CN106372952A - Objective and subjective weight determining multi-model compositional verification-based enterprise credit assessment method and system - Google Patents

Abstract

The invention provides an objective and subjective weight determining multi-model compositional verification-based enterprise credit assessment method and system. An enterprise assessment index model is built with consideration given to growth force, competitiveness, financing capacity, team power, public opinion power, external force and innovation power. According to the enterprise credit assessment method and system, indexes are subjected to quantitative screening operation, combination of objective and subjective weight determining modes is adopted for the indexes, Kendall consistency coefficients are employed for effectively evaluating assigned values, accuracy of index weight assignment can be improved, a combination evaluation method is used for calculation of enterprise credit assessment, evaluation results obtained via a plurality of methods are subjected to consistency check operation, and capability value of each sub-system and capability value of an overall system are evaluated objectively and comprehensively. Via the enterprise credit assessment method and system, index completeness, effectiveness and accuracy can be ensured; a problem that a current conventional enterprise credit assessment index system is insufficient in verification, poor in evaluation effects, deficient in result verification, insufficient in weight evaluation in combination evaluation and the like can be solved.

Description

A kind of based on subjective and objective assign power Multi-Model Combination checking evaluation of enterprises credit and System

Technical field

The present invention relates to data processing and the credit rank assessment technology towards Enterprise Integrated information, particularly take master See and objective Cross-Validation technique is to the growth power of enterprise, competitiveness, financing power, team's power, public opinion power, external force and innovation Be optimized assignment weight Deng quantizating index, and combined by several evaluation methods and cross validation build business standing comprehensively comment Valency.

Background technology

Business standing is carried out with objective overall merit can help people to understand understanding enterprise's current the industry status, future Development potentiality or operator formulate corporate strategy and provide decision-making foundation and carry out equity investment offer for business investor Selection gist.Currently, valuation of enterprise research primarily focuses on and Enterprise Performance Correlative Influence Factors is carried out with proof analysis, index body The structure research of system is less, is less related to the research of evaluation model structure, so lack really carrying out objective synthesis to enterprise The method evaluated.

Currently enterprise is drawn a portrait, can be used to evaluate business standing, by gathering the related data of enterprise, including tax number Pay data etc. according to, merchandise news, labor service, to describe 360 degree of solid images of enterprise.At present credit rank assessment is mainly Using single evaluation methodology, evaluation effect and result are not tested.Application " one as Publication No. 102629296a Plant the evaluation of enterprises credit based on grey fuzzy ", the enterprise's financial data using Real-time Collection carries out credit rank assessment, Carry out grey correlation analysis and fuzzy cluster analysis after data prediction, obtain evaluation result.Due to various main method All there is certain deficiency, the data of acquisition there is also the problems such as quantity is big, dimension is high, noise is big, result in evaluation result One-sidedness and discordance.In addition, the research being at least partly based on method collection is the combination of objective evaluation, rely only on data originally The information of body is although be not easily susceptible to estimator experience itself and the impact of hobby, but does not but make full use of expertise The importance of index is distinguished.Finally, the research based on combination evaluation has often only carried out combination evaluation to evaluation of estimate, does not have Have and other aspects such as weight are combined evaluate.

Therefore, need at present to provide a kind of model to business standing overall merit and method, to avoid single evaluation side The unilateral and inaccurate problem of the evaluation result that method is brought.

Content of the invention

The present invention is directed to credit rank assessment index system establishment checking deficiency, evaluation effect and the result that presently, there are and tests The problems such as in card shortcoming and combination evaluation, weight evaluates less, there is provided one kind assigns power Multi-Model Combination checking based on subjective and objective Evaluation of enterprises credit and system.

What the present invention provided includes following step based on the subjective and objective evaluation of enterprises credit assigning power Multi-Model Combination checking Rapid:

Step 1, sets up the qualitative evaluation index model of enterprise.From growth power, competitiveness, financing power, team's power, public opinion Power, external force and seven aspects of innovation are set up and are had 7 first class index under business evaluation indicator model, overall objective, each There is one group of two-level index under first class index, under 7 first class index, have 32 two-level index.Obtain all two-level index of enterprise Quantized data, for the actual index that cannot obtain quantized data, using all candidate enterprises meansigma methodss as this index Quantized data.

Described enterprise's qualitative evaluation index model is as shown in table 1.

Table 1 enterprise's qualitative evaluation index model

First class index	Two-level index	First class index	Two-level index
				Growth power	Main operating income	Financing power	Total assets
	Net profit		Enterprise's valuation
					On-job employee's number	Team's power	Postgraduate's number accounting
	Operation total time		Undergraduate course number accounting
					Income from main operation rate		Junior college's number accounting
	Return on Total Assets		Junior college's following number accounting
				Competitiveness	Web flow amount	Public opinion power	News media's concern index
	App flow		Microblogging concern index
					Search engine index		Wechat concern index
	Total number of users	External force	Big industry development temperature
					Conversion number of users		The area grade of enterprise region
	Retain number of users	Innovation	Patent of invention quantity
					Number of users market share		Computer copyright quantity
	Every user's average income arpu		Certificate sum
					Competing product quantity		Trade mark sum
	Ranking in similar competing product		Research staff accounts for employee's proportion

Step 2. quantitative target is screened.Quantitative target analysis is carried out to the quantification of targets data extracted in step 1, calculates one Between the two-level index in level index, between first class index, the correlation coefficient of first class index and overall objective and significance level, screening Credit rank assessment index.

(1) correlation test between the two-level index in first class index；

To each first class index, calculate each two-level index and the multiple correlation organizing other two-level index interior in this first class index Coefficient and significance level, delete multiple correlation coefficient and are more than threshold₁And significance level is more than threshold₂Index. Wherein, threshold is set₁Value is not less than 0.9, threshold₂Value is 0.05.

(2) correlation test between first class index；

Evaluation index model includes growth power, competitiveness, financing power, team's power, public opinion power, external force and innovation 7 First class index, calculates the pearson correlation coefficient between first class index two-by-two and significance level, between inspection first class index respectively Related substitutability.Delete pearson correlation coefficient and be more than threshold₃First class index, threshold₃Usual value is not Less than 0.95.

(3) first class index and overall objective correlation test；

All two-level index values are standardized, weighted average as referring generally to scale value, calculate each first class index with The pearson correlation coefficient of overall objective and significance level.Gained pearson correlation coefficient be on the occasion of first class index with total Body index positive correlation, gained pearson correlation coefficient is that the first class index of negative value is negatively correlated with overall objective.

Step 3. is standardized to the index of step 2 quantitative screening processing.If total n enterprise, remaining after screening have The first class index of effect is p, and two-level index has m, sets up enterprise's efficiency index and quantifies value matrix, element x in matrix_ijRepresent the J-th two-level index value of i enterprise, wherein, i=1,2 ..., n, j=1,2 ..., m.Standard is carried out using extreme value facture Change, if x_ij ^*It is to x using extreme value facture_ijIt is standardized the value obtaining.

Two-level index value standardized value under the positively related first class index with overall objective is:

$\begin{array}{ccc}{x_{ij}}^{*}=0.1+\frac{x_{ij}-\underset{i}{min}\left\{x_{ij}\right\}}{\underset{i}{\max }\left\{x_{ij}\right\}-\underset{i}{min}\left\{x_{ij}\right\}}\×0.9,& i=1,2,...,n;& j=1,2,...,m\end{array};$

Two-level index value standardized value under the negatively correlated first class index with overall objective is:

$\begin{array}{ccc}{x_{ij}}^{*}=0.1+\frac{\underset{i}{\max }\left\{x_{ij}\right\}-x_{ij}}{\underset{i}{\max }\left\{x_{ij}\right\}-\underset{i}{min}\left\{x_{ij}\right\}}\×0.9,& i=1,2,...,n;& j=1,2,...,m\end{array};$

Desired value scope after standardization is 0.1≤x_ij ^*≤1.

Step 4. adopts analytic hierarchy process (AHP) (analytic hierarchy process, ahp) that credit rank assessment is referred to Mark carries out subjective weights.Set up the three-decker of decision problem, destination layer is Enterprise Integrated developing ability, rule layer is that one-level refers to Mark layer, a point rule layer is two-level index layer；Rule layer and a point rule layer are drawn up a questionnaire, expert please adopt 1～9 scaling law, structure Produce the multilevel iudge matrix two-by-two of each layer index, the importance comparative result of element representation two indices in judgment matrix；? Parameter weight consistency check afterwards.

If the weight vectors w of the rule layer obtaining₀=(w₁ ⁽⁰⁾,w₂ ⁽⁰⁾,…,w_p ⁽⁰⁾), w₁ ⁽⁰⁾,w₂ ⁽⁰⁾,…,w_p ⁽⁰⁾Represent p The weighted value of individual first class index；Divide the weight vectors w of rule layer₁=(w₁ ⁽¹⁾,w₂ ⁽¹⁾,…,w_m ⁽¹⁾), w₁ ⁽¹⁾,w₂ ⁽¹⁾,…,w_m ⁽¹⁾ Represent the weighted value of m two-level index.

Step 5. carries out Objective Weight to credit rank assessment index.

(1) Objective Weight is carried out using average variance method, obtain the weight vectors table that all two-level index are with respect to destination layer It is shown as: w₂=(w₁ ⁽²⁾,w₂ ⁽²⁾,…,w_m ⁽²⁾)；

(2) Objective Weight is carried out using Information Entropy, obtain all two-level index and represent with respect to the weight vectors of destination layer For: w₃=(w₁ ⁽³⁾,w₂ ⁽³⁾,…,w_m ⁽³⁾)；

(3) Objective Weight is carried out using critic method, obtain the weight vectors table that all two-level index are with respect to destination layer It is shown as: w₄=(w₁ ⁽⁴⁾,w₂ ⁽⁴⁾,…,w_m ⁽⁴⁾).

Step 6. carries out combination weights.

The weight vectors w that three Objective Weight weight vectors that step 5 is obtained are obtained with analytic hierarchy process (AHP) respectively₁Group Close, obtain optimum combination weight vectors, the optimum combination weight vectors being obtained and weight vectors w₁With Objective Weight weight to The sum of square of deviations of amount is minimum.If the optimum combination weight vectors w that analytic hierarchy process (AHP) and average variance method obtain₅=(w₁ ⁽⁵⁾,w₂ ⁽⁵⁾,…,w_m ⁽⁵⁾)；The optimum combination weight vectors w that analytic hierarchy process (AHP) and Information Entropy obtain₆=(w₁ ⁽⁶⁾,w₂ ⁽⁶⁾,…,w_m ⁽⁶⁾)；Layer The optimum combination weight vectors w that fractional analysis is obtained with critic method₇=(w₁ ⁽⁷⁾,w₂ ⁽⁷⁾,…,w_m ⁽⁷⁾).

Step 7. utilizes three kinds of subjective and objective comprehensive weights to credit rank assessment:

$\begin{array}{ccc}{z}_i^{\left(k\right)}=\underset{j=1}{\overset{m}{\σ}}{w_j}^{\left(k\right)}{x}_{ij},& i=1,2,...,n;& k=5,6,7\end{array}$

Enterprise is commented to pass through weight vectors w for i-th_k=(w₁ ^(k),w₂ ^(k),…,w_m ^(k)) in whole m two-level index On comprehensive evaluation value.Now obtain the credit rank assessment result of three kinds of combining weights.

The business standing that step 8. is obtained to the three kinds of combined methods obtaining in step 7 using kendall consistency coefficient Evaluation result carries out consistency check, eliminates the method being unsatisfactory for consistency check；Three kinds of combined methods refer to: analytic hierarchy process (AHP) Combine with critic method with the combined method of Information Entropy, analytic hierarchy process (AHP) with the combined method of average variance method, analytic hierarchy process (AHP) Method.

Step 9. is combined to the method after consistency check in step 8 evaluating enterprise using combined evaluation model Credit.

(1) arithmetic average built-up pattern is adopted to evaluate business standing, specifically: set y_ijRepresenting is commented enterprise to exist i-th Ranking value under jth kind evaluation methodology, i=1,2 ..., n, j=1,2 ..., g, g are evaluation methodology number；Beaten with sequence first The ranking that every kind of method sorts is converted into integer by point-score, if y_ijCorresponding must be divided into r_ij, then calculate score under distinct methods Meansigma methodss, then by meansigma methodss, each enterprise is resequenced；If the average of You Liangge enterprise is identical, calculate in not Tongfang The little person of the standard deviation of FAXIA score, wherein standard deviation is excellent.

(2) factorial analyses built-up pattern is adopted to evaluate business standing, specifically: to set n and commented enterprise under g kind method Score vector iss_ijIt is to be commented score under jth kind method for the enterprise, s i-th_jIt is n Commented score vector under jth kind method for the enterprise；If f=is (f₁,f₂,...,f_l)^tIt is the common factor vector of g kind method, public Common factor number is l；

First, set up the linear relationship between s and f: s^t=af+e, a are factor loads matrix, and e is specific factor vector； Then, the initial factor, certainty factor number and recognition factor are extracted；From PCA extraction factor；According to eigenvalue Criterion certainty factor number；From the variance rule for the treatment of recognition factor in orthogonal rotation；Finally, calculate and commented enterprise j-th Whole comprehensive evaluation valueWherein v is the factor number extracted, β_iFor the value of i-th factor, ψ_iFor variance Contribution rate.

Step 10. carries out consistency check using spearman coefficient of rank correlation to combination evaluation result in step 9, really Determine the final result of credit rank assessment.

Based on described evaluation of enterprises credit, the invention also discloses a kind of assign power Multi-Model Combination based on subjective and objective The credit rank assessment system of checking, including following ingredient:

Distributed data grasping system, is commented the information data of enterprise, and acquired data processing is obtained for crawl Quantification of targets data to enterprise；

Evaluation index screening and standardized module, for the two-level index in first order calculation index between, between first class index, one The correlation coefficient of level index and overall objective and significance level, reject the evaluation index being unsatisfactory for requiring, to the finger filtering out Mark is standardized processing；

Subjective and objective combination weights module, carries out subjective tax initially with analytic hierarchy process (AHP) to the index of credit rank assessment Power, next is respectively adopted average variance method, Information Entropy and critic method and carries out Objective Weight, finally by three Objective Weight weights to Amount is combined with the weight vectors of subjective weights respectively, obtains optimum combination weight vectors；

First credit rank assessment module, for the optimum combination weight vectors being obtained using subjective and objective combination weights module Overall merit is carried out to business standing, then using kendall consistency coefficient, consistency check is carried out to evaluation result, eliminate It is unsatisfactory for the result of consistency check and corresponding combined method；

Second credit rank assessment module, for the evaluation result that the first credit rank assessment module is filtered out, adopts Arithmetic average built-up pattern and factorial analyses built-up pattern carry out overall merit；Then the comprehensive evaluation result obtaining is utilized Spearman coefficient of rank correlation carries out consistency check, determines the final result of credit rank assessment.

Advantages of the present invention with have the active effect that

(1) index quantification screening when, for two-level index between, enter with overall three aspects with first class index between first class index Correlation series and significance level inspection, remove the invalid index of redundancy it is ensured that the integrity of index, effectiveness and accuracy；

(2) method that subjectivity and objective weight assignment combine is taken to the index of credit rank assessment, and adopt Kendall-w carries out efficiency evaluation to assignment, improves the accuracy of index weights assignment；

(3) overall merit is carried out to enterprise using the combination evaluation including analytic hierarchy process (AHP) and three kinds of Objective Weights, and right Multi-method evaluation result carries out concordance, dependency and validity check, each subsystem of objective overall merit enterprise and whole The ability value of body；

(4) adopt combination evaluation method, solve the result inconsistence problems that multi-method evaluation exists；The present invention passes through distribution To mainboard, new three plates, new four plates etc., open enterprise annual reports captures formula data grabber system, and using at Chinese natural language Destructuring annual report data dissection process is obtained structuring quantification of targets data by reason, in conjunction with the inventive method, enterprise is carried out Credit overall merit.Rely on distributed real-time data grasping system and Chinese natural language treatment technology so that business standing is comprehensive Close evaluation can automatically complete in real time, greatly reduce artificial download, process, the cost of parsing annual report.

(5), during index qualitative analyses of the present invention, taken into consideration residing for the big industry of enterprise and sub-industry development temperature, enterprise The key factor such as Location class, news media and microblogging, wechat degree of concern, overall merit business standing.

Brief description

Fig. 1 is the credit rank assessment model of the present invention；

Fig. 2 is the index quantification screening process figure of the present invention；

Fig. 3 be the present invention carry out credit rank assessment model based on subjective and objective combination weights；

Fig. 4 is that the embodiment of the present invention carries out Objective Weight result schematic diagram using three kinds of methods；

Fig. 5 is combination weights result schematic diagram in the embodiment of the present invention.

Specific embodiment

With reference to the accompanying drawings and examples technical scheme is specifically described.

As shown in figure 1, the evaluation of enterprises credit based on subjective and objective tax power Multi-Model Combination checking is whole for the present invention Body flow process, and using the method, credit appraisal is carried out to 15 enterprises at Beijing share exchange center and verify the inventive method.

Step 1., according to Enterprise Integrated information, formulates the enterprise's qualitative evaluation index model shown in table 1.Handed over Beijing equity Easily 15, center enterprise is object of study, obtains the quantification of targets data of these enterprises with the evaluation index model of table 1.Obtain enterprise All two-level index quantized datas of industry are for the actual quantized data index that cannot obtain, flat using all candidate enterprises Average is as the quantized data reference value of index.The Information Number being commented enterprise can be obtained using distributed data grasping system According to.

Enterprise's quantizating index data that step 2. obtains to step 1 carries out quantitative screening.To the index amount extracted in step 1 Change data and carry out quantitative target analysis, as shown in Fig. 2 between the two-level index in first order calculation index, between first class index, one-level refers to The correlation coefficient of mark and overall objective and significance level, screen credit rank assessment index.What significance level reacted refers to Whether there is correlation, what multiple correlation coefficient reacted is the degree of correlation power between index between mark.

(1) correlation test between the two-level index in first class index.According to greatly uncorrelated principle, reflect same subsystem The measurement index of system should be independent, can not be substituted for each other between each index.Calculate each two-level index with group in other two The multiple correlation coefficient of level index and significance level, delete multiple correlation coefficient and are more than threshold₁And significance level is more than threshold₂Index, retain index related should the smaller the better index, if the index meeting is more, as needed Delete the front several of complex correlation coefficient maximum, to obtain the index quantity setting.Wherein threshold₁Generally value is not little In 0.9, this value embodies the dependency of index and other indexs in group, is worth in bigger explanation group and there are other this indexs alternative Index.threshold₂The usual value of threshold value 0.05.

In the embodiment of the present invention, 15, Beijing share exchange center enterprise is tested, as shown in table 2.

Related-coefficient test between the two-level index in table 2 first class index

In the embodiment of the present invention, threshold is set₁Value 0.95, according to each in the result display growth power index of table 2 Between measurement index, dependency is higher, deletes the main operating income of multiple correlation coefficient highest, operation total time；Each in competitiveness indicator Between measurement index, dependency is higher, deletes conversion number of users, retains number of users；Delete junior college's number in team's power index to account for The following number accounting than, junior college.The multiple correlation coefficient of each index in group is recalculated after deleting the high index of multiple correlation coefficient, As shown in table 3, result shows substantially separate between measurement index in each subsystem, can react its corresponding son from different aspect System, meets the requirements.

Multiple correlation coefficient between each index and other indexs of subsystem after table 3 adjustment

(2) correlation test between first class index.Calculate pearson correlation coefficient between first class index two-by-two and notable respectively Property level, inspection first class index between related substitutability.Delete pearson correlation coefficient and be more than threshold₃One-level refer to Mark, threshold₃Generally value is not less than 0.95.

Between first class index, pearson correlation coefficient is shown in Table 4, and significance level is shown in Table 5.Result shows between each first class index Pearson correlation coefficient is all very little, and maximum is again smaller than 0.5, and its significance level is both greater than 0.05, for notable, Illustrate that each first class index is separate, each first class index independently reflects some aspect of enterprise, each one-level comprehensive refers to Mark then can effectively reflect the general status of enterprise, meets the requirement setting up index system.

Pearson correlation coefficient charts between table 4 first class index

	Growth power	Competitiveness	Financing power	Team's power	Public opinion power	External force	Innovation
								Growth power	1
Competitiveness	-0.464	1
								Financing power	-0.309	-0.018	1
Team's power	-0.298	0.15	-0.065	1
								Public opinion power	-0.006	0.124	0.426	-0.072	1
External force	0.089	0.1	0.142	0.09	0.482	1
								Innovation	-0.157	0.296	0.299	-0.185	0.095	0.301	1

Dependency significance level (bilateral) between table 5 first class index

	Growth power	Competitiveness	Financing power	Team's power	Public opinion power	External force	Innovation
								Growth power	1
Competitiveness	0.081	1
								Financing power	0.262	0.95	1
Team's power	0.28	0.594	0.818	1
								Public opinion power	0.983	0.659	0.113	0.799	1
External force	0.751	0.723	0.613	0.75	0.069	1
								Innovation	0.576	0.283	0.278	0.508	0.737	0.276	1

(3) first class index and overall relevance are checked.All two-level index values are standardized, first class index value is group The weighted mean of interior two-level index value, the weighted mean of all first class index values is to refer generally to scale value, first order calculation index Pearson correlation coefficient and significance level with overall objective.To the first class index totally with certain prediction and refer generally to Scale value should have certain dependency, its predictive validity of guarantee.Meanwhile, it is negative first class index with overall objective dependency Group, will take negative sign in evaluation methodology, to be different from other index groups.

First class index the results are shown in Table 6 with the pearson correlation coefficient of overall objective and significance level, in addition to growth power, Remaining first class index is higher with overall relevance, and significance level is all higher, and the evaluation index model pair that the present invention sets up is described Aggregate level has certain predictability.

Table 6 first class index and overall relevance

	Pearson correlation coefficient	Significance level (bilateral)
			Growth power is average	-0.024	0.932
Competitiveness is average	0.68	0.005
			Financing power is average	0.274	0.323
Team's power is average	0.119	0.671
			Public opinion power is average	0.539	0.038
External force is average	0.645	0.009
			Innovation is average	0.598	0.019

If after step 2, in remaining effective evaluation index, first class index has p, and two-level index has m.If Step 2 carries out screening failure, and the index number for example screened does not meet the minimum quantity of setting, then need again to choose evaluation Index model, the selecting index shown in adjustment table 1.Screening proceeds below step after being successfully completed.

Step 3. analytic hierarchy process (AHP) subjective weights.If through step 2, being left effective first class index is p, two-level index M.The index of step 2 quantitative screening is standardized processing, sets up enterprise's efficiency index and quantify value matrix, as table 7 below institute Show.

Table 7 enterprise's efficiency index quantization matrix

Object	Index 1	Index 2	…	Index m
					Enterprise 1	x11	x12		x1m
Enterprise 2	x21	x22		x2m
					…	…	…	…	…
Enterprise n	xn1	xn2		xnm

X in matrix_ijRepresent j-th two-level index value of i-th enterprise.

It is standardized using extreme value facture, positively related desired value is adopted with the following method with overall objective:

$\begin{array}{ccc}{x_{ij}}^{*}=0.1+\frac{x_{ij}-\underset{i}{min}\left\{x_{ij}\right\}}{\underset{i}{\max }\left\{x_{ij}\right\}-\underset{i}{min}\left\{x_{ij}\right\}}\×0.9,& i=1,2,...,n;& j=1,2,...,m\end{array};---\left(1\right)$

Wherein, x_ij ^*For x_ijThe value obtaining after standardization.

The negatively correlated desired value with overall objective, using following methodological standardization:

$\begin{array}{ccc}{x_{ij}}^{*}=0.1+\frac{\underset{i}{\max }\left\{x_{ij}\right\}-x_{ij}}{\underset{i}{\max }\left\{x_{ij}\right\}-\underset{i}{min}\left\{x_{ij}\right\}}\×0.9,& i=1,2,...,n;& j=1,2,...,m\end{array};---\left(2\right)$

Desired value scope after standardization is [0.1,1].

Step 4. carries out subjective weights using analytic hierarchy process (AHP) to the index of credit rank assessment.Set up the three of decision problem Rotating fields, destination layer is Enterprise Integrated developing ability, and rule layer is first class index layer, and a point rule layer is two-level index layer.Design Questionnaire, expert please construct the judgment matrix two-by-two between each layer factor using 1～9 scaling law；Last parameter weight is simultaneously Consistency check.In one judgment matrix element represent two indices importance degree fiducial value, expert adopt 1～9 scale, 1 Represent that two indices are of equal importance, 3 represent that the former is more important than the latter, 5 represent that the former is obvious more important than the latter, 7 represent the former ratio The latter is strongly important, and 9 represent that the former is more extremely important than the latter, and 2,4,6,8 is the intermediate value of above-mentioned adjacent judgement.

In order to set up judgment matrix two-by-two, agriculture products weight, with the expert in industry as respondent, by electronics postal The form of part provides questionnaire, and business evaluation indicator model is estimated.Provide 45 parts of questionnaire altogether, reality withdraws 39 parts, wherein Effectively 35 parts of questionnaire.First class index judgment matrix is shown in Table 8, for save the judgment matrix of length two-level index and data table related from Slightly.

The weight vectors w of the rule layer being obtained by ahp₀=(w₁ ⁽⁰⁾,w₂ ⁽⁰⁾,…,w_p ⁽⁰⁾), w₁ ⁽⁰⁾,w₂ ⁽⁰⁾,…,w_p ⁽⁰⁾Table Show the weighted value of p first class index.Divide the weight vectors w of rule layer₁=(w₁ ⁽¹⁾,w₂ ⁽¹⁾,…,w_m ⁽¹⁾), w₁ ⁽¹⁾,w₂ ⁽¹⁾,…,w_m ⁽¹⁾Represent the weighted value of m two-level index.

Table 8 weight judgment matrix and related data

Calculate through ahp model, the comprehensive weight w of first class index₀For: (0.1818,0.1818,0.0909,0.1818, 0.1818,0.0909,0.0909).Same method can obtain the weight vectors w of two-level index₁.

Step 5. carries out Objective Weight to credit rank assessment index.It is respectively adopted average variance method, Information Entropy and critic Method Objective Weight.

Step 5.1, carries out Objective Weight using average variance method.If k-th first class index has q effectively two-level index, adopt The two-level index being obtained with average variance method in k-th first class index with respect to the weight vectors of this first class index is:

w^(k)=(w_k1 ^(k),w_k2 ^(k),…,w_kq ^(k)), k=1,2 ..., p (3)

Wherein, w_k1 ^(k),w_k2 ^(k),…,w_kq ^(k)Represent the weighted value of each two-level index in k-th first class index.

Weight w of the rule layer that step 4 is obtained_k ⁽⁰⁾Bring formula (3) into calculate, obtain two grades in k-th first class index Index with respect to the weight vectors of destination layer is:

w^(k)′=w_k ⁽⁰⁾×w^(k)=(w_k1 ^(k)′,w_k2 ^(k)′,…,w_kq ^(k)′), k=1,2 ..., p (4)

Using formula (4) after the two-level index in each first class index of acquisition is with respect to the weight of destination layer, can own Two-level index is with respect to the weight vectors w of destination layer₂=(w₁ ⁽²⁾,w₂ ⁽²⁾,…,w_m ⁽²⁾), w₁ ⁽²⁾,w₂ ⁽²⁾,…,w_m ⁽²⁾It is with mean square The m two-level index that difference method calculates is with respect to the weighted value of destination layer.

Step 5.2, carries out Objective Weight using Information Entropy.

Two-level index is calculated with respect to the weight vectors of destination layer by Information Entropy, is expressed as:

w₃=(w₁ ⁽³⁾,w₂ ⁽³⁾,…,w_m ⁽³⁾) (5)

Wherein, w₁ ⁽³⁾,w₂ ⁽³⁾,…,w_m ⁽³⁾Refer to the weighted value of each two-level index with Information Entropy calculating.

Step 5.3, carries out Objective Weight using critic method

Two-level index is calculated with respect to the weight vectors of destination layer by critic method, is expressed as:

w₄=(w₁ ⁽⁴⁾,w₂ ⁽⁴⁾,…,w_m ⁽⁴⁾) (6)

Wherein, w₁ ⁽⁴⁾,w₂ ⁽⁴⁾,…,w_m ⁽⁴⁾Refer to the weighted value of each two-level index with the calculating of critic method.

Embodiment of the present invention result is as shown in figure 4, result shows: first, all in all tax of three kinds of methods weighs distribution relatively Be close, on-job employee's number, total assets, three indexs of enterprise's valuation weight higher；Second, average variance method and critic method Assign power result to be more nearly, analyze its reason be three kinds of methods mathematical principle variant, average variance method and critic method former Reason is directed to the standard deviation of data, to show the size of each sample value gap of same index in the form of standard deviation, and entropy Value method determines weight according to comentropy.

The subjective and objective integrated weighting method of three kinds of step 6. calculates.

Step 4 has obtained subjective weights weight vectors, gets the Objective Weight weight vectors of three kinds of algorithms in step 5. The each index weights vector being obtained by analytic hierarchy process (AHP) in step 4 is w₁, weighed by each index that average variance method obtains in step 5.1 Weight vector is w₂, now set each indicator combination weight vectors being obtained by above two method as w₅=(w₁ ⁽⁵⁾,w₂ ⁽⁵⁾..., w_m ⁽⁵⁾), The principle of optimal weights combination is w₅With w₁And w₂Sum of square of deviations minimum.

Construction optimizing model is accordingly:

$\left\{\begin{array}{c}\min \underset{k=1}{\overset{2}{\σ}}\left|\right|w_5-w_k|{|}^2=\min \underset{k=1}{\overset{2}{\σ}}\underset{i=1}{\overset{m}{\σ}}{(w_i^{\left(5\right)}-w_i^{\left(k\right)})}^2\\ \begin{array}{ccc}s.t.& \underset{i=1}{\overset{m}{\σ}}{w}_i^{\left(k\right)}=1& (k=0,1,2)\end{array}\end{array}\right.---\left(7\right)$

Using lagrange's method of multipliers, the unique solution trying to achieve above formula is:

$\begin{array}{cc}{w}_i^{\left(5\right)}=\frac{1}{2}(w_i^{\left(1\right)}+w_i^{\left(2\right)})+\frac{1}{p}(1-\frac{1}{2}\underset{i=1}{\overset{m}{\σ}}(w_i^{\left(1\right)}+w_i^{\left(2\right)}\left)\right),& i=1,2,...,m\end{array}---\left(8\right)$

The weight vectors w being obtained by above formula₅=(w₁ ⁽⁵⁾,w₂ ⁽⁵⁾,…,w_m ⁽⁵⁾) it is the optimal set of ahp method and average variance method Close weight.

In the same manner, optimum combination weight w of ahp method and Information Entropy can be obtained₆=(w₁ ⁽⁶⁾,w₂ ⁽⁶⁾,…,w_m ⁽⁶⁾), ahp method with Optimum combination weight w of critic method₇=(w₁ ⁽⁷⁾,w₂ ⁽⁷⁾,…,w_m ⁽⁷⁾).

$\begin{array}{cc}{w}_i^{\left(6\right)}=\frac{1}{2}(w_i^{\left(1\right)}+w_i^{\left(3\right)})+\frac{1}{p}(1-\frac{1}{2}\underset{i=1}{\overset{m}{\σ}}(w_i^{\left(1\right)}+w_i^{\left(3\right)})),& i=1,2,...,m\end{array}---\left(9\right)$

$\begin{array}{cc}{w}_i^{\left(7\right)}=\frac{1}{2}(w_i^{\left(1\right)}+w_i^{\left(4\right)})+\frac{1}{p}(1-\frac{1}{2}\underset{i=1}{\overset{m}{\σ}}(w_i^{\left(1\right)}+w_i^{\left(4\right)}\left)\right),& i=1,2,...,m\end{array}---\left(10\right)$

Using three kinds of subjective and objective combination weights methods, the result that the embodiment of the present invention is obtained is as shown in Figure 5.Contrast Fig. 4 and Fig. 5 finds, compared to the result of three kinds of Objective Weights, three based on optimum combined model kind combination weights result more tends to one Cause.

The weight that step 7. is obtained using three kinds of subjective and objective combination weights methods is to business standing overall merit.Using optimal set Close the business standing comprehensive evaluation value that weight calculation obtainsFor:

$\begin{array}{ccc}{z}_i^{\left(k\right)}=\underset{j=1}{\overset{m}{\σ}}{w_j}^{\left(k\right)}{x}_{ij}& i=1,2,...,n;& k=5,6,7\end{array}---\left(11\right)$

Table 9 is to pass through, to 15 enterprises in the embodiment of the present invention, score and the ranking result that three kinds of methods are evaluated, and method 1 refers to Ahp and the combination of average variance method, method 2 refers to the combination of ahp and Information Entropy, and method 3 refers to the combination of ahp and critic method.

The score of 9 three kinds of evaluation methodologys of table and ranking result

Process shown in step 3～step 7 is as shown in Figure 3.

Step 8. carries out consistency check using kendall-w to the multi-method evaluation result in step 7.

Using kendall consistency coefficient to three kinds of methods in step 7: ahp method+average variance method, ahp method+Information Entropy, The credit rank assessment result that ahp method+critic method obtains carries out consistency check.If being commented enterprise to enter with g kind method to n Row is evaluated, y_ijRepresent and commented ranking value under jth kind evaluation methodology for the enterprise, 1≤y i-th_ij≤ n, i=1,2 ..., n；J= 1,2,…,g.Kendall consistency coefficient t computational methods are as follows:

Wherein

r_iRepresent the summation being commented ranking value in each method evaluation result for the enterprise for i-th.

Construction statistic q:q=g (n-1) t, q approximately obey the χ that degree of freedom is n-1²Distribution.

The rating result score value of three kinds of methods to table 9 for the applying equation (12) and ranking result are tested, and are computed Kendall consistency coefficient t=0.884, chi-square value is 26.533, χ²=46.91.Take level of significance α=0.01, table look-up Marginal value χ² _α/2(n-1)=χ² _0.005(14)=31.3193 it is clear that χ²＞ 31.3193, q value is more than marginal value, therefore three kinds methods Evaluation result totally has concordance.If a certain method, not over consistency check, eliminates the method.

Step 9. is combined to the method after consistency check in step 8 evaluating enterprise using combined evaluation model Credit.

Step 9.1, arithmetic average combined evaluation model.

y_ijRepresent and commented ranking value under jth kind evaluation methodology for the enterprise, i=1,2 ..., n, j=1,2 ..., g i-th. First with Ordering and marking method, the ranking that every kind of method sorts is converted into integer, that is, the 1st n divide, kth name obtains n-j+1 and divides, n-th Name obtains 1 point, wherein if any identical ranking, then takes the average mark of these positions, if y_ijCorresponding must be divided into r_ij, Ran Houji Calculate the meansigma methodss of score under distinct methods, computing formula is as follows:

$\stackrel{\&overbar;}{r_i}=\frac{1}{g}\underset{j=1}{\overset{g}{\σ}}{r}_{ij}---\left(13\right)$

By meansigma methodss, each enterprise is resequenced, if the average of You Liangge enterprise is identical,Then calculate not The standard deviation of Tongfang FAXIA score, computing formula is as follows

$\mathrm{\σ}=\sqrt{\frac{1}{g}\underset{j=1}{\overset{g}{\σ}}{(r_{ij}-\stackrel{\&overbar;}{r_i})}^2}---\left(14\right)$

Wherein the little person of standard deviation is excellent.

Step 9.2, factorial analyses combined evaluation model.

IfIt is n and commented score vector under g kind method for the enterprise, s_ijIt is i-th quilt Comment score under jth kind method for the enterprise, s_jIt is n and commented score vector under jth kind method for the enterprise.The embodiment of the present invention The score to n enterprise for three kinds of combined methods can be obtained according to step 7.

If f=is (f₁,f₂,...,f_l)^tIt is the common factor vector of g kind method, common factor number is l.

First, set up the linear relationship between s and f: s^t=af+e, wherein a are factor loads matrix, and e is specific factor Vector.

Then, the initial factor, certainty factor number and recognition factor are extracted.From PCA extraction factor；Cause Sub- number determines according to Eigenvalue Criteria, i.e. the main constituent more than or equal to 1 for the eigenvalue, as the initial factor, abandons eigenvalue little Main constituent in l；From the variance rule for the treatment of recognition factor in orthogonal rotation.

Finally, calculate the comprehensive evaluation value of multiple methods,z_jCommented the final synthesis of enterprise for j-th Evaluation of estimate, v is the factor number extracted, β_iFor the value of i-th factor, ψ_iFor variance contribution ratio.

In the embodiment of the present invention, former variable is three kinds of combined methods described in step 6, now by the side of principal component analysiss Method finds out number and the common factor of the common factor of three kinds of methods.Embodiment of the present invention application factor analytic process is combined commenting Valency, kmo (kaiser-meyer-olkin) assay of factorial analyses is 0.714 (p < 0.0001), and more than 0.5, reliability is divided Analysis result display cronbach ' s alpha is 0.95, more than 0.8, illustrates that factor-analysis approach is feasible, and reliable results.The factor The constituents extraction of analysis is shown in Table 10 and 11, and the computing formula of factor score is: factor score=0.335*critic+0.334* is equal Variance method+Information Entropy * 0.334.

Table 10 results of factor analysis

Table 11 factor score table

The evaluation result of factorial analyses combined evaluation model and arithmetic average combined evaluation model is shown in Table 12.

The evaluation result of table 12 combined method

Step 10. carries out consistency check to combination evaluation result in step 9.

By consistency check, on the one hand check each combined method gained ranking results and original method gained ranking results Between level of intimate, on the other hand check the level of intimate between each combined method gained ranking results.In addition, it is multiple when having During combined method, also can select the most rational combination evaluation methods with it.Using the inspection of spearman coefficient of rank correlation, including step Rapid 10.1～step 10.4.

Step 10.1, combination evaluation result is converted into ranking value.

If to g kind, former method carries out d kind combination, u in step 9_ikRepresent i-th and commented row under the combination of kth kind for the enterprise Sequence value, wherein, i=1,2 .., n；K=1,2 ..., d, 1≤u_ik≤ n, n are to be commented object number.G kind in the embodiment of the present invention Former method refers to: ahp method+average variance method, ahp method+Information Entropy, ahp method+critic method；The combination of d kind refers in step 9 according to g The combination evaluation that the evaluation result of the method for kind is carried out again, including arithmetic average combination and factorial analyses combination.

Table 13 combination evaluation sort result table

Commented enterprise	Combination 1	Combination 2	...	Combination d
					Object 1	u11	u12	...	u1d
Object 2	u21	u22	...	u2d
					...	...	...	...	...
Object n	un1	un2		und

Step 10.2, proposes to assume.Assume that h0: kth kind combined method is unrelated with former g kind evaluation methodology；H1: kth kind group Conjunction method is relevant with former g kind evaluation methodology.

Ask spearman coefficient of rank correlation as follows:

$\begin{array}{ccc}{\mathrm{\&rho;}}_{jk}=1-6\frac{\underset{i=1}{\overset{n}{\&sigma;}}(u_{ik}-y_{ij})}{n(n^2-1)},& j=1,2,...,g;& k=1,2,...,d\end{array}---\left(15\right)$

ρ_jkFor the spearman coefficient of rank correlation of jth kind iotave evaluation method and kth kind combination evaluation method, y_ijFor Commented enterprise's ranking results under jth kind original method for i-th.

$\begin{array}{ccc}{\mathrm{\&rho;}}_{kq}=1-6\frac{\underset{i=1}{\overset{n}{\&sigma;}}(u_{ik}-u_{iq})}{n(n^2-1)},& k=1,2,...,d;& q=1,2,...,d\end{array}---\left(16\right)$

ρ_kqFor the spearman coefficient of rank correlation of kth kind and q kind combination evaluation method, u_iqCommented object for i-th Ranking results under q kind combined method.

Step 10.3, constructs statistic t_k, t_kObey the t-distribution that degree of freedom is n-2.

$\begin{array}{ccc}{t}_k={\mathrm{\&rho;}}_k\sqrt{\frac{n-2}{1-{\mathrm{\&rho;}}_k^2},}& k=1,2,...,d,& {\mathrm{\&rho;}}_k=\frac{1}{g}\underset{j=1}{\overset{g}{\&sigma;}}{\mathrm{\&rho;}}_{jk}\end{array}---\left(17\right)$

Wherein, ρ_jkRepresent the spearman coefficient of rank correlation between kth kind combined method and former jth kind method.

After obtaining the evaluation result of single evaluation model and combined evaluation model respectively, applying equation (15) and (16) are entered Row spearman correlation analysiss, such as table 13.

Table 13 combination evaluation results relevance is analyzed

In table 13, * * represents notable under 1% significance level.

Finally, choose t_kIn the maximum, as best of breed method, the evaluation result of the method is exactly whole business standing The end product evaluated.

According to the size of spearman coefficient of rank correlation, select most suitable combination evaluation method result as finally commenting Valency result.The purpose of combination evaluation methods, it is simply that the shortcoming of single evaluation method should be overcome, absorbs several evaluation methods again Advantage.So, though will not be identical between the result of combination evaluation and the result of original multiple methods, should be very close to.Cause This, select and the immediate combined method of original multiple method is best of breed method.Choose t_kIn the maximum, for optimal Combined method, its result is exactly the end product of whole credit rank assessment.

In the embodiment of the present invention, applying equation (17) logistic average combined evaluation model and factorial analyses combination evaluation respectively Statistic under model, uses t respectively₁And t₂Represent, its result is t₁=38.604, t₂=25.743.Take level of significance α= 0.01, marginal value of tabling look-up to obtainObviously t₁And t₂It is all higher thanTherefore two kinds of combined methods and four kinds of single sides Method is closely related, and the t value under arithmetic average combined evaluation model is maximum, and its ranking results is final ranking results.

Based on evaluation of enterprises credit of the present invention, one kind of the present invention is based on subjective and objective tax and weighs Multi-Model Combination The credit rank assessment system of checking, includes distributed data grasping system, evaluation index is screened and standardized module, subjective and objective Combination weights module, the first credit rank assessment module and the second credit rank assessment module.Each ingredient can be passed through Programming realization on computer.

Distributed data grasping system, is commented the information data of enterprise, and acquired data processing is obtained for crawl Quantification of targets data to enterprise.Evaluation index screening and standardized module, for the two-level index in first order calculation index between, Between first class index, the correlation coefficient of first class index and overall objective and significance level, reject the evaluation index being unsatisfactory for requiring, The index filtering out is standardized process.Subjective and objective combination weights module, initially with analytic hierarchy process (AHP) to business standing The index evaluated carries out subjective weights, and next is respectively adopted average variance method, Information Entropy and critic method and carries out Objective Weight, finally Three Objective Weight weight vectors are combined with the weight vectors of subjective weights respectively, obtains optimum combination weight vectors.First Credit rank assessment module, for being entered to business standing using the optimum combination weight vectors that subjective and objective combination weights module obtains Row overall merit, then carries out consistency check to evaluation result using kendall consistency coefficient, eliminates and is unsatisfactory for concordance The result of inspection and corresponding combined method.Second credit rank assessment module, for sieving to the first credit rank assessment module The evaluation result selected, carries out overall merit using arithmetic average built-up pattern and factorial analyses built-up pattern；Then to To comprehensive evaluation result carry out consistency check using spearman coefficient of rank correlation, determine the final of credit rank assessment Result.

Claims (5)

1. a kind of evaluation of enterprises credit based on subjective and objective tax power Multi-Model Combination checking is it is characterised in that include as follows Step:

Step 1, sets up the qualitative evaluation index model of enterprise, has 7 first class index, under each first class index under overall objective There is one group of two-level index, under 7 first class index, have 32 two-level index, obtain the quantized data of all two-level index of enterprise, For the actual index that cannot obtain quantized data, using all candidate enterprises meansigma methodss as this index quantized data； The evaluation index model being adopted is as follows:

First class index: growth power；Two-level index under it has: main operating income, net profit, on-job employee's number, runs total time, Income from main operation rate, Return on Total Assets；

First class index: competitiveness；Two-level index under it has: web flow amount, app flow, search engine index, total number of users, Conversion number of users, retains number of users, number of users market share, every user's average income, competing product quantity, in similar competing product Ranking；

First class index: financing power；Its lower two-level index has: total assets, enterprise's valuation；

First class index: team's power；Its lower two-level index has: postgraduate's number accounting, undergraduate course number accounting, junior college's number accounting, Junior college's following number accounting；

First class index: public opinion power；Its lower two-level index has: news media's concern index, microblogging concern index, wechat concern refers to Number；

First class index: external force；Its lower two-level index has: big industry development temperature, the area grade of enterprise region；

First class index: innovation；Its lower two-level index has: patent of invention quantity, computer copyright quantity, certificate sum, business Mark sum, research staff accounts for employee's proportion；

Step 2, quantitative target is screened；Between the two-level index in first order calculation index, between first class index, first class index with refer generally to Target correlation coefficient and significance level, screen credit rank assessment index；

(1) to each first class index, each two-level index and the multiple correlation organizing other two-level index interior in this first class index are calculated Coefficient and significance level, delete multiple correlation coefficient and are more than threshold₁And significance level is more than threshold₂Two grades Index, is provided with threshold₁Value is not less than 0.9, threshold₂Value is 0.05；

(2) calculate the pearson correlation coefficient between first class index two-by-two and significance level respectively, delete pearson phase relation Number is more than threshold₃First class index, be provided with threshold₃Value is not less than 0.95；

(3) pearson correlation coefficient and the significance level of each first class index and overall objective, gained are calculated respectively Pearson correlation coefficient be on the occasion of first class index and overall objective positive correlation, gained pearson correlation coefficient is the one of negative value Level index is negatively correlated with overall objective；

Step 3, is standardized to the index filtering out processing, if total n enterprise, remaining first class index p after screening, two Level index m, sets up enterprise's efficiency index and quantifies value matrix, element x in matrix_ijRepresent j-th two-level index of i-th enterprise Value, wherein, i=1,2 ..., n, j=1,2 ..., m；If x_ij ^*It is to x using extreme value facture_ijIt is standardized the value obtaining；

Two-level index value under the positively related first class index with overall objective is standardized as:

${x_{ij}}^{*}=0.1+\frac{x_{ij}-\underset{i}{\min }\left\{x_{ij}\right\}}{\underset{i}{\max }\left\{x_{ij}\right\}-\underset{i}{\min }\left\{x_{ij}\right\}}\&times;0.9,i=1,2,...,n;j=1,2,...,m;$

Two-level index value under the negatively correlated first class index with overall objective is standardized as:

${x_{ij}}^{*}=0.1+\frac{\underset{i}{\max }\left\{x_{ij}\right\}-x_{ij}}{\underset{i}{\max }\left\{x_{ij}\right\}-\underset{i}{\min }\left\{x_{ij}\right\}}\&times;0.9,i=1,2,...,n;j=1,2,...,m;$

Step 4, carries out subjective weights using analytic hierarchy process (AHP) to credit rank assessment index；Set up the three-layered node of decision problem Structure, destination layer is Enterprise Integrated developing ability, and rule layer is first class index layer, and a point rule layer is two-level index layer；To rule layer Draw up a questionnaire with a point rule layer, construct the multilevel iudge matrix two-by-two of each layer index, element representation two indices in judgment matrix Importance comparative result；Then the weight vectors of calculation criterion layer and point rule layer；

If the weight vectors w of the rule layer obtaining₀=(w₁ ⁽⁰⁾,w₂ ⁽⁰⁾,…,w_p ⁽⁰⁾), w₁ ⁽⁰⁾,w₂ ⁽⁰⁾,…,w_p ⁽⁰⁾Represent p one-level The weighted value of index；Divide the weight vectors w of rule layer₁=(w₁ ⁽¹⁾,w₂ ⁽¹⁾,…,w_m ⁽¹⁾), w₁ ⁽¹⁾,w₂ ⁽¹⁾,…,w_m ⁽¹⁾Represent m The weighted value of two-level index；

Step 5, carries out Objective Weight to credit rank assessment index；

(1) Objective Weight is carried out using average variance method, if obtain all two-level index representing with respect to the weight vectors of destination layer For: w₂=(w₁ ⁽²⁾,w₂ ⁽²⁾,…,w_m ⁽²⁾)；

(2) Objective Weight is carried out using Information Entropy, obtain all two-level index and be expressed as with respect to the weight vectors of destination layer: w₃ =(w₁ ⁽³⁾,w₂ ⁽³⁾,…,w_m ⁽³⁾)；

(3) Objective Weight is carried out using critic method, obtains all two-level index and be expressed as with respect to the weight vectors of destination layer: w₄=(w₁ ⁽⁴⁾,w₂ ⁽⁴⁾,…,w_m ⁽⁴⁾)；

Step 6, carries out combination weights；

The weight vectors w that three Objective Weight weight vectors are obtained with analytic hierarchy process (AHP) respectively₁Combination, obtains optimum combination power Weight vector, the optimum combination weight vectors being obtained and weight vectors w₁With the sum of square of deviations of Objective Weight weight vectors Little；If the optimum combination weight vectors w that analytic hierarchy process (AHP) and average variance method obtain₅=(w₁ ⁽⁵⁾,w₂ ⁽⁵⁾,…,w_m ⁽⁵⁾)；Level divides The optimum combination weight vectors w that analysis method and Information Entropy obtain₆=(w₁ ⁽⁶⁾,w₂ ⁽⁶⁾,…,w_m ⁽⁶⁾)；Analytic hierarchy process (AHP) and critic method The optimum combination weight vectors w obtaining₇=(w₁ ⁽⁷⁾,w₂ ⁽⁷⁾,…,w_m ⁽⁷⁾)；

Step 7, carries out overall merit using optimum combination weight vectors to business standing, as follows:

$z_i^{\left(k\right)}=\underset{j=1}{\overset{m}{\mathrm{\&sigma;}}}{w}_j^{\left(k\right)}{x}_{\mathrm{ij}}i=\mathrm{1,2},\&centerdot;\&centerdot;\&centerdot;,n;k=\mathrm{5,6,7}$

Enterprise is commented to pass through weight vectors w for i-th_kComprehensive evaluation value on whole m two-level index；

Step 8, the credit rank assessment three kinds of combined methods obtaining in step 7 being obtained using kendall consistency coefficient Result carries out consistency check, eliminates the method being unsatisfactory for consistency check；Three kinds of combined methods refer to: analytic hierarchy process (AHP) with all The combined method of the combined method of the combined method of variance method, analytic hierarchy process (AHP) and Information Entropy, analytic hierarchy process (AHP) and critic method；

Step 9, is combined to the method after consistency check in step 8 evaluating business standing, using the following two kinds side Method is carrying out；

(1) arithmetic average built-up pattern is adopted to evaluate business standing, specifically: set y_ijRepresent and commented enterprise for i-th in jth kind Ranking value under evaluation methodology, i=1,2 ..., n, j=1,2 ..., g, g are evaluation methodology number；First will with Ordering and marking method The ranking of every kind of method sequence is converted into integer, if y_ijCorresponding must be divided into r_ij, then calculate distinct methods under score average Value, then by meansigma methodss, each enterprise resequences；If the average of You Liangge enterprise is identical, calculates and obtain under distinct methods The standard deviation divided, the wherein little person of standard deviation are excellent；

(2) factorial analyses built-up pattern is adopted to evaluate business standing, specifically: to set n and commented score under g kind method for the enterprise Vector iss_ijIt is to be commented score under jth kind method for the enterprise, s i-th_jIt is n to be commented Score vector under jth kind method for the enterprise；If f=is (f₁,f₂,...,f_l)^tBe g kind method common factor vector, public because Sub- number is l；First, set up the linear relationship between s and f: s^t=af+e, a be factor loads matrix, e be specific factor to Amount；Then, the initial factor, certainty factor number and recognition factor are extracted；From PCA extraction factor；According to feature Value criterion certainty factor number；From the variance rule for the treatment of recognition factor in orthogonal rotation；Finally, calculate and commented enterprise j-th Final comprehensive evaluation valueWherein v is the factor number extracted, β_iFor the value of i-th factor, ψ_iFor side Difference contribution rate；

Step 10, carries out consistency check using the combination evaluation result that spearman coefficient of rank correlation obtains to step 9, really Determine the final result of credit rank assessment.

2. a kind of evaluation of enterprises credit based on subjective and objective tax power Multi-Model Combination checking according to claim 1, It is characterized in that, in described step 6, ask for the optimum combination weight vectors w that analytic hierarchy process (AHP) is obtained with average variance method₅When, Construction optimizing model is as follows:

$\left\{\begin{array}{c}min\underset{k=1}{\overset{2}{\&sigma;}}\underset{i=1}{\overset{m}{\&sigma;}}{(w_i^{\left(5\right)}-w_i^{\left(k\right)})}^2\\ \begin{array}{cc}s.t.& \underset{i=1}{\overset{m}{\&sigma;}}{w}_i^{\left(k\right)}=1,k=0,1,2\end{array}\end{array}\right.$

Using lagrange's method of multipliers, the unique solution trying to achieve above formula is:

$w_i^{\left(5\right)}=\frac{1}{2}(w_i^{\left(1\right)}+w_i^{\left(2\right)})+\frac{1}{p}(1-\frac{1}{2}\underset{i=1}{\overset{m}{\&sigma;}}(w_i^{\left(1\right)}+w_i^{\left(2\right)}\left)\right),i=1,2,...,m;$

Equally, construction optimizing model solves and obtains:

$w_i^{\left(6\right)}=\frac{1}{2}(w_i^{\left(1\right)}+w_i^{\left(3\right)})+\frac{1}{p}(1-\frac{1}{2}\underset{i=1}{\overset{m}{\&sigma;}}(w_i^{\left(1\right)}+w_i^{\left(3\right)}\left)\right),i=1,2,...,m;$

$w_i^{\left(7\right)}=\frac{1}{2}(w_i^{\left(1\right)}+w_i^{\left(4\right)})+\frac{1}{p}(1-\frac{1}{2}\underset{i=1}{\overset{m}{\&sigma;}}(w_i^{\left(1\right)}+w_i^{\left(4\right)}\left)\right),i=1,2,...,m.$

3. a kind of evaluation of enterprises credit based on subjective and objective tax power Multi-Model Combination checking according to claim 1, It is characterized in that, in described step 8, calculate kendall consistency coefficient t as follows:

If being commented enterprise to evaluate with g kind method to n, y_ijRepresent and commented row under jth kind evaluation methodology for the enterprise i-th Sequence value, 1≤y_ij≤ n, i=1,2 ..., n, j=1,2 ..., g；Calculated according to formula below:

Wherein

Construction statistic q:q=g (n-1) t, q approximately obey the χ that degree of freedom is n-1²Distribution.

4. a kind of evaluation of enterprises credit based on subjective and objective tax power Multi-Model Combination checking according to claim 1, It is characterized in that, in described step 10, the method carrying out consistency check to the combination evaluation result in step 9 is:

Step 10.1, if step 9 is to g kind, and former method carries out d kind combination, combination evaluation result is converted into ranking value, u_ikRepresent Commented ranking value under the combination of kth kind for the enterprise, i=1,2 .., n i-th；K=1,2 ..., d, 1≤u_ik≤ n, d are combination Method number；

Step 10.2, proposes to assume；Assume that h0: kth kind combined method is unrelated with former g kind evaluation methodology；H1: kth kind combination side Method is relevant with former g kind evaluation methodology；Ask spearman coefficient of rank correlation as follows:

The spearman coefficient of rank correlation ρ of the former evaluation methodology of jth kind and kth kind combination evaluation method_jkFor:

${\mathrm{\&rho;}}_{jk}=1-6\frac{\underset{i=1}{\overset{n}{\&sigma;}}(u_{ik}-y_{ij})}{n(n^2-1)},j=1,2,...,g;k=1,2,...,d;$

y_ijCommented enterprise's ranking results under the former method of jth kind for i-th；

The spearman coefficient of rank correlation ρ of kth kind and q kind combination evaluation method_kqFor:

${\mathrm{\&rho;}}_{kq}=1-6\frac{\underset{i=1}{\overset{n}{\&sigma;}}(u_{ik}-u_{iq})}{n(n^2-1)},k=1,2,...,d;q=1,2,...,d;$

Step 10.3, constructs statistic t_k, t_kObey the t-distribution that degree of freedom is n-2, as follows:

Wherein

Finally, choose t_kIn the maximum, as best of breed method, the evaluation result of best of breed method is exactly credit rank assessment End product.

5. a kind of described commented based on the subjective and objective business standing assigning power Multi-Model Combination checking based on Claims 1 to 4 is arbitrary The credit rank assessment system of valency method is it is characterised in that this system includes following ingredient:

Distributed data grasping system, is commented the information data of enterprise, and acquired data processing is looked forward to for crawl The quantification of targets data of industry；

Evaluation index screening and standardized module, for the two-level index in first order calculation index between, between first class index, one-level refers to The correlation coefficient of mark and overall objective and significance level, reject the evaluation index being unsatisfactory for requiring, the index filtering out are entered Row standardization；

Subjective and objective combination weights module, carries out subjective weights initially with analytic hierarchy process (AHP) to the index of credit rank assessment, its Secondary average variance method, Information Entropy and the critic method of being respectively adopted carries out Objective Weight, finally divides three Objective Weight weight vectors Do not combine with the weight vectors of subjective weights, obtain optimum combination weight vectors；

First credit rank assessment module, for the optimum combination weight vectors that obtained using subjective and objective combination weights module to enterprise Industry credit carries out overall merit, then carries out consistency check to evaluation result using kendall consistency coefficient, eliminates discontented The result of sufficient consistency check and corresponding combined method；

Second credit rank assessment module, for the evaluation result that the first credit rank assessment module is filtered out, using counting Meansigma methodss built-up pattern and factorial analyses built-up pattern carry out overall merit；Then the comprehensive evaluation result obtaining is utilized Spearman coefficient of rank correlation carries out consistency check, determines the final result of credit rank assessment.