CN109272312A - Method and apparatus for transaction risk detecting real-time - Google Patents
Method and apparatus for transaction risk detecting real-time Download PDFInfo
- Publication number
- CN109272312A CN109272312A CN201710584386.9A CN201710584386A CN109272312A CN 109272312 A CN109272312 A CN 109272312A CN 201710584386 A CN201710584386 A CN 201710584386A CN 109272312 A CN109272312 A CN 109272312A
- Authority
- CN
- China
- Prior art keywords
- transaction
- historical trading
- transaction feature
- single account
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q20/00—Payment architectures, schemes or protocols
- G06Q20/38—Payment protocols; Details thereof
- G06Q20/382—Payment protocols; Details thereof insuring higher security of transaction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q20/00—Payment architectures, schemes or protocols
- G06Q20/38—Payment protocols; Details thereof
Landscapes
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Engineering & Computer Science (AREA)
- Finance (AREA)
- Strategic Management (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Security & Cryptography (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
Abstract
The present invention relates to on-line payment technologies, in particular to the computer readable storage medium of the method for detecting real-time transaction risk, the server and client side of implementation this method and the computer program comprising implementing this method during online transaction.It is comprised the steps of according to the method for transaction risk detecting real-time of one aspect of the invention and establishes the historical trading model for corresponding to the single account based on historical trading associated with single account；And historical trading model is provided so that it judges the risk of the single account currently traded to client.
Description
Technical field
The present invention relates to on-line payment technologies, in particular to the method for detecting real-time transaction risk during online transaction,
Implement the server and client side of this method and the computer readable storage medium of the computer program comprising implementing this method.
Background technique
On-line payment is due to its convenience and with the close correlation of daily life and by the favor of consumer, and
Transaction payment mode as mainstream.However, on-line payment also results in more transaction swindling risks.Industry is mainly adopted at present
The transaction swindling risk of on-line payment is coped with rule-based method and based on the method for disaggregated model.
But all there is clearly disadvantageous place in above two method.For example, rule-based method depends on expertise,
Subjective factor is stronger, and the accuracy of risk judgment and the ability level of expert are closely related.Disaggregated model mainly passes through
Obtained from for statistical analysis to the transaction feature of a large amount of accounts, it is preferably objective to have compared with the rule that expert works out
Property, but disaggregated model is substantially a kind of statistical model, between different user otherness (in many cases, this
Species diversity is apparent and very important) then seem helpless.In addition, when many transactions occur simultaneously, if this
A little transaction require to handle in real time, then system will be made to bear huge processing pressure, and lead to the reduction of processing speed.
Therefore there is an urgent need to provide the method and dress of a kind of prevention transaction risk that can overcome above-mentioned various disadvantages
It sets.
Summary of the invention
It is an object of the present invention to provide a kind of methods for detecting real-time transaction risk, with calculating speed
Fastly, the advantages that recognition accuracy is high.
It is comprised the steps of according to the method for transaction risk detecting real-time of one aspect of the invention
The historical trading model for corresponding to the single account is established based on historical trading associated with single account；And
Historical trading model is provided to client so that it judges the risk of the single account currently traded.
Preferably, in the above-mentioned methods, the step of establishing the historical trading model for corresponding to single account is completed beyond the clouds.
Preferably, in the above-mentioned methods, further comprise the following steps:
Historical trading model is periodically or non-periodically updated using the historical transaction record that single account increases newly.
Preferably, in the above-mentioned methods, the historical trading model is hidden Markov model, establishes historical trading model
The step of include:
The first transaction sequence for indicating the observable behavior state of historical trading of single account is generated as hidden Ma Erke
The observation state set of husband's model；And
The hidden Markov model is trained using the first transaction sequence to establish the historical trading model of the single account.
Preferably, in the above-mentioned methods, the step of the first transaction sequence of generation includes:
The transaction feature vector of every historical trading of single account is generated to obtain multiple transaction feature vectors；
To obtained multiple transaction feature vectors progress clustering processings to obtain one or more transaction feature classifications,
In each transaction feature classification correspond to an observable behavior state；And
Transaction feature classification belonging to every historical trading is determined according to respective transaction feature vector to obtain the list
First transaction sequence of a account.
Preferably, in the above-mentioned methods, the transaction feature classification belonging to determining as follows:
Calculate the transaction feature vector of every historical trading of single account and the similitude of each transaction feature classification；With
And
The transaction feature classification for corresponding to maximum comparability is determined as belonging to the transaction feature vector of this historical trading
Transaction feature classification.
Preferably, in the above-mentioned methods, hidden Markov model is trained to include the following steps: using the first transaction sequence
Setting corresponds to the quantity of the hidden state of the hidden Markov model of single account；
Set the initial value of the parameter of the hidden Markov model, wherein the parameter includes the transfer between hidden state
The initial probability distribution of probability matrix, the probability matrix of hidden state to observation state and hidden state；
For the first transaction sequence of the single account, based on the maximum optimization mesh of probability of occurrence for making the first transaction sequence
The optimal value to determine the parameter is marked, the historical trading model for corresponding to the single account is thus established.
Preferably, in the above-mentioned methods, the initial probability distribution of the transition probability matrix between hidden state and hidden state
Initial value be set as probability values, the initial value of the probability matrix of hidden state to observation state is according to transaction feature classification
Distribution determines.
Preferably, in the above-mentioned methods, the historical trading corresponding to single account is saved using non-relational database
Model.
It is comprised the steps of according to the method for detecting transaction risk of another aspect of the invention
Client obtains the historical trading model for corresponding to single account from cloud, which is based on and the list
A associated historical trading of account and establish；
Client is judged using the risk currently traded of the historical trading model to the single account；And
Export the judging result for the risk currently traded.
Preferably, in the above-mentioned methods, the historical trading model is hidden Markov model, the hidden Markov model
Observation state collection be combined into the first transaction sequence for indicating the observable behavior state of historical trading of single account, utilize first
Transaction sequence trains the hidden Markov model to establish the historical trading model of the single account.
Preferably, in the above-mentioned methods, the observable behavior state of historical trading determines as follows:
The transaction feature vector of every historical trading of single account is generated to obtain multiple transaction feature vectors；
To obtained multiple transaction feature vectors progress clustering processings to obtain one or more transaction feature classifications,
In each transaction feature classification correspond to an observable behavior state；And
Transaction feature classification belonging to every historical trading is determined according to respective transaction feature vector.
Preferably, in the above-mentioned methods, using the historical trading model to the risk currently traded of single account into
Row judgement includes the following steps:
Determine transaction feature classification belonging to the transaction feature vector currently traded；
It is handed over by replacing the first of the single account with transaction feature classification belonging to the transaction feature vector currently traded
Transaction feature classification belonging to the transaction feature vector of historical trading in easy sequence and generate the second transaction sequence；
Determine the probability of occurrence of the second transaction sequence；And
By being compared to the probability of occurrence of the probability of occurrence of the second transaction sequence and the first transaction sequence to the list
The risk of a account currently traded is judged.
Preferably, in the above-mentioned methods, the historical trading being replaced in the first transaction sequence corresponds to the first transaction sequence
The earliest historical trading of middle exchange hour.
Preferably, in the above-mentioned methods, the transaction feature classification belonging to determining as follows:
Calculate the transaction feature vector of single account currently traded and the similitude of each transaction feature classification；And
The transaction feature classification for corresponding to maximum comparability is determined as friendship belonging to the transaction feature vector currently traded
Easy feature classification.
Preferably, in the above-mentioned methods, the similitude is characterized with one of following index: the transaction currently traded
Euclidean distance between feature vector and each transaction feature class center vector, the transaction feature vector currently traded and each
COS distance and the transaction feature vector currently traded and each transaction feature class between transaction feature class center vector
Other Jie Kade similarity.
Preferably, in the above-mentioned methods, transaction feature classification belonging to current transaction, and each friendship are determined by client
Easy feature classification and corresponding center vector are stored in client in the form of relation database table.
It is also an object of the present invention to provide a kind of for detecting the server of transaction risk, has identification accurate
Spend the advantages that high.
Server according to another aspect of the invention includes memory, processor and is stored on the memory simultaneously
The computer program that can be run on the processor is to execute method as described above.
It is also an object of the present invention to provide a kind of for detecting the client of transaction risk, has identification accurate
Spend the advantages that high.
Client according to another aspect of the invention includes memory, processor and is stored on the memory simultaneously
The computer program that can be run on the processor is to execute method as described above.
It is also an object of the present invention to provide a kind of computer readable storage mediums, store computer program thereon,
The program realizes method as described above when being executed by processor.
Detailed description of the invention
Above-mentioned and/or other aspects and advantage of the invention will be become by the description of the various aspects below in conjunction with attached drawing
It is more clear and is easier to understand, the same or similar unit, which is adopted, in attached drawing is indicated by the same numeral.Attached drawing includes:
Fig. 1 is the flow chart according to the method for transaction risk detecting real-time of one embodiment of the invention.
Fig. 2 is the flow chart that can be applied to the first transaction sequence generation method of embodiment illustrated in fig. 1.
Fig. 3 is the flow chart that can be applied to the hidden Markov model training method of embodiment illustrated in fig. 1.
Fig. 4 is the flow chart according to the method for detecting transaction risk of another embodiment of the present invention.
