CN109389424A - Flow allocation method, device, electronic equipment and storage medium - Google Patents
Flow allocation method, device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN109389424A CN109389424A CN201811102772.0A CN201811102772A CN109389424A CN 109389424 A CN109389424 A CN 109389424A CN 201811102772 A CN201811102772 A CN 201811102772A CN 109389424 A CN109389424 A CN 109389424A
- Authority
- CN
- China
- Prior art keywords
- works
- presented
- maximum
- light exposure
- global gain
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0269—Targeted advertisements based on user profile or attribute
- G06Q30/0271—Personalized advertisement
-
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0276—Advertisement creation
-
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0277—Online advertisement
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Economics (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
This application provides a kind of flow allocation method, device, electronic equipment and storage mediums, comprising: obtains the specified cold start-up light exposure of each works to be presented;Interaction data is estimated according to what the user behavior data generated in real time obtained each works to be presented;Under the conditions of global gain is maximum, the maximum matching value for estimating interaction data and the specified cold start-up light exposure for obtaining user sends in preset time period request data, the works to be presented distributes each works to be presented corresponding flow according to the matching value.Scheme provided by the present application has taken into account the nominal exposure amount of user experience and works to be presented, and each works to be presented is made to obtain matched request data, more rationally and accurate to the assignment of traffic of each works to be presented.
Description
Technical field
This application involves internet application field, especially a kind of flow allocation method, device, electronic equipment and storage are situated between
Matter.
Background technique
New works needs with certain flow to go that it is helped to be cold-started after generating, and very little cold start-up flow will cause
Author is lost, and can also miss many high-quality works, and excessive cold start-up flow will affect whole recommendation experience.
In the related technology, such as: by SHALE algorithm, the offline part that solves is directed to each advertisement solution dual variable,
The advertising aggregator and the determining final advertisement to be presented of dual variable solved offline that line service is hit according to current request,
Guarantee that each advertisement and guarantor are measured contract and can be met.The program only considered whether advertisement contract is met, and not account for seeing
Crowd's experience.In the related technology, if only considering viewer experience, distribution flow is removed fully according to hobby of the spectators to works, most
Will lead to a small number of works eventually and summarize most flows, tail portion author hardly results in cold start-up flow, cause part works without
Method reaches it and arranges light exposure.To sum up, the relevant technologies can not distribute accurately flow for cold start-up works.
Summary of the invention
To overcome the problems in correlation technique, the application provide a kind of flow allocation method, device, electronic equipment and
Storage medium, to solve the problems, such as to cannot be considered in terms of the specified cold start-up light exposure and user preferences of works.
According to the embodiment of the present application in a first aspect, providing a kind of flow allocation method, comprising:
Obtain the specified cold start-up light exposure of each works to be presented;
Interaction data is estimated according to what the user behavior data generated in real time obtained each works to be presented;
Under the conditions of global gain is maximum, the request data, described to be presented that user sends in preset time period is obtained
The maximum matching value for estimating interaction data and the specified cold start-up light exposure of works distributes each according to the matching value
The corresponding flow of works to be presented.
Optionally, described under the conditions of global gain is maximum, obtain preset time period in user send request data,
The step of maximum matching value for estimating interaction data and the specified cold start-up light exposure of the works to be presented, comprising:
According to the expectation displaying value estimated interaction data and determine each works to be presented;
In the case where the light exposure of each works to be presented reaches the constraint condition of specified cold start-up light exposure, based on it is described each to
It shows the expectation displaying value of works and its shows that probability obtains global gain;
When global gain maximum, the request data, the works to be presented that user sends in preset time period are obtained
Estimate the maximum matching value of interaction data and the specified cold start-up light exposure.
Optionally, the expectation displaying value and its displaying probability based on each works to be presented obtains global gain
The step of, comprising:
The expectation displaying value for calculating each works to be presented shows the product of probability with it, will be described in each works to be presented
Result of product carries out cumulative acquisition global gain.
Optionally, the result of product by each works to be presented carry out cumulative the step of obtaining global gain it
Afterwards, further includes:
Optimize the accumulation result so that the global gain of accumulation result characterization is maximum.
Optionally, the global gain maximum for optimizing the accumulation result so that the accumulation result characterizes, description are as follows:
Wherein, subscript j indicates request, and subscript j indicates works to be presented, wijIndicate user and works j in i-th request
Expect displaying value, xijIndicate whether i-th request shows works j, djIndicate the specified cold start-up light exposure of works j, pjIt indicates
Works probability to be presented, λ are the regular coefficients of regular terms.
Optionally, optimize the accumulation result so that the accumulation result characterization global gain maximum step, comprising:
Expect that the displaying probability of displaying value and the works is full with works to be presented described in Lagrange duality algorithm description
Sufficient preset condition obtains the corresponding dual variable of each works to be presented by iteration of interlocking;
Optimize the accumulation result so that global gain is maximum based on the corresponding dual variable of each works.
It is optionally, described the accumulation result to be optimized based on the corresponding dual variable of each works so that global gain is maximum,
Description are as follows:
Wherein, subscript i indicates request, and subscript j indicates works to be presented, wijIndicate user and works j in i-th request
Expect displaying value, xijIndicate whether i-th request shows works j, djIndicate the specified cold start-up light exposure of works j, pjIt indicates
Works probability to be presented, λ are the regular coefficient of regular terms, βiIt is the dual variable of flow side, αjIt is works side to mutation
Amount, γijIt is probability xijDual variable.
Optionally, described under the conditions of global gain is maximum, obtain preset time period in user send request data,
Before the step of maximum matching value for estimating interaction data and the specified cold start-up light exposure of the works to be presented, also
Include:
All works to be presented are divided into multiple works subsets, wherein each works subset is in the preset time period
The works inside received show that number of requests is identical.
Optionally, the interaction data of estimating includes estimating clicking rate, estimating the rate of thumbing up, estimate concern rate.
Optionally, described under the conditions of global gain is maximum, obtain preset time period in user send request data,
The step of maximum matching value for estimating interaction data and the specified cold start-up light exposure of the works to be presented, comprising:
Obtain works to be presented estimates interaction data, determines each works to be presented according to the interaction data of estimating
Expect displaying value;
The request data received in expectation displaying value and preset time period based on each works to be presented carry out from
Lines matching obtains the corresponding dual variable of each works to be presented;
The request data being currently received is carried out online in conjunction with the expectation displaying value and dual variable of each works to be presented
Matching obtains the works to be presented to match with the request data.
According to the second aspect of the embodiment of the present application, a kind of flow distribution device is provided, comprising:
Light exposure module is obtained, is configured as obtaining the specified cold start-up light exposure of each works to be presented;
Data module is obtained, is configured as obtaining the pre- of each works to be presented according to the user behavior data generated in real time
Estimate interaction data;
Assignment of traffic module is obtained, is configured as under the conditions of global gain is maximum, user in preset time period is obtained
The maximum matching for estimating interaction data and the specified cold start-up light exposure of the request data of transmission, the works to be presented
Value distributes each works to be presented corresponding flow according to the matching value.
According to the third aspect of the embodiment of the present application, a kind of electronic equipment, including processor are provided;For storage processor
The memory of executable instruction;Wherein, the processor is configured to the step of above-mentioned flow allocation method.
According to the fourth aspect of the embodiment of the present application, a kind of non-transitorycomputer readable storage medium is provided, when described
When instruction in storage medium is executed by the processor of electronic equipment, so that electronic equipment is able to carry out above-mentioned flow allocation method
The step of.
The 5th aspect that embodiment is disclosed according to the application, provides a kind of computer program product, including computer program
Code, the computer program include program instruction, when described program instruction is computer-executed, execute the computer
The step of above-mentioned flow allocation method.
The technical solution that embodiments herein provides can include the following benefits:
This patent provides a kind of flow allocation method, by obtaining the request data, institute that user sends in preset time period
The maximum matching value for estimating interaction data and the specified cold start-up light exposure for stating works to be presented, according to the matching value
Distribute each works to be presented corresponding flow, scheme provided by the present application has taken into account the nominal exposure amount and use of works to be presented
Family experience recommends works to be presented for user's request data for it, and as each works distribution to be presented is interesting to its
Customer flow, so that the assignment of traffic to each works to be presented is more rationally and accurate;Be conducive to high-quality works to be mined out
Come, and then works creator is motivated to share out by more high-quality works, realizes multi-party benign cycle.
It should be understood that above general description and following detailed description be only it is exemplary and explanatory, not
The application can be limited.
Detailed description of the invention
The drawings herein are incorporated into the specification and forms part of this specification, and shows the implementation for meeting the application
Example, and together with specification it is used to explain the principle of the application.
Fig. 1 is a kind of flow chart of flow allocation method shown according to an exemplary embodiment.
Fig. 2 is the flow chart of the step S130 shown in FIG. 1 provided according to an exemplary embodiment.
Fig. 3 is shown according to an exemplary embodiment to convert multiple works subsets for the works set of millions dimension
Form, to each subset respectively carry out matching solution process schematic.
Fig. 4 is the block diagram of flow distribution device shown according to an exemplary embodiment.
Fig. 5 is the block diagram of a kind of electronic equipment shown according to an exemplary embodiment.
Specific embodiment
Example embodiments are described in detail here, and the example is illustrated in the accompanying drawings.Following description is related to
When attached drawing, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements.Following exemplary embodiment
Described in embodiment do not represent all embodiments consistent with the application.On the contrary, they be only with it is such as appended
The example of some aspects be described in detail in claims, the application consistent device and flow allocation method.
Fig. 1 is a kind of flow chart of flow allocation method shown according to an exemplary embodiment, as shown in Figure 1, flow
Distribution method is in server, comprising the following steps:
S110: the specified cold start-up light exposure of each works to be presented is obtained.
