WO2019015499A1 - 用于交易风险实时侦测的方法和装置 - Google Patents

用于交易风险实时侦测的方法和装置 Download PDF

Info

Publication number
WO2019015499A1
WO2019015499A1 PCT/CN2018/094883 CN2018094883W WO2019015499A1 WO 2019015499 A1 WO2019015499 A1 WO 2019015499A1 CN 2018094883 W CN2018094883 W CN 2018094883W WO 2019015499 A1 WO2019015499 A1 WO 2019015499A1
Authority
WO
WIPO (PCT)
Prior art keywords
transaction
historical
single account
model
feature
Prior art date
Application number
PCT/CN2018/094883
Other languages
English (en)
French (fr)
Inventor
李旭瑞
邱雪涛
赵金涛
胡奕
Original Assignee
中国银联股份有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 中国银联股份有限公司 filed Critical 中国银联股份有限公司
Publication of WO2019015499A1 publication Critical patent/WO2019015499A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q20/00Payment architectures, schemes or protocols
    • G06Q20/38Payment protocols; Details thereof
    • G06Q20/382Payment protocols; Details thereof insuring higher security of transaction
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q20/00Payment architectures, schemes or protocols
    • G06Q20/38Payment protocols; Details thereof

Definitions

  • the present invention relates to online payment techniques, and more particularly to a method for detecting transaction risk in real time during an online transaction process, a server and client implementing the method, and a computer readable storage medium containing the computer program implementing the method.
  • Online payment has been favored by consumers because of its convenience and close correlation with daily life, and has become the mainstream transaction payment method.
  • online payments have also led to more transaction fraud risks.
  • the industry mainly adopts a rule-based approach and a classification model-based approach to deal with the risk of transaction fraud for online payments.
  • the rule-based approach relies on expert experience, subjective factors are strong, and the accuracy of risk judgment is closely related to the skill level of the expert.
  • the classification model is mainly obtained by statistical analysis of the transaction characteristics of a large number of accounts, and has better objectivity than the rules formulated by experts, but the classification model is essentially a statistical model for the difference between different users. Sex (in many cases, this difference is obvious and cannot be ignored) is seemingly powerless.
  • Sex in many cases, this difference is obvious and cannot be ignored
  • the system will be subjected to huge processing pressure and the processing speed will be reduced.
  • a method for real-time detection of transaction risk in accordance with an aspect of the present invention includes the following steps:
  • a historical transaction model is provided to the client for its determination of the risk of the current transaction for that single account.
  • the step of establishing a historical transaction model corresponding to a single account is completed in the cloud.
  • the method further comprises the following steps:
  • the historical transaction model is updated periodically or irregularly with historical transaction records added by a single account.
  • the historical transaction model is a hidden Markov model
  • the steps of establishing a historical transaction model include:
  • the hidden Markov model is trained using the first transaction sequence to establish a historical transaction model for the single account.
  • the step of generating the first transaction sequence comprises:
  • a transaction feature category to which each historical transaction belongs is determined according to a respective transaction feature vector to obtain a first transaction sequence for the single account.
  • the belonging transaction feature category is determined in the following manner:
  • the transaction feature category corresponding to the maximum similarity is determined as the transaction feature category to which the transaction feature vector of the historical transaction belongs.
  • training the hidden Markov model with the first transaction sequence comprises the following steps:
  • the parameter includes a transition probability matrix between hidden states, a probability matrix of the hidden state to the observed state, and an initial probability distribution of the hidden state;
  • an optimization value for the parameter is determined based on an optimization goal that maximizes the probability of occurrence of the first transaction sequence, thereby establishing a historical transaction model corresponding to the single account.
  • the initial values of the transition probability matrix between the hidden states and the initial probability distribution of the hidden state are set to equal probability values, and the initial values of the probability matrix of the hidden state to the observed state are determined according to the distribution of the transaction feature categories.
  • a non-relational database is employed to hold a historical transaction model corresponding to a single account.
  • a method for detecting transaction risk in accordance with another aspect of the present invention includes the following steps:
  • the client obtains from the cloud a historical transaction model corresponding to a single account, the historical transaction model being established based on historical transactions associated with the single account;
  • the client uses the historical transaction model to determine the risk of the current transaction for the individual account
  • the historical transaction model is a hidden Markov model
  • the observation state set of the hidden Markov model is a first transaction sequence indicating an observable behavior state of a historical transaction of a single account
  • a transaction sequence is used to train the hidden Markov model to establish a historical transaction model for the single account.
  • the observable behavioral state of the historical transaction is determined in the following manner:
  • the transaction feature categories to which each historical transaction belongs are determined according to the respective transaction feature vectors.
  • determining the risk of the current transaction of the single account by using the historical transaction model comprises the following steps:
  • the risk of the current transaction of the single account is determined by comparing the probability of occurrence of the second transaction sequence to the probability of occurrence of the first transaction sequence.
  • the historical transaction that is replaced in the first transaction sequence corresponds to the historical transaction with the oldest transaction time in the first transaction sequence.
  • the belonging transaction feature category is determined in the following manner:
  • the transaction feature category corresponding to the maximum similarity is determined as the transaction feature category to which the transaction feature vector of the current transaction belongs.
  • the similarity is characterized by one of the following indicators: the Euclidean distance between the transaction feature vector of the current transaction and the central vector of each transaction feature category, and the transaction feature vector of the current transaction and The cosine distance between the center vectors of each transaction feature category and the transaction feature vector of the current transaction and the Jaked similarity for each transaction feature category.
  • the transaction feature category to which the current transaction belongs is determined by the client, and each transaction feature category and the corresponding center vector are saved in the form of a relational data table at the client.
  • a server in accordance with another aspect of the present invention includes a memory, a processor, and a computer program stored on the memory and operative on the processor to perform the method as described above.
