WO2022105177A1 - 信用卡客户的用户画像方法、装置、设备及介质 - Google Patents
信用卡客户的用户画像方法、装置、设备及介质 Download PDFInfo
- Publication number
- WO2022105177A1 WO2022105177A1 PCT/CN2021/096978 CN2021096978W WO2022105177A1 WO 2022105177 A1 WO2022105177 A1 WO 2022105177A1 CN 2021096978 W CN2021096978 W CN 2021096978W WO 2022105177 A1 WO2022105177 A1 WO 2022105177A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- analyzed
- target
- transaction data
- data
- historical transaction
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 108
- 239000011159 matrix material Substances 0.000 claims abstract description 116
- 230000007704 transition Effects 0.000 claims abstract description 111
- 230000008569 process Effects 0.000 claims abstract description 53
- 230000002787 reinforcement Effects 0.000 claims abstract description 52
- 238000007476 Maximum Likelihood Methods 0.000 claims abstract description 46
- 230000006870 function Effects 0.000 claims description 90
- 238000004364 calculation method Methods 0.000 claims description 30
- 238000004590 computer program Methods 0.000 claims description 16
- 238000000605 extraction Methods 0.000 claims 1
- 230000006399 behavior Effects 0.000 abstract description 24
- 238000013473 artificial intelligence Methods 0.000 abstract description 4
- 230000007774 longterm Effects 0.000 description 9
- 230000008859 change Effects 0.000 description 7
- 230000009286 beneficial effect Effects 0.000 description 6
- 238000005065 mining Methods 0.000 description 3
- 230000007423 decrease Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000008685 targeting Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
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
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/04—Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2462—Approximate or statistical queries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2465—Query processing support for facilitating data mining operations in structured databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- 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
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/03—Credit; Loans; Processing thereof
Definitions
- the present application relates to the field of artificial intelligence technology, and in particular, to a user portrait method, device, device and medium of a credit card customer.
- the purpose is to solve the technical problems that the existing technology has low accuracy of user portraits of credit card customers, it is difficult to meet the needs of marketing scenarios, and it is difficult to track the behavior of credit card customers for long-term training.
- the main purpose of this application is to provide a method, device, equipment and medium for user portraits of credit card customers, aiming to solve the problem that the user portrait of credit card customers in the prior art is not highly accurate, it is difficult to meet the needs of marketing scenarios, and it is difficult to track credit cards.
- the technical issues of long-term cultivation of customer behavior are not highly accurate, it is difficult to meet the needs of marketing scenarios, and it is difficult to track credit cards.
- the present application proposes a user portrait method of a credit card customer, the method comprising:
- the target historical transaction data is divided into business categories and life cycle matching is performed, and a plurality of historical transaction data to be analyzed and their corresponding target business categories and target life cycles are determined;
- Markov decision process and maximum likelihood inverse reinforcement learning are used to calculate the state transition probability matrix according to each of the historical transaction data to be analyzed and the corresponding target customer value mean data, to obtain each of the historical transaction data to be analyzed.
- the target state transition probability matrix of transaction data
- target state transition probability matrix of each of the historical transaction data to be analyzed and the target customer value mean data respectively, determine a plurality of target sub-category customer value data
- the application also proposes a computer device, comprising a memory and a processor, the memory stores a computer program, and the processor implements the following method steps when executing the computer program:
- Perform business category division and life cycle matching on the target historical transaction data determine a plurality of historical transaction data to be analyzed and their corresponding target business categories and target life cycles; perform business category division and life cycle matching;
- Markov decision process and maximum likelihood inverse reinforcement learning are used to calculate the state transition probability matrix according to each of the historical transaction data to be analyzed and the corresponding target customer value mean data, to obtain each of the historical transaction data to be analyzed.
- the target state transition probability matrix of transaction data
- target state transition probability matrix of each of the historical transaction data to be analyzed and the target customer value mean data respectively, determine a plurality of target sub-category customer value data
- the present application also proposes a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the following method steps are implemented:
- Perform business category division and life cycle matching on the target historical transaction data determine a plurality of historical transaction data to be analyzed and their corresponding target business categories and target life cycles; perform business category division and life cycle matching;
- Markov decision process and maximum likelihood inverse reinforcement learning are used to calculate the state transition probability matrix according to each of the historical transaction data to be analyzed and the corresponding target customer value mean data, to obtain each of the historical transaction data to be analyzed.
- the target state transition probability matrix of transaction data
- target state transition probability matrix of each of the historical transaction data to be analyzed and the target customer value mean data respectively, determine a plurality of target sub-category customer value data
- the user portrait method, device, equipment and medium of the credit card customer of the present application divide the target historical transaction data into business categories and match the life cycle, and determine a plurality of historical transaction data to be analyzed and their corresponding target business categories and target life cycles.
- the target state transition probability matrix of the data and the mean value data of target customers determine the customer value data of multiple target subdivisions, carry out user portraits of credit card customers according to the customer value data of multiple target subdivisions, determine the target user portraits, Markov decision-making ability
- the business category, life cycle, and historical transaction data change the behavior of credit card customers is fully exploited, which is conducive to improving the accuracy of the user portrait of credit card customers.
- the maximum likelihood inverse reinforcement learning is used to realize the autonomy of credit card customers' behavior. Learning improves the generalization ability of user portraits of credit card customers, so as to meet the needs of marketing scenarios, and is conducive to tracking the behavior of credit card customers for long-term training.
- FIG. 1 is a schematic flowchart of a user portrait method of a credit card customer according to an embodiment of the application
- FIG. 2 is a schematic block diagram of the structure of a user portrait device of a credit card customer according to an embodiment of the application;
- FIG. 3 is a schematic structural block diagram of a computer device according to an embodiment of the present application.
- the present application proposes a user portrait method of credit card customers, The method is applied to the technical field of artificial intelligence, and the method is further applied to the technical field of predictive analysis of artificial intelligence.
- the user portrait method of credit card customers is divided into business categories and life cycle matching, and then the average value of customers in each sub-category is obtained from the division results, and the Markov decision process and maximum likelihood inverse reinforcement learning are used.
- the average value of customer value of each sub-category determines the state transition probability matrix of each sub-category, and the potential value of customers is determined according to the state transition probability matrix of each sub-category and the average value of customer value of each sub-category.
- the whole process fully considers the customer status, It improves the accuracy of the potential value of the identified target customers, provides accurate data support for precision marketing, and is conducive to maximizing the value of credit card customers.
- an embodiment of the present application provides a user portrait method of a credit card customer, and the method includes:
- S2 Divide the target historical transaction data into business categories and match life cycles, and determine a plurality of historical transaction data to be analyzed and their corresponding target business categories and target life cycles;
- S3 Obtain customer value mean data according to the respective target business category and target life cycle corresponding to each of the historical transaction data to be analyzed, and obtain target customer value average data of each of the historical transaction data to be analyzed;
- S4 Use Markov decision process and maximum likelihood inverse reinforcement learning to calculate the state transition probability matrix according to each of the historical transaction data to be analyzed and the corresponding target customer value mean data, to obtain each of the to-be-analyzed transaction data Analyze the target state transition probability matrix of historical transaction data;
- S5 Determine a plurality of target sub-category customer value data according to the target state transition probability matrix of each historical transaction data to be analyzed and the target customer value mean data;
- S6 Perform user portraits of credit card customers according to the plurality of target subdivided customer value data, and determine target user portraits.
- the target historical transaction data is divided into business categories and the life cycle is matched, and a plurality of historical transaction data to be analyzed and their corresponding target business categories and target life cycles are determined, respectively, according to the target business corresponding to each historical transaction data to be analyzed.
- Category and target life cycle to obtain customer value mean data obtain the target customer value mean data of each historical transaction data to be analyzed, and use Markov decision process and maximum likelihood inverse reinforcement learning to analyze each historical transaction data and their respective data.
- target state transition probability matrix and target customer value mean data of each historical transaction data to be analyzed , determine multiple target subdivision customer value data, carry out user portrait of credit card customers according to multiple target subdivided customer value data, determine target user portrait, Markov decision-making can be used when the business category, life cycle, and historical transaction data change Fully mining the behavior of credit card customers, which is beneficial to improve the accuracy of the user portrait of credit card customers, and then realizes the self-learning of the behavior of credit card customers through maximum likelihood inverse reinforcement learning, and improves the generalization ability of the user portrait of credit card customers. , so as to meet the needs of marketing scenarios and help track the behavior of credit card customers for long-term training.
- a request for determining the customer value of a credit card sent by the user may be obtained, or it may be a request for determining the customer value of a credit card sent by the application system.
- the credit card customer value determination request refers to a request to determine the customer potential value of a credit card customer.
- the target historical transaction data is the historical credit card transaction data of credit card customers within the first preset time period of the current time as the end time.
- Historical credit card transaction data that is, credit card transaction data.
- Credit card transaction data includes but is not limited to: business category, transaction amount, transaction time.
- Business categories include: basic consumption, personal installment, merchant installment, and cash withdrawal.
- Basic consumption is the consumption by credit card customers other than instalments at merchants.
- Personal installments include: bill installments and cash withdrawal installments.
- Merchant instalment refers to the instalment customized for the merchant's products, and the credit card customer chooses the instalment payment when purchasing the merchant's products.
- Cash withdrawal is a consumption in which a credit card customer withdraws cash from the credit limit but does not choose to install the withdrawn cash.
- the target historical transaction data is sequentially divided into business categories and life cycle matching is performed to obtain a plurality of historical transaction data to be analyzed, that is, the data in the same historical transaction data to be analyzed belong to the same business category and the same life cycle .
- the respective corresponding target business categories and target life cycles refer to respective target business categories and target life cycles corresponding to a plurality of historical transaction data to be analyzed.
- the business category used for dividing the historical transaction data to be analyzed is used as the target business category corresponding to the historical transaction data to be analyzed.
- the life cycle in which the historical transaction data to be analyzed is divided is used as the target life cycle corresponding to the historical transaction data to be analyzed.
- Life cycle refers to the stages of the life cycle of a credit card customer.
- the life cycle includes: customer acquisition period, contact period, growth period, maturity period, decline period, and silence period. It is understood that the life cycle can also have other division methods and naming methods, which are not specifically limited here.
- the customer status includes: consumption time, consumption frequency, and consumption amount since the present.
- the consumption time since the present includes: the consumption time since the first consumption, and the consumption time since the last consumption.
- the consumption frequency refers to the consumption times within the second preset duration with the current time as the end time.
- the obtained client value mean data is used as the target client value mean data of the extracted historical transaction data to be analyzed; the above steps are repeated until all the target client value mean data of the to-be-analyzed historical transaction data are determined.
- Target customer value average data that is, customer value average data.
- the customer value mean data is a time series. That is to say, the average customer value data is the average customer value of credit card customers in the same business category and the same life cycle.
- the average customer value data includes, but is not limited to: data on the average gross profit of credit card customers and data on the average net profit of credit card customers.
- the Markov decision process is used to extract the utility function according to each of the historical transaction data to be analyzed and the corresponding target customer value mean data, to obtain a plurality of utility function sets; then use the maximum likelihood inverse reinforcement Learning to perform parameter estimation on each of the utility function sets, respectively, to obtain a target state transition probability matrix of each of the historical transaction data to be analyzed.
- Each of the historical transaction data to be analyzed corresponds to a target state transition probability matrix.
- the target state transition probability matrix that is, the state transition probability matrix.
- the state transition probability matrix is the matrix of the probability of customer state transition.
- each of the historical transaction data to be analyzed corresponds to one of the target sub-category customer value data.
- the target state transition probability matrix is expressed as:
- the target customer value mean data G is: [G1G2G3G4G5], the target state transition probability matrix is multiplied by the target customer value mean data, and the matrix J is obtained by multiplying P and G, and the element values of all elements in the matrix J are calculated.
- the calculation result is used as the target segmented customer value data, which is not specifically limited in this example.
- the target sub-category customer value data is the customer value of the credit card customer corresponding to the target historical transaction data in the target business category in the target life cycle. That is, the amount of target segment customer value data is the same as the amount of business category.
- the multiple target sub-category customer value data is formed into a matrix by time, and the matrix is the user portrait of the credit card customer, and the user portrait of the credit card customer is used as the target user portrait.
- the target user portrait is used to describe whether the credit card customer corresponding to the credit card customer value determination request conducts consumption at a time point in each business category of the credit card and the expected customer value brought by the consumption.
- the expected customer value brought by consumption refers to the gross profit value
- the expected customer value brought by consumption refers to Net profit value
- the target user profile can be expressed as:
- the first line expresses the target business category as basic consumption
- the second line expresses the target business category as personal installment
- the third line expresses the target business category as merchant installment
- the fourth line expresses the target business category as cash withdrawal
- the first column expresses The first time, the second column describes the second time, the third column describes the third time, and the fourth column describes the fourth time;
- the third time of the target user portrait indicates that the credit card customer will not consume, and the credit card customer will not need to be targeted for marketing at the third time, which is not specifically limited in this example.
- the above-mentioned steps of dividing the target historical transaction data into business categories and matching life cycles to determine a plurality of historical transaction data to be analyzed and their corresponding target business categories and target life cycles include:
- S21 Divide the target historical transaction data by using the business category, and obtain a plurality of historical transaction data to be analyzed and their corresponding target business categories;
- S23 Use the life cycle division standard data to perform life cycle matching according to the current consumption time, consumption frequency, consumption amount, and the target business category of each of the historical transaction data to be analyzed, and obtain each of the to-be-analyzed transaction data The target life cycle corresponding to the historical transaction data.
- the target historical transaction data is firstly divided according to the business category and the life cycle.
- the life cycle is determined according to the historical transaction data to be analyzed and the target business category, and the business category, life cycle, and historical transaction data are fully considered. , which provides a data basis for the subsequent determination of target and subdivided customer value data.
- the data of the same business category in the target historical transaction data is put into a set, each business category corresponds to a set, and the data of each set is a piece of historical transaction data to be analyzed.
- the business category corresponding to the collection is used as the target business category corresponding to the historical transaction data to be analyzed.
- the life cycle division standard data can be obtained from the database.
- Life cycle division standard data including: business category, life cycle, standard value of consumption time since the present, standard value of consumption frequency, and standard value of consumption amount.
- the business category includes: basic consumption, personal installment, merchant installment, and cash withdrawal
- the life cycle includes: customer acquisition period, contact period, growth period, maturity period, decline period, and silent period.
- the target historical transaction data of credit card customer A is adopted.
- the business category is divided, and the historical transaction data of basic consumption is A1, the historical transaction data of personal installment is A2, the historical transaction data of merchant installment is A3, and the historical transaction data of cash withdrawal is A4, according to the historical consumption of A1.
- the target life cycle of A1 is the mature period
- the life cycle of A2 is divided into the life cycle according to the consumption duration, consumption frequency, and consumption amount of A2, and the target life cycle of A2 is the growth period.
- the target life cycle of A3 is the contact period
- the life cycle of A4 is divided according to the current consumption time, consumption frequency and consumption amount, and the target life cycle of A4 is contact. period, which is not specifically limited in this example.
- the average value data of customers is obtained according to the corresponding target business category and target life cycle of each historical transaction data to be analyzed, and the average value data of target customers of each historical transaction data to be analyzed is obtained. steps, including:
- S32 Obtain customer value average data according to the target business category and the target life cycle corresponding to the extracted historical transaction data to be analyzed, and obtain the target customer value average data of each historical transaction data to be analyzed ;
- This embodiment realizes the determination of target customer value average data for each of the historical transaction data to be analyzed, and provides a data basis for the subsequent determination of target subdivided customer value data.
- the customer value average data found in the average value list is used as the target customer value average data corresponding to the extracted historical transaction data to be analyzed.
- the average customer value list includes: business category, life cycle, and customer value average data.
- steps S31 and S32 are repeatedly executed until the target customer value mean data of all the historical transaction data to be analyzed is determined.
- the above-mentioned Markov decision process and maximum likelihood inverse reinforcement learning are used to calculate the state transition probability matrix according to each of the historical transaction data to be analyzed and the corresponding target customer value mean data, respectively, to obtain Before each step of the target state transition probability matrix of the historical transaction data to be analyzed, it also includes:
- S42 Use the average value data of the target customer value of the historical transaction data to be analyzed as the average value data of the customers to be analyzed in the time series of the transaction data to be analyzed corresponding to the historical transaction data to be analyzed;
- S43 Use Markov decision process and maximum likelihood inverse reinforcement learning to calculate the state transition probability matrix according to each time series of the transaction data to be analyzed and the corresponding mean value data of the customer to be analyzed, to obtain each Describe the target state transition probability matrix of the historical transaction data to be analyzed.
- This embodiment realizes the state transition probability matrix calculation according to each of the historical transaction data to be analyzed and the corresponding target customer value mean value data, and fully considers the historical transaction data and the customer value average value data of each sub-category. Determining Target Segments Customer Value Data provides accurate data support.
- time series construction is performed on each of the historical transaction data to be analyzed in chronological order, that is, a time series of transaction data to be analyzed is finally generated for each historical transaction data to be analyzed in chronological order.
- the time granularity of the customer value mean data is the same as the time granularity of the transaction data time series to be analyzed.
- the time granularity of the customer value average data is daily
- the time granularity of the transaction data time series to be analyzed is daily, that is, each element of the transaction data time series to be analyzed represents the daily transaction data, then the time granularity of the customer value average data
- the granularity is the same as the time granularity of the time series of the transaction data to be analyzed, and is not specifically limited by this example.
- the average value data of the target customer of each historical transaction data to be analyzed is the same as the average value data of the customer to be analyzed in the time series of the transaction data to be analyzed corresponding to itself.
- a Markov decision process is used to extract a utility function according to each time series of the transaction data to be analyzed and the corresponding mean value data of the customer to be analyzed, and a plurality of utility function sets are obtained. Then use maximum likelihood inverse reinforcement learning to perform parameter estimation on each of the utility function sets respectively, and obtain the state transition probability matrix of each of the time series of the transaction data to be analyzed; The state transition probability matrix is used as the corresponding target state transition probability matrix of the historical transaction data to be analyzed. That is to say, each time series of the transaction data to be analyzed corresponds to a utility function set.
- the utility function is the state-value function of the Markov decision process.
- the Markov decision process and maximum likelihood inverse reinforcement learning are used to calculate the state transition probability matrix according to each time series of the transaction data to be analyzed and the corresponding mean value data of the customers to be analyzed.
- the steps of obtaining the target state transition probability matrix of each of the historical transaction data to be analyzed include:
- S431 Using a Markov decision-making process to extract a utility function according to each time series of the transaction data to be analyzed and the corresponding mean value data of the customers to be analyzed, to obtain the utility of each time series of the transaction data to be analyzed collection of functions;
- S432 Perform parameter estimation on each of the utility function sets by maximum likelihood inverse reinforcement learning to obtain the target state transition probability matrix of each of the historical transaction data to be analyzed.
- This embodiment implements the use of Markov decision-making process and maximum likelihood inverse reinforcement learning to calculate the state transition probability matrix according to each time series of the transaction data to be analyzed and the corresponding mean value data of the customer to be analyzed, which is sufficient Considering the historical credit card transaction data and the average customer value data of each sub-category, it can fully explore the behavior of credit card customers when the business category, life cycle, and historical transaction data change, and improve the accuracy of the target sub-category customer value data. It is beneficial to improve the accuracy of user portraits of credit card customers.
- the decision-making process establishes the relationship between state, behavior and utility function, and then optimizes and solves the utility function corresponding to the extracted time series of the transaction data to be analyzed.
- the number of utility functions in the utility function set is the same as the number of elements in the transaction data time series to be analyzed corresponding to the utility function set.
- a utility function set is sequentially extracted from all the utility function sets; when the maximum likelihood inverse reinforcement learning is performed on the extracted utility function set, the utility function in the extracted utility function set is linearly superimposed.
- Carry out integration use maximum entropy inverse reinforcement learning to estimate the parameters of the integration results, and obtain a state transition probability matrix after the parameter estimation is completed, and use the obtained state transition probability matrix as the extracted utility function set. Describe the target state transition probability matrix.
- the above step of performing maximum likelihood inverse reinforcement learning on each of the utility function sets to perform parameter estimation to obtain the target state transition probability matrix of each of the historical transaction data to be analyzed includes:
- S4321 Perform linear superposition of the utility functions in each of the utility function sets, respectively, to obtain a plurality of individual utility functions to be estimated;
- S4323 Use the maximum entropy inverse reinforcement learning method to perform parameter estimation on each of the normalized personal utility functions, respectively, to obtain the target state transition probability matrix of each of the historical transaction data to be analyzed.
- This embodiment realizes the determination of the target state transition probability matrix according to the utility function set, fully considers the historical transaction data and the customer value mean data of each sub-category, and realizes the self-learning of the behavior of credit card customers through maximum likelihood inverse reinforcement learning.
- the generalization ability of user profiles of credit card customers has been improved.
- the utility function set is expressed as ⁇ U 1 , U 2 , U 3 , ... U n ⁇ , and the utility functions in the utility function set are linearly superimposed to obtain the to-be-estimated personal utility function U d , specifically expressed as:
- p 1 , p 2 , p 3 ?? p n are parameters that need to be estimated.
- the Softmax function is a normalized exponential function, which "compresses" a K-dimensional vector z containing any real number into another K-dimensional real vector ⁇ (z), so that each element is in the range of (0, 1 ), and the sum of all elements is 1.
- the method for normalizing the to-be-estimated personal utility function by using the softmax function can be selected from the prior art, and details are not described here.
- the result of parameter estimation is the target state transition probability matrix. That is, p 1 , p 2 , p 3 ?? p n form the target state transition probability matrix.
- the above-mentioned Markov decision-making process is used to extract the utility function according to each time series of the transaction data to be analyzed and the corresponding average value data of the customers to be analyzed, to obtain each transaction to be analyzed.
- the steps of the utility function collection for the data time series including:
- S4311 Use Markov decision-making process to construct a calculation formula for the maximum customer value total value of sub-categories according to each time series of the transaction data to be analyzed and the corresponding average data of the customer value to be analyzed, and obtain the maximum customer value of multiple sub-categories The formula for calculating the total value;
- S4312 Use the dynamic programming method to iteratively optimize and solve the calculation formula of the maximum customer value total value of each sub-category, and obtain the calculation formula of the maximum customer value total value of multiple target sub-categories;
- S4313 Extract a utility function from the calculation formula of the maximum customer value total value of each target sub-category, respectively, to obtain a utility function set of each time series of the transaction data to be analyzed.
- the utility function set for each time series of the transaction data to be analyzed is determined according to each time series of the transaction data to be analyzed and the corresponding average value data of the customers to be analyzed, and the historical transaction data is fully considered. It can fully mine the behavior of credit card customers when the business category, life cycle, and historical transaction data change, and improve the accuracy of customer value data of target sub-categories, which is conducive to improving the credit card customers’ value. User portrait accuracy.
- the time series of the transaction data to be analyzed is used as the state set, the average value of the customer value data to be analyzed is used as the behavior set, and the method for constructing the calculation formula of the maximum customer value total value of the subdivision by using the Markov decision process can be obtained from the prior art. selection, which will not be repeated here.
- the calculation formula of the maximum customer value total value of each sub-category is optimized and solved, which means finding an optimal strategy so that the interaction process of each state feature in the transaction data time series to be analyzed is always better than other Strategies pay more.
- the optimization solution is to maximize the value of the calculation formula of the maximum customer value total value of the subdivision category, and the utility function extracted when the value of the calculation formula of the maximum customer value total value of the subdivision category is the maximum value is the utility function of the optimal value.
- the method of iteratively using the dynamic programming method to optimize and solve the calculation formula of the maximum customer value total value of the sub-categories can be selected from the prior art, which will not be described in detail here.
- the Bellman equation is used to iteratively and optimally solve the calculation formula of the maximum customer value total value of the sub-category by using a dynamic programming method.
- the present application also proposes a user portrait device of a credit card customer, the device comprising:
- the request acquisition module 100 is configured to acquire a request for determining the value of a credit card customer, where the request for determining the value of a credit card customer carries target historical transaction data;
- the historical transaction data to be analyzed determination module 200 is used for classifying the target historical transaction data by business category and life cycle matching, and determining a plurality of historical transaction data to be analyzed and their corresponding target business categories and target life cycles;
- the target customer value average data determination module 300 is used to obtain customer value average data according to the target business category and target life cycle corresponding to each of the historical transaction data to be analyzed, and obtain the target of each historical transaction data to be analyzed.
- Customer value mean data
- the target state transition probability matrix determination module 400 is configured to use Markov decision process and maximum likelihood inverse reinforcement learning to perform state transition probability according to each of the historical transaction data to be analyzed and the corresponding target customer value mean data respectively. Matrix calculation to obtain the target state transition probability matrix of each of the historical transaction data to be analyzed;
- the target sub-category customer value data determination module 500 is configured to determine a plurality of target sub-category customer value data according to the target state transition probability matrix of each of the historical transaction data to be analyzed and the target customer value mean data;
- the target user portrait determination module 600 is configured to perform user portraits of credit card customers according to the plurality of target sub-category customer value data, and determine the target user portraits.
- the target historical transaction data is divided into business categories and the life cycle is matched, and a plurality of historical transaction data to be analyzed and their corresponding target business categories and target life cycles are determined, respectively, according to the target business corresponding to each historical transaction data to be analyzed.
- Category and target life cycle to obtain customer value mean data obtain the target customer value mean data of each historical transaction data to be analyzed, and use Markov decision process and maximum likelihood inverse reinforcement learning to analyze each historical transaction data and their respective data.
- target state transition probability matrix and target customer value mean data of each historical transaction data to be analyzed , determine multiple target subdivision customer value data, carry out user portrait of credit card customers according to multiple target subdivided customer value data, determine target user portrait, Markov decision-making can be used when the business category, life cycle, and historical transaction data change Fully mining the behavior of credit card customers, which is beneficial to improve the accuracy of the user portrait of credit card customers, and then realizes the self-learning of the behavior of credit card customers through maximum likelihood inverse reinforcement learning, and improves the generalization ability of the user portrait of credit card customers. , so as to meet the needs of marketing scenarios and help track the behavior of credit card customers for long-term training.
- an embodiment of the present application further provides a computer device.
- the computer device may be a server, and its internal structure may be as shown in FIG. 3 .
- the computer device includes a processor, memory, a network interface, and a database connected by a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities.
- the memory of the computer device includes a non-volatile storage medium, an internal memory.
- the nonvolatile storage medium stores an operating system, a computer program, and a database.
- the memory provides an environment for the execution of the operating system and computer programs in the non-volatile storage medium.
- the database of the computer equipment is used for storing data such as user portrait methods of credit card customers.
- the network interface of the computer device is used to communicate with an external terminal through a network connection.
- the computer program when executed by the processor, implements a method for user portraiture of a credit card customer.
- the method for user portraits of credit card customers includes: acquiring a request for determining the value of a credit card customer, the request for determining the value of a credit card customer carrying target historical transaction data; dividing the target historical transaction data by business category and life cycle matching, and determining multiple transactions.
- the customer value data is subdivided by the target; the user portrait of the credit card customer is performed according to the plurality of target subdivided customer value data, and the target user portrait is determined.
- the target historical transaction data is divided into business categories and the life cycle is matched, and a plurality of historical transaction data to be analyzed and their corresponding target business categories and target life cycles are determined, respectively, according to the target business corresponding to each historical transaction data to be analyzed.
- Category and target life cycle to obtain customer value mean data obtain the target customer value mean data of each historical transaction data to be analyzed, and use Markov decision process and maximum likelihood inverse reinforcement learning to analyze each historical transaction data and their respective data.
- target state transition probability matrix and target customer value mean data of each historical transaction data to be analyzed , determine multiple target subdivision customer value data, carry out user portrait of credit card customers according to multiple target subdivided customer value data, determine target user portrait, Markov decision-making can be used when the business category, life cycle, and historical transaction data change Fully mining the behavior of credit card customers, which is beneficial to improve the accuracy of the user portrait of credit card customers, and then realizes the self-learning of the behavior of credit card customers through maximum likelihood inverse reinforcement learning, and improves the generalization ability of the user portrait of credit card customers. , so as to meet the needs of marketing scenarios and help track the behavior of credit card customers for long-term training.
- An embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored.
- a user portrait method of a credit card customer is implemented, including the steps of: obtaining a request for determining the value of a credit card customer;
- the credit card customer value determination request carries the target historical transaction data;
- the target historical transaction data is divided into business categories and life cycle matching is performed to determine a plurality of historical transaction data to be analyzed and their corresponding target business categories and target life cycles; respectively;
- Markov decision process and Maximum likelihood inverse reinforcement learning calculates the state transition probability matrix according to each of the historical transaction data to be analyzed and the corresponding target customer value mean data, and obtains the target state transition probability of each historical transaction data to be analyzed.
- target state transition probability matrix of each of the historical transaction data to be analyzed and the target customer value mean data, determine a plurality of target sub-categories customer value data; according to the multiple target sub-category customer value data User portraits of credit card customers to determine target user portraits.
- the target historical transaction data is divided into business categories and life cycle matching, and a plurality of historical transaction data to be analyzed and their corresponding target business categories and target life cycles are determined.
- the target business category and target life cycle corresponding to the historical transaction data are used to obtain the customer value mean data, and the target customer value average data of each historical transaction data to be analyzed is obtained.
- Matrix and target customer value mean data determine multiple target sub-category customer value data, carry out user portrait of credit card customers according to multiple target sub-category customer value data, determine target user portrait, Markov decision-making can be in business category, life cycle .
- the behavior of credit card customers is fully explored, which is beneficial to improve the accuracy of the user portrait of credit card customers.
- the generalization ability of user portraits can meet the needs of marketing scenarios and help track the behavior of credit card customers for long-term training.
- the computer-readable storage medium may be non-volatile or volatile.
- Nonvolatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
- Volatile memory may include random access memory (RAM) or external cache memory.
- RAM is available in various forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Mathematical Physics (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Fuzzy Systems (AREA)
- Development Economics (AREA)
- Technology Law (AREA)
- Strategic Management (AREA)
- Computational Linguistics (AREA)
- Marketing (AREA)
- Economics (AREA)
- General Business, Economics & Management (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Computing Systems (AREA)
- Algebra (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
本申请涉及人工智能技术领域,揭示了一种信用卡客户的用户画像方法、装置、设备及介质,其中方法包括:将目标历史交易数据进行业务类别划分和生命周期匹配确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期;根据目标业务类别和目标生命周期获取客户价值均值数据得到目标客户价值均值数据;采用马尔科夫决策过程和最大似然逆强化学习分别根据每个待分析历史交易数据和各自对应的目标客户价值均值数据进行状态转移概率矩阵计算得到目标状态转移概率矩阵;分别根据每个待分析历史交易数据的目标状态转移概率矩阵和目标客户价值均值数据确定目标用户画像。在客户数据发生变化时充分挖掘信用卡客户的行为,提高用户画像的准确度。
Description
本申请要求于2020年11月17日提交中国专利局、申请号为2020112877471,发明名称为“信用卡客户的用户画像方法、装置、设备及介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
本申请涉及到人工智能技术领域,特别是涉及到一种信用卡客户的用户画像方法、装置、设备及介质。
用户画像,作为一种勾画目标用户、联系用户诉求与设计方向的有效工具,用户画像在各领域得到了广泛的应用。发明人意识到在信用卡场景中,客户资源的激烈争夺导致客户状态的难确定性和多变性,客户个性化需求的多样化意味着客户状态的难确定性和多变性,对信用卡客户的准确用户画像对针对信用卡客户的精准营销显得尤为重要。传统的用户画像只能对用户在单一场景下进行分析,不能针对信用卡客户的业务类别、生命周期、历史交易数据进行改变,从而导致对信用卡客户的用户画像的准确度不高,难以满足营销场景的需求,难以追踪信用卡客户的行为进行长期培养。
旨在解决现有技术对信用卡客户的用户画像的准确度不高,难以满足营销场景的需求,难以追踪信用卡客户的行为进行长期培养的技术问题。
本申请的主要目的为提供一种信用卡客户的用户画像方法、装置、设备及介质,旨在解决现有技术对信用卡客户的用户画像的准确度不高,难以满足营销场景的需求,难以追踪信用卡客户的行为进行长期培养的技术问题。
为了实现上述发明目的,本申请提出一种信用卡客户的用户画像方法,所述方法包括:
获取信用卡客户价值确定请求,所述信用卡客户价值确定请求携带有目标历史交易数据;
将所述目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期;
分别根据每个所述待分析历史交易数据各自对应的目标业务类别及目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的目标客户价值均值数据;
采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析历史交易数据和各自对应的所述目标客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵;
分别根据每个所述待分析历史交易数据的目标状态转移概率矩阵和所述目标客户价值均值数据,确定多个目标细分类客户价值数据;
根据所述多个目标细分类客户价值数据进行信用卡客户的用户画像,确定目标用户画像。
本申请还提出了一种计算机设备,包括存储器和处理器,所述存储器存储有 计算机程序,所述处理器执行所述计算机程序时实现如下方法步骤:
获取信用卡客户价值确定请求,所述信用卡客户价值确定请求携带有目标历史交易数据;
将所述目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期;进行业务类别划分和生命周期匹配;
分别根据每个所述待分析历史交易数据各自对应的目标业务类别及目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的目标客户价值均值数据;
采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析历史交易数据和各自对应的所述目标客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵;
分别根据每个所述待分析历史交易数据的目标状态转移概率矩阵和所述目标客户价值均值数据,确定多个目标细分类客户价值数据;
根据所述多个目标细分类客户价值数据进行信用卡客户的用户画像,确定目标用户画像。
本申请还提出了一种计算机可读存储介质,其上存储有计算机程序,所述计算机程序被处理器执行时实现如下方法步骤:
获取信用卡客户价值确定请求,所述信用卡客户价值确定请求携带有目标历史交易数据;
将所述目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期;进行业务类别划分和生命周期匹配;
分别根据每个所述待分析历史交易数据各自对应的目标业务类别及目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的目标客户价值均值数据;
采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析历史交易数据和各自对应的所述目标客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵;
分别根据每个所述待分析历史交易数据的目标状态转移概率矩阵和所述目标客户价值均值数据,确定多个目标细分类客户价值数据;
根据所述多个目标细分类客户价值数据进行信用卡客户的用户画像,确定目标用户画像。
本申请的信用卡客户的用户画像方法、装置、设备及介质,将目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期,分别根据每个待分析历史交易数据对应的目标业务类别和目标生命周期获取客户价值均值数据,得到每个待分析历史交易数据的目标客户价值均值数据,采用马尔科夫决策过程和最大似然逆强化学习分别根据每个待分析历史交易数据和各自对应的目标客户价值均值数据进行状态转移概率矩阵计算,得到每个待分析历史交易数据的目标状态转移概率矩阵,分别根据每个待分析历史交易数据的目标状态转移概率矩阵和目标客户价值均值数据,确定多个目标细分类客户价值数据,根据多个目标细分类客户价值数据进行 信用卡客户的用户画像,确定目标用户画像,马尔科夫决策能在业务类别、生命周期、历史交易数据发生变化时充分挖掘信用卡客户的行为,从而有利于提高信用卡客户的用户画像的准确度,然后通过最大似然逆强化学习实现了对信用卡客户的行为的自主学习,提高了信用卡客户的用户画像的泛化能力,从而满足营销场景的需求,有利于追踪信用卡客户的行为进行长期培养。
图1为本申请一实施例的信用卡客户的用户画像方法的流程示意图;
图2为本申请一实施例的信用卡客户的用户画像装置的结构示意框图;
图3为本申请一实施例的计算机设备的结构示意框图。
本申请目的实现、功能特点及优点将结合实施例,参照附图做进一步说明。
为了使本申请的目的、技术方案及优点更加清楚明白,以下结合附图及实施例,对本申请进行进一步详细说明。应当理解,此处描述的具体实施例仅仅用以解释本申请,并不用于限定本申请。
为了解决现有技术对信用卡客户的用户画像的准确度不高,难以满足营销场景的需求,难以追踪信用卡客户的行为进行长期培养的技术问题,本申请提出了一种信用卡客户的用户画像方法,所述方法应用于人工智能技术领域,所述方法进一步应用于人工智能的预测分析技术领域。所述信用卡客户的用户画像方法通过进行业务类别划分和生命周期匹配,然后对划分结果获取每个细分类的客户价值均值,采用马尔科夫决策过程和最大似然逆强化学习根据划分结果和每个细分类的客户价值均值确定每个细分类的状态转移概率矩阵,根据每个细分类的状态转移概率矩阵和每个细分类的客户价值均值确定客户潜在价值,整个过程充分考虑了客户状态,提高了确定的目标客户潜在价值的准确性,为精准营销提供了准确的数据支持,有利于实现信用卡客户的价值最大化。
参照图1,本申请实施例中提供一种信用卡客户的用户画像方法,所述方法包括:
S1:获取信用卡客户价值确定请求,所述信用卡客户价值确定请求携带有目标历史交易数据;
S2:将所述目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期;
S3:分别根据每个所述待分析历史交易数据各自对应的目标业务类别及目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的目标客户价值均值数据;
S4:采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析历史交易数据和各自对应的所述目标客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵;
S5:分别根据每个所述待分析历史交易数据的目标状态转移概率矩阵和所述目标客户价值均值数据,确定多个目标细分类客户价值数据;
S6:根据所述多个目标细分类客户价值数据进行信用卡客户的用户画像,确定目标用户画像。
本实施例将目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期,分别根据每个 待分析历史交易数据对应的目标业务类别和目标生命周期获取客户价值均值数据,得到每个待分析历史交易数据的目标客户价值均值数据,采用马尔科夫决策过程和最大似然逆强化学习分别根据每个待分析历史交易数据和各自对应的目标客户价值均值数据进行状态转移概率矩阵计算,得到每个待分析历史交易数据的目标状态转移概率矩阵,分别根据每个待分析历史交易数据的目标状态转移概率矩阵和目标客户价值均值数据,确定多个目标细分类客户价值数据,根据多个目标细分类客户价值数据进行信用卡客户的用户画像,确定目标用户画像,马尔科夫决策能在业务类别、生命周期、历史交易数据发生变化时充分挖掘信用卡客户的行为,从而有利于提高信用卡客户的用户画像的准确度,然后通过最大似然逆强化学习实现了对信用卡客户的行为的自主学习,提高了信用卡客户的用户画像的泛化能力,从而满足营销场景的需求,有利于追踪信用卡客户的行为进行长期培养。
对于S1,可以获取用户发送的信用卡客户价值确定请求,也可以是应用系统发送的信用卡客户价值确定请求。
信用卡客户价值确定请求,是指确定信用卡客户的客户潜在价值的请求。
目标历史交易数据,是以当前时间为结束时间第一预设时长内的信用卡客户的历史信用卡交易数据。
历史信用卡交易数据,也就是信用卡交易数据。信用卡交易数据包括但不限于:业务类别、交易金额、交易时间。
业务类别包括:基础消费、个人分期、商户分期、提取现金。基础消费,是信用卡客户在商户分期以外的刷卡消费。个人分期包括:账单分期、取现分期。商户分期,是指为商户商品定制的分期,信用卡客户购买该商户的商品时选择了分期的刷卡消费。提取现金,是信用卡客户将信用额度提取现金但是没有将提取的现金选择分期的消费。
对于S2,将所述目标历史交易数据依次进行业务类别划分和生命周期匹配,得到多个待分析历史交易数据,也就是说,同一待分析历史交易数据中的数据属于相同业务类别和相同生命周期。
所述各自对应的目标业务类别及目标生命周期,是指多个待分析历史交易数据各自对应的目标业务类别及目标生命周期。
将划分待分析历史交易数据所用的业务类别作为待分析历史交易数据对应的目标业务类别。将划分待分析历史交易数据所处的生命周期作为待分析历史交易数据对应的目标生命周期。
生命周期,是指信用卡客户的生命周期阶段。生命周期包括:获客期、接触期、成长期、成熟期、衰退期、沉默期,可以理解的是,生命周期还可以有其他划分方式和命名方式,在此不做具体限定。
客户状态包括:距今消费时长、消费频次、消费金额。距今消费时长包括:首次消费的距今消费时长、最后一次消费的距今消费时长。消费频次是指以当前时间为结束时间第二预设时长内的消费次数。
对于S3,依次从多个待分析历史交易数据中提取出待分析历史交易数据;根据提取出的所述待分析历史交易数据对应的所述目标业务类别和所述目标生命周期从数据库中获取客户价值均值数据,将获取的客户价值均值数据作为提取出的待分析历史交易数据的目标客户价值均值数据;重复上述步骤直至确定所有所述待分析历史交易数据的目标客户价值均值数据。
目标客户价值均值数据,也就是客户价值均值数据。客户价值均值数据是个 时间序列。也就是说,客户价值均值数据,是同一个业务类别和同一个生命周期的信用卡客户的客户价值的平均值。
客户价值均值数据包括但不限于:信用卡客户毛利润均值数据、信用卡客户净利润均值数据。
对于S4,采用马尔科夫决策过程分别根据每个所述待分析历史交易数据和各自对应的所述目标客户价值均值数据进行效用函数提取,得到多个效用函数集合;然后采用最大似然逆强化学习分别对每个所述效用函数集合进行进行参数估计,得到每个所述待分析历史交易数据的目标状态转移概率矩阵。每个所述待分析历史交易数据对应一个目标状态转移概率矩阵。
目标状态转移概率矩阵,也就是状态转移概率矩阵。状态转移概率矩阵,是客户状态转移的概率的矩阵。
对于S5,分别将每个所述待分析历史交易数据的目标状态转移概率矩阵和各自对应的所述目标客户价值均值数据进行相乘计算,得到多个待分析客户价值矩阵;分别将每个所述待分析客户价值矩阵的元素值进行求和计算,得到所述多个目标细分类客户价值数据。也就是说,每个所述待分析历史交易数据对应一个所述目标细分类客户价值数据。
比如,目标状态转移概率矩阵表述为:
目标客户价值均值数据G为:[G1G2G3G4G5],目标状态转移概率矩阵和目标客户价值均值数据相乘,将P和G相乘得到矩阵J,将矩阵J中所有元素的元素值进行求和计算,将计算结果作为目标细分类客户价值数据,在此举例不做具体限定。
目标细分类客户价值数据,就是目标历史交易数据对应的信用卡客户在目标业务类别在目标生命周期的客户价值。也就是说,目标细分类客户价值数据的数量和业务类别的数量相同。
对于S6,可选的,将所述多个目标细分类客户价值数据按时间形成矩阵将该矩阵也就是信用卡客户的用户画像,将该信用卡客户的用户画像作为目标用户画像。
目标用户画像,用于描述所述信用卡客户价值确定请求对应的信用卡客户在信用卡各业务类别的在时间点是否进行消费及消费带来的预期客户价值。
当客户价值均值数据是信用卡客户毛利润均值数据时,消费带来的预期客户价值是指毛利润数值;当客户价值均值数据是信用卡客户净利润均值数据时,消费带来的预期客户价值是指净利润数值。
比如,当客户价值均值数据是信用卡客户毛利润均值数据时,目标用户画像可以表述为:
其中,第一行表述目标业务类别为基础消费,第二行表述目标业务类别为个人分期,第三行表述目标业务类别为商户分期,第四行表述目标业务类别为提取现金,第一列表述第一时间、第二列表述第二时间、第三列表述第三时间、第四列表述第四时间;
目标用户画像的第一时间表述在个人分期和提取现金将进行消费,在第一时间针对该信用卡客户应该将精准营销定位在个人分期和提取现金,又由于个人分期带来的预期客户价值500元大于提取现金带来的预期客户价值100元,需要进一步将精准营销定位在个人分期;
目标用户画像的第三时间表述该信用卡客户不会进行消费,在第三时间将不需要对该信用卡客户进行精准营销,在此举例不做具体限定。
在一个实施例中,上述将所述目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期的步骤,包括:
S21:采用所述业务类别对所述目标历史交易数据进行划分,得到多个待分析历史交易数据和各自对应的所述目标业务类别;
S22:获取生命周期划分标准数据;
S23:采用所述生命周期划分标准数据分别根据每个所述待分析历史交易数据的距今消费时长、消费频次、消费金额、所述目标业务类别进行生命周期匹配,得到每个所述待分析历史交易数据对应的目标生命周期。
本实施例实现了先按业务类别和生命周期对所述目标历史交易数据进行划分,生命周期是根据待分析历史交易数据和目标业务类别确定的,充分考虑了业务类别、生命周期、历史交易数据,为后续确定目标细分类客户价值数据提供了数据基础。
对于S21,将所述目标历史交易数据中同一个业务类别的数据放在一个集合中,每个业务类别对应一个集合,每个集合的数据是一个待分析历史交易数据。将集合对应的业务类别作为待分析历史交易数据对应的目标业务类别。
对于S22,可以从数据库中获取生命周期划分标准数据。
生命周期划分标准数据,包括:业务类别、生命周期、距今消费时长标准值、消费频次标准值、消费金额标准值。
对于S23,依次从多个待分析历史交易数据中提取出待分析历史交易数据;将提取出的所述待分析历史交易数据的距今消费时长、消费频次、消费金额、所述目标业务类别在所述生命周期划分标准数据中进行匹配,将匹配到的生命周期作为提取出的所述待分析历史交易数据对应的目标生命周期;循环执行上述步骤直至确定所有所述待分析历史交易数据的所述目标生命周期。
比如,业务类别包括:基础消费、个人分期、商户分期、提取现金,生命周期包括:获客期、接触期、成长期、成熟期、衰退期、沉默期,信用卡客户A的目标历史交易数据采用所述业务类别进行划分,得到基础消费的历史交易数据为A1、个人分期的历史交易数据为A2、商户分期的历史交易数据为A3、提取现金的历史交易数据为A4,根据A1的距今消费时长、消费频次、消费金额进行生命周期划分得到A1的目标生命周期为成熟期,根据A2的距今消费时长、消费频次、消费金额进行生命周期划分得到A2的目标生命周期为成长期,根据A3的距今消费时长、消费频次、消费金额进行生命周期划分得到A3的目标生命周期为接触期,根据A4的距今消费时长、消费频次、消费金额进行生命周期划分 得到A4的目标生命周期为接触期,在此举例不做具体限定。
在一个实施例中,上述分别根据每个所述待分析历史交易数据各自对应的目标业务类别及目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的目标客户价值均值数据的步骤,包括:
S31:依次从所述多个待分析历史交易数据中提取出所述待分析历史交易数据;
S32:根据提取出的所述待分析历史交易数据对应的所述目标业务类别和所述目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的所述目标客户价值均值数据;
S33:重复上述步骤直至确定所有所述待分析历史交易数据的所述目标客户价值均值数据。
本实施例实现了对每个所述待分析历史交易数据确定目标客户价值均值数据,为后续确定目标细分类客户价值数据提供了数据基础。
对于S31,依次从所述多个待分析历史交易数据中提取出一个所述待分析历史交易数据的所有数据。
对于S32,获取客户价值均值列表;将提取出的所述待分析历史交易数据对应的所述目标业务类别和所述目标生命周期在所述客户价值均值列表中进行查找,将在所述客户价值均值列表中查找到的客户价值均值数据作为提取出的所述待分析历史交易数据对应的所述目标客户价值均值数据。
客户价值均值列表包括:业务类别、生命周期、客户价值均值数据。
对于S33,重复执行步骤S31和步骤S32直至确定所有所述待分析历史交易数据的所述目标客户价值均值数据。
在一个实施例中,上述采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析历史交易数据和各自对应的所述目标客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵的步骤之前,还包括:
S41:分别对每个所述待分析历史交易数据进行时间序列构建,得到多个待分析交易数据时间序列;
S42:将所述待分析历史交易数据的所述目标客户价值均值数据,作为所述待分析历史交易数据对应的所述待分析交易数据时间序列的待分析客户价值均值数据;
S43:采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵。
本实施例实现了根据每个所述待分析历史交易数据和各自对应的所述目标客户价值均值数据进行状态转移概率矩阵计算,充分考虑了历史交易数据和各细分类的客户价值均值数据,为确定目标细分类客户价值数据提供了准确的数据支持。
对于S41,分别对每个所述待分析历史交易数据按时间顺序进行时间序列构建,也就是每个所述待分析历史交易数据按时间顺序最终生成一个待分析交易数据时间序列。
客户价值均值数据的时间粒度与待分析交易数据时间序列的时间粒度相同。 比如,客户价值均值数据的时间粒度为每天,待分析交易数据时间序列的时间粒度为每天,也就是待分析交易数据时间序列每个元素代表的是每天的交易数据,则客户价值均值数据的时间粒度与待分析交易数据时间序列的时间粒度相同,在此举例不做具体限定。
对于S42,每个所述待分析历史交易数据的目标客户价值均值数据与自身对应的所述待分析交易数据时间序列的待分析客户价值均值数据相同。
对于S43,采用马尔科夫决策过程采用马尔科夫决策过程分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行效用函数提取,得到多个效用函数集合;然后采用最大似然逆强化学习分别对每个所述效用函数集合进行进行参数估计,得到每个所述待分析交易数据时间序列的状态转移概率矩阵;将所述待分析交易数据时间序列的状态转移概率矩阵作为对应的所述待分析历史交易数据的目标状态转移概率矩阵。也就是说,每个所述待分析交易数据时间序列对应一个效用函数集合。
效用函数,也就是马尔科夫决策过程的状态价值函数。
在一个实施例中,上述采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵的步骤,包括:
S431:采用马尔科夫决策过程分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行效用函数提取,得到每个所述待分析交易数据时间序列的效用函数集合;
S432:分别对每个所述效用函数集合进行最大似然逆强化学习进行参数估计,得到每个所述待分析历史交易数据的所述目标状态转移概率矩阵。
本实施例实现了采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行状态转移概率矩阵计算,充分考虑了历史信用卡交易数据和各细分类的客户价值均值数据,能在业务类别、生命周期、历史交易数据发生变化时充分挖掘信用卡客户的行为,提高了目标细分类客户价值数据的准确性,从而有利于提高信用卡客户的用户画像的准确度。
对于S431,依次从多个待分析交易数据时间序列中提取待分析交易数据时间序列;根据提取出的所述待分析交易数据时间序列和对应的所述待分析客户价值均值数据,基于马尔科夫决策过程建立状态、行为、效用函数的关系,然后对提取出的所述待分析交易数据时间序列对应的效用函数进行优化求解,根据提取出的所述待分析交易数据时间序列对应的优化求解结果确定提取出的所述待分析交易数据时间序列的效用函数集合;重复上述过程直至确定所有的所述待分析交易数据时间序列的效用函数集合。
效用函数集合中效用函数的数量与该效用函数集合对应的所述待分析交易数据时间序列中元素的个数相同。
对于S432,从所有所述效用函数集合中依次提取出效用函数集合;对提取出的效用函数集合进行最大似然逆强化学习时,采用线性叠加的方式将提取出的效用函数集合中的效用函数进行整合,采用最大熵逆强化学习对整合结果进行参数估计,参数估计完成得到状态转移概率矩阵,将得到的状态转移概率矩阵作为提取出的效用函数集合对应的所述待分析历史交易数据的所述目标状态转移概 率矩阵。
在一个实施例中,上述分别对每个所述效用函数集合进行最大似然逆强化学习进行参数估计,得到每个所述待分析历史交易数据的所述目标状态转移概率矩阵的步骤,包括:
S4321:分别对每个所述效用函数集合中的效用函数进行线性叠加,得到多个待估计个人效用函数;
S4322:采用softmax函数分别对每个所述待估计个人效用函数进行归一化处理,得到多个归一化个人效用函数;
S4323:采用最大熵逆强化学习方法分别对每个所述归一化个人效用函数进行参数估计,得到每个所述待分析历史交易数据的所述目标状态转移概率矩阵。
本实施例实现了根据效用函数集合确定目标状态转移概率矩阵,充分考虑了历史交易数据和各细分类的客户价值均值数据,通过最大似然逆强化学习实现了对信用卡客户的行为的自主学习,提高了信用卡客户的用户画像的泛化能力。
对于S4321,将所述效用函数集合表述为{U
1,U
2,U
3,……U
n},将对所述效用函数集合中的效用函数进行线性叠加,得到所述待估计个人效用函数U
d,具体表述为:
U
d=p
1U
1+p
2U
2+p
3U
3+……+p
nU
n
其中,p
1,p
2,p
3……p
n是需要估计的参数。
对于S4322,Softmax函数是归一化指数函数,将一个含任意实数的K维向量z“压缩”到另一个K维实向量σ(z)中,使得每一个元素的范围都在(0,1)之间,并且所有元素的和为1。
采用softmax函数对所述待估计个人效用函数进行归一化处理的方法可以从现有技术中选择,在此不做赘述。
对于S4323,采用最大熵逆强化学习方法分别对每个所述归一化个人效用函数进行参数估计的方法可以从现有技术中选择,在此不做赘述。
参数估计的结果即为目标状态转移概率矩阵。也就是将p
1,p
2,p
3……p
n组成目标状态转移概率矩阵。
在一个实施例中,上述采用马尔科夫决策过程分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行效用函数提取,得到每个所述待分析交易数据时间序列的效用函数集合的步骤,包括:
S4311:采用马尔科夫决策过程分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据构建细分类最大客户价值总值计算公式,得到多个细分类最大客户价值总值计算公式;
S4312:采用动态规划方法迭代分别对每个所述细分类最大客户价值总值计算公式进行优化求解,得到多个目标细分类最大客户价值总值计算公式;
S4313:分别从每个所述目标细分类最大客户价值总值计算公式中提取效用函数,得到每个所述待分析交易数据时间序列的效用函数集合。
本实施例实现了根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据确定每个所述待分析交易数据时间序列的效用函数集合,充分考虑了历史交易数据和各细分类的客户价值均值数据,能在业务类别、生命周期、历史交易数据发生变化时充分挖掘信用卡客户的行为,提高了目标细分类客户价值数据的准确性,从而有利于提高信用卡客户的用户画像的准确度。
对于S4311,从所有所述待分析交易数据时间序列依次提取出所述待分析交易数据时间序列;采用马尔科夫决策过程根据提取出的所述待分析交易数据时间序列和该所述待分析交易数据时间序列对应的所述待分析客户价值均值数据构建细分类最大客户价值总值计算公式,得到提取出的所述待分析交易数据时间序列对应的细分类最大客户价值总值计算公式;重复上述步骤直至确定所有所述待分析交易数据时间序列对应的细分类最大客户价值总值计算公式。
将所述待分析交易数据时间序列作为状态集,将所述待分析客户价值均值数据作为行为集,采用马尔科夫决策过程构建细分类最大客户价值总值计算公式的方法可以从现有技术中选择,在此不做赘述。
对于S4312,对每个所述细分类最大客户价值总值计算公式进行优化求解,意味着寻找一个最优的策略让所述待分析交易数据时间序列中各个状态特征的交互过程中获得始终比其它策略都要多的收获。优化求解就是使所述细分类最大客户价值总值计算公式的值最大,所述细分类最大客户价值总值计算公式的值最大时提取的效用函数是最优价值的效用函数。
采用动态规划方法迭代对所述细分类最大客户价值总值计算公式进行优化求解的方法可以从现有技术中选择,在此不做赘述。
可选的,采用贝尔曼方程对所述细分类最大客户价值总值计算公式采用动态规划方法迭代进行优化求解。
对于S4313,从所有所述细分类最大客户价值总值计算公式提取出所述细分类最大客户价值总值计算公式;从提取出的所述细分类最大客户价值总值计算公式中提取出效用函数,将提取出的效用函数放入集合,将该集合作为提取出的所述细分类最大客户价值总值计算公式对应的所述效用函数集合。
参照图2,本申请还提出了一种信用卡客户的用户画像装置,所述装置包括:
请求获取模块100,用于获取信用卡客户价值确定请求,所述信用卡客户价值确定请求携带有目标历史交易数据;
待分析历史交易数据确定模块200,用于将所述目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期;
目标客户价值均值数据确定模块300,用于分别根据每个所述待分析历史交易数据各自对应的目标业务类别及目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的目标客户价值均值数据;
目标状态转移概率矩阵确定模块400,用于采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析历史交易数据和各自对应的所述目标客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵;
目标细分类客户价值数据确定模块500,用于分别根据每个所述待分析历史交易数据的目标状态转移概率矩阵和所述目标客户价值均值数据,确定多个目标细分类客户价值数据;
目标用户画像确定模块600,用于根据所述多个目标细分类客户价值数据进行信用卡客户的用户画像,确定目标用户画像。
本实施例将目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期,分别根据每个待分析历史交易数据对应的目标业务类别和目标生命周期获取客户价值均值数 据,得到每个待分析历史交易数据的目标客户价值均值数据,采用马尔科夫决策过程和最大似然逆强化学习分别根据每个待分析历史交易数据和各自对应的目标客户价值均值数据进行状态转移概率矩阵计算,得到每个待分析历史交易数据的目标状态转移概率矩阵,分别根据每个待分析历史交易数据的目标状态转移概率矩阵和目标客户价值均值数据,确定多个目标细分类客户价值数据,根据多个目标细分类客户价值数据进行信用卡客户的用户画像,确定目标用户画像,马尔科夫决策能在业务类别、生命周期、历史交易数据发生变化时充分挖掘信用卡客户的行为,从而有利于提高信用卡客户的用户画像的准确度,然后通过最大似然逆强化学习实现了对信用卡客户的行为的自主学习,提高了信用卡客户的用户画像的泛化能力,从而满足营销场景的需求,有利于追踪信用卡客户的行为进行长期培养。
参照图3,本申请实施例中还提供一种计算机设备,该计算机设备可以是服务器,其内部结构可以如图3所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中,该计算机设计的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机程序和数据库。该内存器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。该计算机设备的数据库用于储存信用卡客户的用户画像方法等数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机程序被处理器执行时以实现一种信用卡客户的用户画像方法。所述信用卡客户的用户画像方法,包括:获取信用卡客户价值确定请求,所述信用卡客户价值确定请求携带有目标历史交易数据;将所述目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期;分别根据每个所述待分析历史交易数据各自对应的目标业务类别及目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的目标客户价值均值数据;采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析历史交易数据和各自对应的所述目标客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵;分别根据每个所述待分析历史交易数据的目标状态转移概率矩阵和所述目标客户价值均值数据,确定多个目标细分类客户价值数据;根据所述多个目标细分类客户价值数据进行信用卡客户的用户画像,确定目标用户画像。
本实施例将目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期,分别根据每个待分析历史交易数据对应的目标业务类别和目标生命周期获取客户价值均值数据,得到每个待分析历史交易数据的目标客户价值均值数据,采用马尔科夫决策过程和最大似然逆强化学习分别根据每个待分析历史交易数据和各自对应的目标客户价值均值数据进行状态转移概率矩阵计算,得到每个待分析历史交易数据的目标状态转移概率矩阵,分别根据每个待分析历史交易数据的目标状态转移概率矩阵和目标客户价值均值数据,确定多个目标细分类客户价值数据,根据多个目标细分类客户价值数据进行信用卡客户的用户画像,确定目标用户画像,马尔科夫决策能在业务类别、生命周期、历史交易数据发生变化时充分挖掘信用卡客户的行为,从而有利于提高信用卡客户的用户画像的准确度,然后通过最大似然逆强化学习实现了对信用卡客户的行为的自主学习,提高了信用卡客户的用户画 像的泛化能力,从而满足营销场景的需求,有利于追踪信用卡客户的行为进行长期培养。
本申请一实施例还提供一种计算机可读存储介质,其上存储有计算机程序,计算机程序被处理器执行时实现一种信用卡客户的用户画像方法,包括步骤:获取信用卡客户价值确定请求,所述信用卡客户价值确定请求携带有目标历史交易数据;将所述目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期;分别根据每个所述待分析历史交易数据各自对应的目标业务类别及目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的目标客户价值均值数据;采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析历史交易数据和各自对应的所述目标客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵;分别根据每个所述待分析历史交易数据的目标状态转移概率矩阵和所述目标客户价值均值数据,确定多个目标细分类客户价值数据;根据所述多个目标细分类客户价值数据进行信用卡客户的用户画像,确定目标用户画像。
上述执行的信用卡客户的用户画像方法,将目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期,分别根据每个待分析历史交易数据对应的目标业务类别和目标生命周期获取客户价值均值数据,得到每个待分析历史交易数据的目标客户价值均值数据,采用马尔科夫决策过程和最大似然逆强化学习分别根据每个待分析历史交易数据和各自对应的目标客户价值均值数据进行状态转移概率矩阵计算,得到每个待分析历史交易数据的目标状态转移概率矩阵,分别根据每个待分析历史交易数据的目标状态转移概率矩阵和目标客户价值均值数据,确定多个目标细分类客户价值数据,根据多个目标细分类客户价值数据进行信用卡客户的用户画像,确定目标用户画像,马尔科夫决策能在业务类别、生命周期、历史交易数据发生变化时充分挖掘信用卡客户的行为,从而有利于提高信用卡客户的用户画像的准确度,然后通过最大似然逆强化学习实现了对信用卡客户的行为的自主学习,提高了信用卡客户的用户画像的泛化能力,从而满足营销场景的需求,有利于追踪信用卡客户的行为进行长期培养。
所述计算机可读存储介质可以是非易失性,也可以是易失性。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程,是可以通过计算机程序来指令相关的硬件来完成,所述的计算机程序可存储于一非易失性计算机可读取存储介质中,该计算机程序在执行时,可包括如上述各方法的实施例的流程。其中,本申请所提供的和实施例中所使用的对存储器、存储、数据库或其它介质的任何引用,均可包括非易失性和/或易失性存储器。非易失性存储器可以包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限,RAM以多种形式可得,诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双速据率SDRAM(SSRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
需要说明的是,在本文中,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、装置、物品或者方法不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、装置、物品或者方法所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、装置、物品或者方法中还存在另外的相同要素。
以上所述仅为本申请的优选实施例,并非因此限制本申请的专利范围,凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换,或直接或间接运用在其他相关的技术领域,均同理包括在本申请的专利保护范围内。
Claims (20)
- 一种信用卡客户的用户画像方法,其中,所述方法包括:获取信用卡客户价值确定请求,所述信用卡客户价值确定请求携带有目标历史交易数据;将所述目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期;进行业务类别划分和生命周期匹配;分别根据每个所述待分析历史交易数据各自对应的目标业务类别及目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的目标客户价值均值数据;采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析历史交易数据和各自对应的所述目标客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵;分别根据每个所述待分析历史交易数据的目标状态转移概率矩阵和所述目标客户价值均值数据,确定多个目标细分类客户价值数据;根据所述多个目标细分类客户价值数据进行信用卡客户的用户画像,确定目标用户画像。
- 根据权利要求1所述的信用卡客户的用户画像方法,其中,所述将所述目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期的步骤,包括:采用所述业务类别对所述目标历史交易数据进行划分,得到多个待分析历史交易数据和各自对应的所述目标业务类别;获取生命周期划分标准数据;采用所述生命周期划分标准数据分别根据每个所述待分析历史交易数据的距今消费时长、消费频次、消费金额、所述目标业务类别进行生命周期匹配,得到每个所述待分析历史交易数据对应的目标生命周期。
- 根据权利要求1所述的信用卡客户的用户画像方法,其中,所述分别根据每个所述待分析历史交易数据各自对应的目标业务类别及目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的目标客户价值均值数据的步骤,包括:依次从所述多个待分析历史交易数据中提取出所述待分析历史交易数据;根据提取出的所述待分析历史交易数据对应的所述目标业务类别和所述目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的所述目标客户价值均值数据;重复上述步骤直至确定所有所述待分析历史交易数据的所述目标客户价值均值数据。
- 根据权利要求1所述的信用卡客户的用户画像方法,其中,所述采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析历史交易数据和各自对应的所述目标客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵的步骤之前,还包括:分别对每个所述待分析历史交易数据进行时间序列构建,得到多个待分析交易数据时间序列;将所述待分析历史交易数据的所述目标客户价值均值数据,作为所述待分析 历史交易数据对应的所述待分析交易数据时间序列的待分析客户价值均值数据;采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵。
- 根据权利要求4所述的信用卡客户的用户画像方法,其中,所述采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵的步骤,包括:采用马尔科夫决策过程分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行效用函数提取,得到每个所述待分析交易数据时间序列的效用函数集合;分别对每个所述效用函数集合进行最大似然逆强化学习进行参数估计,得到每个所述待分析历史交易数据的所述目标状态转移概率矩阵。
- 根据权利要求5所述的信用卡客户的用户画像方法,其中,所述分别对每个所述效用函数集合进行最大似然逆强化学习进行参数估计,得到每个所述待分析历史交易数据的所述目标状态转移概率矩阵的步骤,包括:分别对每个所述效用函数集合中的效用函数进行线性叠加,得到多个待估计个人效用函数;采用softmax函数分别对每个所述待估计个人效用函数进行归一化处理,得到多个归一化个人效用函数;采用最大熵逆强化学习方法分别对每个所述归一化个人效用函数进行参数估计,得到每个所述待分析历史交易数据的所述目标状态转移概率矩阵。
- 根据权利要求5所述的信用卡客户的用户画像方法,其中,所述采用马尔科夫决策过程分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行效用函数提取,得到每个所述待分析交易数据时间序列的效用函数集合的步骤,包括:采用马尔科夫决策过程分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据构建细分类最大客户价值总值计算公式,得到多个细分类最大客户价值总值计算公式;采用动态规划方法迭代分别对每个所述细分类最大客户价值总值计算公式进行优化求解,得到多个目标细分类最大客户价值总值计算公式;分别从每个所述目标细分类最大客户价值总值计算公式中提取效用函数,得到每个所述待分析交易数据时间序列的效用函数集合。
- 一种信用卡客户的用户画像装置,其中,所述装置包括:请求获取模块,用于获取信用卡客户价值确定请求,所述信用卡客户价值确定请求携带有目标历史交易数据;待分析历史交易数据确定模块,用于将所述目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期;目标客户价值均值数据确定模块,用于分别根据每个所述待分析历史交易数据各自对应的目标业务类别及目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的目标客户价值均值数据;目标状态转移概率矩阵确定模块,用于采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析历史交易数据和各自对应的所述目标客户价 值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵;目标细分类客户价值数据确定模块,用于分别根据每个所述待分析历史交易数据的目标状态转移概率矩阵和所述目标客户价值均值数据,确定多个目标细分类客户价值数据;目标用户画像确定模块,用于根据所述多个目标细分类客户价值数据进行信用卡客户的用户画像,确定目标用户画像。
- 一种计算机设备,包括存储器和处理器,所述存储器存储有计算机程序,其中,所述处理器执行所述计算机程序时实现如下方法步骤:获取信用卡客户价值确定请求,所述信用卡客户价值确定请求携带有目标历史交易数据;将所述目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期;进行业务类别划分和生命周期匹配;分别根据每个所述待分析历史交易数据各自对应的目标业务类别及目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的目标客户价值均值数据;采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析历史交易数据和各自对应的所述目标客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵;分别根据每个所述待分析历史交易数据的目标状态转移概率矩阵和所述目标客户价值均值数据,确定多个目标细分类客户价值数据;根据所述多个目标细分类客户价值数据进行信用卡客户的用户画像,确定目标用户画像。
- 根据权利要求9所述的计算机设备,其中,所述将所述目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期的步骤,包括:采用所述业务类别对所述目标历史交易数据进行划分,得到多个待分析历史交易数据和各自对应的所述目标业务类别;获取生命周期划分标准数据;采用所述生命周期划分标准数据分别根据每个所述待分析历史交易数据的距今消费时长、消费频次、消费金额、所述目标业务类别进行生命周期匹配,得到每个所述待分析历史交易数据对应的目标生命周期。
- 根据权利要求9所述的计算机设备,其中,所述分别根据每个所述待分析历史交易数据各自对应的目标业务类别及目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的目标客户价值均值数据的步骤,包括:依次从所述多个待分析历史交易数据中提取出所述待分析历史交易数据;根据提取出的所述待分析历史交易数据对应的所述目标业务类别和所述目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的所述目标客户价值均值数据;重复上述步骤直至确定所有所述待分析历史交易数据的所述目标客户价值均值数据。
- 根据权利要求9所述的计算机设备,其中,所述采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析历史交易数据和各自对应的所 述目标客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵的步骤之前,还包括:分别对每个所述待分析历史交易数据进行时间序列构建,得到多个待分析交易数据时间序列;将所述待分析历史交易数据的所述目标客户价值均值数据,作为所述待分析历史交易数据对应的所述待分析交易数据时间序列的待分析客户价值均值数据;采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵。
- 根据权利要求12所述的计算机设备,其中,所述采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵的步骤,包括:采用马尔科夫决策过程分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行效用函数提取,得到每个所述待分析交易数据时间序列的效用函数集合;分别对每个所述效用函数集合进行最大似然逆强化学习进行参数估计,得到每个所述待分析历史交易数据的所述目标状态转移概率矩阵。
- 根据权利要求13所述的计算机设备,其中,所述分别对每个所述效用函数集合进行最大似然逆强化学习进行参数估计,得到每个所述待分析历史交易数据的所述目标状态转移概率矩阵的步骤,包括:分别对每个所述效用函数集合中的效用函数进行线性叠加,得到多个待估计个人效用函数;采用softmax函数分别对每个所述待估计个人效用函数进行归一化处理,得到多个归一化个人效用函数;采用最大熵逆强化学习方法分别对每个所述归一化个人效用函数进行参数估计,得到每个所述待分析历史交易数据的所述目标状态转移概率矩阵。
- 一种计算机可读存储介质,其上存储有计算机程序,其中,所述计算机程序被处理器执行时实现如下方法步骤:获取信用卡客户价值确定请求,所述信用卡客户价值确定请求携带有目标历史交易数据;将所述目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期;进行业务类别划分和生命周期匹配;分别根据每个所述待分析历史交易数据各自对应的目标业务类别及目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的目标客户价值均值数据;采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析历史交易数据和各自对应的所述目标客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵;分别根据每个所述待分析历史交易数据的目标状态转移概率矩阵和所述目标客户价值均值数据,确定多个目标细分类客户价值数据;根据所述多个目标细分类客户价值数据进行信用卡客户的用户画像,确定目标用户画像。
- 根据权利要求15所述的计算机可读存储介质,其中,所述将所述目标历史交易数据进行业务类别划分和生命周期匹配,确定多个待分析历史交易数据和各自对应的目标业务类别及目标生命周期的步骤,包括:采用所述业务类别对所述目标历史交易数据进行划分,得到多个待分析历史交易数据和各自对应的所述目标业务类别;获取生命周期划分标准数据;采用所述生命周期划分标准数据分别根据每个所述待分析历史交易数据的距今消费时长、消费频次、消费金额、所述目标业务类别进行生命周期匹配,得到每个所述待分析历史交易数据对应的目标生命周期。
- 根据权利要求15所述的计算机可读存储介质,其中,所述分别根据每个所述待分析历史交易数据各自对应的目标业务类别及目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的目标客户价值均值数据的步骤,包括:依次从所述多个待分析历史交易数据中提取出所述待分析历史交易数据;根据提取出的所述待分析历史交易数据对应的所述目标业务类别和所述目标生命周期获取客户价值均值数据,得到每个所述待分析历史交易数据的所述目标客户价值均值数据;重复上述步骤直至确定所有所述待分析历史交易数据的所述目标客户价值均值数据。
- 根据权利要求15所述的计算机可读存储介质,其中,所述采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析历史交易数据和各自对应的所述目标客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵的步骤之前,还包括:分别对每个所述待分析历史交易数据进行时间序列构建,得到多个待分析交易数据时间序列;将所述待分析历史交易数据的所述目标客户价值均值数据,作为所述待分析历史交易数据对应的所述待分析交易数据时间序列的待分析客户价值均值数据;采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵。
- 根据权利要求18所述的计算机可读存储介质,其中,所述采用马尔科夫决策过程和最大似然逆强化学习分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行状态转移概率矩阵计算,得到每个所述待分析历史交易数据的目标状态转移概率矩阵的步骤,包括:采用马尔科夫决策过程分别根据每个所述待分析交易数据时间序列和各自对应的所述待分析客户价值均值数据进行效用函数提取,得到每个所述待分析交易数据时间序列的效用函数集合;分别对每个所述效用函数集合进行最大似然逆强化学习进行参数估计,得到每个所述待分析历史交易数据的所述目标状态转移概率矩阵。
- 根据权利要求19所述的计算机可读存储介质,其中,所述分别对每个所述效用函数集合进行最大似然逆强化学习进行参数估计,得到每个所述待分析历史交易数据的所述目标状态转移概率矩阵的步骤,包括:分别对每个所述效用函数集合中的效用函数进行线性叠加,得到多个待估计个人效用函数;采用softmax函数分别对每个所述待估计个人效用函数进行归一化处理,得到多个归一化个人效用函数;采用最大熵逆强化学习方法分别对每个所述归一化个人效用函数进行参数估计,得到每个所述待分析历史交易数据的所述目标状态转移概率矩阵。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011287747.1 | 2020-11-17 | ||
CN202011287747.1A CN112395331B (zh) | 2020-11-17 | 2020-11-17 | 信用卡客户的用户画像方法、装置、设备及介质 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022105177A1 true WO2022105177A1 (zh) | 2022-05-27 |
Family
ID=74606194
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2021/096978 WO2022105177A1 (zh) | 2020-11-17 | 2021-05-28 | 信用卡客户的用户画像方法、装置、设备及介质 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN112395331B (zh) |
WO (1) | WO2022105177A1 (zh) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116186351A (zh) * | 2023-04-23 | 2023-05-30 | 武汉想不到网络科技有限公司 | 一种基于加密传输的数据优化系统 |
CN116797341A (zh) * | 2023-08-17 | 2023-09-22 | 北京信大融金教育科技有限公司 | 基于运营平台的积分产生和管理方法、装置、设备及介质 |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112395331B (zh) * | 2020-11-17 | 2023-10-10 | 平安科技(深圳)有限公司 | 信用卡客户的用户画像方法、装置、设备及介质 |
CN115578157B (zh) * | 2022-11-13 | 2023-10-24 | 南京数族信息科技有限公司 | 一种基于路由策略的助贷平台动态路由方法 |
CN117408531B (zh) * | 2023-10-26 | 2024-05-07 | 北京赛博数智科技有限公司 | 一种大数据智能匹配的客户信息管理方法及系统 |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106603291A (zh) * | 2016-12-19 | 2017-04-26 | 中国联合网络通信集团有限公司 | 用户画像信息处理方法及装置 |
CN107767259A (zh) * | 2017-09-30 | 2018-03-06 | 平安科技(深圳)有限公司 | 贷款风险控制方法、电子装置及可读存储介质 |
CN110704730A (zh) * | 2019-09-06 | 2020-01-17 | 中国平安财产保险股份有限公司 | 基于大数据的产品数据推送方法、系统及计算机设备 |
CN112395331A (zh) * | 2020-11-17 | 2021-02-23 | 平安科技(深圳)有限公司 | 信用卡客户的用户画像方法、装置、设备及介质 |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105824912A (zh) * | 2016-03-15 | 2016-08-03 | 平安科技(深圳)有限公司 | 基于用户画像的个性化推荐方法和装置 |
CN107862413A (zh) * | 2017-11-16 | 2018-03-30 | 四维口袋科技(北京)有限公司 | 基于马尔科夫链的预测方法和装置 |
WO2019143370A1 (en) * | 2018-01-22 | 2019-07-25 | Visa International Service Association | Method, system, and computer program product for real-time data aggregation |
CN109300050A (zh) * | 2018-08-31 | 2019-02-01 | 平安科技(深圳)有限公司 | 基于用户画像的保险推送方法、装置及存储介质 |
CN111709765A (zh) * | 2020-03-25 | 2020-09-25 | 中国电子科技集团公司电子科学研究院 | 一种用户画像评分方法、装置和存储介质 |
CN111597348B (zh) * | 2020-04-27 | 2024-02-06 | 平安科技(深圳)有限公司 | 用户画像方法、装置、计算机设备和存储介质 |
-
2020
- 2020-11-17 CN CN202011287747.1A patent/CN112395331B/zh active Active
-
2021
- 2021-05-28 WO PCT/CN2021/096978 patent/WO2022105177A1/zh active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106603291A (zh) * | 2016-12-19 | 2017-04-26 | 中国联合网络通信集团有限公司 | 用户画像信息处理方法及装置 |
CN107767259A (zh) * | 2017-09-30 | 2018-03-06 | 平安科技(深圳)有限公司 | 贷款风险控制方法、电子装置及可读存储介质 |
CN110704730A (zh) * | 2019-09-06 | 2020-01-17 | 中国平安财产保险股份有限公司 | 基于大数据的产品数据推送方法、系统及计算机设备 |
CN112395331A (zh) * | 2020-11-17 | 2021-02-23 | 平安科技(深圳)有限公司 | 信用卡客户的用户画像方法、装置、设备及介质 |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116186351A (zh) * | 2023-04-23 | 2023-05-30 | 武汉想不到网络科技有限公司 | 一种基于加密传输的数据优化系统 |
CN116797341A (zh) * | 2023-08-17 | 2023-09-22 | 北京信大融金教育科技有限公司 | 基于运营平台的积分产生和管理方法、装置、设备及介质 |
CN116797341B (zh) * | 2023-08-17 | 2023-11-21 | 北京信大融金教育科技有限公司 | 基于运营平台的积分产生和管理方法、装置、设备及介质 |
Also Published As
Publication number | Publication date |
---|---|
CN112395331B (zh) | 2023-10-10 |
CN112395331A (zh) | 2021-02-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2022105177A1 (zh) | 信用卡客户的用户画像方法、装置、设备及介质 | |
Ribeiro et al. | Transfer learning with seasonal and trend adjustment for cross-building energy forecasting | |
Cimadomo et al. | Nowcasting with large Bayesian vector autoregressions | |
Schorfheide et al. | Real-time forecasting with a mixed-frequency VAR | |
Xu et al. | Composite quantile regression neural network with applications | |
Ghysels et al. | Estimating MIDAS regressions via OLS with polynomial parameter profiling | |
Banerjee et al. | Forecasting with factor-augmented error correction models | |
Xu et al. | A novel UMIDAS-SVQR model with mixed frequency investor sentiment for predicting stock market volatility | |
CN112381154A (zh) | 预测用户概率的方法、装置和计算机设备 | |
CN110223137A (zh) | 产品组合推荐方法、装置、计算机设备和存储介质 | |
Cipiloglu Yildiz et al. | A portfolio construction framework using LSTM‐based stock markets forecasting | |
CN111951045A (zh) | 一种基于大数据的电费核算方法及装置 | |
Wang et al. | Multivariate time series prediction based on optimized temporal convolutional networks with stacked auto-encoders | |
Xu et al. | Group penalized unrestricted mixed data sampling model with application to forecasting US GDP growth | |
Liu et al. | Non-homogeneous volatility correlations in the bivariate multifractal model | |
Gupta et al. | Has the SARB become more effective post inflation targeting? | |
Shamshad et al. | Forecasting and Trading of the Stable Cryptocurrencies with Machine Learning and Deep Learning Algorithms for Market Analytics | |
Kočenda et al. | Nowcasting real GDP growth: comparison between old and new EU countries | |
Kim et al. | Likelihood inference for dynamic linear models with Markov switching parameters: on the efficiency of the Kim filter | |
Berg | Multivariate forecasting with BVARs and DSGE models | |
CN116800831A (zh) | 一种业务数据推送方法、装置、存储介质及处理器 | |
Loonat et al. | Learning zero-cost portfolio selection with pattern matching | |
Meligkotsidou et al. | Detecting structural breaks in multivariate financial time series: evidence from hedge fund investment strategies | |
Nonejad | Predicting equity premium using dynamic model averaging. Does the state–space representation matter? | |
Boccanfuso et al. | Parametric and nonparametric income distribution estimators in CGE micro-simulation modeling |
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: 21893332 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: 21893332 Country of ref document: EP Kind code of ref document: A1 |