The cold start-up works referred in the application are and are not shown in the new works in face of the public, therefore claim the cold start-up
Works are works to be presented, the current information of each works to be presented are loaded and count, to obtain required for the works to be presented
Nominal exposure amount.In a kind of embodiment, the contract information of works to be presented is obtained according to current information, the contract information is parsed and obtains
The corresponding specified starting light exposure of the works to be presented.
S120 estimates interaction data according to what the user behavior data generated in real time obtained each works to be presented.
In a kind of embodiment, transfers feed on line and flow log, feed stream is the orderly diffusible message pushed in real time
Stream parses feed on the line and flows log acquisition user behavior data, carries out classification processing to the user behavior data, utilize
Neural network even depth learning algorithm establishes the model of different behavioral datas, acquires the behavioral data of user in real time, continues to optimize
The model of the difference behavioral data, so that obtain works to be presented according to the different behavioral data model estimates interaction
Data, improve and obtain the accuracy that works to be presented estimate interaction data, and each works to be presented estimate interaction data
It include: pCTR (predicted click through rate, the clicking rate estimated): pLTR (predicted like
Through rate, that estimates thumbs up rate), pFTR (predicted follow through rate, the concern rate estimated).
Accordingly, it establishes and estimates clicking rate model, estimates and thumb up rate model, estimate concern rate model, estimated according to after optimization
Clicking rate model is estimated and thumbs up rate model, estimates concern rate model, and estimating for each works to be presented can be quickly and accurately obtained
Clicking rate estimates the rate of thumbing up and estimates concern rate, is conducive to the subsequent expectation displaying value for each works and obtains the works pair
Answer the speed of flow.
In a kind of embodiment, works and number of users to be presented are huge, occupied by the data bulk and data for needing to acquire
Memory space it is big, collected user related data and works related data to be presented are converted to low latitudes by the present embodiment
Embeding layer (embedding) indicates, is conducive to more efficiently estimate concern for what user behavior data resolved to works to be presented
Template.
S130, under the conditions of global gain is maximum, obtain preset time period in user send request data, it is described to
The maximum matching value for estimating interaction data and the specified cold start-up light exposure for showing works is distributed according to the matching value
The corresponding flow of each works to be presented.
The global gain maximum refers in the case where meeting each works nominal exposure amount to be presented, fully considers every
A works to be presented estimate interaction data, and the interaction data of estimating can characterize user to the hobby of the works to be presented,
The maximum return of allocation plan is obtained according to the feature for the request data for estimating interaction data and user's transmission.The present embodiment is taken into account
The specified cold start-up light exposures of user preferences and each works to be presented, it is contemplated that the interests of works provider and user side,
And the data model of big data foundation is utilized, the data of acquisition are more accurate and timely.
Assignment of traffic scheme provided by the embodiments of the present application maximises while providing cold start-up flow for new works
User experience keeps the distribution for being cold-started flow more accurate, is conducive to that author end is motivated to generate more works, more high-quality works
It is mined out by being cold-started flow.The situation of Profit for optimizing flow distributing system totality can utmostly optimize flow money
The utilization efficiency and value in source.
In a kind of embodiment, the flow diagram of step S130 is as shown in Fig. 2, described in the maximum condition of global gain
Under, obtain the request data, the works to be presented that user sends in preset time period estimates interaction data and the volume
Surely the step of being cold-started the maximum matching value of light exposure, comprising:
S210, according to the expectation displaying value estimated interaction data and determine each works.
Estimate interaction data using what step S120 obtained each works to be presented, according to it is described estimate interaction data determine it is every
The expectation displaying value w of a works to be presentedij, wijIndicate expectation displaying value of the user to works j in i-th request, the expectation
Displaying value characterizes hobby of the user to the expectation degree of each works to be presented and user to the works, described in a kind of embodiment
Expect displaying value is indicated in the form of score value, it is described expect displaying value can by each works to be presented estimate clicking rate, in advance
Estimate the rate of thumbing up and estimate the acquisition of concern rate, such as estimating each works to be presented clicking rate, estimate the rate of thumbing up, estimate concern rate phase
Add and obtain corresponding expectation displaying value, or will estimate that clicking rate is corresponding with the rate of thumbing up is estimated to be estimated click volume and thumb up with estimating
Amount, which is multiplied, obtains the expectation displaying value etc. of corresponding works.
S220, in the case where the light exposure of each works to be presented reaches the constraint condition of specified cold start-up light exposure, based on described
The expectation displaying value and its displaying probability of each works to be presented obtain global gain.
The expectation displaying value characterization user of the works to be presented is meeting each work to be presented to the fancy grade of the works
In the case where the specified cold start-up light exposure of product, the displaying probability by adjusting each works obtains different global gains.For every
A works, in the case where meeting its constraint condition, based on user to the corresponding displaying probability of expectation displaying value distribution of the works, at this time
The global gain of acquisition is maximum.
In a kind of embodiment, the global gain is to be shown by calculating the expectation displaying value of each works to be presented with it
The product of probability adds up the result of product of each works to be presented.In a kind of embodiment, based on user to the works
The distribution of expectation displaying value it is corresponding show probability, other influence factor can also be added, such as: the appreciation level of user, root
Displaying value, quantized value of user's appreciation level etc., which calculate global gain, to be expected to the works according to user.
In preset time period, the request data distribution that user sends keeps stablizing, such as in 5 minutes, 10 minutes, Yong Hufa
Acute variation will not occur for the request data distribution sent.It can refer in stable fixed time period in the preset time period,
The variable time period for obtaining default request data can also be characterized, such as preset time period refers to acquisition nearest 60,000 or 100,000
In matching solves the period drastic change does not occur for the period of secondary request data, in the case, the distribution of user's request data, can
With the request data of the user of realization truly and the On-line matching of corresponding works.
Since interconnection user on the network's dimension is in hundred million ranks, in ten million rank, the application is mentioned works quantity dimension to be presented
For in a kind of embodiment, using whole works to be presented as a works set, using default partitioning scheme by the works
Set-partition is at multiple works subsets, wherein the works that each works subset receives in the preset time period show request
Quantity is identical;Each works subset is obtained under global gain maximal condition most using the scheme that step S110 to S130 is provided
Big matching value, and then obtain the optimum flow distribution of entire works set.Scheme provided in this embodiment is by number in terms of necessarily
Works to be presented are split, so that the application has very strong scalability, and can be coped with more on a large scale by machine expansion
The problem of, such as: Rapid matching is carried out using the request data obtained in nearest a period of time, and works to be presented are divided
The matching problem of hundred million grades of users and ten million works to be presented are converted into multiple 100,000 requests and 60,000 to be presented by means, the application
The subproblem of works, respectively solves subproblem, the final best match for obtaining hundred million grades of users with ten million works to be presented.
Scheme provided in this embodiment says that the operation of global high-volume data is converted to the processing of multiple local small lot data, reduces
Calculation scale and computation complexity reduce the system resource that calculating process occupies.Scheme provided in this embodiment: by it is whole to
Show that the works set as a works set, is divided into multiple works subsets using default partitioning scheme, obtained by works
Maximum matching value of each works subset under global gain maximal condition is obtained, and then obtains the optimum flow of entire works set
Distribution, in a kind of embodiment, the process schematic of the program is as shown in figure 3, convert the works set of millions dimension to more
A 60,000 works subset to be presented, carries out matching solution to each subset respectively, and the n in Fig. 3 is the positive integer greater than 1, such as: whole
The works amount of a works set is 600,000, then n is 9, and the matching solution procedure of each subset is as follows: obtaining the pre- of works to be presented
Interaction data is estimated, according to the expectation displaying value estimated interaction data and determine each works to be presented, such as above-described embodiment institute
It states, expects that displaying value can carry out score value by that will estimate interaction data, and by the interaction data of estimating with preset algorithm
(such as by each works to be presented estimate clicking rate, estimate the rate of thumbing up, estimate concern rate be added) calculate, in the form of score value
It indicates;Offline is carried out using the request data received in the expectation displaying value and preset time period of each works to be presented
Match, such as 100,000 obtained recently time request data based on real-time update is matched offline, should be from by optimization algorithm optimization
Lines matching process obtains the corresponding dual variable α of each works to be presented;In conjunction with the expectation displaying value of each works to be presented and right
Even variable α carries out the On-line matching for the request data being currently received, and shows that the sum of probability is 1 constraint condition in works
Under, the corresponding dual variable β of the request data is obtained, works to be presented are corresponding with works by the dual variable β of request data
Dual variable α is determined, is determined according to the corresponding dual variable of determining request data and works and is recommended works to be presented, it is described to
Show the possible more than one of works, accordingly, each works to be presented are obtained with the optimum flow point of user's request data characterization
Match.
S230 obtains the request data, the work to be presented that user sends in preset time period when global gain maximum
The maximum matching value for estimating interaction data and the specified cold start-up light exposure of product.
In a kind of embodiment, the result of product of each works to be presented is being subjected to cumulative the step of obtaining global gain
Later, further includes: optimize the accumulation result so that the global gain of accumulation result characterization is maximum.
In a kind of embodiment, estimating for the request data, the works to be presented that user sends in preset time period is obtained
The matching value of interaction data and the specified cold start-up light exposure, in the matching value maximum, according to the maximum matching
Value distributes each works to be presented corresponding flow, and the global gain is maximum, refers to and is guaranteeing that each works to be presented can
In the case where obtaining specified cold start-up light exposure, according to user to the matching value point for expecting displaying value and works of works to be presented
With flow, there is infinite combinations, each combination in the other works to be presented of the user of hundred million ranks and millions in dimension between the two
A matching value is corresponded to, but the application finds maximum matching value, according to the corresponding combination of maximum matching value, is embodied as every time
User's request data shows the works most to match therewith, and the flow for each works distribution to be presented is the most accurate, this implementation
The scheme that example provides, which can use following optimization problem, indicates that the optimization problem is the optimization problem of belt restraining, and constraint condition is
Each works to be presented can obtain specified cold start-up light exposure, and optimization aim is the flow for each works distribution to be presented
Global gain is maximum, and as each works to be presented find top quality customer flow, and works to be presented, which are showed in, it
The user terminal of interest can be write as following form:
Wherein, subscript i indicates request, and subscript j indicates works to be presented, wijIndicate user and works j in i-th request
Expect that displaying value, the calculation method for expecting displaying value can be estimated as described in above-described embodiment based on deep learning
PCTR (estimating clicking rate) etc. is calculated.xijIndicate whether i-th request shows works j, riIndicate the return number of i-th request,
The works quantity shown, djIndicate the specified cold start-up light exposure of works j.Problem has had changed into request side and works at this time
The bipartite graph matching problem of side utilizes max-flow or the Hungarian Method bipartite graph matching problem, wijFor bipartite graph side
On weight.X hereinijFor 0-1 discrete variable, solution is more difficult, and the objective function of original optimization problem is xijIt is primary
Function, the process for being iterated solution are complicated.For this problem, the application does following setting: xijAs continuous probability point
Cloth, and secondary regular terms is added in objective function, these settings make the strong convex function of objective function, the strong convex function
It is as follows:
Wherein, pjIndicate works probability and d to be presentedjRelated, λ is the regular coefficient of regular terms, at this time objective function
Have been converted into strong convex function.Direct solution original optimization problem, obtained only one group of offline xijSolution, while direct solution
Convex optimization problem needs to use convex optimized algorithm and solves lib (outside relies on), and solving speed is slow, and system dependence is also compared
Weight.So primal problem is finally converted into Lagrange duality by the application, solved by way of iteration of interlocking, Lagrange
Antithesis dual equation is as follows:
Wherein, βiIt is the dual variable of flow side, αjIt is the dual variable of works side, γijIt is probability xijDual variable.
Suitable dual variable value is iterated to calculate, by the analysis of the K.K.T condition to corresponding dual problem, dual variable is brought into
Former problem further derives final allocation probability parameter: fixed α acquires the β of the condition of satisfaction, and then fixing Beta, which acquires, meets item
The α of part, iterative solution.X in the present embodimentijMeet following condition:
xij=max (0, (αj-βi+wij)/λ+pj)
Wherein, α is the dual variable of works side, unrelated with flow, and β is the dual variable of flow side, is had with current request
It closes.The β for determining request for 1 can be summed it up according to probability when fixed α.Fixing Beta can be solved according to the traffic constraints condition of works
To α.After offline solution obtains α and β, temporarily only retain the state α of works side.According to the user of present flow rate when servicing on line
With the dual variable α of works pCTR value and works, dichotomy solves the β value of present flow rate.When receiving request data, then foundation
X in above formulaijExpression formula, determine that current data requests the β of corresponding flow according to probability adduction for 1i, and then solve and currently ask
The probability distribution x for the works to be presented askedi.The user characteristics are directed to, the displaying for obtaining the works to be presented of current request is general
Rate.
The global gain maximum can guarantee the case where each works to be presented can obtain specified cold start-up light exposure
Under, the matching value of displaying value and works, which distributes flow, to be expected to works to be presented according to user, is provided using above-described embodiment
Scheme obtain maximum global gain, corresponding maximum matching value when maximum global gain is obtained, according to the maximum matching value
It distributes each works to be presented corresponding flow, shows the works to be presented to match for each request data.
In a kind of embodiment, the acquisition for dual variable α, can use PID control, (ratio, integral, differential control are adjusted
Section) technology, according to the dual variable α of the cold start-up flow growth rate dynamic regulation works side of works to be presented, simulation matching
The process of solution, scheme provided in this embodiment reduce the complicated journey of calculating compared to the matching solution procedure of above-described embodiment
Degree, reduces the resource occupation of calculating process.
In a kind of embodiment, after the probability distribution for obtaining specific works to be presented, can be generated using reservoir sampling plus
Power does not put back to the result (sample without replacement) of sampling, obtains final output.Probability distribution is put back to
In database for accommodating limited probability distribution, constantly the probability data of acquisition is put into, when reaching the receiving upper limit,
It determines whether the probability distribution is put into the database with preset condition, recycles the process for being constantly put into, replacing.
User is expected displaying value as a part of objective function in optimization problem, with the volume of works to be presented by the application
Surely cold start-up light exposure is constraint condition, has taken into account the light exposure of user experience and works to be presented, while using optimization algorithm
The optimal case of objective function is obtained, so that assignment of traffic is more rationally and accurate.
The application uses SIMD+AVX (single-instruction multiple-data stream (SIMD)), this technology will be by base in practice process
Plinth Video coding and transcoding performance promote 2 to 4 times, improve the speed of coding and transcoding, can be realized 60,000 using this technology
PCTR, pLTR, pFTR data of works to be presented are estimated and can be completed in 20ms, and the solution of matching process can be at 10 minutes
Interior completion realizes that the process whole process of processing request is unimpeded.In a kind of embodiment, the partial data in the application is used
The advanced data structures such as CuckooHash (cuckoo hash), occupied space is few, and inquiry velocity is fast.
Fig. 4 is a kind of flow distribution device block diagram shown according to an exemplary embodiment.Referring to Fig. 4, the assignment of traffic
Device includes obtaining light exposure module 410, obtain data module 420 and obtaining assignment of traffic module 430.
Light exposure module 410 is obtained, is configured as obtaining the specified cold start-up light exposure of each works to be presented;
Data module 420 is obtained, is configured as obtaining the pre- of each works to be presented according to the user behavior data on line
Estimate interaction data;
Assignment of traffic module 430 is obtained, is configured as under the conditions of global gain is maximum, obtains in preset time period and uses
Maximum for estimating interaction data and the specified cold start-up light exposure of request data, the works to be presented that family is sent
With value, distribute each works to be presented corresponding flow according to the matching value.
In a kind of embodiment, obtains assignment of traffic module 430 and comprise determining that expectation shows value cell, computing unit, matching
Unit.
The determining expectation shows value cell, is configured as estimating the expectation that interaction data determines each works according to
Displaying value;The computing unit is configured as under the constraint condition that the light exposure of each works reaches specified cold start-up light exposure,
Expectation displaying value and its displaying probability based on each works to be presented obtain global gain;The matching unit is matched
It is set to when global gain maximum, obtains estimating for the request data, the works to be presented that user sends in preset time period
The maximum matching value of interaction data and the specified cold start-up light exposure.
About the flow distribution device in above-described embodiment, wherein modules, unit have executed the concrete mode of operation
Through being described in detail in the embodiment in relation to the flow allocation method, no detailed explanation will be given here.
Fig. 5 is a kind of block diagram of electronic equipment 500 for assignment of traffic shown according to an exemplary embodiment.Example
Such as, electronic equipment 500 may be provided as a server.Referring to Fig. 5, electronic equipment 500 includes processing component 522, into one
Step includes one or more processors, and the memory resource as representated by memory 532, and being used to store can be by processing component
The instruction of 522 execution, such as application program.The application program stored in memory 532 may include one or more
Each correspond to one group of instruction module.In addition, processing component 522 is configured as executing instruction, to execute above-mentioned flow
Distribution method.
Electronic equipment 500 can also include that a power supply module 526 is configured as executing the power supply pipe of electronic equipment 500
Reason, a wired or wireless network interface 550 are configured as electronic equipment 500 being connected to network and an input and output (I/
O) interface 558.Electronic equipment 500 can be operated based on the operating system for being stored in memory 532, such as Windows
ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or similar.
A kind of computer program product, including computer program code, the computer program include program instruction, work as institute
When stating program instruction and being computer-executed, the computer is made to execute above-mentioned flow allocation method.
Those skilled in the art consider specification and practice it is disclosed herein after, will readily occur to other implementations of the application
Scheme.This application is intended to cover any variations, uses, or adaptations of the application, these modifications, purposes or adaptations
Property variation follow the general principle of the application and including the undocumented common knowledge in the art of the application or usual
Technological means.The description and examples are only to be considered as illustrative, and the true scope and spirit of the application are wanted by following right
It asks and points out.
It should be understood that the application is not limited to the precise structure that has been described above and shown in the drawings, and
And various modifications and changes may be made without departing from the scope thereof.Scope of the present application is only limited by the accompanying claims.
Claims (10)
1. a kind of flow allocation method characterized by comprising
Obtain the specified cold start-up light exposure of each works to be presented;
Interaction data is estimated according to what the user behavior data generated in real time obtained each works to be presented;
Under the conditions of global gain is maximum, the request data, the works to be presented that user sends in preset time period are obtained
The maximum matching value for estimating interaction data and the specified cold start-up light exposure, according to the matching value distribute each wait open up
It is shown as the corresponding flow of product.
2. flow allocation method according to claim 1, which is characterized in that it is described under the conditions of global gain is maximum,
Obtain the request data, the works to be presented that user sends in preset time period estimates interaction data and described specified cold
The step of starting the maximum matching value of light exposure, comprising:
According to the expectation displaying value estimated interaction data and determine each works to be presented;
In the case where the light exposure of each works to be presented reaches the constraint condition of specified cold start-up light exposure, based on described each to be presented
The expectation displaying value and its displaying probability of works obtain global gain;
When global gain maximum, estimating for the request data, the works to be presented that user sends in preset time period is obtained
The maximum matching value of interaction data and the specified cold start-up light exposure.
3. flow allocation method according to claim 2, which is characterized in that described based on each works to be presented
Expect displaying value and its show the step of probability obtains global gain, comprising:
The expectation displaying value for calculating each works to be presented shows the product of probability with it, by the product of each works to be presented
As a result cumulative acquisition global gain is carried out.
4. flow allocation method according to claim 3, which is characterized in that the product by each works to be presented
As a result after the step of carrying out cumulative acquisition global gain, further includes:
Optimize the accumulation result so that the global gain of accumulation result characterization is maximum.
5. flow allocation method according to claim 4, which is characterized in that the optimization
The accumulation result is so that the global gain maximum that the accumulation result characterizes, description are as follows:
Wherein, subscript i indicates request, and subscript j indicates works to be presented, wijIndicate the expectation of user and works j in i-th request
Displaying value, xijIndicate whether i-th request shows works j, djIndicate the specified cold start-up light exposure of works j, pjIndicate works
Probability to be presented, λ are the regular coefficients of regular terms.
6. flow allocation method according to claim 4, which is characterized in that the optimization accumulation result is so that described
The step of the global gain maximum of accumulation result characterization, comprising:
Expect that the displaying probability of displaying value and the works meets in advance with works to be presented described in Lagrange duality algorithm description
If condition, the corresponding dual variable of each works to be presented is obtained by iteration of interlocking;
Optimize the accumulation result so that global gain is maximum based on the corresponding dual variable of each works.
7. flow allocation method according to claim 6, which is characterized in that described to be based on the corresponding dual variable of each works
Optimize the accumulation result so that global gain is maximum, description are as follows:
Wherein, subscript i indicates request, and subscript j indicates works to be presented, wijIndicate the expectation of user and works j in i-th request
Displaying value, xijIndicate whether i-th request shows works j, djIndicate the specified cold start-up light exposure of works j, pjIndicate works
Probability to be presented, λ are the regular coefficient of regular terms, βiIt is the dual variable of flow side, αjIt is the dual variable of works side,
γijIt is probability xijDual variable.
8. a kind of flow distribution device characterized by comprising
Light exposure module is obtained, is configured as obtaining the specified cold start-up light exposure of each works to be presented;
Data module is obtained, be configured as obtaining each works to be presented in real time according to the user behavior data that generates estimates friendship
Mutual data;
Assignment of traffic module is obtained, is configured as under the conditions of global gain is maximum, user in preset time period is obtained and sends
Request data, the works to be presented the maximum matching value for estimating interaction data and the specified cold start-up light exposure,
Distribute each works to be presented corresponding flow according to the matching value.
9. a kind of electronic equipment characterized by comprising
Processor;
Memory for storage processor executable instruction;
Wherein, the processor is configured to flow allocation method as described in any one of claims 1 to 7 claim
Step.
10. a kind of non-transitorycomputer readable storage medium, when the instruction in the storage medium is by the processing of electronic equipment
When device executes, so that the step of electronic equipment is able to carry out the flow allocation method as described in any one of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811102772.0A CN109389424B (en) | 2018-09-20 | 2018-09-20 | Flow distribution method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811102772.0A CN109389424B (en) | 2018-09-20 | 2018-09-20 | Flow distribution method and device, electronic equipment and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109389424A true CN109389424A (en) | 2019-02-26 |
CN109389424B CN109389424B (en) | 2021-08-06 |
Family
ID=65418835
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811102772.0A Active CN109389424B (en) | 2018-09-20 | 2018-09-20 | Flow distribution method and device, electronic equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109389424B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111047009A (en) * | 2019-11-21 | 2020-04-21 | 腾讯科技(深圳)有限公司 | Event trigger probability pre-estimation model training method and event trigger probability pre-estimation method |
CN111192602A (en) * | 2019-12-03 | 2020-05-22 | 广州荔支网络技术有限公司 | White noise audio content value evaluation method based on audio content portrait system |
CN112214665A (en) * | 2019-07-12 | 2021-01-12 | 北京字节跳动网络技术有限公司 | Content display method and device, electronic equipment and computer readable storage medium |
CN112288233A (en) * | 2020-09-29 | 2021-01-29 | 南京大学 | Network taxi appointment order distribution method suitable for time-varying passenger waiting time distribution |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104299151A (en) * | 2014-09-28 | 2015-01-21 | 新浪网技术(中国)有限公司 | Online advertising flow distribution system and method |