  • Still another object of the present invention is to provide a client for detecting transaction risk, which has the advantages of high recognition accuracy and the like.
  • a client in accordance with another aspect of the present invention includes a memory, a processor, and a computer program stored on the memory and operative on the processor to perform the method as described above.
  • FIG. 1 is a flow chart of a method for real-time detection of transaction risk in accordance with one embodiment of the present invention.
  • FIG. 2 is a flow chart of a first transaction sequence generation method applicable to the embodiment of FIG. 1.
  • FIG. 3 is a flow chart of a hidden Markov model training method applicable to the embodiment shown in FIG. 1.
  • FIG. 4 is a flow chart of a method for detecting a transaction risk in accordance with another embodiment of the present invention.
  • FIG. 5 is a flow chart of a method for judging a transaction risk based on a hidden Markov model applicable to the embodiment shown in FIG. 4.
  • FIG. 6 is a block diagram of a server for real-time detection of transaction risk in accordance with another embodiment of the present invention.
  • FIG. 7 is a block diagram of a client for real-time detection of transaction risk in accordance with another embodiment of the present invention.
  • a historical transaction model corresponding to the account is established based on a historical transaction associated with a single account, and in the online payment process, the established historical transaction model is utilized to determine whether the current transaction of the account is at risk . That is to say, each account can have a historical transaction model or classification model tailored to it, so that the transaction characteristics of the account can be perfectly portrayed, which effectively improves the accuracy of risk judgment.
  • the historical transaction model added by the account may be used to update the corresponding historical transaction model periodically or irregularly to further improve the accuracy of the judgment.
  • the historical transaction model is a hidden Markov model
  • the observation state set of the hidden Markov model is a first transaction sequence representing an observable behavior state of a historical transaction of a single account
  • the hidden Markov model is trained using the first transaction sequence to establish a historical transaction model for the account.
  • a hidden Markov model can usually be described by a five-tuple (Q, O, A, B, ⁇ ), where Q is a set of hidden states, O is a set of observation states, and A is a transition probability matrix between hidden states. B is the probability matrix from the hidden state to the observed state, and ⁇ is the initial probability distribution of the hidden state.
  • the hidden Markov model when utilized to characterize the historical transaction characteristics of the account, the hidden state corresponds to the trading behavior state, and the observed state corresponds to the observable trading behavior state of each transaction.
  • the establishment of the historical transaction model is done in the cloud, and the transaction risk judgment based on the historical transaction model is completed on the client side.
  • This approach can not only take advantage of the powerful computing power of the cloud, but also reduce the operating pressure of the back-end system during large-scale concurrent transactions, thus ensuring fast real-time risk detection capabilities.
  • FIG. 1 is a flow chart of a method for real-time detection of transaction risk in accordance with one embodiment of the present invention.
  • the method illustrated in Figure 1 can be performed at a cloud server or a background transaction processing system.
  • a first transaction sequence is generated that represents an observable behavioral state of a historical transaction for a single account, which may be used as a set of observation states for a hidden Markov model.
  • step 120 the hidden Markov model is trained using the first transaction sequence generated in step 110 to establish a historical transaction model corresponding to the single account.
  • the client is provided with a historical transaction model for its determination of the risk of the current transaction for the single account.
  • a non-relational database can be employed to hold a historical transaction model corresponding to each account.
  • the method process of this embodiment further includes step 140.
  • step 140 historical transaction models are updated periodically or irregularly using historical transactions added by a single account.
  • FIG. 2 is a flow chart of a first transaction sequence generation method applicable to the embodiment of FIG. 1.
  • the method illustrated in Figure 2 can be performed at a cloud server or a background transaction processing system.
  • a transaction feature vector for each historical transaction of a single account is generated to obtain a plurality of transaction feature vectors.
  • the transaction feature vector refers to a vector consisting of one or more observable transaction features of an account. Examples of observable transaction characteristics include, but are not limited to, transaction amount, transaction location, transaction time, and consumption type. It should be noted that in this embodiment, the structure of the transaction feature vector (vector dimension and component type) of different accounts may be the same or different.
  • transaction features with continuity values they can be mapped to discretized values.
  • each transaction characteristic in each transaction is useful for the analysis of transaction risk, but the basic hidden Markov model cannot handle more than one marker feature.
  • multiple observable transaction features of each transaction are mapped into a single transaction feature tag or transaction feature category (hereinafter this mapping operation is also referred to as clustering of transaction feature vectors). deal with).
  • Each of the transaction feature categories obtained by the clustering process corresponds to one of the observable behavior states of the hidden Markov model.
  • step 220 clustering multiple transaction feature vectors of a single account to obtain one or more transaction feature categories, wherein each category represents one Groups have similar patterns of trading behavior.
  • the set of transaction feature categories is denoted as ⁇ C 1 , C 2 ... C k ⁇ , where k is the number of categories, and each category has a corresponding category center.
  • the K-means algorithm may be used to cluster the transaction feature vectors.
  • the transaction feature categories obtained by the clustering process and the coordinates of the corresponding category centers may be stored in the form of a relational data table. Because the data table occupies less storage space, it can be downloaded to the appropriate client when the user installs or updates the application.
  • each transaction feature vector of a single account is mapped to a corresponding transaction feature category, thereby obtaining a first transaction sequence for the account.
  • a transaction feature vector may be mapped to a corresponding transaction feature category in the following manner: first, the similarity of the transaction feature vector to each transaction feature category is calculated, and then the transaction feature category corresponding to the maximum similarity is determined as the The transaction feature category to which the transaction feature vector belongs.
  • the similarity is characterized by one of the following indicators: the Euclidean distance between the transaction feature vector and the center vector of each transaction feature category, the cosine distance between the transaction feature vector and the center vector of each transaction feature category. And the similarity of the transaction feature vector to each of the transaction feature categories.
