CN113537728A - Intelligent recommendation system and recommendation method based on business recruitment in industrial park - Google Patents
Intelligent recommendation system and recommendation method based on business recruitment in industrial park Download PDFInfo
- Publication number
- CN113537728A CN113537728A CN202110706177.3A CN202110706177A CN113537728A CN 113537728 A CN113537728 A CN 113537728A CN 202110706177 A CN202110706177 A CN 202110706177A CN 113537728 A CN113537728 A CN 113537728A
- Authority
- CN
- China
- Prior art keywords
- index
- weight
- layer
- module
- industry
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 190
- 230000007115 recruitment Effects 0.000 title claims abstract description 63
- 238000011156 evaluation Methods 0.000 claims abstract description 93
- 230000008569 process Effects 0.000 claims abstract description 62
- 238000004364 calculation method Methods 0.000 claims abstract description 51
- 238000012216 screening Methods 0.000 claims abstract description 25
- 238000013210 evaluation model Methods 0.000 claims abstract description 10
- 210000002569 neuron Anatomy 0.000 claims description 106
- 239000011159 matrix material Substances 0.000 claims description 82
- 238000013528 artificial neural network Methods 0.000 claims description 53
- 230000006870 function Effects 0.000 claims description 47
- 230000008901 benefit Effects 0.000 claims description 35
- 230000004913 activation Effects 0.000 claims description 25
- 239000013598 vector Substances 0.000 claims description 22
- 238000013135 deep learning Methods 0.000 claims description 18
- 238000005457 optimization Methods 0.000 claims description 13
- 238000012546 transfer Methods 0.000 claims description 13
- 238000004519 manufacturing process Methods 0.000 claims description 12
- 238000011161 development Methods 0.000 claims description 10
- 230000018109 developmental process Effects 0.000 claims description 10
- 238000010606 normalization Methods 0.000 claims description 9
- 230000011218 segmentation Effects 0.000 claims description 9
- 238000011160 research Methods 0.000 claims description 7
- 230000000052 comparative effect Effects 0.000 claims description 6
- 238000009795 derivation Methods 0.000 claims description 6
- 238000007726 management method Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 238000004451 qualitative analysis Methods 0.000 claims description 5
- 238000004445 quantitative analysis Methods 0.000 claims description 5
- 230000008447 perception Effects 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 230000005540 biological transmission Effects 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 abstract description 10
- 238000004422 calculation algorithm Methods 0.000 description 9
- 238000004587 chromatography analysis Methods 0.000 description 8
- 210000004205 output neuron Anatomy 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 238000013139 quantization Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 230000003247 decreasing effect Effects 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003032 molecular docking Methods 0.000 description 1
- 238000011144 upstream manufacturing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Abstract
The invention discloses an intelligent recommendation system based on industry park recruitment, which comprises an industry evaluation screening module, a park impression module, an enterprise impression module, a recruitment strategy module and an intelligent recommendation module; the industry evaluation screening module is used for establishing a combined empowerment evaluation model; the park impression module is used for introducing the bar information of the industrial park in an all-round way; the enterprise impression module is used for accurately inquiring enterprise information in an enterprise big database; the recruiter strategy module is used for providing a model basis for the intelligent recommendation module; the intelligent recommendation module is used for accurately recommending high-quality enterprises for business recruitment in an industrial park; also discloses a recommendation method for the intelligent recommendation system for the business recruiters in the industrial park. The invention utilizes the AHP and the entropy weight method to combine weighting to carry out index weight calculation, thereby not only eliminating the defect of stronger subjectivity of the calculation result of the analytic hierarchy process, but also avoiding the situation that the index weight is not accordant with the actual situation, leading the calculation result of the index weight to be more scientific and accurate, and improving the efficiency of attracting business in industrial parks.
Description
Technical Field
The invention relates to the technical field of intelligent business recruitment, in particular to an intelligent recommendation system and a recommendation method based on business recruitment in an industrial park.
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 of the method is the accurate recruiting problem, and how to select the most suitable enterprises for the best subdivided industries to recruit is the most concerned. However, due to the influence of factors such as performance and information asymmetry, various regions blindly recruit companies, the appropriateness between the accuracy of the recruit companies and the industry is not fully considered, and finally, the recruit enterprises cannot drive the development of the local economy, even need 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.
The accurate business recruiter has the following basic characteristics that firstly, the accurate business recruiter is expressed as industrial positioning project, business recruiter object point location, information docking specialization and the like; secondly, the accurate business recruitment has the characteristic of personalized business recruitment service; finally, the accurate recruitment result can be measured by the industrial structure upgrading and the traditional enterprise transformation condition, the financial and tax growth condition, the employment condition and the like. Therefore, investment can be better attracted and economy can be developed through accurate business recruitment.
A patent No.: 201910366550.8, discloses a system and a method for evaluating investment intention of accurate business enterprise of recruiter based on big data, which comprises a city determining module for selecting and determining the city of the recruiter; a recruiter target determination module for selecting and analyzing to determine a recruiter target; the accurate enrollment industry determining module is used for analyzing an enrollment target proposed by the government in the enterprise big database, screening the enterprise big database and determining the accurate enrollment industry; the investment intention index determining module is used for determining each investment intention index of each enterprise in the accurate business recruitment industry; the investment intention index weighting module is used for carrying out weight analysis on each investment intention index and weighting each index; and the enterprise investment intention assessment module is used for calculating an enterprise investment intention assessment index and determining an accurate business inviting enterprise. The remarkable effects are as follows: the efficiency of attracting business is greatly improved, can help local government to carry out the industry location, realize accurate attracting business.
However, the above patent nos.: 201910366550.8, the invention mainly applies the TOPSIS distance method, the logic of which is relatively single, and the normative decision matrix is more complex, so it is not easy to find the positive ideal solution and the negative ideal solution.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides an intelligent recommendation system and a recommendation method based on industrial park recruitment, after the intelligent recommendation system and the recommendation method are used, index weight calculation is carried out by combining weighting based on the minimum deviation principle of AHP (analytic hierarchy process) and entropy weight method, and the intelligent recommendation system and the recommendation method have obvious advantages compared with a single weighting method, so that the defect of strong subjectivity of the result calculated by the analytic hierarchy process can be eliminated, the condition that the index weight possibly does not accord with the actual condition can be avoided, and the intelligent recommendation module has the advantages of two evaluation methods, so that the calculation result of the index weight is more scientific and accurate, the intelligent recommendation module for the recruitment builds a neural network for deep learning by combining various models and indexes such as industrial advantages, regional advantages, policy advantages, talent advantages and the like through a multilayer perceptron MLP (Multi-level perception processor), more effective accurate business recruitment in the industrial park.
In order to solve the technical problems, the invention adopts the technical scheme that:
the intelligent recommendation system based on industry park business recruitment is provided, and comprises an industry evaluation screening module, a park impression module, an enterprise impression module, a business recruitment strategy module and an intelligent recommendation module;
the industry evaluation screening module is used for establishing a combined weighted evaluation model, calculating the combined weight of the garden industry evaluation indexes based on the minimum deviation principle by adopting an AHP (analytic hierarchy process) and an entropy weight method, and determining the optimal industry segmentation;
the park impression module is used for introducing the bar information of the industrial park in an all-round way;
the enterprise impression module is used for carrying out user-defined accurate query on enterprise information in an enterprise big database;
the business recruitment strategy module defines business recruitment strategies through the industry evaluation screening module and the campus impression module and is used for providing a model basis for the intelligent recommendation module;
the intelligent recommendation module is used for building a neural network MLP multi-layer perceptron deep learning framework through the business recruitment strategy module and the enterprise impression module, and accurately recommending high-quality enterprises for the business recruitment in the industrial park.
In order to solve the technical problem, the invention adopts the further technical scheme that:
further, in the campus impression module, the parcel information includes a regional dominance, a development strategy, an industry cluster, a campus policy, talent introduction, enterprise services, innovative operations, a featured building, and success cases.
Further, in the enterprise impression module, the enterprise information includes enterprise basic information, asset scale, revenue level, enterprise development, enterprise high-management social relationship, enterprise location policy index, and enterprise location industry structure index.
Further, in the recruiter strategy module, the recruiter strategy comprises an industry chain recruiter strategy, an area dominance strategy, a policy dominance strategy, and a talent strategy.
Further, the specific calculation steps of the campus industry evaluation index combination weight in the step 1 are as follows:
deducing a subdivision industry suitable for an industrial park through 25 indexes including five-dimensional locational areas, policies, talents, space and industries;
for example, in an industrial park a, the dominant industry is the internet of things industry, and the internet of things subdivision industry suitable for the local is derived through an industry evaluation module as follows: smart homes, smart sports equipment, smart cars, smart city management equipment;
the main algorithm is AHP chromatography and entropy weight method to screen the optimal subdivision industry;
the basic idea of AHP chromatography and entropy weight method combined weighting is as follows:
the AHP chromatographic analysis method and the entropy weight method are combined weighting methods of a minimum deviation principle analytic hierarchy process and an entropy weight method, and are combined weighting optimization models which are established based on the minimum deviation between weights obtained by the combined weighting and analytic hierarchy process and the entropy weight method. The analytic hierarchy process is a method for decomposing indexes related to an evaluation object into layers such as a layer target, a criterion and an index for qualitative and quantitative analysis; the entropy weight method is a commonly used objective weighting method for determining the weight value of an index by an evaluator according to the information quantity provided by the index in evaluation;
compared with a single weighting method, the AHP chromatographic analysis method and entropy weight method combined weighting based on the minimum deviation principle has obvious advantages in index weight calculation, can eliminate the defect of strong subjectivity of a result calculated by an analytic hierarchy process, can avoid the condition that the index weight is possibly inconsistent with the actual condition, and has the advantages of two evaluation methods, so that the index weight calculation result is more scientific and accurate;
step 1.1: calculating evaluation index weight by an AHP analytic hierarchy process;
the traditional analytic hierarchy process adopts a nine-scale method to determine a judgment matrix, and when the relative importance degree between indexes is judged by adopting the scale, accurate judgment is difficult to make due to more choices; the traditional analytic hierarchy process also has the defects of large calculated amount, easy error occurrence, complicated judgment matrix consistency inspection and the like. Therefore, the patent improves the calculation weight of the analytic hierarchy process, a three-scale method is adopted, namely the comparison result of the index i and the index j is {0,1,2}, so that the judgment difficulty is reduced; b. the optimal transfer matrix is utilized to construct a judgment matrix, consistency check of the judgment matrix is avoided, index weight is directly calculated, and calculation amount is reduced;
step 1.1 the specific calculation steps of the AHP analytic hierarchy process for calculating the evaluation index weight are as follows:
step 1.1.1: establishing a hierarchical structure, wherein the hierarchical structure comprises a target layer, a criterion layer and an index layer, and in an industry evaluation screening module, a first-level index in a combined empowerment evaluation model is used as the target layer, a second-level index is used as the criterion layer, and a third-level index is used as the index layer;
step 1.1.2: scaling by adopting a three-degree index method, and constructing a comparison matrix A reflecting the relative importance of each index in the step 1.1.1;
wherein aij represents the importance of the ith index relative to the jth index, and n is the order of the comparison matrix;
aij is 0, and the j index is more important than the i index;
aij is 1, the ith index is more important than the jth index;
aij is 2, and the ith index is more important than the jth index;
step 1.1.3: calculating a ranking index hi;
calculating the comparison result of the ith index and other indexes in the step 1.1.2, and summing up the comparison results respectively to obtain a ranking index, wherein the ranking index is represented by hi;
step 1.1.4: constructing a judgment matrix B;
constructing a judgment matrix B according to the calculated ranking index hi, wherein elements bij in the judgment matrix B can be obtained by the following formula:
wherein:
step 1.1.5: solving a pseudo-optimal matrix B' of the judgment matrix B in the step 1.1.4;
according to the traditional analytic hierarchy process, the consistency of a judgment matrix is checked, and because the consistency check is complex and has various defects, a quasi-optimal matrix of the judgment matrix is constructed, and the characteristic value and the characteristic vector of the quasi-optimal matrix are solved to obtain the weight value of an index;
if the pseudo-optimal matrix of the judgment matrix B is B ', the element bij ' in B ' is obtained by the following formula: first of all, the calculation of Cij,
substituting Cij into the following formula to obtain bij',
wherein the matrix formed by Cij is an optimal transfer matrix C;
step 1.1.6: solving the maximum eigenvalue of the quasi-optimal matrix B' and the corresponding eigenvector thereof, and carrying out normalization processing to obtain the weight value of each index;
the weight of a certain level index relative to the previous level index can be obtained through the steps from step 1.1.2 to step 1.1.6, the hierarchical structure is three layers, if the weight of the third level index relative to the first level index, namely the weight of the third level index relative to the first level index is calculated, the calculation formula is as follows:
wherein m represents a secondary index, and n represents the number of tertiary indexes; wj represents the weight of the jth index in the secondary indexes relative to the primary indexes, and wij represents the weight of the ith index in the tertiary indexes relative to the jth index in the secondary indexes; if the ith index in the three-level indexes has no relation with the jth index in the two-level indexes, wij is 0;
step 1.2: calculating the weight of the evaluation index by an entropy weight method;
when the weight is calculated by the entropy weight method, the final weight of each index is determined by integrating the importance of the index and the information quantity provided by the index;
step 1.2 the entropy weight method calculates the evaluation index weight with the following specific calculation steps:
step 1.2.1: setting m evaluation objects and n evaluation indexes to form the following original data matrix X,
x=(xij)m×n
wherein xij is the average value of the ith evaluation object on the jth index;
in the matrix x ═ xij)m×nFor the j-th evaluation index, the larger the variation degree of xij is, the larger the information amount provided by the index is, and the larger the weight value is;
in an evaluation index system of industrial evaluation, because different indexes have different meanings and corresponding quantization units are different, empowerment calculation cannot be directly carried out, and in order to eliminate incomparability caused by different index dimensions, evaluation indexes need to be subjected to non-quantization processing to obtain a standardized matrix;
step 1.2.2: standardizing the original data matrix X to obtain a standardized matrix Y;
Y=(xij’)m×n
wherein: xij' represents the normalized value of the ith evaluation object on the jth evaluation index, and xij represents the original data of the ith evaluation object on the jth evaluation index;
step 1.2.3: calculating index weight;
firstly, calculating the proportion of an index standardization value;
the normalization of each evaluation index in the normalization matrix Y is subjected to proportion transformation, the contribution Zij of the jth index and the ith evaluation object is calculated,
then, calculating the information entropy ej of each index;
wherein: when Zij is defined as 0, ZijlnZij is defined as 0;
then, calculating the information entropy redundancy dj;
dj=1-ej
finally, directly calculating the index weight w;
step 1.3: constructing a combined weighting model of an analytic hierarchy process and an entropy weight method based on a minimum deviation principle to calculate a combined weight array of each index;
step 1.3, the concrete calculation steps of calculating the combined weight matrix of each index by constructing the combined weighting model of the analytic hierarchy process and the entropy weight method based on the minimum deviation principle are as follows:
step 1.3.1: let the weight vector of n indexes obtained by AHP analytic hierarchy process be,
let the weight vector of n indices obtained by the entropy weight method be,
in the combined weighting model, the weight vectors obtained by the k methods are set as fk, and the weight vectors obtained by the two weighting methods are set as fk,
f=(f1,f2)Tand f is1+f2=1;
Step 1.3.2: constructing a single-target optimization model and a corresponding Lagrangian function, and solving the values of the weights f1 and f2 and an optimal combined weight vector of the two methods;
a common combination method of subjective and objective weighting is to linearly weight the two weighting methods, i.e.Wherein alpha is a subjective preference coefficient, alpha is more than or equal to 0 and less than or equal to 1, 1-alpha is an objective preference coefficient, and the alpha value is generally determined according to experience and has great subjectivity, so the weighting result also has great subjectivity; to avoid this in the future, a combined weighted optimization model based on the principle of minimum deviation is used, in which the index weight vector ω is (ω ═ ω)1,ω2,…,ωn)TThe total deviation of the calculated result and the weighted value calculated by the analytic hierarchy process and the entropy weight method respectively is minimum;
for this purpose, a single-objective optimization model is constructed,
constructing corresponding Lagrange function, and calculating extremum,
the above function is derived and L (f, λ)' -0, i.e.:
wherein k is 1,2, forming an equation solution containing 3 unknowns and 3 equations, and the coefficient matrix is not equal to zero, as known from the gram's law, the equation set has a unique non-zero solution, the combination of the two methods is in the combined weighting calculation, when s is equal to 1 and 2, the index weights solved by the AHP analytic hierarchy process and the entropy weight method are substituted into the above formula, and the values of the weights f1 and f2 of the two methods and the optimal combined weight vector f can be calculated1ωu+f2ωvAnd deducing the optimal segmentation industry chain sequencing suitable for the park to obtain the optimal segmentation industry category.
Further, the specific deep learning step of the neural network MLP multi-layer perceptron deep learning framework in step 5 is as follows:
MLP multilayer perceptron basic idea:
MLP (multi-layer perceptron) is based on a neural network deep learning model; a typical MLP comprises three layers: the MLP neural network comprises an input layer, a hidden layer and an output layer, wherein different layers of the MLP neural network are fully connected (any neuron in the upper layer is connected with all neurons in the lower layer);
neural networks have three main elements: weight, bias and activation functions;
and (3) weighting: the connection strength between neurons is represented by a weight, and the magnitude of the weight represents the magnitude of the probability;
biasing: the bias is set for correctly classifying the samples and is an important parameter in the model, namely ensuring that an output value calculated by inputting cannot be activated randomly;
activation function: the nonlinear mapping function is realized, the output amplitude of the neuron can be limited within a certain range, and is generally limited to (-1) or (0-1); the most common activation function is the Sigmoid function, which can map a (- ∞, + ∞) number into a range of (0-1);
step 5.1: calculating a first stage output based on the MLP neural network;
calculating a trigger value of each neuron of each layer, wherein the trigger value is obtained by calculating the sum of products of values of all neurons of a previous layer which link the current neuron and corresponding weights;
the activation function is used for normalizing the output of each neuron, the activation function is frequently appeared in the analysis of a perceptron, the output of the neuron is calculated in the neural network layer by layer until the output layer obtains a plurality of output values, the output values are random at the beginning and have no relation with our target value, but the output values are the beginning of a reverse transfer algorithm;
step 5.1 the specific calculation steps for calculating the first stage output based on the MLP neural network are as follows:
step 5.1.1: a weighted sum hj is calculated which is,
wherein hj is the weighted sum input by the current node; wij is the weight from each neuron in the previous layer to the current neuron, namely the weight of the neuron j;
step 5.1.2: the neuron outputs aj of the hidden layer are calculated,
wherein i is used as a subscript of a neuron of a previous layer or is an output layer node; j is used as the subscript of the current neuron or a hidden layer neuron; aj represents the output value of the hidden layer neuron; g () represents an activation function; wij represents the weight from each neuron in the upper layer to the current neuron, namely the weight of the neuron j; x is input, w0jx0j represents an offset node; when aj ═ xjk, it means that the output value of the current layer neuron is equal to the input value of the next layer neuron;
step 5.1.3: the input value y of the output layer is calculated,
y represents an input value of the output layer; hk represents the output weighted sum of output layer neuron k; wjk represents the weight of the current neuron and each neuron in the next layer, namely the weight of the neuron k;
step 5.1.4: normalizing the output value of each neuron using the activation functions Sigmoid' and g (h),
Sigmoid’=σ′(x)=σ(x)[1-σ(x)];
by substituting aj ═ g (hj) into the above equation,
g′(h)=dj(1-dj);
step 5.1.5: using the error square and the calculated loss function E,
squaring (y-t) is adopted to avoid mutual offset of error points at two ends of the hyperplane (y-t has positive and negative);
the front coefficient 1/2 is used to solve the problem that when the gradient is decreased later, the coefficient 2 after square derivation can be cancelled when the gradient (partial derivative) is calculated;
step 5.2: calculating a second stage backward pass based on the MLP neural network;
calculating a local gradient descent delta for each neuron from the output back to the input layer;
the back-propagation algorithm uses a delta gradient descent rule to calculate the local gradient descent of each neuron starting from the output neuron and going back to the input layer; to calculate the delta of the output neuron, firstly, the error of each output neuron is obtained, because the multilayer perceptron is a supervised training network, the error is the difference between the output of the neural network and the actual output, and the part is important in the delta rule and is the essence of a reverse transfer algorithm; by passing the derivative values back, the preceding neurons will know how much the weights are to change to better fit the output of the neural network to the actual output, all starting from the difference between the output of the neural network and the actual output;
step 5.2 the specific calculation steps for calculating the second-stage backward transfer based on the MLP neural network are as follows:
step 5.2.1: reverse transmission, updating the weight;
the gradient descent is adopted to solve the optimal solution, namely to solve the partial derivative of the loss function E with respect to the weight w,
the right side of the equation of the above formula can be understood as: if we want to know how the error E of the output changes when the weight w changes, we can observe how the error E changes with the input value h of the activation function, and how the input value h of the activation function changes with the weight w;
where hk represents the sum of all input weights for output layer neuron k, i.e., the input values of activation function g (h);
the second term on the right of the above equation, the following equation can be finally derived, i.e. the output value aj of the upper layer neuron,
step 5.3: recalculating to perform third-stage weight adjustment;
after calculating the gradient descent deltas of all neurons, starting the calculation of the last phase;
step 5.3 the concrete calculation steps of recalculating and adjusting the weight of the third stage are as follows:
step 5.3.1: updating the output layer increment delta 0;
the gain of the output layer is obtained through the derivation by a chain method,
step 5.3.2: updating the weight wjk of the output layer;
the weights are updated using a gradient descent method for the loss function,
to obtain a formula which is expressed by the following formula,
wjk=wjk-ηδo(k)ai;
wherein ai is an output value of a previous layer, namely an input value xj of an output layer;
step 5.3.3: updating the hidden layer increment delta h;
where wjk is the weight between the current hidden neuron j and the output layer neuron k;
step 5.3.4: updating hidden layer weight vj;
aj=g(hj);
∑aijwij=hj;
wherein aj is the output value of the current neuron; ai is the input value of the current neuron, i.e., the output value of the previous layer.
The invention also provides a recommendation method based on the intelligent recommendation system for the recruiters in the industrial park, which comprises the following steps:
step 1: screening the optimal subdivision industry through an industry evaluation screening module, establishing a combined weighted evaluation model, calculating the combined weight of the garden industry evaluation indexes based on the minimum deviation principle by adopting an AHP (analytic hierarchy process) and an entropy weight method, and determining the optimal subdivision industry;
step 2: the strip information of the industrial park is introduced in an all-around way through the park impression module, and a fine picture of the industrial park is created;
and step 3: matching the macro-level industrial map of the industrial park after the large enterprise database passes through the ETL process, and performing user-defined accurate query on the input enterprise information in the large enterprise database through an enterprise impression module;
and 4, step 4: an industry evaluation screening module and a park impression module define a recruitment strategy, and a model basis is provided for an intelligent recommendation module through the recruitment strategy module;
and 5: a neural network MLP multi-layer perception machine deep learning framework is built through a business recruitment strategy module and an enterprise impression module, and high-quality enterprises are accurately recommended to be recruited for an industrial park through an intelligent recommendation module.
In order to solve the technical problem, the invention adopts the further technical scheme that:
further, in step 1, the optimal industry segment suitable for the industrial park is derived through 5 first-class indexes and 25 second-class indexes;
the 5 indexes are location, policy, talent, space and industry;
the 25 second-class indexes are regional status, regional resources, living environment, production environment, ecological environment, land policy, talent policy, financial policy, innovation policy, special policy, labor population ratio, labor recruitment difficulty, labor productivity, scientific research institutions, talent introduction, production type buildings, living type buildings, service type buildings, basic supporting facilities, online government indexes, industrial foundations, comparative advantages, industrial association, industrial contribution and ecological benefits;
and the 5 first-class indexes correspond to the 25 second-class indexes one by one, and the method comprises the following steps:
location: regional status, regional resources, living environment, production environment, ecological environment;
policy: land policy, talent policy, financial policy, innovation policy, special policy;
talents: labor population proportion, difficulty of recruitment, labor productivity, scientific research institutions and talent introduction;
space: production type buildings, living type buildings, service type buildings, foundation supporting facilities and online government indexes;
industry: industrial basis, comparative advantages, industrial relevance, industrial contribution and ecological benefits.
Further, in step 1, the AHP analytic hierarchy process + entropy weight method is a combined weighting method based on the minimum deviation principle, and a combined weighting optimization model is established based on the minimum deviation between the combined weighting and the weights obtained by the AHP analytic hierarchy process and the entropy weight method, wherein the AHP analytic hierarchy process decomposes indexes related to an evaluation object into an objective, a criterion and an index, and performs qualitative and quantitative analysis; the entropy weight method will determine the indicator weight value based on the amount of information the evaluator provides in the evaluation according to the indicator.
Further, in step 5, the neural network MLP multi-layer perceptron deep learning framework has 6 neural network layers, where the 6 neural network layers are 1 input layer, 4 hidden layers and 1 output layer respectively; the input layer is input by a neural network, the input number of the enterprise portrait is equivalent to the number of neurons, and the input layer is accurate portrait of the enterprise; the hidden layer is positioned between the input layer and the output layer, the hidden layer maps input to output, the hidden layer comprises 4 recruitment strategies, and the 4 recruitment strategies comprise an industrial chain recruitment strategy, an area advantage strategy, a policy advantage strategy and a talent strategy; the output layer is a high-quality enterprise for attracting business.
The invention has the beneficial effects that:
the method has the advantages that the method has obvious advantages compared with a single empowerment method when the combination of the AHP analytic hierarchy process and the entropy weight method based on the minimum deviation principle is utilized to carry out index weight calculation, the defect that the subjectivity of the result calculated by the analytic hierarchy process is high can be eliminated, the situation that the index weight possibly does not accord with the actual situation can be avoided, and the advantages of two evaluation methods are achieved, so that the index weight calculation result is more scientific and accurate, the intelligent business recruitment recommending module builds a neural network for deep learning through a multi-layer perceptron MLP and various models and indexes such as industry advantages, regional advantages, policy advantages and talent advantages, and algorithm logic is enriched, and accurate business recruitment of an industry park is carried out more effectively;
secondly, the calculation weight of the AHP analytic hierarchy process is improved, a three-scale method is adopted, namely the comparison result of the index i and the index j is {0,1,2}, and the judgment difficulty is reduced; b. the optimal transfer matrix is utilized to construct the judgment matrix, so that the consistency check of the judgment matrix is avoided, the index weight is directly calculated, the calculated amount is reduced, the problem that the traditional analytic hierarchy process determines the judgment matrix by adopting a nine-scale method, and when the scale is adopted to judge the relative importance degree of each index, the accurate judgment is difficult to make due to more choices is solved, and the technical defects that the traditional analytic hierarchy process is large in calculated amount, easy to generate errors, complicated in the consistency check of the judgment matrix and the like are overcome;
according to the method, a full industrial chain map is automatically generated according to the industrial chain subdivision field combined and combed by the AHP and the entropy weight method, and the information and the distribution condition of the enterprises in the upstream, the downstream, the derivation and the matching of the industrial chain are displayed, so that the method is used as a powerful basis for industrial development planning of the industrial park, and the high-quality enterprise directory is accurately and efficiently recommended to the industrial park on the basis of MLP neural network deep learning according to the recruitment strategy, so that the recruitment efficiency based on the industrial park is greatly improved.
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 recommendation system based on industry park recruiters according to the present invention;
FIG. 2 is an architecture diagram of a campus impression module according to the present invention;
FIG. 3 is an architecture diagram of an enterprise impression module in accordance with the present invention;
FIG. 4 is an architecture diagram of a solicitor policy module in accordance with the present invention;
FIG. 5 is a schematic diagram of a neural network MLP multi-layer perceptron framework according to the present invention;
FIG. 6 is a schematic flow chart of a recommendation method based on an intelligent recommendation system for industry park recruiters according to the present invention;
the parts in the drawings are marked as follows:
the system comprises an industry evaluation screening module 1, a park impression module 2, an enterprise impression module 3, a business recruitment strategy module 4 and an intelligent recommendation module 5.
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 recommendation system based on industry park business recruitment, as shown in fig. 1-6, comprising: the system comprises an industry evaluation screening module 1, a park impression module 2, an enterprise impression module 3, a business recruitment strategy module 4 and an intelligent recommendation module 5;
the industry evaluation screening module is used for establishing a combined weighted evaluation model, calculating the combined weight of the garden industry evaluation indexes based on the minimum deviation principle by adopting an AHP (analytic hierarchy process) and an entropy weight method, and determining the optimal industry segmentation;
the park impression module is used for introducing the bar information of the industrial park in an all-round way;
the enterprise impression module is used for carrying out user-defined accurate query on enterprise information in an enterprise big database;
the business recruitment strategy module defines business recruitment strategies through the industry evaluation screening module and the campus impression module and is used for providing a model basis for the intelligent recommendation module;
the intelligent recommendation module is used for building a neural network MLP multi-layer perceptron deep learning framework through the business recruitment strategy module and the enterprise impression module, and accurately recommending high-quality enterprises for the business recruitment in the industrial park.
In the campus impression module, the parcel information includes location dominance, development strategy, industry cluster, campus policy, talent introduction, enterprise service, innovation operations, featured buildings, and success cases.
In the enterprise impression module, the enterprise information comprises enterprise basic information, asset scale, revenue level, enterprise development, enterprise high-management social relationship, enterprise location region policy index and enterprise location region industry structure index.
In the recruiter strategy module, the recruiter strategy comprises an industry chain recruiter strategy, an area dominance strategy, a policy dominance strategy and a talent strategy.
The concrete calculation steps of the combined weight of the garden industry evaluation indexes in the step 1 are as follows:
deducing a subdivision industry suitable for an industrial park through 25 indexes including five-dimensional locational areas, policies, talents, space and industries;
for example, in an industrial park a, the dominant industry is the internet of things industry, and the internet of things subdivision industry suitable for the local is derived through an industry evaluation module as follows: smart homes, smart sports equipment, smart cars, smart city management equipment;
the main algorithm is AHP chromatography and entropy weight method to screen the optimal subdivision industry;
the basic idea of AHP chromatography and entropy weight method combined weighting is as follows:
the AHP chromatographic analysis method and the entropy weight method are combined weighting methods of a minimum deviation principle analytic hierarchy process and an entropy weight method, and are combined weighting optimization models which are established based on the minimum deviation between weights obtained by the combined weighting and analytic hierarchy process and the entropy weight method. The analytic hierarchy process is a method for decomposing indexes related to an evaluation object into layers such as a layer target, a criterion and an index for qualitative and quantitative analysis; the entropy weight method is a commonly used objective weighting method for determining the weight value of an index by an evaluator according to the information quantity provided by the index in evaluation;
compared with a single weighting method, the AHP chromatographic analysis method and entropy weight method combined weighting based on the minimum deviation principle has obvious advantages in index weight calculation, can eliminate the defect of strong subjectivity of a result calculated by an analytic hierarchy process, can avoid the condition that the index weight is possibly inconsistent with the actual condition, and has the advantages of two evaluation methods, so that the index weight calculation result is more scientific and accurate;
step 1.1: calculating evaluation index weight by an AHP analytic hierarchy process;
the traditional analytic hierarchy process adopts a nine-scale method to determine a judgment matrix, and when the relative importance degree between indexes is judged by adopting the scale, accurate judgment is difficult to make due to more choices; the traditional analytic hierarchy process also has the defects of large calculated amount, easy error occurrence, complicated judgment matrix consistency inspection and the like. Therefore, the patent improves the calculation weight of the analytic hierarchy process, a three-scale method is adopted, namely the comparison result of the index i and the index j is {0,1,2}, so that the judgment difficulty is reduced; b. the optimal transfer matrix is utilized to construct a judgment matrix, consistency check of the judgment matrix is avoided, index weight is directly calculated, and calculation amount is reduced;
step 1.1 the specific calculation steps of the AHP analytic hierarchy process for calculating the evaluation index weight are as follows:
step 1.1.1: establishing a hierarchical structure, wherein the hierarchical structure comprises a target layer, a criterion layer and an index layer, and in an industry evaluation screening module, a first-level index in a combined empowerment evaluation model is used as the target layer, a second-level index is used as the criterion layer, and a third-level index is used as the index layer;
step 1.1.2: scaling by adopting a three-degree index method, and constructing a comparison matrix A reflecting the relative importance of each index in the step 1.1.1;
wherein aij represents the importance of the ith index relative to the jth index, and n is the order of the comparison matrix;
aij is 0, and the j index is more important than the i index;
aij is 1, the ith index is more important than the jth index;
aij is 2, and the ith index is more important than the jth index;
step 1.1.3: calculating a ranking index hi;
calculating the comparison result of the ith index and other indexes in the step 1.1.2, and summing up the comparison results respectively to obtain a ranking index, wherein the ranking index is represented by hi;
step 1.1.4: constructing a judgment matrix B;
constructing a judgment matrix B according to the calculated ranking index hi, wherein elements bij in the judgment matrix B can be obtained by the following formula:
wherein:
step 1.1.5: solving a pseudo-optimal matrix B' of the judgment matrix B in the step 1.1.4;
according to the traditional analytic hierarchy process, the consistency of a judgment matrix is checked, and because the consistency check is complex and has various defects, a quasi-optimal matrix of the judgment matrix is constructed, and the characteristic value and the characteristic vector of the quasi-optimal matrix are solved to obtain the weight value of an index;
if the pseudo-optimal matrix of the judgment matrix B is B ', the element bij ' in B ' is obtained by the following formula: first of all, the calculation of Cij,
substituting Cij into the following formula to obtain bij',
wherein the matrix formed by Cij is an optimal transfer matrix C;
step 1.1.6: solving the maximum eigenvalue of the quasi-optimal matrix B' and the corresponding eigenvector thereof, and carrying out normalization processing to obtain the weight value of each index;
the weight of a certain level index relative to the previous level index can be obtained through the steps from step 1.1.2 to step 1.1.6, the hierarchical structure is three layers, if the weight of the third level index relative to the first level index, namely the weight of the third level index relative to the first level index is calculated, the calculation formula is as follows:
wherein m represents a secondary index, and n represents the number of tertiary indexes; wj represents the weight of the jth index in the secondary indexes relative to the primary indexes, and wij represents the weight of the ith index in the tertiary indexes relative to the jth index in the secondary indexes; if the ith index in the three-level indexes has no relation with the jth index in the two-level indexes, wij is 0;
step 1.2: calculating the weight of the evaluation index by an entropy weight method;
when the weight is calculated by the entropy weight method, the final weight of each index is determined by integrating the importance of the index and the information quantity provided by the index;
step 1.2 the entropy weight method calculates the evaluation index weight with the following specific calculation steps:
step 1.2.1: setting m evaluation objects and n evaluation indexes to form the following original data matrix X,
X=(xij)m×n
wherein xij is the average value of the ith evaluation object on the jth index;
in matrix X ═ Xij)m×nFor the j-th evaluation index, the larger the variation degree of xij is, the larger the information amount provided by the index is, and the larger the weight value is;
in an evaluation index system of industrial evaluation, because different indexes have different meanings and corresponding quantization units are different, empowerment calculation cannot be directly carried out, and in order to eliminate incomparability caused by different index dimensions, evaluation indexes need to be subjected to non-quantization processing to obtain a standardized matrix;
step 1.2.2: standardizing the original data matrix X to obtain a standardized matrix Y;
Y=(xij’)m×n
wherein: xij' represents the normalized value of the ith evaluation object on the jth evaluation index, and xij represents the original data of the ith evaluation object on the jth evaluation index;
step 1.2.3: calculating index weight;
firstly, calculating the proportion of an index standardization value;
the normalization of each evaluation index in the normalization matrix Y is subjected to proportion transformation, the contribution Zij of the jth index and the ith evaluation object is calculated,
then, calculating the information entropy ej of each index;
wherein: when Zij is defined as 0, ZijlnZij is defined as 0;
then, calculating the information entropy redundancy dj;
dj=1-ej
finally, directly calculating the index weight w;
step 1.3: constructing a combined weighting model of an analytic hierarchy process and an entropy weight method based on a minimum deviation principle to calculate a combined weight array of each index;
step 1.3, the concrete calculation steps of calculating the combined weight matrix of each index by constructing the combined weighting model of the analytic hierarchy process and the entropy weight method based on the minimum deviation principle are as follows:
step 1.3.1: let the weight vector of n indexes obtained by AHP analytic hierarchy process be,
let the weight vector of n indices obtained by the entropy weight method be,
in the combined weighting model, the weight vectors obtained by the k methods are set as fk, and the weight vectors obtained by the two weighting methods are set as fk,
f=(f1,f2)Tand f is1++2=1;
Step 1.3.2: constructing a single-target optimization model and a corresponding Lagrangian function, and solving the values of the weights f1 and f2 and an optimal combined weight vector of the two methods;
a common combination method of subjective and objective weighting is to linearly weight the two weighting methods, i.e.Wherein alpha is a subjective preference coefficient, alpha is more than or equal to 0 and less than or equal to 1, 1-alpha is an objective preference coefficient, and the alpha value is generally determined according to experience and has great subjectivity, so the weighting result also has great subjectivity; to avoid this in the future, a combined weighted optimization model based on the principle of minimum deviation is used, in which the index weight vector ω is (ω ═ ω)1,ω2,…,ωn)TThe total deviation of the calculated result and the weighted value calculated by the analytic hierarchy process and the entropy weight method respectively is minimum;
for this purpose, a single-objective optimization model is constructed,
constructing corresponding Lagrange function, and calculating extremum,
the above function is derived and L (f, λ) is made 0, i.e.:
wherein k is 1,2, forming an equation solution containing 3 unknowns and 3 equations, and the coefficient matrix is not equal to zero, as known from the gram's law, the equation set has a unique non-zero solution, the combination of the two methods is in the combined weighting calculation, when s is equal to 1 and 2, the index weights solved by the AHP analytic hierarchy process and the entropy weight method are substituted into the above formula, and the values of the weights f1 and f2 of the two methods and the optimal combined weight vector f can be calculated1ωu+f2ωvAnd deducing the optimal segmentation industry chain sequencing suitable for the park to obtain the optimal segmentation industry category.
MLP multilayer perceptron basic idea:
MLP (multi-layer perceptron) is based on a neural network deep learning model; a typical MLP comprises three layers: the MLP neural network comprises an input layer, a hidden layer and an output layer, wherein different layers of the MLP neural network are fully connected (any neuron in the upper layer is connected with all neurons in the lower layer);
neural networks have three main elements: weight, bias and activation functions;
and (3) weighting: the connection strength between neurons is represented by a weight, and the magnitude of the weight represents the magnitude of the probability;
biasing: the bias is set for correctly classifying the samples and is an important parameter in the model, namely ensuring that an output value calculated by inputting cannot be activated randomly;
activation function: the nonlinear mapping function is realized, the output amplitude of the neuron can be limited within a certain range, and is generally limited to (-1) or (0-1); the most common activation function is the Sigmoid function, which can map a (- ∞, + ∞) number into a range of (0-1);
step 5.1: calculating a first stage output based on the MLP neural network;
calculating a trigger value of each neuron of each layer, wherein the trigger value is obtained by calculating the sum of products of values of all neurons of a previous layer which link the current neuron and corresponding weights;
the activation function is used for normalizing the output of each neuron, the activation function is frequently appeared in the analysis of a perceptron, the output of the neuron is calculated in the neural network layer by layer until the output layer obtains a plurality of output values, the output values are random at the beginning and have no relation with our target value, but the output values are the beginning of a reverse transfer algorithm;
step 5.1 the specific calculation steps for calculating the first stage output based on the MLP neural network are as follows:
step 5.1.1: a weighted sum hj is calculated which is,
wherein hj is the weighted sum input by the current node; wij is the weight from each neuron in the previous layer to the current neuron, namely the weight of the neuron j;
step 5.1.2: the neuron outputs aj of the hidden layer are calculated,
wherein i is used as a subscript of a neuron of a previous layer or is an output layer node; j is used as the subscript of the current neuron or a hidden layer neuron; aj represents the output value of the hidden layer neuron; g () represents an activation function; wij represents the weight from each neuron in the upper layer to the current neuron, namely the weight of the neuron j; x is input, w0jx0j represents an offset node; when aj ═ xjk, it means that the output value of the current layer neuron is equal to the input value of the next layer neuron;
step 5.1.3: the input value y of the output layer is calculated,
y represents an input value of the output layer; hk represents the output weighted sum of output layer neuron k; wjk represents the weight of the current neuron and each neuron in the next layer, namely the weight of the neuron k;
step 5.1.4: normalizing the output value of each neuron using the activation functions Sigmoid' and g (h),
Sigmoid’=σ′(x)=σ(x)[1-σ(x)]:
by substituting aj ═ g (hj) into the above equation,
g′(h)=aj(1-aj);
step 5.1.5: using the error square and the calculated loss function E,
squaring (y-t) is adopted to avoid mutual offset of error points at two ends of the hyperplane (y-t has positive and negative);
the front coefficient 1/2 is used to solve the problem that when the gradient is decreased later, the coefficient 2 after square derivation can be cancelled when the gradient (partial derivative) is calculated;
step 5.2: calculating a second stage backward pass based on the MLP neural network;
calculating a local gradient descent delta for each neuron from the output back to the input layer;
the back-propagation algorithm uses a delta gradient descent rule to calculate the local gradient descent of each neuron starting from the output neuron and going back to the input layer; to calculate the delta of the output neuron, firstly, the error of each output neuron is obtained, because the multilayer perceptron is a supervised training network, the error is the difference between the output of the neural network and the actual output, and the part is important in the delta rule and is the essence of a reverse transfer algorithm; by passing the derivative values back, the preceding neurons will know how much the weights are to change to better fit the output of the neural network to the actual output, all starting from the difference between the output of the neural network and the actual output;
step 5.2 the specific calculation steps for calculating the second-stage backward transfer based on the MLP neural network are as follows:
step 5.2.1: reverse transmission, updating the weight;
the gradient descent is adopted to solve the optimal solution, namely to solve the partial derivative of the loss function E with respect to the weight w,
the right side of the equation of the above formula can be understood as: if we want to know how the error E of the output changes when the weight w changes, we can observe how the error E changes with the input value h of the activation function, and how the input value h of the activation function changes with the weight w;
where hk represents the sum of all input weights for output layer neuron k, i.e., the input values of activation function g (h);
the second term on the right of the above equation, the following equation can be finally derived, i.e. the output value aj of the upper layer neuron,
step 5.3: recalculating to perform third-stage weight adjustment;
after calculating the gradient descent deltas of all neurons, starting the calculation of the last phase;
step 5.3 the concrete calculation steps of recalculating and adjusting the weight of the third stage are as follows:
step 5.3.1: updating the output layer increment delta 0;
the gain of the output layer is obtained through the derivation by a chain method,
step 5.3.2: updating the weight wjk of the output layer;
the weights are updated using a gradient descent method for the loss function,
to obtain a formula which is expressed by the following formula,
wjk=wjk-ηδ0(k)ai;
wherein ai is an output value of a previous layer, namely an input value xj of an output layer;
step 5.3.3: updating the hidden layer increment delta h;
where wjk is the weight between the current hidden neuron j and the output layer neuron k;
step 5.3.4: updating hidden layer weight vj;
aj=g(hj);
∑aijwij=hj;
wherein aj is the output value of the current neuron; ai is the input value of the current neuron, i.e., the output value of the previous layer.
Example 2: a recommendation method based on an intelligent recommendation system for recruiting business in an industrial park, as shown in fig. 1-6, includes the following steps:
step 1: screening the optimal subdivision industry through an industry evaluation screening module, establishing a combined weighted evaluation model, calculating the combined weight of the garden industry evaluation indexes based on the minimum deviation principle by adopting an AHP (analytic hierarchy process) and an entropy weight method, and determining the optimal subdivision industry;
step 2: the strip information of the industrial park is introduced in an all-around way through the park impression module, and a fine picture of the industrial park is created;
and step 3: matching the macro-level industrial map of the industrial park after the large enterprise database passes through the ETL process, and performing user-defined accurate query on the input enterprise information in the large enterprise database through an enterprise impression module;
and 4, step 4: an industry evaluation screening module and a park impression module define a recruitment strategy, and a model basis is provided for an intelligent recommendation module through the recruitment strategy module;
and 5: a neural network MLP multi-layer perception machine deep learning framework is built through a business recruitment strategy module and an enterprise impression module, and high-quality enterprises are accurately recommended to be recruited for an industrial park through an intelligent recommendation module.
In step 1, the optimal subdivision industry suitable for an industrial park is deduced through 5 first-class indexes and 25 second-class indexes;
the 5 indexes are location, policy, talent, space and industry;
the 25 second-class indexes are regional status, regional resources, living environment, production environment, ecological environment, land policy, talent policy, financial policy, innovation policy, special policy, labor population ratio, labor recruitment difficulty, labor productivity, scientific research institutions, talent introduction, production type buildings, living type buildings, service type buildings, basic supporting facilities, online government indexes, industrial foundations, comparative advantages, industrial association, industrial contribution and ecological benefits;
and the 5 first-class indexes correspond to the 25 second-class indexes one by one, and the method comprises the following steps:
location: regional status, regional resources, living environment, production environment, ecological environment;
policy: land policy, talent policy, financial policy, innovation policy, special policy;
talents: labor population proportion, difficulty of recruitment, labor productivity, scientific research institutions and talent introduction;
space: production type buildings, living type buildings, service type buildings, foundation supporting facilities and online government indexes;
industry: industrial basis, comparative advantages, industrial relevance, industrial contribution and ecological benefits.
In step 1, the AHP analytic hierarchy process + entropy weight method is a combined weighting method based on a minimum deviation principle, and a combined weighting optimization model is established based on the minimum deviation between the combined weighting and the weights obtained by the AHP analytic hierarchy process and the entropy weight method, wherein the AHP analytic hierarchy process decomposes indexes related to an evaluation object into a layer target, a criterion and an index and performs qualitative and quantitative analysis; the entropy weight method will determine the indicator weight value based on the amount of information the evaluator provides in the evaluation according to the indicator.
In step 5, the neural network MLP multi-layer perceptron deep learning framework has 6 neural network layers, wherein the 6 neural network layers are 1 input layer, 4 hidden layers and 1 output layer respectively; the input layer is input by a neural network, the input number of the enterprise portrait is equivalent to the number of neurons, and the input layer is accurate portrait of the enterprise; the hidden layer is positioned between the input layer and the output layer, the hidden layer maps input to output, the hidden layer comprises 4 recruitment strategies, and the 4 recruitment strategies comprise an industrial chain recruitment strategy, an area advantage strategy, a policy advantage strategy and a talent strategy; the output layer is a high-quality enterprise for attracting business.
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 (10)
1. An intelligent recommendation system based on industry park business recruitment is characterized by comprising: the system comprises an industry evaluation screening module (1), a park impression module (2), an enterprise impression module (3), a business recruitment strategy module (4) and an intelligent recommendation module (5);
the industry evaluation screening module is used for establishing a combined weighted evaluation model, calculating the combined weight of the garden industry evaluation indexes based on the minimum deviation principle by adopting an AHP (analytic hierarchy process) and an entropy weight method, and determining the optimal industry segmentation;
the park impression module is used for introducing the bar information of the industrial park in an all-round way;
the enterprise impression module is used for carrying out user-defined accurate query on enterprise information in an enterprise big database;
the business recruitment strategy module defines business recruitment strategies through the industry evaluation screening module and the campus impression module and is used for providing a model basis for the intelligent recommendation module;
the intelligent recommendation module is used for building a neural network MLP multi-layer perceptron deep learning framework through the business recruitment strategy module and the enterprise impression module, and accurately recommending high-quality enterprises for the business recruitment in the industrial park.
2. The intelligent recommendation system based on industry park recruiter according to claim 1, wherein: in the campus impression module, the parcel information includes location dominance, development strategy, industry cluster, campus policy, talent introduction, enterprise service, innovation operations, featured buildings, and success cases.
3. The intelligent recommendation system based on industry park recruiter according to claim 1, wherein: in the enterprise impression module, the enterprise information comprises enterprise basic information, asset scale, revenue level, enterprise development, enterprise high-management social relationship, enterprise location region policy index and enterprise location region industry structure index.
4. The intelligent recommendation system based on industry park recruiter according to claim 1, wherein: in the recruiter strategy module, the recruiter strategy comprises an industry chain recruiter strategy, an area dominance strategy, a policy dominance strategy and a talent strategy.
5. The intelligent recommendation system based on industry park recruiter according to claim 1, wherein: the concrete calculation steps of the combined weight of the garden industry evaluation indexes in the step 1 are as follows:
step 1.1: calculating evaluation index weight by an AHP analytic hierarchy process;
step 1.1 the specific calculation steps of the AHP analytic hierarchy process for calculating the evaluation index weight are as follows:
step 1.1.1: establishing a hierarchical structure, wherein the hierarchical structure comprises a target layer, a criterion layer and an index layer, and in an industry evaluation screening module, a first-level index in a combined empowerment evaluation model is used as the target layer, a second-level index is used as the criterion layer, and a third-level index is used as the index layer;
step 1.1.2: scaling by adopting a three-degree index method, and constructing a comparison matrix A reflecting the relative importance of each index in the step 1.1.1;
wherein aij represents the importance of the ith index relative to the jth index, and n is the order of the comparison matrix;
aij is 0, and the j index is more important than the i index;
aij is 1, the ith index is more important than the jth index;
aij is 2, and the ith index is more important than the jth index;
step 1.1.3: calculating a ranking index hi;
calculating the comparison result of the ith index and other indexes in the step 1.1.2, and summing up the comparison results respectively to obtain a ranking index, wherein the ranking index is represented by hi;
step 1.1.4: constructing a judgment matrix B;
constructing a judgment matrix B according to the calculated ranking index hi, wherein elements bij in the judgment matrix B can be obtained by the following formula:
wherein:
step 1.1.5: solving a pseudo-optimal matrix B' of the judgment matrix B in the step 1.1.4;
if the pseudo-optimal matrix of the judgment matrix B is B ', the element bij ' in B ' is obtained by the following formula: first of all, the calculation of Cij,
substituting Cij into the following formula to obtain bij',
wherein the matrix formed by Cij is an optimal transfer matrix C;
step 1.1.6: solving the maximum eigenvalue of the quasi-optimal matrix B' and the corresponding eigenvector thereof, and carrying out normalization processing to obtain the weight value of each index;
the weight of a certain level index relative to the previous level index can be obtained through the steps from step 1.1.2 to step 1.1.6, the hierarchical structure is three layers, if the weight of the third level index relative to the first level index, namely the weight of the third level index relative to the first level index is calculated, the calculation formula is as follows:
wherein m represents a secondary index, and n represents the number of tertiary indexes; wj represents the weight of the jth index in the secondary indexes relative to the primary indexes, and wij represents the weight of the ith index in the tertiary indexes relative to the jth index in the secondary indexes; if the ith index in the three-level indexes has no relation with the jth index in the two-level indexes, wij is 0;
step 1.2: calculating the weight of the evaluation index by an entropy weight method;
step 1.2 the entropy weight method calculates the evaluation index weight with the following specific calculation steps:
step 1.2.1: setting m evaluation objects and n evaluation indexes to form the following original data matrix X,
X=(xij)m×n
wherein xij is the average value of the ith evaluation object on the jth index;
step 1.2.2: standardizing the original data matrix X to obtain a standardized matrix Y;
Y=(xij′)m×n
wherein: xij' represents the normalized value of the ith evaluation object on the jth evaluation index, and xij represents the original data of the ith evaluation object on the jth evaluation index;
step 1.2.3: calculating index weight;
firstly, calculating the proportion of an index standardization value;
the normalization of each evaluation index in the normalization matrix Y is subjected to proportion transformation, the contribution Zij of the jth index and the ith evaluation object is calculated,
then, calculating the information entropy ej of each index;
wherein: when Zij is defined as 0, ZijlnZij is defined as 0;
then, calculating the information entropy redundancy dj;
dj=1-ej
finally, directly calculating the index weight w;
step 1.3: constructing a combined weighting model of an analytic hierarchy process and an entropy weight method based on a minimum deviation principle to calculate a combined weight array of each index;
step 1.3, the concrete calculation steps of calculating the combined weight matrix of each index by constructing the combined weighting model of the analytic hierarchy process and the entropy weight method based on the minimum deviation principle are as follows:
step 1.3.1: let the weight vector of n indexes obtained by AHP analytic hierarchy process be,
Let the weight vector of n indices obtained by the entropy weight method be,
In the combined weighting model, the weight vectors obtained by the k methods are set as fk, and the weight vectors obtained by the two weighting methods are set as fk,
f=(f1,f2)Tand f is1+f2=1;
Step 1.3.2: constructing a single-target optimization model and a corresponding Lagrangian function, and solving the values of the weights f1 and f2 and an optimal combined weight vector of the two methods;
a single-target optimization model is constructed,
constructing corresponding Lagrange function, and calculating extremum,
the above function is derived and L (f, λ)' -0, i.e.:
wherein k is 1,2, forming an equation solution containing 3 unknowns and 3 equations, and the coefficient matrix is not equal to zero, as known from the claime rule, the equation set has a unique non-zero solution, the combination of the two methods is in the combined weighting calculation, when s is equal to 1 and 2, the index weights solved by the AHP analytic hierarchy process and the entropy weight method are substituted into the above formula, and the weights f1 and f1 of the two methods can be obtained through calculationValue of f2 and optimal combining weight vector f1ωu+f2ωvAnd deducing the optimal segmentation industry chain sequencing suitable for the park to obtain the optimal segmentation industry category.
6. The intelligent recommendation system based on industry park recruiter according to claim 1, wherein: step 5, the specific deep learning steps of the neural network MLP multi-layer perceptron deep learning framework are as follows:
step 5.1: calculating a first stage output based on the MLP neural network;
calculating a trigger value of each neuron of each layer, wherein the trigger value is obtained by calculating the sum of products of values of all neurons of a previous layer which link the current neuron and corresponding weights;
step 5.1 the specific calculation steps for calculating the first stage output based on the MLP neural network are as follows:
step 5.1.1: a weighted sum hj is calculated which is,
wherein hj is the weighted sum input by the current node; wij is the weight from each neuron in the previous layer to the current neuron, namely the weight of the neuron j;
step 5.1.2: the neuron outputs aj of the hidden layer are calculated,
wherein i is used as a subscript of a neuron of a previous layer or is an output layer node; j is used as the subscript of the current neuron or a hidden layer neuron; aj represents the output value of the hidden layer neuron; g () represents an activation function; wij represents the weight from each neuron in the upper layer to the current neuron, namely the weight of the neuron j; x is input, w0jx0j represents an offset node; when aj ═ xjk, it means that the output value of the current layer neuron is equal to the input value of the next layer neuron;
step 5.1.3: the input value y of the output layer is calculated,
y represents an input value of the output layer; hk represents the output weighted sum of output layer neuron k; wjk represents the weight of the current neuron and each neuron in the next layer, namely the weight of the neuron k;
step 5.1.4: normalizing the output value of each neuron using the activation functions Sigmoid' and g (h),
Sigmoid’=σ′(x)=σ(x)[1-σ(x)];
by substituting aj ═ g (hj) into the above equation,
g′(h)=aj(1-aj);
step 5.1.5: using the error square and the calculated loss function E,
step 5.2: calculating a second stage backward pass based on the MLP neural network;
calculating a local gradient descent delta for each neuron from the output back to the input layer;
step 5.2 the specific calculation steps for calculating the second-stage backward transfer based on the MLP neural network are as follows:
step 5.2.1: reverse transmission, updating the weight;
the gradient descent is adopted to solve the optimal solution, namely to solve the partial derivative of the loss function E with respect to the weight w,
the right side of the equation of the above formula can be understood as: if we want to know how the error E of the output changes when the weight w changes, we can observe how the error E changes with the input value h of the activation function, and how the input value h of the activation function changes with the weight w;
where hk represents the sum of all input weights for output layer neuron k, i.e., the input values of activation function g (h);
the second term on the right of the above equation, the following equation can be finally derived, i.e. the output value aj of the upper layer neuron,
step 5.3: recalculating to perform third-stage weight adjustment;
after calculating the gradient descent deltas of all neurons, starting the calculation of the last phase;
step 5.3 the concrete calculation steps of recalculating and adjusting the weight of the third stage are as follows:
step 5.3.1: updating the output layer increment delta 0;
the gain of the output layer is obtained through the derivation by a chain method,
step 5.3.2: updating the weight wjk of the output layer;
the weights are updated using a gradient descent method for the loss function,
to obtain a formula which is expressed by the following formula,
wjk=wjk-ηδo(k)ai;
wherein ai is an output value of a previous layer, namely an input value xj of an output layer;
step 5.3.3: updating the hidden layer increment delta h;
where wjk is the weight between the current hidden neuron j and the output layer neuron k;
step 5.3.4: updating hidden layer weight vj;
aj=g(hj);
∑aijwij=hj;
wherein aj is the output value of the current neuron; ai is the input value of the current neuron, i.e., the output value of the previous layer.
7. A recommendation method for the intelligent recommendation system based on industry park recruiter according to claim 1, characterized by: the method comprises the following steps:
step 1: screening the optimal subdivision industry through an industry evaluation screening module, establishing a combined weighted evaluation model, calculating the combined weight of the garden industry evaluation indexes based on the minimum deviation principle by adopting an AHP (analytic hierarchy process) and an entropy weight method, and determining the optimal subdivision industry;
step 2: the strip information of the industrial park is introduced in an all-around way through the park impression module, and a fine picture of the industrial park is created;
and step 3: matching the macro-level industrial map of the industrial park after the large enterprise database passes through the ETL process, and performing user-defined accurate query on the input enterprise information in the large enterprise database through an enterprise impression module;
and 4, step 4: an industry evaluation screening module and a park impression module define a recruitment strategy, and a model basis is provided for an intelligent recommendation module through the recruitment strategy module;
and 5: a neural network MLP multi-layer perception machine deep learning framework is built through a business recruitment strategy module and an enterprise impression module, and high-quality enterprises are accurately recommended to be recruited for an industrial park through an intelligent recommendation module.
8. The recommendation method based on the intelligent recommendation system for the recruiter in the industrial park as claimed in claim 7, wherein: in step 1, the optimal subdivision industry suitable for an industrial park is deduced through 5 first-class indexes and 25 second-class indexes;
the 5 indexes are location, policy, talent, space and industry;
the 25 second-class indexes are regional status, regional resources, living environment, production environment, ecological environment, land policy, talent policy, financial policy, innovation policy, special policy, labor population ratio, labor recruitment difficulty, labor productivity, scientific research institutions, talent introduction, production type buildings, living type buildings, service type buildings, basic supporting facilities, online government indexes, industrial foundations, comparative advantages, industrial association, industrial contribution and ecological benefits;
and the 5 first-class indexes correspond to the 25 second-class indexes one by one, and the method comprises the following steps:
location: regional status, regional resources, living environment, production environment, ecological environment;
policy: land policy, talent policy, financial policy, innovation policy, special policy;
talents: labor population proportion, difficulty of recruitment, labor productivity, scientific research institutions and talent introduction;
space: production type buildings, living type buildings, service type buildings, foundation supporting facilities and online government indexes;
industry: industrial basis, comparative advantages, industrial relevance, industrial contribution and ecological benefits.
9. The recommendation method based on the intelligent recommendation system for the recruiter in the industrial park as claimed in claim 7, wherein: in step 1, the AHP analytic hierarchy process + entropy weight method is a combined weighting method based on a minimum deviation principle, and a combined weighting optimization model is established based on the minimum deviation between the combined weighting and the weights obtained by the AHP analytic hierarchy process and the entropy weight method, wherein the AHP analytic hierarchy process decomposes indexes related to an evaluation object into a layer target, a criterion and an index and performs qualitative and quantitative analysis; the entropy weight method will determine the indicator weight value based on the amount of information the evaluator provides in the evaluation according to the indicator.
10. The recommendation method based on the intelligent recommendation system for the recruiter in the industrial park as claimed in claim 7, wherein: in step 5, the neural network MLP multi-layer perceptron deep learning framework has 6 neural network layers, wherein the 6 neural network layers are 1 input layer, 4 hidden layers and 1 output layer respectively; the input layer is input by a neural network, the input number of the enterprise portrait is equivalent to the number of neurons, and the input layer is accurate portrait of the enterprise; the hidden layer is positioned between the input layer and the output layer, the hidden layer maps input to output, the hidden layer comprises 4 recruitment strategies, and the 4 recruitment strategies comprise an industrial chain recruitment strategy, an area advantage strategy, a policy advantage strategy and a talent strategy; the output layer is a high-quality enterprise for attracting business.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110706177.3A CN113537728A (en) | 2021-06-24 | 2021-06-24 | Intelligent recommendation system and recommendation method based on business recruitment in industrial park |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110706177.3A CN113537728A (en) | 2021-06-24 | 2021-06-24 | Intelligent recommendation system and recommendation method based on business recruitment in industrial park |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113537728A true CN113537728A (en) | 2021-10-22 |
Family
ID=78096684
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110706177.3A Pending CN113537728A (en) | 2021-06-24 | 2021-06-24 | Intelligent recommendation system and recommendation method based on business recruitment in industrial park |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113537728A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114757730A (en) * | 2022-03-24 | 2022-07-15 | 联通(山东)产业互联网有限公司 | Supplier recommendation method based on intelligent supply chain platform |
CN114819590A (en) * | 2022-04-20 | 2022-07-29 | 平安科技(深圳)有限公司 | Intelligent strategy recommendation method, device, equipment and storage medium |
CN115238753A (en) * | 2022-09-21 | 2022-10-25 | 西南交通大学 | Self-adaptive SHM data cleaning method based on local outlier factor |
CN116383511A (en) * | 2023-05-29 | 2023-07-04 | 四川邕合科技有限公司 | Method, system, terminal and medium for recommending campus recruits based on industry chain analysis |
CN116720792A (en) * | 2023-08-10 | 2023-09-08 | 北京正开科技有限公司 | Incubator public management service system based on data processing |
CN117520864A (en) * | 2024-01-08 | 2024-02-06 | 四川易利数字城市科技有限公司 | Multi-feature fusion intelligent matching method for data elements |
CN117575370A (en) * | 2024-01-16 | 2024-02-20 | 浙江省发展规划研究院 | Project recommendation method and device based on park material flow |
-
2021
- 2021-06-24 CN CN202110706177.3A patent/CN113537728A/en active Pending
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114757730A (en) * | 2022-03-24 | 2022-07-15 | 联通(山东)产业互联网有限公司 | Supplier recommendation method based on intelligent supply chain platform |
CN114819590A (en) * | 2022-04-20 | 2022-07-29 | 平安科技(深圳)有限公司 | Intelligent strategy recommendation method, device, equipment and storage medium |
CN114819590B (en) * | 2022-04-20 | 2023-07-18 | 平安科技(深圳)有限公司 | Policy intelligent recommendation method, device, equipment and storage medium |
CN115238753A (en) * | 2022-09-21 | 2022-10-25 | 西南交通大学 | Self-adaptive SHM data cleaning method based on local outlier factor |
CN115238753B (en) * | 2022-09-21 | 2022-12-06 | 西南交通大学 | Self-adaptive SHM data cleaning method based on local outlier factor |
CN116383511A (en) * | 2023-05-29 | 2023-07-04 | 四川邕合科技有限公司 | Method, system, terminal and medium for recommending campus recruits based on industry chain analysis |
CN116383511B (en) * | 2023-05-29 | 2023-08-04 | 四川邕合科技有限公司 | Method, system, terminal and medium for recommending campus recruits based on industry chain analysis |
CN116720792A (en) * | 2023-08-10 | 2023-09-08 | 北京正开科技有限公司 | Incubator public management service system based on data processing |
CN117520864A (en) * | 2024-01-08 | 2024-02-06 | 四川易利数字城市科技有限公司 | Multi-feature fusion intelligent matching method for data elements |
CN117520864B (en) * | 2024-01-08 | 2024-03-19 | 四川易利数字城市科技有限公司 | Multi-feature fusion intelligent matching method for data elements |
CN117575370A (en) * | 2024-01-16 | 2024-02-20 | 浙江省发展规划研究院 | Project recommendation method and device based on park material flow |
CN117575370B (en) * | 2024-01-16 | 2024-04-12 | 浙江省发展规划研究院 | Project recommendation method and device based on park material flow |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113537728A (en) | Intelligent recommendation system and recommendation method based on business recruitment in industrial park | |
CN106951976B (en) | Bus passenger flow prediction method based on mode classification | |
Bonaccorsi et al. | One or more growth processes? Evidence from new Italian firms | |
CN111160750A (en) | Distribution network analysis and investment decision method based on association rule mining | |
CN112232561B (en) | Power load probability prediction method based on constrained parallel LSTM fractional regression | |
CN113408869A (en) | Power distribution network construction target risk assessment method | |
CN110751355A (en) | Scientific and technological achievement assessment method and device | |
CN113822499B (en) | Train spare part loss prediction method based on model fusion | |
Zhang et al. | A novel normal wiggly hesitant fuzzy multi-criteria group decision making method and its application to electric vehicle charging station location | |
Chen | Evaluating sustainable liveable city via multi-MCDM and hopfield neural network | |
CN110968651A (en) | Data processing method and system based on grey fuzzy clustering | |
Wang et al. | E-commerce credit risk assessment based on fuzzy neural network | |
CN115660170A (en) | Multidimensional index weight collaborative optimization data asset management effect differentiation evaluation method and system | |
CN110956541A (en) | Stock tendency classification prediction method based on intelligent fusion calculation | |
CN116485021A (en) | Coal enterprise technical skill person sentry matching prediction method and system | |
CN116629904A (en) | Client hierarchical matching method based on big data | |
CN115860552A (en) | Rural energy Internet comprehensive benefit evaluation method | |
Reig-Mullor et al. | Sustainability performance assessment with intuitionistic fuzzy composite metrics and its application to the motor industry | |
CN112465253B (en) | Method and device for predicting links in urban road network | |
Ding et al. | Implementation of dynamic credit rating method based on clustering and classification technology | |
Petersohn | Assessment of clusteranalysis and self-organizing maps | |
El‐Sawalhi et al. | Forecasting contractor performance using a neural network and genetic algorithm in a pre‐qualification model | |
CN112116238A (en) | Satisfaction evaluation method based on index weight system design | |
Milosavljević et al. | The selection of optimal transport vehicle using multi criteria decision making methods | |
Haji Mohamad Ali Jahromi et al. | Designing the strategy map with combined approach of Anp and Dematel |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20211022 |