CN106327320A - Price mismatching tax evasion behavior identification method based on tax payer benefit association network - Google Patents

Abstract

The invention discloses a price mismatching tax evasion behavior identification method based on a tax payer benefit association network. The price mismatching tax evasion behavior identification method uses data of a tax bureau, combines with inspection cases, brings forwards a price mismatching tax evasion behavior, extracts a price mismatching group, and finally positions an enterprise associated with the tax evasion. The price mismatching tax evasion behavior identification method solves a problem that the tax evasion between associated enterprises is hard to find.

Description

Price mispairing based on taxpayer's interests related network is evaded the tax Activity recognition method

Technical field:

The present invention relates to a kind of based on taxpayer's interests related network (Taxpayer Profit Interactive Network, TPIN) price mispairing evades the tax Activity recognition method, knows for solving the behavior of evading the tax between current affiliated enterprise Not difficulty, the problem that Tax Check inefficiency, ways of check are single.

Background technology:

Along with the foundation of tax information platform, tax data amount presents the trend of quickly growth.But the intelligence of taxation supervision Energyization but yet suffers from deficiency, and China's tradition tax audit and inspection artificially analyze object independently to pay taxes, dependence manually select case, Computer selects the approach such as case, report, and efficiency is low, the degree of depth and range critical constraints, it is difficult to identify in interests associations group with interests The behavior of evading the tax that conveying is characterized.Transaction tax rate difference is wherein utilized to carry out profit shifting to reach tax evasion by price mispairing The mode of tax is the most common, and two taxpayers of the direct or indirect incidence relation of this existence carry out inhomogeneity with third party simultaneously The sale of type or when buying transaction, by adjusting the transfer price between different type of transaction, it is achieved the reduction of overall burden, from And the purpose behavior that reaches to evade the tax is a difficult problem for China's Tax Check.Therefore, by tax data mining identification and analysis price The mispairing behavior of evading the tax has become as the emphasis of taxation authority's present stage research and a difficult problem urgently to be resolved hurrily.

For how efficiently identifying price mispairing evading the tax behavior, following patent and paper provide technical scheme:

Chinese patent literature 201310293435.5 discloses a kind of tax evasion based on taxpayer's interests related network model The customhouse's connection Corporate Identity method, it is provided that taxpayer's interests of a kind of oriented closed loop collection based on maximum betweenness constraint associate group Whether discrimination method, associate to exist in group according to the trading activity weighted judgment taxpayer's interests between enterprise simultaneously and evade the tax Suspicion；

Document " correlation rule data mining application in the tax inspection system " (Xu Shengang, economic supervision 2011,13, Pp:43-44) use association rule mining to evade the tax enterprise's linked character potential in case, provide certainly for Tax Check personnel Plan is supported.

But method described in document above is primarily present problems with: 1, document 1 associates with enterprise about enterprise attributes There is distortion phenomenon in the data of relation, it is impossible to is applicable to taxpayer's interests related network modeling of price mispairing pattern；Based on Macro ring betweenness algorithm is not associated with practical business, it is impossible to be applied to the Activity recognition of evading the tax of price mispairing pattern.2、 Based on association rule mining in document 2, it is only capable of identifying the incidence relation existed between illegal and undisciplined means of evading the tax, it is impossible to application Mode excavation in taxpayer's interests related network.

Summary of the invention:

It is an object of the invention to price mispairing based on taxpayer's interests related network evade the tax Activity recognition method, should Method utilizes tax bureau's data, in conjunction with inspection case, proposes price mispairing and evades the tax behavior, by message passing mechanism and static state Incidence relation chain extraction price mispairing group, evades the tax affiliated enterprise based on association tax index location.Finally solve association Evade the tax between enterprise the indiscoverable problem of behavior.

For reaching above-mentioned purpose, the present invention adopts the following technical scheme that and is achieved:

Price mispairing based on taxpayer's interests related network is evaded the tax Activity recognition method, and the method utilizes tax inning According to, in conjunction with inspection case, proposing price mispairing and evade the tax behavior, and extract price mispairing group, evade the tax association in final location Enterprise.

The present invention is further improved by, and specifically includes following steps:

1) first, taxpayer's interests related network is built；

2) secondly, evade the tax behavioral pattern based on inspection case extraction price mispairing；

3) last, based on 1) in taxpayer's interests related network and 2) in price mispairing evade the tax behavior pattern recognition price Mispairing is evaded the tax behavior.

The present invention is further improved by, and builds taxpayer's interests related network, specific as follows:

Taxpayer's interests related network represents the contact between enterprise and enterprise, investor and legal representative, by entity, The network that the relation of inter-entity, entity attribute and attribute of a relation four key element enough become；Wherein, entity includes enterprise, investor, method People represents, and the relation of inter-entity includes transaction relationship, investment relation, control planning, and entity attribute includes enterprise name, industry class Not, attribute of a relation includes ratio between investments, dealing money；

Taxpayer's interests related network is expressed as two tuples:

TPIN={ (V, VD), (E, ED) }

Wherein V={v_p| p=1,2 ..., N_pRepresent node set, wherein N_pRepresent the node number in network, each Node is uniquely identified by the numeral of 1 to p, all set that there is limit in E expression figure, and makes E={e_pq}={ (v_p,v_q) | 0 ＜ P, q ＜ N_p, wherein e_pq=(v_p,v_q) representing the oriented line from the node to the node being numbered q being numbered p, VD is expressed as Nodal community, is denoted as:

VD={Type, ID, Name}

WhereinRepresenting the type of node, Vnsr represents taxpayer, Vfddbr representation Determining representative, Vtzf represents investor, represents with triad number respectively, binary number 001, and decimal number 1 represents pays taxes People；Binary number 010, decimal number 2 represents legal representative；Binary number 100, decimal number 4 represents investor, if joint Point has multiple types and then carries out binary system or computing；ID={sfzh, nsrsbh} represent the identification number of node, are enterprise and method The unique identifying number that people represents, wherein enterprise Taxpayer Identification Number nsrsbh uniquely identifies, and legal representative is by identification card number Sfzh uniquely identifies；Name represents the Chinese of node, and the attribute list on limit is shown as:

ED={CT, IV, TD}

Wherein CT={w_pq| 0 ＜ p, q ＜ N_pRepresent legal person and the control weight on limit, the human world of paying taxes, only comprise two kinds of controls Relation { controls, do not control }, IV={w_pq| 0 ＜ p, q ＜ N_p∈ (0,1] represent limit (v_p,v_q) investment weight, for investor v_p Place enterprise v_qThe size of control ratio, TD={w_pq| 0 ＜ p, q ＜ N_p∈ (0,1] represent limit (v_p,v_q) transaction weight, for Enterprise v_pWith enterprise v_qTurnover accounts for enterprise v_qRatio.

The present invention is further improved by, based on inspection case price mispairing evade the tax behavioral pattern extraction, specifically As follows:

From business case, taking out the cohort model with tax evasion suspicion, obtain point without concrete business and The topology diagram on limit, makes the excavation of the pattern of evading the tax is converted into the search of incidence relation chain in TPIN.

The present invention is further improved by, and price mispairing is evaded the tax behavior pattern recognition, and concrete grammar is as follows:

Evade the tax enterprise based on TPIN and price mispairing behavior pattern recognition of evading the tax, be divided into four parts, first, set up figure Message propagation model RHSF, extracts static association relation chain secondly based on RHSF, excavates valency based on static association relation chain afterwards Lattice mispairing suspicion group, finally carries out tax index calculating and analysis, and exports the enterprise that evades the tax, the most such as suspicion enterprise Under:

(1) RHSF message mechanism of transmission builds

Based on RHSF figure message propagation model, RHSF model uses the figure computation model centered by node, node updates The processing stage that function being divided into four continuous print: information collecting step Receive, signal processing stages Handle, information are distributed Stage Send and filtration stage Filter, wherein Receive/Send, be called for short RS, and function is with single edge for operation granularity, Handle/Filter function is with single node for operation granularity, as a example by node i, the step of RHSF model is described:

Step1: information collecting step, is led to all of its neighbor node of i-node by one with the information on the limit being connected Collect by self-defining function:

$\mathrm{\Σ}\←{\⊕}_{v\∈b\[i\]}R({\mathrm{Attr}}_i,{\mathrm{Attr}}_{(i,v)},{\mathrm{Attr}}_v)$

Wherein Attr_i、Attr_v、Attr_(i,v)Be respectively node i, node v and from node i to the limit of node v letter Breath, v ∈ b [i] represents all nodes being connected with node i, and R () represents the reception process of message, User-Defined FunctionsCome The definition process to message preliminary treatment；Σ means that node receives the information on adjacent node and limit the knot of preliminary pretreatment Really；

In Step2: the Information application stage, the Σ receiving in Step1 and calculating being applied to node i, change node is certainly The attribute of body:

${\mathrm{Attr}}_i^{new}\←H({\mathrm{Attr}}_i,\mathrm{\Σ})$

WhereinProperty value after updating for node i, H () represents the processing procedure of message；

Step3: information distribution phase, by information new on i-node and all initial informations being connected on limit with node i It is distributed to other adjacent nodes by User-Defined Functions:

$\∀v\∈b\[i\]:\left({\mathrm{Attr}}_v\right)\←S({\mathrm{Attr}}_i^{new},{\mathrm{Attr}}_{(i,v)})$

WhereinRepresenting and travel through each node being connected with node i, S () represents the distribution procedure of message；

Step 4: filtration stage, by the node required for User Defined Rules Filtering next round iteration and limit, deletes Epicycle iteration not receiving node and its limit being connected of message, whole graph structure being updated, if figure still having surplus Remaining node, return Step1:

Graph_new←F(vertices,edges)

Wherein vertices represents Step3 Point Set, and edges represents the limit collection in Step3, and F () represents filter process, Graph_newRepresent the graph structure after updating；

(2) static association relation chain extraction

First two concepts are defined:

1) static association relational network

Static association relational network is the sub-network only comprising investment and control planning in taxpayer's interests related network；

2) static association relation chain

Static association relation chain is to control, invest the collection of end to end path, limit chain in static association relational network Close, if path chain here is defined as investment enterprise C₁, C here_iRepresent enterprise, indirectly controlled by one or many enterprises or Investment enterprise C_n+1, then C is claimed₁And C_n+1Between end to end path be path chain, and useRepresent a static state Incidence relation chain；

Static association relation chain is the mostly important ingredient being carried behavioral pattern topological model of evading the tax, therefore, logical Crossing periodically structure off-line static association relation chain storehouse to avoid repeating to travel through static association relational network, its building process is as follows:

Step1: based on static association Relation extraction static association relational network from TPIN；

Step2: for node v each in network_pAdd local path aggregate attribute LP_p, wherein, p=1 ..., N_P, v_pRepresent Node, LP_pRepresent node v_pIn local path attribute, its structure is Set [Seq [P]], and wherein Set represents set, uses symbol { } represents, Seq represents sequence, represents with symbol<>, and P represents that the element type of sequence is node serial number type；

Step3: for each node v_pA path sequence ＜ v only comprising own node numbering is added in local path set_p ＞, the most each node v_pLocal path set is initialized as { ＜ v_p＞ }, meanwhile, it is empty for initializing global path set GP in network；

Step4: based on RHSF message distribution mechanism, each node v in network_pAlong static association relation edge direction, by v_p's Current local path set is sent to its adjacent node v_ps(1),v_ps(2),...,v_psM (), wherein m represents with v_pFor initial joint The quantity of the adjacent node of point, v_psI () represents v_pNode passes through static association relation chainThe all of its neighbor node being connected, Wherein, 0 ＜ i≤m；

Step5: based on RHSF message collection and treatment mechanism, each node v in network_pTo the local path collection received Close LP_pr(1),LP_pr(2),...,LP_prN () does union, wherein n is with v_pFor the quantity of the adjacent node of terminal node, obtain Set of paths LP_pr, travel through LP_prIn each paths, if path comprises current vertex self numbering p, then delete this path；No Then, adding current vertex self numbering p at this end, path, the set of paths finally given is designated as LP '_pr；

Step6: by each node v_pLocal path set be newly defined as LP '_pr, and by LP '_prAdd global path to In set GP；

Step7: based on RHSF strobe utility, deletes the node being not received by message and the limit with it as source point, it is judged that Current static incidence relation nodes quantity, if number of nodes is zero, then terminates,

After above-mentioned steps terminates, output global path set is static association relation chain storehouse；

(3) price mispairing pattern extraction

First related notion is defined:

1) dynamic associations network

Dynamic associations network refers to only comprise in taxpayer's interests related network the sub-network of transaction relationship；

2) dynamic associations chain

Dynamic associations chain refers to the transaction limit in dynamic associations network, and comprises static association in nodal community Relation chain set；

3) reverse bilateral dynamic associations chain

Reverse bilateral dynamic associations chain is by pointing to same node or point out by same node two dynamically associate The set of relation chain；

Being mainly characterized by of price mispairing behavioral pattern comprises a reverse bilateral dynamic associations chain, its pattern recognition Core be excavate two Opposite direction connections dynamic associations limit, and identify be connected limit two ends whether there is " static association Relation chain or bidirectional static incidence relation chain "；

The step of price mispairing behavior patterns mining is as follows:

Step1: extract dynamic associations network, its node serial number set based on dynamic associations extraction from TPIN For Q；

Step2: to global path set GP, wherein GP is static association relation storehouse, compiles according to each path termination in set Number q is polymerized, and obtains terminal numbering and key-value pair (q, the path of set of paths composition_q), wherein q ∈ Q, path_qFor terminal q Corresponding set of paths；

Step3: for node v each in network_qAdd local path attribute LP_q, its value is set to path_q；

Step4: send mechanism, each node v in network based on RHSF message_qBy dynamic associations limit by its local road Footpath information LP_qIn conjunction with the flag bit flag representing edge direction_q, it is combined into message Msg_qPass to adjacent node, flag_qRepresent two Direction similarities and differences flag bit between bar dynamic associations chain, if direction of transfer is identical with edge direction, then flag_qIt is set to 1；Instead It, be set to-1；

Step5: each node v in network_qAccording to flag_qBe polymerized, obtain two massage set MsgSet1 and MsgSet2:

MsgSet1={Msg_i|flag_i=1, i ∈ Q}

MsgSet2={Msg_j|flag_j=-1, j ∈ Q}

Step6: arbitrary Msg_i∈ MsgSet1, Msg_j∈ MsgSet1, it is assumed that Msg_iAnd Msg_jCorresponding set of paths is divided Wei MP_iAnd MP_j,Wherein MP_iAnd MP_jIt is all MsgSet1 In the set in all paths,Representing with p as start node, r is the static association relation chain of terminal node, Representing with t as start node, s is the static association relation chain of terminal node, if there is p=t, then static association relation chainWithIn conjunction with dynamic associations limitWithConstitute the suspicion group of a price mispairing transaction；

Step7: arbitrary Msg_i∈ MsgSet2, Msg_j∈ MsgSet2, it is assumed that Msg_iAnd Msg_jCorresponding set of paths is divided Wei MP_iAnd MP_j,Wherein MP_iAnd MP_jIt is all MsgSet2 In the set in all paths,Representing with p as start node, r is the static association relation chain of terminal node, Representing with t as start node, s is the static association relation chain of terminal node, if there is p=t, then static association relation chainWithIn conjunction with dynamic associations limitWithConstitute the suspicion group of a price mispairing transaction；

(4) affiliated enterprise that evades the tax based on tax index identifies

3. the price mispairing suspicion group obtained step proceeds as follows:

Step1: extract the node that type type is taxpayer nsr, its node from all of price mispairing suspicion group Collection is combined into M；

Step2: to each node i (i ∈ M), calculates three the specific tax indexs relevant to price mispairing pattern, respectively For taxpayer's stock turnover rate and income from sales rate of change coefficient of elasticity, be designated as ISEC, enterprise's end of term accounts receivable rate of change with Income from sales rate of change coefficient of elasticity, is designated as ASEC, taxpayer's period expense rate of change and main business income rate of change elasticity Coefficient, is designated as PMEC；

Wherein the formula of ISEC, ASEC, PMEC is respectively as follows:

Step 3: be analyzed the index calculated, utilizes following span to evaluate these business indicatorses, and to often Individual enterprise i forms corresponding result of determination, if desired value is in the range of " existing the behavior of evading the tax ", then the existence of this enterprise is evaded the tax Behavior, if desired value is in the range of " in the range of belonging to warning ", then this enterprise evades the tax suspicion, if desired value " is just belonging to In the range of Chang " in the range of, then this enterprise is without the suspicion of evading the tax:

Wherein, IT represents that stock turnover rate, ST represent the income from sales rate of change, and AR represents the end of term accounts receivable rate of change, SI represents the income from sales rate of change, and PC represents the period expense rate of change, and MB represents the main business income rate of change；

Step4: if having ways of going about tax evasion according to formula ruling, then the static association of output enterprise i and i association closes TethersWithDynamic associations limitWithIn all enterprises, there are the interests evaded the tax between them Association, tax law department carries out emphasis inspection according to tax law principle.

Compared with prior art, the inventive method has the advantage that

1, the mode using triad coding represents legal representative, investor, the combination of taxpayer's node type Situation, simplifies the expression of node type.

2, transportable property is good, extracts price mispairing pattern, make the method spending business take out from true inspection case As going out static association relation chain and dynamic associations chain, the method can be widely used in other connected transaction pattern extractions；

3, efficiency is high, and RHSF figure message passing mechanism has the advantages that speed is fast, simplify amount of calculation, and the method adds figure Point Set and the renewal of limit collection, make every iterative computation amount next time less；The definition of static association relation chain, solves often The problem that transaction traversal incidence relation network causes high time complexity.Price mispairing group is gathered by node messages in excavating The mode closed, improves nodal parallel and calculates speed, improve the efficiency of excavation；

4, recognition accuracy is high, and this method filters out the specific tax index location relevant to price mispairing pattern and evades the tax Enterprise, by the screening of three coefficient of elasticity, can the most effectively be located through the enterprise that price mispairing behavior is evaded the tax.

Accompanying drawing illustrates:

Below in conjunction with the drawings and the specific embodiments, the present invention is described in further detail.

Fig. 1 is that the inventive method relates to price mispairing based on taxpayer's interests related network and evades the tax the whole of Activity recognition Body schematic flow sheet.

Fig. 2 is taxpayer's interests related network legend and an example.

Fig. 3 is that price mispairing pattern checks case legend.

Fig. 4 is price mispairing model legend.

Fig. 5 is the legend of static association relation chain extraction；Wherein, Fig. 5-1 is taxpayer's interests related network case diagram, figure 5-2 is static association network, and Fig. 5-3 is static association relation chain extraction process figure, and Fig. 5-4 is the extraction of static association relation chain Result figure.

Fig. 6 is the legend that price mispairing suspicion group excavates；Wherein, Fig. 6-1 is to dynamically associate network, and Fig. 6-2 is State incidence relation chain calculates procedure chart, and Fig. 6-3 is price mispairing suspicion cluster results figure.

Detailed description of the invention:

Below in conjunction with accompanying drawing, price mispairing based on taxpayer's interests related network to the present invention is evaded the tax Activity recognition Particular content does careful description.A kind of price mispairing that the present invention relates to evades the tax behavior pattern recognition process as it is shown in figure 1, have Body process is as follows:

(1) structure of taxpayer's interests related network

Taxpayer's interests related network represents the contact between enterprise and enterprise, investor and legal representative, and it is by reality Body (including enterprise, investor, legal representative), the relation (transaction relationship, investment relation, control planning) of inter-entity, entity belong to The network that property (enterprise name, category of employment etc.) enough becomes with attribute of a relation (ratio between investments, dealing money etc.) four key element.

Taxpayer's interests related network is expressed as two tuples by this method:

TPIN={ (V, VD), (E, ED) }

Wherein V={v_p| p=1,2 ..., N_pRepresent node set, wherein N_pRepresent the node number in network, each Node is uniquely identified by the numeral of 1 to p, all set that there is limit in E expression figure, and makes E={e_pq}={ (v_p,v_q) | 0 ＜ P, q ＜ N_p, wherein e_pq=(v_p,v_q) representing the oriented line from the node to the node being numbered q being numbered p, VD is expressed as Nodal community, this method is denoted as:

VD={Type, ID, Name}

WhereinRepresenting the type of node, Vnsr represents taxpayer, Vfddbr representation Determining representative, Vtzf represents investor, and they represent with triad number respectively, and binary number 001 (i.e. 1) represents pays taxes People；Binary number 010 (i.e. 2) represents legal representative；Binary number 100 (i.e. 4) represents investor.If node has multiple Type is then carried out or computing, such as, be that taxpayer invests again then binary number 001 | 100=101 (i.e. 5)；ID={sfzh, Nsrsbh} represents the identification number of node, and it is the unique identifying number of enterprise and legal representative, wherein enterprise's Taxpayer Identification Number Nsrsbh uniquely identifies, and legal representative is uniquely identified by identification card number sfzh；Name represents the Chinese of node.The attribute on limit It is expressed as:

ED={CT, IV, TD}

Wherein CT={w_pq| 0 ＜ p, q ＜ N_pRepresent legal person and the control weight on limit, the human world of paying taxes, only comprise two kinds of controls Relation { controls, do not control }, IV={w_pq| 0 ＜ p, q ＜ N_p∈ (0,1] represent limit (v_p,v_q) investment weight, for investor v_p Place enterprise v_qThe size of control ratio, TD={w_pq| 0 ＜ p, q ＜ N_p∈ (0,1] represent limit (v_p,v_q) transaction weight, for Enterprise v_pWith enterprise v_qTurnover accounts for enterprise v_qRatio.

As in figure 2 it is shown, describe is that enterpriser L1 controls mining firm C1 by the holding ratio of 51% and 80% respectively Invested, with the ratio between investments of 20%, the friendship that cement processing enterprise C3, C3 occupy the 20% of C2 with program of real estate enterprise C2, enterprise C1 Easily share.

(2) based on inspection case price mispairing evade the tax behavioral pattern extraction

Existing tax evasion case much derives from the running of affiliated enterprise, such as between investment enterprise and invested enterprise, Having between the enterprise of common holding people, this method has group's mould of tax evasion suspicion taking out from business case Type, obtains the point without concrete business and the topology diagram on limit, makes the excavation of the pattern of evading the tax is converted in TPIN association The search of relation chain.

1) price mispairing mode business example is as it is shown on figure 3, certain is engaged in the company A that printer produces, for dominating the market And implement low price marketing strategy, and the subsidiary B set up by it produces and direct to third party's distributor's sale at low prices printer. The printer produced due to B company has its special printer material consumption, and therefore company A is engaged in printer material consumption by its another family The subsidiary C produced, sells goods at a high figure the printer material consumptions such as print cartridge to this third party distributor.Pass through pricing strategy, it is achieved company B With C overall burden less than reality.Such as B company is suitable for the tax rate of 33%, and C company is suitable for the tax rate of 15%, and D company must prop up to B Pay 600,000 to buy printer, and the dedicated printer volume consumptive material of 100,000 must be bought to C company, and pass through the association between A and D Business, can make in B payment printer expense 600,000 to take 500,000 and transfer in printer material consumption expense, and real trade amount is not Become, finally make D company pay 100,000 yuan of B company, pay C company 600,000.Each company amount of tax to be paid is calculated as follows:

B company Ying Na income tax=60 × 33%=19.98 (ten thousand yuan)

C company Ying Na income tax=10 × 15%=1.5 (ten thousand yuan)

B, C two company altogether pay income tax=19.98+1.5=21.48 (ten thousand yuan)

But for tax avoidance purpose, by price mispairing, the part of my profit of B company high for the tax rate can be transferred to C public Department.Computation of tax amount after price mispairing is:

B company Ying Na income tax=10 × 33%=3.3 (ten thousand yuan)

C company Ying Na income tax=60 × 15%=9 (ten thousand yuan)

B, C two company pay income tax altogether=3.3+9=12.3 (ten thousand yuan)

It will thus be seen that before the transaction value mispairing of B, C two companies, profit summation is: 60+10=70 (ten thousand yuan), profit After transfer, gross profit is: 10+60=70 (ten thousand yuan).I.e. before and after profit shifting, the gross profit of two companies is equal, simply Using after transfer pricing, two companies they the income tax that should pay decrease 21.48-12.3=9.18 (ten thousand yuan), become Merit is evaded taxes.Therefore, there is the subsidiary that two tax rates are different in a company, can by must purchase commodities to Liang Ge subsidiary simultaneously the Tripartite company reaches the purpose of tax evasion.

2) price mispairing pattern extraction result is as shown in Figure 4, above-mentioned case can be summarized as enterprise A control its subsidiary B, , there is third party enterprise D in C, enterprise B, C are simultaneously as the side of purchasing of D, or enterprise B, C are simultaneously as the pin side of D, in this feelings Under condition, it is possible to there is the behavior of price mispairing between B, C, D, this method meet above-mentioned pattern point set A, B, C, D} and Limit collectionThe group definition of composition is price mispairing suspicion group.

a.C₁By control/investment chainControl C_n, C_m+nEnterprise, C_n, C_m+nBy transaction LimitWith C₀Transaction；

b.C₁By control/investment chainControl C_nEnterprise, C₁, C_nBy transaction limitWith C₀Hand over Easily；

c.C₁By control/investment chainControl C_n, C_m+nEnterprise, C_n, C_m+nBy transaction LimitWith C₀Transaction；

d.C₁By control/investment chainControl C_nEnterprise, C₁, C_nBy transaction limitWith C₀Hand over Easily.

(3) price mispairing is evaded the tax behavior pattern recognition

This section is evaded the tax enterprise based on TPIN and price mispairing behavior pattern recognition of evading the tax, and is broadly divided into four parts.First First, set up figure message propagation model RHSF, extract static association relation chain secondly based on RHSF, afterwards based on static association relation Chain excavates price mispairing suspicion group, suspicion enterprise finally carries out tax index calculating and analysis, and exports the enterprise that evades the tax.

1. RHSF message mechanism of transmission builds

This method proposes RHSF figure message propagation model, and RHSF model uses the figure computation model centered by node, joint The processing stage that some renewal function being divided into four continuous print: information collecting step (Receive), signal processing stages (Handle), information distribution phase (Send) and filtration stage (Filter), wherein Receive/Send (being called for short RS) function is With single edge for operation granularity, Handle/Filter function is with single node for operation granularity.As a example by node i, illustrate The step of RHSF model:

Step1: information collecting step, is led to all of its neighbor node of i-node by one with the information on the limit being connected Collect by self-defining function:

$\mathrm{\&Sigma;}\&LeftArrow;{\&CirclePlus;}_{v\&Element;b\&lsqb;i\&rsqb;}R({\mathrm{Attr}}_i,{\mathrm{Attr}}_{(i,v)},{\mathrm{Attr}}_v)$

Wherein Attr_i、Attr_v、Attr_(i,v)Be respectively node i, node v and from node i to the limit of node v letter Breath, v ∈ b [i] represents all nodes being connected with node i, and R () represents the reception process of message, and user can be with self-defining functionDefine the process to message preliminary treatment；Σ means that node receives the information on adjacent node and limit preliminary pretreatment Result.

In Step2: the Information application stage, the Σ receiving in Step1 and calculating being applied to node i, change node is certainly The attribute of body:

${\mathrm{Attr}}_i^{new}\&LeftArrow;H({\mathrm{Attr}}_i,\mathrm{\&Sigma;})$

WhereinProperty value after updating for node i, H () represents the processing procedure of message.

Step3: information distribution phase, by information new on i-node and all initial informations being connected on limit with node i It is distributed to other adjacent nodes by User-Defined Functions:

$\&ForAll;v\&Element;b\&lsqb;i\&rsqb;:\left({\mathrm{Attr}}_v\right)\&LeftArrow;S({\mathrm{Attr}}_i^{new},{\mathrm{Attr}}_{(i,v)})$

WhereinRepresenting and travel through each node being connected with node i, S () represents the distribution procedure of message.

Step 4: filtration stage, by the node required for User Defined Rules Filtering next round iteration and limit, deletes Epicycle iteration not receiving node and its limit being connected of message, whole graph structure being updated, if figure still having surplus Remaining node, return Step1:

Graph_new←F(vertices,edges)

Wherein vertices represents Step3 Point Set, and edges represents the limit collection in Step3, and F () represents filter process, Graph_newRepresent the graph structure after updating.

2. static association relation chain extraction

First two concepts are defined:

1) static association relational network

Static association relational network is the sub-network only comprising investment and control planning in taxpayer's interests related network.

2) static association relation chain

Static association relation chain is to control, invest the collection of end to end path, limit chain in static association relational network Close.If path chain here is defined as investment enterprise C₁(C here_iRepresent enterprise) by one or many enterprises control indirectly or Investment enterprise C_n+1, then C is claimed₁And C_n+1Between end to end path be path chain, and useRepresent a static state Incidence relation chain.

Static association relation chain is the mostly important ingredient being carried behavioral pattern topological model of evading the tax, therefore, and can To avoid repeating to travel through static association relational network by periodically structure off-line static association relation chain storehouse, its building process is such as Under:

Step1: based on static association Relation extraction static association relational network from TPIN；

Step2: for node v each in network_p(p=1 ..., N_P) add local path aggregate attribute LP_p, wherein v_pRepresent Node, LP_pRepresent node v_pIn local path attribute, its structure is Set [Seq [P]], and wherein Set represents set, uses symbol { } represents, Seq represents sequence, represents with symbol<>, and P represents that the element type of sequence is node serial number type；

Step3: for each node v_pA path sequence ＜ v only comprising own node numbering is added in local path set_p ＞, the most each node v_pLocal path set is initialized as { ＜ v_p＞ }.Meanwhile, it is empty for initializing global path set GP in network；

Step4: based on RHSF message distribution mechanism, each node v in network_pAlong static association relation edge direction, by v_p's Current local path set is sent to its adjacent node v_ps(1),v_ps(2),...,v_psM (), wherein m represents with v_pFor initial joint The quantity of the adjacent node of point, v_psI () (0 ＜ i≤m) represents v_pNode passes through static association relation chainBe connected is all Adjacent node；

Step 5: based on RHSF message collection and treatment mechanism, each node v in network_pTo the local path received Set LP_pr(1),LP_pr(2),...,LP_prN () does union, wherein n is with v_pFor the quantity of the adjacent node of terminal node, obtain To set of paths LP_pr, travel through LP_prIn each paths, if path comprises current vertex self numbering p, then delete this path； Otherwise, current vertex self numbering p is added at this end, path.The set of paths finally given is designated as LP '_pr；

Step6: by each node v_pLocal path set be newly defined as LP '_pr, and by LP '_prAdd global path to In set GP；

Step7: based on RHSF strobe utility, deletes the node being not received by message and the limit with it as source point, it is judged that Current static incidence relation nodes quantity, if number of nodes is zero, then terminates.

After above-mentioned steps terminates, output global path set is static association relation chain storehouse.

The structure example in static association relation storehouse is as shown in Figure 5.

Step1: build TPIN.Fig. 5-1 is an example of TPIN, and wherein A, B are investor and enterprise, and C, D are Enterprise, A enterprise investment B enterprise, B enterprise investment C enterprise, there is transaction relationship between A and D, D and C, B and D, direction is that the former refers to Latter.

Step2: extract former piece network.Fig. 5-2 is the former piece network extracted in TPIN, only closes containing investment relation and control System, without transaction relationship.Limit is invested as figure extracts AB, BC two.

Step3: first round iteration: former piece network of iteration, generates former piece path.For the first time in iteration, each node will be from The ID of body sends to adjacent node along former piece network edge, and the numbering received is stored in genus plus self numbering by the node receiving number information In property, in Fig. 5-3 (left), node A will to node B, B by the transmission of numbering AAs Article one, former piece path is stored in self attributes, and node B will to node C, C by the transmission of numbering B simultaneously It is stored in self attributes as a former piece path.At this moment, in order to prevent from receiving repetition message, reject and do not receive letter The node A of breath, the iteration after the entrance of remaining figure.

Step4: the 2～n wheel iteration: on the basis of the figure of Step3, loop iteration former piece network, until generate final before Part path.In 2～n wheel iteration, the path attribute of self is sent to adjacent node by each node along former piece network edge, receives letter The information that receives is polymerized by the node of breath plus own node numbering and self original attribute, such as Fig. 5-3 (in), node B will belong to Property (A, B) send to node C, C will Rejecting the node B not receiving information in epicycle iteration, remaining figure continues executing with Step4, if Point in figure is the most all rejected or has not been had the shop that can reject in figure, exits circulation, such as Fig. 5-3 (right), third round In iteration, C node does not receive the information from adjacent node, rejects C node, has not had remaining node, now move back in figure Go out circulation.

Step5: obtain a result.Such as Fig. 5-4, with before with this node as terminal node in the most each nodal community Part path, now, for follow-up mode discovery, adds own node numbering, such as A nodal community on the attribute of each point { (A) }, B node attribute { (B), (A, B) }, C nodal community { (C), (B, C), (A, B, C) }.

3. price mispairing pattern extraction

First related notion is defined:

1) dynamic associations network

Dynamic associations network refers to only comprise in taxpayer's interests related network the sub-network of transaction relationship

2) dynamic associations chain

Dynamic associations chain refers to the transaction limit in dynamic associations network, and comprises static association in nodal community Relation chain set.

3) reverse bilateral dynamic associations chain

Reverse bilateral dynamic associations chain is by pointing to same node or point out by same node two dynamically associate The set of relation chain.

Being mainly characterized by of price mispairing behavioral pattern comprises a reverse bilateral dynamic associations chain, its pattern recognition Core be excavate two Opposite direction connections dynamic associations limit, and identify be connected limit two ends whether there is " static association Relation chain or bidirectional static incidence relation chain ".

The step of price mispairing behavior patterns mining is as follows:

Step1: extract dynamic associations network, its node serial number set based on dynamic associations extraction from TPIN For Q；

Step2: to global path set GP, wherein GP is static association relation storehouse, compiles according to each path termination in set Number q (q ∈ Q) is polymerized, and obtains terminal numbering and key-value pair (q, the path of set of paths composition_q), wherein path_qFor terminal q Corresponding set of paths；

Step3: for node v each in network_qAdd local path attribute LP_q, its value is set to path_q；

Step4: send mechanism, each node v in network based on RHSF message_qBy dynamic associations limit by its local road Footpath information LP_qIn conjunction with the flag bit flag representing edge direction_q, it is combined into message Msg_qPass to adjacent node.flag_qRepresent two Direction similarities and differences flag bit between bar dynamic associations chain, if direction of transfer is identical with edge direction, then flag_qIt is set to 1；Instead It, be set to-1.

Step5: each node v in network_qAccording to flag_qBe polymerized, obtain two massage set MsgSet1 and MsgSet2:

MsgSet1={Msg_i|flag_i=1, i ∈ Q}

MsgSet2={Msg_j|flag_j=-1, j ∈ Q}

Step 6: arbitrary Msg_i∈ MsgSet1, Msg_j∈ MsgSet1, it is assumed that Msg_iAnd Msg_jCorresponding set of paths It is respectively MP_iAnd MP_j,Wherein MP_iAnd MP_jIt is all MsgSet1 In the set in all paths,Representing with p as start node, r is the static association relation chain of terminal node, Representing with t as start node, s is the static association relation chain of terminal node, if there is p=t, then static association relation chainWithIn conjunction with dynamic associations limitWithConstitute the suspicion group of a price mispairing transaction.

Step 7: arbitrary Msg_i∈ MsgSet2, Msg_j∈ MsgSet2, it is assumed that Msg_iAnd Msg_jCorresponding set of paths It is respectively MP_iAnd MP_j,Wherein MP_iAnd MP_jIt is all The set in all paths in MsgSet2,Representing with p as start node, r is the static association relation chain of terminal node,Representing with t as start node, s is the static association relation chain of terminal node, if there is p=t, then static association closes TethersWithIn conjunction with dynamic associations limitWithConstitute the suspicion group of a price mispairing transaction.

Price mispairing mode construction example is as shown in Figure 6.Step is as follows:

Step1: extract trade network.Fig. 6-1 is an example of TPIN, and wherein A, B are investor and enterprise, C, D is enterprise, A enterprise investment B enterprise, B enterprise investment C enterprise, has transaction relationship between A and D, D and C, B and D, and direction is for the former Pointing to the latter, after removing two investments limit (grey) in TPIN exemplary plot, the remaining figure containing only transaction limit is trade network.

Step2: trade network, former piece path attribute are polymerized.In order to calculated in trade network co-controlling people or The suspicion group of joint investment side, needs to add to trade network the path attribute calculated in former piece network before, is formed Such as the network of Fig. 6-2 (left), the point without former piece path attribute is assigned to sky, and the attribute such as Fig. 6-2 (left) interior joint D is Empty.

Step3: trade network node of graph attribute transmits.The attribute of self is transferred to adjoin by each node by transaction limit Node, adds flag bit when of propagation, add flag=1 in the message during along the transmission of transaction limit, during inverse transaction limit transmission Message is added flag=-1, such as Fig. 6-2 (right), A node will (A), 1} is transferred to D node, B node incite somebody to action (B), (A, B), 1} is transferred to D node, and C node will { (C), (B, C), (A, B, C) ,-1} be transferred to D node.

Step4: node messages calculates.After Step3, each point with the former piece path of adjacent transaction node, so Parallel in Step4 calculating suspicion group at each intra-node, all message received are resolved into by the most each node Two queues:

List1={Msg_i| flag=1}

List2={Msg_i| flag=-1}

Wherein Msg_iRepresent a piece of news passed over from node i, such as in Step3, B node passes to D node { (B), (A, B), 1} is indicated as a piece of news；List1 be all flag bits be the set of the message of 1, represent in List1 All informed sources are in the selling party of node；List2 be all flag bits be the set of the message of-1, represent the institute in List2 Have informed source in the purchaser of node, such as Fig. 6-2 (right) D node by receive three message category be

List1={{ (A), 1}, { (B), (A, B), 1}}

List2={{ (C), (B, C), (A, B, C) ,-1}}

Owing in price mispairing pattern two transaction limits are identical two transaction limits, direction, so traveling through Msg respectively_m ∈ List1, Msg_n∈ List1, Msg_p∈ List2, Msg_q∈ List2,

$\{\&ForAll;{\mathrm{Msg}}_m\&Element;List1,\&ForAll;{\mathrm{Msg}}_n\&Element;List1\}\&DoubleRightArrow;\{{\mathrm{Msg}}_m,{\mathrm{Msg}}_n\}$

$\{\&ForAll;{\mathrm{Msg}}_p\&Element;List2,\&ForAll;{\mathrm{Msg}}_q\&Element;List2\}\&DoubleRightArrow;\{{\mathrm{Msg}}_p,{\mathrm{Msg}}_q\}$

Such as { (A), 1} is with { 1} combines for (B), (A, B).

Step5: suspicion mode excavation.Extract the former piece path in each combination in Step4, as (A), and 1} with (B), (A, B), extracts Part1={ (A) in 1} combination }, Part2={ (B), (A, B) }, travel through all of Route_m∈ Part1, Route_n∈ Part2, if Route_m.head=Route_n.head, (head represents first node in path), then represent Route_m、Route_nThere is same juristic person to control source or same investor invests source, such as path (A) and Part2 Road in Part1 Footpath (A, B) start node is all A, extracts this two former piece paths, and own node ID forms a price mispairing suspicion group, Such as the suspicion group of Fig. 6-3, each suspicion group table is shown as { id, vertex={V₁,V₂,...V_n, edge={E₁,E₂, ...E_n, wherein vertex set of all nodes in being suspicion group, edge is the set on all limits in suspicion group.

4. the affiliated enterprise that evades the tax based on tax index identifies

3. the price mispairing suspicion group obtained step proceeds as follows:

Step1: extract, from all of price mispairing suspicion group, the node that type (type) is taxpayer (nsr), its Node set is M；

Step2: to each node i (i ∈ M), calculates three the specific tax indexs relevant to price mispairing pattern, respectively For taxpayer's stock turnover rate and income from sales rate of change coefficient of elasticity (being designated as ISEC), enterprise's end of term accounts receivable rate of change with Income from sales rate of change coefficient of elasticity (being designated as ASEC), taxpayer's period expense rate of change are elastic with the main business income rate of change Coefficient (is designated as PMEC).

Wherein the formula of ISEC, ASEC, PMEC is respectively as follows:

Step 3: be analyzed the index calculated, utilizes following span to evaluate these business indicatorses, and to often Individual enterprise i forms corresponding result of determination, if desired value is in the range of " existing the behavior of evading the tax ", then the existence of this enterprise is evaded the tax Behavior, if desired value is in the range of " in the range of belonging to warning ", then this enterprise evades the tax suspicion, if desired value " is just belonging to In the range of Chang " in the range of, then this enterprise is without the suspicion of evading the tax:

Wherein, IT represents that stock turnover rate, ST represent the income from sales rate of change, and AR represents the end of term accounts receivable rate of change, SI represents the income from sales rate of change, and PC represents the period expense rate of change, and MB represents the main business income rate of change；

Step4: if having ways of going about tax evasion according to formula ruling, then relevant for output enterprise i and enterprise i static pass Connection relation chainWithDynamic associations limitWithIn all enterprises, exist between them and evade the tax Interests associate, and tax law department can carry out emphasis inspection according to tax law principle.

Claims (5)

1. price mispairing based on taxpayer's interests related network is evaded the tax Activity recognition method, it is characterised in that the method profit Use tax bureau's data, in conjunction with inspection case, propose price mispairing and evade the tax behavior, and extract price mispairing group, finally position Evade the tax affiliated enterprise.

Price mispairing based on taxpayer's interests related network the most according to claim 1 is evaded the tax Activity recognition method, It is characterized in that, specifically include following steps:

1) first, taxpayer's interests related network is built；

2) secondly, evade the tax behavioral pattern based on inspection case extraction price mispairing；

3) last, based on 1) in taxpayer's interests related network and 2) in price mispairing evade the tax behavior pattern recognition price mispairing Evade the tax behavior.

Price mispairing based on taxpayer's interests related network the most according to claim 2 is evaded the tax Activity recognition method, It is characterized in that, build taxpayer's interests related network, specific as follows:

Taxpayer's interests related network represents the contact between enterprise and enterprise, investor and legal representative, by entity, entity Between the network that enough becomes of relation, entity attribute and attribute of a relation four key element；Wherein, entity includes enterprise, investor, Fa Rendai Table, the relation of inter-entity includes transaction relationship, investment relation, control planning, and entity attribute includes enterprise name, category of employment, Attribute of a relation includes ratio between investments, dealing money；

Taxpayer's interests related network is expressed as two tuples:

TPIN={ (V, VD), (E, ED) }

Wherein V={v_p| p=1,2 ..., N_pRepresent node set, wherein N_pRepresent the node number in network, each node The unique mark of numeral by 1 to p, all set that there is limit in E expression figure, and make E={e_pq}={ (v_p,v_q) | 0 ＜ p, q ＜ N_p, wherein e_pq=(v_p,v_q) representing the oriented line from the node to the node being numbered q being numbered p, VD is expressed as node and belongs to Property, it is denoted as:

VD={Type, ID, Name}

WhereinRepresenting the type of node, Vnsr represents that taxpayer, Vfddbr represent authorised representative People, Vtzf represents investor, represents with triad number respectively, binary number 001, and decimal number 1 represents taxpayer；Two enter Making several 010, decimal number 2 represents legal representative；Binary number 100, decimal number 4 represents investor, if node has Multiple types then carries out binary system or computing；ID={sfzh, nsrsbh} represent the identification number of node, are enterprise and legal representative Unique identifying number, wherein enterprise Taxpayer Identification Number nsrsbh uniquely identifies, and legal representative is unique by identification card number sfzh Mark；Name represents the Chinese of node, and the attribute list on limit is shown as:

ED={CT, IV, TD}

Wherein CT={w_pq| 0 ＜ p, q ＜ N_pRepresent legal person and the control weight on limit, the human world of paying taxes, only comprise two kinds of control planning { controlling, do not control }, IV={w_pq| 0 ＜ p, q ＜ N_p∈ (0,1] represent limit (v_p,v_q) investment weight, for investor v_pPlace Enterprise v_qThe size of control ratio, TD={w_pq| 0 ＜ p, q ＜ N_p∈ (0,1] represent limit (v_p,v_q) transaction weight, for enterprise v_pWith enterprise v_qTurnover accounts for enterprise v_qRatio.

Price mispairing based on taxpayer's interests related network the most according to claim 2 is evaded the tax Activity recognition method, It is characterized in that, based on inspection case price mispairing evade the tax behavioral pattern extraction, specific as follows:

From business case, take out the cohort model with tax evasion suspicion, obtain the point without concrete business and limit Topology diagram, makes the excavation of the pattern of evading the tax is converted into the search of incidence relation chain in TPIN.

Price mispairing based on taxpayer's interests related network the most according to claim 2 is evaded the tax Activity recognition method, It is characterized in that, price mispairing is evaded the tax behavior pattern recognition, and concrete grammar is as follows:

Evade the tax enterprise based on TPIN and price mispairing behavior pattern recognition of evading the tax, be divided into four parts, first, set up figure message Propagation model RHSF, extracts static association relation chain secondly based on RHSF, excavates price based on static association relation chain afterwards wrong Join suspicion group, finally suspicion enterprise is carried out tax index calculating and analysis, and exports the enterprise that evades the tax, specific as follows:

(1) RHSF message mechanism of transmission builds

Based on RHSF figure message propagation model, RHSF model uses the figure computation model centered by node, node updates function The processing stage of being divided into four continuous print: information collecting step Receive, signal processing stages Handle, information distribution phase Send and filtration stage Filter, wherein Receive/Send, be called for short RS, and function is with single edge for operation granularity, Handle/Filter function is with single node for operation granularity, as a example by node i, the step of RHSF model is described:

Step1: information collecting step, by all of its neighbor node of i-node and the information on the limit being connected by one general from Defined function collects:

$\mathrm{\&Sigma;}\&LeftArrow;{\&CirclePlus;}_{v\&Element;b\&lsqb;i\&rsqb;}R({\mathrm{Attr}}_i,{\mathrm{Attr}}_{(i,v)},{\mathrm{Attr}}_v)$

Wherein Attr_i、Attr_v、Attr_(i,v)Be respectively node i, node v and from node i to the limit of node v information, v ∈ b [i] represents all nodes being connected with node i, and R () represents the reception process of message, User-Defined FunctionsDefine and offset The process of breath preliminary treatment；Σ means that node receives the information on adjacent node and limit the result of preliminary pretreatment；

In Step2: the Information application stage, the Σ receiving in Step1 and calculating is applied to node i, change node self Attribute:

Attr_i ^new←H(Attr_i,Σ)

Wherein Attr_i ^newProperty value after updating for node i, H () represents the processing procedure of message；

Step3: information distribution phase, passes through information new on i-node and all initial informations being connected on limit with node i User-Defined Functions is distributed to other adjacent nodes:

$\&ForAll;v\&Element;b\&lsqb;i\&rsqb;:\left({\mathrm{Attr}}_v\right)\&LeftArrow;S({\mathrm{Attr}}_i^{new},{\mathrm{Attr}}_{(i,v)})$

WhereinRepresenting and travel through each node being connected with node i, S () represents the distribution procedure of message；

Step 4: filtration stage, by the node required for User Defined Rules Filtering next round iteration and limit, deletes epicycle Iteration does not receive node and its limit being connected of message, whole graph structure is updated, if figure still has residue joint Point, return Step1:

Graph_new←F(vertices,edges)

Wherein vertices represents Step3 Point Set, and edges represents the limit collection in Step3, and F () represents filter process, Graph_newRepresent the graph structure after updating；

(2) static association relation chain extraction

First two concepts are defined:

1) static association relational network

Static association relational network is the sub-network only comprising investment and control planning in taxpayer's interests related network；

2) static association relation chain

Static association relation chain is to control, invest the set of end to end path, limit chain in static association relational network, this If path chain in is defined as investment enterprise C₁, C here_iRepresent enterprise, indirectly controlled by one or many enterprises or invest enterprise Industry C_n+1, then C is claimed₁And C_n+1Between end to end path be path chain, and useRepresent that a static association closes Tethers；

Static association relation chain is the mostly important ingredient being carried behavioral pattern topological model of evading the tax, therefore, by fixed Phase builds off-line static association relation chain storehouse and avoids repeating to travel through static association relational network, and its building process is as follows:

Step1: based on static association Relation extraction static association relational network from TPIN；

Step2: for node v each in network_pAdd local path aggregate attribute LP_p, wherein, p=1 ..., N_P, v_pRepresent node, LP_pRepresent node v_pIn local path attribute, its structure is Set [Seq [P]], and wherein Set represents set, with symbol { } table Showing, Seq represents sequence, represents with symbol<>, and P represents that the element type of sequence is node serial number type；

Step3: for each node v_pA path sequence ＜ v only comprising own node numbering is added in local path set_p＞, i.e. Each node v_pLocal path set is initialized as { ＜ v_p＞ }, meanwhile, it is empty for initializing global path set GP in network；

Step4: based on RHSF message distribution mechanism, each node v in network_pAlong static association relation edge direction, by v_pCurrent Local path set is sent to its adjacent node v_ps(1),v_ps(2),...,v_psM (), wherein m represents with v_pFor start node The quantity of adjacent node, v_psI () represents v_pNode passes through static association relation chainThe all of its neighbor node being connected, its In, 0 ＜ i≤m；

Step5: based on RHSF message collection and treatment mechanism, each node v in network_pTo the local path set received LP_pr(1),LP_pr(2),...,LP_prN () does union, wherein n is with v_pFor the quantity of the adjacent node of terminal node, obtain road Footpath set LP_pr, travel through LP_prIn each paths, if path comprises current vertex self numbering p, then delete this path；Otherwise, Adding current vertex self numbering p at this end, path, the set of paths finally given is designated as LP '_pr；

Step6: by each node v_pLocal path set be newly defined as LP '_Zr, and by LP '_prAdd global path set to In GP；

Step7: based on RHSF strobe utility, deletes the node being not received by message and the limit with it as source point, it is judged that current Static association relational network interior joint quantity, if number of nodes is zero, then terminates,

After above-mentioned steps terminates, output global path set is static association relation chain storehouse；

(3) price mispairing pattern extraction

First related notion is defined:

1) dynamic associations network

Dynamic associations network refers to only comprise in taxpayer's interests related network the sub-network of transaction relationship；

2) dynamic associations chain

Dynamic associations chain refers to the transaction limit in dynamic associations network, and comprises static association relation in nodal community Chain set；

3) reverse bilateral dynamic associations chain

Reverse bilateral dynamic associations chain is by pointing to same node or two dynamic associations pointed out by same node The set of chain；

Being mainly characterized by of price mispairing behavioral pattern comprises a reverse bilateral dynamic associations chain, the core of its pattern recognition The heart is the dynamic associations limit excavating two Opposite direction connections, and identifies whether the two ends on the limit that is connected exist " static association relation Chain or bidirectional static incidence relation chain "；

The step of price mispairing behavior patterns mining is as follows:

Step1: extracting dynamic associations network based on dynamic associations extraction from TPIN, its node serial number collection is combined into Q；

Step2: to global path set GP, wherein GP is static association relation storehouse, enters according to each path termination numbering q in set Row polymerization, obtains terminal numbering and key-value pair (q, the path of set of paths composition_q), wherein q ∈ Q, path_qCorresponding for terminal q Set of paths；

Step3: for node v each in network_qAdd local path attribute LP_q, its value is set to path_q；

Step4: send mechanism, each node v in network based on RHSF message_qBy dynamic associations limit, its local path is believed Breath LP_qIn conjunction with the flag bit flag representing edge direction_q, it is combined into message Msg_qPass to adjacent node, flag_qRepresent that two are moved Direction similarities and differences flag bit between state incidence relation chain, if direction of transfer is identical with edge direction, then flag_qIt is set to 1；Otherwise, put For-1；

Step5: each node v in network_qAccording to flag_qIt is polymerized, obtains two massage set MsgSet1 and MsgSet2:

MsgSet1={Msg_i|flag_i=1, i ∈ Q}

MsgSet2={Msg_j|flag_j=-1, j ∈ Q}

Step6: arbitrary Msg_i∈ MsgSet1, Msg_j∈ MsgSet1, it is assumed that Msg_iAnd Msg_jCorresponding set of paths is respectively MP_iAnd MP_j,Wherein MP_iAnd MP_jIt it is all institute in MsgSet1 There is the set in path,Representing with p as start node, r is the static association relation chain of terminal node,Represent With t as start node, s is the static association relation chain of terminal node, if there is p=t, then static association relation chain WithIn conjunction with dynamic associations limitWithConstitute the suspicion group of a price mispairing transaction；

Step7: arbitrary Msg_i∈ MsgSet2, Msg_j∈ MsgSet2, it is assumed that Msg_iAnd Msg_jCorresponding set of paths is respectively MP_iAnd MP_j,Wherein MP_iAnd MP_jIt is all in MsgSet2 The set in path,Representing with p as start node, r is the static association relation chain of terminal node,Represent with t For start node, s is the static association relation chain of terminal node, if there is p=t, then static association relation chainWithIn conjunction with dynamic associations limitWithConstitute the suspicion group of a price mispairing transaction；

(4) affiliated enterprise that evades the tax based on tax index identifies

3. the price mispairing suspicion group obtained step proceeds as follows:

Step1: extract the node that type type is taxpayer nsr, its node set from all of price mispairing suspicion group For M；

Step2: to each node i (i ∈ M), calculate three the specific tax indexs relevant to price mispairing pattern, respectively receive Tax people's stock turnover rate and income from sales rate of change coefficient of elasticity, be designated as ISEC, enterprise's end of term accounts receivable rate of change and sale Income change rate coefficient of elasticity, is designated as ASEC, taxpayer's period expense rate of change and main business income rate of change coefficient of elasticity, It is designated as PMEC；

Wherein the formula of ISEC, ASEC, PMEC is respectively as follows:

Step 3: be analyzed the index calculated, utilizes following span to evaluate these business indicatorses, and to each enterprise Industry i forms corresponding result of determination, if desired value is in the range of " existing the behavior of evading the tax ", then this enterprise exists row of evading the tax For, if desired value is in the range of " in the range of belonging to warning ", then this enterprise evades the tax suspicion, if desired value " is belonging to normal In the range of " in the range of, then this enterprise is without the suspicion of evading the tax:

Wherein, IT represents that stock turnover rate, ST represent the income from sales rate of change, and AR represents the end of term accounts receivable rate of change, SI table Showing the income from sales rate of change, PC represents the period expense rate of change, and MB represents the main business income rate of change；

Step4: if having ways of going about tax evasion according to formula ruling, then the static association relation chain of output enterprise i and i associationWithDynamic associations limitWithIn all enterprises, there are the interests evaded the tax between them and close Connection, tax law department carries out emphasis inspection according to tax law principle.