  • FIG. 3 is a flow chart of a hidden Markov model training method applicable to the embodiment shown in FIG. 1.
  • the hidden Markov model is trained using the aforementioned first transaction sequence.
  • the method illustrated in Figure 3 can be performed at a cloud server or a background transaction processing system.
  • the flow shown in FIG. 3 begins at step 310.
  • the number of hidden states of the hidden Markov model corresponding to a single account is set. Assume that the hidden state is S.
  • the parameters include a transition probability matrix A between hidden states, a probability matrix B from a hidden state to an observed state, and an initial probability distribution ⁇ of a hidden state.
  • the initial value of the transition probability matrix between the hidden states and the initial probability distribution of the hidden state is set to an equal probability value, that is, the initial probability of each hidden state is set to 1/S, and is transferred from a hidden state.
  • the probability of going to another hidden state is also 1/S (including the case where the hidden state is transferred to itself).
  • the initial value of the probability matrix B for the hidden state to the observed state may be determined according to the distribution of the transaction feature categories. Specifically, for a single account, the ratio of the number of transactions corresponding to each transaction feature category to the total number of transactions may be set as the probability of each hidden state to the observation state corresponding to the transaction feature category.
  • step 330 in which the parameters A, B and ⁇ are optimized based on the first transaction sequence of each account based on the optimization target that maximizes the probability of occurrence of the first transaction sequence, thereby establishing corresponding to the account Historical trading model.
  • the Baum-Welch algorithm can be employed to optimize the parameter ⁇ (A, B, ⁇ ).
  • the above hidden Markov model can be trained by sequentially performing the following steps.
  • Step b Calculate the probability ⁇ t (i, j) when the sequence t is in the state q i and the sequence t+1 is in the state q j using the following equation (3):
  • Step c Calculate the probability ⁇ t (i) when the sequence t is in the state q i using the following equation (4):
  • Step e Re-estimate the transition probability matrix by using the following equation (6)
  • Step f re-estimating the output probability matrix, wherein Q is emitted from the observation state state q j o K Q with a desired state q j reaches the desired ratio:
  • the first transaction sequence for each account is designed as a circular queue with space R.
  • the value of R can be adjusted according to the actual application.
  • the cloud server or background transaction processing system may periodically perform an update operation on the database storing the first transaction sequence for each account, wherein only the first transaction sequence of the account with the most recent transaction record change is updated at a time.
  • a time stamp may be attached to the transaction feature category corresponding to each transaction record to indicate the update time.
  • FIG. 4 is a flow chart of a method for real-time detection of transaction risk in accordance with another embodiment of the present invention. Preferably, but not necessarily, the method illustrated in Figure 4 can be performed at the client.
  • the flow of the method illustrated in FIG. 4 begins at step 410.
  • the client obtains a historical transaction model corresponding to a single account from the cloud.
  • the manner in which the historical transaction model is established has been fully described above with reference to Figures 1-3 and will not be described again here.
  • step 420 the client uses the historical transaction model acquired in step 410 to determine the risk of the current transaction for the single account.
  • step 430 the client outputs a determination result of the risk of the current transaction to the cloud server or the background transaction processing system.
  • the client can also present the result of the judgment on its display interface.
  • FIG. 5 is a flow chart of a method for judging a transaction risk based on a hidden Markov model applicable to the embodiment shown in FIG. 4.
  • the method illustrated in Figure 5 can be performed at the client.
  • the transaction feature category to which the transaction feature vector of the current transaction of an account belongs is determined.
  • the corresponding transaction feature category may be determined in the following manner: first, the similarity between the transaction feature vector of the current transaction and each transaction feature category is calculated, and then the transaction feature category corresponding to the maximum similarity is determined as the transaction feature vector belongs to.
  • Transaction feature category Preferably, the similarity is characterized by one of the following indicators: the Euclidean distance between the transaction feature vector and the center vector of each transaction feature category, the cosine distance between the transaction feature vector and the center vector of each transaction feature category. And the similarity of the transaction feature vector to each of the transaction feature categories.
  • a second transaction sequence is generated by replacing the transaction feature category to which the historical transaction in the first transaction sequence of the account belongs in the transaction feature category to which the transaction feature category of the current transaction belongs.
  • the historical transaction that is replaced in the first transaction sequence corresponds to the historical transaction with the earliest trading time in the first transaction sequence.
  • the probability of occurrence of the second transaction sequence ⁇ ' P (o 2 , o 3 ... o R + 1
  • ⁇ ) and the probability of occurrence of the first transaction sequence ⁇ P (o 1 , o 2 ...o R
  • the rate of change of the probability of occurrence can be defined As the basis for the judgment of trading risk. Specifically, if ⁇ ⁇ ⁇ , it means that the current transaction has a large difference from the historical transaction model, so it is judged as a suspicious transaction; if ⁇ ⁇ ⁇ , it is judged that the current transaction is a normal transaction, where ⁇ is a preset threshold. .
  • step 550 the client sends the determination result of step 540 to the cloud server or the background transaction processing system.
  • FIG. 6 is a block diagram of a server for real-time detection of transaction risk in accordance with another embodiment of the present invention.
  • the server 60 shown in FIG. 6 includes a memory 610, a processor 620, and a computer program 630 stored on the memory 610 and executable on the processor 620, wherein the computer program 630 is executable by operating on the processor 620
  • FIG. 7 is a block diagram of a client for real-time detection of transaction risk in accordance with another embodiment of the present invention.
  • the client 70 shown in FIG. 7 includes a memory 710, a processor 720, and a computer program 730 stored on the memory 710 and executable on the processor 720, wherein the computer program 730 is executable by operating on the processor 720 as above
  • the client may be a POS machine or a scanner of a mobile phone or an acquirer.
  • a computer readable storage medium having stored thereon a computer program that, when executed by a processor, implements the method of the embodiment described with reference to Figures 1-3.
  • a computer readable storage medium having stored thereon a computer program that, when executed by a processor, implements the method of the embodiment described with reference to Figures 4 and 5.

