CN114529038A

CN114529038A - Intelligent matching business recruitment strategy system and method based on enterprise demands

Info

Publication number: CN114529038A
Application number: CN202111678574.0A
Authority: CN
Inventors: 毛蔚赢; 章岩; 孙志奎; 赵立杰; 刘敏
Original assignee: Shanghai Afa Analydi Data Technology Co ltd
Current assignee: Shanghai Afa Analydi Data Technology Co ltd
Priority date: 2021-12-31
Filing date: 2021-12-31
Publication date: 2022-05-24

Abstract

The invention discloses an intelligent matching business recruitment strategy system based on enterprise requirements, which comprises: the enterprise demand evaluation module acquires enterprise demand matching data through LM-BP neural network deep learning; the business recruitment strategy module is used for establishing an enterprise demand scoring model based on random forest improvement and classifying and predicting enterprise demand characteristics; the industry matching module selects the industry suitable for the enterprise through an industry matching library; the policy matching module selects a policy suitable for an enterprise through the policy matching library; the regional space matching module is used for selecting regional spaces suitable for enterprises by matching the regional space library; the building matching module selects buildings suitable for enterprises through a matching planning building library; a method for an enterprise intelligent matching recruiter policy system is also disclosed. According to the method, the BP neural network and the random forest algorithm are optimized, improved and the demand data are obtained to build an enterprise demand scoring model, the enterprise demand evaluation is calculated more effectively, and accurate recruitment is ensured through comprehensive evaluation of the enterprise.

Description

Intelligent matching business recruitment strategy system and method based on enterprise demands

Technical Field

The invention relates to the technical field of intelligent business recruitment, in particular to an intelligent matching business recruitment strategy system and method based on enterprise requirements.

Background

The accurate recruiter is the inevitable choice for improving the regional economic core competitiveness and adapting to the regularity of the recruiter activity. How to improve the recruitment work of our recruiters, implement accurate recruiters and optimize the industrial structure is the key point of research at present. The most important thing is the accurate business recruitment problem, how to evaluate enterprise requirements to obtain industry, policy, location and building structural data matched with the enterprise requirements, establish an enterprise requirement scoring model through the data, and classify and predict the characteristic conditions of the enterprise requirements, which are the most concerned matters.

In the real society, there are still many influences due to factors such as performance and information asymmetry, which lead to blind recruitment in various regions, and the fitness of the precision of the recruitment and the industry is not fully considered, so that the recruitment enterprise can not drive the development of the local economy, even needs the support of the government, and the development of the local economy is hindered. The core content of accurate recruiting is to improve the pertinence of the recruiting, avoid the randomness of the recruiting activities, reduce the risk of the recruiting and enable the recruited enterprises to better meet the requirements of local economic development.

An application No.: 201811058230.8, discloses an enterprise site selection system based on investment cooperation carrier environment evaluation index, comprising site selection information collection unit: for guiding the user to enter the desired address information, the GIS unit: the carrier query unit is used for generating a map of a recommended enterprise address selection area: the system is used for providing other carrier information inquiry services except for a recommended addressing scheme, and comprises an addressing unit: the method is used for analyzing the enterprise site selection area, and the industrial cluster analysis and evaluation database comprises the following steps: the system is used for storing the industrial cluster analysis and evaluation index, and comprises an address selection element database: the system is used for storing site selection element scores, and an enterprise basic information database comprises: the system is used for storing the industry door types, the enterprise sales amounts and the enterprise tax amounts of the enterprises according to the specified divided areas. The remarkable effects are as follows: six investment modes of investment, skill, talent investment, intelligence investment, platform investment and joint investment can be divided through the carrier environment index calculation unit, the cluster evaluation module and the carrier resource evaluation module are respectively designed, if the investment mode is investment or skill investment, the industry cluster evaluation index is calculated through the cluster evaluation module, if the investment mode is talent investment, skill investment, platform investment or cooperation investment, the operator environment analysis index is calculated through the carrier resource evaluation module, the industry cluster evaluation index or the operator environment analysis index is sorted, 3 carrier environments with high scores are selected to recommend each enterprise, and the site selection scheme of the enterprise is more scientific and accurate.

However, the above application No.: 201811058230.8, the open source algorithm is mainly applied, the logic of the algorithm is relatively single, and when the standard decision is solved, the linear derivation is not easy to solve the positive ideal solution and the negative ideal solution, so that the enterprise matching data can not be rapidly and accurately obtained, and the efficient establishment of the business recruitment strategy scoring model is difficult.

Disclosure of Invention

Aiming at the problems in the prior art, the invention provides an intelligent matching business recruitment strategy system and method based on enterprise requirements, after the intelligent matching business recruitment strategy system and method are used, in the aspect of obtaining accurate matching data of enterprise requirements, an LM-BP neural network algorithm is optimized and improved based on a BP neural network, and a local area industrial value characteristic analysis, a matching area policy characteristic analysis, a matching area location advantage characteristic analysis and a matching area building adaptation characteristic analysis are matched through an enterprise requirement image, a local area industrial value, policy, location advantage, building resources, policy advantage, location advantage and building advantage model and index are obtained through an improved random forest algorithm to build a more efficient matching enterprise requirement scoring model, and enterprise requirement evaluation can be calculated more effectively by utilizing rich algorithm logic, and ensuring accurate enrollment through comprehensive evaluation of the enrollment enterprises.

