projects, they can be provided with job rotation conditions. Fine-tune the retention strategies recommended by the machine to best meet the personalized needs of employees.
[0206] Before the personalized retention plan is implemented, it is necessary to conduct preliminary simulations to evaluate its feasibility and effectiveness. You need to put yourself in the shoes of the employees. If I were an employee, would this plan impress me? Compared with the employees' expectations, is there room for improvement? Quantify the expected effect of the proposed plan and calculate the success rate of retention. If the success rate is lower than expected, the plan needs to be optimized and adjusted again to ensure that the most thoughtful and sincere plan is presented to employees.
[0207] Finally, the employee's direct superior will be the representative in the retention negotiation. Through face-to-face and frank communication, the organization's attention and expectations will be conveyed to the employee, and the sincerity of retention will be demonstrated. The employee's doubts will be answered, their concerns will be eliminated, and their thoughts of leaving will be dispelled. The employee will feel that he or she is an indispensable part of the organization, and his or her sense of identity and achievement will be enhanced. The success or failure of the retention negotiation is crucial to the employee's stay. Only sincere communication and personalized solutions can maximize the success rate of retention.
[0208] Personalized retention plans are a winning magic weapon for organizations to retain key talents. On the one hand, it starts with employee portraits, fully draws on successful experiences, finds the best combination of retention strategies for employees, and improves the accuracy of retention. On the other hand, it closely combines the personalized needs of employees, listens to their voices, resolves their dissatisfaction with actions, touches their hearts with sincerity, and improves the success rate of retention. Personalized retention plans are employee-centric and flexibly adapt to the unique characteristics and needs of each employee. It replaces the traditional one-size-fits-all approach and provides targeted and customized incentives. This makes employees feel truly valued and appreciated, and greatly enhances their sense of belonging and loyalty to the organization. Minimize the organization's talent turnover rate, enhance human capital advantages, and create sustainable competitiveness for the organization.
[0209] Example 2
[0210] Based on Example 1, this example provides a system for constructing an employee resignation portrait model based on the fusion of multi-source heterogeneous data. Figure 4 As shown, including:
[0211] Data collection module: used to collect employees’ basic attribute data and behavioral log data generated in their daily work from multiple data sources, as well as employees’ text data on various media channels;
[0212] Data processing module: used to structure the collected unstructured basic attribute data and associate and integrate the behavior log data from different sources according to employee ID and timestamp;
[0213] Employee behavior profile building module: used to characterize key employee behavior characteristics based on behavior log data, and combine it with employee attribute characteristics to build a multi-dimensional employee behavior profile;
[0214] Employee resignation risk level classification module: used to set the initial resignation risk level of employees based on their basic attribute data; set the risk weights of key behavioral characteristics based on employee behavior profiles, and perform weighted corrections on the initial resignation risk level based on the risk weights to form a second resignation risk level;
[0215] Sentiment analysis module: used to build an employee sentiment dictionary based on the text data of employees on various media channels, judge the sentiment tendency of the text data, obtain employee sentiment analysis results, and calculate the negative sentiment density based on the employee sentiment analysis results;
[0216] Employee value assessment module: used to quantify the value contribution of employees to the organization from multiple dimensions, generate multi-dimensional contribution indicators, perform weighted summation of the contribution indicators of each dimension to form an employee comprehensive value score, normalize the comprehensive value scores of all employees, and generate an employee value percentage ranking as an employee value index;
[0217] Resignation risk assessment module: used to form a comprehensive resignation risk score based on the employee's second resignation risk level, negative emotion density and employee value index, and to characterize the resignation risk matrix;
[0218] Personalized retention plan generation module: used to match the best personalized retention plan for high-risk and high-value employees.
[0219] Example 3
[0220] This embodiment discloses an electronic device, including a memory, a central processing unit, and a computer program stored in the memory and executable on the central processing unit. When the central processing unit executes the computer program, the above-mentioned method for constructing an employee resignation portrait model based on the fusion of multi-source heterogeneous data is implemented.
[0221] Since the electronic device introduced in this embodiment is an electronic device used to implement the method for building an employee resignation portrait model based on the fusion of multi-source heterogeneous data in the embodiment of this application, based on the method for building an employee resignation portrait model based on the fusion of multi-source heterogeneous data introduced in the embodiment of this application, the technical personnel in this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not introduced in detail here. As long as the technical personnel in this field implement the electronic device used in the method for building an employee resignation portrait model based on the fusion of multi-source heterogeneous data in the embodiment of this application, it belongs to the scope of protection of this application.