Abstract

本发明涉及在线支付技术,特别涉及在线交易过程中实时侦测交易风险的方法、实施该方法的服务器和客户端以及包含实施该方法的计算机程序的计算机可读存储介质。按照本发明一个方面的用于交易风险实时侦测的方法包含下列步骤:基于与单个账户相关联的历史交易建立对应于该单个账户的历史交易模型;以及向客户端提供历史交易模型以供其对该单个账户的当前交易的风险进行判断。

Description

用于交易风险实时侦测的方法和装置 技术领域
本发明涉及在线支付技术,特别涉及在线交易过程中实时侦测交易风险的方法、实施该方法的服务器和客户端以及包含实施该方法的计算机程序的计算机可读存储介质。
背景技术
在线支付由于其便捷性和与日常生活的紧密相关性而受到消费者的青睐,并已经成为主流的交易支付方式。然而,在线支付也导致了更多的交易欺诈风险。目前业界主要采用基于规则的方法和基于分类模型的方法来应对在线支付的交易欺诈风险。
但上述两种方法都存在明显的不足之处。例如,基于规则的方法依赖于专家经验,主观因素较强,而且风险判断的准确性与专家的能力水平密切相关。分类模型主要是通过对大量账户的交易特征进行统计分析而得到的,与专家制订的规则相比具有更好的客观性,但是分类模型本质上是一种统计模型,对于不同用户之间的差异性(在很多情况下,这种差异是明显的和不可忽视的)则显得无能为力。此外,当同时发生很多笔交易时,如果这些交易都需要实时处理,则将使系统承受巨大的处理压力,并导致处理速度的降低。
由上可见,迫切需要提供一种能够克服上述各种缺点的防范交易风险的方法和装置。
发明内容
本发明的一个目的是提供一种用于实时侦测交易风险的方法,其具有计算速度快、识别准确度高等优点。
按照本发明一个方面的用于交易风险实时侦测的方法包含下列步骤:
基于与单个账户相关联的历史交易建立对应于该单个账户的历 史交易模型;以及
向客户端提供历史交易模型以供其对该单个账户的当前交易的风险进行判断。
优选地,在上述方法中,建立对应于单个账户的历史交易模型的步骤在云端完成。
优选地,在上述方法中,进一步包括下列步骤:
利用单个账户新增的历史交易记录对历史交易模型进行定期或不定期地更新。
优选地,在上述方法中,所述历史交易模型为隐马尔可夫模型,建立历史交易模型的步骤包括:
生成表示单个账户的历史交易的可观察行为状态的第一交易序列作为隐马尔可夫模型的观察状态集合;以及
利用第一交易序列来训练该隐马尔可夫模型以建立该单个账户的历史交易模型。
优选地,在上述方法中,生成第一交易序列的步骤包括:
生成单个账户的每笔历史交易的交易特征向量以得到多个交易特征向量;
对所得到的多个交易特征向量进行聚类处理以得到一个或多个交易特征类别,其中每个所述交易特征类别对应于一个可观察行为状态;以及
根据各自的交易特征向量确定每笔历史交易所属的交易特征类别从而得到该单个账户的第一交易序列。
优选地,在上述方法中,按照下列方式确定所属的交易特征类别:
计算单个账户的每笔历史交易的交易特征向量与每个交易特征类别的相似性;以及
将对应于最大相似性的交易特征类别确定为该笔历史交易的交易特征向量所属的交易特征类别。
优选地,在上述方法中,利用第一交易序列来训练隐马尔可夫模 型包括下列步骤:
设定对应于单个账户的隐马尔可夫模型的隐藏状态的数量;
设定该隐马尔可夫模型的参数的初始值,其中,所述参数包括隐藏状态间的转移概率矩阵、隐藏状态到观察状态的概率矩阵和隐藏状态的初始概率分布;
对于该单个账户的第一交易序列,基于使第一交易序列的出现概率最大的优化目标来确定所述参数的优化值,由此建立对应于该单个账户的历史交易模型。
优选地,在上述方法中,隐藏状态间的转移概率矩阵和隐藏状态的初始概率分布的初始值设定为等概率值,隐藏状态到观察状态的概率矩阵的初始值根据交易特征类别的分布确定。
优选地,在上述方法中,采用非关系型数据库来保存对应于单个账户的历史交易模型。
按照本发明另一个方面的用于侦测交易风险的方法包含下列步骤:
客户端从云端获取对应于单个账户的历史交易模型,该历史交易模型基于与该单个账户相关联的历史交易而建立;
客户端利用历史交易模型对该单个账户的当前交易的风险进行判断;以及
输出当前交易的风险的判断结果。
优选地,在上述方法中,所述历史交易模型为隐马尔可夫模型,该隐马尔可夫模型的观察状态集合为表示单个账户的历史交易的可观察行为状态的第一交易序列,利用第一交易序列来训练该隐马尔可夫模型以建立该单个账户的历史交易模型。
优选地,在上述方法中,历史交易的可观察行为状态按照下列方式确定:
生成单个账户的每笔历史交易的交易特征向量以得到多个交易特征向量;
对所得到的多个交易特征向量进行聚类处理以得到一个或多个交易特征类别,其中每个所述交易特征类别对应于一个可观察行为状态;以及
根据各自的交易特征向量确定每笔历史交易所属的交易特征类别。
优选地,在上述方法中,利用所述历史交易模型对单个账户的当前交易的风险进行判断包括下列步骤:
确定当前交易的交易特征向量所属的交易特征类别;
通过以当前交易的交易特征向量所属的交易特征类别替换该单个账户的第一交易序列中的历史交易的交易特征向量所属的交易特征类别而生成第二交易序列;
确定第二交易序列的出现概率;以及
通过将第二交易序列的出现概率与第一交易序列的出现概率进行比较来对该单个账户的当前交易的风险进行判断。
优选地,在上述方法中,第一交易序列中被替换的历史交易对应于第一交易序列中交易时间最早的历史交易。
优选地,在上述方法中,按照下列方式确定所属的交易特征类别:
计算单个账户的当前交易的交易特征向量与每个交易特征类别的相似性;以及
将对应于最大相似性的交易特征类别确定为当前交易的交易特征向量所属的交易特征类别。
优选地,在上述方法中,所述相似性以下列指标中的一种来表征:当前交易的交易特征向量与每个交易特征类别中心向量之间的欧氏距离、当前交易的交易特征向量与每个交易特征类别中心向量之间的余弦距离以及当前交易的交易特征向量与每个交易特征类别的杰卡德相似度。
优选地,在上述方法中,由客户端确定当前交易所属的交易特征类别,并且每个交易特征类别以及相应的中心向量以关系数据表的形 式保存在客户端。
本发明的还有一个目的是提供一种用于侦测交易风险的服务器,其具有识别准确度高等优点。
按照本发明另一个方面的服务器包含存储器、处理器以及存储在所述存储器上并可在所述处理器上运行的计算机程序以执行如上所述的方法。
本发明的还有一个目的是提供一种用于侦测交易风险的客户端,其具有识别准确度高等优点。
按照本发明另一个方面的客户端包含存储器、处理器以及存储在所述存储器上并可在所述处理器上运行的计算机程序以执行如上所述的方法。
本发明的还有一个目的是提供一种计算机可读存储介质,其上存储计算机程序,该程序被处理器执行时实现如上所述的方法。
附图说明
本发明的上述和/或其它方面和优点将通过以下结合附图的各个方面的描述变得更加清晰和更容易理解,附图中相同或相似的单元采用相同的标号表示。附图包括:
图1为按照本发明一个实施例的用于交易风险实时侦测的方法的流程图。
图2为可应用于图1所示实施例的第一交易序列生成方法的流程图。
图3为可应用于图1所示实施例的隐马尔可夫模型训练方法的流程图。
图4为按照本发明另一个实施例的用于侦测交易风险的方法的流程图。
图5为可应用于图4所示实施例的基于隐马尔可夫模型的交易风险判断方法的流程图。
图6为按照本发明另一个实施例的用于交易风险实时侦测的服务器的框图。
图7为按照本发明另一个实施例的用于交易风险实时侦测的客户端的框图。
具体实施方式
下面参照其中图示了本发明示意性实施例的附图更为全面地说明本发明。但本发明可以按不同形式来实现,而不应解读为仅限于本文给出的各实施例。给出的上述各实施例旨在使本文的披露全面完整,以将本发明的保护范围更为全面地传达给本领域技术人员。
在本说明书中,诸如“包含”和“包括”之类的用语表示除了具有在说明书和权利要求书中有直接和明确表述的单元和步骤以外,本发明的技术方案也不排除具有未被直接或明确表述的其它单元和步骤的情形。
按照本发明的一个方面,基于与单个账户相关联的历史交易建立对应于该账户的历史交易模型,并且在在线支付过程中,利用所建立的历史交易模型来判断该账户的当前交易是否存在风险。也就是说,每个账户都可拥有为其专门定制的历史交易模型或分类模型,使得能够完美刻画该账户的交易特征,这有效提高了风险判断的准确性。优选地,可以利用账户新增的历史交易记录来定期或不定期地更新相应的历史交易模型,进一步提高判断的准确性。
按照本发明的另一个方面,上述历史交易模型为隐马尔可夫模型,其中,该隐马尔可夫模型的观察状态集合为表示单个账户的历史交易的可观察行为状态的第一交易序列,并且利用第一交易序列来训练该隐马尔可夫模型以建立该账户的历史交易模型。
一个隐马尔可夫模型通常可由一个五元组(Q,O,A,B,π)来描述,其中,Q为隐藏状态集合,O为观察状态集合,A为隐藏状态间的转移概率矩阵,B为隐藏状态到观察状态的概率矩阵,π为隐藏状态的 初始概率分布。按照本发明的还有一个方面,当利用隐马尔可夫模型来表征账户的历史交易特性时,隐藏状态对应于交易行为状态,观察状态对应于每笔交易的可观察的交易行为状态。
按照本发明的还有一个方面,历史交易模型的建立在云端完成,而基于历史交易模型所作的交易风险判断则在客户端完成。这种方式既可以发挥云端强大的计算能力,又能够减轻大规模并发交易时后台系统的运行压力,从而确保提供快速的实时风险侦测能力。
图1为按照本发明一个实施例的用于交易风险实时侦测的方法的流程图。优选地但非必须地,图1所示的方法可在云端服务器或后台交易处理系统处执行。
图1所示的方法的流程开始于步骤110。在该步骤中,生成表示单个账户的历史交易的可观察行为状态的第一交易序列,该第一交易序列可作为隐马尔可夫模型的观察状态集合。
随后进入步骤120,利用步骤110生成的第一交易序列来训练该隐马尔可夫模型,从而建立对应于该单个账户的历史交易模型。
接着进入步骤130,向客户端提供历史交易模型以供其对该单个账户的当前交易的风险进行判断。优选地,可采用非关系型数据库来保存对应于每个账户的历史交易模型。
可选地,本实施例的方法流程还包含步骤140。在该步骤中,利用单个账户新增的历史交易记录对其历史交易模型进行定期或不定期地更新。
图2为可应用于图1所示实施例的第一交易序列生成方法的流程图。优选地但非必须地,图2所示的方法可在云端服务器或后台交易处理系统处执行。
如图2所示,在步骤210,生成单个账户的每笔历史交易的交易特征向量以得到多个交易特征向量。在本实施例中,交易特征向量指的是由一个账户的一个或多个可观察的交易特征构成的向量。可观察的交易特征的例子包括但不限于交易金额、交易地点、交易时间和消 费类型等。需要指出的是,在本实施例中,不同账户的交易特征向量的结构(向量维数和分量的类型)可以相同,也可以不同。
优选地,对于具有连续性取值的交易特征,可以将其映射为离散化的数值。
需要指出的是,每笔交易中的多个交易特征对于交易风险的分析都是有用的,但是基本的隐马尔可夫模型无法处理多于一个标记的特征。针对这种情况,在本实施例中将每笔交易的多个可观察的交易特征映射为单个交易特征标记或交易特征类别(以下将这种映射操作又称为对交易特征向量的聚类化处理)。经过聚类化处理得到的交易特征类别的每一个对应于隐马尔可夫模型的其中一个可观察行为状态。
为聚类化处理的目的,图2所示的流程进入步骤220,对单个账户的多个交易特征向量进行聚类化处理,从而得到一个或多个交易特征类别,其中,每个类别代表一组具有相似模式的交易行为。以下将交易特征类别的集合记为{C 1,C 2…C k},其中k为类别的数量,每个类别都具有相应的类别中心。优选地,在本实施例中可以采用K-means算法对交易特征向量进行聚类化处理。
优选地,聚类处理所得到的交易特征类别及其对应的类别中心的坐标可以关系型数据表的形式存储。由于数据表占据的存储空间较小,因此其可以在用户安装或更新应用程序时下载到相应的客户端。
随后进入步骤230,将单个账户的每个交易特征向量都映射至相应的交易特征类别,由此得到该账户的第一交易序列。
优选地,可以按照下列方式将一个交易特征向量映射至相应的交易特征类别:首先计算该交易特征向量与每个交易特征类别的相似性,然后将对应于最大相似性的交易特征类别确定为该交易特征向量所属的交易特征类别。更好地,相似性以下列指标中的一种来表征:交易特征向量与每个交易特征类别中心向量之间的欧氏距离、交易特征向量与每个交易特征类别中心向量之间的余弦距离以及交易特征向量与每个交易特征类别的杰卡德相似度。
图3为可应用于图1所示实施例的隐马尔可夫模型训练方法的流程图。在图3所示的实施例中,利用前述第一交易序列来训练隐马尔可夫模型。优选地但非必须地,图3所示的方法可在云端服务器或后台交易处理系统处执行。
图3所示的流程开始于步骤310。在该步骤中,设定对应于单个账户的隐马尔可夫模型的隐藏状态的数量。假设隐藏状态为S个。
随后进入步骤320,设定隐马尔可夫模型的参数的初始值。在本实施例中,参数包括隐藏状态间的转移概率矩阵A、隐藏状态到观察状态的概率矩阵B和隐藏状态的初始概率分布π。
优选地,隐藏状态间的转移概率矩阵和隐藏状态的初始概率分布的初始值设定为等概率值,即,将每个隐藏状态的初始概率都设定为1/S,从一个隐藏状态转移到另一个隐藏状态的概率也为1/S(包括隐藏状态转移到其自身的情形)。
优选地,对于隐藏状态到观察状态的概率矩阵B的初始值,其可根据交易特征类别的分布来确定。具体而言,对于单个账户,可将每个交易特征类别所对应的交易次数占总交易次数的比例设定为每个隐藏状态到该交易特征类别所对应的观察状态的概率。
需要指出的是,隐藏状态的数量的设定值以及上述参数的初始值仅影响模型训练的效率,但不影响模型训练的有效性。
接着进入步骤330,在该步骤中,利用每个账户的第一交易序列,基于使第一交易序列的出现概率最大的优化目标对参数A、B和π进行优化,由此建立对应于该账户的历史交易模型。
优选地,可以采用Baum-Welch算法来优化参数λ(A,B,π)。具体而言,可通过依次执行下列步骤对上述隐马尔可夫模型进行训练。
步骤a:对于给定的观察序列或第一交易序列O={o 1,o 2...o T},根据当前的参数λ,利用下式(1)和(2)计算向前变量α t(i)和向后变量β t+1(i):
α t(i)=P(O 1,O 2...O t,q t=i|λ)      (1)
β t+1(i)=P(O t+1,O t+2...O T|q t=i,λ)     (2)
步骤b:利用下式(3)计算当序列t位于状态q i并且序列t+1位于状态q j时的概率ξ t(i,j):
Figure PCTCN2018094883-appb-000001
步骤c:利用下式(4)计算当序列t位于状态q i时的概率γ t(i):
Figure PCTCN2018094883-appb-000002
步骤d:利用下式(5)重新估计初始状态(t=1时刻隐藏状态q i的概率):
Figure PCTCN2018094883-appb-000003
步骤e:利用下式(6)重新估计转移概率矩阵
Figure PCTCN2018094883-appb-000004
Figure PCTCN2018094883-appb-000005
步骤f:重新估计输出概率矩阵,其中,
Figure PCTCN2018094883-appb-000006
为Q从状态q j发出观察状态o k的期望与Q到达状态q j的期望的比值:
Figure PCTCN2018094883-appb-000007
其中
Figure PCTCN2018094883-appb-000008
重复执行上述步骤a-f直到
Figure PCTCN2018094883-appb-000009
收敛,从而得到优化参数λ={A,B,π},即,完成隐马尔可夫模型的训练。
由于每个账户所对应的参数中的矩阵A、B和π在格式上都不是固定的,需要在训练阶段作调整优化,因此在本实施例中,优选地可采用诸如MongoDB和Hbase之类的非关系型数据库来存储上述优化参数λ={A,B,π}。
为了节省存储空间,对于每个账户,仅存储最近发生的R笔交易所对应的交易特征类别作为第一交易序列或观察序列。也就是说,每个账户的第一交易序列被设计为一个空间为R的循环队列。R的取值可以根据实际应用场合进行调整。云端服务器或后台交易处理系统可以周期性地对存储各个账户的第一交易序列的数据库执行更新操作,其中,每次仅对最近有交易记录变化的账户的第一交易序列进行更新。优选地,可以为每笔交易记录所对应的交易特征类别附接时间戳以指示更新时间。云端服务器在执行更新操作时,将删除时间戳大于时长参数Pe的账户记录,这不仅可以避免数据库存储空间的不足,还可以使得账户的历史交易模型更多地反映近期交易行为。
图4为按照本发明另一个实施例的用于交易风险实时侦测的方法的流程图。优选地但非必须地,图4所示的方法可在客户端处执行。
图4所示的方法的流程开始于步骤410。在该步骤中,客户端从云端获取对应于单个账户的历史交易模型。有关历史交易模型的建立方式已经在上面借助图1-3作了充分的描述,此处不再赘述。
随后进入步骤420,客户端利用步骤410获取的历史交易模型对该单个账户的当前交易的风险进行判断。
最后在步骤430,客户端向云端服务器或后台交易处理系统输出当前交易的风险的判断结果。可选地,客户端还可在其显示界面上呈现判断结果。
图5为可应用于图4所示实施例的基于隐马尔可夫模型的交易风险判断方法的流程图。优选地但非必须地,图5所示的方法可在客户 端处执行。
图5所示的流程开始于步骤510。在该步骤中,确定一个账户的当前交易的交易特征向量所属的交易特征类别。优选地,可以按照下列方式确定相应的交易特征类别:首先计算当前交易的交易特征向量与每个交易特征类别的相似性,然后将对应于最大相似性的交易特征类别确定为该交易特征向量所属的交易特征类别。更好地,相似性以下列指标中的一种来表征:交易特征向量与每个交易特征类别中心向量之间的欧氏距离、交易特征向量与每个交易特征类别中心向量之间的余弦距离以及交易特征向量与每个交易特征类别的杰卡德相似度。
随后进入步骤520,通过以当前交易的交易特征类别所属的交易特征类别替换该账户的第一交易序列中的历史交易所属的交易特征类别而生成第二交易序列。以下将第一交易序列记为O={o 1,o 2...o R},其中,序列中的各个元素按照时间先后顺序排列。
在本实施例中,优选地,第一交易序列中被替换的历史交易对应于第一交易序列中交易时间最早的历史交易。因此第二交易序列可记为O′={o 2,o 3...o R+1},其中最早的历史交易的交易特征类别O 1被当前交易的交易特征类别O R+1替代。
接着在步骤530,确定第二交易序列O’的出现概率ρ′=P(o 2,o 3...o R+1|λ)。
随后进入步骤540,将第二交易序列的出现概率ρ′=P(o 2,o 3...o R+1|λ)与第一交易序列的出现概率ρ=P(o 1,o 2...o R|λ)进行比较来对该单个账户的当前交易的风险进行判断。优选地,可以定义出现概率的变化率
Figure PCTCN2018094883-appb-000010
作为交易风险判断的依据。具体而言,如果Δρ≥θ,则说明当前交易与历史交易模型存在较大差别,因此判断为可疑交易;如果Δρ<θ,则判断当前交易为正常交易,这里θ为一个预先设定的阈值。
最后在步骤550,客户端将步骤540的判断结果发送给云端服务 器或后台交易处理系统。
图6为按照本发明另一个实施例的用于交易风险实时侦测的服务器的框图。
图6所示的服务器60包含存储器610、处理器620以及存储在存储器610上并可在处理器620上运行的计算机程序630,其中,计算机程序630通过在处理器620上运行以可执行如上借助图1-3所述实施例的方法。
图7为按照本发明另一个实施例的用于交易风险实时侦测的客户端的框图。
图7所示的客户端70包含存储器710、处理器720以及存储在存储器710上并可在处理器720上运行的计算机程序730,其中,计算机程序730通过在处理器720上运行以可执行如上借助图4和5所述实施例的方法。在本实施例中,客户端可以为手机、收单机构的POS机或扫码器。
按照本发明的一个方面,提供一种计算机可读存储介质,其上存储计算机程序,该程序被处理器执行时实现借助图1-3所述实施例的方法。
按照本发明的一个方面,提供一种计算机可读存储介质,其上存储计算机程序,该程序被处理器执行时实现借助图4和5所述实施例的方法。
与现有技术相比,本发明的上述实施例具有下列优点:
1、通过采用针对每个账户定制的历史交易模型来判断交易风险,有效避免了通用模型适应性差等不足。
2、通过利用隐马尔可夫算法建立账户级的历史交易模型并且根据当前交易与交易历史模型的概率差异来判断交易风险,提高了判断的准确性。
3、由于可以将交易风险的判断操作安排在客户端处完成,避免了大规模交易并发时对后台交易处理系统的运行压力,同时也提高了 判别速度。
提供本文中提出的实施例和示例,以便最好地说明按照本技术及其特定应用的实施例,并且由此使本领域的技术人员能够实施和使用本发明。但是,本领域的技术人员将会知道,仅为了便于说明和举例而提供以上描述和示例。所提出的描述不是意在涵盖本发明的各个方面或者将本发明局限于所公开的精确形式。
鉴于以上所述,本公开的范围通过以下权利要求书来确定。

Claims (22)

  1. 一种用于交易风险实时侦测的方法,其特征在于,包含下列步骤:
    基于与单个账户相关联的历史交易建立对应于该单个账户的历史交易模型;以及
    向客户端提供历史交易模型以供其对该单个账户的当前交易的风险进行判断。
  2. 如权利要求1所述的方法,其中,建立对应于单个账户的历史交易模型的步骤在云端完成。
  3. 如权利要求1所述的方法,其中,进一步包括下列步骤:
    利用单个账户新增的历史交易记录对历史交易模型进行定期或不定期地更新。
  4. 如权利要求1-3中任意一项所述的方法,其中,所述历史交易模型为隐马尔可夫模型,建立历史交易模型的步骤包括:
    生成表示单个账户的历史交易的可观察行为状态的第一交易序列作为隐马尔可夫模型的观察状态集合;以及
    利用第一交易序列来训练该隐马尔可夫模型以建立该单个账户的历史交易模型。
  5. 如权利要求4所述的方法,其中,生成第一交易序列的步骤包括:
    生成单个账户的每笔历史交易的交易特征向量以得到多个交易特征向量;
    对所得到的多个交易特征向量进行聚类处理以得到一个或多个交易特征类别,其中每个所述交易特征类别对应于一个可观察行为状态;以及
    根据各自的交易特征向量确定每笔历史交易所属的交易特征类别从而得到该单个账户的第一交易序列。
  6. 如权利要求5所述的方法,其中,按照下列方式确定所属的交易特征类别:
    计算单个账户的每笔历史交易的交易特征向量与每个交易特征类别的相似性;以及
    将对应于最大相似性的交易特征类别确定为该笔历史交易的交易特征向量所属的交易特征类别。
  7. 如权利要求5所述的方法,其中,利用第一交易序列来训练隐马尔可夫模型包括下列步骤:
    设定对应于单个账户的隐马尔可夫模型的隐藏状态的数量;
    设定该隐马尔可夫模型的参数的初始值,其中,所述参数包括隐藏状态间的转移概率矩阵、隐藏状态到观察状态的概率矩阵和隐藏状态的初始概率分布;
    对于该单个账户的第一交易序列,基于使第一交易序列的出现概率最大的优化目标来确定所述参数的优化值,由此建立对应于该单个账户的历史交易模型。
  8. 如权利要求7所述的方法,其中,隐藏状态间的转移概率矩阵和隐藏状态的初始概率分布的初始值设定为等概率值,隐藏状态到观察状态的概率矩阵的初始值根据交易特征类别的分布确定。
  9. 如权利要求4所述的方法,其中,采用非关系型数据库来保存对应于单个账户的历史交易模型。
  10. 一种用于侦测交易风险的方法,其特征在于,包含下列步骤:
    客户端从云端获取对应于单个账户的历史交易模型,该历史交易模型基于与该单个账户相关联的历史交易而建立;
    客户端利用历史交易模型对该单个账户的当前交易的风险进行判断;以及
    输出当前交易的风险的判断结果。
  11. 如权利要求10所述的方法,其中,所述历史交易模型为隐马尔可夫模型,该隐马尔可夫模型的观察状态集合为表示单个账户的 历史交易的可观察行为状态的第一交易序列,利用第一交易序列来训练该隐马尔可夫模型以建立该单个账户的历史交易模型。
  12. 如权利要求11所述的方法,其中,历史交易的可观察行为状态按照下列方式确定:
    生成单个账户的每笔历史交易的交易特征向量以得到多个交易特征向量;
    对所得到的多个交易特征向量进行聚类处理以得到一个或多个交易特征类别,其中每个所述交易特征类别对应于一个可观察行为状态;以及
    根据各自的交易特征向量确定每笔历史交易所属的交易特征类别。
  13. 如权利要求12所述的方法,其中,利用所述历史交易模型对单个账户的当前交易的风险进行判断包括下列步骤:
    确定当前交易的交易特征向量所属的交易特征类别;
    通过以当前交易的交易特征向量所属的交易特征类别替换该单个账户的第一交易序列中的历史交易的交易特征向量所属的交易特征类别而生成第二交易序列;
    确定第二交易序列的出现概率;以及
    通过将第二交易序列的出现概率与第一交易序列的出现概率进行比较来对该单个账户的当前交易的风险进行判断。
  14. 如权利要求13所述的方法,其中,第一交易序列中被替换的历史交易对应于第一交易序列中交易时间最早的历史交易。
  15. 如权利要求13所述的方法,其中,按照下列方式确定所属的交易特征类别:
    计算单个账户的当前交易的交易特征向量与每个交易特征类别的相似性;以及
    将对应于最大相似性的交易特征类别确定为当前交易的交易特征向量所属的交易特征类别。
  16. 如权利要求15所述的方法,其中,所述相似性以下列指标中的一种来表征:当前交易的交易特征向量与每个交易特征类别中心向量之间的欧氏距离、当前交易的交易特征向量与每个交易特征类别中心向量之间的余弦距离以及当前交易的交易特征向量与每个交易特征类别的杰卡德相似度。
  17. 如权利要求15所述的方法,其中,由客户端确定当前交易所属的交易特征类别,并且每个交易特征类别以及相应的中心向量以关系数据表的形式保存在客户端。
  18. 一种服务器,包含存储器、处理器以及存储在所述存储器上并可在所述处理器上运行的计算机程序,其特征在于,执行如权利要求1-9中任意一项所述的方法。
  19. 一种客户端,包含存储器、处理器以及存储在所述存储器上并可在所述处理器上运行的计算机程序,其特征在于,执行如权利要求10-17中任意一项所述的方法。
  20. 如权利要求19所述的客户端,其中,所述客户端为手机、收单机构的POS机或扫码器。
  21. 一种计算机可读存储介质,其上存储计算机程序,其特征在于,该程序被处理器执行时实现如权利要求1-9中任意一项所述的方法。
  22. 一种计算机可读存储介质,其上存储计算机程序,其特征在于,该程序被处理器执行时实现如权利要求10-17中任意一项所述的方法。
PCT/CN2018/094883 2017-07-18 2018-07-06 用于交易风险实时侦测的方法和装置 WO2019015499A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710584386.9 2017-07-18
CN201710584386.9A CN109272312B (zh) 2017-07-18 2017-07-18 用于交易风险实时侦测的方法和装置

Publications (1)

Publication Number Publication Date
WO2019015499A1 true WO2019015499A1 (zh) 2019-01-24

Family

ID=65015362

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/094883 WO2019015499A1 (zh) 2017-07-18 2018-07-06 用于交易风险实时侦测的方法和装置

Country Status (3)

Country Link
CN (1) CN109272312B (zh)
TW (1) TWI734920B (zh)
WO (1) WO2019015499A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110458576A (zh) * 2019-07-31 2019-11-15 同济大学 一种融合事前预测和事中检测的网络交易反欺诈方法
CN110991871A (zh) * 2019-11-29 2020-04-10 深圳前海微众银行股份有限公司 风险监测方法、装置、设备与计算机可读存储介质

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111311408B (zh) * 2020-02-10 2021-08-03 支付宝(杭州)信息技术有限公司 电子交易属性识别方法及装置
CN116611829B (zh) * 2023-07-21 2023-11-14 山东美丽乡村云计算有限公司 一种基于区块链的消费监管系统

Citations (3)

* Cited by examiner, † Cited by third party
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
CN105809502A (zh) * 2014-12-30 2016-07-27 阿里巴巴集团控股有限公司 交易风险检测方法和装置
CN106251214A (zh) * 2016-08-02 2016-12-21 东软集团股份有限公司 账户监控方法及装置

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB0013011D0 (en) * 2000-05-26 2000-07-19 Ncr Int Inc Method and apparatus for determining one or more statistical estimators of customer behaviour
CN103577413B (zh) * 2012-07-20 2017-11-17 阿里巴巴集团控股有限公司 搜索结果排序方法及系统、搜索结果排序优化方法及系统
CN106485396A (zh) * 2016-09-09 2017-03-08 北京科技大学 一种安全生产隐患排查系统
CN106485348A (zh) * 2016-09-22 2017-03-08 中国银联股份有限公司 一种交易数据的预测方法及装置

Patent Citations (3)

* Cited by examiner, † Cited by third party
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
CN105809502A (zh) * 2014-12-30 2016-07-27 阿里巴巴集团控股有限公司 交易风险检测方法和装置
CN106251214A (zh) * 2016-08-02 2016-12-21 东软集团股份有限公司 账户监控方法及装置

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110458576A (zh) * 2019-07-31 2019-11-15 同济大学 一种融合事前预测和事中检测的网络交易反欺诈方法
CN110458576B (zh) * 2019-07-31 2022-12-20 同济大学 一种融合事前预测和事中检测的网络交易反欺诈方法
CN110991871A (zh) * 2019-11-29 2020-04-10 深圳前海微众银行股份有限公司 风险监测方法、装置、设备与计算机可读存储介质

Also Published As

Publication number Publication date
CN109272312A (zh) 2019-01-25
TW201909047A (zh) 2019-03-01
TWI734920B (zh) 2021-08-01
CN109272312B (zh) 2021-07-13

Similar Documents

Publication Publication Date Title
WO2019015499A1 (zh) 用于交易风险实时侦测的方法和装置
US11343156B2 (en) Compute intensive stream processing with context data routing
CN108399381A (zh) 行人再识别方法、装置、电子设备和存储介质
US8731523B1 (en) Push notification delivery system with feedback analysis
WO2020073714A1 (zh) 训练样本获取方法,账户预测方法及对应装置
CN108629413A (zh) 神经网络模型训练、交易行为风险识别方法及装置
CN107609147B (zh) 一种从日志流中自动提取特征的方法和系统
JP2013137769A (ja) ユーザ間の親密度に基づいて友人の友人を推薦するソーシャルネットワークサービスシステム及びその提供方法
CN111340246A (zh) 用于企业智能决策分析的处理方法、装置和计算机设备
CN111932268B (zh) 企业风险识别方法及装置
CN112799708B (zh) 联合更新业务模型的方法及系统
WO2019075599A1 (zh) 数据填补方法和装置
WO2020077888A1 (zh) 贷款用户信用评分的计算方法、装置和计算机设备
CN110991871A (zh) 风险监测方法、装置、设备与计算机可读存储介质
CN110796089A (zh) 用于训练换脸模型的方法和设备
JP2019096227A (ja) 操業改善効果算出装置、操業改善効果算出方法、操業改善効果算出プログラム、および記録媒体
CN109871866B (zh) 用于医院内感染预测的模型训练方法、装置、设备及介质
CN113590807B (zh) 一种基于大数据挖掘的科技企业信用评价方法
US11971873B2 (en) Real-time anomaly determination using integrated probabilistic system
CN110796379A (zh) 业务渠道的风险评估方法、装置、设备及存储介质
CN108564458B (zh) 一种基于蒙特卡洛模拟的消费预测方法及系统
CN109493065A (zh) 一种基于行为增量更新的欺诈交易检测方法
CN110008972A (zh) 用于数据增强的方法和装置
CN108830603A (zh) 交易识别方法及装置
CN114298825A (zh) 还款积极度评估方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18835904

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18835904

Country of ref document: EP

Kind code of ref document: A1