Labor service outsourcing team performance intelligent evaluation system based on big data

Abstract

The invention provides a labor service outsourcing team performance intelligent evaluation system based on big data, relates to the technical field of performance intelligent evaluation, and aims at solving the technical problems that existing labor service outsourcing team performance evaluation is single in dimension, high in subjectivity, poor in dynamic adaptability and low in evaluation accuracy. The system comprises a multi-source data acquisition module, a data preprocessing module, a feature extraction module, an intelligent evaluation module, a dynamic adjustment module, a result output module and a data security module. Various data of a labor outsourcing team are collected in a multi-source mode, core performance characteristics are extracted after self-adaptive preprocessing, a self-created comprehensive performance evaluation formula and a team and first party adaptation degree coefficient formula are combined, a BP neural network and a random forest fusion model are adopted to calculate comprehensive performance scores, parameter weights are optimized through a dynamic adjustment module, and a comprehensive performance evaluation result is obtained. The system achieves the multi-dimensional, intelligent and precise evaluation of the performance, and has the functions of performance prediction, risk early warning and data safety guarantee.

Description

Technical Field

[0001] This invention belongs to the field of intelligent performance evaluation technology, and more specifically, it relates to an intelligent performance evaluation system for labor outsourcing teams based on big data. Background Technology

[0002] With the rapid development of the labor outsourcing industry, outsourcing teams are expanding in size and becoming increasingly diverse. Performance evaluation, as a core aspect of outsourcing management, directly impacts service quality, cost control, and the stability of cooperation between the client and the outsourcing company. Currently, existing performance evaluation technologies for labor outsourcing teams still have many shortcomings and are insufficient to meet the industry's development needs.

[0003] Firstly, the evaluation dimensions are too narrow. Existing technologies mostly focus on simple quantitative evaluation of task completion, without integrating multi-dimensional data such as client feedback, cost control, team fit, and staff turnover. The evaluation results are one-sided and cannot fully reflect the comprehensive capabilities of the outsourced team.

[0004] Secondly, it is highly subjective. Most evaluation methods rely on manual scoring, lack scientific quantitative formulas and intelligent models to support them, resulting in large human errors and insufficient fairness in the evaluation results.

[0005] Third, the data processing capabilities are weak. There is no suitable preprocessing scheme designed for multi-source heterogeneous data (structured, semi-structured, and unstructured) from labor outsourcing. The outlier removal uses a fixed threshold, which is prone to accidentally deleting valid data, resulting in low data quality and affecting the accuracy of the assessment.

[0006] Fourth, it has poor dynamic adaptability. The evaluation model parameters and weights are fixed and cannot be dynamically optimized according to changes in client needs, adjustments to task types, fluctuations in the industry environment, etc., making it difficult to adapt to the core needs of labor outsourcing teams that are highly mobile and serve multiple clients.

[0007] Fifth, it lacks a closed-loop design for the entire process, fails to achieve a closed loop of evaluation-feedback-optimization, and lacks effective performance prediction and risk warning functions, making it impossible to avoid the risk of failing to meet performance targets in advance. At the same time, the data security protection mechanism is imperfect, which easily leads to the leakage of sensitive data.

[0008] Therefore, in response to the shortcomings of existing technologies, developing an intelligent performance evaluation system for labor outsourcing teams that can integrate multi-source big data, achieve accurate quantitative assessment, dynamically adapt to changes in scenarios, and possess a closed-loop process and data security guarantees has become an urgent technical problem to be solved. Summary of the Invention

[0009] To address the aforementioned technical problems, this invention provides a big data-based intelligent performance evaluation system for labor outsourcing teams. This system solves the technical problems of existing labor outsourcing team performance evaluations, such as limited dimensions, strong subjectivity, poor dynamic adaptability, low evaluation accuracy, weak data processing capabilities, lack of a complete closed-loop process, and inadequate data security.

[0010] A big data-based intelligent performance evaluation system for labor outsourcing teams includes:

[0011] The data acquisition module is used to collect basic information data, task execution data, client feedback data, cost control data, and environmental impact data from multiple sources for the labor outsourcing team. The multi-source data includes structured data, semi-structured data, and unstructured data.

[0012] The data preprocessing module is used to clean, denoise, normalize, and align features of the collected multi-source data. It uses an adaptive outlier removal algorithm to remove invalid data and outputs a standardized dataset.

[0013] The feature extraction module is used to extract core performance evaluation features from a standardized dataset. These core features include team fit features, task completion quality features, collaboration efficiency features, cost optimization features, and risk warning features.

[0014] The intelligent evaluation module has a built-in performance evaluation model. Based on the extracted core features, the model calculates the comprehensive performance score of the outsourcing team by combining the self-created comprehensive performance evaluation formula, so as to realize multi-dimensional intelligent evaluation of the performance of the labor outsourcing team.

[0015] The dynamic adjustment module is used to dynamically optimize the parameters and feature weights of the performance evaluation model based on historical evaluation data, real-time task changes, and client requirements.

[0016] The results output module is used to output the performance scores, ratings, shortcomings analysis and optimization suggestions output by the intelligent evaluation module to the user terminal in a visual form.

[0017] The comprehensive performance evaluation formula used by the intelligent evaluation module is a self-created formula (1):

[0018] (1);

[0019] Where: P is the comprehensive performance score of the labor outsourcing team, with a value range of [0, 100];

[0020] T represents the task completion score, with a value range of [0, 100], which indicates the timeliness and completeness of the outsourced team's task delivery.

[0021] F is the compatibility coefficient between the team and the client, with a value range of [0.6, 1.2], calculated using the self-created formula (2);

[0022] C represents the score for collaboration efficiency, with a value range of [0, 100], which indicates the smoothness of collaboration within the team and with the client.

[0023] S is the score for the cost optimization dimension, with a value range of [0, 100], representing the team's cost control and resource utilization efficiency;

[0024] D is a dynamic adjustment factor with a value range of [0.9, 1.1], which is dynamically output by the dynamic adjustment module according to the real-time scenario.

