The content of the invention
The present invention, which provides one kind, can be based on enterprise and employee's risk conduct the relation, be capable of the enterprise of accurate evaluation business risk
Industry appraisal procedure, comprises the following steps,
Nodal information is obtained, establishes risk conducting networks;
Obtain the risk initial value of node;
Definition node risk propagation function and relational risk propagation function;
Training risk conduction model.
Further, the risk conduction model is multiple that each risk conduction model includes a Centroid;
The risk conduction model is multilayer network structure;
It establishes risk conducting networks to specifically include, risk conduction is established according to the equity relation between node and personnel's relation
Network;
Nodal information includes the one or more in the equity information, personal information, operation information of node.
Further, the risk initial value of node is obtained, including the news information according to node, financial information manages letter
One or more acquisition risk initial values in breath.
Further, definition node risk propagation function and relational risk propagation function include,
Definition node B risk propagation functions RAB=fAB(rAB),For the value-at-risk that node A is inputted to node B, fABFor section
Point risk propagation function, RABThe risk exported for node A to the node B that the node B value-at-risks inputted are brought;
Relational risk propagation function r between definition node A and node BAB=FAB(RA), rABIt is inputted for node A to node B
Value-at-risk, FABIt is A to the relational risk propagation function of B, RAFor the risk of node A outputs;
Complete each node risk propagation function and the definition of each relationships between nodes risk propagation function in risk conducting networks.
Further, training risk conduction model includes,
To each risk conduction model, between the risk initial value, each node risk propagation function, each node according to each node
Relational risk propagation function calculates Centroid value-at-risk;
Obtain the AUC value of the set of value-at-risk obtained using each node as Centroid, adjust model parameter until
It obtains AUC value and is more than preset value and record cast parameter.
In order to ensure the implementation of the above method, the present invention also provides a kind of enterprise evaluation systems based on risk conduction model
System, including with lower unit,
Risk conducting networks establish unit, for obtaining nodal information, establish risk conducting networks, obtain the risk of node
Initial value;
Risk conduction model establishes unit, for definition node risk propagation function and relational risk propagation function;
Training unit, for training risk conduction model.
Further, the risk conduction model is multiple that each risk conduction model includes a Centroid;
The risk conduction model is multilayer network structure;
Risk conducting networks, which establish unit and establish risk conducting networks, to be specifically included, according to the equity relation between node and
Personnel's relation establishes risk conducting networks;
Nodal information includes the one or more in the equity information, personal information, operation information of node.
Further, risk conducting networks establish the risk initial value that unit obtains node, including the news according to node
Information, financial information, one or more in operation information obtain risk initial values.
Further, risk conduction model establishes unit definition node risk propagation function and relational risk propagation function bag
It includes,
Definition node B risk propagation functions RAB=fAB(rAB),For the value-at-risk that node A is inputted to node B, fABFor section
Point risk propagation function, RABThe risk exported for node A to the node B that the node B value-at-risks inputted are brought;
Relational risk propagation function r between definition node A and node BAB=FAB(RA), rABIt is inputted for node A to node B
Value-at-risk, FABIt is A to the relational risk propagation function of B, RAFor the risk of node A outputs;
Complete each node risk propagation function and the definition of each relationships between nodes risk propagation function in risk conducting networks.
Further, training unit training risk conduction model includes,
To each risk conduction model, between the risk initial value, each node risk propagation function, each node according to each node
Relational risk propagation function calculates Centroid value-at-risk;
Obtain the AUC value of the set of value-at-risk obtained using each node as Centroid, adjust model parameter until
It obtains AUC value and is more than preset value and record cast parameter.
The beneficial effects of the invention are as follows:
The present invention obtains preferably model parameter, Jin Erke by establishing network risks conduction model, and to model iteration
During business risk is assessed, to obtain accurate business risk value.
Specific embodiment
As shown in Figure 1, the present invention provides a kind of enterprise evaluation method based on risk conduction model, comprise the following steps,
Nodal information is obtained, establishes risk conducting networks;
Obtain the risk initial value of node;
Definition node risk propagation function and relational risk propagation function;
Training risk conduction model.
Further, the risk conduction model is multiple that each risk conduction model includes a Centroid;
The risk conduction model is multilayer network structure;
It establishes risk conducting networks to specifically include, risk conduction is established according to the equity relation between node and personnel's relation
Network;
Nodal information includes the one or more in the equity information, personal information, operation information of node.
One business connection network can be established by equity relation between enterprise, shareholder, employee relationship, for each
A enterprise can establish a risk conducting networks figure centered on it, and it is network first tier to have direct relation with it, remove
First layer and center enterprise can be considered as the second layer with what first layer had a direct relation, and so on can establish with centromere
Multitiered network relational graph centered on point;
Conducting calculating in risk, the enterprise of the invention for defining outer layer can be still interior with inner layer enterprise conduction risk in the process
Risk is not conducted in outer layers enterprise for layer enterprise.
Further, the risk initial value of node is obtained, including the news information according to node, financial information manages letter
One or more acquisition risk initial values in breath.
The company information that the risk initial value of nodal information can be captured by network obtains, such as new from news website crawl
It hears, from sale website such as day cat or 1688 crawl operation information, captures equity relation information from industrial and commercial website and shareholder holds a post and closes
It is information, employee relationship's information is captured from professional social network sites such as Linkedin, and passes through information and assignment is carried out to node risk,
Assignment can have artificial progress, can also bring the calculating of assignment model by computer, assignment scope is between 0-100.
Marketing risk value can be defined in an embodiment of the present invention is
XS=K*XJ
Wherein XS is marketing risk value, and K is marketing risk model coefficient, and XJ declines numerical value for sales volume ring ratio, and K is risk
One parameter of conduction model, as the target of final optimization pass, constantly adjusts in model iterative process.
Profit value-at-risk can be defined in an embodiment of the present invention is
LR=T*KS
Wherein LR is profit value-at-risk, and T is profit risk model coefficient, and KS is amount of loss, and T is the one of risk conduction model
A parameter as the target of final optimization pass, constantly adjusts in model iterative process.
Further, definition node risk propagation function and relational risk propagation function include,
Definition node B risk propagation functions RAB=fAB(rAB),For the value-at-risk that node A is inputted to node B, fABFor section
Point risk propagation function, RABThe risk exported for node A to the node B that the node B value-at-risks inputted are brought;
In an embodiment of the present inventionThe risk for losing backward node B conduction is sent for node A, it is clear that RABWith node B
Management data especially total profit have apparent relation.Therefore the present invention mirrors parameter W in this each node function
W is a relevant parameter of management data with B, be risk conduction model a parameter as final optimization pass
Target constantly adjusts in model iterative process.
T is time variable, and α is fAB(rAB) include one with the parameter of time correlation, reaction A sends the backward section of loss
The risk of point B conduction changes over time, and α is target of the parameter of risk conduction model as final optimization pass, is changed in model
Constantly adjustment during generation.
The relation that risk changes over time has been reacted in the addition of alpha parameter, more accurately embodies the risk between node
Conduct the relation.
Relational risk propagation function r between definition node A and node BAB=FAB(RA), rABIt is inputted for node A to node B
Value-at-risk, FABIt is A to the relational risk propagation function of B, RAFor the risk of node A outputs;
Company B holds the 20% of the total shares of company A in an embodiment of the present invention,
Company A, which goes into the red, quantifies value-at-risk as both RA=100, then the value-at-risk that company A is exported to company B
rAB=100*0.2=20
In another embodiment, the president of company B company A hold a post general manager, this relation chain risk conduct
Relation is apparent unlike equity relation, therefore the present invention is in FAB() introduces a parameter Q, and company A, which goes into the red, quantifies value-at-risk
For both RA=100, then the value-at-risk that company A is exported to company B
rAB=Q*100
Q is target of the parameter of risk conduction model as final optimization pass, is constantly adjusted in model iterative process.
Complete each node risk propagation function and the definition of each relationships between nodes risk propagation function in risk conducting networks.
The risk that a certain node externally exports in risk conduction model passes through for the initial risks of itself and other nodes
Relational risk propagation function summarizes total to the risk that the risk that it is conducted is exported into the node risk propagation function for crossing the node
Value.
Further, training risk conduction model includes,
To each risk conduction model, between the risk initial value, each node risk propagation function, each node according to each node
Relational risk propagation function calculates Centroid value-at-risk;
Fig. 4 is one embodiment of the invention risk conducting networks schematic diagram, and node centered on A in figure, BCD is risky with A
The second layer network node of relation, EFGH are the third layer network node with the risky relation of second layer network node, are being calculated
During the risk of node A, using following steps:
The first step:It brings the initial risk values of third layer network node EFGH into, and passes through each node of third layer network and
Relational risk propagation function between each node of double layer network calculates the risk that third layer network is conducted to second layer network B CD,
And BCD receives the risk exported after these risks by respective node risk propagation function.
Second step:The risk that second layer network B CD in the first step is exported by respective node risk propagation function and its
The risk of itself calculates each second of network node and is passed by the risk between each node of the second layer network and Centroid after summarizing
The risk and Centroid that derived function is conducted to Centroid receive the section that each node input risk of the second layer passes through Centroid
Put the Centroid overall risk after the total value that risk propagation function exports summarizes with Centroid initial risks.
Centered on each node, a risk relations conducting networks can be established, with reference in risk relations network
Relational risk propagation function between the node risk propagation function and node of each node forms risk conduction model, will be each
The initial value of node risk brings after risk network model the value-at-risk that Centroid can be calculated into.
Obtain the AUC value of the set of value-at-risk obtained using each node as Centroid, adjust model parameter until
AUC value is more than preset value and record cast parameter.
Adjust the ginseng for the relational risk propagation function that model parameter is included between adjustment node risk propagation function and node
The parameter of number and the function that each node initial risk values are quantified.
AUC (Area Under roc Curve) is an a kind of standard for being used for measuring disaggregated model quality, and AUC value is
The region area that ROC curve is covered, it is clear that AUC is bigger, and grader classifying quality is better.
AUC=1 is perfect grader, during using this prediction model, no matter set what threshold value can draw it is perfect pre-
It surveys.The occasion of overwhelming majority prediction, there is no perfect graders.
0.5<AUC<1, better than random guess.This grader (model) properly if given threshold, can there is predictive value.
The AUC=0.5, (example as random guess:Lose copper coin), model is not previously predicted value.
AUC<0.5, it is also poorer than random guess;But as long as always instead predicting and going, just it is better than random guess.
Preset value is pre-set by operating personnel according to project demand, is constantly adjusted risk conduction parameter, is realized AUC value
It continues to optimize, when AUC value is greater than or equal to preset value, represents risk conduction model and reached default quality requirement.At this time
Record each parameter in risk conduction model.
When needing to calculate Target Enterprise risk, node establishes risk conduction model centered on Target Enterprise, and brings into
The risk of the risk initial value of network Zhong Ge enterprises and the model parameter calculation Target Enterprise of record.
Risk conduction model the present invention is based on Complex Networks Theory be can study in the network that enterprise and people are formed when
The influence that other nodes in network are generated after some node occurrence risks.
The risk of enterprise is more than the business circumstance correlation with itself, is also looked forward to his direct correlation enterprise and indirect association
Industry is related.An extremely complex network can be formed between enterprise and enterprise, between enterprise and people.The present invention looks forward to the whole nation
Industry, enterprise administrator and the relation between them are configured to a very huge network, possess about 1,000,000,000 nodes,
About 4,000,000,000 sides.For the network of such complexity, the theory and technology research that the present invention is conducted using complex network and risk is looked forward to
Risk transmission relation between industry and enterprise quantifies an enterprise and generates risk to the much influences of other enterprises generation.
The present invention can be realized from Research on network structure business risk.
The present invention can realize the angle research business risk conducted from risk.
The present invention establishes multi-level network model, quantify after network interior joint occurrence risk to other directly or
The influence for the node being indirectly connected with.The foundation of risk conduction model solves business risk assessment risk and quantifies, and risk is at any time
Between variation and node between risk transmission the problems such as.