Task allocation method for flexible employment

Abstract

The invention relates to the technical field of task allocation, in particular to a task allocation method for flexible employment. According to the technical scheme, the method comprises the following steps that a database of related information of tasks and employees is constructed, wherein the database comprises detailed information of the tasks and comprehensive information of the employees; respectively constructing a task model and an employee portrait based on the information in the database; and matching the task with the employee through an intelligent matching algorithm, wherein the intelligent matching algorithm comprises matching algorithm selection and optimization, real-time matching and dynamic adjustment, and multi-objective optimization and tradeoff. According to the method, multiple matching algorithms are combined for optimization, weights are selected and adjusted according to tasks and employee characteristics, accurate matching is achieved, a task model and employee portrait multi-dimension consideration adaptability is achieved, adaptive changes are matched in real time, tasks and multiple aspects of employees are optimized and balanced in a multi-objective mode, the development of rights and interests of the employees is considered, and the matching quality, the working enthusiasm and the stability are improved; flexible employment resources are managed more effectively, and the task allocation effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of task allocation, and in particular to a method for allocating tasks for flexible employment. Background Art

[0002] In today's fast-paced business environment, the flexible employment model is popular because of its advantages such as being able to quickly respond to changes in business needs and reduce costs. However, this model also brings complexity and challenges to task allocation.

[0003] When allocating existing tasks, the diversity of task requirements does not match the skills of employees. Different tasks have different requirements for skills, experience and knowledge, and traditional allocation methods make it difficult to accurately match employees with tasks, resulting in low quality or low efficiency in completing some tasks. For example, some project tasks that require professional technical knowledge may be assigned to employees who lack relevant experience, thus affecting project progress and results.

[0004] In addition, business volume fluctuates periodically or randomly, and traditional allocation models make it difficult to flexibly adjust labor resources according to real-time task volume, which can easily lead to idle or overly tight human resources. In the peak season, some employees may have too much workload due to unreasonable task allocation, affecting work quality and employee satisfaction; in the off-season, it may cause a waste of human resources.

[0005] In summary, in order to more effectively manage flexible labor resources and improve the effect of task allocation, we propose a task allocation method for flexible labor. Summary of the invention

[0006] The purpose of the present invention is to propose a flexible labor task allocation method to solve the problem that existing business tasks cannot be allocated flexibly and effectively in the background technology.

[0007] The technical solution of the present invention is a method for allocating tasks for flexible employment, comprising the following steps:

[0008] Build a database of task and employee related information, which contains detailed information of tasks and comprehensive information of employees;

[0009] constructing a task model and an employee profile respectively based on the information in the database;

[0010] Matching tasks with employees through intelligent matching algorithms, which include matching algorithm selection and optimization, real-time matching and dynamic adjustment, multi-objective optimization and trade-offs, to achieve effective matching between tasks and employees;

[0011] Make task allocation decisions based on the matching results, monitor the task execution process, and conduct quality assessment and feedback after the task is completed to continuously optimize the task allocation method.

[0012] Optionally, the step of constructing a database of task and employee related information specifically includes the following steps:

[0013] Task information collection: Collect task information through various internal enterprise systems, tools and communication channels with external customers. This information covers task description, task type, task priority, task cycle, required skills and knowledge, and quality requirements.

[0014] Employee information collection: Collect information on flexible workers, including basic information such as name, age, education, work-related information such as work experience, professional skills, performance evaluation records, and work preferences, including working hours, work location, and project type preferences. At the same time, employees are encouraged to actively update information and have a deeper understanding of their potential abilities through interviews and communication.

[0015] Optionally, when constructing a task model, natural language processing technology and data analysis algorithms are used to deeply process the task description, extract key feature words, assign corresponding weights to tasks based on the importance and relevance of the feature words, and convert them into quantifiable feature vectors. Then, machine learning algorithms are used to group tasks with similar feature vectors into one category to construct a task model, and the task database is continuously updated to optimize the accuracy and adaptability of the task model.

[0016] When constructing employee profiles, we use a variety of assessment methods to evaluate employee skills, including but not limited to online skill tests, actual project operation assessments, and peer reviews. The assessment results are converted into quantitative skill indicators to form the employee's skill vector. The employee profile is constructed based on the employee's basic information, skill vector, and work preferences, and the relevant parameters of the employee profile are regularly updated based on changes in the employee's skill learning, project performance, and work preferences.

[0017] Optionally, the matching algorithm selection and optimization specifically includes the following steps:

[0018] Matching algorithms include but are not limited to collaborative filtering algorithms, content-based recommendation algorithms, and multi-layer perceptron algorithms in deep learning algorithms, which are selected based on the characteristics of the task model and employee profile;

[0019] The collaborative filtering algorithm:

[0020] Construct a user-project-task rating matrix R, where rows represent employees and columns represent tasks. ij It represents the performance score of employee i on task j. The similarity between employees is calculated by a specific similarity calculation method. The specific similarity calculation method uses the Pearson correlation coefficient to measure the similarity between employees. The formula is as follows:

[0021]

[0022] Among them, x and y represent two different employees, R xi and R yi are the scores of employee x and employee y on task i, and are the average ratings of employee x and employee y respectively.

[0023] Find a neighbor set N with high similarity to the target employee. N usually sets a similarity threshold, and employees exceeding the threshold are included in the neighbor set.

[0024] Predict the target employee's score for the unperformed task. For the target employee and the unperformed task, the predicted score is calculated using the following formula:

[0025]

[0026] in, is the average rating of target employee u, ρ uv is the similarity between the target employee u and the neighbor employee v, R vt is the rating of the neighbor employee v on task t, is the average rating of neighbor employee v.

[0027] Tasks are sorted according to the predicted scores, and employees who have completed tasks with high predicted scores and performed well, i.e., have high scores, are recommended as candidates for the current task.

[0028] Optionally, optimization strategies are also included:

[0029] Time weights are introduced. Considering that employees’ skills and work performance may change over time, time weights are added to historical scoring data. Tasks completed more recently have larger scoring weights, while tasks completed more distantly have smaller scoring weights. An exponential decay function is used to set the time weight w. t , where t is the difference between the task completion time and the current time, and the formula is w t =e -λt , λ is a parameter for adjusting the time decay speed, which can be adjusted according to the actual data. When calculating the similarity and predicting the score, the original score is multiplied by the corresponding time weight;

[0030] Add a penalty factor. For employees who perform poorly on tasks, that is, when their scores are below a certain threshold, add a penalty factor when calculating similarity to reduce their contribution to similarity calculation, so as to improve the accuracy of recommendations. The penalty factor can be set according to specific business conditions. The score R of employee x on task i is xi If it is lower than the threshold T, then ρ xyMultiply by (1-p), where p is the penalty ratio, which is 0.2.

[0031] Optionally, the content-based recommendation algorithm: matches the task feature vector and the employee skill vector according to the similarity between them, the task feature vector and the employee skill vector are obtained through the task analysis and modeling and employee portrait construction process mentioned above, and the similarity calculation method adopts cosine similarity;

[0032] The specific implementation steps include:

[0033] For the task feature vector T = (t1, t2..., t n ) and employee skill vector E = (e1, e2…, e n ), where n is the feature dimension;

[0034] Cosine similarity calculation

[0035]

[0036] Among them, θ is the angle between the task vector and the employee vector. The closer the cosine value is to 1, the higher the similarity. According to the calculated similarity, the employees are sorted and the employees with higher similarity are selected as candidates for the task.

[0037] Optionally, the multi-layer perceptron algorithm in the deep learning algorithm specifically includes:

[0038] Data preparation: The task feature vector and the employee skill vector are concatenated to form the input data matrix X. At the same time, the matching result of each task-employee pair, whether the task is successfully completed, and the quality score of the completed task are used as label data y;

[0039] Network structure design:

[0040] Input layer: The number of nodes is equal to the sum of the dimensions of the task feature vector and the employee skill vector;

[0041] Hidden layer: Set multiple hidden layers, and the number of nodes in each layer is adjusted based on experience and experiments. The number of hidden layer nodes gradually increases from half of the number of input layer nodes, and then gradually decreases, specifically: [input_dim / 2, input_dim, input_dim / 2]. The hidden layer uses the activation function, ReLU (Rectified Linear Unit) function f(x) = max(0, x), to increase the nonlinear expression ability of the network;

[0042] Output layer: The number of nodes is 1, which indicates the predicted value of the matching degree between the task and the employee. The predicted value is a continuous value, which is a probability value between 0 and 1, indicating the possibility of successful matching. The predicted value also includes a binary classification value after threshold processing, where 0 indicates no match and 1 indicates a match.

[0043] Optionally, also include:

[0044] Model training:

[0045] Define the loss function and use the mean square error loss function for continuous value output. The formula is: L =

[0046]

[0047] Where m is the number of samples, y i is the true label of the i-th sample, is the model prediction value; for binary classification problems, the cross entropy loss function is used;

[0048] Select the optimization algorithm, stochastic gradient descent, Adam, to update the parameters of the network. During the training process, the weights and biases of the network are continuously adjusted to minimize the loss function.

[0049] Divide the training set and the test set, perform multiple iterations of training on the training set, and evaluate the performance of the model on the test set, including the accuracy and recall rate indicators, to adjust the network structure and hyperparameters until satisfactory performance is achieved;

[0050] Optimization strategy

[0051] Dropout regularization: During the training process, some neurons are randomly dropped and the output of the neurons is set to 0 with a certain probability to prevent overfitting. The Dropout probability is generally between 0.2-0.5.

[0052] Early Stopping: During the training process, monitor the performance indicators of the model on the validation set. When the performance of the model on the validation set no longer improves, and the validation set loss no longer decreases or the accuracy no longer improves after several consecutive epochs, stop training to avoid overfitting.

[0053] Learning rate adjustment: Use a learning rate decay strategy, including exponential decay or phased decay. As training progresses, gradually reduce the learning rate so that the model converges to the optimal solution more stably in the later stages of training. The learning rate is calculated according to the formula lr = lr0 × e -αt where lr0 is the initial learning rate, α is the decay rate, and t is the number of training steps or epochs.

[0054] Optionally, the real-time matching and dynamic adjustment specifically include the following steps:

[0055] Real-time matching mechanism, establishes task queues and monitors employee status information in real time. Employee status information includes but is not limited to whether the employee is currently idle, the progress of the task being executed, the estimated completion time, etc. The task queue contains information such as the urgency and priority of the task. When a new task enters the system, the matching algorithm is triggered according to the urgency and priority of the task. The matching score between each employee and the task is calculated by comprehensively considering the results of multiple matching algorithms and adjusting the weights according to the actual situation of the task, and the employee candidates are screened out based on the matching score;

[0056] Dynamically adjust strategies and monitor task requirement changes: Establish a task change monitoring mechanism to obtain task requirement change information in real time through communication channels with the project management system or task initiator, including changes in task objectives, content, skill requirements, etc., re-extract features and model tasks, and evaluate the impact of task requirement changes on current task allocation;

[0057] Employee work status monitoring: continuously monitor the work status of employees in the process of performing tasks, covering aspects such as work progress, work quality, and problems encountered. Information is obtained through work reports submitted by employees, task progress update records, and communication with employees, and the reasons for employees' work difficulties are analyzed;

[0058] Rematching and resource adjustment: decide whether to dynamically adjust task assignments based on changes in task requirements and employee work status monitoring results. If necessary, restart the matching algorithm to screen employees who are more suitable for current task requirements from the remaining available employees. Reallocate replaced employees to other suitable tasks based on their skills and current task queue conditions. At the same time, adjust task resources based on the urgency and importance of the task, add additional time or manpower support, or arrange relevant training to improve employee skills to adapt to task changes.

[0059] Optionally, the multi-objective optimization and trade-off specifically include:

[0060] Multi-objective optimization model construction:

[0061] Objective function definition:

[0062] Assume that the task set is T = {T1, T2…, T m}, the employee set is E = {E1, E2…, E m}, Task T i Assigned to employee E j The decision variable is x iy , x iy =1 means allocation, x iy =0 means no allocation;

[0063] Task completion quality objective function:

[0064]

[0065] Among them, q ij Represents employee E j Complete the task T i The quality assessment value is determined based on the employee's past performance record, skill level and the matching degree of task requirements, with the goal of maximizing the overall task completion quality;

[0066] Task completion time objective function:

[0067]

[0068] Among them, t ij Represents employee E j Complete the task T i The estimated time required, the goal is to minimize the total time to complete all tasks and ensure that tasks can be completed on time;

[0069] Employee satisfaction objective function:

[0070]

[0071] Among them, s ij Represents employee E j For task T i The satisfaction evaluation value of the employee can take into account the employee's job preference, which includes the type of work, working hours, the matching degree between the work location and the task, and the matching degree between the task difficulty and the employee's ability. The goal is to maximize the overall satisfaction of the employee;

[0072] Cost control objective function:

[0073]

[0074] Among them, c ij Represents task T i Assigned to employee E j The costs incurred include employee salaries and training costs, with the goal of minimizing the total cost of task allocation;

[0075] Constraint setting:

[0076] Task assignment uniqueness constraint: Each task can only be assigned to one employee, that is for all i=1, 2, ..., m;

[0077] Employee task capacity constraint: The number of tasks that each employee can perform simultaneously cannot exceed its maximum capacity. j The maximum task capacity is K j ,but for all j=1, 2, ..., n;

[0078] Task skill requirement constraint: Employees must have the skills required to complete the task. i The required skill set is S i , employee E j The skill set is S j , then only when When x ij To take the value of 1;

[0079] Task urgency constraint: For urgent tasks, they need to be completed within a specified time. Let task T i The urgency level is e i The completion time limit is The time it takes the employee assigned to the task to complete the task Must be satisfied;

[0080] Weight coefficient determination:

[0081] Determine weights using the analytic hierarchy process:

[0082] A hierarchical model is constructed, with the target layer set as the comprehensive optimization target of task allocation, the criterion layer including four targets: task completion quality, task completion time, employee satisfaction and cost control, and the scheme layer as various allocation combinations of tasks and employees.

[0083] By means of expert scoring or questionnaire survey, the importance of each goal in the criterion layer relative to the goal layer is compared pairwise, and the judgment matrix A is constructed. The element a of the judgment matrix ij It indicates the importance of the i-th goal relative to the j-th goal, and is quantified using a 1-9 scale, where 1 means that the two goals are equally important, 9 means that one goal is extremely more important than the other, and the intermediate values ​​represent the corresponding transitional degrees;

[0084] Calculate the eigenvector and maximum eigenvalue of the judgment matrix, and obtain the weight coefficient of each target after normalizing the eigenvector. The specific calculation method adopts the power method or the sum-product method.

[0085] Perform consistency check on the judgment matrix and calculate the consistency index where λ max is the maximum eigenvalue of the judgment matrix, n is the order of the judgment matrix, and finds the corresponding average random consistency index RI, obtains it according to the matrix order, and calculates the consistency ratio When CR<0.1, the weight coefficient is considered reasonable.

[0086] Compared with the prior art, the present invention has at least one of the following beneficial technical effects:

[0087] By combining and optimizing multiple matching algorithms, and selecting and adjusting weights based on the different characteristics of tasks and employees, tasks and employees can be matched more accurately. For example, collaborative filtering algorithms use historical data to find similar employee performances for recommendations, content-based recommendation algorithms match similarities between task features and employee skill features, and deep learning algorithms can handle complex nonlinear relationships. Together, they can adapt to various types of tasks and employee situations and improve matching accuracy.

[0088] When constructing task models and employee portraits, we fully consider various aspects of the task information so that the matching process can consider the compatibility of tasks and employees from multiple dimensions, rather than just being limited to a single factor, thereby improving the overall matching quality.

[0089] The real-time matching mechanism can respond to new tasks in a timely manner and quickly determine suitable employee candidates based on the urgency and priority of the task. This enables task allocation to adapt to the ever-changing characteristics of tasks and employee status in a flexible employment environment, ensuring the smooth progress of tasks.

[0090] In the multi-objective optimization and trade-off part, multiple objectives such as task completion quality, task completion time, employee satisfaction and cost control are taken into consideration, the rights and development of employees are taken into consideration, and the work enthusiasm and stability of employees are improved.

[0091] The present invention combines and optimizes multiple matching algorithms, selects and adjusts weights according to task and employee characteristics, accurately matches, builds task models and employee portraits to consider adaptability in multiple dimensions, adapts to changes in real time, and balances tasks and employees in multi-objective optimization, taking into account the development of employee rights and interests, improving matching quality, work enthusiasm and stability, more effectively managing flexible employment resources, and improving the effectiveness of task allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 The figure is a flowchart of a task allocation method for flexible employment. DETAILED DESCRIPTION

[0093] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0094] This embodiment provides a method for assigning tasks to flexible employees. The overall framework includes four main parts: building a database of task and employee related information, building a task model and employee portrait based on the database, matching tasks with employees through an intelligent matching algorithm, and making task assignment decisions and execution monitoring based on the matching results, and performing quality evaluation and feedback. Figure 1 , each part is elaborated in detail below.

[0095] 1. Construct a database of information related to tasks and employees. The database contains detailed information of tasks and comprehensive information of employees. Constructing a database of information related to tasks and employees specifically includes:

[0096] Task information collection: Collect task information through various internal enterprise systems, tools and communication channels with external customers. This information covers task description, task type, task priority, task cycle, required skills and knowledge, and quality requirements.

[0097] Task information is collected through various internal enterprise systems, such as enterprise resource planning system ERP, project management system, tools, task management software, and communication channels with external customers, including emails and customer demand documents. This information includes task descriptions, task types, such as R&D tasks, marketing tasks, etc., task priorities, which can be divided into three levels of high, medium, and low or use numbers 1-5 to indicate the degree of priority, task cycle start and end time or estimated duration; required skills and knowledge, such as programming languages, marketing channel knowledge; quality requirements such as accuracy of more than 95%, and delivery results that meet specific industry standards.

[0098] In terms of data storage, we use relational databases (such as MySQL) to store task information. We establish different data tables to store basic task information, required skills, quality requirements, etc., and use database constraints (such as primary keys, foreign keys, unique constraints, etc.) to ensure data integrity and consistency. At the same time, we use data encryption technology, such as the AES encryption algorithm, to encrypt and store sensitive information and task content involving customer confidentiality to ensure data security.

[0099] Employee information collection: collect information about flexible workers, including basic information such as name, age, education, work-related information such as work experience, professional skills, performance evaluation records, and work preferences, including working hours, work location, and project type preferences. Employees are encouraged to actively update their information, and their potential abilities are deeply understood through interviews and communication. In database storage, employee information is stored in a special employee information table and associated with the task information table (such as through employee numbers). For employee skill information, a many-to-many relationship table is used to store the association between employees and skills for easy query and management.

[0100] 2. constructing a task model and an employee profile based on the information in the database;

[0101] When building the task model, natural language processing techniques such as the word vector model Word2Vec and data analysis algorithms such as principal component analysis PCA are used to reduce the dimension and perform in-depth processing on the task description. When extracting key feature words, the task description is first segmented, and then the semantic vector of each word is determined based on the pre-trained word vector model. The key feature words are selected by calculating the relevance between the word and the task topic, such as cosine similarity. The tasks are assigned corresponding weights based on the importance and relevance of the feature words. For example, a higher weight of 0.8-1.0 is assigned to words that are directly related to the core objectives of the task, and a lower weight of 0.1-0.3 is assigned to auxiliary words, which are converted into quantifiable feature vectors.

[0102] The K-Means clustering algorithm is selected to group tasks with similar feature vectors into one category to build a task model. The K-Means algorithm is selected because it is simple and efficient, suitable for processing large-scale data, and has a good clustering effect on numerical data such as task feature vectors. In the process of building the task model, the task database is continuously updated to optimize the accuracy and adaptability of the task model. For example, after adding 100 task data each time, the clustering algorithm is re-run to adjust the task model;

[0103] When building employee portraits, we use a variety of assessment methods to assess employee skills, including online skill tests accounting for 40% of the total skill assessment weight, actual project operation assessment accounting for 30%, and peer evaluation accounting for 30%. For online skill tests, we standardize the scores of employees and convert 0-100 points into values ​​between 0-1; the actual project operation assessment is comprehensively scored based on multiple factors such as project results and work efficiency. By establishing an evaluation index system, each index is given a certain weight to calculate the total score; peer evaluation collects colleagues' evaluation scores of employees through anonymous questionnaires and then calculates the average. These evaluation results are converted into quantitative skill vectors according to their respective weights.

[0104] Build an employee profile by combining the employee's basic information, skill vector, and work preference. For the quantitative representation of work preference, a coding method is used, such as working time preference: full-time is 1, part-time is 0.5, and flexible working hours is 0.8; work location preference: remote is 0, local office is 1; project type preference: innovative project is 1, conventional project is 0.6. Every six months, the relevant parameters of the employee profile are updated according to the employee's skill learning, obtaining certificates through new skills or completing relevant training courses, project performance, performance evaluation results, and changes in work preferences.

[0105] 3. Matching tasks with employees through intelligent matching algorithms, which include matching algorithm selection and optimization, real-time matching and dynamic adjustment, multi-objective optimization and trade-offs, so as to achieve effective matching between tasks and employees;

[0106] Select a matching algorithm based on the characteristics of the task model and employee profile. If the task type in the task model is relatively clear and the skill distribution in the employee profile is relatively concentrated, the collaborative filtering algorithm is given priority; if the characteristics of the task mainly rely on text descriptions (such as creative tasks) and the employee skills are more reflected through the features in the skill vector, the content-based recommendation algorithm may be more appropriate; for situations where the relationship between tasks and employees is complex and difficult to describe with simple rules, the multi-layer perceptron algorithm in the deep learning algorithm is more advantageous.

[0107] The performance benchmarks for different matching algorithms are root mean square error (RMSE) and recall. Different algorithms are run on a test set containing historical task assignment data, and RMSE is calculated to measure the error between the predicted score and the actual score. Recall is used to evaluate the proportion of suitable employees that the algorithm can correctly recommend. The initial matching algorithm is selected based on these performance indicators, and is continuously monitored and adjusted during actual operation.

[0108] In the matching algorithm selection and optimization part of the intelligent matching algorithm:

[0109] Matching algorithms include but are not limited to collaborative filtering algorithms, content-based recommendation algorithms, and multi-layer perceptron algorithms in deep learning algorithms, which are selected based on the characteristics of the task model and employee profile;

[0110] The collaborative filtering algorithm:

[0111] Construct a user-project-task rating matrix R, where rows represent employees and columns represent tasks. ij It represents the performance score of employee i on task j. The similarity between employees is calculated by a specific similarity calculation method. The specific similarity calculation method uses the Pearson correlation coefficient to measure the similarity between employees. The formula is as follows:

[0112]

[0113] Among them, x and y represent two different employees, R xi and R yi are the scores of employee x and employee y on task i, and are the average ratings of employee x and employee y respectively.

[0114] Find a neighbor set N with high similarity to the target employee. N usually sets a similarity threshold, and employees exceeding the threshold are included in the neighbor set.

[0115] Predict the target employee's score for the unperformed task. For the target employee and the unperformed task, the predicted score is calculated using the following formula:

[0116]

[0117] in, is the average rating of target employee u, ρ uv is the similarity between the target employee u and the neighbor employee v, R vt is the rating of the neighbor employee v on task t, is the average rating of neighbor employee v.

[0118] The tasks are sorted according to the predicted scores, and the employees who have completed the tasks with high predicted scores and performed well, i.e., have high scores, are recommended as candidates for the current tasks.

[0119] This embodiment also includes an optimization strategy: introducing time weights, taking into account that the skills and work performance of employees may change over time, adding time weights to historical scoring data, with tasks completed more recently having larger scoring weights and tasks completed more distant having smaller scoring weights, and using an exponential decay function to set the time weight w t , where t is the difference between the task completion time and the current time, and the formula is w t =e -λt , λ is a parameter for adjusting the time decay speed, which can be adjusted according to the actual data. When calculating the similarity and predicting the score, the original score is multiplied by the corresponding time weight;

[0120] Add a penalty factor. For employees who perform poorly on a task, that is, if their score is below a certain threshold, add a penalty factor when calculating the similarity to reduce their contribution to the similarity calculation to improve the accuracy of the recommendation. The penalty factor can be set according to the specific business situation. The score R of employee x on task i is xi If it is lower than the threshold T, then ρ xy Multiply by (1-p), where p is the penalty ratio, which is 0.2.

[0121] Among them, the content-based recommendation algorithm: matches the task feature vector and the employee skill vector according to the similarity between them. The task feature vector and the employee skill vector are obtained through the task analysis and modeling and employee portrait construction process mentioned above. The similarity calculation method uses cosine similarity.

[0122] The specific implementation steps include:

[0123] For the task feature vector T = (t1, t2..., t n ) and employee skill vector E = (e1, e2…, en ), where n is the feature dimension;

[0124] Cosine similarity calculation

[0125]

[0126] Among them, θ is the angle between the task vector and the employee vector. The closer the cosine value is to 1, the higher the similarity. According to the calculated similarity, the employees are sorted and the employees with higher similarity are selected as candidates for the task. In order to optimize the algorithm, the feature weighting method is adopted to assign different weights to the task features and employee skill features according to the importance of the task and the criticality of different skills in the task. For example, for the core task feature dimensions including the key technical requirements in the R&D task, a larger weight of 1.5-2.0 is given, and for the general feature dimensions, a standard weight of 1.0 is given. At the same time, the algorithm is further optimized by adding extended features to the task feature vector and the employee skill vector, such as adding the degree of innovation requirements of the task to the task feature vector with a value between 0 and 1, and adding the learning speed of the employee to the employee skill vector, which is evaluated as a value between 0 and 1 according to the time and difficulty of learning new skills.

[0127] In addition, the multi-layer perceptron algorithm in the deep learning algorithm specifically includes:

[0128] Data preparation: The task feature vector and the employee skill vector are concatenated to form the input data matrix X. At the same time, the matching result of each task-employee pair, whether the task is successfully completed, and the quality score of the completed task are used as label data y;

[0129] Network structure design:

[0130] Input layer: The number of nodes is equal to the sum of the dimensions of the task feature vector and the employee skill vector;

[0131] Hidden layer: Set multiple hidden layers, and the number of nodes in each layer is adjusted based on experience and experiments. The number of hidden layer nodes gradually increases from half of the number of input layer nodes, and then gradually decreases, specifically: [input_dim / 2, input_dim, input_dim / 2]. The hidden layer uses the activation function, ReLU (Rectified Linear Unit) function f(x) = max(0, x), to increase the nonlinear expression ability of the network;

[0132] Output layer: The number of nodes is 1, which indicates the predicted value of the matching degree between the task and the employee. The predicted value is a continuous value, which is a probability value between 0 and 1, indicating the possibility of successful matching. The predicted value also includes a binary classification value after threshold processing, where 0 indicates no match and 1 indicates a match.

[0133] In this embodiment, model training is also included:

[0134] Define the loss function and use the mean square error loss function for continuous value output. The formula is: L =

[0135]

[0136] Where m is the number of samples, yi is the true label of the i-th sample, is the model prediction value; for binary classification problems, the cross entropy loss function is used;

[0137] Select the optimization algorithm, stochastic gradient descent, Adam, to update the parameters of the network. During the training process, the weights and biases of the network are continuously adjusted to minimize the loss function.

[0138] Divide the training set and the test set, perform multiple iterations of training on the training set, and evaluate the performance of the model on the test set, including the accuracy and recall rate indicators, to adjust the network structure and hyperparameters until satisfactory performance is achieved;

[0139] Optimization strategy

[0140] Dropout regularization: During the training process, some neurons are randomly dropped and the output of the neurons is set to 0 with a certain probability to prevent overfitting. The Dropout probability is generally between 0.2-0.5.

[0141] Early Stopping: During the training process, monitor the performance indicators of the model on the validation set. When the performance of the model on the validation set no longer improves, and the validation set loss no longer decreases or the accuracy no longer improves after several consecutive epochs, stop training to avoid overfitting.

[0142] Learning rate adjustment: Use a learning rate decay strategy, including exponential decay or phased decay. As training progresses, gradually reduce the learning rate so that the model converges to the optimal solution more stably in the later stages of training. The learning rate is calculated according to the formula lr = lr0 × e -αt where lr0 is the initial learning rate, α is the decay rate, and t is the number of training steps or epochs.

[0143] In this embodiment, the intelligent matching algorithm also includes real-time matching and dynamic adjustment, which specifically includes the following steps:

[0144] Real-time matching mechanism, establishes task queues and monitors employee status information in real time. Employee status information includes but is not limited to whether the employee is currently idle, the progress of the task being executed, the estimated completion time, etc. The task queue contains information such as the urgency and priority of the task. When a new task enters the system, the matching algorithm is triggered according to the urgency and priority of the task. The matching score between each employee and the task is calculated by comprehensively considering the results of multiple matching algorithms and adjusting the weights according to the actual situation of the task, and the employee candidates are screened out based on the matching score;

[0145] Dynamically adjust strategies and monitor task requirement changes: Establish a task change monitoring mechanism to obtain task requirement change information in real time through communication channels with the project management system or task initiator, including changes in task objectives, content, skill requirements, etc., re-extract features and model tasks, and evaluate the impact of task requirement changes on current task allocation;

[0146] Employee work status monitoring: continuously monitor the work status of employees in the process of performing tasks, covering aspects such as work progress, work quality, and problems encountered. Information is obtained through work reports submitted by employees, task progress update records, and communication with employees, and the reasons for employees' work difficulties are analyzed;

[0147] Rematching and resource adjustment: decide whether to dynamically adjust task assignments based on changes in task requirements and employee work status monitoring results. If necessary, restart the matching algorithm to screen employees who are more suitable for current task requirements from the remaining available employees. Reallocate replaced employees to other suitable tasks based on their skills and current task queue conditions. At the same time, adjust task resources based on the urgency and importance of the task, add additional time or manpower support, or arrange relevant training to improve employee skills to adapt to task changes.

[0148] The multi-objective optimization and trade-off specifically include:

[0149] Multi-objective optimization model construction:

[0150] Objective function definition:

[0151] Assume that the task set is T = {T1, T2…, T m}, the employee set is E = {E1, E2…, E m}, Task T i Assigned to employee E j The decision variable is x iy , x iy =1 means allocation, x iy =0 means no allocation;

[0152] Task completion quality objective function:

[0153]

[0154] Among them, q ij Represents employee E j Complete the task T i The quality assessment value is determined based on the employee's past performance record, skill level and the matching degree of task requirements, with the goal of maximizing the overall task completion quality;

[0155] Task completion time objective function:

[0156]

[0157] Among them, t ij Represents employee E j Complete the task T i The estimated time required, the goal is to minimize the total time to complete all tasks and ensure that tasks can be completed on time;

[0158] Employee satisfaction objective function:

[0159]

[0160] Among them, s ij Represents employee E j For task T i The satisfaction evaluation value of the employee can take into account the employee's job preference, which includes the type of work, working hours, the matching degree between the work location and the task, and the matching degree between the task difficulty and the employee's ability. The goal is to maximize the overall satisfaction of the employee;

[0161] Cost control objective function:

[0162]

[0163] Among them, c ij Represents task T i Assigned to employee E j The costs incurred include employee salaries and training costs, with the goal of minimizing the total cost of task allocation;

[0164] Constraint setting:

[0165] Task assignment uniqueness constraint: Each task can only be assigned to one employee, that is for all i=1, 2, ..., m;

[0166] Employee task capacity constraint: The number of tasks that each employee can perform simultaneously cannot exceed its maximum capacity. j The maximum task capacity is K j ,but for all j=1, 2, ..., n;

[0167] Task skill requirement constraint: Employees must have the skills required to complete the task. i The required skill set is S i , employee E j The skill set is S j , then only when When x ij To take the value of 1;

[0168] Task urgency constraint: For urgent tasks, they need to be completed within a specified time. Let task T i The urgency level is e i The completion time limit is The time it takes the employee assigned to the task to complete the task Must be satisfied;

[0169] Weight coefficient determination:

[0170] Determine weights using the analytic hierarchy process:

[0171] A hierarchical model is constructed, with the target layer set as the comprehensive optimization target of task allocation, the criterion layer including four targets: task completion quality, task completion time, employee satisfaction and cost control, and the scheme layer as various allocation combinations of tasks and employees.

[0172] By means of expert scoring or questionnaire survey, the importance of each goal in the criterion layer relative to the goal layer is compared pairwise, and the judgment matrix A is constructed. The element a of the judgment matrix ij It indicates the importance of the i-th goal relative to the j-th goal, and is quantified using a 1-9 scale, where 1 means that the two goals are equally important, 9 means that one goal is extremely more important than the other, and the intermediate values ​​represent the corresponding transitional degrees;

[0173] Calculate the eigenvector and maximum eigenvalue of the judgment matrix, and obtain the weight coefficient of each target after normalizing the eigenvector. The specific calculation method adopts the power method or the sum-product method.

[0174] Perform consistency check on the judgment matrix and calculate the consistency index where λ max is the maximum eigenvalue of the judgment matrix, n is the order of the judgment matrix, and finds the corresponding average random consistency index RI, obtains it according to the matrix order, and calculates the consistency ratio When CR<0.1, the weight coefficient is considered reasonable.

[0175] 4. Make task allocation decisions based on the matching results, monitor the task execution process, and conduct quality assessment and feedback after the task is completed to continuously optimize the task allocation method, including:

[0176] Task allocation decision: Based on the output of the intelligent matching algorithm, the final task allocation plan is determined in combination with the actual business situation and management decisions of the enterprise, taking into account human factors and special circumstances such as employees’ temporary leave and special requirements of customers. Task allocation notifications are sent to employees and relevant managers to clarify the task content, requirements, completion time, responsibilities and authority, and provide necessary support and resources;

[0177] Task execution and monitoring: Employees execute tasks according to the task assignment requirements, use project management tools and workflow systems to monitor tasks in real time, record task progress, working hours, and work results, establish a communication mechanism to encourage employees to report problems, and regularly organize task progress reporting meetings to coordinate and resolve various problems and conflicts that arise during task execution;

[0178] Quality assessment and feedback: After the task is completed, the quality of the task is evaluated according to the pre-set quality standards. The evaluation results serve as one of the important bases for employee performance evaluation. At the same time, feedback from employees and customers on the task execution process is collected and used to optimize the task model, employee portrait and matching algorithm to continuously improve the effectiveness and accuracy of the task allocation method.

[0179] The present invention combines and optimizes multiple matching algorithms (collaborative filtering algorithm, content-based recommendation algorithm, multi-layer perceptron algorithm in deep learning algorithm), and selects and adjusts weights according to different characteristics of tasks and employees, so as to match tasks with employees more accurately. For example, the collaborative filtering algorithm uses historical data to find the performance of similar employees for recommendation, the content-based recommendation algorithm matches based on the similarity of task features and employee skill features, and the deep learning algorithm can handle complex nonlinear relationships. Together, they can adapt to various types of tasks and employee situations and improve the accuracy of matching.

[0180] When constructing task models and employee portraits, we fully consider various aspects of the task information (such as task description, type, priority, cycle, required skills and knowledge, quality requirements, etc.) and comprehensive information of the employee (basic information, work-related information, work preferences, etc.), so that the matching process can consider the compatibility of tasks and employees from multiple dimensions, rather than being limited to a single factor, thereby improving the overall matching quality.

[0181] The real-time matching mechanism of the present invention can respond to the arrival of new tasks in a timely manner and quickly determine suitable employee candidates based on the urgency and priority of the tasks. At the same time, the dynamic adjustment strategy can timely re-match or adjust resources during the task execution process according to changes in task requirements (such as changes in goals, content, and skill requirements) and employee work conditions (such as progress lags, quality failures, difficulties encountered, etc.). This enables task allocation to adapt to the characteristics of constantly changing tasks and employee status in a flexible employment environment, ensuring the smooth progress of tasks.

[0182] It is worth noting that in the multi-objective optimization and trade-off part, multiple objectives such as task completion quality, task completion time, employee satisfaction and cost control are taken into consideration, and a balance is achieved between these objectives by building a multi-objective optimization model and determining reasonable weight coefficients. For example, on the premise of meeting the requirements of task completion quality and time, employee satisfaction is improved as much as possible and cost is controlled. This balance helps to maximize the overall benefits of the enterprise, while also taking into account the rights and development of employees, and improving the work enthusiasm and stability of employees.

[0183] Furthermore, the detailed data collection process covers all aspects of information about tasks and employees, and quantifies this information (such as task feature vectors, employee skill vectors, job preference quantification, etc.), providing data support for task allocation. Data-based decisions are more objective and scientific than traditional subjective judgments, reducing unreasonable allocations caused by human factors.

[0184] The entire task allocation process of the present invention is a continuous optimization cycle. Through the quality assessment and feedback mechanism, the evaluation results after the task is completed and the feedback from employees and customers are used to optimize the task model, employee portrait and matching algorithm, so that the system can continuously learn and improve and adapt to the changes in the business development of the enterprise and the flexible employment market.

[0185] The above specific embodiments are only several optional embodiments of the present invention. Based on the technical solutions of the present invention and the relevant inspirations of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A method for allocating tasks for flexible employment, characterized in that: The following steps are involved: Build a database of task and employee related information, which contains detailed information of tasks and comprehensive information of employees; constructing a task model and an employee profile respectively based on the information in the database; Matching tasks with employees through intelligent matching algorithms, which include matching algorithm selection and optimization, real-time matching and dynamic adjustment, multi-objective optimization and trade-offs, to achieve effective matching between tasks and employees; Make task allocation decisions based on the matching results, monitor the task execution process, and conduct quality assessment and feedback after the task is completed to continuously optimize the task allocation method.

2. A method for allocating tasks for flexible employment according to claim 1, characterized in that: The database of task and employee related information construction specifically includes the following steps: Task information collection: Collect task information through various internal enterprise systems, tools and communication channels with external customers. This information covers task description, task type, task priority, task cycle, required skills and knowledge, and quality requirements. Employee information collection: Collect information on flexible workers, including basic information such as name, age, education, work-related information such as work experience, professional skills, performance evaluation records, and work preferences, including working hours, work location, and project type preferences. At the same time, employees are encouraged to actively update information and have a deeper understanding of their potential abilities through interviews and communication.

3. A flexible employment task allocation method according to claim 2, characterized in that: Build a task model, use natural language processing technology and data analysis algorithms to deeply process task descriptions, extract key feature words, assign corresponding weights to tasks based on the importance and relevance of feature words, convert them into quantifiable feature vectors, and then use machine learning algorithms to classify tasks with similar feature vectors into one category to build a task model. Continuously update the task database to optimize the accuracy and adaptability of the task model; Build employee portraits and evaluate employee skills using a variety of assessment methods, including but not limited to online skill tests, actual project operation assessments, and peer reviews. Convert the assessment results into quantitative skill indicators to form the employee's skill vector. Build employee portraits based on the employee's basic information, skill vector, and work preferences, and regularly update the relevant parameters of the employee portrait based on changes in the employee's skill learning, project performance, and work preferences.

4. A method for allocating tasks for flexible employment according to claim 1, characterized in that: The matching algorithm selection and optimization specifically includes the following steps: Matching algorithms include but are not limited to collaborative filtering algorithms, content-based recommendation algorithms, and multi-layer perceptron algorithms in deep learning algorithms, which are selected based on the characteristics of the task model and employee profile; The collaborative filtering algorithm: Construct a user-project-task rating matrix R, where rows represent employees and columns represent tasks. ij It represents the performance score of employee i on task j. The similarity between employees is calculated by a specific similarity calculation method. The specific similarity calculation method uses the Pearson correlation coefficient to measure the similarity between employees. The formula is as follows: Among them, x and y represent two different employees, R xi and R yi are the scores of employee x and employee y on task i, and are the average ratings of employee x and employee y respectively. Find a neighbor set N with high similarity to the target employee. N usually sets a similarity threshold, and employees exceeding the threshold are included in the neighbor set. Predict the target employee's score for the unperformed task. For the target employee and the unperformed task, the predicted score is calculated using the following formula: in, is the average rating of target employee u, ρ uv is the similarity between the target employee u and the neighbor employee v, R vt is the rating of the neighbor employee v on task t, is the average rating of neighbor employee v. Tasks are sorted according to the predicted scores, and employees who have completed tasks with high predicted scores and performed well, i.e., have high scores, are recommended as candidates for the current task.

5. A method for allocating tasks for flexible employment according to claim 4, characterized in that: Also includes optimization strategies: Time weights are introduced. Considering that employees’ skills and work performance may change over time, time weights are added to historical scoring data. Tasks completed more recently have larger scoring weights, while tasks completed more distantly have smaller scoring weights. An exponential decay function is used to set the time weight w. t , where t is the difference between the task completion time and the current time, and the formula is w t =e -λt , λ is a parameter for adjusting the time decay speed, which can be adjusted according to the actual data. When calculating the similarity and predicting the score, the original score is multiplied by the corresponding time weight; Add a penalty factor. For employees who perform poorly on tasks, that is, when their scores are below a certain threshold, add a penalty factor when calculating similarity to reduce their contribution to similarity calculation, so as to improve the accuracy of recommendations. The penalty factor can be set according to specific business conditions. The score R of employee x on task i is xi If it is lower than the threshold T, then ρ xy Multiply by (1-p), where p is the penalty ratio, which is 0.

2.

6. A method for allocating tasks for flexible employment according to claim 5, characterized in that: The content-based recommendation algorithm: matches the task feature vector and the employee skill vector according to the similarity between them. The task feature vector and the employee skill vector are obtained through the task analysis and modeling and employee portrait construction process mentioned above. The similarity calculation method adopts cosine similarity. The specific implementation steps include: For the task feature vector T = (t1, t2..., t n ) and employee skill vector E = (e1, e2…, e n ), where n is the feature dimension; Cosine similarity calculation Among them, θ is the angle between the task vector and the employee vector. The closer the cosine value is to 1, the higher the similarity. According to the calculated similarity, the employees are sorted and the employees with higher similarity are selected as candidates for the task.

7. A method for allocating tasks for flexible employment according to claim 6, characterized in that: The multi-layer perceptron algorithm in the deep learning algorithm specifically includes: Data preparation: The task feature vector and the employee skill vector are concatenated to form the input data matrix X. At the same time, the matching result of each task-employee pair, whether the task is successfully completed, and the quality score of the completed task are used as label data y; Network structure design: Input layer: The number of nodes is equal to the sum of the dimensions of the task feature vector and the employee skill vector; Hidden layer: Set multiple hidden layers, and the number of nodes in each layer is adjusted based on experience and experiments. The number of hidden layer nodes gradually increases from half of the number of input layer nodes, and then gradually decreases, specifically: [input_dim / 2, input_dim, input_dim / 2]. The hidden layer uses the activation function, ReLU (Rectified Linear Unit) function f(x) = max(0, x), to increase the nonlinear expression ability of the network; Output layer: The number of nodes is 1, which indicates the predicted value of the matching degree between the task and the employee. The predicted value is a continuous value, which is a probability value between 0 and 1, indicating the possibility of successful matching. The predicted value also includes a binary classification value after threshold processing, where 0 indicates no match and 1 indicates a match.

8. A method for allocating tasks for flexible employment according to claim 7, characterized in that: Also includes: Model training: Define the loss function and use the mean square error loss function for continuous value output. The formula is: Where m is the number of samples, y i is the true label of the i-th sample, is the model prediction value; for binary classification problems, the cross entropy loss function is used; Select the optimization algorithm, stochastic gradient descent, Adam, to update the parameters of the network. During the training process, the weights and biases of the network are continuously adjusted to minimize the loss function. Divide the training set and the test set, perform multiple iterations of training on the training set, and evaluate the performance of the model on the test set, including the accuracy and recall rate indicators, to adjust the network structure and hyperparameters until satisfactory performance is achieved; Optimization strategy Dropout regularization: During the training process, some neurons are randomly dropped and the output of the neurons is set to 0 with a certain probability to prevent overfitting. The Dropout probability is generally between 0.2-0.

5. Early Stopping: During the training process, monitor the performance indicators of the model on the validation set. When the performance of the model on the validation set no longer improves, and the validation set loss no longer decreases or the accuracy no longer improves after several consecutive epochs, stop training to avoid overfitting. Learning rate adjustment: Use a learning rate decay strategy, including exponential decay or phased decay. As training progresses, gradually reduce the learning rate so that the model converges to the optimal solution more stably in the later stages of training. The learning rate is calculated according to the formula lr = lr o ×e -αt where lr0 is the initial learning rate, α is the decay rate, and t is the number of training steps or epochs.

9. A method for allocating tasks for flexible employment according to claim 1, characterized in that: The real-time matching and dynamic adjustment specifically include the following steps: Real-time matching mechanism, establishes task queues and monitors employee status information in real time. Employee status information includes but is not limited to whether the employee is currently idle, the progress of the task being executed, the estimated completion time, etc. The task queue contains information such as the urgency and priority of the task. When a new task enters the system, the matching algorithm is triggered according to the urgency and priority of the task. The matching score between each employee and the task is calculated by comprehensively considering the results of multiple matching algorithms and adjusting the weights according to the actual situation of the task, and the employee candidates are screened out based on the matching score; Dynamically adjust strategies and monitor task requirement changes: Establish a task change monitoring mechanism to obtain task requirement change information in real time through communication channels with the project management system or task initiator, including changes in task objectives, content, skill requirements, etc., re-extract features and model tasks, and evaluate the impact of task requirement changes on current task allocation; Employee work status monitoring: continuously monitor the work status of employees in the process of performing tasks, covering aspects such as work progress, work quality, and problems encountered. Information is obtained through work reports submitted by employees, task progress update records, and communication with employees, and the reasons for employees' work difficulties are analyzed; Rematching and resource adjustment: decide whether to dynamically adjust task assignments based on changes in task requirements and employee work status monitoring results. If necessary, restart the matching algorithm to screen employees who are more suitable for current task requirements from the remaining available employees. Reallocate replaced employees to other suitable tasks based on their skills and current task queue conditions. At the same time, adjust task resources based on the urgency and importance of the task, add additional time or manpower support, or arrange relevant training to improve employee skills to adapt to task changes.

10. A method for allocating tasks for flexible employment according to claim 1, characterized in that: The multi-objective optimization and trade-off specifically include: Multi-objective optimization model construction: Objective function definition: Assume that the task set is T = {T1, T2…, T m }, the employee set is E = {E1, E2…, E m }, Task T i Assigned to employee E j The decision variable is x iy , x iy =1 means allocation, x iy =0 means no allocation; Task completion quality objective function: Among them, q ij Represents employee E j Complete the task T i The quality assessment value is determined based on the employee's past performance record, skill level and the matching degree of task requirements, with the goal of maximizing the overall task completion quality; Task completion time objective function: Among them, t ij Represents employee E j Complete the task T i The estimated time required, the goal is to minimize the total time to complete all tasks and ensure that tasks can be completed on time; Employee satisfaction objective function: Among them, s ij Represents employee E j For task T i The satisfaction evaluation value of the employee can take into account the employee's job preference, which includes the type of work, working hours, the matching degree between the work location and the task, and the matching degree between the task difficulty and the employee's ability. The goal is to maximize the overall satisfaction of the employee; Cost control objective function: Among them, c ij Represents task T i Assigned to employee E j The costs incurred include employee salaries and training costs, with the goal of minimizing the total cost of task allocation; Constraint setting: Task assignment uniqueness constraint: Each task can only be assigned to one employee, that is for all i=1, 2, ..., m; Employee task capacity constraint: The number of tasks that each employee can perform simultaneously cannot exceed its maximum capacity. j The maximum task capacity is K j ,but for all j = 1, 2, ..., n; Task skill requirement constraint: Employees must have the skills required to complete the task. i The required skill set is S i , employee E j The skill set is S j , then only when When x ij To take the value of 1; Task urgency constraint: For urgent tasks, they need to be completed within a specified time. Let task T i The urgency level is e i The completion time limit is The time it takes the employee assigned to the task to complete the task Must be satisfied; Weight coefficient determination: Determine weights using the analytic hierarchy process: A hierarchical model is constructed, with the target layer set as the comprehensive optimization target of task allocation, the criterion layer including four targets: task completion quality, task completion time, employee satisfaction and cost control, and the scheme layer as various allocation combinations of tasks and employees. By means of expert scoring or questionnaire survey, the importance of each goal in the criterion layer relative to the goal layer is compared pairwise, and the judgment matrix A is constructed. The element a of the judgment matrix ij It indicates the importance of the i-th goal relative to the j-th goal, and is quantified using a 1-9 scale, where 1 means that the two goals are equally important, 9 means that one goal is extremely more important than the other, and the intermediate values ​​represent the corresponding transitional degrees; Calculate the eigenvector and maximum eigenvalue of the judgment matrix, and obtain the weight coefficient of each target after normalizing the eigenvector. The specific calculation method adopts the power method or the sum-product method. Perform consistency check on the judgment matrix and calculate the consistency index where λ max is the maximum eigenvalue of the judgment matrix, n is the order of the judgment matrix, and finds the corresponding average random consistency index RI, obtains it according to the matrix order, and calculates the consistency ratio When CR<0.1, the weight coefficient is considered reasonable.

Images