[0222] Example 4
[0223] This embodiment discloses a computer-readable storage medium having a computer program stored thereon. When the computer program is executed, the method for constructing an employee resignation portrait model based on the fusion of multi-source heterogeneous data is implemented.
[0224] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters, weights and thresholds in the formula are set by technicians in this field according to actual conditions.
[0225] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center through a wired network or a wireless network. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD) or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.
[0226] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.
[0227] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0228] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic, for example, the division of the units is only one, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.
[0229] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0230] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
[0231] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
[0232] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention. The collection of data and personal information involved in this application has been fully authorized and explicitly agreed to by the parties and related objects; and the collection, use and processing of all data strictly abide by the laws, regulations and relevant standards of relevant countries and regions to ensure information security and compliance.
Claims
1. A method for constructing an employee resignation portrait model based on the fusion of multi-source heterogeneous data, characterized in that: include: Collect employees' basic attribute data and set their initial risk level of resignation based on the basic attribute data; Collect the behavioral log data generated by employees in their daily work, build a multi-dimensional employee behavior profile based on the basic attribute data and behavioral log data, and modify the initial resignation risk level based on the employee behavior profile to form a second resignation risk level; Collect employees' text data on various media channels, build employee sentiment dictionaries based on the text data, judge the sentiment tendency of the text data, and obtain employee sentiment analysis results; calculate negative sentiment density based on employee sentiment analysis results, set hierarchical negative sentiment density thresholds, and trigger sentiment warnings when employees' negative sentiment density exceeds the density threshold of the level for multiple consecutive cycles; Calculate the employee value index and form a comprehensive turnover risk score based on the employee's second turnover risk level, negative emotion density and employee value index; Based on the comprehensive turnover risk score, a two-dimensional matrix of employee turnover risk is drawn; for high-risk and high-value employees, the best personalized retention plan is matched.
2. The method for constructing an employee resignation portrait model based on integrating multi-source heterogeneous data according to claim 1 is characterized in that: The method comprises: Obtain the basic attribute data of employees, structure the unstructured basic attribute data, convert it into numerical variables, and form a structured employee attribute table; based on the employee attribute table, construct employee attribute characteristics, divide employees into initial resignation risks according to employee attribute characteristics, and set the initial resignation risk level of employees; Collect the behavior log data generated by employees in their daily work, characterize the key behavior characteristics of employees based on the behavior log data, and combine it with the employee attribute characteristics to build a multi-dimensional employee behavior portrait; According to the employee behavior profile, risk weights for key behavior characteristics are set, and the initial resignation risk level is weighted and modified according to the risk weights to form a second resignation risk level; Set a warning threshold for the second risk level of resignation. When the second risk level of resignation exceeds the warning threshold, an alert is triggered and a list of key employees is generated, prompting the HR department and direct supervisor to pay attention and communicate in a timely manner.
3. The method for constructing an employee resignation portrait model based on integrating multi-source heterogeneous data according to claim 2 is characterized in that: The method comprises: The employee attribute table contains categorical variables and numerical variables; Generate employee attribute categorical features from categorical variables; Generate employee attribute numerical features from numerical variables; Generate new characteristics of employee career based on employee attribute table; Combine employee attribute category characteristics, employee attribute numerical characteristics and employee career new characteristics to form employee attribute characteristics; Filter out a subset of employee attribute characteristics whose correlation with the risk of resignation is greater than a relevant threshold from the employee attribute characteristics; Construct an initial resignation risk warning model, predict the resignation risk of all employees based on the employee attribute feature subset and the initial resignation risk warning model, output the resignation risk prediction probability, and divide the resignation risk prediction probability into three levels: high, medium, and low to form the initial resignation risk level of employees; High risk: predicted probability of resignation risk c≥0.6; Medium risk: predicted probability of leaving the company is 0.3 <c<0.6; Low risk: The predicted probability of resignation risk c≤0.
3.
4. The method for constructing an employee resignation portrait model based on integrating multi-source heterogeneous data according to claim 2 is characterized in that: The method comprises: Collect the behavior log data generated by employees in their daily work from multiple data sources, associate and integrate the behavior log data from different sources according to employee ID and timestamp to form a structured employee behavior data set; Identify key behavioral factors that influence employee performance and turnover from employee behavior datasets; The employee behavior dataset is divided into behavior time series data according to the time window. Based on the behavior time series data, the frequent subsequences of employee behavior evolution are found as the sequence pattern of employee behavior. Extract the key influencing factors of turnover risk from the key behavioral factors and quantify them into key behavioral characteristics; Extract the evolutionary trajectory of employee behavior from the sequence pattern and summarize it into the characteristics of behavior change trend; Combine employee attribute characteristics, key behavior characteristics and behavior change trend characteristics to construct a multi-dimensional employee behavior portrait.
5. The method for constructing an employee resignation portrait model based on integrating multi-source heterogeneous data according to claim 2 is characterized in that: The method of weighting and modifying the initial resignation risk level according to the risk weight to form a second resignation risk level includes: in, For the The second risk level of employee turnover is For the The initial turnover risk level of an employee, n is the number of key behavioral characteristics, j is the index of the key behavioral characteristics, and w is the j is the risk weight of the jth key behavioral characteristic, μ j is the mean value of the jth key behavior feature, For the The performance value of an employee on the jth key behavioral characteristic, σ j is the standard deviation of the jth key behavior feature, K is the adjustment coefficient, and round(·) is the rounding function.
6. The method for constructing an employee resignation portrait model based on integrating multi-source heterogeneous data according to claim 1 is characterized in that: The method for calculating the negative emotion density based on the employee emotion analysis results includes: Among them, T i is the timestamp of the ith sentence, T max is the end time of the time window, S i is the sentiment score of the i-th sentence, N is the total number of sentences in the time window, and w i is the time weight of the i-th sentence, λ is the decay coefficient, D neg is the density of negative emotions; δ(S i ) is the negative sentiment indicator function of the i-th sentence, if S i <0, then δ(S i )=1, otherwise δ(S i )=0.
7. The method for constructing an employee resignation portrait model based on integrating multi-source heterogeneous data according to claim 1 is characterized in that: The method comprises: Quantify the value contribution of employees to the organization from multiple dimensions, generate multi-dimensional contribution indicators, perform weighted summation of the contribution indicators of each dimension to form an employee comprehensive value score, normalize the comprehensive value scores of all employees, generate an employee value percentage ranking as the employee value index; The second resignation risk level, negative emotion density and employee value index are standardized and unified to form three risk sub-factors; Fit the relationship between the second turnover risk level, negative emotion density, employee value index and actual employee turnover rate, and estimate the regression coefficients of the three risk sub-factors; The regression coefficients of the three risk sub-factors are used as weights, and the risk sub-factors are weighted and summed to generate a comprehensive turnover risk score; The comprehensive turnover risk score is divided into three turnover risk levels: high, medium, and low, which serve as the horizontal axis of the matrix; The employee value index is divided into three employee value levels: high, medium, and low, which serve as the vertical axis of the matrix; The employee value levels and turnover risk levels are combined in pairs to form a 3×3 turnover risk matrix. Employees in the upper right corner of the matrix are high-risk and high-value employees.
8. A system for constructing an employee resignation portrait model based on the fusion of multi-source heterogeneous data, which is used to implement the method for constructing an employee resignation portrait model based on the fusion of multi-source heterogeneous data according to any one of claims 1 to 7, characterized in that: The system comprises: Data collection module: used to collect employees’ basic attribute data and behavioral log data generated in their daily work from multiple data sources, as well as employees’ text data on various media channels; Data processing module: used to structure the collected unstructured basic attribute data and associate and integrate the behavior log data from different sources according to employee ID and timestamp; Employee behavior profile building module: used to characterize key employee behavior characteristics based on behavior log data, and combine it with employee attribute characteristics to build a multi-dimensional employee behavior profile; Employee resignation risk level classification module: used to set the initial resignation risk level of employees based on their basic attribute data; set the risk weights of key behavioral characteristics based on employee behavior profiles, and perform weighted corrections on the initial resignation risk level based on the risk weights to form a second resignation risk level; Sentiment analysis module: used to build an employee sentiment dictionary based on the text data of employees on various media channels, judge the sentiment tendency of the text data, obtain employee sentiment analysis results, and calculate the negative sentiment density based on the employee sentiment analysis results; Employee value assessment module: used to quantify the value contribution of employees to the organization from multiple dimensions, generate multi-dimensional contribution indicators, perform weighted summation of the contribution indicators of each dimension to form an employee comprehensive value score, normalize the comprehensive value scores of all employees, and generate an employee value percentage ranking as an employee value index; Resignation risk assessment module: used to form a comprehensive resignation risk score based on the employee's second resignation risk level, negative emotion density and employee value index, and to characterize the resignation risk matrix; Personalized retention plan generation module: used to match the best personalized retention plan for high-risk and high-value employees.
9. An electronic device comprising a memory, a central processing unit, and a computer program stored in the memory and executable on the central processing unit, characterized in that: When the central processing unit executes the computer program, the method for constructing an employee resignation portrait model based on the fusion of multi-source heterogeneous data described in any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed, implements the method for constructing an employee resignation portrait model based on the fusion of multi-source heterogeneous data according to any one of claims 1 to 7.