Fig. 5 is the stream that can be applied to the transaction risk judgment method based on hidden Markov model of embodiment illustrated in fig. 4
Cheng Tu.
Fig. 6 is the block diagram according to the server for transaction risk detecting real-time of another embodiment of the present invention.
Fig. 7 is the block diagram according to the client for transaction risk detecting real-time of another embodiment of the present invention.
Specific embodiment
Referring to which illustrates the attached drawings of illustrative examples of the present invention to more fully illustrate the present invention.But this hair
It is bright to be realized by different form, and be not construed as being only limitted to each embodiment given herein.The above-mentioned each implementation provided
Example is intended to make the disclosure of this paper comprehensively complete, and protection scope of the present invention is more fully communicated to those skilled in the art
Member.
In the present specification, the term of such as "comprising" and " comprising " etc indicates to want in addition to having in specification and right
Asking has in book directly and other than the unit clearly stated and step, technical solution of the present invention be also not excluded for having not by directly or
The situation of the other units clearly stated and step.
According to one aspect of the present invention, it is established based on historical trading associated with single account and corresponds to the account
Historical trading model, and during on-line payment, utilize established historical trading model to judge the current of the account
Transaction whether there is risk.That is, historical trading model or classification mould that each account can possess for its custom-made
Type makes it possible to the perfect transaction feature for portraying the account, this effectively increases the accuracy of risk judgment.It preferably, can be with
Corresponding historical trading model is periodically or non-periodically updated using the historical transaction record that account increases newly, further increases and sentences
Disconnected accuracy.
Other side according to the invention, above-mentioned historical trading model are hidden Markov model, wherein the hidden Ma Er
Can the observation state collection of husband's model be combined into the first transaction sequence for indicating the observable behavior state of historical trading of single account,
And the hidden Markov model is trained using the first transaction sequence to establish the historical trading model of the account.
One hidden Markov model can usually be described by a five-tuple (Q, O, A, B, π), wherein Q is to hide shape
State set, O are observation state set, and transition probability matrix of the A between hidden state, B is hidden state to the general of observation state
Rate matrix, π are the initial probability distribution of hidden state.According to the invention there are one aspects, when utilization hidden Markov mould
When historical trading characteristic of the type to characterize account, hidden state corresponds to trading activity state, and observation state corresponds to every friendship
The trading activity state of easy observable.
According to the invention there are one aspect, the foundation of historical trading model is completed beyond the clouds, and is based on historical trading
The judgement of transaction risk made by model is then completed in client.This mode can not only play the powerful computing capability in cloud, but also
The operating pressure of background system when can reduce large-scale concurrent transaction, so that it is guaranteed that providing quickly, risk detects energy in real time
Power.
Fig. 1 is the flow chart according to the method for transaction risk detecting real-time of one embodiment of the invention.Preferably
But not necessarily, method shown in FIG. 1 can execute at server or backstage transaction processing system beyond the clouds.
The process of method shown in FIG. 1 starts from step 110.In this step, generating indicates that the history of single account is handed over
First transaction sequence of easy observable behavior state, first transaction sequence can be used as the observation state of hidden Markov model
Set.
Step 120 is subsequently entered, trains the hidden Markov model using the first transaction sequence that step 110 generates,
To establish the historical trading model for corresponding to the single account.
Step 130 is subsequently entered, provides historical trading model to client for its current transaction to the single account
Risk judged.Preferably, non-relational database can be used to save the historical trading model corresponding to each account.
Optionally, the method flow of the present embodiment also includes step 140.In this step, it is increased newly using single account
Historical transaction record periodically or non-periodically updates its historical trading model.
Fig. 2 is the flow chart that can be applied to the first transaction sequence generation method of embodiment illustrated in fig. 1.Preferably but it is non-must
Must ground, method shown in Fig. 2 can execute at server or backstage transaction processing system beyond the clouds.
As shown in Fig. 2, the transaction feature vector for generating every historical trading of single account is more to obtain in step 210
A transaction feature vector.In the present embodiment, transaction feature vector is referred to by one or more observables of an account
The vector that transaction feature is constituted.When the example of the transaction feature of observable includes but is not limited to transaction amount, loco, transaction
Between and consumption type etc..It should be pointed out that in the present embodiment, structure (the vector dimension of the transaction feature vector of different accounts
Several and component type) it may be the same or different.
Preferably for the transaction feature with continuity value, the numerical value of discretization can be mapped as.
It should be pointed out that multiple transaction features in every transaction are all useful for the analysis of transaction risk, but
It is the feature that basic hidden Markov model can not handle more than one label.In response to this, in the present embodiment will
The transaction feature of multiple observables of every transaction is mapped as single transaction feature label or transaction feature classification (below will be this
Map operation is also known as the clustering processing to transaction feature vector).Obtained transaction feature classification is handled by clustering
Each corresponds to one of observable behavior state of hidden Markov model.
For the purpose of clustering processing, process shown in Fig. 2 enters step 220, to multiple transaction features of single account
Vector carries out clustering processing, to obtain one or more transaction feature classifications, wherein each classification, which represents one group, has phase
The trading activity of antitype.The set of transaction feature classification is denoted as { C below1,C2…Ck, wherein k is the quantity of classification, often
A classification all has corresponding class center.Preferably, in the present embodiment can using K-means algorithm to transaction feature to
Amount carries out clustering processing.
Preferably, the coordinate of the obtained transaction feature classification of clustering processing and its corresponding class center can be with relationship type
The form of tables of data stores.Since the memory space that tables of data occupies is smaller, in user installation or application can be updated
Corresponding client is downloaded to when program.
Step 230 is subsequently entered, each transaction feature vector of single account is mapped into corresponding transaction feature class
Not, the first transaction sequence of the account is thus obtained.
It preferably, can be as follows by a transaction feature DUAL PROBLEMS OF VECTOR MAPPING to corresponding transaction feature classification: first
The similitude of the transaction feature vector Yu each transaction feature classification is first calculated, it is then that the transaction for corresponding to maximum comparability is special
Sign classification is determined as transaction feature classification belonging to the transaction feature vector.Preferably, similitude is with one of following index
To characterize: Euclidean distance, transaction feature vector between transaction feature vector and each transaction feature class center vector and every
The outstanding of COS distance and transaction feature vector and each transaction feature classification between a transaction feature class center vector blocks
Moral similarity.
Fig. 3 is the flow chart that can be applied to the hidden Markov model training method of embodiment illustrated in fig. 1.It is shown in Fig. 3
Embodiment in, train hidden Markov model using aforementioned first transaction sequence.Preferably, but not necessarily, shown in Fig. 3
Method can execute at server or backstage transaction processing system beyond the clouds.
Process shown in Fig. 3 starts from step 310.In this step, setting corresponds to the hidden Markov of single account
The quantity of the hidden state of model.Assuming that hidden state is S.
Step 320 is subsequently entered, the initial value of the parameter of hidden Markov model is set.In the present embodiment, parameter packet
Include transition probability matrix A, the probability matrix B of hidden state to observation state and the probability of hidden state between hidden state
It is distributed π.
Preferably, the initial value of the initial probability distribution of the transition probability matrix between hidden state and hidden state is set as
Probability values, that is, the probability of each hidden state is set to 1/S, it is hidden to be transferred to another from a hidden state
Hiding shape probability of state is also 1/S (situation of its own is transferred to including hidden state).
It, can be according to transaction feature classification preferably for the initial value of the probability matrix B of hidden state to observation state
Distribution determine.Specifically, transaction count corresponding to each transaction feature classification can be accounted for total friendship for single account
The ratio of easy number is set as the probability of observation state corresponding to each hidden state to the transaction feature classification.
It should be pointed out that the setting value of the quantity of hidden state and the initial value of above-mentioned parameter only influence model training
Efficiency, but do not influence the validity of model training.
Step 330 is subsequently entered, in this step, using the first transaction sequence of each account, based on making the first transaction
The maximum optimization aim of the probability of occurrence of sequence optimizes parameter A, B and π, thus establishes the history friendship for corresponding to the account
Easy model.
It preferably, can be using Baum-Welch algorithm come Optimal Parameters λ (A, B, π).Specifically, can be by successively holding
Row the following steps are trained above-mentioned hidden Markov model.
Step a: for given observation sequence or the first transaction sequence O={ o1,o2…oT, according to current parameter lambda,
Variable α forward is calculated using following formula (1) and (2)t(i) and backward variable βt+1(i):
αt(i)=P (O1,O2…Ot,qt=i | λ) (1)
βt+1(i)=P (Ot+1,Ot+2…OT|qt=i, λ) (2)
Step b: it is calculated using following formula (3) when sequence t is located at state qiAnd sequence t+1 is located at state qjWhen probability ξt
(i, j):
Step c: it is calculated using following formula (4) when sequence t is located at state qiWhen probability γt(i):
Step d: original state (t=1 moment hidden state q is reevaluated using following formula (5)iProbability):
Step e: transition probability matrix is reevaluated using following formula (6)
Step f: output probability matrix is reevaluated, whereinIt is Q from state qjIssue observation state okExpectation
State q is reached with QjDesired ratio:
Wherein
Repeat above-mentioned steps a-f untilConvergence, so that Optimal Parameters λ={ A, B, π } is obtained,
That is, completing the training of hidden Markov model.
Matrix A, B and π in the parameter as corresponding to each account be not on format it is fixed, need in training
Stage adjusts optimization, therefore in this embodiment, it is preferred that the non-relational of such as MongoDB and Hbase etc can be used
Database stores above-mentioned Optimal Parameters λ={ A, B, π }.
In order to save memory space, for each account, transaction feature corresponding to the R transaction occurred recently is only stored
Classification is as the first transaction sequence or observation sequence.That is, the first transaction sequence of each account is designed to a sky
Between be R round-robin queue.The value of R can be adjusted according to practical application.Cloud server or backstage trading processing
System periodically can execute update operation by the database to the first transaction sequence for storing each account, wherein every time only
First transaction sequence of the account for having transaction record to change recently is updated.Preferably, institute can be recorded for every transaction
Corresponding transaction feature classification attachment timestamp is to indicate renewal time.Cloud server will be deleted when executing update operation
Timestamp be greater than duration parameters Pe Account History, this not only can to avoid the deficiency of storage space of database, it is also possible that
The historical trading model of account more reflects Recent Activity behavior.
Fig. 4 is the flow chart according to the method for transaction risk detecting real-time of another embodiment of the present invention.It is preferred that
Ground but not necessarily, method shown in Fig. 4 can execute at client.
The process of method shown in Fig. 4 starts from step 410.In this step, client is obtained from cloud corresponds to list
The historical trading model of a account.The mode of establishing of historical relevance Trading Model has been described above and has made adequately by Fig. 1-3
Description, details are not described herein again.
Step 420 is subsequently entered, the historical trading model that client is obtained using step 410 is to the current of the single account
The risk of transaction is judged.
Finally in step 430, client exports the risk currently traded to cloud server or backstage transaction processing system
Judging result.Optionally, also judging result can be presented on its display interface in client.
Fig. 5 is the stream that can be applied to the transaction risk judgment method based on hidden Markov model of embodiment illustrated in fig. 4
Cheng Tu.Preferably, but not necessarily, method shown in fig. 5 can execute at client.
Process shown in fig. 5 starts from step 510.In this step, determine that the transaction currently traded an of account is special
Levy transaction feature classification belonging to vector.Preferably, corresponding transaction feature classification can be determined as follows: being counted first
The similitude for calculating the transaction feature vector and each transaction feature classification currently traded, then will correspond to the friendship of maximum comparability
Easy feature classification is determined as transaction feature classification belonging to the transaction feature vector.Preferably, similitude is in following index
One kind characterizes: Euclidean distance, transaction feature vector between transaction feature vector and each transaction feature class center vector
COS distance and transaction feature vector and each transaction feature classification between each transaction feature class center vector
Jie Kade similarity.
Step 520 is subsequently entered, by should with the replacement of transaction feature classification belonging to the transaction feature classification currently traded
Transaction feature classification belonging to historical trading in first transaction sequence of account and generate the second transaction sequence.Below by first
Transaction sequence is denoted as O={ o1,o2…oR, wherein each element in sequence is arranged according to chronological order.
In this embodiment, it is preferred that the historical trading being replaced in the first transaction sequence corresponds to the first transaction sequence
The earliest historical trading of middle exchange hour.Therefore the second transaction sequence can be denoted as O={ o2,o3…oR+1, wherein earliest history
The transaction feature classification O of transaction1The transaction feature classification O currently to be tradedR+1Substitution.
Then in step 530, probability of occurrence ρ '=P (o of the second transaction sequence O ' is determined2,o3…oR+1|λ)。
Step 540 is subsequently entered, by the probability of occurrence ρ ' of the second transaction sequence=P (o2,o3…oR+1| λ) it trades with first
Probability of occurrence ρ=P (o of sequence1,o2…oR| λ) it is compared to judge the risk of the single account currently traded.
Preferably, the change rate of probability of occurrence can be definedFoundation as transaction risk judgement.Specifically, such as
Fruit Δ ρ >=θ, then there are bigger differences with historical trading model for the current transaction of explanation, therefore are judged as suspicious transaction；If Δ ρ <
θ judges currently transaction then as arm's length dealing, and θ is a preset threshold value here.
Finally in step 550, the judging result of step 540 is sent to cloud server or backstage trading processing by client
System.
Fig. 6 is the block diagram according to the server for transaction risk detecting real-time of another embodiment of the present invention.
Server 60 shown in fig. 6 includes memory 610, processor 620 and is stored on memory 610 and can locate
The computer program 630 run on reason device 620, wherein computer program 630 is by running on processor 620 can be performed
As above by the method for embodiment described in Fig. 1-3.
Fig. 7 is the block diagram according to the client for transaction risk detecting real-time of another embodiment of the present invention.
Client 70 shown in Fig. 7 includes memory 710, processor 720 and is stored on memory 710 and can locate
The computer program 730 run on reason device 720, wherein computer program 730 is by running on processor 720 can be performed
As above by the method for embodiment described in Figure 4 and 5.In the present embodiment, client can be the POS machine of mobile phone, acquirer
Or code reader.
According to one aspect of the present invention, a kind of computer readable storage medium is provided, stores computer program thereon, it should
The method by embodiment described in Fig. 1-3 is realized when program is executed by processor.
According to one aspect of the present invention, a kind of computer readable storage medium is provided, stores computer program thereon, it should
The method by embodiment described in Figure 4 and 5 is realized when program is executed by processor.
Compared with prior art, the above embodiment of the present invention has the advantage that
1, transaction risk is judged by using the historical trading model customized for each account, effectively prevent general
The deficiencies of Model suitability is poor.
2, by establishing the historical trading model of account levels using hidden Markov algorithm and according to current transaction and friendship
The probability difference of easy historical models judges transaction risk, improves the accuracy of judgement.
3, it is completed due to that can operate the judgement of transaction risk to be arranged at client, it is concurrent to avoid extensive transaction
When to the operating pressure of backstage transaction processing system, while also improving distinguishing speed.
Embodiments and examples set forth herein is provided, to be best described by the reality according to this technology and its specific application
Example is applied, and thus enables those skilled in the art to implement and using the present invention.But those skilled in the art will
Know, provides above description and example only for the purposes of illustrating and illustrating.The description proposed is not intended to cover the present invention
Various aspects or limit the invention to disclosed precise forms.
In view of the above, the scope of the present disclosure is determined by following claims.
Claims (22)
1. a kind of method for transaction risk detecting real-time, which is characterized in that comprise the steps of
The historical trading model for corresponding to the single account is established based on historical trading associated with single account；And
Historical trading model is provided to client so that it judges the risk of the single account currently traded.
2. the step of the method for claim 1, wherein establishing the historical trading model for corresponding to single account is beyond the clouds
It completes.
3. the method for claim 1, wherein further comprising the following steps:
Historical trading model is periodically or non-periodically updated using the historical transaction record that single account increases newly.
4. the method as described in any one of claim 1-3, wherein the historical trading model is hidden Markov mould
Type, the step of establishing historical trading model include:
The first transaction sequence for indicating the observable behavior state of historical trading of single account is generated as hidden Markov mould
The observation state set of type；And
The hidden Markov model is trained using the first transaction sequence to establish the historical trading model of the single account.
5. method as claimed in claim 4, wherein generate the first transaction sequence the step of include:
The transaction feature vector of every historical trading of single account is generated to obtain multiple transaction feature vectors；
Clustering processing is carried out to obtain one or more transaction feature classifications, wherein often to obtained multiple transaction feature vectors
A transaction feature classification corresponds to an observable behavior state；And
Transaction feature classification belonging to every historical trading is determined according to respective transaction feature vector to obtain the single account
First transaction sequence at family.
6. method as claimed in claim 5, wherein the transaction feature classification belonging to determining as follows:
Calculate the transaction feature vector of every historical trading of single account and the similitude of each transaction feature classification；And
The transaction feature classification for corresponding to maximum comparability is determined as friendship belonging to the transaction feature vector of this historical trading
Easy feature classification.
7. method as claimed in claim 5, wherein it includes following for training hidden Markov model using the first transaction sequence
Step:
Setting corresponds to the quantity of the hidden state of the hidden Markov model of single account；
Set the initial value of the parameter of the hidden Markov model, wherein the parameter includes the transition probability between hidden state
The initial probability distribution of matrix, the probability matrix of hidden state to observation state and hidden state；
For the first transaction sequence of the single account, based on make the maximum optimization aim of probability of occurrence of the first transaction sequence come
It determines the optimal value of the parameter, thus establishes the historical trading model for corresponding to the single account.
8. the method for claim 7, wherein the probability of transition probability matrix and hidden state between hidden state
The initial value of distribution is set as probability values, and the initial value of the probability matrix of hidden state to observation state is according to transaction feature class
Other distribution determines.
9. method as claimed in claim 4, wherein save the history corresponding to single account using non-relational database
Trading Model.
10. a kind of method for detecting transaction risk, which is characterized in that comprise the steps of
Client obtains the historical trading model for corresponding to single account from cloud, which is based on and the single account
The associated historical trading in family and establish；
Client is judged using the risk currently traded of the historical trading model to the single account；And
Export the judging result for the risk currently traded.
11. method as claimed in claim 10, wherein the historical trading model is hidden Markov model, the hidden Ma Er
Can the observation state collection of husband's model be combined into the first transaction sequence for indicating the observable behavior state of historical trading of single account,
The hidden Markov model is trained using the first transaction sequence to establish the historical trading model of the single account.
12. method as claimed in claim 11, wherein the observable behavior state of historical trading determines as follows:
The transaction feature vector of every historical trading of single account is generated to obtain multiple transaction feature vectors；
Clustering processing is carried out to obtain one or more transaction feature classifications, wherein often to obtained multiple transaction feature vectors
A transaction feature classification corresponds to an observable behavior state；And
Transaction feature classification belonging to every historical trading is determined according to respective transaction feature vector.
13. method as claimed in claim 12, wherein utilize the historical trading model currently trading to single account
Risk carries out judgement and includes the following steps:
Determine transaction feature classification belonging to the transaction feature vector currently traded；
By the first transaction sequence for replacing the single account with transaction feature classification belonging to the transaction feature vector currently traded
Transaction feature classification belonging to the transaction feature vector of historical trading in column and generate the second transaction sequence；
Determine the probability of occurrence of the second transaction sequence；And
By being compared to the probability of occurrence of the probability of occurrence of the second transaction sequence and the first transaction sequence to the single account
The risk currently traded at family is judged.
14. method as claimed in claim 13, wherein the historical trading being replaced in the first transaction sequence corresponds to first and hands over
The earliest historical trading of exchange hour in easy sequence.
15. method as claimed in claim 13, wherein the transaction feature classification belonging to determining as follows:
Calculate the transaction feature vector of single account currently traded and the similitude of each transaction feature classification；And
The transaction feature classification for corresponding to maximum comparability is determined as the spy of transaction belonging to the transaction feature vector currently traded
Levy classification.
16. method as claimed in claim 15, wherein the similitude is characterized with one of following index: current to hand over
Euclidean distance between easy transaction feature vector and each transaction feature class center vector, the transaction feature currently traded to
COS distance and the transaction feature vector currently traded and each friendship between amount and each transaction feature class center vector
The Jie Kade similarity of easy feature classification.
17. method as claimed in claim 15, wherein transaction feature classification belonging to current transaction is determined by client, and
And each transaction feature classification and corresponding center vector are stored in client in the form of relation database table.
18. a kind of server comprising memory, processor and is stored on the memory and can transport on the processor
Capable computer program, which is characterized in that execute the method as described in any one of claim 1-9.
19. a kind of client comprising memory, processor and is stored on the memory and can transport on the processor
Capable computer program, which is characterized in that execute the method as described in any one of claim 10-17.
20. client as claimed in claim 19, wherein the client is the POS machine or barcode scanning of mobile phone, acquirer
Device.
21. a kind of computer readable storage medium, stores computer program thereon, which is characterized in that the program is held by processor
The method as described in any one of claim 1-9 is realized when row.
22. a kind of computer readable storage medium, stores computer program thereon, which is characterized in that the program is held by processor
The method as described in any one of claim 10-17 is realized when row.
Priority Applications (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710584386.9A CN109272312B (en) | 2017-07-18 | 2017-07-18 | Method and device for real-time detection of transaction risk |
| PCT/CN2018/094883 WO2019015499A1 (en) | 2017-07-18 | 2018-07-06 | Method and device for real-time detection of transaction risk |
| TW107124221A TWI734920B (en) | 2017-07-18 | 2018-07-13 | Method and device for real-time detection of transaction risk |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710584386.9A CN109272312B (en) | 2017-07-18 | 2017-07-18 | Method and device for real-time detection of transaction risk |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109272312A true CN109272312A (en) | 2019-01-25 |
| CN109272312B CN109272312B (en) | 2021-07-13 |
Family
ID=65015362
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710584386.9A Active CN109272312B (en) | 2017-07-18 | 2017-07-18 | Method and device for real-time detection of transaction risk |
Country Status (3)
| Country | Link |
|---|---|
| CN (1) | CN109272312B (en) |
| TW (1) | TWI734920B (en) |
| WO (1) | WO2019015499A1 (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111311408A (en) * | 2020-02-10 | 2020-06-19 | 支付宝(杭州)信息技术有限公司 | Electronic transaction attribute identification method and device |
| CN116611829A (en) * | 2023-07-21 | 2023-08-18 | 山东美丽乡村云计算有限公司 | Consumption supervision system based on block chain |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110458576B (en) * | 2019-07-31 | 2022-12-20 | 同济大学 | Network transaction anti-fraud method integrating advance prediction and in-process detection |
| CN110991871A (en) * | 2019-11-29 | 2020-04-10 | 深圳前海微众银行股份有限公司 | Risk monitoring method, device, equipment and computer readable storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020099594A1 (en) * | 2000-05-26 | 2002-07-25 | Nicholas Heard | Method and apparatus for determining one or more statistical estimators of customer behavior |
| CN105809502A (en) * | 2014-12-30 | 2016-07-27 | 阿里巴巴集团控股有限公司 | Transaction risk detection method and apparatus |
| CN106485396A (en) * | 2016-09-09 | 2017-03-08 | 北京科技大学 | A kind of safety in production hidden troubles removing system |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110112869A1 (en) * | 2009-11-09 | 2011-05-12 | Revolutionary E-Commerce Systems, Inc. | Online transaction hosting apparatus and method |
| CN103577413B (en) * | 2012-07-20 | 2017-11-17 | 阿里巴巴集团控股有限公司 | Search result ordering method and system, search results ranking optimization method and system |
| CN106251214A (en) * | 2016-08-02 | 2016-12-21 | 东软集团股份有限公司 | account monitoring method and device |
| CN106485348A (en) * | 2016-09-22 | 2017-03-08 | 中国银联股份有限公司 | A kind of Forecasting Methodology of transaction data and device |
-
2017
- 2017-07-18 CN CN201710584386.9A patent/CN109272312B/en active Active
-
2018
- 2018-07-06 WO PCT/CN2018/094883 patent/WO2019015499A1/en active Application Filing
- 2018-07-13 TW TW107124221A patent/TWI734920B/en active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020099594A1 (en) * | 2000-05-26 | 2002-07-25 | Nicholas Heard | Method and apparatus for determining one or more statistical estimators of customer behavior |
| CN105809502A (en) * | 2014-12-30 | 2016-07-27 | 阿里巴巴集团控股有限公司 | Transaction risk detection method and apparatus |
| CN106485396A (en) * | 2016-09-09 | 2017-03-08 | 北京科技大学 | A kind of safety in production hidden troubles removing system |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111311408A (en) * | 2020-02-10 | 2020-06-19 | 支付宝(杭州)信息技术有限公司 | Electronic transaction attribute identification method and device |
| CN116611829A (en) * | 2023-07-21 | 2023-08-18 | 山东美丽乡村云计算有限公司 | Consumption supervision system based on block chain |
| CN116611829B (en) * | 2023-07-21 | 2023-11-14 | 山东美丽乡村云计算有限公司 | Consumption supervision system based on block chain |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109272312B (en) | 2021-07-13 |
| TWI734920B (en) | 2021-08-01 |
| WO2019015499A1 (en) | 2019-01-24 |
| TW201909047A (en) | 2019-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109272312A (en) | Method and apparatus for transaction risk detecting real-time | |
| CN112035541A (en) | Client image drawing method and device, computer readable storage medium and terminal equipment | |
| US20240144374A1 (en) | Mapping consumer ownership of financial assets to geographic localities and computer-implemented methods and computer systems thereof | |
| CN110175850A (en) | A kind of processing method and processing device of Transaction Information | |
| CN107657500A (en) | Stock recommends method and server | |
| CN111882420A (en) | Generation method of response rate, marketing method, model training method and device | |
| CN108932646A (en) | User tag verification method, device and electronic equipment based on operator | |
| CN109934700A (en) | A kind of method and device of arbitrage detecting | |
| Lee et al. | Hidden markov models for forex trends prediction | |
| CN111695938A (en) | Product pushing method and system | |
| CN113144624B (en) | Data processing method, device, equipment and storage medium | |
| US20200364537A1 (en) | Systems and methods for training and executing a recurrent neural network to determine resolutions | |
| CN105023041B (en) | Data clustering method and B2B platform customer priorities acquisition methods, device | |
| CN107844874A (en) | Enterprise operation problem analysis system and its method | |
| CN116187808A (en) | Electric power package recommendation method based on virtual power plant user-package label portrait | |
| CN104636489B (en) | The treating method and apparatus of attribute data is described | |
| CN110796379B (en) | Risk assessment method, device and equipment of business channel and storage medium | |
| CN108446985A (en) | The share system and method for Portfolio Selection | |
| CN108389134A (en) | The monitoring system and method for Portfolio Selection | |
| CN114298825A (en) | Method and device for extremely evaluating repayment volume | |
| CN108305174A (en) | Method for processing resource, device, storage medium and computer equipment | |
| CN111898708A (en) | Transfer learning method and electronic equipment | |
| CN109272373B (en) | Brand recommendation method based on computer | |
| CN111598671A (en) | Commodity recommendation method based on human-computer interaction | |
| CN107092599A (en) | A kind of method and apparatus for being used to provide the user knowledge information |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 40003681 Country of ref document: HK |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |