CN109670927A - The method of adjustment and its device of credit line, equipment, storage medium - Google Patents
The method of adjustment and its device of credit line, equipment, storage medium Download PDFInfo
- Publication number
- CN109670927A CN109670927A CN201710952035.9A CN201710952035A CN109670927A CN 109670927 A CN109670927 A CN 109670927A CN 201710952035 A CN201710952035 A CN 201710952035A CN 109670927 A CN109670927 A CN 109670927A
- Authority
- CN
- China
- Prior art keywords
- data information
- output result
- neural network
- target user
- input
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/03—Credit; Loans; Processing thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Technology Law (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the present invention provides the method for adjustment and its device, equipment, storage medium of a kind of credit line, wherein the described method includes: the identification information of the target user based on acquisition, obtains the data information of the target user;The state vector that the target user is determined according to the data information of the target user determines the corresponding tactful income vector of the state vector according to the state vector of the target user;State vector described in the tactful income vector sum is input to the first full Connection Neural Network, determines the adjustment direction and adjusted value of credit line;The target credit line of the target user is determined according to the current credit degree of the adjustment direction, adjusted value and the target user.
Description
Technical field
The present invention relates to the method for adjustment and its device of Internet technical field more particularly to a kind of credit line, equipment,
Storage medium.
Background technique
Currently, with the development of economy, the basic guarantee that people are no longer content with eating one's fill and wearing warm clothes often has some advanced
Consumption, and these excessive consumptions needs are got a bank loan.Loan on personal security is bank or other financial institutions to having sure credibility
Borrower provide without the RMB fiduciary loan tendered guarantee, and credit line is foundation credit side's income, business revenue, debt
Than condition elements such as, occupation, properties, by the amount of money of debit's unilateral decision.This means that debit will have higher credit line,
The various conditions of debit all must be quite excellent, and credit side can just be ready to provide higher credit amount.Traditional automatic tune volume system
System, is all based on the rule manually customized, can not automatically adapt to the variation of current financial situation and User Status.
Summary of the invention
In view of this, an embodiment of the present invention is intended to provide a kind of method of adjustment of credit line and its device, equipment, storages
Medium, the adjustment for solving credit line in prior art can not automatically adapt to current financial situation and User Status
The problem of variation, can deeply study utilize the multi-dimensional data information of user, and it is corresponding according to the state vector of user
With the corresponding relationship of tactful income vector, the adjustable strategies of credit line are determined, to be adjusted automatically to credit line.
The technical solution of the embodiment of the present invention is achieved in that
The embodiment of the present invention provides a kind of method of adjustment of credit line, which comprises
The identification information of target user based on acquisition obtains the data information of the target user;
The state vector that the target user is determined according to the data information of the target user, according to the target user
State vector determine the corresponding tactful income vector of the state vector;
State vector described in the tactful income vector sum is input to the first full Connection Neural Network, determines credit line
Adjustment direction and adjusted value;
Determine the target user's according to the current credit degree of the adjustment direction, adjusted value and the target user
Target credit line.
The embodiment of the present invention provides a kind of adjustment device of credit line, and described device includes: the first acquisition module, first
Determining module, the second determining module and third determining module, in which:
The first acquisition module obtains the target user's for the identification information of the target user based on acquisition
Data information;
First determining module, for determining the state of the target user according to the data information of the target user
Vector determines the corresponding tactful income vector of the state vector according to the state vector of the target user;
Second determining module, for state vector described in the tactful income vector sum to be input to the first full connection
Neural network determines the adjustment direction and adjusted value of credit line;
The third determining module, for according to the adjustment direction, adjusted value and the current credit of the target user
Amount determines the target credit line of the target user.
The embodiment of the present invention provides a kind of adjustment equipment of credit line, and the equipment includes at least: memory, communication are total
Line and processor, in which:
The memory, for storing the adjustment programme of credit line;
The communication bus, for realizing the connection communication between processor and memory;
The processor, for executing the adjustment programme of the credit line stored in memory, to perform the steps of
The identification information of target user based on acquisition obtains the data information of the target user;
The state vector that the target user is determined according to the data information of the target user, according to the target user
State vector determine the corresponding tactful income vector of the state vector;
State vector described in the tactful income vector sum is input to the first full Connection Neural Network, determines credit line
Adjustment direction and adjusted value;
Determine the target user's according to the current credit degree of the adjustment direction, adjusted value and the target user
Target credit line.
The embodiment of the present invention provides a kind of computer readable storage medium, is stored on the computer readable storage medium
The adjustment programme of the adjustment programme of credit line, the credit line realizes credit line as described above when being executed by processor
Method of adjustment the step of.
The embodiment of the present invention provides the method for adjustment and its device, equipment, storage medium of a kind of credit line, wherein first
The first identification information of the target user based on acquisition obtains the data information of the target user;Then it is used according to the target
The data information at family determines the state vector of the target user, and determines the shape according to the state vector of the target user
The corresponding tactful income vector of state vector;State vector described in the tactful income vector sum is input to the first full connection mind again
Through network, the adjustment direction and adjusted value of credit line are determined;Finally used according to the adjustment direction, adjusted value and the target
The current credit degree at family determines the target credit line of the target user, so, it is possible deeply study and utilizes user
Multi-dimensional data information determine line of credit and according to the corresponding corresponding relationship with tactful income vector of the state vector of user
The adjustable strategies of degree, to be adjusted automatically to credit line.
Detailed description of the invention
Fig. 1 is the schematic diagram of the implementation environment of the method for adjustment of credit line provided in an embodiment of the present invention;
Fig. 2-1 is the implementation process schematic diagram of the method for adjustment of credit line of the embodiment of the present invention;
Fig. 2-2 is the schematic diagram of the full neural network of the embodiment of the present invention first;
Fig. 3 is the implementation process schematic diagram of the method for adjustment of credit line of the embodiment of the present invention;
Fig. 4 is that the composed structure of automated intelligent tune volume system of the embodiment of the present invention based on user's various dimensions information is illustrated
Figure;
Fig. 5 is state aware of embodiment of the present invention module diagram;
Fig. 6 is the course of work schematic diagram of strategy of embodiment of the present invention income iteration module;
Fig. 7 is the course of work schematic diagram of strategy of embodiment of the present invention output module;
Fig. 8 is the composed structure schematic diagram of the adjustment device of credit line of the embodiment of the present invention;
Fig. 9 is the composed structure schematic diagram of the adjustment equipment of credit line of the embodiment of the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, the specific technical solution of invention is described in further detail.The following examples are intended to illustrate the invention, but does not have to
To limit the scope of the invention.
Fig. 1 is the schematic diagram of the implementation environment of the method for adjustment of credit line provided in an embodiment of the present invention, such as Fig. 1 institute
Show, which includes credit line adjustment platform 101, loan platform 102, social application server 103 and target user
Terminal 104, wherein credit line adjusts between platform 101 and loan platform 102 and credit line adjusts platform 101 and society
It hands between application server 103, loan platform 102 and target terminal user 104, target terminal user 104 and social application take
It is all to be interacted by being connected to the network row information of going forward side by side between business device 103.
Credit line adjusts platform 101, for obtaining the data information of target user, and root from social application server
The credit line adjustable strategies that target user is determined according to data information, further according to the credit line adjustable strategies and mesh of target user
The current credit line of mark user determines target credit line, and target credit line is sent to loan platform 102.Line of credit
Degree adjustment platform 101 can be a server, or the server cluster consisted of several servers or a cloud
Calculate service centre.
Loan platform 102, for being offered loans according to target credit line for user.Loan platform 102 can be one
Server, or the server cluster consisted of several servers or a cloud computing service center.
Social application server 103 is the corresponding server of social application program installed in target terminal user 104.It can
To be a server, or the server cluster consisted of several servers or a cloud computing service center.
Social application program is installed, which refers to the Tencent similar to China in target terminal user 104
The QQ and wechat of scientific and technological (Shenzhen) Co., Ltd, the Michat of Chinese Beijing Xiaomi Technology Co., Ltd. and China it is new
The application programs such as the Sina weibo of unrestrained network technology limited liability company.The target terminal user can be smart phone, plate
Computer, pocket computer on knee and desktop computer etc..
Schematic diagram as shown in connection with fig. 1, below to each of the adjustment device of the method for adjustment of credit line and credit line
Embodiment is illustrated.
Embodiment in order to better understand the present invention, it is various to deep learning and the present embodiments relate to what is arrived here
Neural network is explained.
Deep learning (Deep Learning) has overturned image recognition, natural language understanding, language since two thousand six
The fields such as sound identification and synthesis have driven the third time wave of artificial intelligence so that very much " impossible tasks " becomes " possibility "
Tide.Neural network can efficiently indicate high dimensional data, be mapped as low-dimensional data.Deep neural network, i.e., study is more
The neural network that data are indicated during layer is abstract has greatly improved speech recognition, Object identifying, object detection, prediction medicine
Object molecular activity and other many technologies.Deep learning indicates (supervised learning, unsupervised learning, strong by the way that building is distributed
Chemistry is practised) discovery labyrinth is concentrated in large data.
General artificial intelligence requires model to emphasize have the same multiple intelligent behavior of people, including perception, decision, reasoning with
Planning and exchange and conmmunication.Most important of which is that learning ability.Model should have an independent learning ability, for example people is from small
Suffered supervised learning is less, is more continuous autonomous learning after contacting with environment;By exchange, communication and feedback come
Study.Currently, this some learning ability is all that deep learning is short of.Interaction with environment is autonomous learning, self evolves
Basis, generally entail specific target;Intensified learning (Reinforcement Learning), namely pass through trial and error, simple
The study formula that ground is completed by reward or punishment.Intensified learning wishes to be learned according to the interaction and feedback with external environment
Acquistion is to an optimal strategy;Intensified learning is the only way which must be passed for realizing strong artificial intelligence.Using deep learning to extraneous ring
The powerful expression ability combination intensified learning in border realizes strong artificial intelligence, possesses huge potentiality.
General artificial intelligence be to create it is a kind of without manually program oneself association solve the problems, such as various intelligent bodies, most
Whole target is to realize class people the rank even intelligence of superman's rank.The behavior of intelligent body can be attributed to the interaction with the world.
Intelligent body observes this world, then according to the observation and the state output of itself movement, this world can therefore and change,
Intelligent body is returned to form feedback.General artificial intelligence has been achieved with huge breakthrough in fields such as game AI, gos, together
Sample also has many valuable researchs in other field, and then substitutes the existing work needed by being accomplished manually.
Recognition with Recurrent Neural Network: a type of neural network, the node between hidden layer is no longer connectionless but has connection
, and it further includes the output of last moment hidden layer that the input of hidden layer, which not only includes the output of input layer,.
Hidden layer: other each layers in addition to input layer and output layer are called hidden layer, that is to say, that hidden layer is not direct
Receive extraneous signal, also sends signal not directly to the external world.
Input layer: only serve input signal is fanned out to effect.So be not credited to when calculating the number of plies of neural network, it should
Layer is responsible for receiving the information from network-external, is referred to as the 0th layer.
Output layer: being the last layer of neural network, and the maximum level number with the network is responsible for the meter of output nerve network
Calculate result.
Convolutional neural networks: by full-mesh layer (corresponding classical neural network) group of one or more convolutional layers and top
At, while also including associated weights and pond layer (pooling layer).This structure enables convolutional neural networks to utilize
The two-dimensional structure of input data.
Full Connection Neural Network: for n-1 layers and n-layer, n-1 layers any one node, all with all nodes of n-th layer
There is connection.That is for each node of n-th layer when being calculated, the input of activation primitive is the output of n-1 layers of all nodes
It is weighted with weight.
The embodiment of the present invention provides a kind of method of adjustment of credit line, and Fig. 2-1 is credit line of the embodiment of the present invention
The implementation process schematic diagram of method of adjustment the described method comprises the following steps as shown in Fig. 2-1:
Step S201, the identification information of the target user based on acquisition obtain the data information of the target user.
In the present embodiment, the identification information of target user can be unique distinguishable using user's having in social networks
The information for the property known, including text (such as user name), image (such as two dimensional code), naturally it is also possible to specific using other of target user
Identify such as ID card No., phone number, email address.
User can generate various types of behavioral datas in social networks, such as pay close attention to the dynamic of good friend, browse good friend master
Page issues the Social behaviors data such as personal dynamic, message;Information as user class being obtained by crawler software.
Step S202 determines the state vector of the target user according to the data information of the target user, according to institute
The state vector for stating target user determines the corresponding tactful income vector of the state vector.
It in the present embodiment, may include the data information of multidimensional in the data information of the target user, such as can be with
Including but not limited to interesting data information, consumption data information and behavioral data information, in which:
Interesting data information can reflect the information such as the hobby of user, can read in application and obtain from social category
's.
Consumption data information can be after lend-borrow action occurs for target user, the data information of the consumption of generation and return
The also data information of loaning bill principal and interest.Consumption data information can be from loan platform acquisition.
Behavioral data information can be user in the behavioral data of social platform, can be regarded as being made of several behaviors
Stream data, reflect the Behavior preference of user, and behavioral data information can be to be obtained from social application server.
During realization, determine that the state vector of the target user is logical according to the data information of the target user
It crosses and the data information of each dimension is input in a Recognition with Recurrent Neural Network respectively, the corresponding output letter for obtaining each dimension
The output information of each dimension, is then input in a full Connection Neural Network, last output state vector by breath again.At this time
Obtained state vector is that the data information for each dimension that deeply has learnt target user obtains, so that final letter
It is more accurate, comprehensive with the adjustable strategies of amount.
The state vector according to the target user determines the corresponding tactful income vector of the state vector in reality
During now, the income probability distribution letter for determining obtain under tune volume strategy different under the state vector first can be
Then breath carries out maximum pondization operation, at this time completion an iteration process, by iteration preset times, determine the state to
Measure corresponding tactful income vector.
In the present embodiment, state vector can reflect the hobby of target user, consumption habit, Behavior preference it is comprehensive
Conjunction state, the element number for including in the state vector are pre-set, and the state vector of reality output is by the number that inputs
According to decision.What tactful income vector reflected is the corresponding relationship of tune volume strategy and income, wherein volume strategy is adjusted to can be primary tune
Volume such as raises 100 yuan, can also be the combination for repeatedly adjusting volume, such as can be 100 yuan of up-regulation, 300 yuan of up-regulation, lowers 50 yuan
The combination of volume is adjusted three times.The number for the element for including in tactful income vector is also to pre-set.
State vector described in the tactful income vector sum is input to the first full Connection Neural Network, really by step S203
Determine the adjustment direction and adjusted value of credit line.
Fig. 2-2 is the schematic diagram of the full neural network of the embodiment of the present invention first, as shown in Fig. 2-2, in the first full nerve net
In network, 221 be input layer, and 222 be the 1st layer of hidden layer, and 223 be the 2nd layer of hidden layer, and 224 be output layer, wherein the 1st layer is hidden
Any one node of layer 222, such as node 2221, all have connection with the 2nd layer of all node of hidden layer 223.
In the present embodiment, adjustment direction includes that upper reconciliation is lowered.Up-regulation is to increase credit line, and lowering is to reduce
Credit line.Adjusted value is generally as unit of member, such as adjusted value can be 500 yuan, 1000 yuan etc..
Step S204, according to the determination of the current credit degree of the adjustment direction, adjusted value and the target user
The target credit line of target user.
In the present embodiment, it is determined according to the current credit degree of the adjustment direction, adjusted value and the target user
The target credit line of the target user can obtain the current credit degree of target user first in the actual implementation process,
Then if adjustment direction is up-regulation, current credit degree is obtained into the target line of credit of the target user plus adjusted value
Degree;If adjustment direction is to lower, current credit degree is subtracted into adjusted value and obtains the target credit line of the target user.
In the method for adjustment of credit line provided in this embodiment, it is primarily based on the mark letter of the target user of acquisition
Breath, obtains the data information of the target user;Then the target user is determined according to the data information of the target user
State vector, and the corresponding tactful income vector of the state vector is determined according to the state vector of the target user;Again
State vector described in the tactful income vector sum is input to the first full Connection Neural Network, determines the adjustment side of credit line
To and adjusted value;The mesh is finally determined according to the current credit degree of the adjustment direction, adjusted value and the target user
The target credit line for marking user so, it is possible the multi-dimensional data information that deeply study utilizes user, and according to user
The corresponding corresponding relationship with tactful income vector of state vector, the adjustable strategies of credit line are determined, thus automatically to credit
Amount is adjusted.
Based on embodiment above-mentioned, the embodiment of the present invention provides a kind of method of adjustment of credit line again, and Fig. 3 is the present invention
The implementation process schematic diagram of the method for adjustment of embodiment credit line, as shown in figure 3, the described method comprises the following steps:
Step S301, the identification information of the target user based on acquisition obtain the data information of the target user.
Here, the data information includes at least the first dimension data information, the second dimension data information and third dimension
Data information, wherein the first dimension data information is in interesting data information, behavioral data information and consumption data information
Any one;The second dimension data information is to remove in interesting data information, behavioral data information and consumption data information
Go any one except the first dimension data information;The third dimension data information is interesting data information, behavior
It is removed in data information and consumption data information except the first dimension data information and the second dimension data information
Data information.
In this example, it is assumed that the first dimension data information is interesting data information;Second dimension data information is to disappear
Take data information;Third dimension data information is behavioral data information.
Interesting data information, consumption data information and behavioral data information are all data sequences, and each moment is one group corresponding
Data information.
The first dimension data information input to first circulation neural network is obtained the first output knot by step S302
Fruit.
Here, in the actual implementation process, the step S302 can be realized by following steps:
Step S3021 is obtained by the first dimension data information input at the first moment to the first circulation neural network
First one-dimensional output result;
The first dimension data information and (k-1) one-dimensional output result at kth moment are input to described by step S3022
First circulation neural network obtains the one-dimensional output result of kth, wherein the k=2,3 ..., N, N be first dimension data
It is total at the time of in information;
The one-dimensional output result of N is determined as the first output result by step S3023.
In the embodiment where step S3021 to step S3023, the first output result is entirely according to the first number of dimensions
It is believed that breath obtained, the data information of other dimensions is not inputted.
The second dimension data information input to second circulation neural network is obtained the second output knot by step S303
Fruit.
Here, the step S303 can be realized by following steps:
The second dimension data information at the first moment and the first one-dimensional output result are input to described by step S3031
Two Recognition with Recurrent Neural Network obtain the first two dimension output result;
Step S3032, the one-dimensional output result of the second dimension data information, kth at kth moment and (k-1) two dimension is defeated
Result is input to the second circulation neural network out, obtain kth two dimension output result, wherein the k=2,3 ..., N;
N two dimension output result is determined as the second output result by step S3033.
In the embodiment where step S3031 to step S3033, input second circulation neural network not only has second
Dimension data information further includes one-dimensional output result in the same time.
The third dimension data information is input to third Recognition with Recurrent Neural Network by step S304, obtains third output knot
Fruit.
Here, the step S304 can be realized by following steps:
Step S3041 exports the third dimension data information at the first moment, the first one-dimensional output result and the first two dimension
As a result it is input to the third Recognition with Recurrent Neural Network, obtains the first three-dimensional output result;
The third dimension data information at kth moment, the one-dimensional output result of kth, kth two dimension are exported result by step S3042
It is input to the third Recognition with Recurrent Neural Network with (k-1) three-dimensional output result, obtains kth three-dimensional output result, wherein described
K=2,3 ..., N;
N three-dimensional output result is determined as third output result by step S3043.
In the embodiment where step S3041 to step S3043, input third Recognition with Recurrent Neural Network not only has third
Dimension data information further includes one-dimensional output result and two dimension output result in the same time.
In other embodiments of the present invention, it if the data information further includes fourth dimension degree data information, also needs
Fourth dimension degree data information is input in the 4th Recognition with Recurrent Neural Network, obtains the 4th output as a result, and inputting the 4th mind
Not only there is fourth dimension degree data information through network, further includes one-dimensional output result, two dimension output result and three-dimensional in the same time
Export result.If the data information further includes the 5th dimension data information, and so on.
The first output result, the second output result and third output result are input to by step S305
Second full Connection Neural Network, obtains the state vector of the target user.
Here, first output result, second output result and third output result be all vector form output as a result, its
In include element number pre-set, and these three output results in include element number can be it is identical
, it is also possible to different.
The first output result, the second output result and third output result are input to the second full connection
Neural network carries out a comprehensive state analysis, obtains the state vector of the target user, the state vector obtained at this time
It is that the data information of each dimension that deeply has learnt target user obtains, so that the adjustment plan of final credit line
It is slightly more accurate, comprehensive.
Step S306 obtains workable credit line adjustable strategies.
Here, the workable credit line adjustable strategies are pre-set, for example include: up-regulation or downward G
A unit, wherein unit can be monetary unit, be also possible to other virtual credit units, such as bit coin, Q coin, credit
Coin etc. can raise 100 yuan (RMB), 200 dollars of up-regulation, 500 Euros of up-regulation, 1000 ports of up-regulation by taking monetary unit as an example
Member lowers 100 dollars, lowers 200 Euros, lower 500 Hongkong dollars, lower 1000 yuan etc..
Step S307 determines that first income of the state vector under the workable credit line adjustable strategies is general
Rate distributed intelligence.
Here, using the principle of convolutional neural networks, determine the state vector in the workable credit line tune
The process of the first income probability distribution information under whole strategy does convolution operation similar to convolution kernel on picture, according to the target
The state vector of user enumerates the probability distribution information of acquired income under various tune volume strategies.
Step S308, according to the first income probability distribution information determine the corresponding tactful income of the state vector to
Amount.
Here, the step S308 can be realized by following steps:
Step S3081 carries out maximum pondization operation to the first income probability distribution information, obtains the first sampling knot
Fruit;
Step S3082 carries out strategy to h sampled result and enumerates to obtain (h+1) income probability distribution information, and to institute
The operation of (h+1) income probability distribution information maximum pondization is stated, h sampled result, h=1,2 ..., M-1 are obtained, M is preset
The number of iterations;
Step S3083 determines the income of the workable credit line adjustable strategies according to (M-1) sampled result;
Step S3084 is adjusted according to the workable credit line adjustable strategies and the workable credit line
The income of strategy determines the corresponding tactful income vector of the state vector.
State vector described in the tactful income vector sum is input to the first full Connection Neural Network, really by step S309
Determine the adjustment direction and adjusted value of credit line.
Step S310, according to the determination of the current credit degree of the adjustment direction, adjusted value and the target user
The target credit line of target user.
In other embodiments, the adjustment direction of credit line and adjusted value are determined as adjustable strategies, in the step
After S309, the method also includes:
Step 41, the estimated revenue value and actual gain value of the adjustable strategies are determined;
Step 42, feedback information is determined according to the estimated revenue value and the actual gain value;
Step 43, back-propagation algorithm training is carried out to the feedback information, adjusts the first circulation neural network, institute
Second circulation neural network, the third Recognition with Recurrent Neural Network, the first full Connection Neural Network and described second is stated to connect entirely
Connect the network parameter of neural network.
In the embodiment where step 41 to step 43, the network parameter may include biasing, weight.These networks
Parameter is to carry out back-propagation algorithm instruction by the feedback information of estimated revenue value and the determination of actual gain value to adjustable strategies
It gets.It is the first circulation neural network, the second circulation neural network, the third Recognition with Recurrent Neural Network, described
The initial network parameter of first full Connection Neural Network and the second full Connection Neural Network is random number, is needed by anti-
Feedforward information is trained and further adjusts.
Back-propagation algorithm (Back propagation) is to be used to train artificial neural network (Artificial at present
Neural Network, ANN) the most frequently used and most effective algorithm, comprising the following steps:
1) training set data is input to the input layer of ANN, by hidden layer, finally reach output layer and export as a result,
This is the propagated forward process of ANN;
2) since the output result of ANN and actual result have error, then the error between estimated value and actual value is calculated, and
By the error from output layer to hidden layer backpropagation, until traveling to input layer;
3) during backpropagation, according to the value of error transfer factor various parameters;The continuous iteration above process, until receiving
It holds back.
In the method for adjustment of credit line provided in an embodiment of the present invention, it is primarily based on the mark of the target user of acquisition
Information obtains the data information of the target user;The data information includes at least the first dimension data information, the second dimension
Data information and third dimension data information;Then by the first dimension data information input to first circulation neural network,
Obtain the first output result;By the second dimension data information input to second circulation neural network, the second output knot is obtained
Fruit;The third dimension data information is input to third Recognition with Recurrent Neural Network, obtains third output result;And by described first
Output result, the second output result and third output result are input to the second full Connection Neural Network, obtain described
The state vector of target user;Workable credit line adjustable strategies are obtained again;Determine that the state vector makes described
The first income probability distribution information under credit line adjustable strategies is simultaneously true according to the first income probability distribution information
Determine the corresponding tactful income vector of the state vector;It is complete that state vector described in the tactful income vector sum is input to first
Connection Neural Network determines the adjustment direction and adjusted value of credit line;According to the adjustment direction, adjusted value and the target
The current credit degree of user determines the target credit line of the target user, is capable of the more of deep learning target user in this way
Dimension data information obtains the comprehensive state of target user, and then realizes automatic tune volume according to comprehensive state, is able to ascend entirety
Efficiency, and the network parameter for carrying out each neural network of deep learning is the estimated revenue and practical receipts according to adjustable strategies
What the feedback signal that benefit determines was adjusted, it ensure that the accuracy of deep learning.
The embodiment of the present invention first provides a kind of adjustment system of credit line.Credit line provided in an embodiment of the present invention
Adjustment system is the thought based on deep learning and intensified learning, and deep learning is used to provide powerful user's various dimensions information
The mechanism of study and expression, and intensified learning provides the target of study for deep learning.
Fig. 4 is that the composed structure of automated intelligent tune volume system of the embodiment of the present invention based on user's various dimensions information is illustrated
Figure, as shown in figure 4, the system 400 is made of three modules: state aware module 401, tactful 402 and of income iteration module
Tactful output module 403.
The working method of three modules and effect are illustrated below.
State aware module 401, the data information for the various dimensions according to user generates comprehensive state expression, at it
In his embodiment, state expression can be state vector.
User can generate the data of multiple dimensions after lend-borrow action occurs;Such as user behavior data can be generated, disappeared
Take with refund data and interesting data, and these data are all the forms of data sequence, be referred to as user behavior sequence,
Consumption and refund sequence and sequence of interest.Wherein, user behavior sequence is behavioral data of the user in social platform, is considered as
It is the stream data being made of several behaviors, reflects the Behavior preference of user;Consumption is that debt-credit row occurs for user with refund sequence
For later, the consumer behavior of generation and the behavior for the principal and interest that pay back the borrowed money;Sequence of interest is to read to apply from social category
Middle acquisition.Reflect the hobby of user.
State aware module comprehensively utilizes the data information of multiple dimensions using the Recognition with Recurrent Neural Network of multilayer.Three dimensions
Sequence distribution be input in different Recognition with Recurrent Neural Network, but three networks are not completely independent, this layer of neural network
Hiding layer unit input not only comprising this layer of neural network hidden layer of last moment output but also include upper one layer of mind
Output through network current time hidden layer.Based on this, the output of the neural network hidden layer of upper layer network may include multiple
The information of dimension.
Fig. 5 be state aware of embodiment of the present invention module diagram, as shown in figure 5, first layer Recognition with Recurrent Neural Network 501 by
Input of the interesting data as this layer of neural network, t moment hide the output of layer unitIt can indicate are as follows:
In formula (1-1), σ is sigmoid function, Wi1And Si1It is the parameter of function, xtFor the defeated of t moment interesting data
Enter,Carving copy layer hides the output of layer unit, b when for t-1iFor the bias vector of first layer.
Sigmoid function is the function of a common S type in biology, also referred to as S sigmoid growth curve.In Information Center
It, will since singly properties, the sigmoid function such as increasing and the increasing of inverse function list are often used as the threshold function table of neural network for it in
Variable mappings are between 0,1.The advantages of sigmoid, is that output area is limited, so data are not easy during transmitting
Diverging.
Input of the second layer Recognition with Recurrent Neural Network 502 by the consumption and refund data of user as this layer of neural network, t
The input of moment neural network not only includes the consumption and refund data at current time, but also includes first layer circulation nerve net
The hidden layer of the t moment of network exports, so second layer Recognition with Recurrent Neural Network t moment, which hides layer unit output, to be indicated are as follows:
In formula (1-2), xtFor the input of t moment consumption and refund data, bi2For the bias vector of the second layer,
The output of layer unit is hidden for this layer of t-1 moment,The output of layer unit is hidden for first layer Recognition with Recurrent Neural Network t moment, σ is
Sigmoid function, Wi2、Ti2And Si2It is the parameter of sigmoid function.
Input of the third layer Recognition with Recurrent Neural Network 503 by the behavior sequence data of user as network, t moment nerve net
Network input not only include t moment behavioral data, but also include the t moment of first layer Recognition with Recurrent Neural Network hidden layer it is defeated
Out and the hidden layer of second layer Recognition with Recurrent Neural Network t moment exports, so the hiding layer unit of third layer Recognition with Recurrent Neural Network
Output can indicate are as follows:
In formula (1-3), xtFor the input of t moment behavior sequence data, bi3For the bias vector of third layer,For
T-1 moment third layer Recognition with Recurrent Neural Network hides the output of layer unit,For first layer Recognition with Recurrent Neural Network t moment hidden layer list
The output of member,The output σ that layer unit is hidden for second layer Recognition with Recurrent Neural Network t moment is sigmoid function, Wi3、Ti3、Ui3
And Si3It is the parameter of sigmoid function.
First layer Recognition with Recurrent Neural Network, second layer Recognition with Recurrent Neural Network and third layer Recognition with Recurrent Neural Network the last one when
The state expression for generating synthesis is done in being input to for quarter in the neural network 504 connected entirely;Sigmoid letter in above-mentioned formula
Number parameter can be obtained by backpropagation algorithm training.
Tactful income iteration module 402, for according to current state, iteration calculates the income of the acquirement under the state
Value is until convergence.Enumeration strategy can generate the financial value generated under different decisions, a decision pair according to the state of user
Answer the probability distribution and acquired income of a dbjective state.
Fig. 6 is the course of work schematic diagram of strategy of embodiment of the present invention income iteration module, as shown in fig. 6, strategy is enumerated
Process do convolution operation on picture 602 similar to convolution kernel 601, strategy, which is enumerated, to be enumerated according to the state of active user various
Tune volume strategy policyiThe probability distribution reward of lower acquired incomei, wherein i=0,1 ..., n;For benefit distribution
rewardiIn include 12 states, M0,M1,...,M11, M0User is there is no overdue under the tune volume strategy for representative, M11It represents
User refunds overdue 11 phases of generation under the tune volume strategy, and overdue will stop as user's granting loan of more than phase refunding more than 11 occurs
Money;MiIncome under state is usedIt indicates;Sampling operation is done using the method in maximum pond, strategy is enumerated and most
Great Chiization operation completes the process of an iteration;Multiple income is realized by each tactful enumeration layer of heap overlay and sample level
Iterative calculation generates optimal value;This process can be indicated with formula (1-4):
rewardsum=max { rewardi+rewardnext} (1-4);
In formula (1-4), rewardsumTo adjust volume strategy policyiOptimal financial value, max () be maximizing letter
Number, rewardnextThe financial value generated for last iteration;
The number of plies that strategy is enumerated and sampled is network parameter, can test adjustment in actual application, to obtain most
Excellent effect.
Tactful output module 403, for determining optimal tune volume strategy according to tactful income vector.
For tactful income iteration module after the sampling of multilayer convolution sum, it is defeated that the tactful income vector of output is input to strategy
Out in module, Fig. 7 is the course of work schematic diagram of strategy of embodiment of the present invention output module, as shown in fig. 7, tactful output module
It is made of the full Connection Neural Network 701 of multilayer;The parameter of its network can be obtained by back-propagation algorithm training.In training
In the process, Attention mechanism is introduced, the status information 702 that state aware module is exported is used as additional information, so that network
The status information of user can be efficiently used.
It should be noted that the income for adjusting volume strategy to generate that tactful income iteration module generates, usually practical with strategy
There are errors for the actual gain generated after, the error can be known as feedback signal here.Tactful output module is outside
The feedback signal L of portion's environment adjusts the parameter in all neural networks used in three modules by back-propagation algorithm;
In practical applications, feedback signal can be determined according to formula (1-5), which indicates the income of the strategy prediction of application
With square of actual gain difference.
L=(rewardtraget-rewardprediction)2(1-5);
In formula (1-5), rewardpredictionFor the prediction income for adjusting volume strategy, rewardtragetTo adjust volume strategy
Actual gain.
Feedback signal is passed into tactful income iteration module and state aware module simultaneously, the two modules utilize feedback
Signal learns the parameter of this module by back-propagation algorithm.
In embodiments of the present invention, the thought based on deep learning and intensified learning is applied to the adjustment system of credit line
In system, wherein the block of state in the adjustment system of credit line generates synthesis according to the data information of the various dimensions of user
State vector, tactful income iteration module determine tactful income vector, last strategy output mould further according to the state vector of user
Root tuber determines optimal tune volume strategy according to tactful income vector, in this way, the adjustment system energy of credit line provided in this embodiment
The multilayer circulation neural network for enough utilizing various dimensions user information realizes User Status perception, and then according to the synthesis shape of user
State automatically generates optimal tune volume strategy and can aid in the whole efficiency of promotion and provide differentiation financial service.
The embodiment of the present invention provides a kind of adjustment device of credit line, and Fig. 8 is line of credit provided in an embodiment of the present invention
The composed structure schematic diagram of the adjustment device of degree, as shown in figure 8, described device 800 includes: that the first acquisition module 801, first is true
Cover half block 802, the second determining module 803 and third determining module 804, in which:
The first acquisition module 801 obtains the target user for the identification information of the target user based on acquisition
Data information;
First determining module 802, for determining the target user's according to the data information of the target user
State vector determines the corresponding tactful income vector of the state vector according to the state vector of the target user;
Second determining module 803 is complete for state vector described in the tactful income vector sum to be input to first
Connection Neural Network determines the adjustment direction and adjusted value of credit line;
The third determining module 804, for according to the adjustment direction, adjusted value and the current letter of the target user
The target credit line of the target user is determined with amount.
In other embodiments, the data information includes at least: the first dimension data information, the second dimension data information
With third dimension data information, accordingly, first determining module 802 includes:
First input unit, for by the first dimension data information input to first circulation neural network, obtaining the
One output result;
Second input unit, for by the second dimension data information input to second circulation neural network, obtaining the
Two output results;
Third input unit obtains for the third dimension data information to be input to third Recognition with Recurrent Neural Network
Three output results;
4th input unit, for tying the first output result, the second output result and third output
Fruit is input to the second full Connection Neural Network, obtains the state vector of the target user.
In other embodiments, first input unit is also used to:
By the first dimension data information input at the first moment to the first circulation neural network, it is one-dimensional defeated to obtain first
Result out;
The first dimension data information and (k-1) one-dimensional output result at kth moment are input to the first circulation mind
Through network, obtain the one-dimensional output result of kth, wherein the k=2,3 ..., N, N be the first dimension data information in when
Carve sum;
The one-dimensional output result of N is determined as the first output result.
In other embodiments, second input unit, be also used to the second dimension data information at the first moment and
First one-dimensional output result is input to the second circulation neural network, obtains the first two dimension output result;By the of the kth moment
The one-dimensional output result of two-dimensions data information, kth and (k-1) two dimension output result are input to the second circulation nerve net
Network obtains kth two dimension output result, wherein the k=2,3 ..., N;N two dimension output result is determined as the second output knot
Fruit.
In other embodiments, the third input unit is also used to the third dimension data information at the first moment,
One one-dimensional output result and the first two dimension output result are input to the third Recognition with Recurrent Neural Network, obtain the first three-dimensional output knot
Fruit;The third dimension data information at kth moment, the one-dimensional output result of kth, kth two dimension output result and (k-1) three-dimensional is defeated
Result is input to the third Recognition with Recurrent Neural Network out, obtains kth three-dimensional output result, wherein the k=2,3 ..., N;It will
N three-dimensional output result is determined as third output result.
In other embodiments, first determining module 802 further include:
First acquisition unit, for obtaining workable credit line adjustable strategies;
First determination unit, for determining the state vector under the workable credit line adjustable strategies
One income probability distribution information;
Second determination unit, for determining the corresponding plan of the state vector according to the first income probability distribution information
Slightly income vector.
In other embodiments, second determination unit is also used to: being carried out to the first income probability distribution information
Maximum pondization operation, obtains the first sampled result;Strategy is carried out to h sampled result to enumerate to obtain (h+1) income probability point
Cloth information, and (h+1) income probability distribution information maximum pondization is operated, obtain h sampled result, h=1,2 ...,
M-1, M are preset the number of iterations;The receipts of the workable credit line adjustable strategies are determined according to (M-1) sampled result
Benefit;Institute is determined according to the income of the workable credit line adjustable strategies and the workable credit line adjustable strategies
State the corresponding tactful income vector of state vector.
The description of apparatus above embodiment, be with the description of above method embodiment it is similar, have same embodiment of the method
Similar beneficial effect.For undisclosed technical detail in apparatus of the present invention embodiment, embodiment of the present invention method is please referred to
Description and understand.
The embodiment of the present invention provides a kind of adjustment equipment of credit line, and Fig. 9 is the tune of credit line of the embodiment of the present invention
The composed structure schematic diagram of finishing equipment, as shown in figure 9, the equipment 900 at least one processor 901, at least one communication are total
Line 902, user interface 903, at least one external communication interface 904 and memory 905.Wherein, communication bus 902 is configured to reality
Connection communication between these existing components.Wherein, user interface 903 may include display screen, and external communication interface 904 can wrap
Include the wireline interface and wireless interface of standard.The wherein processor 901, is configured that
The identification information of target user based on acquisition obtains the data information of the target user;
The state vector that the target user is determined according to the data information of the target user, according to the target user
State vector determine the corresponding tactful income vector of the state vector;
State vector described in the tactful income vector sum is input to the first full Connection Neural Network, determines credit line
Adjustment direction and adjusted value;
Determine the target user's according to the current credit degree of the adjustment direction, adjusted value and the target user
Target credit line.
In other embodiments, the data information includes at least: the first dimension data information, the second dimension data information
With third dimension data information, accordingly, the state vector that the target user is determined according to the data information is wrapped
It includes:
By the first dimension data information input to first circulation neural network, the first output result is obtained;
By the second dimension data information input to second circulation neural network, the second output result is obtained;
The third dimension data information is input to third Recognition with Recurrent Neural Network, obtains third output result;
The first output result, the second output result and third output result are input to the second full connection
Neural network obtains the state vector of the target user.
In other embodiments, described that the first dimension data information input to first circulation neural network is obtained
One output result, comprising:
By the first dimension data information input at the first moment to the first circulation neural network, it is one-dimensional defeated to obtain first
Result out;
The first dimension data information and (k-1) one-dimensional output result at kth moment are input to the first circulation mind
Through network, obtain the one-dimensional output result of kth, wherein the k=2,3 ..., N, N be the first dimension data information in when
Carve sum;
The one-dimensional output result of N is determined as the first output result.
In other embodiments, described that the second dimension data information input to second circulation neural network is obtained
Two output results, comprising:
The second dimension data information at the first moment and the first one-dimensional output result are input to the second circulation nerve
Network obtains the first two dimension output result;
By the one-dimensional output result of the second dimension data information, kth at kth moment and the output result input of (k-1) two dimension
To the second circulation neural network, kth two dimension output result is obtained, wherein the k=2,3 ..., N;
N two dimension output result is determined as the second output result.
In other embodiments, described the third dimension data information is input to third Recognition with Recurrent Neural Network to obtain
Three output results, comprising:
The third dimension data information at the first moment, the first one-dimensional output result and the first two dimension output result are input to
The third Recognition with Recurrent Neural Network obtains the first three-dimensional output result;
The third dimension data information at kth moment, the one-dimensional output result of kth, kth two dimension are exported into result and (k-1)
Three-dimensional output result is input to the third Recognition with Recurrent Neural Network, obtains kth three-dimensional output result, wherein the k=2,
3,…,N;
N three-dimensional output result is determined as third output result.
In other embodiments, it is described according to the state vector determine the corresponding tactful income of the state vector to
Amount, comprising:
Obtain workable credit line adjustable strategies;
Determine first income probability distribution letter of the state vector under the workable credit line adjustable strategies
Breath;
The corresponding tactful income vector of the state vector is determined according to the first income probability distribution information.
In other embodiments, described to determine that the state vector is corresponding according to the first income probability distribution information
Tactful income vector, comprising:
Maximum pondization operation is carried out to the first income probability distribution information, obtains the first sampled result;
It carries out strategy to h sampled result to enumerate to obtain (h+1) income probability distribution information, and to (h+1)
The operation of income probability distribution information maximum pondization, obtains h sampled result, h=1,2 ..., M-1, M is preset the number of iterations;
The income of the workable credit line adjustable strategies is determined according to (M-1) sampled result;
According to the income of the workable credit line adjustable strategies and the workable credit line adjustable strategies
Determine the corresponding tactful income vector of the state vector.
In other embodiments, the processor 901, is configured that
The adjustment direction and adjusted value are determined as adjustable strategies;
Determine the estimated revenue value and actual gain value of the adjustable strategies;
Feedback information is determined according to the estimated revenue value and the actual gain value;
Back-propagation algorithm training is carried out to the feedback information, adjusts the first circulation neural network, described second
Recognition with Recurrent Neural Network, the third Recognition with Recurrent Neural Network, the first full Connection Neural Network and the second full connection nerve
The network parameter of network.
In the embodiment of the present invention, if realizing the method for adjustment of above-mentioned credit line in the form of software function module,
And when sold or used as an independent product, it also can store in a computer readable storage medium.Based in this way
Understanding, substantially the part that contributes to existing technology can be produced the technical solution of the embodiment of the present invention in other words with software
The form of product embodies, which is stored in a storage medium, including some instructions are used so that one
Platform computer equipment (can be personal computer, server or network equipment etc.) executes described in each embodiment of the present invention
The all or part of method.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (ROM, Read Only
Memory), the various media that can store program code such as magnetic or disk.In this way, the embodiment of the present invention is not limited to appoint
What specific hardware and software combines.
The embodiment of the present invention provides a kind of computer storage medium, and being stored with computer in the computer storage medium can
It executes instruction, which is used to execute the method for adjustment of credit line provided in an embodiment of the present invention.
It should be understood that " one embodiment " or " embodiment " that specification is mentioned in the whole text mean it is related with embodiment
A particular feature, structure, or characteristic is included at least one embodiment of the present invention.Therefore, occur everywhere in the whole instruction
" in one embodiment " or " in one embodiment " not necessarily refer to identical embodiment.In addition, these specific features, knot
Structure or characteristic can combine in any suitable manner in one or more embodiments.It should be understood that in various implementations of the invention
In example, magnitude of the sequence numbers of the above procedures are not meant that the order of the execution order, the execution sequence Ying Yiqi function of each process
It can determine that the implementation process of the embodiments of the invention shall not be constituted with any limitation with internal logic.The embodiments of the present invention
Serial number is for illustration only, does not represent the advantages or disadvantages of the embodiments.
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that the process, method, article or the device that include a series of elements not only include those elements, and
And further include other elements that are not explicitly listed, or further include for this process, method, article or device institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to do
There is also other identical elements in the process, method of element, article or device.
In several embodiments provided herein, it should be understood that disclosed device and method can pass through it
Its mode is realized.Apparatus embodiments described above are merely indicative, for example, the division of the unit, only
A kind of logical function partition, there may be another division manner in actual implementation, such as: multiple units or components can combine, or
It is desirably integrated into another system, or some features can be ignored or not executed.In addition, shown or discussed each composition portion
Mutual coupling or direct-coupling or communication connection is divided to can be through some interfaces, the INDIRECT COUPLING of equipment or unit
Or communication connection, it can be electrical, mechanical or other forms.
Above-mentioned unit as illustrated by the separation member, which can be or may not be, to be physically separated, aobvious as unit
The component shown can be or may not be physical unit;Both it can be located in one place, and may be distributed over multiple network lists
In member;Some or all of units can be selected to achieve the purpose of the solution of this embodiment according to the actual needs.
In addition, each functional unit in various embodiments of the present invention can be fully integrated in one processing unit, it can also
To be each unit individually as a unit, can also be integrated in one unit with two or more units;It is above-mentioned
Integrated unit both can take the form of hardware realization, can also realize in the form of hardware adds SFU software functional unit.
Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above method embodiment can pass through
The relevant hardware of program instruction is completed, and program above-mentioned can store in computer-readable storage medium, which exists
When execution, step including the steps of the foregoing method embodiments is executed;And storage medium above-mentioned includes: movable storage device, read-only deposits
The various media that can store program code such as reservoir (Read Only Memory, ROM), magnetic or disk.
If alternatively, the above-mentioned integrated unit of the present invention is realized in the form of software function module and as independent product
When selling or using, it also can store in a computer readable storage medium.Based on this understanding, the present invention is implemented
Substantially the part that contributes to existing technology can be embodied in the form of software products the technical solution of example in other words,
The computer software product is stored in a storage medium, including some instructions are used so that computer equipment (can be with
It is personal computer, server or network equipment etc.) execute all or part of each embodiment the method for the present invention.
And storage medium above-mentioned includes: various Jie that can store program code such as movable storage device, ROM, magnetic or disk
Matter.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (15)
1. a kind of method for adjusting credit line, which is characterized in that the described method includes:
The identification information of target user based on acquisition obtains the data information of the target user;
The state vector that the target user is determined according to the data information of the target user, according to the shape of the target user
State vector determines the corresponding tactful income vector of the state vector;
State vector described in the tactful income vector sum is input to the first full Connection Neural Network, determines the tune of credit line
Perfect square to and adjusted value;
The target of the target user is determined according to the current credit degree of the adjustment direction, adjusted value and the target user
Credit line.
2. method according to claim 1, which is characterized in that the data information includes at least: the first dimension data
Information, the second dimension data information and third dimension data information, it is accordingly, described that the mesh is determined according to the data information
Mark the state vector of user, comprising:
By the first dimension data information input to first circulation neural network, the first output result is obtained;
By the second dimension data information input to second circulation neural network, the second output result is obtained;
The third dimension data information is input to third Recognition with Recurrent Neural Network, obtains third output result;
The first output result, the second output result and third output result are input to the second full connection nerve
Network obtains the state vector of the target user.
3. method according to claim 2, which is characterized in that described by the first dimension data information input to
One Recognition with Recurrent Neural Network obtains the first output result, comprising:
By the first dimension data information input at the first moment to the first circulation neural network, the first one-dimensional output knot is obtained
Fruit;
The first dimension data information and (k-1) one-dimensional output result at kth moment are input to the first circulation nerve net
Network obtains the one-dimensional output result of kth, wherein the k=2,3 ..., N, N be the first dimension data information at the time of it is total
Number;
The one-dimensional output result of N is determined as the first output result.
4. method according to claim 2, which is characterized in that described by the second dimension data information input to
Two Recognition with Recurrent Neural Network obtain the second output result, comprising:
The second dimension data information at the first moment and the first one-dimensional output result are input to the second circulation neural network,
Obtain the first two dimension output result;
The one-dimensional output result of second dimension data information, kth at kth moment and (k-1) two dimension output result are input to institute
Second circulation neural network is stated, obtains kth two dimension output result, wherein the k=2,3 ..., N;
N two dimension output result is determined as the second output result.
5. method according to claim 2, which is characterized in that described that the third dimension data information is input to
Three Recognition with Recurrent Neural Network obtain third output result, comprising:
The third dimension data information at the first moment, the first one-dimensional output result and the first two dimension output result are input to described
Third Recognition with Recurrent Neural Network obtains the first three-dimensional output result;
By the third dimension data information at kth moment, the one-dimensional output result of kth, kth two dimension output result and (k-1) three-dimensional
Output result be input to the third Recognition with Recurrent Neural Network, obtain kth three-dimensional output result, wherein the k=2,3 ..., N;
N three-dimensional output result is determined as third output result.
6. method according to claim 1, which is characterized in that it is described according to the state vector determine the state to
Measure corresponding tactful income vector, comprising:
Obtain workable credit line adjustable strategies;
Determine first income probability distribution information of the state vector under the workable credit line adjustable strategies;
The corresponding tactful income vector of the state vector is determined according to the first income probability distribution information.
7. method according to claim 6, which is characterized in that described true according to the first income probability distribution information
Determine the corresponding tactful income vector of the state vector, comprising:
Maximum pondization operation is carried out to the first income probability distribution information, obtains the first sampled result;
It carries out strategy to h sampled result to enumerate to obtain (h+1) income probability distribution information, and to (h+1) income
The operation of probability distribution information maximum pondization, obtains h sampled result, wherein h=1,2 ..., (M-1), M is preset iteration time
Number;
The income of the workable credit line adjustable strategies is determined according to (M-1) sampled result;
It is determined according to the income of the workable credit line adjustable strategies and the workable credit line adjustable strategies
The corresponding tactful income vector of the state vector.
8. method according to claim 1, which is characterized in that the method also includes:
The adjustment direction and adjusted value are determined as adjustable strategies;
Determine the estimated revenue value and actual gain value of the adjustable strategies;
Feedback information is determined according to the estimated revenue value and the actual gain value;
Back-propagation algorithm training is carried out to the feedback information, adjusts the first circulation neural network, the second circulation
Neural network, the third Recognition with Recurrent Neural Network, the first full Connection Neural Network and the second full Connection Neural Network
Network parameter.
9. a kind of adjustment device of credit line, which is characterized in that described device includes: the first acquisition module, the first determining mould
Block, the second determining module and third determining module, in which:
The first acquisition module obtains the data of the target user for the identification information of the target user based on acquisition
Information;
First determining module, for determined according to the data information of the target user state of the target user to
Amount determines the corresponding tactful income vector of the state vector according to the state vector of the target user;
Second determining module, for state vector described in the tactful income vector sum to be input to the first full connection nerve
Network determines the adjustment direction and adjusted value of credit line;
The third determining module, for the current credit degree according to the adjustment direction, adjusted value and the target user
Determine the target credit line of the target user.
10. device according to claim 9, which is characterized in that the data information includes at least: the first dimension data
Information, the second dimension data information and third dimension data information, accordingly, first determining module includes:
First input unit, for first circulation neural network, it is defeated to be obtained first for the first dimension data information input
Result out;
Second input unit, for second circulation neural network, it is defeated to be obtained second for the second dimension data information input
Result out;
It is defeated to obtain third for the third dimension data information to be input to third Recognition with Recurrent Neural Network for third input unit
Result out;
4th input unit, for the first output result, the second output result and third output result is defeated
Enter to the second full Connection Neural Network, obtains the state vector of the target user.
11. device according to claim 10, which is characterized in that first determining module further include:
First acquisition unit, for obtaining workable credit line adjustable strategies;
First determination unit, for determining first receipts of the state vector under the workable credit line adjustable strategies
Beneficial probability distribution information;
Second determination unit, for determining that the corresponding strategy of the state vector is received according to the first income probability distribution information
Beneficial vector.
12. a kind of adjustment equipment of credit line, which is characterized in that the equipment includes at least: memory, communication bus and place
Manage device, in which:
The memory, for storing the adjustment programme of credit line;
The communication bus, for realizing the connection communication between processor and memory;
The processor, for executing the adjustment programme of the credit line stored in memory, to perform the steps of
The identification information of target user based on acquisition obtains the data information of the target user;
The state vector that the target user is determined according to the data information of the target user, according to the shape of the target user
State vector determines the corresponding tactful income vector of the state vector;
State vector described in the tactful income vector sum is input to the first full Connection Neural Network, determines the tune of credit line
Perfect square to and adjusted value;
The target of the target user is determined according to the current credit degree of the adjustment direction, adjusted value and the target user
Credit line.
13. equipment according to claim 12, which is characterized in that the data information includes at least: the first number of dimensions
It is believed that breath, the second dimension data information and third dimension data information, accordingly, described according to data information determination
The state vector of target user, comprising:
By the first dimension data information input to first circulation neural network, the first output result is obtained;
By the second dimension data information input to second circulation neural network, the second output result is obtained;
The third dimension data information is input to third Recognition with Recurrent Neural Network, obtains third output result;
The first output result, the second output result and third output result are input to the second full connection nerve
Network obtains the state vector of the target user.
14. equipment according to claim 12, which is characterized in that described to determine the state according to the state vector
The corresponding tactful income vector of vector, comprising:
Obtain workable credit line adjustable strategies;
Determine first income probability distribution information of the state vector under the workable credit line adjustable strategies;
The corresponding tactful income vector of the state vector is determined according to the first income probability distribution information.
15. a kind of computer readable storage medium, which is characterized in that be stored with line of credit on the computer readable storage medium
It realizes when the adjustment programme of the adjustment programme of degree, the credit line is executed by processor such as any one of claims 1 to 8 institute
The step of method of adjustment for the credit line stated.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710952035.9A CN109670927A (en) | 2017-10-13 | 2017-10-13 | The method of adjustment and its device of credit line, equipment, storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710952035.9A CN109670927A (en) | 2017-10-13 | 2017-10-13 | The method of adjustment and its device of credit line, equipment, storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109670927A true CN109670927A (en) | 2019-04-23 |
Family
ID=66138859
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710952035.9A Pending CN109670927A (en) | 2017-10-13 | 2017-10-13 | The method of adjustment and its device of credit line, equipment, storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109670927A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110348726A (en) * | 2019-07-02 | 2019-10-18 | 北京淇瑀信息科技有限公司 | A kind of user's amount method of adjustment, device and electronic equipment based on social networks network |
CN110807527A (en) * | 2019-09-30 | 2020-02-18 | 北京淇瑀信息科技有限公司 | Line adjusting method and device based on guest group screening and electronic equipment |
CN111353872A (en) * | 2019-12-20 | 2020-06-30 | 上海淇玥信息技术有限公司 | Credit granting processing method and device based on financial performance value and electronic equipment |
CN112288436A (en) * | 2020-10-27 | 2021-01-29 | 上海淇馥信息技术有限公司 | Triggered resource quota adjusting method, device and system |
TWI768512B (en) * | 2019-12-26 | 2022-06-21 | 日商樂天集團股份有限公司 | Credit calculation system, credit calculation method and program product |
-
2017
- 2017-10-13 CN CN201710952035.9A patent/CN109670927A/en active Pending
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110348726A (en) * | 2019-07-02 | 2019-10-18 | 北京淇瑀信息科技有限公司 | A kind of user's amount method of adjustment, device and electronic equipment based on social networks network |
CN110807527A (en) * | 2019-09-30 | 2020-02-18 | 北京淇瑀信息科技有限公司 | Line adjusting method and device based on guest group screening and electronic equipment |
CN110807527B (en) * | 2019-09-30 | 2023-11-14 | 北京淇瑀信息科技有限公司 | Credit adjustment method and device based on guest group screening and electronic equipment |
CN111353872A (en) * | 2019-12-20 | 2020-06-30 | 上海淇玥信息技术有限公司 | Credit granting processing method and device based on financial performance value and electronic equipment |
TWI768512B (en) * | 2019-12-26 | 2022-06-21 | 日商樂天集團股份有限公司 | Credit calculation system, credit calculation method and program product |
CN112288436A (en) * | 2020-10-27 | 2021-01-29 | 上海淇馥信息技术有限公司 | Triggered resource quota adjusting method, device and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109670927A (en) | The method of adjustment and its device of credit line, equipment, storage medium | |
Chandra | Competition and collaboration in cooperative coevolution of Elman recurrent neural networks for time-series prediction | |
Chen et al. | Mining the customer credit using hybrid support vector machine technique | |
Hsu | A hybrid procedure with feature selection for resolving stock/futures price forecasting problems | |
CN109300029A (en) | Borrow or lend money fraud detection model training method, debt-credit fraud detection method and device | |
CN109902222A (en) | Recommendation method and device | |
CN109766454A (en) | A kind of investor's classification method, device, equipment and medium | |
Li | Research on WNN modeling for gold price forecasting based on improved artificial bee colony algorithm | |
CN109389494A (en) | Borrow or lend money fraud detection model training method, debt-credit fraud detection method and device | |
CN110276679A (en) | A kind of network individual credit fraud detection method towards deep learning | |
Kumar | Machine Learning Quick Reference: Quick and essential machine learning hacks for training smart data models | |
CN110599336A (en) | Financial product purchase prediction method and system | |
JP7059458B2 (en) | Generating hostile neuropil-based classification systems and methods | |
CN109034228A (en) | A kind of image classification method based on difference privacy and level relevance propagation | |
Chen et al. | Research on credit card default prediction based on k-means SMOTE and BP neural network | |
CN107481135A (en) | A kind of personal credit evaluation method and system based on BP neural network | |
Li et al. | Stock trading strategies based on deep reinforcement learning | |
Zhang et al. | The empirical analysis of bitcoin price prediction based on deep learning integration method | |
Negi et al. | Cryptocurrency Price Analysis using Deep Learning | |
CN110110628A (en) | A kind of detection method and detection device of frequency synthesizer deterioration | |
Lv et al. | Detecting fraudulent bank account based on convolutional neural network with heterogeneous data | |
Booth | Automated algorithmic trading: Machine learning and agent-based modelling in complex adaptive financial markets | |
Fajr et al. | An improved particle swarm optimization algorithm for global multidimensional optimization | |
Szwabe et al. | Decision trees as interpretable bank credit scoring models | |
CN108876589A (en) | Methods of risk assessment, device, equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |