CN108647714A - Acquisition methods, terminal device and the medium of negative label weight - Google Patents
Acquisition methods, terminal device and the medium of negative label weight Download PDFInfo
- Publication number
- CN108647714A CN108647714A CN201810436265.4A CN201810436265A CN108647714A CN 108647714 A CN108647714 A CN 108647714A CN 201810436265 A CN201810436265 A CN 201810436265A CN 108647714 A CN108647714 A CN 108647714A
- Authority
- CN
- China
- Prior art keywords
- negative
- node
- sample data
- client
- label
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
-
- 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
- G06Q40/02—Banking, e.g. interest calculation or account maintenance
Abstract
The present invention is suitable for Internet technical field, provides a kind of acquisition methods, terminal device and the medium of negative label weight, this method includes：If detecting, customer relationship network topology updates, and obtains the corresponding client's sample data of plurality of node institute；Based on client's sample data, builds and train neural network model；The Figure Characteristics of each pending client in business approval system are inputted into neural network model, to export weighted value of each pending client respectively in each negative tag types；According to the average value of the identical each weighted value of negative tag types, the label weight of the negative tag types is calculated.The present invention realizes the automation update of label weight, reduces operation complexity；Meanwhile also avoiding the single problem of label weight analysis dimension and occurring, improve the calculating accuracy of label weight；It ensure that business approval system can get the negative label weight of real-time update.
Description
Technical field
The invention belongs to Internet technical field more particularly to a kind of acquisition methods, the terminal devices of negative label weight
And computer readable storage medium.
Background technology
Can be that client stamps different types of label currently, being analyzed by personal characteristics' attribute to client.It is existing
Have in technology, be normally based on the mode of business rule to determine client's label, that is, as long as detecting that personal characteristics attribute meets
Preset business rule will be that the client stamps a label corresponding to business rule.For example, if client meets " reality
The time refund more than the predetermined refund time " this business rule, then stamp overdue label for the client；If client, which meets, " to be had
This business rule of Claims Resolution record ", then stamp Claims Resolution label etc. for the client.Since above-mentioned business rule is provided to determine visitor
Whether family has negative report record, and therefore, above-mentioned label is negative label.For every negative label of one kind, according to such
Negatively client's sum of label ratio shared in all clients, can calculate the label weight of the negative label, to indicate
The influence degree size that overdue refund event occurs in such negative label.Hereafter, which will the person's of being managed input
In approval system of providing a loan, using as loan review process in a reference factor.
However, above-mentioned label weight can only be determined according to each client characteristic attribute of itself, association is had ignored
Influencing each other between client thus reduces the accuracy of negative label weight and the property of can refer to.
Invention content
In view of this, an embodiment of the present invention provides a kind of acquisition methods, terminal device and the calculating of negative label weight
Machine readable storage medium storing program for executing, to solve the accuracy of negative label weight in the prior art and the property of can refer to is more low asks
Topic.
The first aspect of the embodiment of the present invention provides a kind of acquisition methods of negative label weight, including：
If detecting, customer relationship network topology updates, and obtains the corresponding client's sample of plurality of node institute
Notebook data, client's sample data include Figure Characteristics, negative tag types and each negative tag types power
Weight values；
Based on client's sample data, builds and train neural network model；
The Figure Characteristics of each pending client in business approval system are inputted into the neural network model, with defeated
Go out weighted value of each pending client respectively in each negative tag types；
According to the identical each weighted value of the negative tag types, the label power of the negative tag types is calculated
Weight.
The second aspect of the embodiment of the present invention provides a kind of terminal device, including memory and processor, described to deposit
The computer program that can be run on the processor is stored in reservoir, the processor executes real when the computer program
Existing following steps：
If detecting, customer relationship network topology updates, and obtains the corresponding client's sample of plurality of node institute
Notebook data, client's sample data include Figure Characteristics, negative tag types and each negative tag types power
Weight values；
Based on client's sample data, builds and train neural network model；
The Figure Characteristics of each pending client in business approval system are inputted into the neural network model, with defeated
Go out weighted value of each pending client respectively in each negative tag types；
According to the identical each weighted value of the negative tag types, the label power of the negative tag types is calculated
Weight.
The third aspect of the embodiment of the present invention provides a kind of computer readable storage medium, the computer-readable storage
Media storage has computer program, the computer program to realize following steps when being executed by processor：
If detecting, customer relationship network topology updates, and obtains the corresponding client's sample of plurality of node institute
Notebook data, client's sample data include Figure Characteristics, negative tag types and each negative tag types power
Weight values；
Based on client's sample data, builds and train neural network model；
The Figure Characteristics of each pending client in business approval system are inputted into the neural network model, with defeated
Go out weighted value of each pending client respectively in each negative tag types；
According to the identical each weighted value of the negative tag types, the label power of the negative tag types is calculated
Weight.
In the embodiment of the present invention, the neural network model of the weighted value by training for exporting each negative tag types,
And entire neural network model is applied to business approval system, it ensure that user without after counting all kinds of negative label weights
Manually inputting again realizes the automation update of label weight in system, reduces operation complexity.Due to neural network mould
The training sample of type is client's sample data in customer relationship network, and has higher pass between every client's sample data
Connection property, therefore, the weighted value finally exported according to neural network model calculate the label weight of each negative tag types, are
The obtained result of calculation that influences each other for having considered other negative label clients avoids label weight analysis dimension list
One the problem of, occurs, and improves the calculating accuracy of label weight.When being updated due to customer relationship network topology, nerve net
Network model also can dynamically update, therefore ensure that business approval system can get the label weight of real-time update, so that
Approving person based on the label weight come to carry out business approval when, the higher approval results of accuracy can be obtained.
Description of the drawings
It to describe the technical solutions in the embodiments of the present invention more clearly, below will be to embodiment or description of the prior art
Needed in attached drawing be briefly described, it should be apparent that, the accompanying drawings in the following description be only the present invention some
Embodiment for those of ordinary skill in the art without having to pay creative labor, can also be according to these
Attached drawing obtains other attached drawings.
Fig. 1 is the implementation flow chart of the acquisition methods of negative label weight provided in an embodiment of the present invention；
Fig. 2 is the specific implementation flow chart of the acquisition methods S101 of negative label weight provided in an embodiment of the present invention；
Fig. 3 is the specific implementation flow chart of the acquisition methods S102 of negative label weight provided in an embodiment of the present invention；
Fig. 4 is the specific implementation flow chart of the acquisition methods S1024 of negative label weight provided in an embodiment of the present invention；
Fig. 5 is the structure diagram of the acquisition device of negative label weight provided in an embodiment of the present invention；
Fig. 6 is the schematic diagram of terminal device provided in an embodiment of the present invention.
Specific implementation mode
In being described below, for illustration and not for limitation, it is proposed that such as tool of particular system structure, technology etc
Body details, to understand thoroughly the embodiment of the present invention.However, it will be clear to one skilled in the art that there is no these specific
The present invention can also be realized in the other embodiments of details.In other situations, it omits to well-known system, device, electricity
The detailed description of road and method, in case unnecessary details interferes description of the invention.
In order to illustrate technical solutions according to the invention, illustrated below by specific embodiment.
Fig. 1 shows the implementation process of the acquisition methods of negative label weight provided in an embodiment of the present invention, this method stream
Journey includes step S101 to S104.The specific implementation principle of each step is as follows：
S101：If detecting, customer relationship network topology updates, and it is corresponding to obtain plurality of node institute
Client's sample data, client's sample data include Figure Characteristics, negative tag types and each negative tag class
The weighted value of type.
According to the customer data for collecting obtained multiple users in advance, customer relationship network topology is built.Customer relationship net
Include multiple nodes in network topology, each node corresponds to the customer data of a user.Wherein, customer data includes that portrait is special
The weighted value of sign, negative tag types and each negative tag types.Figure Characteristics include but not limited to unique mark of user
Know the label value of the labels such as symbol, work unit, age, address, supplier, credits side, debt side and business scope.Above-mentioned visitor
User data can be collected from the personal information of history loan application user, also can be from the history for other operation systems docked in advance
It is collected in order information.
If the Figure Characteristics of user meet preset negative label rule, it is right to stamp the negative label rule for user
The negative label of one kind answered, such negative label is the negative tag types of user.For example, overdue label is with Claims Resolution label
Different negative tag types.The influence degree size that the negative label of certain class generates a certain user overdue refund event is the use
Weighted value of the family in the negative tag types.If it is regular that user is unsatisfactory for any negative label, negative in customer data
The attribute value of tag types is null value.
In the embodiment of the present invention, in each node for being included from customer relationship network topology, multiple nodes are extracted, and read
The customer data corresponding to each node of extraction is taken, then the customer data read is client's sample data at current time.
In the embodiment of the present invention, if detecting newly-increased customer data, customer relationship network topology is updated.At this point, needing
It repeats in each node for being included from customer relationship network topology, extracts multiple nodes, and read extraction
The step of customer data corresponding to each node, to ensure the real-time update of client's sample data.
As an embodiment of the present invention, Fig. 2 shows the acquisitions of negative label weight provided in an embodiment of the present invention
The specific implementation flow of method S101, details are as follows：
S1011：Updated customer relationship network topology is divided into multiple sub-network topologys.
In the embodiment of the present invention, since customer relationship network topology generally comprised interstitial content a little, will be newest
When inscribe obtained customer relationship network topology random division into multiple sub-networks topology, and make wherein each sub-network topology
Including interstitial content be less than predetermined threshold value.
S1012：In each sub-network topology, each node for carrying negative label is found out, and calculate lookup
First accounting value of each node gone out in the sub-network topology.
In each sub-network topology that segmentation obtains, the node total number that sub-network topology is included is counted.It obtains successively
The customer data corresponding to each node in the sub-network topology is taken, and judges the attribute of negative tag types in its customer data
Whether value is null value.
If in the customer data corresponding to node, the attribute value of negative tag types is non-null value, then carries out the node
It chooses, and the statistical value of negative interstitial content is made to add one.
If the equal poll of each node in sub-network topology finishes, obtains the final statistical value of negative interstitial content and be somebody's turn to do
The node total number that sub-network topology is included, and the ratio of the two is calculated, it is the above-mentioned each section found out by ratio output
First accounting value of the point in the sub-network topology.
S1013：Obtain the second accounting value of each node that negative label is carried in the customer relationship network topology.
Similarly, for the customer relationship network topology before undivided, counting the customer relationship network topology is included
Node total number.The customer data in customer relationship network topology corresponding to each node is obtained successively, and judges its customer data
In the attribute values of negative tag types whether be null value.If judging result is no, which is chosen, and keeps it negative
The statistical value of interstitial content adds one.
If the equal poll of each node in customer relationship network topology finishes, the most finish-unification of its negative interstitial content is obtained
The node total number that evaluation is included with customer relationship network topology, and the ratio of the two is calculated, it is above-mentioned visitor by ratio output
The second accounting value of each node of negative label is carried in the relational network topology of family.
S1014：If the difference of the first accounting value and the second accounting value of any sub-network topology is less than
The corresponding customer data of each node institute in the sub-network topology is then determined as the client by the first predetermined threshold value
Sample data.
In the embodiment of the present invention, according to aforesaid way, a first accounting value of each sub-network topology can be obtained.For
Each first accounting value, calculates the difference of itself and the second accounting value.If the difference is less than predetermined threshold value, according to the sub-network
Each node that topology is included, by wherein each node corresponding customer data be determined as the visitor needed for current time
Family sample data.
Preferably, if there are the differences of the first accounting value of multiple sub-networks topology and the second accounting value to be respectively less than default threshold
Value then obtains the sub-network topology of wherein difference minimum.By the corresponding client's number of each node institute in the sub-network topology
According to being determined as client's sample data.
Preferably as one embodiment of the present of invention, above-mentioned S1014 can also include：If any sub-network is opened up
The difference of the first accounting value flutterred and the second accounting value is less than predetermined threshold value, then obtains the sub-network topology and correspond to respectively
The first node of a negative tag types is distributed and the customer relationship network topology corresponds to each negative label
The second node of type is distributed；If the similarity of the first node distribution and second node distribution is more than the second default threshold
The corresponding customer data of each node institute in the sub-network topology is then determined as client's sample data by value.
In the embodiment of the present invention, to any sub-network topology, in the first accounting value and the client for determining sub-network topology
When the difference of second accounting value of relational network topology is less than predetermined threshold value, further judge that the sub-network topology corresponds to each institute
The first node for stating negative tag types is distributed the second node of each negative tag types corresponding with customer relationship network topology
Whether distribution is similar.
Specifically, according to preset each negative tag types, each negative tag types in sub-network topology are obtained
Third node accounting and the fourth node accounting for obtaining each negative tag types in customer relationship network topology.If any negative
The third node accounting of face tag types is identical as fourth node accounting, then is distributed the distribution of above-mentioned first node with second node
Similarity increase a preset value.In addition, to any negative tag types, according to its third node accounting and fourth node accounting
Difference, determine with the matched calculated value of the difference, and by above-mentioned first node distribution with second node distribution similarity drop
The low calculated value.Wherein, third node accounting and the difference of fourth node accounting are bigger, are got over the matched calculated value of the difference
Greatly.
Illustratively, it if preset negative tag types include A, B and C three types, is negatively marked in sub-network topology
The ratio for signing the sum and the sub-network topology interior joint sum of the node that type is A is above-mentioned third node accounting Q1, client
The sum and the ratio of customer relationship network topology interior joint sum for the node that negative tag types are A in relational network topology be
Above-mentioned fourth node accounting Q2.If Q1=Q2, then first node distribution and the similarity of second node distribution are increased into a；If Q1≠
Q2, Q1-Q2=q, q ∈ [1%, 2%], and be b, then first be currently calculated with section [1%, 2%] matched calculated value
The similarity S=a-b of Node distribution and second node distribution.So analogize, until each negative tag types are to similarity
After the completion of impact factor determines, obtains first node distribution and be distributed the similarity being accumulated by with second node.
Judge whether the similarity that first node distribution is finally accumulated by with second node distribution is more than second in advance
If threshold value.If the determination result is YES, then it represents that sub-network topology Node distribution corresponding in each negative tag types with
The Node distribution of customer relationship network topology is as similar, therefore, each node institute in the sub-network topology is corresponding
Customer data is determined as client's sample data.
In the embodiment of the present invention, since client's sample data is the training number for building and training neural network model
According to therefore, by the way that customer relationship network topology is divided into multiple sub-networks topology, and being included with wherein sub-network topology
Customer data is used as client's sample data, can reduce the data volume of training data, to improve neural network model
Training speed.First accounting of each node of negative label in the sub-network topology is carried by calculating in sub-network topology
Value and calculate customer relationship network topology in carry negative label each node the second accounting value, by the first accounting value with
The sub-network topology that the difference of second accounting value is less than predetermined threshold value is chosen, and ensure that sub-network topology and customer relationship net
Network topology has larger similitude in data distribution so that is trained based on client's sample data in the sub-network topology
When neural network model, the higher model training effect of accuracy can be obtained.
S102：Based on client's sample data, builds and train neural network model.
In the embodiment of the present invention, by preset neural network algorithm, the above-mentioned client's sample data determined is carried out
Training obtains the neural network model of the weighted value for exporting each negative tag types with training.Above-mentioned neural network algorithm
Including but not limited to convolutional neural networks algorithm, recurrent neural network algorithm and reverse transfer (Back Propagation,
BP) neural network algorithm etc..
As an embodiment of the present invention, Fig. 3 shows the acquisition of negative label weight provided in an embodiment of the present invention
The specific implementation flow of method S102, details are as follows：
S1021：It obtains the negative sample data for carrying the negative label in client's sample data and carries just
The positive sample data of face label.
S1022：According to the corresponding negative label section of the negative sample data and the positive sample data institute
Point and front label node are searched and the negative label node and all have incidence relation with the front label node
Test node, and obtain the test sample data corresponding to each test node.
S1023：Client's sample data, the negative sample data and the test sample data are inputted respectively
Input layer, output layer and the hidden layer of the neural network model of initialization.
S1024：Based on preset side right weight and threshold value is put, passes through back-propagation algorithm, the training neural network model.
The embodiment of the present invention filters out negative sample data and proves sample number in the client's sample data determined
According to.Wherein, negative sample data refers to carrying the customer data of negative label, and positive sample data refers to carrying front label
Customer data.
Specifically, it if above-mentioned analysis is it is found that the Figure Characteristics of user meet preset negative label rule, is beaten for user
The negative label of one kind corresponding to the upper negative label rule.Therefore, in the customer data corresponding to any node, if it is negative
The attribute value of face tag types is non-null value, then this customer data is negative sample data；If the category of its negative tag types
Property value be null value, then this customer data is positive sample data.
In the embodiment of the present invention, according to the corresponding node of negative sample data and positive sample data institute, respectively
Determine the negative label node in network topology and front label node.Wherein, above-mentioned network topology refers to client's sample
Network topology belonging to data.In the network topology, orient and meanwhile with a negative label node and a front label section
Point all has the intermediate node of incidence relation, which is known as test node.Each test node is read out to distinguish
Corresponding customer data, then these customer datas are test sample data.
The model parameter value of neural network model is obtained, and sets every model parameter value to neural network model first
Under beginning state.Above-mentioned model parameter value includes the side right weight and point threshold value of neural network model.With the above-mentioned visitor determined
Family sample data is used as the input layer data of the neural network model of initialization, with the test sample number corresponding to test node
According to implicit layer data is used as, output layer data is used as with negative sample data.By using preset back-propagation algorithm,
To the weight on each side in neural network model and point threshold value be adjusted, until neural network model output data with it is above-mentioned
When the error of negative sample data is less than predetermined threshold value, just stop adjustment.At this point, after the completion of side right weight and point threshold value are determined
Neural network model be determined as it is final needed for be applied to neural network model in business approval system.
As an embodiment of the present invention, as shown in figure 4, above-mentioned steps S1024 is specifically included：
S10241：By preset gradient descent method, respectively to the weight on each node and each side in training layer
Value is iterated adjustment, and records the iterations at current time；The training layer includes the input layer, output layer and hidden
Containing layer.
S10242：When the iterations reach third predetermined threshold value, alternatively, when neural network model output
When the weighted value of any negative tag types reaches four predetermined threshold values, complete to train the neural network model；It is no
Then, it returns described in executing through preset gradient descent method, respectively to the weight on each node and each side in training layer
Value is iterated adjustment, and records the operation of the iterations at current time.
In the embodiment of the present invention, using gradient descent method come to input layer in neural network model, output layer and implicit
The side right weight and point threshold value of layer are adjusted.After the completion of each adjustment, the number of record current time iteration adjustment, and
It returns and executes above-mentioned client's sample data based on input, each client that acquisition neural network model is exported is respectively negative
The step of weighted value in tag types.Whether the iterations at detection current time have reached third predetermined threshold value, alternatively, refreshing
The 4th predetermined threshold value whether is had reached through any weighted value that network model is exported.If any of the above-described judging result is yes,
It then determines that neural network model has had smaller error amount, therefore, stops carrying out parameter adjustment, output to neural network model
Neural network model after the completion of training.
S103：The Figure Characteristics of each pending client in business approval system are inputted into the neural network mould
Type, to export weighted value of each pending client respectively in each negative tag types.
In the embodiment of the present invention, the neural network model after the completion of training is published on business approval system, so that industry
The client can be determined as current time by business approval system when detecting the loan application request that any client is initiated
Pending user, and calculation process is carried out to the Figure Characteristics of the pending user using the neural network model, to make
Weighted value of the pending client respectively in preset each negative tag types can be exported by obtaining neural network model.
Illustratively, for a pending user, the pending client that neural network model is exported is in negative label
Weighted value on type A is a, and the weighted value on negative tag types B is b etc..
S104：According to the identical each weighted value of the negative tag types, the mark of the negative tag types is calculated
Sign weight.
After being respectively processed to the Figure Characteristics of multiple pending clients using neural network model, it can obtain every
The one pending client weighted value in each negative tag types respectively, it is therefore, different for identical negative tag types
Pending client be corresponding with different weighted values respectively.
As the implementation example of the present invention, being averaged for each weighted value of corresponding identical negative tag types is calculated
Value, the then average value obtained are the label weight of the negative tag types.
Pending for example, if weighted value of the pending client 1 on negative tag types A of neural network model output is a
Weighted value of the core client 2 on negative tag types A is b, and weighted value of the pending client 3 on negative tag types A is c,
Then the label weight of finally obtained negative tag types A is (a+b+c)/3.
Another as the present invention implements example, is exporting different pending clients in same negative tag types
After corresponding different weighted value, further include：
According to the Figure Characteristics of each pending client, the customer grade of each pending client is determined.Above-mentioned client etc.
Grade indicates influence power size of the pending client to other clients.It obtains and preset adds with the matched label weight of the Class Type
Weight coefficient.Wherein, label Weight coefficient is directly proportional to customer grade.
To identical negative tag types, by the power of each pending client corresponding output in the negative tag types
Weight values and its label Weight coefficient be multiplied after processing, then calculates the product of each pending client and be averaged
Value, the then average value obtained are the label weight of the negative tag types.
For example, in the examples described above, if the customer grade of pending client 1 is VIP, the customer grade of pending client 2
For VVIP, and it is k that customer grade VIP and VVIP, which distinguish matched label Weight coefficient,1And k2, neural network model
Weighted value of the pending client 1 of output on negative tag types A is a, and pending client 2 is on negative tag types A
Weighted value is b, then the label weight of finally obtained negative tag types A is (ak1+bk2)/2。
In the embodiment of the present invention, the neural network model of the weighted value by training for exporting each negative tag types,
And entire neural network model is applied to business approval system, it ensure that user without after counting all kinds of negative label weights
Manually inputting again realizes the automation update of label weight in system, reduces operation complexity.Due to neural network mould
The training sample of type is client's sample data in customer relationship network, and has higher pass between every client's sample data
Connection property, therefore, the weighted value finally exported according to neural network model calculate the label weight of each negative tag types, are
The obtained result of calculation that influences each other for having considered other negative label clients avoids label weight analysis dimension list
One the problem of, occurs, and improves the calculating accuracy of label weight.When being updated due to customer relationship network topology, nerve net
Network model also can dynamically update, therefore ensure that business approval system can get the label weight of real-time update, so that
Approving person based on the label weight come to carry out business approval when, the higher approval results of accuracy can be obtained.
It should be understood that the size of the serial number of each step is not meant that the order of the execution order in above-described embodiment, each process
Execution sequence should be determined by its function and internal logic, the implementation process without coping with the embodiment of the present invention constitutes any limit
It is fixed.
Corresponding to the prediction technique of client's purchase intention described in foregoing embodiments, Fig. 5 shows that the embodiment of the present invention carries
The structure diagram of the acquisition device of the negative label weight supplied.For convenience of description, portion related to the present embodiment is illustrated only
Point.
With reference to Fig. 5, which includes：
If acquiring unit 51 obtains plurality of node institute for detecting that customer relationship network topology updates
Corresponding client's sample data, client's sample data include Figure Characteristics, negative tag types and each described
The weighted value of negative tag types.
Training unit 52, for being based on client's sample data, building and training neural network model.
Output unit 53, for the Figure Characteristics of each pending client in business approval system to be inputted the god
Through network model, to export weighted value of each pending client respectively in each negative tag types.
Computing unit 54, for according to the identical each weighted value of the negative tag types, calculating the negative mark
Sign the label weight of type.
Optionally, the acquiring unit 51 includes：
Divide subelement, for updated customer relationship network topology to be divided into multiple sub-network topologys.
Computation subunit, in each sub-network topology, finding out each node for carrying negative label, and
Calculate first accounting value of each node found out in the sub-network topology.
First obtains subelement, for obtaining each node for carrying negative label in the customer relationship network topology
Second accounting value.
Determination subelement, if the first accounting value for any sub-network topology and the second accounting value
Difference is less than the first predetermined threshold value, then is determined as the corresponding customer data of each node institute in the sub-network topology
Client's sample data.
Optionally, the determination subelement is specifically used for：
If the difference of the first accounting value of any sub-network topology and the second accounting value is less than default threshold
Value then obtains the sub-network topology and corresponds to the first node distribution of each negative tag types and the customer relationship net
Network topology corresponds to the second node distribution of each negative tag types；
If the similarity of the first node distribution and second node distribution is more than the second predetermined threshold value, by the son
The corresponding customer data of each node institute is determined as client's sample data in network topology.
Optionally, the training unit 52 includes：
Second obtains subelement, for obtaining the negative sample number for carrying the negative label in client's sample data
According to this and carry front label positive sample data.
Subelement is searched, for corresponding negative according to the negative sample data and the positive sample data institute
Face label node and front label node are searched with the negative label node and are all had with the front label node
The test node of incidence relation, and obtain the test sample data corresponding to each test node.
Subelement is inputted, is used for client's sample data, the negative sample data and the test sample number
According to input layer, output layer and the hidden layer of the neural network model of input initialization respectively.
Training subelement, for based on preset side right weight and point threshold value, passing through back-propagation algorithm, the training nerve
Network model.
Optionally, the trained subelement is specifically used for：
By preset gradient descent method, change respectively to each node in training layer and the weighted value on each side
Generation adjustment, and record the iterations at current time；The training layer includes the input layer, output layer and hidden layer；
When the iterations reach third predetermined threshold value, alternatively, when any institute of neural network model output
When stating the weighted values of negative tag types and reaching four predetermined threshold values, complete to train the neural network model；Otherwise, it returns
By preset gradient descent method described in executing, change respectively to each node in training layer and the weighted value on each side
Generation adjustment, and record the operation of the iterations at current time.
Fig. 6 is the schematic diagram for the terminal device that one embodiment of the invention provides.As shown in fig. 6, the terminal of the embodiment is set
Standby 6 include：Processor 60 and memory 61 are stored with the calculating that can be run on the processor 60 in the memory 61
Machine program 62, for example, negative label weight acquisition program.The processor 60 is realized when executing the computer program 62
State the step in the acquisition methods embodiment of each negative label weight, such as step 101 shown in FIG. 1 is to 104.Alternatively, institute
The function that each module/unit in above-mentioned each device embodiment is realized when processor 60 executes the computer program 62 is stated, such as
The function of unit 51 to 54 shown in Fig. 5.
Illustratively, the computer program 62 can be divided into one or more module/units, it is one or
Multiple module/units are stored in the memory 61, and are executed by the processor 60, to complete the present invention.Described one
A or multiple module/units can be the series of computation machine program instruction section that can complete specific function, which is used for
Implementation procedure of the computer program 62 in the terminal device 6 is described.
The terminal device 6 can be that the calculating such as desktop PC, notebook, palm PC and cloud server are set
It is standby.The terminal device may include, but be not limited only to, processor 60, memory 61.It will be understood by those skilled in the art that Fig. 6
The only example of terminal device 6 does not constitute the restriction to terminal device 6, may include than illustrating more or fewer portions
Part either combines certain components or different components, such as the terminal device can also include input-output equipment, net
Network access device, bus etc..
Alleged processor 60 can be central processing unit (Central Processing Unit, CPU), can also be
Other general processors, digital signal processor (Digital Signal Processor, DSP), application-specific integrated circuit
(Application Specific Integrated Circuit, ASIC), ready-made programmable gate array (Field-
Programmable Gate Array, FPGA) either other programmable logic device, discrete gate or transistor logic,
Discrete hardware components etc..General processor can be microprocessor or the processor can also be any conventional processor
Deng.
The memory 61 can be the internal storage unit of the terminal device 6, such as the hard disk of terminal device 6 or interior
It deposits.The memory 61 can also be to be equipped on the External memory equipment of the terminal device 6, such as the terminal device 6
Plug-in type hard disk, intelligent memory card (Smart Media Card, SMC), secure digital (Secure Digital, SD) card dodge
Deposit card (Flash Card) etc..Further, the memory 61 can also both include the storage inside list of the terminal device 6
Member also includes External memory equipment.The memory 61 is for storing needed for the computer program and the terminal device
Other programs and data.The memory 61 can be also used for temporarily storing the data that has exported or will export.
In addition, each functional unit in each embodiment of the application can be integrated in a processing unit, it can also
It is that each unit physically exists alone, it can also be during two or more units be integrated in one unit.Above-mentioned integrated list
The form that hardware had both may be used in member is realized, can also be realized in the form of SFU software functional unit.
If the integrated unit is realized in the form of SFU software functional unit and sells or use as independent product
When, it can be stored in a computer read/write memory medium.Based on this understanding, the technical solution of the application is substantially
The all or part of the part that contributes to existing technology or the technical solution can be in the form of software products in other words
It embodies, which is stored in a storage medium, including some instructions are used so that a computer
Equipment (can be personal computer, server or the network equipment etc.) executes the complete of each embodiment the method for the application
Portion or part steps.And storage medium above-mentioned includes：USB flash disk, mobile hard disk, read-only memory (Read-Only Memory,
ROM), random access memory (Random Access Memory, RAM), magnetic disc or CD etc. are various can store program
The medium of code.
The above, above example are only to illustrate the technical solution of the application, rather than its limitations；Although with reference to before
Embodiment is stated the application is described in detail, it will be understood by those of ordinary skill in the art that：It still can be to preceding
The technical solution recorded in each embodiment is stated to modify or equivalent replacement of some of the technical features；And these
Modification or replacement, the spirit and scope of each embodiment technical solution of the application that it does not separate the essence of the corresponding technical solution.
Claims (10)
1. a kind of acquisition methods of negative label weight, which is characterized in that including：
If detecting, customer relationship network topology updates, and obtains the corresponding client's sample number of plurality of node institute
According to, client's sample data include Figure Characteristics, negative tag types and each negative tag types weighted value；
Based on client's sample data, builds and train neural network model；
The Figure Characteristics of each pending client in business approval system are inputted into the neural network model, it is every to export
The one pending client weighted value in each negative tag types respectively；
According to the identical each weighted value of the negative tag types, the label weight of the negative tag types is calculated.
2. the acquisition methods of negative label weight as described in claim 1, which is characterized in that if described detect customer relationship
Network topology updates, then obtains the corresponding client's sample data of plurality of node institute, including：
Updated customer relationship network topology is divided into multiple sub-network topologys；
In each sub-network topology, find out each node for carrying negative label, and calculate find out it is described each
First accounting value of a node in the sub-network topology；
Obtain the second accounting value of each node that negative label is carried in the customer relationship network topology；
If the first accounting value of any sub-network topology and the difference of the second accounting value are less than the first default threshold
The corresponding customer data of each node institute in the sub-network topology is then determined as client's sample data by value.
3. the acquisition methods of negative label weight as claimed in claim 2, which is characterized in that if any sub-network
The first accounting value of topology is less than the first predetermined threshold value with the difference of the second accounting value, then will be in the sub-network topology
The corresponding customer data of each node institute is determined as client's sample data, including：
If the first accounting value of any sub-network topology and the difference of the second accounting value are less than predetermined threshold value,
Obtain the sub-network topology correspond to each negative tag types first node distribution and the customer relationship network open up
Flutter the second node distribution of corresponding each negative tag types；
If the similarity of the first node distribution and second node distribution is more than the second predetermined threshold value, by the sub-network
The corresponding customer data of each node institute is determined as client's sample data in topology.
4. the acquisition methods of negative label weight as described in claim 1, which is characterized in that described to be based on client's sample
Data build and train neural network model, including：
It obtains the negative sample data for carrying the negative label in client's sample data and carries front label just
Face sample data；
According to the corresponding negative label node of the negative sample data and the positive sample data institute and front
The test section of incidence relation is searched with the negative label node and all had with the front label node to label node
Point, and obtain the test sample data corresponding to each test node；
Client's sample data, the negative sample data and the test sample data are distinguished to the god of input initialization
Input layer, output layer through network model and hidden layer；
Based on preset side right weight and threshold value is put, passes through back-propagation algorithm, the training neural network model.
5. the acquisition methods of negative label weight as described in claim 1, which is characterized in that described to be based on preset side right weight
The neural network model is trained by back-propagation algorithm with a threshold value, including：
By preset gradient descent method, tune is iterated to each node in training layer and the weighted value on each side respectively
It is whole, and record the iterations at current time；The training layer includes the input layer, output layer and hidden layer；
When the iterations reach third predetermined threshold value, alternatively, working as any described negative of neural network model output
When the weighted value of face tag types reaches four predetermined threshold values, complete to train the neural network model；Otherwise, it returns and executes
It is described by preset gradient descent method, tune is iterated to each node and the weighted value on each side in training layer respectively
It is whole, and record the operation of the iterations at current time.
6. a kind of terminal device, including memory and processor, it is stored with and can transports on the processor in the memory
Capable computer program, which is characterized in that the processor realizes following steps when executing the computer program：
If detecting, customer relationship network topology updates, and obtains the corresponding client's sample number of plurality of node institute
According to, client's sample data include Figure Characteristics, negative tag types and each negative tag types weighted value；
Based on client's sample data, builds and train neural network model；
The Figure Characteristics of each pending client in business approval system are inputted into the neural network model, it is every to export
The one pending client weighted value in each negative tag types respectively；
According to the identical each weighted value of the negative tag types, the label weight of the negative tag types is calculated.
7. terminal device as claimed in claim 6, which is characterized in that if described detect that customer relationship network topology occurs more
Newly, then the step of obtaining plurality of node institute corresponding client's sample data, specifically includes：
Updated customer relationship network topology is divided into multiple sub-network topologys；
In each sub-network topology, find out each node for carrying negative label, and calculate find out it is described each
First accounting value of a node in the sub-network topology；
Obtain the second accounting value of each node that negative label is carried in the customer relationship network topology；
If the first accounting value of any sub-network topology and the difference of the second accounting value are less than the first default threshold
The corresponding customer data of each node institute in the sub-network topology is then determined as client's sample data by value.
8. terminal device as claimed in claim 7, which is characterized in that if described the first of any sub-network topology
Accounting value and the difference of the second accounting value are less than the first predetermined threshold value, then by each node institute in the sub-network topology
Corresponding customer data is determined as the step of client's sample data, specifically includes：
If the first accounting value of any sub-network topology and the difference of the second accounting value are less than predetermined threshold value,
Obtain the sub-network topology correspond to each negative tag types first node distribution and the customer relationship network open up
Flutter the second node distribution of corresponding each negative tag types；
If the similarity of the first node distribution and second node distribution is more than the second predetermined threshold value, by the sub-network
The corresponding customer data of each node institute is determined as client's sample data in topology.
9. terminal device as claimed in claim 6, which is characterized in that it is described to be based on client's sample data, it builds and instructs
It the step of practicing neural network model, specifically includes：
It obtains the negative sample data for carrying the negative label in client's sample data and carries front label just
Face sample data；
According to the corresponding negative label node of the negative sample data and the positive sample data institute and front
The test section of incidence relation is searched with the negative label node and all had with the front label node to label node
Point, and obtain the test sample data corresponding to each test node；
Client's sample data, the negative sample data and the test sample data are distinguished to the god of input initialization
Input layer, output layer through network model and hidden layer；
Based on preset side right weight and threshold value is put, passes through back-propagation algorithm, the training neural network model.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, feature to exist
In when the computer program is executed by processor the step of any one of such as claim 1 to 5 of realization the method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810436265.4A CN108647714A (en) | 2018-05-09 | 2018-05-09 | Acquisition methods, terminal device and the medium of negative label weight |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810436265.4A CN108647714A (en) | 2018-05-09 | 2018-05-09 | Acquisition methods, terminal device and the medium of negative label weight |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108647714A true CN108647714A (en) | 2018-10-12 |
Family
ID=63749710
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810436265.4A Pending CN108647714A (en) | 2018-05-09 | 2018-05-09 | Acquisition methods, terminal device and the medium of negative label weight |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108647714A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109919235A (en) * | 2019-03-13 | 2019-06-21 | 北京邮电大学 | A kind of deep learning image classification model training method based on manual intervention sample set weight |
| CN112070224A (en) * | 2020-08-26 | 2020-12-11 | 成都品果科技有限公司 | Revision system and method of sample for neural network training |
| CN112184143A (en) * | 2020-09-07 | 2021-01-05 | 支付宝(杭州)信息技术有限公司 | Model training method, device and equipment in compliance audit rule |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080275861A1 (en) * | 2007-05-01 | 2008-11-06 | Google Inc. | Inferring User Interests |
| US20100023469A1 (en) * | 2008-07-28 | 2010-01-28 | Charles River Analytics, Inc | Objective decision making application using bias weighting factors |
| US20100269067A1 (en) * | 2009-03-05 | 2010-10-21 | Virginie De Bel Air | User interface to render a user profile |
| CN106529729A (en) * | 2016-11-18 | 2017-03-22 | 同济大学 | Method and system for forecasting default of credit card user based on BP_Adaboost model |
| CN106874922A (en) * | 2015-12-11 | 2017-06-20 | 腾讯科技（深圳）有限公司 | A kind of method and device for determining service parameter |
| CN107133782A (en) * | 2017-05-03 | 2017-09-05 | 扬州大学 | Electronic mail network evolution method based on user profile feature |
| CN107392310A (en) * | 2016-05-16 | 2017-11-24 | 北京陌上花科技有限公司 | neural network model training method and device |
| CN107871277A (en) * | 2017-07-25 | 2018-04-03 | 平安普惠企业管理有限公司 | The method and computer-readable recording medium that server, customer relationship are excavated |
-
2018
- 2018-05-09 CN CN201810436265.4A patent/CN108647714A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080275861A1 (en) * | 2007-05-01 | 2008-11-06 | Google Inc. | Inferring User Interests |
| US20100023469A1 (en) * | 2008-07-28 | 2010-01-28 | Charles River Analytics, Inc | Objective decision making application using bias weighting factors |
| US20100269067A1 (en) * | 2009-03-05 | 2010-10-21 | Virginie De Bel Air | User interface to render a user profile |
| CN106874922A (en) * | 2015-12-11 | 2017-06-20 | 腾讯科技（深圳）有限公司 | A kind of method and device for determining service parameter |
| CN107392310A (en) * | 2016-05-16 | 2017-11-24 | 北京陌上花科技有限公司 | neural network model training method and device |
| CN106529729A (en) * | 2016-11-18 | 2017-03-22 | 同济大学 | Method and system for forecasting default of credit card user based on BP_Adaboost model |
| CN107133782A (en) * | 2017-05-03 | 2017-09-05 | 扬州大学 | Electronic mail network evolution method based on user profile feature |
| CN107871277A (en) * | 2017-07-25 | 2018-04-03 | 平安普惠企业管理有限公司 | The method and computer-readable recording medium that server, customer relationship are excavated |
Non-Patent Citations (1)
| Title |
|---|
| 丁峰 等: "《基于情感词识别网络负面舆情的信息通报技术研究》", vol. 2, pages 145 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109919235A (en) * | 2019-03-13 | 2019-06-21 | 北京邮电大学 | A kind of deep learning image classification model training method based on manual intervention sample set weight |
| CN109919235B (en) * | 2019-03-13 | 2021-08-20 | 北京邮电大学 | Deep learning image classification model training method based on manual intervention sample set weight |
| CN112070224A (en) * | 2020-08-26 | 2020-12-11 | 成都品果科技有限公司 | Revision system and method of sample for neural network training |
| CN112070224B (en) * | 2020-08-26 | 2024-02-23 | 成都品果科技有限公司 | Revision system and method of samples for neural network training |
| CN112184143A (en) * | 2020-09-07 | 2021-01-05 | 支付宝(杭州)信息技术有限公司 | Model training method, device and equipment in compliance audit rule |
| CN112184143B (en) * | 2020-09-07 | 2022-04-29 | 支付宝(杭州)信息技术有限公司 | Model training method, device and equipment in compliance audit rule |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110956273A (en) | Credit scoring method and system integrating multiple machine learning models | |
| CN108665159A (en) | A kind of methods of risk assessment, device, terminal device and storage medium | |
| CN109685635A (en) | Methods of risk assessment, air control server-side and the storage medium of financial business | |
| CN107818344A (en) | The method and system that user behavior is classified and predicted | |
| CN104915879A (en) | Social relationship mining method and device based on financial data | |
| CN107169768A (en) | The acquisition methods and device of abnormal transaction data | |
| CN107230108A (en) | The processing method and processing device of business datum | |
| CN108647714A (en) | Acquisition methods, terminal device and the medium of negative label weight | |
| CN110147389A (en) | Account number treating method and apparatus, storage medium and electronic device | |
| CN108734587A (en) | The recommendation method and terminal device of financial product | |
| CN106874943A (en) | Business object sorting technique and system | |
| CN109829721A (en) | Online trading multiagent behavior modeling method based on heterogeneous network representative learning | |
| CN111882420A (en) | Generation method of response rate, marketing method, model training method and device | |
| CN111159241B (en) | Click conversion estimation method and device | |
| CN113313538A (en) | User consumption capacity prediction method and device, electronic equipment and storage medium | |
| CN115545886A (en) | Overdue risk identification method, overdue risk identification device, overdue risk identification equipment and storage medium | |
| CN112950347B (en) | Resource data processing optimization method and device, storage medium and terminal | |
| CN110458600A (en) | Portrait model training method, device, computer equipment and storage medium | |
| CN113988459A (en) | Small and medium-sized enterprise growth assessment method and system based on electric power marketing data | |
| CN109711849A (en) | Ether house address portrait generation method and device, electronic equipment and storage medium | |
| CN111291936B (en) | Product life cycle prediction model generation method and device and electronic equipment | |
| CN112435068A (en) | Malicious order identification method and device, electronic equipment and storage medium | |
| CN110264311B (en) | Business promotion information accurate recommendation method and system based on deep learning | |
| CN109767333A (en) | Select based method, device, electronic equipment and computer readable storage medium | |
| CN111291838B (en) | Method and device for interpreting entity object classification result |
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 |