CN111383116A

CN111383116A - Method and device for determining transaction relevance

Info

Publication number: CN111383116A
Application number: CN202010470522.3A
Authority: CN
Inventors: 曾威龙; 钱隽夫; 王膂; 刘丹丹
Original assignee: Alipay Hangzhou Information Technology Co Ltd
Current assignee: Alipay Hangzhou Information Technology Co Ltd
Priority date: 2020-05-28
Filing date: 2020-05-28
Publication date: 2020-07-07

Abstract

In the method, main feature vectors of a transaction initiator main body and a transaction receiver main body related to a transaction are obtained from a main feature vector set, a relation feature vector between the transaction initiator main body and the transaction receiver main body is obtained from a relation feature vector set, the main feature vector set and the relation feature vector set belong to the same vector space, and the main feature vector set and the relation feature vector set are determined based on a knowledge graph. And determining the transaction relevance of the transaction based on the obtained main body feature vector and the relation feature vector.

Description

Method and device for determining transaction relevance

Technical Field

The embodiment of the specification relates to the technical field of artificial intelligence, in particular to a method and a device for determining transaction relevance.

Background

Trading is ubiquitous in daily life, small trades include trading of commodities for individuals, large trades include equity trades, import-export trades, securities trades, etc. for organizations or institutions. The transaction is completed by the transaction initiator and the transaction receiver together. For example, the Business model B2B (Business-to-Business) is a Business model in which businesses exchange and transfer data information with each other via a network to conduct transactions.

Generally, there is no association between two transaction parties, and if there is an association between two transaction parties, the association affects the transaction association of the transactions of the two transaction parties. The closer the association relationship between the two parties to the transaction is, the greater the transaction association of the transactions between the two parties can be considered. And the transaction relevance can influence the authenticity of the transaction, and when the transaction relevance reaches a certain degree, the transaction can be regarded as a false transaction, and then the false transaction is processed correspondingly.

Disclosure of Invention

In view of the foregoing, embodiments of the present specification provide a method and apparatus for determining transaction relevance. In the method and the device, the transaction relevance is determined by using the main body characteristic vector and the relation characteristic vector of the transaction initiator main body and the transaction receiver main body, so that the determined transaction relevance aims at the relevance represented by the relation characteristic vector, and under the condition that the transaction relevance only aims at a specific relevance, the influence of other relevance on the transaction relevance is avoided, and the accuracy of the transaction relevance aiming at the specific relevance is improved. In addition, the transaction relevance of the transaction can be evaluated in real time by utilizing the main feature vector set and the relation feature vector set, and the timeliness of the transaction evaluation is improved.

According to an aspect of an embodiment of the present specification, there is provided a method for determining transaction relevance, including: acquiring main feature vectors of a transaction initiator main body and a transaction receiver main body related to a transaction from a main feature vector set, and acquiring a relation feature vector between the transaction initiator main body and the transaction receiver main body from a relation feature vector set; and determining a transaction relevance of the transaction based on the obtained subject feature vector and relationship feature vector, wherein the subject feature vector set and the relationship feature vector set belong to the same vector space, and the subject feature vector set and the relationship feature vector set are determined based on a knowledge graph for reflecting an association relationship between subjects, the knowledge graph includes nodes and edges, the nodes characterize the subjects, and the edges characterize the association relationship between the subjects.

Optionally, in an example of the foregoing aspect, each relationship feature vector included in the relationship feature vector set corresponds to an association relationship, and determining the transaction relevance of the transaction based on the obtained subject feature vector and the relationship feature vector includes: for each incidence relation, determining the relation incidence degree of the incidence relation between the transaction initiator main body and the transaction receiver main body based on the main body characteristic vector of one transaction main body and the sum vector of the main body characteristic vector of the other end of transaction main body and the relation characteristic vector corresponding to the incidence relation; and determining the transaction relevance of the transaction according to the determined relationship relevance.

Optionally, in an example of the above aspect, for each association, determining a degree of relationship association between the transaction initiator main body and the transaction receiver main body for the association based on the subject feature vector of one transaction main body and the sum vector of the subject feature vector of another pair of transaction main bodies and the relationship feature vector corresponding to the association comprises: and determining a vector distance between the main body characteristic vector of one transaction main body and the sum vector of the main body characteristic vector of the other end of transaction main body and the relation characteristic vector corresponding to the association relation as the relation association degree between the transaction initiator main body and the transaction receiver main body aiming at the association relation aiming at each association relation.

Optionally, in one example of the above aspect, the vector distance comprises one of a euclidean distance, a mahalanobis distance, a hamming distance, a manhattan distance, a chebyshev distance, a correlation distance, a jaccard distance, and a minkowski distance.

Optionally, in an example of the above aspect, the association relationship includes at least one of the following association relationships: legal relation, job position relation, stock holding relation, fund relation, medium relation, address relation, communication mode relation, app friend relation and family member relation.

Optionally, in an example of the above aspect, determining the transaction relevance of the transaction according to the determined relationship relevance comprises: determining that the transaction is a related transaction when at least one of the determined relationship association degrees is greater than a first threshold.

Optionally, in an example of the above aspect, the first threshold corresponding to each association is a threshold matching the association.

Optionally, in an example of the above aspect, determining the transaction relevance of the transaction according to the determined relationship relevance comprises: and when all the relation association degrees in the determined relation association degrees are not larger than the first threshold, if the difference value between the relation association degrees which are more than the specified number and the first threshold is smaller than the difference threshold, determining that the transaction is the associated transaction.

Optionally, in one example of the above aspect, the transaction initiator body comprises at least two transaction bodies, and/or the transaction recipient body comprises at least two transaction bodies; the determined relationship association degree comprises: and the relationship association degree of each transaction main body in the transaction initiator main body and each transaction main body in the transaction receiver main body aiming at various association relations.

Optionally, in an example of the above aspect, the subject feature vector set and the relationship feature vector set are obtained by: constructing a set of triple samples based on the knowledge-graph, each triple sample in the set of triple samples including two subjects and an association between the two subjects; and performing vector characterization learning by using the triple sample set to obtain the main feature vector set and the relationship feature vector set.

Optionally, in an example of the above aspect, the loss function used for the vector characterization learning is obtained based on a subject feature vector of the subject and a vector distance between a sum vector of the subject feature vector of the corresponding subject and a relationship feature vector of the association, where the subject, the corresponding subject and the association belong to the same triple sample.

Optionally, in an example of the foregoing aspect, performing vector feature learning by using the triple sample set to obtain the subject feature vector set and the relationship feature vector set includes: and performing vector characterization learning by using the triple sample set and a derivative triple sample set of the triple sample set to obtain the subject feature vector set and the relationship feature vector set, wherein the derivative triple sample in the derivative triple sample set is obtained by transforming an association relationship or a subject in the triple sample, and the association relationship in the derivative triple sample does not exist between two subjects in the derivative triple sample.

According to another aspect of embodiments herein, there is also provided an apparatus for determining transaction relevance, including: the system comprises a characteristic vector acquisition unit, a transaction processing unit and a transaction processing unit, wherein the characteristic vector acquisition unit is used for acquiring main characteristic vectors of a transaction initiator main body and a transaction receiver main body related to a transaction from a main characteristic vector set and acquiring a relation characteristic vector between the transaction initiator main body and the transaction receiver main body from a relation characteristic vector set; and a transaction relevance determination unit that determines a transaction relevance of the transaction based on the acquired subject feature vector and relationship feature vector, wherein the subject feature vector set and the relationship feature vector set belong to the same vector space, and the subject feature vector set and the relationship feature vector set are determined based on a knowledge graph for reflecting an association relationship between subjects, the knowledge graph includes nodes and edges, the nodes represent subjects, and the edges represent an association relationship between subjects.

Optionally, in an example of the foregoing aspect, each relationship feature vector included in the relationship feature vector set corresponds to an association relationship, and the transaction relevance determining unit includes: a relation association degree determining module, configured to determine, for each association relation, a relation association degree between the transaction initiator main body and the transaction receiver main body for the association relation based on a main body feature vector of one transaction main body and a sum vector of a main body feature vector of another peer transaction main body and a relation feature vector corresponding to the association relation; and the transaction relevance determining module is used for determining the transaction relevance of the transaction according to the determined relation relevance.

Optionally, in an example of the above aspect, the relationship association degree determining module: and determining a vector distance between the main body characteristic vector of one transaction main body and the sum vector of the main body characteristic vector of the other end of transaction main body and the relation characteristic vector corresponding to the association relation as the relation association degree between the transaction initiator main body and the transaction receiver main body aiming at the association relation aiming at each association relation.

Optionally, in one example of the above aspect, the transaction relevance determination module: determining that the transaction is a related transaction when at least one of the determined relationship association degrees is greater than a first threshold.

Optionally, in one example of the above aspect, the transaction relevance determination module: and when all the relation association degrees in the determined relation association degrees are not larger than the first threshold, if the difference value between the relation association degrees which are more than the specified number and the first threshold is smaller than the difference threshold, determining that the transaction is the associated transaction.

Optionally, in an example of the above aspect, further comprising: the triple sample set constructing unit constructs a triple sample set based on the knowledge graph, wherein each triple sample in the triple sample set comprises two subjects and an incidence relation between the two subjects; and the vector characterization learning unit performs vector characterization learning by using the triple sample set to obtain the main feature vector set and the relationship feature vector set.

Optionally, in one example of the above aspect, the vector characterization learning unit: and performing vector characterization learning by using the triple sample set and a derivative triple sample set of the triple sample set to obtain the subject feature vector set and the relationship feature vector set, wherein the derivative triple sample in the derivative triple sample set is obtained by transforming an association relationship or a subject in the triple sample, and the association relationship in the derivative triple sample does not exist between two subjects in the derivative triple sample.

According to another aspect of embodiments herein, there is also provided an electronic device, including: at least one processor; and a memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform a method for determining transaction relevance as described above.

According to another aspect of embodiments herein, there is also provided a machine-readable storage medium storing executable instructions that, when executed, cause the machine to perform a method for determining transaction relevance as described above.

Drawings

A further understanding of the nature and advantages of the present disclosure may be realized by reference to the following drawings. In the drawings, similar components or features may have the same reference numerals. The accompanying drawings, which are included to provide a further understanding of the embodiments of the specification and are incorporated in and constitute a part of this specification, illustrate embodiments of the specification and together with the detailed description serve to explain the embodiments of the specification, but are not intended to limit the embodiments of the specification.

FIG. 1 illustrates a flow chart of one example of a method for determining transaction relevance of embodiments of the present description.

FIG. 2 shows a schematic diagram of one example of a trading scenario.

FIG. 3 shows a schematic diagram of one example of a knowledge-graph of an embodiment of the present description.

FIG. 4 illustrates a flow chart of another example of a method for determining transaction relevance of embodiments of the present description.

FIG. 5 illustrates a schematic diagram of another example of a transaction scenario of an embodiment of the present specification.

Fig. 6 is a block diagram showing an example of a transaction relevance determination apparatus according to an embodiment of the present specification.

Fig. 7 is a block diagram showing an example of a transaction relevance determination unit of the embodiment of the present specification.

FIG. 8 shows a block diagram of an electronic device implementing a method for determining transaction relevance of embodiments of the present description.

The subject matter described herein will be discussed with reference to example embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and thereby implement the subject matter described herein, and are not intended to limit the scope, applicability, or examples set forth in the claims. Changes may be made in the function and arrangement of elements discussed without departing from the scope of the disclosure. Various examples may omit, substitute, or add various procedures or components as needed. In addition, features described with respect to some examples may also be combined in other examples.

As used herein, the term "include" and its variants mean open-ended terms in the sense of "including, but not limited to. The term "based on" means "based at least in part on". The terms "one embodiment" and "an embodiment" mean "at least one embodiment". The term "another embodiment" means "at least one other embodiment". The terms "first," "second," and the like may refer to different or the same object. Other definitions, whether explicit or implicit, may be included below. The definition of a term is consistent throughout the specification unless the context clearly dictates otherwise.

As shown in fig. 1, at block 110, body feature vectors of a transaction initiator body and a transaction recipient body involved in a transaction are obtained from a body feature vector set, and a relationship feature vector between the transaction initiator body and the transaction recipient body is obtained from a relationship feature vector set.

It should be noted that the operation of obtaining the body feature vector and the operation of obtaining the relationship feature vector do not limit the execution order.

In embodiments of the present specification, transactions may include commodity transactions, currency transactions, equity transactions, stock transactions, security transactions, and the like. The transaction body involved in each transaction may include: a transaction initiator and a transaction recipient. The transaction initiator is the party that initiates the transaction, and the transaction recipient is the other party that transacts with the transaction initiator.

FIG. 2 shows a schematic diagram of one example of a trading scenario. As shown in fig. 2, a transaction involves a transaction initiator and a transaction recipient, and the transaction is conducted between the transaction initiator and the transaction recipient.

The transaction initiator may comprise at least one of a type of individual, business, organization, etc., and the transaction recipient may also comprise at least one of a type of individual, business, organization, etc. For one subject, different transaction subjects can be considered in different transactions. For example, a business may be a transaction initiator in one security transaction and a transaction recipient in another currency transaction.

The subject feature vector set may include a plurality of subject feature vectors, each subject feature vector corresponding to a subject, and the subject feature vector corresponding to each subject is used to characterize the subject. In one example, the set of subject feature vectors may include subject feature vectors corresponding to transaction subjects involved in historical transactions.

The historical transactions related to the subject feature vector set may be all the historical transactions that can be acquired, or may be the historical transactions that occur within a specified time period. For example, the subject feature vector set relates to historical transactions in the last year, and the subject feature vectors corresponding to the transaction subjects of the historical transactions in the last year are included in the subject feature vector set.

The historical transactions to which the set of subject feature vectors relates may also be transactions of a specified type. The specified type may be a transaction subject type, for example, the specified type of transaction may be a business-to-business transaction. The specified type may also be a trading object type, for example, the specified type of trade may be a stock exchange, or the like.

In another example, the subject feature vector set may further include subject feature vectors corresponding to other subjects not involved in the above historical transactions. The other main body may be a main body in which no transaction occurs, or may be a transaction main body in which other historical transactions than the above-described historical transactions occur.

For example, the historical transactions related to the subject feature vector set are the transactions in the last year, that is, the subject feature vector set includes the subject feature vectors corresponding to the subjects of the historical transactions in the last year, and the subject feature vector set also includes the subject feature vectors of the subject a, and the historical transactions a performed by the subject a do not belong to the historical transactions in the last year.

The transactions in the embodiments of the present specification may be transactions in historical transactions related to the subject feature vector set, or may not belong to historical transactions related to the subject feature vector set. At this time, the transaction initiator body and the transaction receiver body of the transaction are the bodies characterized by the body feature vectors included in the body feature vector set.

The relation feature vector set may include at least one relation feature vector, each relation feature vector corresponds to an association relation, and the relation feature vector corresponding to each association relation is used to characterize the association relation.

In one example, the set of relationship feature vectors may include relationship feature vectors corresponding to associations involved in historical transactions, the associations involved in transactions being associations between two transaction subjects for the transactions. In another example, the set of relationship feature vectors may further include relationship feature vectors corresponding to associations not involved in the historical transactions.

For example, only two associations are involved in historical transactions: the relationship feature vector set comprises relationship feature vectors corresponding to the relationship between the legal person and the holdings, and relationship feature vectors corresponding to the relationship between the family members.

In the embodiments of the present specification, the subject feature vector set and the relationship feature vector set belong to the same vector space. The dimension of the subject feature vector in the subject feature vector set is consistent with the dimension of the relationship feature vector in the relationship feature vector set. The dimensions of the subject feature vector and the relationship feature vector may be specified. In this way, each subject feature vector and each associated feature vector can be operated, and the operation result can be used to represent the relationship association degree of the corresponding subject for the corresponding association relationship.

In embodiments of the present specification, the set of subject feature vectors and the set of relational feature vectors may be determined based on a knowledge-graph. The knowledge graph can be used for reflecting the incidence relation between the main bodies, the knowledge graph can comprise nodes and edges, the nodes can represent the main bodies, and the edges can represent the incidence relation between the main bodies.

Each edge represents an incidence relation, and when various incidence relations exist between the two main bodies, a plurality of edges are correspondingly arranged between the nodes corresponding to the two main bodies.

FIG. 3 shows a schematic diagram of one example of a knowledge-graph of an embodiment of the present description. As shown in FIG. 3, the knowledge-graph includes a subject A, a subject B, a subject C, a subject D and a subject E, wherein there is only one association between the subject A and the subject B, between the subject B and the subject D and between the subject D and the subject E, and accordingly, there is only one edge on the knowledge-graph between the node of the subject A and the node of the subject B, between the node of the subject B and the node of the subject D and between the node of the subject D and the node of the subject E. There are three associations between principal B and principal C: the relationship of holdings, family membership, and app friends, there are three edges on the knowledge graph between the nodes of principal B and principal C. There are two associations between principal a and principal D: and carrying out holdup relation and fund relation, wherein two edges exist between the nodes of the main body A and the nodes of the main body D on the knowledge graph.

The specific process of determining the set of subject feature vectors and the set of relational feature vectors based on the knowledge-graph is as follows.

After the knowledge-graph is obtained, first, a sample set of triples may be constructed based on the knowledge-graph. The constructed triple sample set comprises all subjects and incidence relations in the knowledge-graph, each triple sample in the triple sample set comprises two subjects and incidence relations between the two subjects, and the form of the triple sample can be as follows: (principal, associative relation, principal).

Taking the above fig. 3 as an example, there are three triad samples including a subject B and a subject C: (principal B, holdings relationship, principal C), (principal B, family membership, principal C), and (principal B, app friends relationship, principal C).

Then, after the triple sample set is constructed, vector characterization learning can be performed by using the triple sample set to obtain a main feature vector set and a relationship feature vector set.

Specifically, all subjects and all association relations in the triple sample set may be traversed, and a subject feature vector set and a relationship feature vector set are correspondingly constructed based on all the traversed subjects and all the association relations, where the subject feature vector set includes all the subjects, and the relationship feature vector set includes all the association relations.

Initialization processing can be performed on the feature vectors corresponding to the various subjects and the various association relationships. Specifically, a subject feature vector of each subject and a dimension of a relationship feature vector of each association relationship are determined, the dimension of the subject feature vector and the dimension of the relationship feature vector are consistent, and the dimension can be specified. And assigning values to the components of each dimension in each main body characteristic vector and the relation characteristic vector, and after assigning values to all the components, normalizing the assignments of all the components in each characteristic vector.

In one example, the assignments of the various components may be arbitrary values. In another example, the assignments for the components may be limited to a specified range of values, so that the values of the components in a feature vector may be equalized without much difference. For example, the specified numerical range is

Where n represents the dimensions of the subject feature vector and the relationship feature vector.

After initialization of each subject feature vector and each relation feature vector is completed, performing vector characterization learning on the subject feature vectors and the relation feature vectors, wherein a loss function adopted by the vector characterization learning is obtained based on the subject feature vectors of the subjects and the vector distance between the subject feature vectors of the corresponding subjects and the sum vectors of the relation feature vectors of the association relations, and the subjects, the corresponding subjects and the association relations belong to the same triple sample.

Specifically, in each cycle of vector characterization learning, current each main feature vector and each relation feature vector are determined, and the loss function of the cycle is calculated by using the following formula:

wherein the content of the first and second substances,

represents the gradient computation for each triplet sample, where the gradient computation is the difference between the vector distance corresponding to the current triplet sample and the vector distance corresponding to that triplet sample in the last cycle.

The representation comprises a current set of subject feature vectors and relationship feature vectors,

respectively represent the feature vectors of the subject,

the feature vector of the relationship is represented,

to represent

And the vector distance between the sum vector of (1) and t.

In order to be a complementary value to the residual error,

may be 0 or a specified value other than 0.

After the loss function is obtained, each component in each subject feature vector and each component in each relationship feature vector may be updated in a gradient descent manner.

Specifically, for each component in the subject feature vector and the relationship feature vector, an updated value of the component is calculated by using the following formula, and the updated value of each component can be used as the value of the component in the next cycle:

wherein the content of the first and second substances,

represents the current value of the ith component in the subject feature vector or the relationship feature vector,

represents the updated value of the ith component in the subject feature vector or the relationship feature vector,

it is indicated that the learning rate is,

may be specified.

After the updated values of the components in the main feature vector and the relation feature vector are obtained through calculation, the updated main feature vector and the updated relation feature vector can be obtained, and when the circulation end condition is not met, the updated main feature vector and the updated relation feature vector are used as the main feature vector and the relation feature vector in the next circulation. Wherein, the cycle end condition may include: the number of cycles reaches a predetermined number and/or the loss function is less than a predetermined threshold.

In one example, after the initialization of each of the subject feature vector and the relationship feature vector is completed, a triplet sample may be selected from the initialized triplet sample set, and the selected triplet sample may be some or all of the triplet samples in the triplet sample set.

And then, replacing the incidence relation in the selected triple sample with a pseudo incidence relation, and obtaining a derived triple sample after replacement. There is no pseudo-associative relationship between two subjects in the derived triple sample.

In the vector characterization learning, the triple samples in the triple sample set can be used as positive samples, and the derived triple samples can be used as negative samples. Therefore, in each cycle of the vector characterization learning, the loss function of the current cycle can be calculated by using the following formula:

wherein the content of the first and second substances,

a set of samples of the derived triples is represented,

representing a pseudo-associative relationship.

In another example, one or two subjects in the selected triple sample may be replaced to obtain a derivative triple sample, and the replaced subjects are also subjects in the subject feature vector set.

In this example, in each cycle of the vector characterization learning, the loss function for this cycle is calculated using the following formula:

wherein the content of the first and second substances,

representing the subject in the derived triple sample,

and

may be both replaced bodies, or only one of them may be a replaced body.

By the scheme of the example, the derived triple samples are added, and the derived triple samples can be used as negative samples in the vector characterization learning, so that the positive and negative samples can be used in the vector characterization learning, and the effect of the vector characterization learning is improved.

Returning to FIG. 1, after the subject feature vector and the relationship feature vector are obtained, at block 120, a transaction relevance for the transaction may be determined based on the obtained subject feature vector and the relationship feature vector. Therefore, the transaction relevance of the transaction can be evaluated in real time by using the main characteristic vector in the main characteristic vector set and the relation characteristic vector in the relation characteristic vector set, and the timeliness of the transaction evaluation is improved.

In embodiments of the present specification, transaction relevance may be used to evaluate whether a transaction is a related transaction. The higher the degree of transaction relevance of a transaction, the greater the probability that the transaction is a relevant transaction. The association transaction is a transaction between a transaction initiator body and a transaction receiver body in an association relationship.

In some transaction application scenarios, the associated transaction may be considered an abnormal transaction, even an illegal transaction. When the transaction is determined to be the related transaction, the transaction platform can give an alarm of the abnormal transaction and cancel the transaction.

In one example, the transaction relevance is for a relevance corresponding to the relationship feature vector, the transaction relevance may be for one or more relevance, and the transaction relevance may be different for different relevance for the same transaction.

Through the embodiment of the specification, the transaction relevance is determined by using the main body characteristic vector and the relation characteristic vector of the transaction initiator main body and the transaction receiver main body, so that the determined transaction relevance aims at the relevance represented by the relation characteristic vector, and under the condition that the transaction relevance only aims at a specific relevance, the influence of other relevance on the transaction relevance is avoided, and the accuracy of the transaction relevance aiming at the specific relevance is improved.

As shown in fig. 4, at block 410, the subject feature vectors of the transaction initiator subject and the transaction recipient subject involved in the transaction are obtained from the set of subject feature vectors, and the relationship feature vector between the transaction initiator subject and the transaction recipient subject is obtained from the set of relationship feature vectors.

In the example shown in fig. 4, the operation of block 410 is the same as the operation of block 110 in fig. 1 and will not be described again here. Only the differences will be described in detail below.

After the body feature vector and the relationship feature vector are obtained, at block 420, for each association relationship, a relationship association degree for the association relationship between the transaction initiator body and the transaction receiver body may be determined based on the body feature vector of one transaction body and the sum vector of the body feature vector of the other pair of transaction bodies and the relationship feature vector corresponding to the association relationship.

In this example, the relationship association degree may represent a degree of association between two subjects for one association relationship. The higher the degree of relationship association, the greater the probability that such an association exists between the two subjects. The relationship association degree and the association relationship are in one-to-one correspondence, and the determined relationship association degree may be different for different association relationships.

In this example, the transaction initiator body and the transaction recipient body are opposite end transaction bodies of each other. Specifically, if one transaction main body is a transaction initiator main body, the other end of the transaction main body is a transaction receiver main body; if one transaction body is the transaction receiver body, the other end of the transaction body is the transaction initiator body.

In one example, for each association relationship, after obtaining a subject feature vector of one transaction subject and a sum vector of the subject feature vector of another pair of transaction subjects and a relationship feature vector corresponding to the association relationship, a similarity between the subject feature vector and the sum vector of the transaction subjects may be calculated, and the similarity may be used as a relationship association degree for the association relationship between the transaction initiator subject and the transaction receiver subject. The greater the similarity, the higher the relationship association.

The similarity between vectors can be calculated by any one of calculation methods such as included angle cosine, Jacard similarity coefficient and correlation coefficient.

In another example, for each association relationship, a vector distance between a subject feature vector of one transaction subject and a sum vector of the subject feature vector of another pair of transaction subjects and a relationship feature vector corresponding to the association relationship may also be determined, and the vector distance is used as a relationship association degree for the association relationship between the transaction initiator subject and the transaction receiver subject.

In this example, the vector distance includes any one of an euclidean distance, a mahalanobis distance, a hamming distance, a manhattan distance, a chebyshev distance, a correlation distance, a jackard distance, and a minkowski distance equidistance.

Taking the Euclidean distance as an example, the principal feature vector of the transaction initiator principal H is (H)₁，h₂，…，h_n) The subject feature vector of the subject T of the transaction receiver is (T)₁，t₂，…，t_n) The relation feature vector of the incidence relation R is (R)₁，r₂，…，r_n) Where n represents the dimensions of the subject feature vector and the relationship feature vector. The Euclidean distance between the main characteristic vector of the transaction initiator main body H and the sum vector of the relation characteristic vectors corresponding to the main characteristic vector of the transaction receiver main body T and the incidence relation R is as follows:

wherein the content of the first and second substances,

the Euclidean distance is expressed, n is the dimension of the main characteristic vector and the relation characteristic vector, and i is an integer which is larger than or equal to 1 and smaller than or equal to n.

In one example of the embodiments of the present specification, the association relationship corresponding to the relationship feature vector in the relationship feature vector set may include one or more specified association relationships. The specified association relationship may be determined according to an application scenario. For example, in an application scenario of an anti-money laundering transaction, the specified association relationship may include a holdings relationship, a funding relationship, a co-media relationship, and the like. The association specified in different application scenarios may be different.

In one example, the association corresponding to the relationship feature vector in the relationship feature vector set may include at least one of the following associations: legal relation, job position relation, stock holding relation, fund relation, medium relation, address relation, communication mode relation, app friend relation and family member relation. Of course, the association relationship of the embodiments of the present specification may also include other types of association relationships.

In this example, the job relationship between the transaction initiator and the transaction recipient may be that both parties are working on the same business or institution, e.g., the transaction initiator and the transaction recipient are tasked with any of the directors, supervisors, and senior managers in the same business. The holdings relationship between the transaction initiator body and the transaction recipient body may be that the two party bodies are stakeholders of the same enterprise. The funding relationship between the transaction initiator body and the transaction recipient body may be that the two party bodies are investors or historical investors of the same business or institution. The medium-medium relationship between the transaction initiator main body and the transaction receiver main body may be that the two main bodies use the same device, for example, the transaction initiator main body and the transaction receiver main body use the same mobile phone to perform transaction operation during transaction. The same address relationship between the transaction initiator and the transaction recipient may be the same address of the enterprise or institution in which the two parties are working. The same communication mode relationship between the transaction initiator main body and the transaction receiver main body can comprise that the two main bodies use the mailbox with the same domain name and/or the host numbers of the telephones are the same. The app buddy relationship between the transaction initiator principal and the transaction recipient principal may be that the two parties are buddy relationships in at least one app, which in one example may comprise a financial business app.

For example, the relationship feature vector set includes nine relationship feature vectors, and one association relationship corresponding to each relationship feature vector is respectively: legal person relationship, job position relationship, stock holding relationship, fund relationship, same medium relationship, same address relationship, same communication mode relationship, app friend relationship and family member relationship. According to the operation of block 420, for each association, a corresponding relationship association may be determined, and thus nine corresponding relationship associations may be determined.

Returning to FIG. 4, after determining the relationship relevance, at block 430, a transaction relevance for the transaction may be determined based on the determined relationship relevance.

In one example of an embodiment of the present specification, when there is at least one of the determined degrees of relationship association that is greater than a first threshold, it may be determined that the transaction is a related transaction.

For example, the association relationship corresponding to the relationship feature vector in the relationship feature vector set includes a corporate relationship, a position relationship, a stock holding relationship, and a fund relationship, and the corresponding four relationship association degrees may be determined, and if the relationship association degree corresponding to only the stock holding relationship is greater than the first threshold, it may also be determined that the transaction is an associated transaction.

In one example, the first threshold values corresponding to all the association relations are the same, so that the relation association degrees corresponding to the association relations can be compared with the same threshold value, the comparison operation is facilitated, the efficiency of the comparison operation is improved, and whether the transaction is the association transaction can be evaluated more efficiently.

In another example, the first threshold corresponding to each association is a threshold matching the association, and the first thresholds corresponding to different associations may be different.

And comparing the relationship association degree corresponding to the association relation with a corresponding first threshold value aiming at each association relation to determine whether the relationship association degree is greater than the first threshold value, and if the relationship association degree is greater than the first threshold value, determining that the transaction is an associated transaction.

In this example, the first threshold corresponding to each association may be determined according to the degree of influence of the association in the association transaction evaluation. The higher the influence of the association, the higher the degree of association between the association and the associated transaction.

For example, if the influence of the holdings relationship and the funding relationship in the evaluation of the associated transaction is high, the first threshold corresponding to the holdings relationship and the funding relationship may be set low, so that for the two associated relationships, the transaction may be determined as the associated transaction even if the determined corresponding relationship is low in association. And the influence degree of the same-address relationship, the same-communication-mode relationship and the app friend relationship is low, the first threshold corresponding to the same-address relationship, the same-communication-mode relationship and the app friend relationship can be set to be high, so that for the three association relationships, the transaction can be determined as an associated transaction only if the relationship association degree is high.

The first threshold value matched with each incidence relation is set through the scheme of the example, and the first threshold value corresponding to each incidence relation can reflect the importance degree of the incidence relation in the incidence transaction evaluation, so that the incidence transaction can be evaluated based on the importance degree of each incidence relation, and the accuracy of the incidence transaction evaluation is improved.

In one example of an embodiment of the present specification, when all of the determined relationship degrees of association are not greater than the first threshold value, it may be determined that the transaction is not an associated transaction.

In another example, the determination that the transaction is a related transaction may also be based on a difference of the respective relational relevance degrees from the first threshold. Specifically, the difference between each relationship association degree and the first threshold is calculated, and when there are more than a specified number of relationship association degrees and the difference between the first threshold and the second threshold is less than the difference threshold, it may be determined that the transaction is an associated transaction.

For example, the association relationship includes a corporate relationship, a position relationship, a holdings relationship and a fund relationship, the first threshold corresponding to all the four association relationships is 5, the difference threshold is set to be 0.2, the specified number is 2, the method can determine that the relationship association degree corresponding to the corporate relationship is 4.9, the relationship association degree corresponding to the position relationship is 3, the relationship association degree corresponding to the holdings relationship is 4.95, the relationship association degree corresponding to the fund relationship is 4.85, and the difference between the relationship association degree corresponding to the corporate relationship, the holdings relationship and the fund relationship and the first threshold is less than 0.2, so that the transaction is determined to be an association transaction.

If there is no more than a specified number of relationship relevancy degrees less than the difference threshold from the first threshold, it may be determined that the transaction is not a related transaction.

By the above example, when all the association relations are not up to the standard of determining the association transaction, the policies of various association relations are further integrated to evaluate whether the transaction is an association transaction, so that the evaluation mechanism of the transaction association is perfected.

In one example of an embodiment of the present specification, the transaction initiator body may include at least two transaction bodies, and/or the transaction recipient body may include at least two transaction bodies. At this time, the determined relationship association degree includes: and the relationship association degree of each transaction main body in the transaction initiator main body and each transaction main body in the transaction receiver main body aiming at various association relations.

Taking fig. 5 as an example, fig. 5 shows a schematic diagram of another example of a transaction scenario of an embodiment of the present specification. As shown in fig. 5, the transaction initiator body includes transaction bodies a and B, and the transaction receiver body includes transaction bodies C and D, then the relationship association degree for an association relationship may include: the relationship association degree between the transaction body A and the transaction body C, the relationship association degree between the transaction body A and the transaction body D, the relationship association degree between the transaction body B and the transaction body C, and the relationship association degree between the transaction body B and the transaction body D.

In this example, the relationship association between the transaction initiator body and the transaction recipient body for one association may include a plurality of relationship associations, and then when determining the transaction association according to the relationship association, for this association, when at least one of the plurality of relationship associations is greater than the first threshold, then it may be determined that the transaction is an associated transaction.

Fig. 6 is a block diagram showing an example of a transaction relevance determination apparatus 600 according to an embodiment of the present specification. As shown in fig. 6, the transaction relevance determination apparatus 600 may include a feature vector acquisition unit 610 and a transaction relevance determination unit 620.

The feature vector obtaining unit 610 is configured to obtain the subject feature vectors of the transaction initiator body and the transaction receiver body involved in the transaction from the subject feature vector set, and obtain the relationship feature vector between the transaction initiator body and the transaction receiver body from the relationship feature vector set. The main body feature vector set and the relation feature vector set belong to the same vector space, and the main body feature vector set and the relation feature vector set are determined based on a knowledge graph for reflecting the incidence relation between the main bodies, wherein the knowledge graph comprises nodes and edges, the nodes represent the main bodies, and the edges represent the incidence relation between the main bodies. The operation of the feature vector acquisition unit 610 may refer to the operation of the block 110 described above with reference to fig. 1.

The transaction relevance determination unit 620 is configured to determine the transaction relevance of the transaction based on the obtained subject feature vector and the relationship feature vector. The operation of the transaction relevance determination unit 620 may refer to the operation of block 120 described above with reference to fig. 1.

In one example, the transaction relevance determination apparatus 600 may further include a triple sample set construction unit and a vector characterization learning unit.

The triple sample set construction unit is configured to construct a triple sample set based on the knowledge-graph, each triple sample in the triple sample set including two subjects and an association between the two subjects.

The vector characterization learning unit is configured to perform vector characterization learning by using the triple sample set to obtain a subject feature vector set and a relationship feature vector set. In an example, the vector characterization learning unit may be further configured to perform vector characterization learning by using the triple sample set and a derived triple sample set of the triple sample set to obtain a subject feature vector set and a relationship feature vector set, where a derived triple sample in the derived triple sample set is obtained by transforming an association relationship or a subject in the triple sample, and an association relationship in the derived triple sample does not exist between two subjects in the derived triple sample.

In one example, the loss function employed for the vector characterization learning is derived based on a subject feature vector of a subject and a vector distance between a sum vector of the subject feature vector of the corresponding subject and a relationship feature vector of an association, wherein the subject, the corresponding subject and the association belong to the same triple sample.

Fig. 7 is a block diagram illustrating an example of the transaction relevance determining unit 620 according to the embodiment of the present specification. In this example, each relationship feature vector included in the relationship feature vector set corresponds to an association relationship. The transaction relevance determination unit 620 may include a relationship relevance determination module 621 and a transaction relevance determination module 625.

The relationship association degree determining module 621 is configured to determine, for each association relationship, a relationship association degree between the transaction initiator main body and the transaction receiver main body for the association relationship based on the main body feature vector of one transaction main body and the sum vector of the main body feature vector of the other pair of transaction main bodies and the relationship feature vector corresponding to the association relationship. The operations of the relationship association determination module 621 may refer to the operations of block 420 described above with reference to fig. 4.

In one example, the transaction initiator body comprises at least two transaction bodies, and/or the transaction recipient body comprises at least two transaction bodies; at this time, the determined relationship association degree may include: and the relationship association degree of each transaction main body in the transaction initiator main body and each transaction main body in the transaction receiver main body aiming at various association relations.

In an example, the relationship association degree determining module 621 may be further configured to determine, for each association relationship, a vector distance between the subject feature vector of one transaction subject and the subject feature vector of another pair of transaction subjects and a sum vector of the relationship feature vectors corresponding to the association relationship, as the relationship association degree between the transaction initiator subject and the transaction receiver subject for the association relationship.

In one example, the association may include at least one of the following associations: legal relation, job position relation, stock holding relation, fund relation, medium relation, address relation, communication mode relation, app friend relation and family member relation.

The transaction relevance determination module 625 may be configured to determine a transaction relevance for the transaction according to the determined relationship relevance. The operations of the transaction relevance determination module 625 may refer to the operations of block 430 described above with reference to fig. 4.

In one example, the transaction relevance determination module 625 may be further configured to determine that the transaction is a relevance transaction when at least one of the determined relationship relevance degrees is greater than a first threshold.

In one example, the transaction relevance determination module 625 may be further configured to determine that the transaction is a related transaction if there are more than a specified number of relationship relevance degrees that differ from the first threshold by less than a difference threshold when all of the determined relationship relevance degrees are not greater than the first threshold.

Embodiments of methods and apparatus for determining transaction relevance according to embodiments of the present description are described above with reference to fig. 1-7.

The means for determining transaction relevance of the embodiments of the present specification may be implemented in hardware, or may be implemented in software, or a combination of hardware and software. The software implementation is taken as an example, and is formed by reading corresponding computer program instructions in the storage into the memory for operation through the processor of the device where the software implementation is located as a logical means. In embodiments of the present specification, the means for determining transaction relevance may be implemented, for example, using an electronic device.

Fig. 8 illustrates a block diagram of an electronic device 800 implementing a method for determining transaction relevance of embodiments of the present description.

As shown in fig. 8, electronic device 800 may include at least one processor 810, storage (e.g., non-volatile storage) 820, memory 830, and communication interface 840, and the at least one processor 810, storage 820, memory 830, and communication interface 840 are connected together via a bus 850. The at least one processor 810 executes at least one computer-readable instruction (i.e., the elements described above as being implemented in software) stored or encoded in memory.

In one embodiment, computer-executable instructions are stored in the memory that, when executed, cause the at least one processor 810 to: acquiring main body characteristic vectors of a transaction initiator main body and a transaction receiver main body related to the transaction from the main body characteristic vector set, and acquiring a relation characteristic vector between the transaction initiator main body and the transaction receiver main body from the relation characteristic vector set; and determining a transaction relevance of the transaction based on the obtained subject feature vector and the relationship feature vector.

It should be appreciated that the computer-executable instructions stored in the memory, when executed, cause the at least one processor 810 to perform the various operations and functions described above in connection with fig. 1-7 in the various embodiments of the present description.

According to one embodiment, a program product, such as a machine-readable medium, is provided. A machine-readable medium may have instructions (i.e., elements described above as being implemented in software) that, when executed by a machine, cause the machine to perform various operations and functions described above in connection with fig. 1-7 in the various embodiments of the present specification.

Specifically, a system or apparatus may be provided which is provided with a readable storage medium on which software program code implementing the functions of any of the above embodiments is stored, and causes a computer or processor of the system or apparatus to read out and execute instructions stored in the readable storage medium.

In this case, the program code itself read from the readable medium can realize the functions of any of the above-described embodiments, and thus the machine-readable code and the readable storage medium storing the machine-readable code constitute a part of the embodiments of the present specification.

Computer program code required for the operation of various portions of the present specification may be written in any one or more programming languages, including an object oriented programming language such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C + +, C #, VB, NET, Python, and the like, a conventional programming language such as C, Visual Basic 2003, Perl, COBOL 2002, PHP, and ABAP, a dynamic programming language such as Python, Ruby, and Groovy, or other programming languages. The program code may execute on the user's computer, or on the user's computer as a stand-alone software package, or partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any network format, such as a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet), or in a cloud computing environment, or as a service, such as a software as a service (SaaS).

Examples of the readable storage medium include floppy disks, hard disks, magneto-optical disks, optical disks (e.g., CD-ROMs, CD-R, CD-RWs, DVD-ROMs, DVD-RAMs, DVD-RWs), magnetic tapes, nonvolatile memory cards, and ROMs. Alternatively, the program code may be downloaded from a server computer or from the cloud via a communications network.

The foregoing description has been directed to specific embodiments of this disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.

Not all steps and elements in the above flows and system structure diagrams are necessary, and some steps or elements may be omitted according to actual needs. The execution order of the steps is not fixed, and can be determined as required. The apparatus structures described in the above embodiments may be physical structures or logical structures, that is, some units may be implemented by the same physical entity, or some units may be implemented by a plurality of physical entities, or some units may be implemented by some components in a plurality of independent devices.

The term "exemplary" used throughout this specification means "serving as an example, instance, or illustration," and does not mean "preferred" or "advantageous" over other embodiments. The detailed description includes specific details for the purpose of providing an understanding of the described technology. However, the techniques may be practiced without these specific details. In some instances, well-known structures and devices are shown in block diagram form in order to avoid obscuring the concepts of the described embodiments.

Although the embodiments of the present disclosure have been described in detail with reference to the accompanying drawings, the embodiments of the present disclosure are not limited to the specific details of the embodiments, and various simple modifications may be made to the technical solutions of the embodiments of the present disclosure within the technical spirit of the embodiments of the present disclosure, and all of them fall within the scope of the embodiments of the present disclosure.

The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other variations without departing from the scope of the disclosure. Thus, the description is not intended to be limited to the examples and designs described herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for determining transaction relevance, comprising:

acquiring main feature vectors of a transaction initiator main body and a transaction receiver main body related to a transaction from a main feature vector set, and acquiring a relation feature vector between the transaction initiator main body and the transaction receiver main body from a relation feature vector set; and

determining a transaction relevance for the transaction based on the obtained subject feature vector and relationship feature vector,

the main body feature vector set and the relation feature vector set belong to the same vector space, and the main body feature vector set and the relation feature vector set are determined based on a knowledge graph for reflecting the association relationship between the main bodies, wherein the knowledge graph comprises nodes and edges, the nodes represent the main bodies, and the edges represent the association relationship between the main bodies.

2. The method of claim 1, wherein the set of relational feature vectors includes each relational feature vector corresponding to an association,

determining a transaction relevance of the transaction based on the obtained subject feature vector and relationship feature vector comprises:

for each incidence relation, determining the relation incidence degree of the incidence relation between the transaction initiator main body and the transaction receiver main body based on the main body characteristic vector of one transaction main body and the sum vector of the main body characteristic vector of the other end of transaction main body and the relation characteristic vector corresponding to the incidence relation; and

and determining the transaction relevance of the transaction according to the determined relationship relevance.

3. The method of claim 2, wherein for each association, determining a relationship association degree between the transaction initiator main body and the transaction receiver main body for the association based on a subject feature vector of one transaction main body and a sum vector of the subject feature vector of another pair of transaction main bodies and a relationship feature vector corresponding to the association comprises:

and determining a vector distance between the main body characteristic vector of one transaction main body and the sum vector of the main body characteristic vector of the other end of transaction main body and the relation characteristic vector corresponding to the association relation as the relation association degree between the transaction initiator main body and the transaction receiver main body aiming at the association relation aiming at each association relation.

4. A method as claimed in claim 3, wherein said vector distance comprises one of a euclidean distance, a mahalanobis distance, a hamming distance, a manhattan distance, a chebyshev distance, a correlation distance, a jaccard distance, and a minkowski distance.

5. The method of claim 2, wherein the association comprises at least one of: legal relation, job position relation, stock holding relation, fund relation, medium relation, address relation, communication mode relation, app friend relation and family member relation.

6. The method of claim 5, wherein determining a transaction relevance for the transaction according to the determined relationship relevance comprises:

determining that the transaction is a related transaction when at least one of the determined relationship association degrees is greater than a first threshold.

7. The method of claim 6, wherein the first threshold corresponding to each association is a threshold matching the association.

8. The method of claim 5, wherein determining a transaction relevance for the transaction according to the determined relationship relevance comprises:

and when all the relation association degrees in the determined relation association degrees are not larger than the first threshold, if the difference value between the relation association degrees which are more than the specified number and the first threshold is smaller than the difference threshold, determining that the transaction is the associated transaction.

9. The method of claim 5, wherein the transaction initiator body comprises at least two transaction bodies and/or the transaction recipient body comprises at least two transaction bodies;

the determined relationship association degree comprises: and the relationship association degree of each transaction main body in the transaction initiator main body and each transaction main body in the transaction receiver main body aiming at various association relations.

10. The method of claim 1, wherein the set of subject feature vectors and the set of relationship feature vectors are derived by:

constructing a set of triple samples based on the knowledge-graph, each triple sample in the set of triple samples including two subjects and an association between the two subjects; and

and performing vector characterization learning by using the triple sample set to obtain the main feature vector set and the relationship feature vector set.

11. The method of claim 10, wherein the loss function employed for the learning of the vector characterization is obtained based on a subject feature vector of the subject and a vector distance between a sum vector of the subject feature vector of the corresponding subject and a relational feature vector of the association,

wherein the main body, the corresponding main body and the association relationship belong to the same triple sample.

12. The method of claim 10, wherein performing vector characterization learning using the set of triplet samples to obtain the set of subject feature vectors and the set of relationship feature vectors comprises:

performing vector characterization learning by using the triple sample set and a derived triple sample set of the triple sample set to obtain the subject feature vector set and the relationship feature vector set,

the correlation relation or the main body in the derived triple sample set is obtained by transforming the correlation relation or the main body in the triple sample, and the correlation relation in the derived triple sample does not exist between the two main bodies in the derived triple sample set.

13. An apparatus for determining transaction relevance, comprising:

the system comprises a characteristic vector acquisition unit, a transaction processing unit and a transaction processing unit, wherein the characteristic vector acquisition unit is used for acquiring main characteristic vectors of a transaction initiator main body and a transaction receiver main body related to a transaction from a main characteristic vector set and acquiring a relation characteristic vector between the transaction initiator main body and the transaction receiver main body from a relation characteristic vector set; and

a transaction relevance determination unit that determines a transaction relevance of the transaction based on the acquired subject feature vector and relationship feature vector,

14. The apparatus of claim 13, wherein the set of relational feature vectors includes each relational feature vector corresponding to an association,

the transaction relevance determination unit includes:

a relation association degree determining module, configured to determine, for each association relation, a relation association degree between the transaction initiator main body and the transaction receiver main body for the association relation based on a main body feature vector of one transaction main body and a sum vector of a main body feature vector of another peer transaction main body and a relation feature vector corresponding to the association relation; and

and the transaction relevance determining module is used for determining the transaction relevance of the transaction according to the determined relation relevance.

15. The apparatus of claim 14, wherein the relational relevance determination module is to:

16. The apparatus of claim 14, wherein the association comprises at least one of: legal relation, job position relation, stock holding relation, fund relation, medium relation, address relation, communication mode relation, app friend relation and family member relation.

17. The apparatus of claim 16, wherein the transaction relevance determination module:

18. The apparatus of claim 16, wherein the transaction relevance determination module:

19. The apparatus of claim 16, wherein the transaction initiator body comprises at least two transaction bodies and/or the transaction recipient body comprises at least two transaction bodies;

20. The apparatus of claim 13, further comprising:

the triple sample set construction unit is used for constructing a triple sample set based on the knowledge graph, and each triple sample in the triple sample set comprises two subjects and an incidence relation between the two subjects; and

and the vector characterization learning unit is used for performing vector characterization learning by using the triple sample set to obtain the main feature vector set and the relationship feature vector set.

21. The apparatus of claim 20, wherein the loss function employed for the vector characterization learning is derived based on a subject feature vector of the subject and a vector distance between a sum vector of the subject feature vector of the corresponding subject and a relational feature vector of the association,

22. The apparatus of claim 20, wherein the vector characterization learning unit:

23. An electronic device, comprising:

at least one processor, and

a memory coupled with the at least one processor, the memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform the method of any of claims 1-12.

24. A machine-readable storage medium storing executable instructions that, when executed, cause the machine to perform the method of any one of claims 1 to 12.