CN105227331A (en) * | 2015-10-16 | 2016-01-06 | 上海晶赞科技发展有限公司 | Information sending control method and device |
US20160225015A1 (en) * | 2015-01-30 | 2016-08-04 | Wal-Mart Stores, Inc. | System for optimizing sponsored product listings for seller performance in an e-commerce marketplace and method of using same |
US20160358208A1 (en) * | 2014-02-28 | 2016-12-08 | Tencent Technology (Shenzhen) Company Limited | Method, device, system for displaying media data |
CN106339897A (en) * | 2016-08-18 | 2017-01-18 | 腾讯科技(深圳)有限公司 | Putting strategy determination method and device |
CN107526810A (en) * | 2017-08-23 | 2017-12-29 | 广州优视网络科技有限公司 | Establish method and device, methods of exhibiting and the device of clicking rate prediction model |
-
2018
- 2018-09-20 CN CN201811102772.0A patent/CN109389424B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160358208A1 (en) * | 2014-02-28 | 2016-12-08 | Tencent Technology (Shenzhen) Company Limited | Method, device, system for displaying media data |
CN104299151A (en) * | 2014-09-28 | 2015-01-21 | 新浪网技术(中国)有限公司 | Online advertising flow distribution system and method |
US20160225015A1 (en) * | 2015-01-30 | 2016-08-04 | Wal-Mart Stores, Inc. | System for optimizing sponsored product listings for seller performance in an e-commerce marketplace and method of using same |
CN105227331A (en) * | 2015-10-16 | 2016-01-06 | 上海晶赞科技发展有限公司 | Information sending control method and device |
CN106339897A (en) * | 2016-08-18 | 2017-01-18 | 腾讯科技(深圳)有限公司 | Putting strategy determination method and device |
CN107526810A (en) * | 2017-08-23 | 2017-12-29 | 广州优视网络科技有限公司 | Establish method and device, methods of exhibiting and the device of clicking rate prediction model |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112214665A (en) * | 2019-07-12 | 2021-01-12 | 北京字节跳动网络技术有限公司 | Content display method and device, electronic equipment and computer readable storage medium |
CN111047009A (en) * | 2019-11-21 | 2020-04-21 | 腾讯科技(深圳)有限公司 | Event trigger probability pre-estimation model training method and event trigger probability pre-estimation method |
CN111047009B (en) * | 2019-11-21 | 2023-05-23 | 腾讯科技(深圳)有限公司 | Event trigger probability prediction model training method and event trigger probability prediction method |
CN111192602A (en) * | 2019-12-03 | 2020-05-22 | 广州荔支网络技术有限公司 | White noise audio content value evaluation method based on audio content portrait system |
CN112288233A (en) * | 2020-09-29 | 2021-01-29 | 南京大学 | Network taxi appointment order distribution method suitable for time-varying passenger waiting time distribution |
CN112288233B (en) * | 2020-09-29 | 2023-09-12 | 南京大学 | Network vehicle order allocation method suitable for time-varying passenger waiting time distribution |
Also Published As
Publication number | Publication date |
---|---|
CN109389424B (en) | 2021-08-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US20200293838A1 (en) | Scheduling computation graphs using neural networks | |
US11531926B2 (en) | Method and apparatus for generating machine learning model by using distributed computing framework | |
Zhang et al. | Parallel and reliable probabilistic load forecasting via quantile regression forest and quantile determination | |
CN109389424A (en) | Flow allocation method, device, electronic equipment and storage medium | |
Paliwal et al. | Reinforced genetic algorithm learning for optimizing computation graphs | |
Kaur et al. | Deep‐Q learning‐based heterogeneous earliest finish time scheduling algorithm for scientific workflows in cloud | |
US11941527B2 (en) | Population based training of neural networks | |
US12008445B2 (en) | Black-box optimization using neural networks | |
CN110633796B (en) | Model updating method and device, electronic equipment and storage medium | |
WO2019084560A1 (en) | Neural architecture search | |
JP2007317068A (en) | Recommending device and recommending system | |
CN110663049A (en) | Neural network optimizer search | |
CN112257868A (en) | Method and device for constructing and training integrated prediction model for predicting passenger flow | |
US20240107125A1 (en) | System and method for modelling access requests to multi-channel content sharing platforms | |
Genov et al. | Forecasting flexibility of charging of electric vehicles: Tree and cluster-based methods | |
CN116664232A (en) | Federal personalized commodity recommendation method based on contextual slot machine | |
WO2015158224A1 (en) | Message recommendation method and apparatus | |
Dreuning et al. | mCAP: Memory-Centric Partitioning for Large-Scale Pipeline-Parallel DNN Training | |
CN111027709B (en) | Information recommendation method and device, server and storage medium | |
KR102067013B1 (en) | Device for collective machine learning of things and method thereof | |
Liu et al. | A New Parallel Item-Based Collaborative Filtering Algorithm Based on Hadoop. | |
CN114385359B (en) | Cloud edge task time sequence cooperation method for Internet of things | |
CN117635214A (en) | Revenue prediction method and device, electronic equipment and storage medium | |
Lu et al. | Non-stationary dueling bandits for online learning to rank | |
WO2023195011A1 (en) | System and method for model training in decentralized computing environments |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |