Specific embodiment
It is specific below in conjunction with this specification to keep the purposes, technical schemes and advantages of the embodiment of the present application clearer
The technical solution of the embodiment of the present application is clearly and completely described in embodiment and corresponding attached drawing.Obviously, described reality
Applying example only is this specification a part of the embodiment, instead of all the embodiments.The embodiment of base in this manual, this field
Those of ordinary skill's every other embodiment obtained without making creative work belongs to the application implementation
The range of example protection.
As previously mentioned, the no specific aim of dispensing of marketing equity has a large number of users not after usually marketing activity is promoted at present
Enthusiasm can be generated, causes marketing effectiveness undesirable.
In order to guarantee that most of user can support marketing activity, the way being readily apparent that is to launch high synthesis to user to comment
The marketing equity divided.Comprehensive score height depends on the rate of drawing of marketing equity, and (rate of drawing refers to that user uses marketing equity
Probability) and marketing equity cost of marketing.But in this mode, the draw rate and cost of marketing for equity of marketing have approximately linear
Relationship, and this relationship be not always set up.
Such as under the not abundant scene of totle drilling cost, no normal direction user launches the marketing equity of high value, at this moment just needs
It adjusts cost of marketing and draws weight of the rate in comprehensive score.In weight adjustment, cost of marketing and draws rate and may go out
Existing acute variation, might not obtain ideal trade-off.
In view of this, the application draws two angles of rate from control cost of marketing and user, provides and both a kind of be in
The dispensing scheme of the marketing equity of non-linear relation.
On the one hand, the embodiment of the present application provides a kind of put-on method of equity of marketing, as shown in Figure 1, comprising:
The quantity of step S102, the potential user group based on marketing equity are grouped marketing equity, potential user group
The corresponding marketing equity grouping of each user, any two marketing is weighed in marketing equity grouping of the marketing equity greater than 1
Beneficial mutual exclusion;
For step S102:
Marketing equity is the preferential equity provided a user in marketing activity, for example cash red packet common in online shopping, is returned
Sharp red packet, interest-free certificate, coupons etc..Draw rate height determine user whether approve marketing equity, can withdraw deposit market equity to
The value at family.
Potential user group includes the popularization object that user is marketing activity.This step can be each user of potential user group
Determine that a marketing equity grouping, the marketing equity grouping of each user may include marketing equity and depend on marketing strategy, with
Marketing equity comprising all kinds also may include the marketing equity of Partial Species.
As exemplary introduction, it is assumed that potential user group includes user 1, user 2 and user 3, this marketing activity is thrown
The marketing equity put has tri- kinds of ABC.
If the quantity of each marketing equity is set as infinitely great, the marketing equity grouping of user 1, user 2 and user 3 are equal
It may include tri- kinds of equity ABC of marketing;
If the quantity of each marketing equity is set as preset value, the quantity of difference marketing equity is identical or different, then user
The quantity of the marketing equity of acquisition should not surpass preset value.
Such as marketing equity A quantity only there are two, market equity B quantity only one, marketing equity C quantity have
Three;
Then as one of feasible implementation, the marketing equity grouping of user 1 may include: marketing equity A, battalion
Equity C is sold, the marketing equity grouping of user 2 may include: marketing equity C, and the marketing equity grouping of user 3 may include marketing
Equity A, marketing equity B, marketing equity C;
Or the marketing equity grouping of user 1 may include: marketing equity A and marketing equity C, the marketing equity of user 2
Grouping may include: marketing equity B and marketing equity C, and the marketing equity grouping of user 3 may include marketing equity A and marketing
Equity C.
Since mode is not unique, no longer citing is repeated herein.
Step S104, the corresponding prediction of the cost of marketing upper limit and each marketing equity based on potential user group are drawn
Rate draws the maximum strategy of rate according to the totality of potential user group, in each user corresponding marketing equity grouping selection with
The matched target marketing equity of user.
For step S104:
Each marketing equity is corresponding with respective cost of marketing, it is however generally that, the cost of marketing for equity of marketing is higher, then
Bigger to the preferential equity of user, corresponding to draw rate higher.But the cost of marketing of potential user group is limited, this step
It suddenly is each most suitable marketing of user in potential user group under the premise of the cost of marketing upper limit without departing from potential user group
Equity draws rate to maximize the totality of potential user group, guarantees the marketing benefit of this marketing activity.
Specifically, this step can be based on grouping knapsack algorithm, to being respectively multiple use from multiple candidate marketing equity
Family selects matched target marketing equity.
Grouping knapsack algorithm refers to that hypothesis has the knapsack that N part article and a capacity are V.The expense of i-th article is ci,
Value is wi.These articles are divided into several groups, and the article in every group conflicts mutually, most multiselect one.Which solve object
Product, which are packed into knapsack, can make the expense summation of these articles be no more than knapsack capacity, and aggregate value is maximum.
According to above-mentioned definition, as the article in grouping knapsack algorithm, (quantity of article is infinite to the equity that can will market
Big or set value), using the cost of marketing for equity of marketing as the expense in grouping knapsack algorithm, the corresponding prediction of the equity that will market
Rate is drawn as the value in grouping knapsack algorithm, using the cost of marketing upper limit of potential user group as in grouping knapsack algorithm
Knapsack capacity passes through the corresponding target marketing equity of user each in knapsack grouping algorithm solution potential user group.
Step S106, the user into potential user group launch matched target marketing equity respectively.
For step 106:
Shown in Fig. 2, it can be based on release time decision after launching trigger event and occurring, launch target marketing to user
Equity.Wherein, putting mode is not unique, as exemplary introduction, passive dispensing channel can be used, such as short message, push, card
The modes such as piece, footmark, corset, favour payment launch target marketing equity;Or also can be used and actively launch channel, such as
Homepage bullet screen mode launches target marketing equity.
The put-on method of the embodiment of the present application is under the premise of the cost of marketing upper limit without departing from potential user group, for not
The marketing equity that height draws rate is launched with user individual, realizes maximized marketing benefit.Due to by the battalion of potential user group
The cost upper limit is sold as the constraint condition launched, so that cost of marketing is drawn with prediction avoids the formation of similar linear number between rate
Relationship need to only determine the marketing of marketing activity in practical applications, not need additionally to configure weight parameter for the two
After cost, suitable marketing equity can be matched for different user, whole process executes simple, convenient and is easy to adjust, side
Case has practicability.
Wherein, the application can draw data based on history marketing event, assess all users to the branch of marketing equity
With behavior, rate is drawn to obtain the corresponding prediction of marketing equity.
As exemplary introduction, the embodiment of the present application can construct one and draw model, can be with when executing step S104
By user's representation data of user in market equity and potential user group be input to it is preset draw model, obtain by drawing model
Rate is drawn in the corresponding prediction of the marketing equity of output.
Wherein, drawing model is obtained based on the training of training sample data, and training sample data, which can be, have been marked
History marketing event, including user's Figure Characteristics and marketing event feature.In the training process, by user's Figure Characteristics and marketing
Affair character is integrated, as the feature vector for drawing model.Later, by the training sample data that have marked to drawing mould
Type is trained, and the marketing equity launched with to determine feature vector to marketing event draws weight.
Rate and marketing event and user's oneself factor phase are drawn in the prediction of the marketing equity for drawing model reckoning after training
Association, therefore accuracy with higher.
Exemplary introduction is carried out to the building for drawing model below with reference to an implementation.
As shown in figure 3, the embodiment of the present application can first establish a kind of FTRL (single machine that Google announces of a deep learning
Single thread mode runs program) model, neural network structure schematic diagram is as shown in figure 3, include upper layer functions and embeding layer.
The feature vector of embeding layer low-dimensional can be, but not limited to be special by user identity Figure Characteristics, user's property portrait
Sign, user credit using Figure Characteristics, marketing event temporal characteristics, marketing event frequecy characteristic and marketing equity feature integration and
At.
Wherein, user identity Figure Characteristics can be user base information, for example, the age, gender, education, family, religion,
The information such as occupation.
User's property Figure Characteristics can be wealth information of user, such as savings, consumption, investment, insurance etc..
User credit using Figure Characteristics is the information of user's debt-credit, for example, borrow, flower, monetary lever, housing loan etc..
Marketing event temporal characteristics can be the release time of marketing event, such as one day, the year of one day, the middle of the month in week
In certain moon, the time interval launched of vacation, working day or marketing event and launch frequency (for example marketing behavior be real-time
Frequency and the offline frequency of marketing behavior) etc..
Marketing event frequecy characteristic can be the dispensing frequency of marketing event, for example be frequency marketing event or interval
Property marketing event etc..
Marketing equity feature can market the related feature of equity, for example, equity title, content, classification, launch channel
Deng.
After the feature vector of FTRL model determines, it can be trained based on training sample data to model is drawn, with
Determine the weighted value of feature vector.
Reckoning result due to drawing model determines the dispensing decision of subsequent marketing equity, determines in order to avoid drawing model
Cause marketing equity expection draw rate and equity of marketing actually draw rate there are relatively large deviation, the embodiment of the present application can will
Marketing equity draws rate as the prediction of marketing equity for the average value of all training sample data.
At the same time it can also further be verified to the accuracy for drawing model.For example, being used before executing step 104
Test sample data are drawn model and are tested, and are based on test result (index of the AUC area under the curve of such as test result)
The accuracy of model is drawn in assessment, to be adjusted to the parameter for drawing model.
As exemplary introduction, the embodiment of the present application can accuracy rate based on test result and after drawing model training
The accuracy rate of the training result of acquisition draws whether model over-fitting occurs described in judgement;
If the accuracy rate of training result is higher than the accuracy rate of test result, and the two difference is than or equal to default threshold
Value then illustrates test result and training result there are relatively large deviation, and confirmation draws model and over-fitting occurs.At this time, it may be necessary to branch
It is adjusted with model, to reduce the error of over-fitting.
Specifically, the present embodiment can solve in the following ways overfitting problem:
Mode one:
The feature vector that model frequency of occurrence in training sample data is less than or equal to preset threshold is drawn in removal, it
Afterwards, it is based on training sample data again, is trained to model is drawn.
Mode one can simplify the neural network structure for drawing model, avoid drawing model because of intrinsic nerve network structure mistake
There is the problem of over-fitting after leading to training in degree complexity.
Mode two:
Based on priori knowledge, Regularization is carried out (for example, using L1 regularization to the Partial Feature vector for drawing model
Method or L2 canonical method directly determine out the weight for drawing model part feature vector).Later, it is based on training sample again
Data are trained to model is drawn.
In mode two, by the feature vector of regularization as the constraint condition for drawing model re -training, so that training institute
Error function is more likely to the direction that selection meets the gradient reduction of constraint, and the training effect for drawing model is finally allowed more to connect
Nearly priori knowledge avoids the problem that or over-fitting occurs after training.
Mode three:
Based on training sample data, training is re-started to model is drawn according to dropout strategy;
In mode three, under dropout strategy, the feature vector of model is drawn in every wheel is trained by Random Activation, from
And the influence in training process between feature vector is eliminated to a certain extent, make to draw model and reduced after re -training to intend
The error of conjunction.
Mode four:
After adjusting training number, training sample data are based on, re -training is carried out to the model of drawing.
It under normal circumstances, can deconditioning when can not effectively improve model accuracy again after training.By repeatedly real
Discovery is trampled, when reaching the 1.5~2 of quantity of training sample data times, optimization income starts to be decreased obviously frequency of training, therefore
Frequency of training adjusted is advisable with 1.5~2 times that are no more than the quantity of training sample data.
Wherein, the error that mode four can occur to avoid the training aggravation over-fitting of excessive number.
In addition, the present embodiment is also based on offline evaluation sample data, offline evaluation carried out to rate of drawing, and according to from
Line assessment result draws the parameter of model to optimize.Since offline evaluation is identical as test assessment principle, no longer lifted herein
Example repeats.
After the completion of drawing model and finally constructing, rate is drawn in the prediction that marketing equity can be obtained by drawing model.
Later, using grouping knapsack algorithm, selection is matched with user in the corresponding marketing equity grouping of each user
Target marketing equity.
As exemplary introduction, it is assumed that the sum of user is M, and the type for equity of marketing has K, and total equity number of marketing is N,
In the case where not considering to market equity inventory, the quantity of every kind of marketing equity is $ ni(i=1,2 ... ..., K) is infinitely great, often
The cost of kind equity is ci(i=1,2 ..., K).
Based on grouping knapsack principle, user can be divided into M group, i.e., each user is as a grouping, each grouping
Marketing equity comprising all kinds, i.e. K different types of marketing equity.So, be grouped knapsack algorithm in grouping be
The value of user (user), article, that is, prize (marketing equity) in each grouping, the article in each grouping are exactly that pij (is used
The corresponding marketing equity j's of family i draws rate), if user is averaged, the cost of marketing upper limit is C, and knapsack total capacity is C*M, marketing
Whether equity is chosen for xij(whether the corresponding marketing equity j of user i is selected);
According to formulaThe target marketing equity of each user can be sought out.
Wherein, the definition based on grouping knapsack, above-mentioned formula need to abide by condition defined below:
It is the introduction to the put-on method of the present embodiment above.It should be noted that in this embodiment scheme, for drawing
The deep learning method of rate prediction can draw model substitution with other, as long as in principle in view of the user of identical dimensional is special
The regression model of sign and marketing equity feature can do this prediction, and the only depth model uniquely distinguished is in big data field
Stability and veracity in scape, and these prediction results for drawing model can use the scheme of the embodiment of the present application.
With the put-on method of above-mentioned marketing equity correspondingly, as shown in figure 4, the embodiment of the present application also provides a kind of marketing
The delivery device 400 of equity, comprising:
The quantity of grouping module 41, the potential user group based on marketing equity is grouped marketing equity, the target
The corresponding marketing equity grouping of each user of user group, marketing equity are greater than any two in 1 marketing equity grouping
Marketing equity mutual exclusion;
Selecting module 42, the cost of marketing upper limit and each marketing equity based on the potential user group are corresponding pre-
Rate is drawn in survey, draws the maximum strategy of rate according to the totality of potential user group, in the corresponding marketing equity grouping of each user
Selection and the matched target marketing equity of the user;
Putting module 43, the user in Xiang Suoshu potential user group launch matched target marketing equity respectively.
The delivery device of the embodiment of the present application is under the premise of without departing from the cost of marketing upper limit, for different user personalization
The marketing equity that height draws rate is launched, realizes maximized marketing benefit.Due to using the cost of marketing upper limit as the constraint launched
Condition, so that cost of marketing is drawn with prediction avoids the formation of similar linear mathematical relationship between rate, thus in practical applications,
It does not need additionally to configure weight parameter for the two, can be different user after the cost of marketing that need to only determine marketing activity
Suitable marketing equity is allotted, whole process executes simple, convenient and is easy to adjust, and scheme has practicability.
Wherein, the selecting module 42 will specifically market equity as the article in grouping knapsack algorithm, by equity of marketing
Cost of marketing as it is described grouping knapsack algorithm in expense, the equity that will market it is corresponding prediction draw rate as the grouping
Value in knapsack algorithm, using the cost of marketing upper limit of the potential user group as the knapsack capacity in the knapsack algorithm,
Pass through the corresponding target marketing equity of user each in the knapsack grouping algorithm solution potential user group.
Wherein, the quantity of each marketing equity is set as infinitely great;Alternatively, the quantity of each marketing equity is set as default
The quantity of value, difference marketing equity is identical or different.
Optionally, selecting module 42 is also used to: user's representation data of user in market equity and potential user group is defeated
Enter to preset and draw model, obtains the corresponding prediction of marketing equity and draw rate.
Optionally, the model of drawing is based on the number of training for including user's Figure Characteristics and marketing event feature
It is obtained according to training.
Wherein, user's Figure Characteristics include: user identity Figure Characteristics, user's property Figure Characteristics, user credit
Use Figure Characteristics;
The marketing event feature includes: marketing event temporal characteristics and marketing event frequecy characteristic.
Optionally, the delivery device of the embodiment of the present application further include:
Test module, by test sample data be input to it is described draw model, obtain test result;
Judgment module, accuracy rate based on the test result and described draws the training result obtained after model training
Accuracy rate, draw whether model over-fitting occurs described in judgement;
Training module re-starts training to the model of drawing if drawing model over-fitting occurs, described in reducing
Draw the error of model over-fitting.
Wherein, training module specifically includes following any unit:
First training unit draws model frequency of occurrence in training sample data for removal and is less than or equal to default threshold
The feature vector of value re-starts training to model is drawn based on training sample data later.
Second training unit is based on later for carrying out Regularization to the Partial Feature vector for drawing model
Training sample data re-start training to model is drawn.
Third training unit re-starts instruction to model is drawn according to dropout strategy for being based on training sample data
Practice;Wherein, under dropout strategy, the feature vector of model is drawn in every wheel is trained by Random Activation.
Obviously, the delivery device of the embodiment of the present application can be used as the executing subject of the put-on method of above-mentioned marketing equity,
Therefore technical effect achieved by the put-on method, the delivery device of the embodiment of the present application equally also can be realized.For example, this
Application embodiment delivery device may be implemented put-on method in figs. 1 to 3 shown in function.
Fig. 5 is the structural schematic diagram of one embodiment electronic equipment of the application.Referring to FIG. 5, in hardware view, the electricity
Sub- equipment includes processor, optionally further comprising internal bus, network interface, memory.Wherein, memory may be comprising interior
It deposits, such as high-speed random access memory (Random-Access Memory, RAM), it is also possible to further include non-volatile memories
Device (non-volatile memory), for example, at least 1 magnetic disk storage etc..Certainly, which is also possible that other
Hardware required for business.
Processor, network interface and memory can be connected with each other by internal bus, which can be ISA
(Industry Standard Architecture, industry standard architecture) bus, PCI (Peripheral
Component Interconnect, Peripheral Component Interconnect standard) bus or EISA (Extended Industry Standard
Architecture, expanding the industrial standard structure) bus etc..The bus can be divided into address bus, data/address bus, control always
Line etc..Only to be indicated with a four-headed arrow in Fig. 5, it is not intended that an only bus or a type of convenient for indicating
Bus.
Memory, for storing program.Specifically, program may include program code, and said program code includes calculating
Machine operational order.Memory may include memory and nonvolatile memory, and provide instruction and data to processor.
Processor is from the then operation into memory of corresponding computer program is read in nonvolatile memory, in logical layer
Multimedia play equipment is formed on face.Processor executes the program that memory is stored, and is specifically used for executing following operation:
The quantity of potential user group based on marketing equity is grouped marketing equity, each of described potential user group
User corresponds to a marketing equity grouping, any two marketing equity mutual exclusion in marketing equity grouping of the marketing equity greater than 1;
Rate, root are drawn in the corresponding prediction of the cost of marketing upper limit and each marketing equity based on the potential user group
The maximum strategy of rate is drawn according to the totality of potential user group, selection and the use in the corresponding marketing equity grouping of each user
The matched target marketing equity in family;
User into the potential user group launches matched target marketing equity respectively.
The method that electronic equipment disclosed in the above-mentioned embodiment illustrated in fig. 1 such as the application executes can be applied in processor,
Or it is realized by processor.Processor may be a kind of IC chip, the processing capacity with signal.In the process of realization
In, each step of the above method can be complete by the integrated logic circuit of the hardware in processor or the instruction of software form
At.Above-mentioned processor can be general processor, including central processing unit (Central Processing Unit, CPU),
Network processing unit (Network Processor, NP) etc.;It can also be digital signal processor (Digital Signal
Processor, DSP), it is specific integrated circuit (Application Specific Integrated Circuit, ASIC), existing
Field programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic device are divided
Vertical door or transistor logic, discrete hardware components.It may be implemented or execute and is in the embodiment of the present application disclosed each
Method, step and logic diagram.General processor can be microprocessor or the processor is also possible to any conventional place
Manage device etc..The step of method in conjunction with disclosed in the embodiment of the present application, can be embodied directly in hardware decoding processor and execute
At, or in decoding processor hardware and software module combination execute completion.Software module can be located at random access memory,
This fields such as flash memory, read-only memory, programmable read only memory or electrically erasable programmable memory, register maturation
In storage medium.The storage medium is located at memory, and processor reads the information in memory, completes above-mentioned side in conjunction with its hardware
The step of method.
The method that the electronic equipment can also carry out Fig. 1, and realize delivery device in Fig. 1-embodiment illustrated in fig. 3 function,
Details are not described herein for the embodiment of the present application.
Certainly, other than software realization mode, other implementations are not precluded in the electronic equipment of the application, for example patrol
Collect device or the mode of software and hardware combining etc., that is to say, that the executing subject of following process flow is not limited to each patrol
Unit is collected, hardware or logical device are also possible to.
The embodiment of the present application also proposed a kind of computer readable storage medium, the computer-readable recording medium storage one
A or multiple programs, the one or more program include instruction, and the instruction is when by the portable electronic including multiple application programs
When equipment executes, the method that the portable electronic device can be made to execute embodiment illustrated in fig. 1, and be specifically used for executing following behaviour
Make:
The quantity of potential user group based on marketing equity is grouped marketing equity, each of described potential user group
User corresponds to a marketing equity grouping, any two marketing equity mutual exclusion in marketing equity grouping of the marketing equity greater than 1;
Rate, root are drawn in the corresponding prediction of the cost of marketing upper limit and each marketing equity based on the potential user group
The maximum strategy of rate is drawn according to the totality of potential user group, selection and the use in the corresponding marketing equity grouping of each user
The matched target marketing equity in family;
User into the potential user group launches matched target marketing equity respectively.
Computer-readable medium includes permanent and non-permanent, removable and non-removable media can be by any method
Or technology come realize information store.Information can be computer readable instructions, data structure, the module of program or other data.
The example of the storage medium of computer includes, but are not limited to phase change memory (PRAM), static random access memory (SRAM), moves
State random access memory (DRAM), other kinds of random access memory (RAM), read-only memory (ROM), electric erasable
Programmable read only memory (EEPROM), flash memory or other memory techniques, read-only disc read only memory (CD-ROM) (CD-ROM),
Digital versatile disc (DVD) or other optical storage, magnetic cassettes, tape magnetic disk storage or other magnetic storage devices
Or any other non-transmission medium, can be used for storage can be accessed by a computing device information.As defined in this article, it calculates
Machine readable medium does not include temporary computer readable media (transitory media), such as the data-signal and carrier wave of modulation.
It should be understood that when the computer program in the computer readable storage medium of the embodiment of the present application is executed by processor,
It can be realized put-on method shown in FIG. 1, and function shown in realizing delivery device in figs. 1 to 3, repeats no more herein.
In short, being not intended to limit the protection scope of the application the foregoing is merely the preferred embodiment of the application.
Within the spirit and principles of this application, any modification, equivalent replacement, improvement and so on should be included in the application's
Within protection scope.
It should also be noted that, the terms "include", "comprise" or its any other variant are intended to nonexcludability
It include so that the process, method, commodity or the equipment that include a series of elements not only include those elements, but also to wrap
Include other elements that are not explicitly listed, or further include for this process, method, commodity or equipment intrinsic want
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including described want
There is also other identical elements in the process, method of element, commodity or equipment.
All the embodiments in this specification are described in a progressive manner, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for system reality
For applying example, since it is substantially similar to the method embodiment, so being described relatively simple, related place is referring to embodiment of the method
Part explanation.