[0025] α, β, γ, and δ are the weight coefficients for each dimension, satisfying the following conditions: And α∈[0.3,0.4], β∈[0.2,0.3], γ∈[0.2,0.3], δ∈[0.1,0.2], can be adaptively optimized through the dynamic adjustment module;

[0026] The self-created calculation formula for the compatibility coefficient F between the team and the client is formula (2):

[0027] (2);

[0028] Where: M is the matching degree between the skills of the outsourced team members and the client's task requirements, with a value range of [0,100], which is obtained by comparing the skill tags of team members with the skill requirements of the client's task through big data analysis;

[0029] N represents the degree to which the outsourced team's service response speed matches the client's expectations, with a value range of [0, 100], calculated from the client's feedback data and response time data;

[0030] L represents the mobility of outsourced team members, with a value range of [0,1]. L = Number of employees leaving in the current period / Average number of employees in the current period.

[0031] k1, k2, and k3 are the fitness influence weights, satisfying k1 + k2 + k3 = 2, and k1 ∈ [0.8, 1.0], k2 ∈ [0.6, 0.8], and k3 ∈ [0.2, 0.4].

[0032] ε is a correction factor with a value of 0.01, used to avoid the abnormal case of the denominator being 0 and to ensure the validity of the formula calculation.

[0033] Preferably, the multi-source data of the data acquisition module specifically includes:

[0034] Basic information data: Age, education, skills certificates, years of experience, and job assignments of outsourced team members;

[0035] Task execution data: task assignment records, task start / end time, task completion progress, number of rework attempts, and task acceptance results;

[0036] Client feedback data: Client's rating of task quality, evaluation of team service attitude, complaint records, and satisfaction data on response to changes in requirements;

[0037] Cost control data: team labor costs, consumable costs, management costs, task delivery costs, and cost savings rate data;

[0038] Environmental impact data: industry benchmark performance data, performance data of similar outsourcing teams, environmental change data of the client's industry, and policy adjustment data.

[0039] Preferably, the adaptive outlier removal algorithm of the data preprocessing module is as follows: based on big data statistical analysis, the mean μ and standard deviation σ of each data dimension are calculated, and data exceeding the range of [μ-3σ, μ+3σ] are marked as outliers. At the same time, combined with the characteristics of the labor outsourcing industry, industry thresholds for task completion time and cost fluctuations are set, and the marked outliers are verified a second time to remove invalid outliers, retain reasonable outliers, and make corrections.

[0040] Preferably, among the core features extracted by the feature extraction module, the risk warning features include the risk value of outsourced team personnel turnover, the risk value of task delay, and the risk value of cost overrun. The risk value of personnel turnover is predicted by a machine learning model trained with historical turnover data, current on-duty time of personnel, and salary satisfaction data. The risk value of task delay is calculated by combining the remaining time of the task, the current completion progress, and historical delay data.

[0041] Preferably, the performance evaluation model of the intelligent evaluation module adopts a fusion model trained by big data. The fusion model consists of a BP neural network model and a random forest model. The BP neural network model is used to fit and calculate nonlinear features (collaboration efficiency, fit), and the random forest model is used to assign weights and calculate scores for linear features (task completion, cost optimization). The final performance score is obtained by fusing the output results of the two models.

[0042] Preferably, the parameter optimization logic of the dynamic adjustment module specifically involves: periodically collecting historical evaluation data and actual performance feedback data, and calculating the evaluation error. When E>5, the parameter optimization process is triggered. Based on the gradient descent algorithm, the weight coefficients of α, β, γ, and δ and the values ​​of k1, k2, and k3 in formula (2) are adjusted. At the same time, the extraction weights of each core feature are adjusted in combination with the latest requirements of the client and changes in task type to ensure the accuracy and adaptability of the evaluation results.

[0043] Preferably, the calculation logic for the task completion dimension score T is as follows:

[0044] T = 100 - 5 × number of reworks - 2 × delay duration (hours) / standard duration (hours) × 10. When T < 0, take T = 0.

[0045] The number of reworks refers to the cumulative number of reworks for a single task, and the delay duration is the difference between the actual completion time of the task and the standard completion time. The standard duration is derived from big data analysis of the historical completion times of similar tasks.

[0046] Preferably, the system further includes a data security module for encrypting and storing the collected multi-source data, controlling access, and desensitizing the data. Sensitive data (personnel ID numbers, client business information) is encrypted using an irreversible encryption algorithm, and access control adopts role-based hierarchical management, allowing only authorized users to access the corresponding level of evaluation data and raw data.

[0047] Preferably, the visualization forms of the result output module include line charts (historical performance change trends), radar charts (performance comparisons in various dimensions), heat maps (distribution of shortcomings), and text reports. It also supports exporting the evaluation results to PDF and Excel formats and can automatically push them to the client's terminal to achieve two-way synchronization of the evaluation results.

[0048] Preferably, the intelligent evaluation module also has a performance prediction function. Based on historical performance data, current task progress and dynamic adjustment factor D, it iteratively calculates the performance prediction score of the outsourced team for the next 1-3 months through formula (1), marks risk teams with prediction scores below 60, and pushes early warning information to the user terminal to avoid the risk of performance failure in advance.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The assessment is comprehensive and accurate, addressing the pain point of one-sided assessment in existing technologies: by integrating five categories of heterogeneous data from multiple sources—basic information, task execution, client feedback, cost control, and environmental impact—through a multi-source data collection module, it covers all factors influencing the performance of outsourcing teams. Combined with the core features extracted by the feature extraction module, it achieves a comprehensive assessment of the overall capabilities of labor outsourcing teams. Verified by examples, the assessment accuracy rate can reach over 92%, significantly higher than existing manual and single-dimensional assessment methods (average accuracy rate 76.3%).

[0051] The self-developed comprehensive performance evaluation formula and the team-client fit coefficient formula introduce dynamic adjustment factors, fit influence weights, and correction factors. Combining a BP neural network and random forest fusion model, subjective evaluation is transformed into objective quantitative scores. This completely changes the current situation where existing technologies rely on manual scoring and are highly subjective, improving the fairness and accuracy of evaluation results. The correlation coefficient between the performance score calculated by the self-developed formula and the client's final evaluation score reaches 0.89, which is significantly higher than the existing single formula (correlation coefficient 0.62).

[0052] Through the dynamic adjustment module, the model parameters, formula weights and feature extraction weights are optimized in real time based on the evaluation error. It can adapt to scenarios such as changes in client needs, adjustments to task types, and fluctuations in the industry environment. This solves the shortcomings of existing technical models that are fixed and cannot be dynamically adjusted. For example, when the client's needs change in priority, the weights of each dimension can be quickly adjusted, increasing the fit between the evaluation results and the client's actual needs from 82% to 93%.

[0053] It constructs a closed-loop process of "collection-preprocessing-evaluation-adjustment-output-feedback", and integrates performance prediction and risk warning functions. Based on historical data and current status, it can predict the performance score for the next 1-3 months, mark teams with poor performance and push warning information to help users take optimization measures in advance and avoid the risk of poor performance. Verified by examples, it can effectively improve the actual performance of risk teams from the predicted 58 points to over 65 points.

[0054] An independent data security module is set up, employing the AES-256 irreversible encryption algorithm, role-based access control, data anonymization, and off-site multi-backup mechanisms to provide comprehensive protection for collected sensitive data (personnel information, client business information, cost data), solving the problem of inadequate data security protection in existing technologies and ensuring the security of data storage, transmission, and use.

[0055] Adopting a distributed architecture design, it supports a combination of streaming and batch processing, with a system response latency of ≤5s. It can be adapted to various labor outsourcing teams (technology outsourcing, service outsourcing, production outsourcing, etc.) ranging from 10 to 1000 people. It also supports multi-terminal access and bidirectional synchronization of evaluation results, as well as format export, which improves outsourcing management efficiency, reduces management costs, and has broad application prospects. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the system framework of the present invention. Detailed Implementation

[0057] The present invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention.

[0058] This invention discloses a big data-based intelligent performance evaluation system for labor outsourcing teams. It aims to address the technical pain points of existing labor outsourcing team performance evaluations, such as single evaluation dimensions, strong subjectivity, inability to dynamically adapt to changing scenarios, and low accuracy of evaluation results. Through multi-source big data fusion collection, adaptive preprocessing, precise extraction of core features, self-developed formula-based quantitative evaluation, dynamic parameter optimization, and visualization output, it achieves full-process, multi-dimensional, and intelligent evaluation of labor outsourcing team performance. It also boasts advantages such as high data security, strong scalability, and adaptability to various outsourcing scenarios. The specific implementation method is as follows.

[0059] I. System Overall Architecture Implementation:

[0060] This system adopts a distributed architecture and is deployed on a cloud server cluster. It supports access from multiple terminals (PC, mobile, and client-specific terminals) and can be adapted to various labor outsourcing teams (including technology outsourcing, service outsourcing, and production outsourcing) ranging from 10 to 1000 people. The system is divided into 6 core modules (data acquisition module, data preprocessing module, feature extraction module, intelligent evaluation module, dynamic adjustment module, and result output module) and 1 auxiliary module (data security module, corresponding to claim 8). Each module achieves data communication through a high-speed data interface and adopts a combination of streaming and batch processing to ensure the real-time performance of data transmission and evaluation calculation. The overall system response latency is ≤5s, and the evaluation accuracy is ≥92%.

[0061] II. Specific implementation details of each module:

[0062] 2.1 Implementation of the data acquisition module:

[0063] The data acquisition module adopts a multi-source heterogeneous data acquisition architecture, integrating four acquisition methods: interface acquisition, web crawler acquisition, manual data entry, and third-party data integration. This enables comprehensive and real-time acquisition of multi-source data related to the labor outsourcing team. The specific acquisition content and implementation methods are as follows:

[0064] (1) Data Collection Content: The data collected includes five categories: basic information data, task execution data, client feedback data, cost control data, and environmental impact data. It covers three types of data: structured, semi-structured, and unstructured, as detailed below:

[0065] ① Basic information data: Data is collected through manual entry and interface with the enterprise's HR system. This includes the outsourced team members' age (integer, 18-60 years old), education level (enumerated type: primary school / junior high school / high school / junior college / bachelor's degree / master's degree and above), skills certificates (text type + tagged processing, such as "computer level 2 certificate" "electrician certificate"), years of work experience (floating type, unit: year), and job assignment (enumerated type: technical position / service position / production position / management position). The data is collected once a month, and new / departing employees are added and supplemented in real time.

[0066] ② Task execution data: Data is collected bidirectionally through the outsourcing management system interface and the client's task management system interface, including task assignment records (text type, including task number, assignment time, and responsible team), task start / end time (timestamp format), task completion progress (percentage type, 0%-100%), number of task reworks (integer type, ≥0), and task acceptance results (enumerated type: qualified / unqualified / requires rectification). The collection frequency is once per hour to ensure real-time synchronization of task execution status.

[0067] ③ Client feedback data: Collected through the client's dedicated feedback portal (web page + mini-program), including the client's rating of task quality (integer type, 0-100 points), evaluation of the team's service attitude (text type, unstructured data), complaint records (text type, including complaint time, complaint content, and handling result), and satisfaction with the response to requirement changes (integer type, 0-100 points). The collection frequency is real-time (synchronized immediately after the client submits the data).

[0068] ④ Cost control data: Data is collected through interfaces with the enterprise's financial system and outsourced cost management system, including team labor costs (floating-point type, unit: RMB / month), consumable costs (floating-point type, unit: RMB / month), management costs (floating-point type, unit: RMB / month), task delivery costs (floating-point type, unit: RMB / task), and cost saving rate (percentage type, calculated as: (budgeted cost - actual cost) / budgeted cost × 100%), collected once a week.

[0069] ⑤ Environmental Impact Data: Data is collected using web crawling and integration with third-party industry databases (such as iResearch Consulting and industry association databases). This includes industry benchmark performance data (floating-point, 0-100 points), performance data of similar outsourcing teams (floating-point, 0-100 points), environmental change data of the client's industry (text-based, such as policy adjustments and market fluctuations), and policy adjustment data (text-based, such as labor outsourcing industry regulatory policies and social security policies). The data is collected monthly, with policy / industry changes collected in real time.

[0070] (2) Data collection and adaptation: For structured data (such as age, rating, cost), it is stored directly in the preset format; for semi-structured data (such as skill certificates, task assignment records), it is stored after being tagged; for unstructured data (such as client evaluations, complaint content), natural language processing (NLP) technology is used to segment words and remove stop words, and then it is converted into structured features before storage to ensure compatibility of subsequent data processing.

[0071] 2.2 Implementation of the data preprocessing module:

[0072] The data preprocessing module performs four processing steps sequentially on the collected multi-source heterogeneous data: cleaning, denoising, normalization, and feature alignment. The core optimization point is the adaptive outlier removal algorithm. The specific implementation process is as follows:

[0073] (1) Data cleaning: missing value filling + format standardization are adopted. For missing values, numerical data is filled with the mean of that dimension (e.g., missing age values ​​are filled with the mean of all collected members' ages), text data is filled with "unknown", and enumerated data is filled with the most frequent value. For data with inconsistent formats (e.g., time formats are "YYYY-MM-DD" and "MM / DD / YYYY"), they are uniformly standardized to the format "YYYY-MM-DDHH:MM:SS" to ensure data format uniformity.

[0074] (2) Data denoising: The core adopts an adaptive outlier removal algorithm, and the specific steps are as follows:

[0075] ① Based on big data statistical analysis, for each type of numerical data dimension (such as task quality score, cost, and schedule), calculate its mean μ and standard deviation σ, using the following formula:

[0076] ;

[0077] ;

[0078] Where n is the number of data samples in this dimension. This is the i-th sample data;

[0079] ② Set the initial outlier range to [μ-3σ, μ+3σ], and mark data that exceeds this range as initial outliers (such as task quality scores below 10 points, negative costs, etc.).

[0080] ③ Based on the characteristics of the labor outsourcing industry, set exclusive secondary verification thresholds: abnormal task completion time threshold (more than twice the standard time of similar tasks), abnormal cost fluctuation threshold (single cost fluctuation exceeding 30%), and abnormal rating threshold (below 20 points or above 95 points without reasonable explanation).

[0081] ④ Perform secondary verification on initial outliers: If an outlier exceeds the industry-specific threshold and has no reasonable explanation (e.g., a task delay due to temporary changes in client requirements is a reasonable explanation), it is determined to be an invalid outlier and directly removed; if an outlier exceeds the initial range but does not exceed the industry threshold, or has a reasonable explanation, it is determined to be a reasonable outlier and corrected using linear interpolation (e.g., if cost fluctuation is 35% but caused by raw material price increases, linear correction is performed using two adjacent cost data sets) to ensure the accuracy of data denoising and avoid mistakenly deleting valid data.

[0082] (3) Data normalization: The Min-Max normalization algorithm is used to map all numerical data to the [0,100] interval (adapting to performance scoring scenarios). The normalization formula is:

[0083] Where x is the original data, The minimum value of the data in this dimension. This is the maximum value for the data in this dimension, ensuring that data of different magnitudes (such as costs and ratings) can be fused and calculated subsequently.

[0084] (4) Feature alignment: For features with the same meaning but different expressions in multi-source data (such as “task acceptance qualified” and “acceptance result is qualified”), unified naming and feature mapping are performed, a feature dictionary is established to ensure the consistency of subsequent feature extraction. After preprocessing, a standardized dataset is output and stored in a distributed database (Hadoop cluster) to support subsequent fast access.

[0085] 2.3 Implementation of the feature extraction module:

[0086] The feature extraction module, based on a standardized dataset, employs a three-tiered extraction model of "basic feature screening + core feature enhancement + risk feature prediction" to extract core features for performance evaluation. The specific implementation method is as follows:

[0087] (1) Basic feature selection: The ANOVA algorithm is used to calculate the correlation coefficient between each preprocessed feature and the performance evaluation result. Features with a correlation coefficient ≥ 0.3 are selected (irrelevant features, such as member gender, place of origin, etc., which have little impact on performance) to obtain the basic feature set, including age, number of skill certificates, years of service, task completion rate, client rating, cost saving rate, etc.

[0088] (2) Core Feature Enhancement: Extract the five core features described in claim 1 from the basic feature set. The specific extraction method for each feature is as follows:

[0089] ① Team fit characteristics: Based on basic information data and client feedback data, three sub-features are extracted: "skill matching degree M", "response fit degree N", and "personnel mobility L" (corresponding to the parameters of self-created formula 2). M is obtained by comparing the skill tags of team members with the skill requirements of the client's task through big data (e.g., if the client's task requires "having an electrician's certificate", the percentage of team members with electrician's certificates × 100 is the initial value of M, which is then adjusted based on skill proficiency). N is calculated by weighting the response satisfaction feedback from the client. L is calculated by "the number of people who left during the period / the average number of people in the team during the period".

[0090] ② Task completion quality characteristics: Integrate the number of reworks, acceptance results, and completion time in the task execution data, and extract the "task completion dimension score T" (corresponding to the parameters of the self-created formula 1). The specific calculation logic is implemented according to claim 7.

[0091] ③ Collaboration efficiency characteristics: Based on the task allocation time, cross-team collaboration times, and communication feedback time in the task execution data, the collaboration efficiency score C (corresponding to the parameters of self-created formula 1) is calculated by weighted summation. The weights are: task allocation time (30%), cross-team collaboration smoothness (40%), and communication feedback time (30%).

[0092] ④ Cost optimization features: Based on cost control data, extract the "cost optimization dimension score S" (corresponding to the parameter of self-created formula 1). The calculation logic is: S = cost saving rate × 60 + cost control compliance rate × 40 (cost control compliance rate = number of times the current cost compliance is achieved / total number of times the current task is completed × 100).

[0093] ⑤ Risk warning features: Extract three sub-features: personnel turnover risk value, task delay risk value, and cost overrun risk value. Specific implementation:

[0094] Employee turnover risk value: The logistic regression model is used for prediction. The training data is the historical turnover data of the past 12 months (labeled as "churned" and "not churned"). The input features are on-the-job duration, salary satisfaction, skill matching degree, and industry turnover rate. The model outputs a risk value (0-1, the higher the value, the greater the risk). A risk value ≥0.7 is judged as high risk.

[0095] Task delay risk value: The calculation logic is as follows: delay risk value = (standard duration - current remaining duration) / standard duration × historical delay probability (historical delay probability = number of delayed tasks in the past 12 months / total number of tasks). A risk value ≥ 0.6 is considered high risk.

[0096] Cost overrun risk value: The calculation logic is: (current actual cost - budgeted cost) / budgeted cost × 100%, and a 20% overrun is considered high risk.

[0097] (3) Feature output: The five core features and their corresponding sub-features extracted are organized into feature vectors and output to the intelligent evaluation module for subsequent performance calculation.

[0098] 2.4 Implementation of the intelligent assessment module:

[0099] The intelligent assessment module is the core module of the system. It has a built-in integrated performance assessment model (BP neural network + random forest), which combines the self-created comprehensive performance assessment formula (1) and the team and client fit coefficient formula (2) to realize performance score calculation and performance prediction. The specific implementation method is as follows:

[0100] (1) Training and deployment of the integrated evaluation model:

[0101] ① Training data: Historical performance data of labor outsourcing teams over the past 36 months and multi-source data were collected. More than 10,000 valid samples were selected and divided into training set, validation set and test set in a ratio of 7:2:1.

[0102] ② Model structure: BP neural network model (input layer has 5 core features, hidden layer has 3 layers, output layer is non-linear feature fitting score), used to fit and calculate non-linear features such as collaboration efficiency C and fitness coefficient F; random forest model (100 decision trees, 10 layers deep), used to assign weights and calculate scores for linear features such as task completion score T and cost optimization score S.

[0103] ③ Model training: The gradient descent algorithm is used to optimize the activation function of the BP neural network (using the ReLU function), and the grid search method is used to optimize the number and depth of decision trees in the random forest. During training, the model parameters are adjusted through the validation set to ensure that the model test accuracy is ≥92%. After training, it is deployed to the intelligent evaluation module and supports real-time access.

[0104] ④ Model Fusion: The nonlinear feature fitting scores output by the BP neural network and the linear feature scores output by the random forest are fused with a weight of 4:6 to obtain the initial scores (T, C, S) for each dimension, providing a basis for subsequent formula calculations.

[0105] (2) Specific applications of the self-created formula:

[0106] ① Calculation of the compatibility coefficient F between the team and the client (application of formula 2):

[0107] The formula is The specific values ​​and calculation examples for each parameter are as follows:

[0108] Parameter values: k1=0.9 (skill matching weight), k2=0.7 (response fit weight), k3=0.4 (personnel mobility weight), satisfying k1+k2+k3=2, ε=0.01 (correction factor).

[0109] Calculation Example: For a technology outsourcing team, M=85 (skill matching), N=90 (response fit), and L=0.1 (personnel turnover), then:

[0110] = = =2790. It should be noted here that since M and N have been normalized to [0,100], the calculated F needs to be further normalized to the interval [0.6,1.2]. After normalization, F=1.15 (excellent fit).

[0111] Normalization logic: ,in The minimum F value calculated historically. The maximum F value calculated historically is used to ensure that F ∈ [0.6, 1.2].

[0112] ② Calculation of the overall performance score P (Application of Formula 1):

[0113] The formula is The specific values ​​and calculation examples for each parameter are as follows:

[0114] Parameter values: α=0.35 (task completion weight), β=0.25 (fitness weight), γ=0.25 (collaboration efficiency weight), δ=0.15 (cost optimization weight), satisfying... =1, D=1.05 (Dynamic adjustment factor, output by the dynamic adjustment module, adapted to the client's urgent task scenario).

[0115] Calculation Example: Based on the above outsourcing team, T=88, C=92 (collaboration efficiency score), S=86 (cost optimization score), and F=1.15 (normalized fit coefficient), then:

[0116] Calculation process: 0.35×88=30.8, 0.25×1.15=0.2875, 0.25×92=23, 0.15×86×1.05=13.545, summing up to P≈67.63, which is 68 points (rounded to the nearest integer), corresponding to a performance level of "Good" (level division: ≥90 Excellent, 80-89 Good, 70-79 Satisfactory, <70 Unsatisfactory).

[0117] (3) Implementation of performance forecasting function:

[0118] Based on the trained fusion model, combined with historical performance data (past 12 months), current task progress (e.g., current task completion rate of 60%), and dynamic adjustment factor D, the performance prediction score for the next 1-3 months is iteratively calculated using formula (1). Specific implementation:

[0119] ① Input data: historical performance score sequence, current core feature data, future task planning data (number of tasks, difficulty, client requirements), and predicted value of dynamic adjustment factor D (predicted by the dynamic adjustment module based on scenario changes).

[0120] ② Iterative calculation: Using the time series forecasting algorithm (ARIMA algorithm) and formula (1), the prediction score is calculated once a month to generate the performance trend curve for the next 1-3 months;

[0121] ③ Risk warning: Mark risky teams with a predicted score below 60, generate warning information (including the reason for the warning, such as "high risk of staff turnover leading to decreased fit, predicted performance score of 58 next month"), and push it to the user terminal to avoid the risk of failing to meet performance targets in advance.

[0122] 2.5 Implementation of the dynamic adjustment module:

[0123] The dynamic adjustment module is used to optimize the evaluation model parameters and feature weights in real time to ensure that the evaluation results adapt to changes in the scenario. The specific implementation logic is as follows:

[0124] (1) Data collection: Historical evaluation data (overall performance score P of the past month) and actual performance feedback data (final evaluation score of the client and actual performance score of task completion) are collected regularly once a week to form a comparison dataset.

[0125] (2) Error calculation: Calculate the evaluation error ,in The overall performance score output by the intelligent evaluation module. For the actual performance feedback score, an error threshold of 5 is set (i.e., when E≤5, the evaluation result is qualified; when E>5, the parameter optimization process is triggered).

[0126] (3) Parameter optimization: When E>5, the following optimization operations are performed based on the gradient descent algorithm:

[0127] ① Weight coefficient adjustment: Adjust the values ​​of α, β, γ and δ in formula (1). For example, when the client reports that "task quality priority is higher than cost", increase α (task completion weight) to 0.4 and decrease δ (cost optimization weight) to 0.1 to ensure that the weights are adapted to the client's needs.

[0128] ②Adjustment of fit weight: Adjust the values ​​of k1, k2 and k3 in formula (2). For example, when the mobility of industry personnel increases, increase k3 (personnel mobility weight) to 0.4 and decrease k1 to 0.8 to enhance the accuracy of fit coefficient;

[0129] ③ Model parameter adjustment: Optimize the parameters of the fusion model (BP neural network + random forest), such as adjusting the number of hidden layer nodes in the BP neural network and the decision tree depth in the random forest to ensure the model fit effect;

[0130] ④ Feature weight adjustment: Adjust the extraction weight of core features based on changes in client needs and task types (such as switching from technical outsourcing to service outsourcing). For example, in the service outsourcing scenario, adjust the weight of features that improve collaboration efficiency and response fit.

[0131] (4) Optimization and verification: After the parameters are adjusted, the accuracy of the model is tested using the validation set data to ensure that the accuracy is ≥92%. If the accuracy is not met, repeat the above optimization process until the error E≤5.

[0132] 2.6 Implementation of the Result Output Module:

[0133] The results output module is used to output the evaluation results in a visual format, supporting multi-terminal viewing, exporting, and bidirectional synchronization. The specific implementation method is as follows:

[0134] (1) Output content: including comprehensive performance score (0-100 points), performance level (excellent / good / qualified / unqualified), analysis of shortcomings in each dimension (such as "the collaboration efficiency score is low, mainly because the cross-team communication time is too long"), optimization suggestions (such as "strengthen team communication training and shorten the communication feedback time"), and risk warning information (if any).

[0135] (2) Visualization format:

[0136] ① Line chart: Displays the historical performance trend over the past 6 months, clearly showing the rise and fall of performance;

[0137] ② Radar chart: Displays the score comparison of the five core features, intuitively presenting the strengths and weaknesses of each dimension;

[0138] ③ Heat map: Shows the distribution of shortcomings (such as the severity of shortcomings such as "high risk of staff turnover" and "cost overrun");

[0139] ④ Text Report: Generates detailed performance evaluation reports, including the scoring calculation process, formula application details, weakness analysis and optimization suggestions, and supports custom report templates.

[0140] (3) Output method:

[0141] ① Terminal display: Synchronously output to the PC management backend, mobile APP, and client's dedicated terminal, supporting real-time viewing;

[0142] ② Format export: Supports exporting evaluation results and reports to PDF and Excel formats. The Excel format includes raw data, score calculation process, and feature data, which facilitates subsequent analysis.

[0143] ③ Two-way synchronization: The evaluation results are automatically pushed to the client's terminal. The client can submit feedback online, and the feedback is synchronized to the user's terminal in real time, forming a closed loop of evaluation-feedback-optimization.

[0144] 2.7 Implementation of the Data Security Module:

[0145] The data security module is used to ensure the security of multi-source collected data and prevent data leakage and tampering. The specific implementation method is as follows:

[0146] (1) Data encryption: The AES-256 irreversible encryption algorithm is used to encrypt and store sensitive data (personnel ID numbers, client business information, cost data). The encryption key is dynamically generated and updated once a quarter to ensure encryption security. Non-sensitive data uses conventional encryption algorithms to ensure data transmission and storage security.

[0147] (2) Access control: A role-based access control system is adopted, which is divided into four levels: administrator, operator, client, and viewer.

[0148] ① Administrator role: Has full operation permissions, can modify system parameters and view all data;

[0149] ② Operator role: Has permissions for data collection, evaluation initiation, and result viewing, but no permissions for parameter modification or viewing sensitive data;

[0150] ③ Client role: Can only view the evaluation results of the corresponding outsourcing team and submit feedback, but does not have permission to view the original data;

[0151] ④ View Roles: Can only view the performance scores and levels of a specified team; no other operation permissions are available.

[0152] (3) Data desensitization: Sensitive data collected is desensitized, such as hiding the middle 8 digits of the personnel's ID number (displayed as 110101****1234), and hiding the core data of the client's business information (such as the contract amount being displayed as "500,000 to 1,000,000 RMB"), to ensure that sensitive data is not leaked.

[0153] (4) Data backup: The off-site multi-backup mode is adopted. The database is backed up once a day and incrementally backed up once an hour. The backup data is kept for 6 months to prevent data loss and support rapid recovery in case of data abnormality.

[0154] III. Implementation of the Overall System Workflow:

[0155] The overall workflow of this system, combined with the implementation details of the above modules, is as follows:

[0156] 1. Data Acquisition: Through multi-source acquisition methods, five categories of multi-source data from labor outsourcing teams are collected in real time / regularly, covering structured, semi-structured, and unstructured data;

[0157] 2. Data preprocessing: The collected data is sequentially cleaned, denoised (adaptive outlier removal), normalized, and feature aligned to output a standardized dataset;

[0158] 3. Feature Extraction: Extract five core features and risk warning sub-features from the standardized dataset, generate feature vectors, and output them to the intelligent assessment module;

[0159] 4. Intelligent assessment: The fusion model processes the feature vectors and calculates the comprehensive performance score by combining the self-created formulas (1) and (2), while realizing performance prediction and risk warning for the next 1-3 months;

[0160] 5. Dynamic adjustment: Periodically calculate the evaluation error, and when the error exceeds the threshold, dynamically optimize the model parameters, formula weights, and feature weights to ensure the accuracy of the evaluation;

[0161] 6. Results Output: The evaluation scores, grades, weakness analysis, and optimization suggestions are output to each terminal in a visual format, supporting export and two-way synchronization;

[0162] 7. Data Security: Data is encrypted, access controlled, and anonymized throughout the entire process. Data is backed up regularly to ensure data security.

[0163] 8. Closed-loop optimization: Receive feedback from clients and users, and adjust system parameters and evaluation logic based on the feedback to form a closed-loop process of "collection-preprocessing-evaluation-adjustment-output-feedback".

[0164] IV. Verification using specific implementation examples:

[0165] To verify the effectiveness of this invention, three different types of labor outsourcing teams were selected (Team A: technology outsourcing, 50 people; Team B: service outsourcing, 80 people; Team C: production outsourcing, 120 people). A three-month performance evaluation was conducted using this system, while an existing evaluation method (manual scoring + single-dimensional quantification) was used as a control group. The verification results are as follows:

[0166] (1) Assessment accuracy: The assessment accuracy of this system was 94% for Team A, 93% for Team B, and 92.5% for Team C, with an average accuracy of 93.2%; the assessment accuracy of the control group was 78% for Team A, 76% for Team B, and 75% for Team C, with an average accuracy of 76.3%. The accuracy of this system was significantly higher than that of the control group, which verified the accuracy of the assessment results.

[0167] (2) Dynamic adaptability: When the client's requirements of Team B were adjusted from "service attitude priority" to "task efficiency priority", the system increased the weight of task completion dimension α to 0.4 and decreased the weight of collaboration efficiency γ to 0.2 through the dynamic adjustment module. After the adjustment, the degree of fit between the evaluation results and the client's actual feedback increased from 82% to 93%, which verified the dynamic adaptability of the system.

[0168] (3) Effectiveness of risk warning: The system predicted that Team C's performance score in the second month would be 58 points (lower than 60 points). The warning reason was "high risk of staff turnover (risk value 0.75) and high risk of cost overrun". After the user adjusted the team configuration and strengthened cost control based on the warning information, Team C's actual performance score in the second month was 65 points, successfully avoiding the risk of failing to meet the performance target, thus verifying the effectiveness of the risk warning function.

[0169] (4) Formula validity: The performance score calculated using the self-created formula (1) and (2) has a correlation coefficient of 0.89 with the final evaluation score of Party A, which is significantly higher than the existing single formula (correlation coefficient 0.62), verifying the rationality and validity of the self-created formula.

[0170] The advantages of this invention are mainly reflected in the following aspects:

[0171] 1. Multi-source big data fusion collection and adaptive preprocessing: Existing technologies mostly use a single data source for collection and use a fixed threshold for outlier removal. This invention integrates five categories of multi-source heterogeneous data and designs an adaptive outlier removal algorithm based on the characteristics of the labor outsourcing industry to avoid erroneous deletion of valid data and improve data quality.

[0172] 2. Self-created formula for quantitative evaluation: Existing technologies mostly use simple weighted summation, which is highly subjective. This invention creates a comprehensive performance evaluation formula (1) and a team-client fit coefficient formula (2), and introduces dynamic adjustment factor D, fit influence weight k1-k3, and correction factor ε to achieve accurate quantification of performance and adapt to different outsourcing scenarios.

[0173] 3. Combining the fusion model with dynamic optimization: Existing technology evaluation models are fixed and cannot adapt to changes in the scenario. This invention adopts a fusion model of BP neural network + random forest, combined with a dynamic adjustment module, to optimize parameters and weights in real time based on evaluation error, so as to ensure the accuracy and adaptability of evaluation results.

[0174] 4. Closed-loop design throughout the entire process: It integrates the entire process of "collection-preprocessing-evaluation-adjustment-output-feedback", and has the functions of performance prediction, risk warning and data security protection. It solves the pain points of existing technology evaluation processes such as fragmentation, lack of risk warning and data insecurity, and adapts to the core needs of labor outsourcing teams with high mobility and multiple clients.

[0175] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0176] The embodiments of the present invention are given for illustrative and descriptive purposes only, and are not intended to be exhaustive or to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described in order to better illustrate the principles and practical application of the invention, and to enable those skilled in the art to understand the invention and to design various embodiments with various modifications suitable for a particular purpose.

Claims

1. A big data-based intelligent performance evaluation system for labor outsourcing teams, characterized in that, include: The data acquisition module is used to collect basic information data, task execution data, client feedback data, cost control data, and environmental impact data from multiple sources for the labor outsourcing team. The multi-source data includes structured data, semi-structured data, and unstructured data. The data preprocessing module is used to clean, denoise, normalize, and align features of the collected multi-source data. It uses an adaptive outlier removal algorithm to remove invalid data and outputs a standardized dataset. The feature extraction module is used to extract core performance evaluation features from a standardized dataset. These core features include team fit features, task completion quality features, collaboration efficiency features, cost optimization features, and risk warning features. The intelligent evaluation module has a built-in performance evaluation model. Based on the extracted core features, the model calculates the comprehensive performance score of the outsourcing team by combining the self-created comprehensive performance evaluation formula, so as to realize multi-dimensional intelligent evaluation of the performance of the labor outsourcing team. The dynamic adjustment module is used to dynamically optimize the parameters and feature weights of the performance evaluation model based on historical evaluation data, real-time task changes, and client requirements. The results output module is used to output the performance scores, ratings, shortcomings analysis and optimization suggestions output by the intelligent evaluation module to the user terminal in a visual form. The comprehensive performance evaluation formula used by the intelligent evaluation module is a self-created formula (1): （1）； Where: P is the comprehensive performance score of the labor outsourcing team, with a value range of [0, 100]; T represents the task completion score, with a value range of [0, 100], which indicates the timeliness and completeness of the outsourced team's task delivery. F is the compatibility coefficient between the team and the client, with a value range of [0.6, 1.2], calculated using the self-created formula (2); C represents the score for collaboration efficiency, with a value range of [0, 100], which indicates the smoothness of collaboration within the team and with the client. S is the score for the cost optimization dimension, with a value range of [0, 100], representing the team's cost control and resource utilization efficiency; D is a dynamic adjustment factor with a value range of [0.9, 1.1], which is dynamically output by the dynamic adjustment module according to the real-time scenario. α, β, γ, and δ are the weight coefficients for each dimension, satisfying the following conditions: And α∈[0.3,0.4], β∈[0.2,0.3], γ∈[0.2,0.3], δ∈[0.1,0.2], can be adaptively optimized through the dynamic adjustment module; The self-created calculation formula for the compatibility coefficient F between the team and the client is formula (2): （2）； Where: M is the matching degree between the skills of the outsourced team members and the client's task requirements, with a value range of [0,100], which is obtained by comparing the skill tags of team members with the skill requirements of the client's task through big data analysis; N represents the degree to which the outsourced team's service response speed matches the client's expectations, with a value range of [0, 100], calculated from the client's feedback data and response time data; L represents the mobility of outsourced team members, with a value range of [0,1]. L = Number of employees leaving in the current period / Average number of employees in the current period. k1, k2, and k3 are the fitness influence weights, satisfying k1 + k2 + k3 = 2, and k1 ∈ [0.8, 1.0], k2 ∈ [0.6, 0.8], and k3 ∈ [0.2, 0.4]. ε is a correction factor with a value of 0.01, used to avoid the abnormal case of the denominator being 0 and to ensure the validity of the formula calculation.

2. The system according to claim 1, characterized in that, The multi-source data of the data acquisition module specifically includes: Basic information data: Age, education, skills certificates, years of experience, and job assignments of outsourced team members; Task execution data: task assignment records, task start / end time, task completion progress, number of rework attempts, and task acceptance results; Client feedback data: Client's rating of task quality, evaluation of team service attitude, complaint records, and satisfaction data on response to changes in requirements; Cost control data: team labor costs, consumable costs, management costs, task delivery costs, and cost savings rate data; Environmental impact data: industry benchmark performance data, performance data of similar outsourcing teams, environmental change data of the client's industry, and policy adjustment data.

3. The system according to claim 1, characterized in that, The adaptive outlier removal algorithm of the data preprocessing module is as follows: Based on big data statistical analysis, the mean μ and standard deviation σ of each data dimension are calculated. Data exceeding the range of [μ-3σ, μ+3σ] are marked as outliers. At the same time, combined with the characteristics of the labor outsourcing industry, industry thresholds for task completion time and cost fluctuations are set. The marked outliers are then verified a second time to remove invalid outliers, retain reasonable outliers, and make corrections.

4. The system according to claim 1, characterized in that, Among the core features extracted by the feature extraction module, the risk warning features include the risk value of outsourced team personnel turnover, the risk value of task delay, and the risk value of cost overrun. The risk value of personnel turnover is predicted by a machine learning model trained with historical turnover data, current on-duty time of personnel, and salary satisfaction data. The risk value of task delay is calculated by combining the remaining time of the task, the current completion progress, and historical delay data.

5. The system according to claim 1, characterized in that, The performance evaluation model of the intelligent evaluation module adopts a fusion model trained with big data. The fusion model consists of a BP neural network model and a random forest model. The BP neural network model is used to fit and calculate nonlinear features, and the random forest model is used to assign weights and calculate scores for linear features. The final performance score is obtained by fusing the output results of the two models.

6. The system according to claim 1, characterized in that, The parameter optimization logic of the dynamic adjustment module is as follows: periodically collect historical evaluation data and actual performance feedback data, and calculate the evaluation error. When E>5, the parameter optimization process is triggered. Based on the gradient descent algorithm, the weight coefficients of α, β, γ, and δ and the values ​​of k1, k2, and k3 in formula (2) are adjusted. At the same time, the extraction weights of each core feature are adjusted in combination with the latest requirements of the client and changes in task type to ensure the accuracy and adaptability of the evaluation results.

7. The system according to claim 1, characterized in that, The calculation logic for the task completion dimension score T is as follows: T = 100 - 5 × number of reworks - 2 × delay duration / standard duration × 10. When T < 0, take T = 0. The number of reworks refers to the cumulative number of reworks for a single task, and the delay duration is the difference between the actual completion time of the task and the standard completion time. The standard duration is derived from big data analysis of the historical completion times of similar tasks.

8. The system according to claim 1, characterized in that, The system also includes a data security module, which is used to encrypt and store the collected multi-source data, control access, and de-identify the data. Sensitive data is encrypted using an irreversible encryption algorithm, and access control adopts a role-based hierarchical management system, allowing only authorized users to access the corresponding level of evaluation data and raw data.

9. The system according to claim 1, characterized in that, The visualization formats of the output module include line charts, radar charts, heat maps, and text reports. It also supports exporting evaluation results to PDF and Excel formats and can automatically push them to the client's terminal to achieve two-way synchronization of evaluation results.

10. The system according to claim 1, characterized in that, The intelligent assessment module also has a performance prediction function. Based on historical performance data, current task progress and dynamic adjustment factor D, it iteratively calculates the performance prediction score of the outsourced team for the next 1-3 months through formula (1), marks risk teams with prediction scores below 60, and pushes early warning information to the user terminal to avoid the risk of performance failure in advance.

Images