In order to solve the technical problems, the invention adopts the technical scheme that:

an intelligent matching recruiter strategy system based on enterprise needs is provided, which comprises: the system comprises an enterprise demand evaluation module, a business recruitment strategy module, an industry matching module, a policy matching module, an area space matching module and a building matching module;

the enterprise demand evaluation module is used for obtaining industrial, policy, location and building structured data matched with enterprise demands after deep learning of regional industrial value characteristic analysis, policy characteristic analysis, location characteristic analysis and building characteristic analysis through the LM-BP neural network;

the business recruitment strategy module is used for scoring the enterprise demand figures and the matched local industry, policy, location and building resources, constructing an enterprise demand business recruitment strategy report scoring model, establishing an enterprise demand scoring model based on random forest improvement and classifying and predicting the enterprise demand characteristic conditions;

the industry matching module is used for analyzing the industrial scale of local business recruitment areas, the total number of enterprises related to an industrial chain, the number of the enterprises on the scale, the conditions of customers and suppliers in respective subdivided industries of the enterprises through an industry matching library, and providing an enterprise industry aggregation effect and a supply and marketing relationship effect reference value, so that the optimal industry suitable for the enterprises is matched and selected;

the policy matching module is used for analyzing the policies adapted to the enterprises through the policy matching library, and making business recruitment service personnel follow the specific requirements of the enterprises and recommend marks in the policy matching module through the universal policies adapted to the entrepreneurial period, the growth period and the maturity period of the development stage of the enterprises, the industrial policies adapted to all the subdivided industries of the enterprises and the policy recommendation for cultivating the enterprise foreground, so that the optimal policies suitable for the enterprises are matched and selected;

the regional space matching module is used for matching enterprise demand quantitative evaluation scores after data formatting of the matching regional space library and performing adaptive evaluation by matching macroscopic location, mesoscopic location, microscopic location, land use planning, traffic logistics and life support through enterprise demands, so that the optimal regional space suitable for enterprises is matched and selected;

the building matching module is used for selecting an optimal building carrier suitable for an enterprise through a matching planning building library, and performing adaptability evaluation on the building carrier through building basic information, a building structure, energy conservation, environmental protection, fire prevention, explosion prevention, supporting equipment and survival cost, so that the optimal building planning suitable for the enterprise is matched and selected.

In order to solve the technical problem, the invention adopts the further technical scheme that:

further, in the enterprise requirement evaluation module, the enterprise requirement matched industry structured data includes: industrial scale, value chain customer market, industrial chain complement and enterprise complement; the enterprise requirement matched policy structured data comprises: industry policies, talent policies, and financial policies; the enterprise demand matched zone bit structured data comprises: traffic logistics, supporting resources and planning elements; the enterprise demand matched building structured data comprises: building foundation elements, energy conservation, environmental protection, load, fire and explosion prevention and use cost.

Further, the specific calculation steps of the LM-BP neural network in step 1 are as follows:

step 1.1: initializing network structure parameters, wherein an error allowable value is epsilon, constants u and b, initializing a weight and a threshold vector, enabling k to be 0, u to be u0, and calculating the precision to be epsilon and the maximum learning time M;

step 1.2: inputting training data of an enterprise demand portrait index matrix into an LM-BP neural network as an input vector;

step 1.3: calculating a network output and error index function e;

step 1.4: calculating a Jacobian matrix J [ W (k) ]; wherein, W (k) represents a vector formed by the threshold value and the weight value of the kth neural network iteration;

step 1.5: calculating delta W; wherein Δ W is a threshold change amount;

step 1.6: if e < epsilon, go to step 1.8, otherwise go to step 1.5;

step 1.7: the error function e is calculated with the new weight and the threshold vector W (k +1),

W(k+1)＝W(k)-{J^T[W(k)]J[W(k)]}^-1J[W(k)]e[W(k)]

if e [ W (k +1) ] is smaller than e [ W (k) ], making k ═ k +1 and u ═ b, go to step 1.2, otherwise, go to step 1.5; wherein, W (k) represents a vector formed by the threshold value and the weight value of the kth neural network iteration, and W (k +1) represents a vector formed by the threshold value and the weight value of the new kth iteration + 1;

step 1.8: and the LM-BP neural network calculation is finished.

Further, in step 1.1, the value range of b is: 0< b <1, and when k is 0 and u is u0, the accuracy e and the maximum learning number M are calculated.

Further, in step 1.4 and step 1.7, the Jacobian matrix J [ W (k)]Is deformed to calculate J^T(W), calculate J^TThe formula of (W) is:

further, the step 2 of establishing an enterprise demand scoring model based on random forest improvement comprises the following steps:

step 2.1: preprocessing enterprise demand data;

step 2.2: calculating an enterprise demand matrix;

step 2.3: data weighted sampling;

step 2.4: selecting an optimal requirement characteristic subset of enterprise requirements by a characteristic selection method;

step 2.5: optimizing algorithm parameters;

step 2.6: an evaluation result is generated.

Further, in step 2.1, the enterprise demand data preprocessing comprises a Min-Max standardization process and a Z-Score standardization process;

the Min-Max standardization processing is that offline data in enterprise demand data are subjected to linear transformation, so that the data of the enterprise demand data after the linear transformation are between [0 and 1 ];

the Z-Score normalization process is to convert the business demand data to a Gaussian distribution with a mean of 0 and a standard deviation of 1.

Further, in step 2.2, the specific steps of calculating the enterprise demand matrix are as follows: assuming that X ═ b1, b2, … …, bL } represents a set of L samples of M features, and Y ═ Y1, Y2, … …, yL } represents a set of categories, the enterprise demand data may be constructed as a matrix:

where the size of the matrix L is L (M +1), +1 denotes a set of classes, bi ═ { Xi1, Xi2, … …, XiM } represents the M eigenvalues of the bi table samples, Xij represents the j-th eigenvalue of the sample bi;

in the enterprise demand matrix L, a few samples L ' and a plurality of samples L ' are included, Q samples in the few samples L ' are taken, and the matrix form is as follows:

if there are Q samples in the L 'few samples and L total samples, then there are (L-Q) samples in the L' most samples, then the matrix form is:

further, in step 2.3, the data weighted sampling specifically includes:

step 2.3.1: dividing original enterprise demand data into a training set L and a training set L1;

step 2.3.2: dividing a training set L into two subsets which are a majority sample L 'and a minority sample L' respectively;

step 2.3.3: in the sampling process, firstly, weighted sampling is carried out on most samples L ', samples with similar sizes of a few samples L' are picked out from the most samples L ', the proportion of the picked few samples L' to the proportion L 'is calculated, the proportion of the L' to all the samples L is calculated, and then weighting is carried out on the weight to select the final training sample;

step 2.3.4: repeating the step 2.3.3 for a plurality of times until a balance sample is selected;

step 2.3.5: and picking out balance samples for division into a training set and a testing set.

Further, in step 2.4, the input is the original enterprise demand data set D { (x1, y1), (x2, y2), … …, (xn, yn) }, x_i∈R^mAnd y is_nE { -1, 1 }; setting g1, g 2;

outputting an optimal feature subset f;

the specific selection step of selecting the optimal requirement characteristic subset of the enterprise requirement by the characteristic selection method comprises the following steps:

step 2.4.1: setting M enterprise demand characteristics i to be 1,2,3,4, … and M;

step 2.4.2: calculating a corresponding value of each enterprise demand characteristic by using the following formula;

setting D as a sample data set, x and y as arbitrary attributes of the samples, and n as the number of categories in the data set D, the information entropy of x is:

wherein P (xi) is the probability that the value of the characteristic attribute x is xi;

the conditional entropy of the feature attribute x given by the feature attribute y is:

wherein p (yi) is the probability that the value of the characteristic attribute y is yj, and p (xi | yi) is the probability that the value of the characteristic attribute x is xi under the condition that the value of the characteristic attribute y is yj;

the information entropy obtained by the above formula is:

Gain(x、y)＝Info(x)-Info(x|y)

selecting the feature with the largest information gain as the splitting attribute of the data set D, creating a node, using the feature as a mark, creating a branch for each value of the feature, and dividing the enterprise requirements of the samples according to the branch;

step 2.4.3: respectively calculating an entropy comparison value un of each feature and the category variable yn by using the following formula;

step 2.4.4: if un is greater than or equal to g1, the feature xn is in the selected optimal feature subset f, i.e. x_n∈f；

Sorting the features, measuring the selected features in the set f, and determining a correlation value S between the features xi and xj;

step 2.4.5: when S is less than or equal to g2, deleting the characteristics in the set f according to the information entropy comparison value un in the step (3);

step 2.4.6: and obtaining an optimal feature subset.

Further, in step 2.5, the specific optimization step of the algorithm parameter optimization includes:

step 2.5.1: setting a parameter searching range and step length to be optimized;

step 2.5.2: further calculating the average absolute error value of the two parameters S and C according to the step 2.5.1, and obtaining the number specific range of the two parameters S and C by using the average absolute error value;

step 2.5.3: calculating random forest OOB values by combining S and C according to the value range of the parameter S, C obtained in the step 2.5.2 and utilizing the following process to obtain accuracy;

when the sampling training is carried out on the sample each time, the sample data which is not sampled is marked as a set OOBi, the number of the classification errors of the OOBi in the data set which is not sampled is marked as ErrorNumOOB, and finally the error of the random forest OOB value is defined as:

namely, the generalization error is:

step 2.5.4: and selecting the optimal parameters determined by the S-C combination according to the OOB values, outputting the S-C combination if the random forest OOB values meet the requirements, and otherwise, changing the search range and the step length and continuing searching until the final conditions are met.

Further, in step 2.6, the enterprise demand scoring model based on random forest improvement generates an optimal evaluation result through steps 2.1 to 2.5, and the optimal evaluation result is used as an evaluation reference and provided to an industrial park recruitment worker to judge whether the enterprise is adapted to the park recruitment strategy.

Also provided is a method for an enterprise intelligent matching recruiter policy system, comprising the steps of:

s1: the enterprise demand evaluation module acquires industrial, policy, location and building structured data matched with enterprise demands after deep learning of regional industrial value characteristic analysis, policy characteristic analysis, location characteristic analysis and building characteristic analysis through the LM-BP neural network;

s2: the business recruitment strategy module scores the enterprise demand figures and matched local industries, policies, location and building resources, constructs an enterprise demand business recruitment strategy report scoring model, establishes an enterprise demand scoring model based on random forest improvement, and classifies and predicts the enterprise demand characteristic conditions;

s3: the industry matching module analyzes the industrial scale of local business recruitment areas, the total number of enterprises related to an industrial chain, the number of the enterprises on the scale and the conditions of customers and suppliers in respective subdivided industries of the enterprises through an industry matching library and provides reference values of enterprise industry aggregation effect and supply and marketing relation effect, so that the optimal industry suitable for the enterprises is matched and selected;

s4: the policy matching module analyzes the policies adapted to the enterprises through the policy matching library, and leads business recruiters to follow the specific requirements of the enterprises and recommend marks in the policy matching module through the universal policies adapted to the entrepreneurial period, the growth period and the maturity period of the development stage of the enterprises, all industrial policies adapted to the subdivided industries of the enterprises and the policy recommendation for cultivating the prospects of the enterprises, so as to match and select the optimal policies suitable for the enterprises;

s5: the regional space matching module matches enterprise demand quantitative evaluation scores after data formatting of the matching regional space library and performs adaptive evaluation by matching macroscopic location, mesoscopic location, microscopic location, land planning, traffic logistics and life matching according to enterprise demands, so that an optimal regional space suitable for an enterprise is matched and selected;

s6: the building matching module selects an optimal building carrier suitable for the enterprise through a matching planning building library, and performs adaptive evaluation on the building carrier through building basic information, building structures, energy conservation, environmental protection, fire prevention, explosion prevention, supporting equipment and survival cost, so that the optimal building planning suitable for the enterprise is matched and selected.

The invention has the beneficial effects that:

the intelligent matching business recruitment strategy system and method based on enterprise requirements optimize and improve an LM-BP neural network algorithm based on a BP neural network in the aspect of obtaining accurate matching data of enterprise requirements, match regional industrial value characteristic analysis, match regional policy characteristic analysis, match regional location advantage characteristic analysis and match regional building adaptation characteristic analysis through enterprise requirement images, further obtain various models and indexes of local regional industrial value, policy, location advantage, building resources, policy advantage, location advantage and building advantage through an improved random forest algorithm to build a more efficient matching enterprise requirement scoring model, can more effectively calculate enterprise requirement evaluation by utilizing rich algorithm logic, and ensure accurate business recruitment through comprehensive evaluation of business recruitment;

secondly, the LM-BP neural network algorithm is improved based on the BP neural network for optimization, and the Levenberg-Marquardt (LM for short) algorithm has the advantages of a gradient method and a Newton method at the same time, in order to reduce the singular problem of non-optimal points, when the objective function is close to the optimal point, the characteristic of the second derivative near the extreme point is used for approximating the quadratic property, by the LM-BP neural network improved to the BP neural network algorithm, the optimization convergence process is accelerated, the speed is much higher than that of a gradient method and the BP algorithm, the efficiency of analyzing and processing big data is optimized, after deep learning of regional industrial value characteristic analysis, policy characteristic analysis, location characteristic analysis and building characteristic analysis, the method comprises the steps of obtaining industry, policy, location and building structured data accurately matched with enterprise requirements, and being capable of assisting enterprises to deeply learn and mine enterprise requirements and obtaining adaptive local resources according to the requirements;

thirdly, the main functions of the invention are based on enterprise demand portrait and matched local industry, policy, location and building resources to grade, an enterprise demand business recruitment strategy report grading model is constructed by machine learning and statistics, enterprise demand characteristics are followed, relationships among different characteristics are excavated, a grading model based on random forest improvement is established, the condition of the enterprise demand characteristics is classified and predicted, enterprises with high matching degree of weight indexes are accurate adaptive business recruiters, and the improved random forest algorithm is applied to the construction of the enterprise demand grading model and mainly comprises enterprise demand data preprocessing, an enterprise demand matrix, data weighted sampling and sampling, The method comprises the steps of selecting an optimal enterprise demand characteristic subset by a characteristic selection method, optimizing algorithm parameters, generating an evaluation result, and providing an effective evaluation reference basis for an industrial park recruiter to judge whether an enterprise is matched with a park recruiter strategy or not based on a random forest improved enterprise demand scoring model.

The foregoing description is only an overview of the technical solutions of the present invention, and in order to make the technical solutions of the present invention more clearly understood and to implement them in accordance with the contents of the description, the following detailed description is given with reference to the preferred embodiments of the present invention and the accompanying drawings.

Drawings

FIG. 1 is a functional architecture diagram of an intelligent matching business-inviting strategy system based on enterprise requirements according to the present invention;

FIG. 2 is a block diagram of industrial structured data in the enterprise requirement assessment module according to the present invention (taking first-class and second-class indexes as examples);

FIG. 3 is a schematic diagram of the specific principle and algorithm flow of the BP neural network according to the present invention;

FIG. 4 is a schematic flow chart of an LM-BP neural network algorithm after being improved based on a BP neural network according to the present invention;

FIG. 5 is a schematic diagram of the specific body principle and algorithm flow of the random forest algorithm of the present invention;

FIG. 6 is a schematic flow chart of the method for establishing an enterprise demand scoring model based on random forest improvement according to the present invention;

FIG. 7 is a schematic flow chart of weighted sampling of data according to the present invention;

FIG. 8 is a schematic flow chart of selecting an optimal requirement feature subset of enterprise requirements by the feature selection method according to the present invention;

FIG. 9 is a schematic flow chart of parameter optimization of the random forest algorithm according to the present invention;

FIG. 10 is a flow diagram of a method for an enterprise intelligent matching recruiter policy system according to the present invention;

FIG. 11 is a second schematic flow chart of the present invention for establishing an enterprise demand scoring model based on random forest improvement;

the parts in the drawings are marked as follows:

the system comprises an enterprise demand evaluation module 1, a business recruitment strategy module 2, an industry matching module 3, an industry matching library 31, a policy matching module 4, a policy matching library 41, a regional space matching module 5, a regional space library 51, a building matching module 6 and a planning building library 61.

Detailed Description

The following description of the embodiments of the present invention is provided for illustrative purposes, and the present invention will be described in detail with reference to the accompanying drawings. The invention may be embodied in other different forms, i.e. it is capable of various modifications and changes without departing from the scope of the invention as disclosed.

Example 1:

an intelligent matching business recruitment strategy system based on enterprise needs, as shown in fig. 1, comprises: the system comprises an enterprise demand evaluation module 1, a business recruitment strategy module 2, an industry matching module 3, a policy matching module 4, an area space matching module 5 and a building matching module 6;

the enterprise demand evaluation module 1 is used for obtaining industrial, policy, location and building structured data matched with enterprise demands after deep learning of regional industrial value characteristic analysis, policy characteristic analysis, location characteristic analysis and building characteristic analysis through the LM-BP neural network;

the business recruitment strategy module 2 is used for scoring the enterprise demand figures and the matched local industries, policies, regions and building resources, constructing an enterprise demand business recruitment strategy report scoring model, establishing an enterprise demand scoring model based on random forest improvement, and classifying and predicting the enterprise demand characteristic conditions;

the industry matching module 3 is used for analyzing the industry scale of local business recruitment areas, the total number of enterprises related to an industry chain, the quantity of regular enterprises, the conditions of customers and suppliers in respective subdivided industries of the enterprises through the industry matching library 31 and providing reference values of enterprise industry clustering effect and supply and marketing relationship effect, so that the optimal industry suitable for the enterprises is matched and selected;

the policy matching module 4 is used for analyzing the policies adapted to the enterprises through the policy matching library 41, and making business recruiters follow the specific requirements of the enterprises and recommend marks in the policy matching module 2 through the universal policies adapted to the entrepreneurial period, the growth period and the maturity period of the enterprise in the development stage, all industrial policies adapted to the subdivided industries of the enterprise and the policy recommendation for cultivating the enterprise foreground, so as to match and select the optimal policies suitable for the enterprises;

the regional space matching module 5 is used for matching enterprise demand quantitative evaluation scores after data formatting of the matching regional space library 51 and performing adaptive evaluation by matching macroscopic location, mesoscopic location, microscopic location, land planning, traffic logistics and life support through enterprise demand, so that the optimal regional space suitable for enterprises is matched and selected;

and the building matching module 6 is used for selecting an optimal building carrier suitable for the enterprise through the matching planning building library 61, and performing adaptability evaluation on the building carrier through building basic information, building structures, energy conservation, environmental protection, fire prevention, explosion prevention, supporting equipment and survival cost, so that the optimal building planning suitable for the enterprise is matched and selected.

The enterprise demand evaluation module is mainly used for accurately matching data based on enterprise demands, the enterprise images the demands of local industries, policies, zone bits and buildings, and the enterprise acquires the structured data of the industries, the policies, the zone bits and the buildings, which are accurately matched with the enterprise demands, after deep learning of regional industrial value characteristic analysis, policy characteristic analysis, zone bit characteristic analysis and building characteristic analysis through an LM-BP neural network improved by a BP neural network algorithm.

As shown in fig. 2, in the enterprise requirement evaluation module, the industry structured data matched by the enterprise requirement includes: industrial scale, value chain customer market, industrial chain complement and enterprise complement; the enterprise demand matched policy structured data comprises: industry policies, talent policies, and financial policies; the enterprise demand matched zone bit structured data comprises: traffic logistics, supporting resources and planning elements; the building structured data matched by the enterprise demand comprises: building foundation elements, energy conservation, environmental protection, load, fire and explosion prevention and use cost.

The main first-class and second-class indexes are (the three-class indexes are too much and are not listed at all):

A. industry:

industrial scale: industrial production value, industrial profit and income scales above the scale, industrial chain related enterprises, high and new technology enterprises and marketing enterprises;

value chain customers: the system comprises the following components of enterprises of the same type, high and new technology enterprises of the same type, on-scale enterprises of the same type, evaluation scales of the same type of enterprises, downstream enterprises and downstream enterprise scales;

matching an industrial chain: upstream enterprises, upstream enterprise scales, regional brand promotion platform grades/scales, enterprise research and development centers and industrial matching service platforms;

enterprise matching: talent training institution level/scale, financial support fund scale, professional intermediary service scale, one-stop agent service, one-network communication and office, government and enterprise direct traffic and talent apartment;

B. policy:

industry development policy: an industrial structure policy, an industrial organization policy, and an industrial area layout policy;

tax policies: enterprise income tax policy, personal income tax policy and value-added tax policy;

talent policy: talent drop policy, talent subsidy policy, and talent hold policy;

financial policy: researching and developing a policy subsidy, a innovation policy subsidy and a tax policy subsidy;

C. location:

traffic logistics: distance from airports, distance from high-speed rails/railway stations, distance from national roads/provincial roads/highways, distance from ports/docks and government agencies, and logistics;

and (3) matching resources: residential, commercial, medical, educational, greenfield, lifestyle services;

planning elements: garden building time, land property, floor area, admission industry, volume rate, building attribute, building classification and renting and selling proportion;

D. building:

building foundation elements: total building area, building floor area, parking spaces, single-storey building area, building contents, non-motor vehicle parking spaces, floors, storey height and the like;

energy conservation and environmental protection: safety protection distance, radiation intensity, vibration, noise level limit and corrosion resistance;

loading: permanent load, variable load, accidental load;

fireproof and explosion-proof: fire rating, fire limit, fire spacing;

the use cost is as follows: availability of buildings, rent, property management cost, water consumption of ten thousand yuan production value and energy consumption of ten thousand yuan production value.

Introduction of LM-BP neural network principle for BP neural network algorithm improvement:

the LM algorithm has the advantages of a gradient method and a Newton method, and in order to reduce the singular problem of a non-optimal point, the quadratic property is approximated by the characteristic of a second derivative near an extreme point when an objective function is close to the optimal point, so that the optimization convergence process is accelerated. The method is much faster than the gradient method and the BP algorithm, and the efficiency of big data analysis and processing is optimized.

The specific introduction is as follows:

defining an error function as

Wherein w is a vector formed by the neural network threshold value and the weight; ei is error. The calculation method according to the gauss-newton method includes:

W(k+1)＝W(k)-[J^T(W_k)J(W_k)]^-1J(W_k)e(W_k)

w (k) represents a vector formed by the threshold and the weight of the kth neural network iteration, and W (k +1) represents a vector formed by the threshold and the weight of the new kth iteration + 1.

The LM algorithm is a modified gaussian-newton method, as follows:

W(k+1)＝W(k)-[J^T(w_k)J(w_k)+μ_kI]^-1J(w_k)e(w_k)

where I is the identity matrix, J is the Jacobian matrix, and uk is a scaling factor. The key step of the LM algorithm is the calculation of the Jacobian matrix, and the deformation calculation of the BP algorithm is used, and the method comprises the following steps:

equating to a gauss-newton algorithm if the scaling factor u is 0; if the scale factor is very large, the LM algorithm is close to the gradient algorithm. Each iteration is one step, u is reduced by some. Therefore, when the target is close to the error target, the method is closer to the Gauss-Newton algorithm, and the calculation speed and the calculation precision are high. u is a heuristic parameter, and for a given parameter u, u decreases if the calculated threshold change Δ w is such that the error function e (w) decreases. Otherwise u increases. The LM algorithm is calculated much faster than the gradient from information that utilizes the second derivative. The practical test data proves that the calculation speed of the LM-BP algorithm is reduced by dozens of times compared with the traditional BP gradient.

Since the BP neural network has been disclosed as an open source, the specific principle and the algorithm logic of the BP neural network are not specifically described in the technical scheme of the patent, as shown in fig. 3; as shown in fig. 4, mainly introducing the part of optimization improvement, the specific calculation steps of the LM-BP neural network include the following steps:

step 1.3: calculating a network output and error index function e;

step 1.5: calculating delta W; wherein Δ W is a threshold change amount;

step 1.6: if e < epsilon, go to step 1.8, otherwise go to step 1.5;

W(k+1)＝W(k)-{J^T[W(k)]J[W(k)]}^-1J[W(k)]e[W(k)]

step 1.8: and the LM-BP neural network calculation is finished.

The data of industry, policy, zone and building structure matched with enterprise requirements accurately are calculated through the LM-BP neural network, and indexes and weights matched with enterprise requirement figures are scored through a business recruitment strategy module.

In step 1.1, the value range of b is: 0< b <1, and when k is 0 and u is u0, the accuracy e and the maximum learning number M are calculated.

In step 1.4 and step 1.7, the Jacobian matrix J [ W (k)]Is deformed to calculate J^T(W), calculate J^TThe formula of (W) is:

introduction of the function of the recruiter policy module:

the main functions of the enterprise demand image-based business recruitment strategy evaluation method are based on enterprise demand image and matched local industry, policy, location and building resources to evaluate, machine learning and statistics are used for constructing an enterprise demand business recruitment strategy report evaluation model, enterprise demand characteristics are followed, relationships among different characteristics are excavated, an enterprise demand evaluation model based on random forest improvement is established, enterprise demand characteristic conditions are classified and predicted, and enterprises with high weight index matching degree are accurately matched with business recruitment enterprises.

Introduction of algorithm principle and logic:

as shown in fig. 5, the basic component unit of the random forest algorithm is a decision tree, and the main idea of the algorithm is to repeatedly and randomly extract n samples in an original sample set S in a put-back manner to generate a new sample set, then generate n classification trees according to a self-help sample set to form a random forest, and finally decide according to the voting amount of the classification trees of the classification results.

The random forest algorithm is simple to implement, high in training speed, strong in generalization capability and strong in robustness, so that the improved random forest algorithm is applied to the construction of the enterprise demand scoring model.

The enterprise demand scoring model based on random forest improvement mainly comprises 6 parts: enterprise demand data preprocessing, enterprise demand matrix, data weighted sampling, feature selection method selection of an optimal demand feature subset of enterprise demands, algorithm parameter optimization, and evaluation result generation, as shown in fig. 11.

Step 2, the step of establishing an enterprise demand scoring model based on random forest improvement comprises the following steps:

step 2.1: preprocessing enterprise demand data;

the enterprise demand data is complex, including matching analysis of indexes of local industry, policy, location and construction demand, preliminary matching screening has been performed on an enterprise demand matching module in the last step, and due to the fact that missing, abnormal and redundant data often exist in tests, the data still needs to be preprocessed before a scoring model is established in order to reduce noise data in the enterprise demand data and bring evaluation difficulty to scoring logic and meet calculation requirements and result effectiveness. As shown in fig. 11, the pretreatment method used in the present technical solution mainly includes:

(1) Min-Max standardization

The result is between [0-1] mainly by linear transformation of the off-line data, and the formula is as follows:

where Min _ value is the minimum value of the data sample, Max _ value is the maximum value in the data sample, and new _ value (x) is the new data value of the sample data after Min-Max normalization.

(2) Z-Score normalization

The mean and standard deviation (mean) of the raw data were calculated mainly and then subjected to Z-Score processing. The target converts the data to a Gaussian distribution with mean of 0 and standard deviation of 1. The formula is as follows:

wherein mean of all data sample sets is u (data), standard definition of all data sample sets is o (data), and Z (new _ data) is a data value processed by Z-Score of the data sample set.

Because the enterprise demand data set has non-numerical characteristics, the characteristics are processed by adopting One-Hot coding according to the following formula:

wherein A, B distribution represents two characteristic attributes, ra, b represents the correlation degree of the two characteristic attributes, n is the number of tuples, ai and bi are the values on A, B, Amean and Bmean are the mean values on A, B, aibi is the cross product of AB, σ_Aσ_BThe higher the ra and B values are, the higher the correlation degree of the two characteristics is, and meanwhile, the higher the ra and B values are, the characteristic attribute A or B can be deleted as a redundant characteristic attribute.

Step 2.2: calculating an enterprise demand matrix;

in step 2.2, the specific steps of calculating the enterprise demand matrix are as follows: assuming that X ═ b1, b2, … …, bL } represents a set of L samples of M features, and Y ═ Y1, Y2, … …, yL } represents a set of categories, the enterprise demand data may be constructed as a matrix:

step 2.3: data weighted sampling;

the random forest algorithm adopts a bootstrap sampling method by default, a large error is generated by the method, the structure of original data is damaged, grading is extremely unbalanced, and enterprises with unmatched enterprise demands are mistakenly considered as business recruitment adaption enterprises possibly after the data are unbalanced. In order to improve the accuracy of the scoring model, the original bootstrap sampling method is modified, namely sampling is carried out according to the weight.

Thus, in step 2.3, the specific steps of data weighted sampling include:

After weighted sampling is performed on the data, a majority class L 'of enterprise demand is extracted, and assuming that how many classes of samples L' each class comprises the number of samples H (k1), H (k2), H (k3), …, H (kn), where k1, k2, … kn represent the distribution of how many classes of samples L ', the weight ratio of ki sampled in the majority sample L' is:

and calculating the weight of the sampled few samples in the multiple samples according to the formula, wherein the overall weight of ki sampled in the few samples is as follows:

assuming that the number of the minority samples is Q according to the above complaints, the weighted sampling weight of kj in the multiple samples is:

in step 2.4, the input is the original enterprise demand dataset D { (x1, y1), (x2, y2), … …, (xn, yn) }, x_i∈R^mAnd y is_nE { -1, 1 }; setting g1, g 2;

outputting an optimal feature subset f;

the specific selection step of selecting the optimal requirement feature subset of the enterprise requirement by the feature selection method, as shown in fig. 8, includes:

the information entropy obtained by the above formula is:

Gain(x，y)＝Info(x)-Info(x|y)

step 2.4.6: an optimal feature subset is obtained.

Step 2.5: optimizing algorithm parameters;

the traditional random forest algorithm has the following disadvantages:

(1) the algorithm parameters are complex: the random forest has more set parameters before training, and the set parameters mainly comprise parameters such as n _ estimators (number of decision trees), maximum depth of trees, maximum feature number max _ feature and the like, and due to the fact that the parameters are set in advance, final evaluation precision of enterprise requirements can be seriously influenced if the parameters are not reasonable. If the parameter setting is too small, under-fitting is easy to occur; if the parameter setting is too large, overfitting is easy to occur;

(2) after the number of decision trees is too large, the training time is too long; when the number of the decision trees is too small, the training time is short, and the prediction precision is influenced. How to balance the two relationships is also a challenge in massive data training;

improvement of random forest algorithm parameter selection:

introducing a grid search strategy to optimize n _ estimators (number of decision trees) and max _ feature parameters of the maximum feature number in the random forest; assuming that n _ estimators (number of decision trees) and max _ feature are S, C, respectively, training the random forest classifier with S × C, in step 2.5, a specific optimization step of algorithm parameter optimization, as shown in fig. 9, includes:

step 2.5.3: calculating random forest OOB values by S-C combination according to the value range of the parameter S, C obtained in the step 2.5.2 and using the following process to obtain accuracy;

namely, the generalization error is:

step 2.5.4: and selecting the optimal parameters determined by the S x C combination according to the OOB values, outputting the S x C combination if the random forest OOB values meet the requirements, and otherwise, changing the search range and the step length and continuing searching until final conditions are met.

The grid search strategy is introduced to search for the optimal parameters of the random forest, so that the running time can be reduced, the algorithm complexity is reduced, the algorithm classification precision is provided, and the optimization process is shown in fig. 6.

Step 2.6: generating an evaluation result;

in step 2.6, an optimal evaluation result is generated through the steps 2.1 to 2.5 based on the enterprise demand scoring model improved by the random forest and is used as an evaluation reference for an industrial park recruitment worker to judge whether the enterprise is matched with the park recruitment strategy.

Example 2:

a method for an enterprise intelligent matching recruiter policy system, as shown in fig. 10, comprises the steps of:

the LM-BP neural network principle improved by the BP neural network algorithm is introduced as follows:

the LM algorithm has the advantages of a gradient method and a Newton method, and in order to reduce the singular problem of a non-optimal point, the quadratic property is approximated by the characteristic of a second derivative near an extreme point when an objective function is close to the optimal point, so that the optimization convergence process is accelerated. The speed is much higher than that of a gradient method and a BP algorithm, and the efficiency of big data analysis and processing is optimized;

the specific introduction is as follows:

defining an error function as

W(k+1)＝W(k)-[J^T(W_k)J(W_k)]^-1J(W_k)e(W_k)

The LM algorithm is a modified gaussian-newton method, as follows:

W(k+1)＝W(k)-[J^T(w_k)J(w_k)+μ_kI]^-1J(w_k)e(w_k)

equating to a gauss-newton algorithm if the scaling factor u is 0; if the scale factor is very large, the LM algorithm is close to the gradient algorithm. Each iteration is one step, u is reduced by some. Therefore, when the target is close to the error target, the method is closer to the Gauss-Newton algorithm, and the calculation speed and the calculation precision are high. u is a heuristic parameter, and for a given parameter u, u decreases if the calculated threshold change Δ w is such that the error function e (w) decreases. Otherwise u increases. The LM algorithm is computed much faster than the gradient from information that utilizes the second derivative. The practical test data proves that the calculation speed of the LM-BP algorithm is reduced by dozens of times compared with the traditional BP gradient.

Because the BP neural network has been disclosed as an open source, the specific principle and the algorithm logic of the BP neural network are not specifically described in the technical scheme of the patent, as shown in fig. 3, the optimized and improved part is mainly described, as shown in fig. 4, the specific calculation step of the LM-BP neural network comprises the following steps:

step 1.3: calculating a network output and error index function e;

step 1.5: calculating delta W; wherein Δ W is a threshold change amount;

step 1.6: if e < epsilon, go to step 1.8, otherwise go to step 1.5;

W(k+1)＝W(k)-{J^T[W(k)]J[W(k)]}^-1J[W(k)]e[W(k)]

if e [ W (k +1) ] is smaller than e [ W (k) ], making k ═ k +1 and u ═ b, go to step 1.2, otherwise, go to step 1.5; wherein, W (k) represents a vector formed by the threshold and the weight of the kth neural network iteration, and W (k +1) represents a vector formed by the threshold and the weight of the new kth +1 iteration;

step 1.8: the LM-BP neural network calculation is finished;

introduction of the function of the recruiter policy module:

Introduction of algorithm principle and logic:

The enterprise demand scoring model based on random forest improvement mainly comprises 6 parts: enterprise demand data preprocessing, an enterprise demand matrix, data weighted sampling, feature selection method selection of an optimal demand feature subset of enterprise demands, algorithm parameter optimization and evaluation result generation.

step 2.1: preprocessing enterprise demand data;

the enterprise demand data is complex, including matching analysis of indexes of local industry, policy, location and construction demand, preliminary matching screening has been performed on an enterprise demand matching module in the last step, and due to the fact that missing, abnormal and redundant data often exist in tests, the data still needs to be preprocessed before a scoring model is established in order to reduce noise data in the enterprise demand data and bring evaluation difficulty to scoring logic and meet calculation requirements and result effectiveness. The pretreatment method used in the patent technology mainly comprises the following steps:

(1) Min-Max standardization

where Min _ value is the minimum value of the data sample, Max _ value is the maximum value of the data sample, and new _ value (x) is the new data value of the sample data after Min-Max standardization.

(2) Z-Score normalization

wherein all the data sample sets mean are u (data), all the data sample sets standard deviation are o (data), and Z (new _ data) is a data value processed by the data sample set Z-Score.

wherein A, B distribution represents two characteristic attributes, ra and b represent the degree of association of the two characteristic attributes, n is the number of tuples, ai and bi are values on A, B respectively, Amean and Bmean are mean values on A, B respectively, aibi is the cross product of AB respectively, and σ_Aσ_BThe higher the ra and B values are, the higher the correlation degree of the two characteristics is, and meanwhile, the higher the ra and B values are, the characteristic attribute A or B can be deleted as a redundant characteristic attribute.

Step 2.2: calculating an enterprise demand matrix;

in step 2.2, the concrete steps of calculating the enterprise demand matrix are as follows: assuming that X ═ b1, b2, … …, bL } represents a set of L samples of M features, and Y ═ Y1, Y2, … …, yL } represents a set of categories, the enterprise demand data may be constructed as a matrix:

step 2.3: data weighted sampling;

Thus, in step 2.3, as shown in fig. 7, the specific steps of data weighted sampling include:

outputting an optimal feature subset f;

the information entropy obtained by the above formula is:

Gain(x，，y)＝Info(x)-Info(x|y)

step 2.4.4: if un is greater than etcAt g1, the feature xn is in the selected optimal feature subset f, i.e. x_n∈f；

step 2.4.6: and obtaining an optimal feature subset.

Step 2.5: optimizing algorithm parameters;

the traditional random forest algorithm has the following disadvantages:

improvement of random forest algorithm parameter selection:

introducing a grid search strategy to optimize n _ estimators (number of decision trees) and max _ feature parameters of the maximum feature number in the random forest; assuming that n _ estimators (number of decision trees) and max _ feature are S, C, respectively, training the random forest classifier with S × C, in step 2.5, as shown in fig. 9, the specific optimization step of algorithm parameter optimization includes:

namely, the generalization error is:

Step 2.6: generating an evaluation result;

s6: the building matching module selects an optimal building carrier suitable for the enterprise through a matching planning building library, and performs adaptability evaluation on the building carrier through building foundation information, building structures, energy conservation, environmental protection, fire prevention, explosion prevention, supporting equipment and survival cost, so that optimal building planning suitable for the enterprise is matched and selected.

The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all equivalent structures made by using the contents of the specification and the drawings, or other related technical fields, are encompassed by the present invention.

Claims

1. An intelligent matching business recruitment policy system based on enterprise needs, comprising: the system comprises an enterprise demand evaluation module (1), a business recruitment strategy module (2), an industry matching module (3), a policy matching module (4), an area space matching module (5) and a building matching module (6);

the enterprise demand evaluation module (1) is used for obtaining industry, policy, location and building structured data matched with enterprise demands after deep learning of regional industry value characteristic analysis, policy characteristic analysis, location characteristic analysis and building characteristic analysis through the LM-BP neural network;

the business recruitment strategy module (2) is used for scoring the enterprise demand figures and matched local industries, policies, location and building resources, constructing an enterprise demand business recruitment strategy report scoring model, establishing an enterprise demand scoring model based on random forest improvement, and classifying and predicting the enterprise demand characteristic conditions;

the industry matching module (3) is used for analyzing the industrial scale of local business recruitment areas, the total number of enterprises related to an industrial chain, the quantity of regulated enterprises, the conditions of customers and suppliers in respective subdivided industries of the enterprises through an industry matching library (31) and providing reference values of enterprise industry clustering effect and supply-marketing relationship effect, so that the optimal industry suitable for the enterprises is matched and selected;

the policy matching module (4) is used for analyzing the policies adapted to the enterprises through the policy matching library (41), and making business recruiters follow the specific requirements of the enterprises and recommend marks in the policy matching module (2) through the universal policies adapted to the entrepreneurial period, the growth period and the maturity period of the enterprise, the industrial policies adapted to all the subdivided industries of the enterprise and the policy recommendation for cultivating the enterprise foreground, so as to match and select the optimal policies suitable for the enterprise;

the regional space matching module (5) is used for matching enterprise demand quantitative evaluation scores after data formatting of the matching regional space library (51) and performing adaptive evaluation by matching macroscopic location, mesoscopic location, microscopic location, land planning, traffic logistics and living coordination of enterprise demands, so that the optimal regional space suitable for enterprises is matched and selected;

the building matching module (6) is used for selecting an optimal building carrier suitable for an enterprise through a matching planning building library (61), and performing adaptability evaluation on the building carrier through building basic information, building structures, energy conservation, environmental protection, fire prevention, explosion prevention, supporting equipment and survival cost, so that the optimal building planning suitable for the enterprise is matched and selected.

2. The system of claim 1, wherein the business needs-based intelligent matching recruiter policy system comprises: in the enterprise requirement evaluation module (1), the enterprise requirement matched industry structured data comprises: industrial scale, value chain customer market, industrial chain complement and enterprise complement; the enterprise requirement matched policy structured data comprises: industry policies, talent policies, and financial policies; the enterprise demand matched zone bit structured data comprises: traffic logistics, supporting resources and planning elements; the enterprise demand matched building structured data comprises: building foundation elements, energy conservation, environmental protection, load, fire and explosion prevention and use cost.

3. The system of claim 1, wherein the business needs-based intelligent matching recruiter policy system comprises: the specific calculation steps of the LM-BP neural network in the step 1 are as follows:

step 1.3: calculating a network output and error index function e;

step 1.5: calculating delta W; wherein Δ W is a threshold change amount;

step 1.6: if e is less than epsilon, go to step 1.8, otherwise go to step 1.5;

W(k+1)＝W(k)-{J^T[W(k)]J[W(k)]}^-1J[W(k)]e[W(k)]

step 1.8: and the LM-BP neural network calculation is finished.

4. The system of claim 3, wherein the business needs-based intelligent matching recruiter policy system comprises: in step 1.1, the value range of b is: 0< b <1, and when k is 0 and u is u0, the accuracy ε and the maximum learning time M are calculated.

5. The system of claim 3, wherein the business needs-based intelligent matching recruiter policy system comprises: in step 1.4 and step 1.7, the Jacobian matrix J [ W (k)]Is deformed to calculate J^T(W), calculate J^TThe formula of (W) is:

6. the system of claim 1, wherein the business needs-based intelligent matching recruiter policy system comprises: step 2, the step of establishing the enterprise demand scoring model based on random forest improvement comprises the following steps:

step 2.1: preprocessing enterprise demand data;

step 2.2: calculating an enterprise demand matrix;

step 2.3: data weighted sampling;

step 2.5: optimizing algorithm parameters;

step 2.6: an evaluation result is generated.

7. The system of claim 6, wherein the business demand-based intelligent matching recruiter policy system comprises: in step 2.1, the enterprise demand data preprocessing comprises Min-Max standardization processing and Z-Score standardization processing;

8. The system of claim 6, wherein the business needs-based intelligent matching recruiter policy system comprises: in step 2.2, the concrete steps of calculating the enterprise demand matrix are as follows: assuming that X ═ b1, b2, … …, bL } represents a set of L samples of M features, and Y ═ Y1, Y2, … …, yL } represents a set of categories, the enterprise demand data may be constructed as a matrix:

9. the system of claim 8, wherein the business needs-based intelligent matching recruiter policy system comprises: in step 2.3, the data weighted sampling specifically includes:

10. The system of claim 6, wherein the business needs-based intelligent matching recruiter policy system comprises: in step 2.4, the input is the original enterprise demand dataset D { (x1, y1), (x2, y2), … …, (xn, yn) }, x_i∈R^mAnd y is_nE { -1, 1 }; setting g1, g 2;

outputting an optimal feature subset f;

the information entropy obtained by the above formula is:

Gain(x，y)＝Info(x)-Info(x|y)

step 2.4.6: and obtaining an optimal feature subset.

11. The system of claim 6, wherein the business needs-based intelligent matching recruiter policy system comprises: in step 2.5, the specific optimization step of the algorithm parameter optimization includes:

when the sampling training is carried out on the sample each time, the sample data which is not sampled is marked as a set OOBi, the number of the classified errors of the OOBi in the data set which is not sampled is marked as ErrorNumOOB, and finally the error of the random forest OOB value is defined as:

namely, the generalization error is:

12. The system of claim 6, wherein the business needs-based intelligent matching recruiter policy system comprises: in step 2.6, the enterprise demand scoring model based on random forest improvement generates an optimal evaluation result through steps 2.1 to 2.5, and the optimal evaluation result is used as an evaluation reference and is provided for an industrial park business recruiter to judge whether an enterprise is adapted to a park business recruiting strategy.

13. A method for an enterprise intelligent matching recruiter policy system, comprising: the method comprises the following steps: