CN113722368A

CN113722368A - Data processing method, device, equipment and storage medium

Info

Publication number: CN113722368A
Application number: CN202010443539.XA
Authority: CN
Inventors: 孙莹; 庄福振; 祝恒书; 宋欣; 王鹏
Original assignee: Baidu Online Network Technology Beijing Co Ltd
Current assignee: Baidu Online Network Technology Beijing Co Ltd
Priority date: 2020-05-22
Filing date: 2020-05-22
Publication date: 2021-11-30
Anticipated expiration: 2040-05-22

Abstract

The application discloses a data processing method, a data processing device, data processing equipment and a data processing storage medium for determining skill value, and relates to the technical field of artificial intelligence. The specific implementation scheme is as follows: acquiring target skill data of the target skill of the structured representation and target scene data of the target scene; the target scene has the use requirement on target skills; determining demand closeness data between the target skill data and the target scene data; and determining the skill value of the target skill in the target scene according to the demand compactness data. According to the embodiment of the application, the target skill data and the demand compactness data of the target scene data are introduced to serve as the reference basis for determining the skill value, so that the quantitative calculation of the skill value of the target skill is realized, the value quantification of a single skill is realized, and further, data support is provided for the individual capability assessment or professional planning of a user and the optimization of a salary structure of a skill demand policy on the skill demand.

Description

Data processing method, device, equipment and storage medium

Technical Field

The present application relates to data processing technologies, and in particular, to an artificial intelligence technology, and in particular, to a data processing method, an apparatus, a device, and a storage medium for determining skill value.

Background

In the era of knowledge economy, people continuously invest in skills to improve self-ability and better realize self-value. And through a mode of pricing the working skills, the method can help the individual to effectively evaluate the self ability and guide the individual to carry out short-term or long-term professional planning. In addition, through pricing the working skills, a skill demander can be helped to design a more reasonable salary structure according to the skill requirements.

Therefore, how to use a computer means to process data of complex factors to realize work skill pricing becomes an urgent technical problem to be solved.

Disclosure of Invention

The embodiment of the application provides a data processing method, a data processing device, data processing equipment and a storage medium, so that pricing of single work skills is achieved.

According to a first aspect, an embodiment of the present application provides a data processing method, including:

acquiring target skill data of the target skill of the structured representation and target scene data of the target scene; wherein the target scene has a use requirement for the target skill;

determining demand closeness data between the target skill data and the target scene data;

and determining the skill value of the target skill in the target scene according to the demand closeness data.

According to a second aspect, an embodiment of the present application further provides a data processing apparatus, including:

the structured data acquisition module is used for acquiring target skill data of the target skill represented in a structured mode and target scene data of a target scene; wherein the target scene has a use requirement for the target skill;

the demand closeness data determining module is used for determining demand closeness data between the target skill data and the target scene data;

and the skill value determining module is used for determining the skill value of the target skill in the target scene according to the demand closeness data.

According to a third aspect, an embodiment of the present application further provides an electronic device, including:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a data processing method as provided in the first aspect.

According to a fourth aspect, embodiments of the present application further provide a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute a data processing method provided in the first aspect.

According to the method and the device, the target skill data of the target skill which is structurally represented and the target scene data of the target scene are obtained; the target scene has the use requirement on target skills; determining demand closeness data between the target skill data and the target scene data; and determining the skill value of the target skill in the target scene according to the demand compactness data. The technical scheme is adopted in the embodiment of the application, so that the skill pricing of the target skill is realized.

It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present disclosure, nor do they limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description.

Drawings

The drawings are included to provide a better understanding of the present solution and are not intended to limit the present application. Wherein:

fig. 1 is a flowchart of a data processing method provided in an embodiment of the present application;

FIG. 2A is a flow chart of another data processing method provided by the embodiments of the present application;

FIG. 2B is a block diagram of a skills pricing network provided by an embodiment of the present application;

FIG. 3 is a flow chart of another data processing method provided by an embodiment of the present application;

FIG. 4A is a flow chart of another data processing method provided by the embodiments of the present application;

FIG. 4B is a schematic diagram of a skill graph provided by an embodiment of the present application;

fig. 4C is a block diagram of a skill dominating network according to an embodiment of the present application;

FIG. 5A is a flow chart of another data processing method provided by the embodiments of the present application;

FIG. 5B is a schematic illustration of another skill diagram provided by an embodiment of the present application;

FIG. 5C is a block diagram of a skill pricing portfolio model provided by embodiments of the present application;

FIG. 5D is a block diagram of another skill pricing network provided by embodiments of the present application;

FIG. 5E is a block diagram of another skills management network provided by an embodiment of the present application;

fig. 6 is a block diagram of a data processing apparatus according to an embodiment of the present application;

fig. 7 is a block diagram of an electronic device for implementing the data processing method according to the embodiment of the present application.

Detailed Description

The following description of the exemplary embodiments of the present application, taken in conjunction with the accompanying drawings, includes various details of the embodiments of the application for the understanding of the same, which are to be considered exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present application. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

The embodiment of the application is suitable for determining the skill value of the professional skill on the skill demand side. The vocational skills can include algorithm skills, programming language skills, hardware building skills and the like corresponding to engineering vocational skills; wherein, the skill value can be compensation or partial compensation. Each data processing method provided in the embodiments of the present application may be executed by a data processing apparatus, and the apparatus may be implemented by software and/or hardware and is specifically configured in an electronic device.

Fig. 1 is a flowchart of a data processing method provided in an embodiment of the present application, where the method includes:

s101, acquiring target skill data of a target skill represented in a structured mode and target scene data of a target scene; wherein the target scene has a use requirement for the target skill.

The target skill may be at least one of the professional skills required for a professional position, for example, the target skill may include an algorithmic skill and/or a programming language skill, etc. The target scenario may be a specific post set by a skill demander such as an enterprise and public institution, for example, a specific post set by a company, an enterprise, a public institution, or a research laboratory with a specific attribute.

The target skill data can be understood as structured data obtained by format conversion of attribute data of the target skill. For example, the attribute data of the target skill may include a skill name, and at least one of a familiarity degree, a code development line number, an application project number, and the like.

In order to facilitate subsequent data operation, the target skill data is usually represented in a vector manner. For example, when the target skill name is Java and the proficiency is proficiency, the target skill data of a two-dimensional structure may be Java.

The target scene data can be understood as structured data obtained by converting the format of the attribute data of the demander of the target skill. For example, the attribute data of the target scenario may include at least one of enterprise attributes, historical statistics attributes, skill demand attributes, and the like. The attributes of the enterprise itself may include at least one of information such as unit type, age, city, detailed address, etc. The historical statistical attributes may include historical monthly salary statistical data of the city, for example, at least one of the statistical values of the mean, variance, maximum and minimum values of the historical monthly salaries may be included. The skill requirement attribute may include at least one of information such as work experience and graduation year.

The target scene data may include discrete scene data and/or continuous scene data. The continuous scene data is used for artificially setting the relationship between the scene and the skill, so that the set dimension characteristics can be extracted according to the relationship, namely, the characteristic rule can be searched from the continuous scene data. For example, the internet company B, the recruitment algorithm engineer, a listed company located in a city and having been established for 13 years, has a mean value of the historical monthly salaries of 10000 yuan, a variance of 500 yuan, a highest monthly salary of 12000 yuan, and a lowest monthly salary of 8000 yuan, and then the continuous scene data corresponding to the target scene may be { [10000,500,12000,8000], [ a city, B company, listed company, 13 years, internet class ] }. The discrete scene data is used for limiting the skill requirement of the position, so that the mining of the non-set dimension characteristic can be carried out based on the requirement. For example, an algorithm engineer for company B recruitment requires that the job site be in city C and be assigned a graduate, and the discrete scene data corresponding to the target scene may be [ city C, be assigned a graduate ].

Illustratively, the target skills and target scenes can be extracted from the recruitment data, and can also be determined by the personal ability and skill improvement requirement of the user to be tested. For example, if the user a wants to employ the post B in the company a and has the skill a and the skill B related to the post, the target skill and the target scene can be generated by the above method, and the value of the personal ability can be evaluated by using the data processing method related to the present application. Furthermore, the user A has a plan for learning the skill c and the skill d related to the post at present, and can also generate a target skill and a target scene by combining the skill c and the skill d to be learned, so that the value evaluation of the future personal ability is realized by adopting the data processing method related to the application, and the user A is guided to learn which skills in reverse according to the value evaluation result.

It should be noted that the above examples are only illustrative of the target skill data and the target scenario data by way of example, and are not intended to be limiting. For example, the target skill data and the target scene data may be presented in other manners, and the data content may not be limited to the content itself, and may also be presented in a manner of code identification, for example.

It will be appreciated that in order to improve the accuracy and comprehensiveness of the subsequently determined demand affinity data, the target scene data typically comprises discrete scene data and continuous scene data.

In an alternative embodiment, a skill scene data table corresponding to different skills in each scene may be stored in advance. Accordingly, the target skill data of the target skill represented in the structured representation and the target scene data of the target scene may be obtained by searching the target skill data and the target scene data corresponding to the target skill and the target scene in the skill scene data table. The skill scene data table can be stored in the electronic device locally or other storage devices associated with the electronic device, and the search for the skill scene data table is performed when needed.

In another alternative embodiment, the unstructured target skill attribute data and the unstructured target scene attribute data may be converted into structured target skill data and structured target scene data, respectively, according to a set data conversion rule. The set data transformation rule may be an encoding rule, may be set by a skilled person as desired or empirically, or may be determined by a number of experiments, including but not limited to one-hot (one-hot) encoding.

And S102, determining demand closeness data between the target skill data and the target scene data.

For example, a multi-tier perceptron, factorizer, or other feature extraction network may be employed to perform feature extraction on the target skill data and the target scenario data to determine the demand closeness data between the target skill data and the target scenario data.

In an optional embodiment, in order to improve the efficiency of determining the value skill and the accuracy of the determination result, a pre-trained scene perception network may be adopted to perform feature extraction on the target skill data and the target scene data to obtain the required compactness data. For example, a large amount of sample skill data of the sample skill and sample scene data of a sample scene associated with the sample skill may be used as training samples to be input into a first neural network model constructed in advance, and network parameters of the first neural network model are optimized according to a deviation between a model output result and actual compensation corresponding to the sample skill, so as to obtain a trained scene perception network.

S103, determining the skill value of the target skill in the target scene according to the demand closeness data.

The skill value can be a monetary value of the target skill, and can be, for example, salary in the recruitment data, or a portion of the salary.

The skill value can be a single value, or can be a value interval obtained by combining a value upper limit and a value lower limit, and the value interval is used for limiting a value range so as to realize interval pricing of skills. The upper value limit is not less than the lower value limit, and both the upper value limit and the lower value limit are non-negative data.

Illustratively, a pre-trained second neural network model can be adopted to determine the skill value of the target skill in the target scene according to the demand closeness data. The network structure of the second neural network model and the first neural network model may be the same or different. Illustratively, the second neural network model may be a fully-connected neural network. Alternatively, non-negativity of the determined skill value may be ensured by introducing a non-negative activation function in the second neural network model.

According to the method and the device, the target skill data of the target skill which is structurally represented and the target scene data of the target scene are obtained; the target scene has the use requirement on the target skills; determining demand closeness data between the target skill data and the target scene data; and determining the skill value of the target skill in the target scene according to the demand compactness data. According to the technical scheme, the demand compactness data of the target skill data and the target scene data are introduced to serve as a reference basis for determining the skill value, so that the quantitative calculation of the skill value of the target skill is realized, the value quantification of a single skill is realized, and further, data support is provided for the individual capability assessment or professional planning of a user and the optimization of a salary structure of a skill demand policy on the skill demand.

Fig. 2A is a flowchart of another data processing method provided in an embodiment of the present application, which is improved based on the foregoing technical solutions.

Further, the operation of determining the demand closeness data between the target skill data and the target scene data is refined into the operation of adopting a trained scene perception network to extract the characteristics of the target skill data and the target scene data to obtain the demand closeness data so as to improve the determination mechanism of the demand closeness data.

A data processing method as shown in fig. 2A, comprising:

s201, acquiring target skill data of the target skill in the structured representation and target scene data of a target scene; the target scene has a use requirement for the target skill.

S202, performing feature extraction on the target skill data and the target scene data by adopting a trained scene perception network to obtain the required compactness data.

S203, determining the skill value of the target skill in the target scene according to the demand closeness data.

Optionally, the scene awareness network is constructed based on the first neural network model.

Optionally, the skill value of the target skill in the target scene is determined by using a trained pricing network, namely the second neural network, according to the required compactness data.

For a clear description of the determination process of the demand closeness data and the determination process of the skill value, referring to the structure diagram of the skill pricing network shown in fig. 2B, first, a model training process of the skill pricing network including the scene awareness network and the pricing network is described in detail.

The scene-aware network comprises a skill embedding layer, a scene embedding layer and a feature extraction layer.

In the model training stage, the skill embedding layer is used for performing dimensionality reduction on the sample skill data to obtain a sample skill embedding vector; the scene embedding layer is used for carrying out dimensionality reduction on the sample scene data to obtain a sample scene embedding vector; and the characteristic extraction layer is used for extracting characteristics of at least one of the sample scene data, the sample skill embedding vector and the sample scene embedding vector to obtain sample demand compactness data. And the pricing network is used for determining a value prediction result according to the sample demand compactness data. Correspondingly, network parameters of the scene perception network and the pricing network are adjusted according to the value prediction result and the sample compensation.

It can be understood that the data computation amount when the scene awareness network performs feature extraction can be reduced by determining the embedding vector through the skill embedding layer and the scene embedding layer. The feature extraction layer is used for extracting features based on at least one of the sample scene data, the sample skill embedding vectors and the sample scene embedding vectors, the comprehensiveness and accuracy of the extracted features are improved, and further the performance of the model is improved.

Because the value corresponding to the skill evolves along with time, for example, salaries of 2010 Java engineers and 2020 Java engineers have a significant difference, in order to avoid the influence on the accuracy of the pricing result due to the time lapse, in the model training stage, the skill embedding layer performs the dimension reduction processing on each sample skill data in a time-division manner respectively, so as to obtain the sample skill embedding vector.

In order to reduce the complexity of the model, in an optional embodiment, when the sample skill embedding vector is determined for the sample skill data in a certain time period, the sample skill data in the certain time period is subjected to dimension reduction processing to obtain a sample skill low-rank embedding vector and a shared potential projection matrix respectively; the sample skill embedding vector is determined according to the product of the sample skill low rank embedding vector and the shared potential projection matrix.

In order to avoid higher model complexity caused by the sample skill embedding vectors determined in different time periods and further avoid the occurrence of model overfitting, in an optional embodiment, when network parameters of a scene perception network and a pricing network are adjusted, a loss function can be constructed according to the distance between the sample skill embedding vectors in adjacent time periods; and optimizing the network parameters of the scene-aware network according to the loss function.

It can be understood that the model is constrained by embedding the distance between vectors through sample skills in adjacent time intervals, and the rapid change of the demand compactness data extracted by the model along with time is limited, so that the complexity of the model is reduced, and the over-fitting condition of the model is avoided.

Illustratively, the distance between the sample skill embedding vectors for adjacent time periods is determined by norm calculation. Alternatively, the norm employed may be the F-norm (Frobenius norm).

Optionally, the scene data of the skill includes continuous scene data and/or discrete scene data, and thus the scene embedding layer may include a discrete scene embedding layer, which is used to perform dimension reduction processing on the discrete scene data to obtain a sample discrete scene embedding vector; the scene embedding layer can also comprise a continuous scene embedding layer, and the continuous scene embedding layer is used for carrying out dimension reduction processing on continuous scene data to obtain sample continuous scene embedding vectors.

In order to improve the comprehensiveness and accuracy of the required compactness data and further improve the model precision, in an optional embodiment, the feature extraction layer may extract the required compactness data of different levels between the sample skills and the sample scenes at the same time.

Optionally, determining first-order sample demand compactness data according to the sample skill embedding vector and the sample scene data to determine the direct relation of the single feature of the input data to the skill value; determining second-order sample demand compactness data according to the sample skill embedding vector and the sample scene embedding vector so as to determine the correlation relation of the binary feature combination of the input data to the skill value; and determining high-order sample requirement compactness data according to the sample scene data, the sample scene embedding vector and the sample skill embedding vector so as to determine high-order association of the multivariate feature combination of the input data to the skill value.

Illustratively, the determination of the high-order sample requirement compactness data may be implemented using a deep neural network. For example, the deep neural network may be an MLP (Multilayer Perceptron).

Optionally, performing feature fusion on the sample scene data, the sample scene embedding vector and the sample skill embedding vector by using MLP; and activating the fused features by adopting a plurality of sequentially connected activation layers to finally obtain the high-order sample required compactness data.

Optionally, the pricing network may be provided with at least one non-negative activation function, and the sample demand compactness data including at least one of the first-order sample demand compactness data, the second-order sample demand compactness data, and the high-order sample demand compactness data is activated through the non-negative activation function, so as to obtain a value prediction result. Correspondingly, network parameters in the pricing network and the scene perception network are optimized according to the difference between the value prediction result and the sample compensation.

In an optional embodiment, in order to ensure that the skill value network has a value interval prediction function, two non-negative activation functions can be set in the pricing network, and activation processing is performed on sample demand compactness data respectively to obtain a value upper limit prediction result and a value lower limit prediction result.

Optionally, two non-negative activation functions are set in the pricing network, and activation processing is performed on the sample demand compactness data to obtain a value upper limit prediction result and a value lower limit prediction result, which may be: activating the sample demand compactness data through one of the non-negative activation functions to obtain a value lower limit prediction result; activating the sample required compactness data through another non-negative activation function to obtain the length of a value interval; and taking the sum of the value interval length and the value lower limit prediction result as a value upper limit prediction result.

Correspondingly, network parameters in a pricing network and a scene perception network are optimized according to the difference between the value upper limit prediction result and the sample compensation upper limit and the difference between the value lower limit prediction result and the sample compensation lower limit, and the skill value of the target skill in the target scene can be predicted by adopting the trained skill value network.

It should be noted that the training process and the use process of the skill value network can be implemented by using the same or different electronic devices.

In the using stage of the skill value network, the scene perception network in the skill value network comprises an embedding layer and a feature extraction layer; the embedding layer is used for respectively carrying out dimensionality reduction processing on the target skill data and the target scene data to obtain a skill embedding vector and a scene embedding vector; the feature extraction layer is configured to perform feature extraction on at least one of the target scene data, the skill embedding vector, and the scene embedding vector to obtain the required compactness data.

Specifically, the embedding layer comprises a skill embedding layer and is used for performing dimension reduction processing on target skill data to obtain a skill embedding vector; and the scene embedding layer is used for carrying out dimension reduction processing on the target scene data to obtain a scene embedding vector.

It can be understood that the data computation amount when the scene awareness network performs feature extraction can be reduced by determining the embedding vector through the skill embedding layer and the scene embedding layer. The feature extraction layer is used for extracting features based on at least one of the target scene data, the skill embedding vector and the scene embedding vector, the comprehensiveness and the accuracy of the extracted features are improved, data support is provided for determining a skill value determination result, and meanwhile guarantee is provided for the accuracy of the skill value determination result.

Illustratively, the trained network parameters in the skill embedding layer are adopted to perform dimensionality reduction processing on the target skill data to obtain a skill embedding vector. Optionally, the target skill data may be subjected to dimensionality reduction to obtain a target skill low-rank embedded vector and a target shared potential projection matrix respectively; and determining the skill embedding vector according to the product of the target skill low-rank embedding vector and the target sharing potential projection matrix.

Exemplarily, if the target scene data comprises target discrete scene data, performing dimensionality reduction processing on the target discrete scene data by adopting trained network parameters in a discrete scene embedding layer to obtain discrete scene embedding vectors; and if the target scene data comprises target continuous scene data, performing dimensionality reduction on the target continuous scene data by adopting the trained network parameters in the continuous scene embedding layer to obtain a continuous scene embedding vector.

Illustratively, the trained network parameters in the feature extraction layer are adopted to extract features of at least one of target discrete scene data, target continuous scene data, skill embedding vectors, discrete scene embedding vectors and continuous scene embedding vectors to obtain required compactness data. And the network parameters in the feature extraction layer are parameters after model training is finished.

In order to improve the comprehensiveness and accuracy of the demand closeness data and further improve the accuracy of the skill value determination result, in an optional embodiment, the feature extraction layer may extract the demand closeness data of different levels between the target skill and the target scene at the same time.

Illustratively, the demand compactness data comprises at least one of first order demand compactness data, second order demand compactness data and higher order demand compactness data.

Optionally, determining first-order demand compactness data according to the target discrete scene data and/or the target continuous scene data and the skill embedding vector, wherein the first-order demand compactness data is used for representing the direct relation between the target scene and the single characteristic of the target skill to the skill value; determining second-order demand compactness data for representing the correlation relation of binary feature combinations of the target scene and the target skill to the value skill according to the discrete scene embedding vector and/or the continuous scene embedding vector and the skill embedding vector; and determining high-order demand compactness data according to the target continuous scene data and the continuous scene embedding vector and/or the target discrete scene data and the discrete scene embedding vector and the skill embedding vector, wherein the high-order demand compactness data is used for representing high-order association of the multi-element feature combination of the target skill and the target scene to the skill value. Optionally, the demand compactness data is activated according to at least one non-negative activation function set in the pricing network, so that a skill value of the target skill in the target scene is obtained. And the network parameters in the non-negative activation function are corresponding parameters after the model training is finished.

In an optional embodiment, if the skill value network has a value interval prediction function, correspondingly, the required compactness data can be activated respectively according to two non-negative activation functions set in the pricing network, so as to obtain a value upper limit and a value lower limit of the skill value of the target skill in the target scene.

Optionally, if two non-negative activation functions are set in the pricing network, activation processing can be performed on the required compactness data through one of the non-negative activation functions to obtain a value lower limit of the target skill in the target scene; activating the required compactness data through another non-negative activation function to obtain the length of a value interval; and taking the sum of the length of the value interval and the value lower limit as the value upper limit of the target skill in the target scene.

According to the embodiment of the application, the compactness data is determined and operated, the trained scene perception network is refined, the target skill data and the target scene data are subjected to feature extraction, the demand compactness data is obtained, the determination mechanism of the demand compactness data is perfected, the extraction efficiency of the demand compactness data is improved through the use of a machine learning model, and meanwhile, the accuracy and the comprehensiveness of the extracted demand compactness data are improved.

Fig. 3 is a flowchart of another data processing method provided in the embodiment of the present application, which is improved based on the foregoing technical solutions.

Further, after the operation "determining the skill value of the target skill in the target scene according to the demand closeness data", additionally determining the dominance weight of the target skill in each adjacent skill of the target skill; the dominance weight representing the importance of the target skill in each of the neighboring skills; weighting the skill value according to the dominance weight to update the skill value "to achieve finer grained pricing of the target skill.

A data processing method as shown in fig. 3, comprising:

s301, acquiring target skill data of the target skill in the structured representation and target scene data of a target scene; the target scene has a use requirement for the target skill.

S302, determining demand closeness data between the target skill data and the target scene data.

S303, determining the skill value of the target skill in the target scene according to the demand closeness data.

S304, determining the dominant weight of the target skill in each adjacent skill of the target skill; the dominance weight represents the importance of the target skill in each of the neighboring skills.

It should be noted that, in general, there may be more than one skill required by the skill demander, and there is a certain relationship between the skills. For example, a Java engineer needs to be not only proficient in Java language but also have database processing capability such as MySQL, and the skill values of Java skills and MySQL skills have certain influence on each other. For another example, if a Java engineer with Java skills has C + + skills, the skill values of the Java skills and the C + + skills of the Java engineer will affect each other and have a difference. For another example, a Java engineer having both Java skills and C + + skills has a difference in the skill value corresponding to the same skills from a C + + engineer having both C + + skills and Java skills.

In order to strip the value of skills with mutual influence so as to realize the value determination of a single skill, the dominant weights of different skills can be determined, so that the pricing of the skills with finer granularity on the corresponding skills is realized according to the dominant weights of the skills.

In an optional embodiment, the occurrence frequency of each skill in the recruitment data of the same position can be counted in advance by acquiring a large amount of recruitment data; accordingly, the dominant weight of the target skill in each adjacent skill of the target skill may be determined by using the frequency of occurrence of each skill as the dominant weight of the target skill.

Since the number of adjacent skills of the target skill may be much smaller than the number of co-occurring skills of the target skill when the skills are counted, after the occurrence frequency of each skill in the recruitment data of the same position is counted, the found occurrence frequency of the target skill may be divided by the sum of the occurrence frequency of the target skill and the occurrence frequency of the adjacent skills to obtain the dominance weight of the target skill.

To further improve the accuracy of the determined dominance weights, in an alternative embodiment, demand closeness data of the target skills in the target scenario may also be introduced to assist in determining the dominance weights. Illustratively, a dominance weight for a target skill is determined based on skill association data for the target skill and respective adjacent skills of the target skill, and the demand closeness data. The skill association data may be determined based on the co-occurrence frequency of the target skill and the neighboring skills.

It should be noted that the neighboring skills of the target skill may be acquired in advance before determining the dominant weight, and may be acquired together with the target skill data of the target skill in the structured representation, for example.

S305, weighting the skill value according to the dominance weight so as to update the skill value.

Illustratively, the skill value of the target skill in the target scene is weighted according to the dominance weight, so that the influence of the adjacent skill on the target skill is stripped, and the accuracy of the determined skill value is improved.

When the skill value comprises the upper value limit and the lower value limit, the upper value limit and the lower value limit can be weighted respectively through the domination weight for updating the upper value limit and the lower value limit of the skill value, so that a foundation is laid for the accuracy of the skill value determination result.

It can be understood that in a recruitment requirement, part of skills can improve the lower limit of recruitment compensation, and part of skills can improve the upper limit of recruitment compensation, so that the importance of the upper value limit and the lower value limit of the same skill in the skill value of a single skill may be different, that is, the upper limit dominance weight of the upper value limit and the lower limit dominance weight of the lower value limit may also be different.

To improve the accuracy of the determined dominance weights, when the skill value of the target skill in the target scenario includes an upper value limit and a lower value limit, the determined target value also includes an upper bound dominance weight and a lower bound dominance weight. The upper limit dominance weight and the lower limit dominance weight are the same or different.

In an alternative embodiment, the maximum and minimum values of lower compensation limits including the target skill and the adjacent skill, and the maximum and minimum values of upper compensation limits may be counted when the frequency of occurrence of each skill in the recruitment data for the same position is counted; determining a lower limit adjustment ratio according to the maximum value and the minimum value of the salary lower limit, the difference value between the maximum value and the minimum value of the salary lower limit and the ratio of the minimum value or the maximum value of the salary lower limit; determining an upper limit adjustment ratio according to the difference value between the maximum value and the minimum value of the salary upper limit, the ratio of the maximum value to the minimum value, and the ratio of the maximum value to the minimum value; determining the lower limit domination weight of the target skill according to the product of the lower limit regulation proportion and the occurrence frequency of the target skill; and determining the upper limit dominance weight of the target skill according to the product of the upper limit adjustment proportion and the occurrence frequency of the target skill.

In order to further improve the accuracy of the determined dominance weights, in another alternative embodiment, the upper and lower dominance weights of the target skill may also be determined separately according to the skill association data of the target skill and each adjacent skill of the target skill, and the demand closeness data. The skill association data may be determined based on the co-occurrence frequency of the target skill and the neighboring skills.

Correspondingly, the skill value is weighted according to the domination weight to update the skill value, and the upper limit domination weight can be adopted to weight the upper limit of the value of the skill value to update the upper limit of the value; and weighting the value lower limit of the skill value by adopting the lower limit dominance weight so as to update the value lower limit.

In an optional embodiment, after the skill values of a plurality of target skills in the same target scene are respectively determined, for one target skill, the dominant weight of the target skill is determined by taking other target skills as the adjacent skills of the target skill; and carrying out weighted summation on the corresponding skill values according to the domination weight of each target skill, and taking the final result as the comprehensive compensation corresponding to the plurality of target skills.

It can be understood that the method of comprehensive compensation determination is adopted, and the post compensation determination can be performed aiming at the recruitment data containing at least two skill requirements, so that reference is provided for the compensation standard when the skill demander releases the recruitment data. In addition, aiming at the aspect of personal skill improvement, the value of the user can be effectively evaluated according to the personal skill requirement and the job-entering intention (corresponding skill scene), so that the comprehensive quantification of the skill value of the user is realized, and the user can be guided to carry out short-term or long-term job planning.

The embodiment of the application determines the dominance weight of a target skill in the adjacent skills of the target skill; the dominance weight represents the importance of the target skill in each of the neighboring skills; the skill value is weighted according to the dominance weight to update the skill value. According to the technical scheme, the domination weight of the target skill is introduced, the skill value of the target skill can be stripped from other skills which influence the target skill mutually, so that the skill pricing of a single skill with finer granularity is realized, and the accuracy of the determined skill value is improved.

Fig. 4A is a flowchart of another data processing method provided in the embodiment of the present application, which is improved based on the foregoing technical solutions.

Further, the operation "determining the dominant weight of the target skill in each adjacent skill of the target skill" is refined into "determining the dominant weight of the target skill according to the skill association data of the target skill and each adjacent skill of the target skill and the requirement closeness data" so as to perfect the determination mechanism of the dominant weight.

A data processing method as shown in fig. 4A, comprising:

s401, acquiring target skill data of the target skill structurally represented and target scene data of a target scene; the target scene has a use requirement for the target skill.

S402, determining demand closeness data between the target skill data and the target scene data.

And S403, determining the skill value of the target skill in the target scene according to the demand closeness data.

S404, determining the dominance weight of the target skill according to the skill association data of the target skill and each adjacent skill of the target skill and the demand closeness data.

The dominance weight represents the importance of the target skill in each of the neighboring skills.

The proximity skills characterize the skills that can appear in the same recruitment data at the same time as the target skill, or other skills that are in demand at the same time as the target skill in the same skill scenario.

The skill association data may be a matrix constructed from co-occurrence frequencies of the target skill and the neighboring skills.

Illustratively, the acquisition of a large amount of recruitment data and the construction of a skill map are aimed at in advance. The skill map is described in detail with reference to a skill map structure diagram shown in fig. 4B.

See the skill graph in FIG. 4B, including nodes (corresponding to circles in the graph), edges, and attribute descriptions. Each node corresponds to a skill; the connecting edges of the two nodes indicate that the skills corresponding to the two nodes appear in the same recruitment data, and the skills corresponding to the two nodes are represented as mutually adjacent skills. Each node has a node attribute description that characterizes the node's familiarity with, e.g., proficiency, understanding, etc., the corresponding technology. Each edge corresponds to an edge attribute description for characterizing the co-occurrence frequency of the corresponding skills of the two connected nodes in the mass recruitment data. The co-occurrence frequency is the ratio of the number of recruitment data to the total number of recruitment data for two skills occurring simultaneously.

For example, the skill map may be stored in advance, and accordingly, after the target skill and the adjacent skill of the target skill are acquired, the co-occurrence frequency of the target skill and each adjacent skill may be determined by searching the skill map, and then the matrix including the co-occurrence frequency of each target skill and each adjacent skill may be obtained.

Illustratively, a graph network may be employed to determine a dominance weight for a target skill based on skill association data for the target skill and respective adjacent skills of the target skill, and demand closeness data. Illustratively, the graph network may be a graph convolution neural network or other neural network.

In order to improve the determination efficiency and determination accuracy of the determination result of the dominance weight, in an alternative embodiment, a pre-trained skill dominance network may be adopted, and the dominance weight of the target skill is determined according to the skill association data of the target skill and each adjacent skill of the target skill and the demand closeness data. The skill domination network adopts a large amount of sample skill and sample skill correlation data of each adjacent skill of the sample skill and demand compactness data between the sample skill and a sample scene as training samples to be input into a pre-constructed graph convolution neural network model, and network parameters of the graph convolution neural network model and the neural network models are optimized according to a weighted sum of a model output result and a skill value and a deviation of sample compensation, so that a trained skill domination network and a trained scene perception network are obtained.

As can be seen from the structure diagram of the skill dominating network shown in fig. 4C, the skill dominating network includes a self-influence extraction layer, an interaction-influence extraction layer, and a dominating weight activation layer; the self-influence extraction layer is used for extracting self-influence characteristics related to the target skill in the demand compactness data; the interaction power extraction layer is used for extracting interaction power characteristics among skills in the requirement compactness data according to the skill association data; and the dominance weight activation layer is used for determining the dominance weight of the target skill according to the self influence characteristic and the mutual influence characteristic.

The self influence characteristic represents the influence of the target skill and the target scene on the self importance of the target skill; and the interaction force characteristics are used for representing the influence of the adjacent skills on the importance of the target skills. The dominant weight of the target skill is determined through the self influence force characteristic and the mutual influence force characteristic, the dominant weight of the target skill can be determined from three levels, namely the target skill, a target scene where the target skill is located and adjacent skills related to the target skill, the accuracy of a dominant weight determination result is improved, and the accuracy of the finally determined skill value is improved.

In an alternative embodiment, the self-influence extraction layer may include a multi-layer perceptron, so as to extract self-influence features related to the target skills in the demand compactness data based on the trained network parameters.

In another alternative embodiment, the interaction power extraction layer may include a plurality of layers of perceptors, so as to extract the foreign-involved influence characteristics related to the target skills in the demand compactness data based on the trained network parameters, and use the foreign-involved influence characteristics as the interaction power characteristics. The structure of the multilayer perceptron for extracting the foreign influence force characteristics can be the same as or different from that of the multilayer perceptron for extracting the self influence force characteristics.

In order to improve the comprehensiveness and accuracy of the extracted interaction force features, in yet another optional embodiment, the interaction force extraction layer may further include a graph convolution neural network for extracting local interaction force features between the target skill and each adjacent skill in the demand compactness data according to the extravehicular interaction force features and the skill association data; and taking the local influence characteristic as an interaction influence characteristic, or taking the local influence characteristic and the external influence characteristic as the interaction influence characteristic.

Optionally, determining the dominance weight of the target skill according to the self-influence feature and the mutual-influence feature may be: and determining the dominance weight of the target skill according to the local influence characteristics and the self influence characteristics. Illustratively, the local influence characteristics and the self influence characteristics can be processed by adopting an attention mechanism to obtain a dominance weight of the target skill.

In order to further improve the accuracy of the determined dominance weight, optionally, the dominance weight of the target skill is determined according to the self-influence feature and the mutual-influence feature, and may be: and determining the dominance weight of the target skill according to the local influence characteristics, the foreign-involved influence characteristics and the self influence characteristics. Illustratively, the local and self-influence features may be feature fused; and processing the fused features and the mean value features of the foreign-involved influence features by adopting an attention mechanism to obtain the dominance weight of the target skill.

When the dominating weight includes an upper dominating weight and a lower dominating weight, in a model training phase of the skill dominating network, two different sets of network parameters are trained for the upper dominating weight and the lower dominating weight, respectively, and accordingly, in a model using phase of the skill dominating network, the upper dominating weight and the lower dominating weight are determined by using the network parameters corresponding to the upper dominating weight and the network parameters corresponding to the lower dominating weight, respectively.

S405, weighting the skill value according to the dominance weight so as to update the skill value.

The method and the device have the advantages that the determination operation of the dominance weight is refined into the skill association data of each adjacent skill according to the target skill and the demand compactness data, the dominance weight of the target skill is determined, the determination mechanism of the dominance weight is perfected, the factors such as the target skill, the demand relation between the target skill and the target scene, the mutual influence of the adjacent skills on the target skill and the like are fully considered by introducing the skill association data and the demand compactness data, the influence on the dominance weight of the target skill is improved, the accuracy of the determined dominance weight is improved, guarantee is provided for value stripping of multiple skills in the same target scene, and a foundation is laid for improving the accuracy of the finally determined skill value.

Fig. 5A is a flowchart of another data processing method provided in an embodiment of the present application, and the method provides a preferred implementation manner based on the above technical solutions.

A data processing method as shown in fig. 5A, comprising: a data pre-processing stage 510, a model training stage 520, and a model using stage 530.

A data pre-processing stage 510 comprising:

s511, acquiring a plurality of pieces of recruitment data; the recruitment data includes skill data, job scene data, and surveillance data.

The skill data comprises skill names, skill co-occurrence frequency and skill proficiency level under the same working scene. The skill level can be proficient, skilled, known, etc.

The work scenario data comprises the self attribute, the historical statistical attribute and the skill requirement attribute of the skill demander, for example, the self attribute of the enterprise and public institution, for example, at least one of the listing situation, the type of the work, the establishment period and the like. The historical statistical attribute is historical monthly salary statistical data of a city where the skill demander is located, and the historical statistical attribute comprises at least one of mean, variance, maximum value and minimum value. Skill requirement attributes including at least one of work experience and graduation year, etc.

Supervisory data including a payroll upper limit and a payroll lower limit.

And S512, generating a skill graph according to the skill data.

Referring to the skills graph shown in FIG. 5B, skills are represented by nodes; connecting skills which appear in the same recruitment data at the same time through edges; using the skill proficiency of each skill as description data of the node corresponding to the skill; and taking the co-occurrence frequency of the skills connected through the edge on all the recruitment data as the description data of the edge.

S513, the skill demand attribute and the city code of the skill demand party in the work scene data are coded through one-hot coding to obtain the structured discrete scene vector.

For example, if a city in a certain piece of recruitment data is coded as XXX, requiring 3 years of work experience, the discrete scene vector formed may be [ XXX,1 ]. In this example, the recruitment of the colleague, the work experience is represented by "0"; 1-3 years, the working experience is represented by '1'; and 3-5 years later, the working experience is represented by 2.

And S514, respectively generating two structured continuous scene vectors according to the self attribute and the historical statistical attribute of the skill demander in the working scene data.

Illustratively, if a piece of recruitment data is as follows: the internet company B, a listing company with 13 years of success, located in city a recruits algorithm engineers, and the historical monthly salary statistics for that position are as follows: the mean value is 10000 yuan, the variance is 500 yuan, the highest monthly salary is 12000 yuan, and the lowest monthly salary is 8000 yuan, so that the determined continuous scene vector corresponding to the self attribute of the skill demand party can be [ A city, B company, listed company, 13 years, Internet class ]; the determined continuous scene vector corresponding to the historical statistical attributes may be [10000,500,12000,8000 ]. Of course, the vector elements in the continuous scene vector may also be determined by other encoding methods, and the encoding method may be determined by a skilled person according to needs or experience values.

And S515, generating a structured supervision vector from the supervision data.

If the range of compensation provided in a piece of recruitment data is 8000- > 10000, the generated supervision vector may be [8000,10000 ].

Referring to fig. 5C, a schematic structural diagram of a skill pricing combination model is shown, which is a two-part neural network including a skill pricing network for scene awareness and skill pricing, and a compensation prediction network for skill weight assignment and compensation determination.

A model training phase 520 comprising:

s521, generating a sample skill vector comprising the skill name and the skill proficiency of the sample skill according to the skill map;

s522, inputting the sample skill vector of the sample skill and the sample scene data of the sample scene into a scene perception network as training samples to obtain interactive data of the sample skill and the sample scene.

Specifically, the interaction data of the sample skill and the sample scene is determined according to the following formula:

x＝f(C,s,l|Θ)；

wherein x is interactive data, C is sample scene data including discrete scene vectors and continuous scene vectors, s is a skill name, l is a skill proficiency level, and Θ is a parameter to be trained. Wherein, f () is an information extraction function, and is realized by adopting a scene perception network.

And S523, inputting the data to a pricing network to predict the value interval of the sample skills according to the interactive data to obtain a sample prediction value interval.

Specifically, the sample prediction value interval is determined according to the following formula:

wherein, [ v ]^l,v^u]Predicting a value interval, phi, for the sample_lAnd phi_uFor the parameters to be trained, gl () and gu () are non-negative activation functions and are realized by adopting a pricing network.

Wherein the lower value limit v of the prediction^lUpper limit of sum value v^uThe following conditions are satisfied: v. of^l≥0，v^uIs not less than 0, and v^l≤v^u。

And S524, acquiring co-occurrence frequencies of the sample skill and the sample adjacent skill according to the skill diagram, and generating an adjacency matrix according to the acquired co-occurrence frequencies.

And S525, inputting the adjacency matrix and the interaction data into a skill domination network to obtain a skill weight interval of the sample skill.

Specifically, the skill weight interval of the sample skill is determined by the following formula:

wherein [ d ]^l,d^u]Is a skill weight interval, A is an adjacent matrix, X is partial or whole interactive data or a hidden layer vector when generating interactive data, psi_lAnd Ψ_uTwo sets of parameters to be trained. h () is a weight calculation function, implemented using a skill dominating network.

And S526, weighting the upper limit and the lower limit in the skill weight interval of each sample skill in the same recruitment data to obtain the final prediction value interval of each sample skill.

And S527, summing the final prediction value intervals of the skills of the samples in the data according to each piece of recruitment data to obtain a prediction compensation interval.

Specifically, the predicted compensation interval of each sample skill is determined according to the following formula:

wherein the content of the first and second substances,

a predicted compensation interval corresponding to a certain recruitment data,

a skill weight interval for the ith sample skill in the recruitment data;

and predicting a value interval for the sample of the ith sample skill in the recruitment data, wherein N is the skill quantity contained in the recruitment data.

S528, optimizing network parameters of each network in the skill pricing combination model according to the interval difference between the predicted compensation interval and the actual compensation interval corresponding to the supervision vector of the recruitment data.

Optimizing the network parameters of each network in the skill pricing combination model according to the following loss functions:

wherein L is_sThe value of the loss function of the model training, | J | is the number of pieces of recruitment data of the model training,

the predicted compensation interval corresponding to the recruitment data of the jth item,

lambda is the actual salary interval corresponding to the jth recruitment data_lAnd λ_uIs a hyper-parameter.

In the optimization of the network parameters of each network in the skill pricing combination model, an optimization method can be adopted to find the parameters which enable the function values of the loss functions to reach the local minimum value, and the optimization method includes, but is not limited to, an optimization method such as random gradient descent.

The model training process will be described in detail with reference to the schematic structure of the skill pricing network shown in fig. 5D and the schematic structure of the skill dominating network shown in fig. 5E.

Referring to fig. 5D, the skills pricing network includes a scene aware network and a pricing network. A context aware network, comprising: a skill embedding layer, a discrete scene embedding layer, a continuous scene embedding layer and a feature extraction layer.

Specifically, a skill embedding layer extracts embedding characteristics of skills at different time intervals, and in order to reduce the complexity of the model, the skill embedding is assumed to be combined by low-rank embedding and potential projection matrixes and written as follows:

wherein the content of the first and second substances,

in which a skill embedding vector for the T-th time period is stored, T being the number of time periods, N_sRepresenting the number of sample skills;

is a low rank embedding, W^vs∈R^dl×deRepresenting the potential projection matrix shared for all time periods, de represents the embedded dimensions, and dl is the number of hidden variables. The length of the time period is determined by a skilled person as desired or empirically.

Through the above dynamic modeling mode for different time periods, the ability of the skill pricing model for modeling and evaluating the skill value of the skill in different time periods is improved, but higher model complexity is brought, and model overfitting is easily caused.

In order to avoid model overfitting, a regularization term of a time dimension is introduced into a loss function in the training process of the skill pricing combination model, and the regularization term is specifically as follows:

wherein | | | purple hair_FRepresenting the F-norm, this regularization term constrains the characterization of the technique to not change drastically over time.

Specifically, the discrete scene embedding layer allocates one embedding to each discrete scene vector according to the following formula to obtain discrete scene embedding:

wherein the content of the first and second substances,

for the purpose of discrete scene embedding,

as a discrete scene vector, m_iIs the vector dimension of the discrete scene vector,

is a parameter to be trained; where i ∈ D, where D represents a set of discrete scene vectors.

Specifically, the continuous scene embedding layer allocates one embedding to each continuous scene vector according to the following formula to obtain continuous scene embedding:

wherein the content of the first and second substances,

for the purpose of the continuous scene embedding,

as a continuous scene vector, d_iIs the vector dimension of the continuous scene vector,

and

is a parameter to be trained; where i ∈ C, where C represents a set of consecutive scene vectors.

In order to extract as much information as possible, the skill pricing model extracts deep and shallow interactions between sample scenes and sample skills through a feature extraction layer, first processes discrete scene vectors and continuous scene vectors in different ways, and extracts interaction data of each layer through linear projection, multiplication and MLP. The interactive data includes first order interactive data, second order interactive data, and higher order interactive data.

Specifically, the first-order interaction data is determined by calculation according to the following formula:

wherein h is₁For first order interactive data, e^s∈R^deA skill embedding vector representing the skill of the sample is

The row vector corresponding to the sample skill in the table,

and

as a parameter to be trained, d_o1Is the output dimension.

Specifically, the second-order interaction data is determined by calculation according to the following formula:

wherein, "" indicates element-by-element multiplication, h₂Is second order interactive data.

Specifically, the high-order interaction data is determined by calculation according to the following formula:

wherein, K is the MLP depth,

as a parameter to be trained, x^(k)Represents the output of the k-th layer, σ () represents the activation function, and x |, represents the x connecting the two vectors, which we will eventually output^(K)As high-order interactive data h₃。

Will be oneOrder interaction data h₁Second order interaction data h₂And high-order interactive data h₃The combination of (1) results in interaction data x.

Inputting x into the pricing network, according to v^lAnd v^uThe correlation formula determines the sample prediction value interval, and is not described herein again.

Referring to fig. 5E, a skills governance network, comprising: the device comprises a self-influence extraction layer, an external-involved influence extraction layer, a local influence extraction layer and a domination weight activation layer.

Specifically, a self-influence extraction layer extracts the feature representation related to the importance of each skill by using a multi-layer perceptron to obtain a self-influence feature X_imp∈R^N×dp. Where N represents the number of skills involved and dp is a hyper-parameter, representing the extracted feature dimensions. The multilayer perceptron includes a plurality of sequentially connected linear processing layers and non-linear processing layers.

Specifically, the foreign influence extraction layer extracts the characteristic representation related to the importance of each skill by using another multi-layer perceptron to obtain the self influence characteristic X_inf∈R^N×di. Where dp is a hyper-parameter, representing the extracted feature dimensions. The multilayer perceptron includes a plurality of sequentially connected linear processing layers and non-linear processing layers. Determination of X_impAnd determining X_infThe structures of the multi-layer perceptron can be the same or different.

Specifically, the local influence extraction layer extracts the influence of each skill by the adjacent skill through a graph convolution neural network to obtain the local influence characteristic, and is specifically realized by adopting the following formula:

U＝GCN(X_inf,A)；

u is the local influence feature, A is the adjacency matrix of the sample skills, and GCN () is the function corresponding to the graph convolution neural network.

Specifically, the weight activation layer is dominated, an attention mechanism is introduced, and the skill weight interval of each sample skill is determined by specifically adopting the following formula:

wherein Q is the characteristic X of the influence of the foreign body_infThe mean value characteristic of (1) is used as the global representation of the foreign-involved influence characteristic; [ d^l,d^u]Skill weight interval for predicted sample skills, W_q ^l、W_k ^lAnd W_v ^lTo a lower limit skill weight d^lThe parameters to be trained of the corresponding model; w_q ^u、W_k ^uAnd W_v ^uIs an upper limit skill weight d^uTwo groups of parameters to be trained of the corresponding model are obtained by respectively training the parameters to be trained by adopting the same skill domination network; softmax () is the activation function.

Model usage stage 530, comprising:

s531, acquiring one skill in the recruitment data as a target skill;

s532, inputting the skill data of the target skill and the working scene data of the target scene associated with the target skill into a skill pricing network to obtain the value interval and the interaction data of the target skill.

And S533, determining the adjacency matrix of the target skill by taking other skills except the target skill in the recruitment data as the adjacent skills.

S534, inputting the adjacency matrix and the interaction data of the target skill into a compensation prediction network to obtain a compensation interval of the target skill.

And S535, adding the salary sections corresponding to the skills in the recruitment data to obtain the comprehensive salary corresponding to the recruitment data.

Fig. 6 is a block diagram of a data processing apparatus according to an embodiment of the present application, where the data processing apparatus 600 includes: a structured data acquisition module 601, a demand closeness data determination module 602, and a skill value determination module 603.

A structured data obtaining module 601, configured to obtain target skill data of a target skill represented in a structured manner and target scene data of a target scene; the target scene has a use requirement for the target skill;

a demand closeness data determination module 602, configured to determine demand closeness data between the target skill data and the target scene data;

a skill value determining module 603, configured to determine, according to the demand closeness data, a skill value of the target skill in the target scene.

According to the embodiment of the application, the target skill data of the target skill which is structurally represented and the target scene data of the target scene are obtained through a structured data obtaining module; the target scene has the use requirement on the target skills; determining demand closeness data between the target skill data and the target scene data through a demand closeness data determination module; and determining the skill value of the target skill in the target scene through a skill value determination module according to the demand compactness data. By adopting the technical scheme, the target skill data and the demand compactness data of the target scene data are introduced to serve as the reference basis for determining the skill value, so that the quantitative calculation of the skill value of the target skill is realized, the value quantification of the single skill is realized, and further, the data support is provided for the individual capability evaluation or professional planning of the user and the optimization of the salary structure of the skill demand policy.

Further, the demand compactness data determination module 602 includes:

and the demand compactness data determining unit is used for extracting the characteristics of the target skill data and the target scene data by adopting a trained scene perception network to obtain the demand compactness data.

Further, the scene-aware network comprises an embedding layer and a feature extraction layer;

the embedding layer is used for respectively carrying out dimensionality reduction processing on the target skill data and the target scene data to obtain a skill embedding vector and a scene embedding vector;

the feature extraction layer is configured to perform feature extraction on at least one of the target scene data, the skill embedding vector, and the scene embedding vector to obtain the required compactness data.

Further, the demand compactness data comprises at least one of first order demand compactness data, second order demand compactness data and high order demand compactness data;

correspondingly, the feature extraction layer comprises:

the first order demand compactness data determining unit is used for determining the first order demand compactness data according to the skill embedding vector and the target scene data;

the second-order demand compactness data determining unit is used for determining the second-order demand compactness data according to the skill embedding vector and the scene embedding vector;

and the high-order demand compactness data determining unit is used for determining the high-order demand compactness data according to the target scene data, the scene embedding vector and the skill embedding vector.

The device also comprises a model training module, a model obtaining module and a model obtaining module, wherein the model training module is used for carrying out model training on the scene perception network;

the model training module comprises:

and the sample skill embedding vector determining unit is used for respectively carrying out dimension reduction on each sample skill data in time intervals to obtain the sample skill embedding vector when carrying out dimension reduction on the sample skill data to obtain the sample skill embedding vector.

Further, the model training module includes:

the loss function building unit is used for building a loss function according to the distance between the sample skill embedding vectors in the adjacent time intervals;

and the model training unit is used for optimizing the network parameters of the scene perception network according to the loss function.

Further, the skill value determination module 603 includes:

and the skill value determining unit is used for determining the skill value of the target skill in the target scene according to the demand compactness data by adopting a pre-trained neural network model.

Further, the skill value comprises an upper value limit and a lower value limit;

the upper value limit is not less than the lower value limit.

Further, the apparatus further comprises:

the dominance weight determining module is used for determining the dominance weight of the target skill in each adjacent skill of the target skill after determining the skill value of the target skill in a target scene according to the demand closeness data; the dominance weight representing the importance of the target skill in each of the neighboring skills;

and the skill value updating module is used for weighting the skill value according to the dominance weight so as to update the skill value.

Further, a dominance weight determination module comprising:

and a dominance weight determination unit for determining a dominance weight of the target skill according to the skill association data of the target skill and each adjacent skill of the target skill and the requirement closeness data.

Further, the dominance weight determination unit includes:

and the dominance weight determining subunit is used for determining the dominance weight of the target skill according to the skill association data of the target skill and each adjacent skill of the target skill and the requirement compactness data by adopting the trained skill dominance network.

Further, the skill dominance network comprises a self influence extraction layer, an interaction influence extraction layer and a dominance weight activation layer;

the self influence extraction layer is used for extracting self influence characteristics related to the target skill in the demand compactness data according to the skill association data;

the interaction power extraction layer is used for extracting interaction power characteristics among skills in the demand compactness data;

and the dominance weight activation layer is used for determining the dominance weight of the target skill according to the self influence characteristic and the mutual influence characteristic.

Further, the interacting force extracting layer includes:

the external influence characteristic extraction unit is used for extracting external influence characteristics related to the target skill in the demand compactness data;

a local influence feature extraction unit, configured to extract, according to the foreign-involved influence feature and the skill association data, a local influence feature between the target skill and each of the neighboring skills in the demand closeness data;

and the mutual influence force characteristic extraction unit is used for taking the local influence force characteristic and/or the foreign-involved influence force characteristic as the mutual influence force characteristic.

Further, the dominating weight activation layer includes:

a feature fusion unit for performing feature fusion on the local influence feature and the self-influence feature;

and the dominance weight determining unit is used for processing the fused features and the mean value features of the extravehicular influence force features by adopting an attention mechanism to obtain the dominance weight of the target skill.

Further, the skill association data is a matrix constructed from co-occurrence frequencies of the target skill and the neighboring skills.

Further, if the skill value comprises an upper value limit and a lower value limit, the dominance weight comprises an upper dominance weight and a lower dominance weight;

the upper limit dominance weight and the lower limit dominance weight are the same or different.

Further, the target skills include algorithm skills and/or programming language skills; the target scene comprises an enterprise and public institution; the skill value is compensation.

The data processing device is used for executing any data processing method provided by the embodiment of the application, and has the corresponding functional modules and the beneficial effects of the data processing method.

According to an embodiment of the present application, an electronic device and a readable storage medium are also provided.

Fig. 7 is a block diagram of an electronic device implementing the data processing method according to the embodiment of the present application. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present application that are described and/or claimed herein.

As shown in fig. 7, the electronic apparatus includes: one or more processors 701, a memory 702, and interfaces for connecting the various components, including a high-speed interface and a low-speed interface. The various components are interconnected using different buses and may be mounted on a common motherboard or in other manners as desired. The processor may process instructions for execution within the electronic device, including instructions stored in or on the memory to display graphical information of a GUI on an external input/output apparatus (such as a display device coupled to the interface). In other embodiments, multiple processors and/or multiple buses may be used, along with multiple memories and multiple memories, as desired. Also, multiple electronic devices may be connected, with each device providing portions of the necessary operations (e.g., as a server array, a group of blade servers, or a multi-processor system). In fig. 7, one processor 701 is taken as an example.

The memory 702 is a non-transitory computer readable storage medium as provided herein. The memory stores instructions executable by at least one processor to cause the at least one processor to perform the data processing methods provided herein. The non-transitory computer-readable storage medium of the present application stores computer instructions for causing a computer to execute the data processing method provided by the present application.

The memory 702, which is a non-transitory computer-readable storage medium, may be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as program instructions/modules corresponding to the data processing method in the embodiments of the present application (for example, the structured data acquisition module 601, the demand compactness data determination module 602, and the skill value determination module 603 shown in fig. 6). The processor 701 executes various functional applications of the server and data processing by executing non-transitory software programs, instructions, and modules stored in the memory 702, that is, implements the data processing method in the above-described method embodiment.

The memory 702 may include a program storage area that may store an operating system, an application program required for at least one function, and a data storage area; the storage data area may store data created by use of an electronic device implementing the data processing method, and the like. Further, the memory 702 may include high speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory 702 may optionally include memory located remotely from the processor 701, which may be connected through a network to an electronic device implementing the data processing method. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The electronic device implementing the data processing method may further include: an input device 703 and an output device 704. The processor 701, the memory 702, the input device 703 and the output device 704 may be connected by a bus or other means, and fig. 7 illustrates an example of a connection by a bus.

The input device 703 may receive input numeric or character information and generate key signal inputs related to user settings and function control of an electronic apparatus implementing the data processing method, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, a pointing stick, one or more mouse buttons, a track ball, a joystick, or other input devices. The output devices 704 may include a display device, auxiliary lighting devices (e.g., LEDs), and tactile feedback devices (e.g., vibrating motors), among others. The display device may include, but is not limited to, a Liquid Crystal Display (LCD), a Light Emitting Diode (LED) display, and a plasma display. In some implementations, the display device can be a touch screen.

Various implementations of the systems and techniques described here can be realized in digital electronic circuitry, integrated circuitry, application specific ASICs (application specific integrated circuits), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.

These computer programs (also known as programs, software applications, or code) include machine instructions for a programmable processor, and may be implemented using high-level procedural and/or object-oriented programming languages, and/or assembly/machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus, and/or device (e.g., magnetic discs, optical disks, memory, Programmable Logic Devices (PLDs)) used to provide machine instructions and/or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term "machine-readable signal" refers to any signal used to provide machine instructions and/or data to a programmable processor.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic, speech, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.

The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

According to the technical scheme of the embodiment of the application, the target skill data of the target skill which is structurally represented and the target scene data of the target scene are obtained; the target scene has the use requirement on the target skills; determining demand closeness data between the target skill data and the target scene data; and determining the skill value of the target skill in the target scene according to the demand compactness data. By adopting the technical scheme, the target skill data and the demand compactness data of the target scene data are introduced to serve as the reference basis for determining the skill value, so that the quantitative calculation of the skill value of the target skill is realized, the value quantification of the single skill is realized, and further, the data support is provided for the individual capability evaluation or professional planning of the user and the optimization of the salary structure of the skill demand policy.

It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present application may be executed in parallel, sequentially, or in different orders, and the present invention is not limited thereto as long as the desired results of the technical solutions disclosed in the present application can be achieved.

The above-described embodiments should not be construed as limiting the scope of the present application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A data processing method for determining skill value, the method comprising:

2. The method of claim 1, wherein said determining demand closeness data between said target skill data and said target scenario data comprises:

and performing feature extraction on the target skill data and the target scene data by adopting a trained scene perception network to obtain the required compactness data.

3. The method of claim 2, wherein the scene-aware network comprises an embedding layer and a feature extraction layer;

the embedding layer is used for respectively carrying out dimensionality reduction on the target skill data and the target scene data to obtain a skill embedding vector and a scene embedding vector;

the feature extraction layer is used for performing feature extraction on at least one of the target scene data, the skill embedding vector and the scene embedding vector to obtain the required compactness data.

4. The method of claim 3, wherein the demand compactness data comprises at least one of first order demand compactness data, second order demand compactness data, and higher order demand compactness data;

correspondingly, the performing feature extraction on at least one of the target scene data, the skill embedding vector and the scene embedding vector to obtain the demand compactness data includes:

determining the first-order demand closeness data according to the skill embedding vector and the target scene data;

determining the second-order demand compactness data according to the skill embedding vector and the scene embedding vector; and/or

And determining the high-order demand compactness data according to the target scene data, the scene embedding vector and the skill embedding vector.

5. The method according to claim 3, wherein in the model training phase of the scene-aware network, in performing dimension reduction processing on the sample skill data to obtain the sample skill embedding vector, the method comprises:

and respectively carrying out dimensionality reduction on each sample skill data in time intervals to obtain the sample skill embedding vector.

6. The method of claim 5, further comprising:

constructing a loss function according to the distance between the sample skill embedding vectors in the adjacent time period;

and optimizing the network parameters of the scene-aware network according to the loss function.

7. The method of claim 1, wherein said determining a skill value of said target skill in said target scenario from said demand closeness data comprises:

and determining the skill value of the target skill in the target scene according to the demand compactness data by adopting a pre-trained neural network model.

8. The method of claim 1, wherein the skill value comprises an upper value limit and a lower value limit;

wherein the upper value limit is not less than the lower value limit.

9. The method according to any of claims 1-8, wherein after determining a skill value of a target skill in a target scenario from the demand closeness data, the method further comprises:

determining a dominance weight of the target skill among adjacent skills of the target skill; the dominance weight representing the importance of the target skill in each of the neighboring skills;

weighting the skill value according to the dominance weight to update the skill value.

10. The method of claim 9, wherein the determining a dominant weight of the target skill among the neighboring skills of the target skill comprises:

and determining the dominance weight of the target skill according to the skill association data of the target skill and each adjacent skill of the target skill and the demand closeness data.

11. The method of claim 10, wherein said determining a dominance weight for the target skill based on skill association data for the target skill and each adjacent skill of the target skill and the demand closeness data comprises:

and determining the domination weight of the target skill by adopting a trained skill domination network according to the skill association data of the target skill and each adjacent skill of the target skill and the requirement compactness data.

12. The method of claim 11, wherein the skills dominance network comprises a self-influence extraction layer, an interaction-influence extraction layer, and a dominance weight activation layer;

13. The method of claim 12, wherein said extracting, from said skill association data, interaction force features between skills in said demand closeness data comprises:

extracting the foreign-involved influence characteristics related to the target skill in the demand compactness data;

extracting local influence features between the target skill and each of the adjacent skills in the demand closeness data according to the foreign-involved influence features and the skill association data;

and taking the local influence characteristic and/or the foreign-involved influence characteristic as the mutual influence characteristic.

14. The method of claim 13, wherein determining a dominance weight for the target skill from the self-influence feature and the interaction-influence feature comprises:

performing feature fusion on the local influence feature and the self influence feature;

and processing the fused features and the mean value features of the foreign-involved influence features by adopting an attention mechanism to obtain the dominance weight of the target skill.

15. A method according to claim 10 wherein the skill association data is a matrix constructed from co-occurrence frequencies of the target skill and the neighbouring skills.

16. The method of claim 9, wherein if the skill value comprises an upper value limit and a lower value limit, the dominance weight comprises an upper dominance weight and a lower dominance weight;

17. The method according to any one of claims 1-8, wherein the target skills include algorithmic skills and/or programming language skills; the target scene comprises an enterprise and public institution; the skill value is compensation.

18. A data processing apparatus for determining a skill value, the apparatus comprising:

19. An electronic device, comprising:

at least one processor; and

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a data processing method as claimed in any one of claims 1 to 17.

20. A non-transitory computer readable storage medium storing computer instructions for causing a computer to perform a data processing method according to any one of claims 1 to 17.