Information generation method and device

Abstract

The invention discloses an information generation method and device, and relates to the technical field of information processing, and the method comprises the steps: responding to an obtained user performance prediction request, and extracting the work data of a to-be-predicted user from different data sources; inputting the working data of the to-be-predicted user into a prediction model, and activating a corresponding branch network based on a post type corresponding to the to-be-predicted user in the prediction model; the branch network determines a specified feature; the output specified features are processed and then input into the shared coding network to extract general features; outputting prediction results under different performance indicators of each post through an output head based on the general characteristics; and based on the prediction result, determining scores of different post users under different performance indicators, and based on the scores, determining a performance comprehensive value. And intelligent prediction of the performance is realized.

Description

Technical Field

[0001] The present invention relates to the field of information processing technology, and in particular to an information generation method and device. Background Art

[0002] User performance is usually calculated periodically according to performance criteria. For example, different companies set up performance appraisal rules for different users, and then perform performance calculations manually on the performance calculation day.

[0003] This method cannot judge the user's performance in advance, and therefore cannot intervene in the user in advance. Therefore, how to achieve intelligent prediction of user performance is a technical problem to be solved. Summary of the Invention

[0004] The main purpose of the present invention is to provide a data processing method and device to solve the deficiencies in the related art.

[0005] In order to achieve the above-mentioned purpose, according to a first aspect of the present invention, there is provided an information generation method, comprising: extracting work data of a user to be predicted from different data sources; inputting the work data of the user to be predicted into a prediction model, wherein, in the prediction model, a corresponding branch network is activated based on the job type corresponding to the user to be predicted; and the branch network determines the specified features; the output specified features are processed and input into a shared coding network to extract common features; outputting prediction results under different performance indicators of each position based on the common features through an output head; based on the prediction results, determining scores under different performance indicators of users in different positions, and determining a comprehensive performance value based on the scores.

[0006] Optionally, outputting the prediction results of different performance indicators of each position based on the general features through the output head includes: processing the position type to obtain a position type embedding vector, wherein e j =E·OneHot(j), E is the job embedding matrix, j∈{1, 2, ..., N j} is the job type label; the specified feature h branch Concatenate with the job type embedding vector to obtain the concatenated feature h cond =[h branch ;e j ].

[0007] Optionally, outputting the prediction results of different performance indicators of each position based on the general features through the output head includes: outputting the prediction results of different performance indicators of each position based on the general features through the output head includes: splicing the specified features with the general features to obtain target splicing features; outputting the prediction results based on the target splicing features Among them, h is the target feature, σ=ReLU is the activation function, is the bias term; W is the dynamic weight, W = W base +ΔW·e j , where W base is the basic weight of the shared encoding network, and ΔW is the dynamic adjustment tensor.

[0008] Optionally, when training the prediction model, the method includes: collecting historical multi-source heterogeneous data of users under different job types, wherein different multi-source heterogeneous data are associated with job types, and different multi-source heterogeneous data are labeled with different performance indicator information as training samples; determining the branch network corresponding to the data based on the job type, and using the training samples associated with the job type as the input of the corresponding branch network, and using the corresponding performance indicator information as the output to train the prediction model; wherein, during training, the parameters of each branch network are updated based on the loss function, and the parameters of the shared coding network are updated based on the loss function.

[0009] Optionally, when determining the loss function, the task-specific losses corresponding to different job types j include classification losses Regression Loss Based on the parameters of the shared encoding network, a regularization loss is determined; based on the total task loss and the regularization loss, a loss function is determined: Among them, α and β are the weights of the regression task and the classification task respectively.

[0010] Optionally, the prediction model activates a branch network based on the job type corresponding to the user to be predicted, and the branch network outputs the specified features, including: determining the data types contained in the training samples corresponding to different job types; if the training samples corresponding to the same job type contain both numerical types and text types, then the numerical features and text features extracted by each branch network are fused to obtain the specified features: h fused =w num ·h num +w text ·h text , h num is a numerical feature, h text is the text feature; w num , w text The weights corresponding to different job types, w num is the attention weight corresponding to the numerical feature, w text is the attention weight corresponding to the text feature; during training, based on the job type embedding vector e j Adjust modal fusion weight: w num , w text =MLP(e j ); determining the fusion feature, i.e., the specified feature, based on the modality fusion weight.

[0011] Optionally, the prediction model activates a branch network based on the job type corresponding to the user to be predicted, and the branch network outputs specified features including: if the training samples corresponding to the same job type are of numerical type or text type, then for the training samples whose data type is numerical type, the first branch network is activated, wherein the training samples are normalized by the first branch network and then input into the fully connected layer to extract numerical features; or, for the training samples whose data type is text type, the second branch network is activated, wherein the CLS vector is extracted from the training samples by the second branch network and input into the fully connected layer for dimensionality reduction to obtain text features.

[0012] Optionally, the method also includes: randomly generating an initial development path based on the development actions of each performance indicator in the development action library, wherein the development action includes the name of the development action, the time required, the cost required, the improvement of any performance indicator after completing the development action, and the coverage of the development action for various needs of the enterprise; screening the initial development path based on the objective function to obtain development paths corresponding to different positions and different users, wherein each development path includes three stages: short-term, medium-term and long-term, and each stage includes m development actions.

[0013] Optionally, when constructing the objective function, the method includes: determining the difference between the scores of any user in any position under different performance indicators and the target thresholds of each performance indicator corresponding to the position; calculating the corresponding short board weight based on the difference; in, Based on short board weight And the improvement of any performance indicator after completing the development action Δs i , determine the first objective function Based on the set weights of each enterprise demand The coverage c of development actions in the development path for enterprise demand j j Determine the second objective function The sum of the contributions of all development actions in the development path to enterprise demand j is the coverage c j ; Determine the third objective function based on the total time cost T required for the user to complete all development actions in the development path, and the direct financial cost K paid by the company for the user's development path, f3 = -(α·T+β·k), where α and β are cost coefficients; optimize the objective function composed of the first objective function, the second objective function and the third objective function based on the preset total time and total cost as constraints.

[0014] According to a second aspect of the present invention, a user performance intelligent prediction device is provided, comprising a data acquisition unit, which extracts work data of a user to be predicted from different data sources; a prediction unit, which inputs the work data of the user to be predicted into a prediction model, wherein, in the prediction model, a corresponding branch network is activated based on the job type corresponding to the user to be predicted; and specified features are determined by the branch network; the output specified features are processed and input into a shared coding network to extract common features; the prediction results under different performance indicators of each position are output based on the common features through an output head; and an output unit is used to determine the scores under different performance indicators of users in different positions based on the prediction results, and to determine the comprehensive performance value based on the scores.

[0015] The information generation method and apparatus of this embodiment include, in response to a user performance prediction request, extracting work data of the user to be predicted from different data sources; inputting the work data of the user to be predicted into a prediction model, wherein, in the prediction model, a corresponding branch network is activated based on the job type corresponding to the user to be predicted; and the branch network determines a specified feature; the output specified feature is processed and input into a shared coding network to extract a common feature; an output head outputs prediction results under different performance indicators for each job based on the common feature; based on the prediction results, scores under different performance indicators for users in different jobs are determined, and a comprehensive performance value is determined based on the scores. Intelligent performance prediction is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 is a flow chart of an information generation method according to an embodiment of the present invention;

[0018] Figure 2 This is a schematic diagram of an application of the information generation method according to an embodiment of the invention;

[0019] Figure 3 It is a schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0021] It should be noted that the terms "first," "second," and the like in the specification and claims of the present invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate for the embodiments of the present invention described herein. In addition, the terms "including," "having," and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatuses.

[0022] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0023] According to an embodiment of the present invention, a method for generating information is provided, such as Figure 1 As shown, it includes the following steps 101 to 103:

[0024] Step 101: Extract work data of the user to be predicted from different data sources.

[0025] In this step, users can trigger the performance prediction operation for different users in the terminal interface. When the operation is triggered, a performance prediction request is generated and sent to the server. The server collects data from different data sources, including but not limited to work data in the business system used by the user, such as the patent management business system (this is only exemplary), communication data from the social software used by the user, telephone system call record data, data in different development environments, such as code data in different program development environments, etc. The above data sources are exemplary, and data can be obtained from different sources as needed. Users in different positions need to collect different data, so work data is collected from different data sources for users in different positions. For example, users in sales positions can extract sales-related data such as workload from the business system, and can also obtain telephone record data from the telephone system; for example, users in development positions can collect development code-related data from different development environments, such as code volume, code annotation rate, code error rate, etc.

[0026] Step 102: Input the work data of the user to be predicted into the prediction model, wherein, in the prediction model, the corresponding branch network is activated based on the job type corresponding to the user to be predicted; and the branch network determines the specified features; the output specified features are input into the shared coding network to extract the useful features; and the prediction results under different performance indicators of each job are output through the output head based on the common features.

[0027] Step 103: Based on the prediction results, the scores of different performance indicators for users in different positions are determined, and the comprehensive performance value is determined based on the scores.

[0028] In step 102, after the work data of the user to be predicted is input into the prediction model, different branch networks of the prediction model are activated based on the job type label. The work data of users of different job types has different data types, and different branch networks process different types of work data. Different branch networks collect different features for different types of work data. For example, for sales users, numerical features related to call records, such as call time, number of calls, and call duration, can be extracted from phone records; sales figures from business systems can be extracted; and text features of communication content in social software, such as specified semantic features, can be extracted. After the branch networks process different types of data, features for the same job type are merged to obtain specified features.

[0029] Although performance prediction can be performed only for users in a single position to achieve the purpose of intelligent prediction, the performance of users in a certain position should not only consider their own data. For example, when evaluating the execution efficiency of a sales position, in addition to the order processing speed data of the sales position, the task allocation response time of the management position should also be taken into consideration as a factor for the sales position user.

[0030] Therefore, in this embodiment, for a certain position type, a target position type associated with the position type is determined in advance. The target position type, such as a sales position and a management position, is associated. During prediction, when the work data of the user to be predicted is input into the prediction model, the target position type data associated with the position type is also input into the prediction model. For a user in the sales position, the user data of the management position associated with the user is determined, and the position data of the two users are input into the prediction model. Different branch networks are activated based on different position types. The specified features of different types of positions can be obtained through the branch networks. The features are aligned and spliced ​​into the position type embedding vector and then input into the shared coding network. The universal feature vector output at the shared coding layer is used to represent the common performance indicator features across positions.

[0031] After extracting different features for different job types, the branch network continues to extract common features, such as features related to sales job execution efficiency. The input data of different jobs passes through independent feature extraction branches to generate job-specific features.

[0032] In step 103, the general features and the position-specific features are spliced ​​and input into the output head, and the performance prediction results of the user to be predicted are output through the output head. For example, for a sales position, the predicted sales volume, the predicted data reflecting the execution efficiency, etc. can be used to assign scores based on the data, and the comprehensive score can be determined by assigning different weights based on the assigned scores.

[0033] As an optional implementation of this embodiment, after the output specified features are processed, the following steps are performed: processing the job type to obtain a job type embedding vector, wherein e j =E·OneHot(j), E is the job embedding matrix, j∈{1, 2, ..., N j} is the job type label; the specified feature h branch Concatenate with the job type embedding vector to obtain the concatenated feature h cond =[h branch ;e j ]; Based on the splicing features, they are input into the shared encoding network to extract common features.

[0034] As an optional implementation of this embodiment, outputting the prediction results of different performance indicators of each position based on the general features through the output head includes: splicing the specified features with the general features to obtain the target splicing features; outputting the prediction results based on the target splicing features Among them, h is the target feature, σ=ReLU is the activation function, is the bias term; W is the dynamic weight, W = W base +ΔW·e j , where W base is the basic weight of the shared encoding network, and ΔW is the dynamic adjustment tensor.

[0035] In the above optional implementation, the job embedding plays a guiding role, directing the output head to focus on job-related features. d e is the embedding dimension, N j is the total number of positions. Position embedding vector e j By tensor multiplication ΔW·e j Generate parameter offsets to dynamically adjust the weight of the same layer according to the job type, that is, dynamic weight adjustment, such as the e of the technical position j It will increase the weight of neurons related to collaboration ability and suppress the weight of neurons related to sales. baseThe initial weight of the first layer of the output head is a universal parameter that is independent of the position; the ΔW three-dimensional tensor generates the weight offset through the position embedding vector and dynamically adjusts the basic weight.

[0036] The above optional implementation method combines shared features with job embedding to achieve differentiated output (different positions use independent W and b) through feature splicing and dynamic weight generation mechanism. j , ensuring output spatial isolation), and realizing conditional calculation.

[0037] For example, consider the input data of a developer, the core maintainer of a code module, who has received commit records showing 15 defects fixed in the past 30 days, with an average repair time of 1.8 hours. The model extracts commit record features (low average repair time and high code quality) during processing, and uses the shared coding layer to correlate collaboration and execution efficiency to generate common features. Dynamic weighting can be used to emphasize the contribution of collaboration to repair speed, resulting in a predicted value of 1.5 hours per defect.

[0038] The aforementioned optional implementation, through a chained process of dynamic routing, feature fusion, and conditional weight generation, precisely adapts to technical position requirements and extracts key drivers of performance indicators from heterogeneous data. This approach maintains cross-role generalization, and common features support multi-task joint optimization. Efficient inference meets real-time prediction requirements.

[0039] As an optional implementation method of this embodiment, when training the prediction model, the method includes: collecting historical multi-source heterogeneous data of users under different job types, wherein different multi-source heterogeneous data are associated with job types, and different multi-source heterogeneous data are labeled with different performance indicator information as training samples; determining the branch network corresponding to the data based on the job type, and using the training samples associated with the job type as the input of the corresponding branch network, and using the corresponding performance indicator information as the output to train the prediction model; wherein, during training, the parameters of each branch network are updated based on the loss function, and the parameters of the shared coding network are updated based on the loss function.

[0040] In this optional implementation, different types of positions have different data types, such as structured data such as sales and code submission volume; unstructured data such as text (such as weekly reports, etc.). These data can be used as input to the model. Different positions have different performance indicators, and different positions have different output targets. For example, if it is a sales position, its prediction includes sales forecast, which is predicted using regression tasks; for development positions, its prediction can be code quality level prediction, which uses classification tasks and outputs probability distribution. Different types of data for multiple types of positions (such as sales positions and their associated management positions) are input into different branch networks, and the labeled performance indicators are used as output to train the prediction model.

[0041] During training, different branch networks can be trained using different input data and output targets. The input data is the user's multimodal data (numerical values, text, etc.). Each sample is labeled with its job type and performance indicators. For example, sales positions are labeled with sales (regression task, labels are continuous values). For technical positions, the code quality level is labeled (classification task, labels are category numbers). Management positions are labeled with team effectiveness points (regression) + risk level (classification). The error between the predicted value and the true label is calculated using a loss function (such as MSE, cross entropy), and the model parameters are updated through backpropagation. The error between the predicted value and the true label is calculated using a loss function (such as MSE, cross entropy), and the model parameters are updated through backpropagation.

[0042] For example, the training process includes data preparation and batch processing, grouping data by position type to ensure a balanced representation of samples from different positions within each training batch (e.g., each batch contains 10 samples from sales positions and 10 samples from management positions). The forward propagation process involves branch network activation, selecting a branch network based on the position label (e.g., a fully connected network for sales positions and a GNN for technical positions), and then performing feature extraction. The branch features are concatenated with the position embedding and output through the output header. For example, for sales positions, the regression header outputs sales predictions. The technical position output header outputs the code quality rating logits.

[0043] As an optional implementation of this embodiment, when determining the loss function, the task-specific losses corresponding to different job types j include classification losses Regression Loss The total loss of the task is α, β are task weights; based on the parameters of the shared encoding network, the regularization loss λ||θ|| is determined 2 ; Based on the total task loss and the regularization loss, determine the loss function:

[0044] In this optional implementation, the loss function is determined by the task objective type. For example, for development roles, code quality grading is a classification task, using cross-entropy loss; sales forecasting for sales roles can be a regression task, using mean squared error loss. Each role's output head selects a loss function based on the objective type, for example, the numerical branch for sales roles uses a regression task head and mean squared error loss. This means that each branch of the network processes different types of data, and the output head determines the type of loss function. This allows the model to flexibly adapt to multiple roles and tasks while maintaining clarity and rationality in the training objectives.

[0045] Furthermore, we distinguish between shared parameters and dynamic parameters and apply regularization in a targeted manner. Specifically, we regularize the shared parameters (parameters of the general feature extraction layer θ shared , dynamic weight generation layer parameters θ dynamic ) Apply L2 regularization The shared encoding network layer and dynamic weight layer are jointly updated through this loss function. shared The purpose of setting the regularization loss is to limit the complexity of general features and prevent them from being overly dependent on noise or local features. dynamic The purpose of setting regularization loss is to constrain the fluctuation range of dynamic weights and avoid model instability due to excessive adjustment.

[0046] Initially, set λ to a small value (e.g., 0.01) to avoid interfering with the model's rapid convergence. Later in training, gradually increase λ (e.g., 0.1) to tighten constraints on complex parameters. Adjust λ based on the training stage to achieve a balance between fitting and generalization.

[0047] As an optional implementation of this embodiment, the prediction model activates the branch network based on the job type corresponding to the user to be predicted, and the branch network outputs the specified features, including: determining the data type contained in the training samples corresponding to different job types; if the training samples corresponding to the same job type contain both numerical types and text types, then the numerical features and text features extracted by each branch network are fused to obtain the specified features: h fused =w num ·h num +w text ·h text , h num is a numerical feature, h text is the text feature; w num , w text The weights corresponding to different job types, w num is the attention weight corresponding to the numerical feature, w text is the attention weight corresponding to the text feature; during training, based on the job type embedding vector e j Adjust modal fusion weight: w num , w text =MLP(e j ); determining the fusion feature, i.e., the specified feature, based on the modality fusion weight.

[0048] In this optional implementation, the work data of a position type may include multiple types of data, and different types of data are input into different branch networks and then integrated.

[0049] As an optional implementation method of this embodiment, the prediction model activates a branch network based on the job type corresponding to the user to be predicted, and the branch network outputs specified features including: if the training samples corresponding to the same job type are of numerical type or text type, then for the training samples whose data type is numerical type, the first branch network is activated, wherein the training samples are normalized by the first branch network and then input into the fully connected layer to extract numerical features; or, for the training samples whose data type is text type, the second branch network is activated, wherein the CLS vector is extracted from the training samples by the second branch network and input into the fully connected layer for dimensionality reduction to obtain text features.

[0050] In this optional implementation, feature extraction is performed through different branch networks based on the type of data corresponding to the job type.

[0051] As an optional implementation method of this embodiment, the method also includes: randomly generating an initial development path based on the development actions of each performance indicator in the development action library, wherein the development action includes the name of the development action, the required time, the required cost, the improvement of any performance indicator after completing the development action, and the coverage of the development action for various needs of the enterprise; screening the initial development path based on the objective function to obtain development paths corresponding to different positions and different users, wherein each development path includes three stages: short-term, medium-term and long-term, and each stage includes m development actions.

[0052] In this optional implementation, the system generates a personalized career path for each user based on performance predictions. By analyzing the user's strengths and weaknesses, and integrating them with the company's business needs and the user's personal career goals, the system provides customized training recommendations, skill development plans, and career advancement paths to help the user achieve mutual development for both the individual and the organization.

[0053] Specifically, after performance prediction, a user's performance score can be obtained. Based on the scores of various indicators and the total score, the user's shortcomings, such as insufficient execution efficiency, can be identified. Because companies have different strategic needs, an objective function related to both can be established and optimized to determine the development path for different users.

[0054] Development path planning needs to simultaneously optimize user skill improvement (Goal 1), company demand coverage (Goal 2), and cost control (Goal 3). These three goals may conflict (e.g., rapid skill improvement requires high costs), and an optimal solution is needed.

[0055] First, an action library containing all possible development actions must be constructed. Each action includes an action name (e.g., "Advanced Python Training"), a time cost (e.g., 2 months), a financial cost, associated skills (e.g., "Python"), a skill improvement (e.g., 0.2), and the company's business needs covered by the action (e.g., "Cloud Computing" = 0.0, "AI" = 0.3). Each path is divided into three phases: short-term (<6 months), medium-term (6-18 months), and long-term (>18 months). Each phase contains at least one action (a random number). Generating the initial development path involves initializing an empty path; populating each phase with actions by randomly sampling without replacement from the action library and assigning them to each phase while meeting the time constraints; deleting duplicate actions (e.g., selecting the same training multiple times in the same phase); and regenerating actions if the total time or cost exceeds the limit (e.g., total time > 24 months, total cost > 20,000 yuan). This condition also serves as a constraint in the optimization objective function. Total time refers to the total time cost required for a user to complete all actions in the development path; it represents the direct financial cost incurred by the company for the user's development path.

[0056] Furthermore, the initial development paths are screened based on the constructed objective function.

[0057] As an optional implementation of this embodiment, when constructing the objective function, the method includes: determining the difference between the scores of any user in any position under different performance indicators and the target thresholds of each performance indicator corresponding to any position; calculating the corresponding short board weight based on the difference in, Based on short board weight And the improvement of any performance indicator after completing the development action Δs i , determine the first objective function Based on the set weights of each enterprise demand The coverage c of development actions in the development path for enterprise demand j j Determine the second objective function The sum of the contributions of all development actions in the development path to enterprise demand j is the coverage c j ; Determine the third objective function based on the total time cost T required for the user to complete all development actions in the development path, and the direct financial cost K paid by the company for the user's development path, f3 = -(α·T+β·k), where α and β are cost coefficients; optimize the objective function composed of the first objective function, the second objective function and the third objective function based on the preset total time and total cost as constraints.

[0058] In this optional implementation, the user's skill gap improvement speed is maximized

[0059] The improvement weight of the i-th skill is determined by the gap in the short board Δs i The improvement rate of the i-th skill, such as the improvement rate of "Python training" = 0.2, is determined by the effect of the development action.

[0060] Maximize the matching of skills with the company's business needs c j The coverage of the j-th company's demand is accumulated by the weights of the associated actions in the path, c j =∑(action requirement weight × whether the action is selected). Manual setting. For example, c j The sum of the contributions of all actions in the development path to the jth company demand reflects the degree of matching of the path to the company strategy. For example, the company needs to focus on the development of cloud computing in the future (demand weight is 0.9; actions in the user development path:

[0061] Action 1: Participate in AWS certification training (related to "cloud computing", requirement coverage = 0.6).

[0062] Action 2: Participate in cloud migration projects (associated with "cloud computing", demand coverage = 0.4).

[0063] Coverage calculation: c 云计算 =0.6+0.4=1.0 This path completely covers the cloud computing needs, with a contribution of 100%.

[0064] Furthermore, minimizing the time and financial costs, the cost coefficients α and β of time and expenses, and the penalty intensity of controlling costs in the overall goal can be set as needed.

[0065] Furthermore, the constraints can be that the total time is ≤ 24 months (the upper limit of the long-term path) and the total cost is ≤ the company budget (such as 20,000 yuan / year).

[0066] For example, refer to Figure 2 For technical users, skill weaknesses are "cloud computing" (currently 0.5, target 0.8) and "collaboration" (currently 0.6, target 0.7). The company requires cloud computing (weighted 0.9) and AI (weighted 0.5). Optional development actions include AWS certification training, collaboration workshops, and AI algorithm project practice.

[0067] Path A: AWS certification training → AI algorithm project practice, skill improvement, cloud computing: 0.5 → 0.8 (Δs = 0.3); demand coverage, cloud computing: 0.6, AI: 0.5, total coverage: c 云计算

[0068] +c AI=0.6+0.5=1.1.

[0069] Path B: Collaboration Skills Workshop → AWS Certification Training, Skills Improvement, Collaboration Skills: 0.6 → 0.8 (Δs = 0.2), Cloud Computing: 0.5 → 0.8 (Δs = 0.3); Demand Coverage, Cloud Computing: 0.6, Total Coverage c 云计算 =0.6.

[0070] Optimize your choices. If your company's current priorities are cloud computing > AI > collaboration, then Path A is preferred (covering higher-priority needs). If you need to address shortcomings simultaneously, then Path B is preferred (improving both collaboration and cloud computing).

[0071] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0072] According to an embodiment of the present invention, there is also provided a user performance intelligent prediction device, including a data acquisition unit, for extracting work data of the user to be predicted from different data sources in response to obtaining a user performance prediction request; a prediction unit, for inputting the work data of the user to be predicted into a prediction model, wherein, in the prediction model, a corresponding branch network is activated based on the job type corresponding to the user to be predicted; and the branch network determines the specified features; the output specified features are processed and input into a shared coding network to extract common features; the prediction results under different performance indicators of each position are output based on the common features through an output head; the output unit is used to determine the scores under different performance indicators of users in different positions based on the prediction results, and determine the comprehensive performance value based on the scores.

[0073] According to an embodiment of the present invention, the present invention also provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can implement the method described in any of the above embodiments when executing.

[0074] According to an embodiment of the present invention, the present invention further provides a readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to implement the method described in any of the above embodiments when executed.

[0075] According to an embodiment of the present invention, the present invention further provides a computer program product, which can implement the method described in any of the above embodiments when executed by a processor.

[0076] Figure 3 A schematic block diagram of an example electronic device 300 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices.

[0077] like Figure 3 As shown, the electronic device 300 includes a computing unit 301, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 302 or a computer program loaded from a storage unit 308 into a random access memory (RAM) 303. Various programs and data required for the operation of the electronic device 300 can also be stored in the RAM 303. The computing unit 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0078] Multiple components in the electronic device 300 are connected to the I / O interface 305, including an input unit 306, such as a keyboard, a mouse, etc.; an output unit 307, such as various types of displays, speakers, etc.; a storage unit 308, such as a magnetic disk, an optical disk, etc.; and a communication unit 309, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 309 allows the electronic device 300 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0079] The computing unit 301 can be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 301 performs the various methods and processes described above, such as the object matching method. For example, in some embodiments, the object matching method can be implemented as a computer software program that is tangibly contained in a machine-readable medium, such as the storage unit 308. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 300 via the ROM 302 and / or the communication unit 309. When the computer program is loaded into the RAM 303 and executed by the computing unit 301, one or more steps of the method described above can be performed.

[0080] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0081] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. Such program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0082] In the context of the present invention, machine-readable medium can be a tangible medium that can contain or store a program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

Claims

1. A method for generating information, characterized in that: include: Extract the work data of the user to be predicted from different data sources; Inputting the work data of the user to be predicted into the prediction model, wherein, in the prediction model, a corresponding branch network is activated based on the job type corresponding to the user to be predicted; the branch network determines the specified features; the output specified features are processed and input into the shared coding network to extract common features; and the output head outputs the prediction results under different performance indicators for each job based on the common features; Based on the prediction results, the scores of different performance indicators for users in different positions are determined, and the comprehensive performance value is determined based on the scores.

2. The information generation method according to claim 1, characterized in that When the specified features of the output are processed, the method includes: The job type is processed to obtain the job type embedding vector, where e j =E·OneHot(j), E is the job embedding matrix, j∈{1,2,...,N j } is the position type label; The specified feature hbranch is concatenated with the position type embedding vector to obtain the concatenated feature h cond =[h branch ;e j ].

3. The information generation method according to claim 2, characterized in that The output head outputs the prediction results of different performance indicators for each position based on the common features, including: Splicing the splicing feature with the universal feature to obtain a target splicing feature; outputting a prediction result based on the target splicing feature Among them, h is the target splicing feature, σ=ReLU is the activation function, is the bias term; W is the dynamic weight, W = W base +ΔW·e j , where W base is the basic weight of the shared encoding network, and ΔW is the dynamic adjustment tensor.

4. The information generation method according to claim 3, characterized in that When training the prediction model, the method includes: Collect historical multi-source heterogeneous data of users in different job types, where different multi-source heterogeneous data are associated with job types and are labeled with different performance indicator information as training samples; Determine the corresponding branch network based on the job type, use the training samples associated with the job type as the input of the corresponding branch network, and use the corresponding performance indicator information as the output to train the prediction model; During training, the network parameters of each branch are updated based on the loss function, and the shared coding network parameters are updated based on the loss function.

5. The information generation method according to claim 4, characterized in that When determining the loss function, the task-specific losses corresponding to different job types j include classification losses Regression Loss Determine the regularization loss based on the parameters of the shared encoding network; Based on the total task loss and the regularization loss, a loss function is determined: Among them, α and β are the weights of the regression task and the classification task respectively.

6. The information generating method according to claim 5, characterized in that The prediction model activates the branch network based on the job type corresponding to the user to be predicted, and the branch network outputs the specified features including: Determine the data types contained in the training samples corresponding to different job types; If the training samples corresponding to the same job type contain both numerical and text types, the numerical features and text features extracted by each branch network are fused to obtain the specified features: h fused =w num ·h num +w text ·h text , h num is a numerical feature, h tex t is the text feature; w num , w text The weights corresponding to different job types, w num is the attention weight corresponding to the numerical feature, w text is the attention weight corresponding to the text feature; Among them, during training, based on the job type embedding vector e j Adjust modal fusion weight: w num ,w text =MLP(e i ); A fusion feature, namely the designated feature, is determined based on the modality fusion weight.

7. The information generating method according to claim 6, characterized in that The prediction model activates the branch network based on the job type corresponding to the user to be predicted, and the branch network outputs the specified features including: If the training samples corresponding to the same job type are of numerical type or text type, the first branch network is activated for the training samples of numerical type. The training samples are normalized by the first branch network and then input into the fully connected layer to extract numerical features. Alternatively, for training samples whose data type is text type, the second branch network is activated, wherein the CLS vector is extracted from the training sample through the second branch network and input into the fully connected layer for dimensionality reduction processing to obtain text features.

8. The information generation method according to claim 1, wherein: The method further comprises: Randomly generate initial development paths based on the development actions of each performance indicator in the development action library. The development actions include the name of the development action, the required time, the required cost, the improvement in any performance indicator after completing the development action, and the coverage of the development action for various enterprise needs. The initial development path is screened based on the objective function to obtain the development paths corresponding to different positions and different users. Each development path includes three stages: short-term, medium-term and long-term, and each stage includes m development actions.

9. The information generating method according to claim 8, characterized in that: When constructing the objective function, the method includes: Determine the score of any user in any position under different performance indicators and the difference between the target thresholds of each performance indicator corresponding to any position; calculate the corresponding short board weight based on the difference in, Based on short board weight And the improvement of any performance indicator after completing the development action Δs i , determine the first objective function Based on the set weights of each enterprise demand The coverage c of development actions in the development path for enterprise demand j j Determine the second objective function The sum of the contributions of all development actions in the development path to enterprise demand j is the coverage c j ; The third objective function is determined based on the total time cost T required for the user to complete all development actions in the development path and the direct capital cost K paid by the company for the user's development path: f3 = -(α·T + β·K), where α and β are cost coefficients; The objective function consisting of the first objective function, the second objective function and the third objective function is optimized based on the preset total time and total cost as constraints.

10. A user performance intelligent prediction device, characterized in that: include: A data collection unit extracts the work data of the user to be predicted from different data sources; The prediction unit inputs the work data of the user to be predicted into the prediction model, wherein, in the prediction model, a corresponding branch network is activated based on the job type corresponding to the user to be predicted; the branch network determines the specified features; the output specified features are processed and input into the shared coding network to extract common features; and the output head outputs the prediction results under different performance indicators for each job based on the common features; The output unit is used to determine the scores of different performance indicators of users in different positions based on the prediction results, and to determine the comprehensive performance value based on the scores.

Images