CN106803194A - Online competitive price probabilistic model generation method - Google Patents
Online competitive price probabilistic model generation method Download PDFInfo
- Publication number
- CN106803194A CN106803194A CN201710041394.9A CN201710041394A CN106803194A CN 106803194 A CN106803194 A CN 106803194A CN 201710041394 A CN201710041394 A CN 201710041394A CN 106803194 A CN106803194 A CN 106803194A
- Authority
- CN
- China
- Prior art keywords
- kpi
- interval
- price
- variance
- average
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0273—Determination of fees for advertising
- G06Q30/0275—Auctions
-
- 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
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/08—Auctions
Landscapes
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Strategic Management (AREA)
- Engineering & Computer Science (AREA)
- Development Economics (AREA)
- Marketing (AREA)
- Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Game Theory and Decision Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a kind of online competitive price probabilistic model generation method, by the request won in history record, KPI points according to prediction is interval for several, and each interval is fitted using mixing beta distributions, is fitted the function of each interval KPI and winning price, model is corrected, probabilistic model is finally given, online competitive price can be emulated and simulated by the probabilistic model, be easy to the discovery can problem analysis, explore the bidding strategy of best service configuration, analysis and simulation rival.
Description
Technical field
The invention belongs to Internet technical field, and in particular to a kind of online competitive price probabilistic model generation method.
Background technology
Real time bid (RealTime Bidding, abbreviation RTB) is a kind of utilization third party technology millions of
The technology of bidding for being estimated and bidding for each user displaying behavior on website.The frequency is delivered with making a big purchase in large quantities not
Together, real time bid has been evaded invalid audient and has been reached, and is bought just for significant user.Its core is DSP platform
(party in request's platform), RTB can bring more advertisement sales volumes, realize sales process automation and lower for media
The expenditure of general expenses.And for advertiser and agency, most direct benefit is exactly to improve effect and invest back
Report rate.The appearance of RTB is exactly that Internet advertising enters crowd's real time bid epoch from the advertisement position epoch.
The behavior of online competitive price is an extremely complex problem, in the industry cycle there is no effective model with academia at present.
Although in the real time bid system of reality, there is substantial amounts of data accumulation daily, how effective information is therefrom refined
It is a problem for difficulty to meet actual model for generation.And the model of online competitive price is excellent for control advertiser's cost
Change, and maximize the effect of advertiser's advertisement putting, such as maximized under fixed budget and represent, click on, reach and convert,
Or maximization is delivered the different measurement effects such as crowd's amount of covering and has important meaning.
The content of the invention
The present invention is carried out to solve the above problems, it is therefore intended that providing one kind can produce KPI and triumph valency
Relation between lattice, can produce the online competitive price probabilistic model generation method of the probability distribution of KPI.
The invention provides a kind of online competitive price probabilistic model generation method, it is characterised in that comprise the following steps：Step
1, input is bidded and successfully ask record as sample；
Step 2, the KPI that every request record is predicted based on pre-defined rule, by the sample according to
The KPI of prediction point is interval for several, count the average of the KPI of the statistics in each interval, the variance of KPI, winning price it is equal
The variance of value and winning price, and count the distribution of each interval KPI and the distribution of winning price；
Step 3, using mixing beta fittings of distribution, each interval sample data obtains model profile, and obtains each area
Between calculating KPI average and the variance of KPI；
Step 4, is fitted each interval KPI and winning price obtains pattern function；
Step 5, regenerates each interval one group and meets the new samples of model profile, and unite according to the model profile
The average and variance of the KPI of new samples, the winning price that new samples are obtained according to pattern function are counted, and statistics obtains triumph valency
The average and variance of lattice；
Step 6, judges whether the average and variance of the KPI of new samples in step 5, the average of winning price and variance approach
The variance of the average of the KPI of statistics, the variance of KPI, the average of winning price and winning price, if approached, enters in step 2
Enter next step, if it is not, then return to step 3, and change the quantity of beta distributions in the mixing beta distributions；
Step 7, judge the KPI of new samples in step 5 distribution and winning price distribution whether sample in approximation step 2
The distribution of middle KPI and the distribution of winning price, if approached, model construction is completed, otherwise, return to step 2, and change sample
Interval quantity.
Further, in the online competitive price probabilistic model generation method that the present invention is provided, can also have the feature that：
Wherein, request record is comprising the click behavior before natural quality, medium property, user, bid, knock-down price, click and turns
Change..
Further, in the online competitive price probabilistic model generation method that the present invention is provided, can also have the feature that：
Wherein, KPI is included：Click on cost, click volume, conversion cost, inversion quantity, CTR, CVR.
Further, in the online competitive price probabilistic model generation method that the present invention is provided, can also have the feature that：
Wherein, in step 2, also need to count each interval sample size, if the quantity of single interval sample is less than threshold value,
Needs merge interval with other intervals, and update the average of KPI of the interval statistics after merging, the variance of KPI,
The average of winning price and the variance of winning price, and update the distribution of each interval KPI and dividing for winning price after merging
Cloth.
Advantages of the present invention is as follows：
According to online competitive price probabilistic model generation method involved in the present invention, by the request won in history record, root
It is predicted that KPI point it is interval for several, each interval is fitted using mixing beta distributions, be fitted each interval KPI
With the function of winning price, model is corrected, finally gives probabilistic model, can be to online competitive price by the probabilistic model
Emulated and simulated, be easy to discovery can problem analysis, explore best service configuration, analysis and simulation rival bid plan
Slightly.
Brief description of the drawings
Fig. 1 is the flow chart of online competitive price probabilistic model generation method in the present invention；
Fig. 2 is KPI distributions in the present invention in sample and sample data according to mould
The KPI profiles versus figure of type distribution function generation；
Fig. 3 is the triumph valency that the distribution of the winning price in sample and sample data are generated according to pattern function in the present invention
The comparison diagram of the distribution of lattice.
Specific embodiment
In order that technological means, creation characteristic, reached purpose and effect that the present invention is realized are easy to understand, it is real below
Example combination accompanying drawing is applied to be specifically addressed online competitive price probabilistic model generation method of the present invention.
Before described, the technical term that this patent is related to is illustrated：
KPI：KPI Key Performance Indicator；
CTR：Clicking rate；
CVR：Click on conversion ratio；
pdf：Probability density.
Online competitive price probabilistic model generation method can generate probabilistic model of bidding, the probabilistic model of bidding comprising KPI and
Relation between winning price, the probability distribution of KPI.
As shown in figure 1, in the present embodiment, online competitive price probabilistic model generation method is comprised the steps of：
Step S1, input is bidded and successfully ask record as sample.
One request is recorded as a sample data.Substantial amounts of data are included in sample.Request record comprising natural quality,
Click behavior, bid, knock-down price, click and conversion before medium property, user.Natural quality is included and once bidded (pv)
Time, medium property includes web page address (url), and the click behavior before user is that how much user clicked on certain advertisement in one week
It is secondary.
Step S2, the KPI that every request record is predicted based on pre-defined rule, the KPI according to prediction will be all of
Request record is divided into several intervals, counts average, the variance of KPI, the average of winning price of the KPI of the statistics in each interval
With the variance of winning price, and the distribution of each interval KPI and the distribution of winning price are counted.
KPI is included：Click on cost, click volume, conversion cost, inversion quantity, CTR or CVR.
In the present embodiment, by taking CTR as an example, the CTR of prediction uses existing LR (Logistic Regression) to KPI,
FM (Factorization Machines), GBDT scheduling algorithms are calculated.
Comprising multiple sample datas under the CTR of identical prediction.Sample is divided for several are interval according to the CTR of prediction,
Each interval is the CTR of the prediction of number range, counts the average of the actual KPI in each interval, the variance of KPI, obtains
Win the average of price and the variance of winning price, and count the quantity of each interval sample data.And count each interval
The distribution of KPI and the distribution of winning price.
If the quantity of certain interval sample data is less than threshold value, the interval is merged with other intervals, then
Again the side of average, the variance of KPI, the average of winning price and the winning price of the interval actual KPI after merging is counted
Difference.And the distribution and the distribution of winning price of the interval KPI after counting simultaneously.
Step S3, using mixing beta fittings of distribution, each interval sample data obtains model profile, such as Fig. 2, and obtains
To the average and the variance of KPI of the KPI of the calculating in each interval.
Each interval is distributed with several beta and is fitted, and the mixing beta distribution functions obtained according to fitting are calculated
Obtain the average of each interval KPI and the variance of KPI.
Step S4, is fitted each interval KPI and winning price obtains pattern function, such as Fig. 3.
Step S5, the model point is met by each interval according to regenerating one group in the model profile obtained in step S3
The new samples of cloth, and the average and variance of the KPI of the new samples are counted, by the new samples according to the model letter obtained in step S4
One group of winning price corresponding with the new samples that number is obtained, and count the average and variance for obtaining winning price.
Step S6, judges whether the average and variance of the KPI of new samples in step S5, the average of winning price and variance force
The variance of the average of the KPI of statistics, the variance of KPI, the average of winning price and winning price in nearly step S2, if approached,
Then enter next step, if it is not, then return to step S3, and change the quantity of beta distributions in mixing beta distributions.
Step S7, judges whether the distribution of the KPI of new samples in step S5 and the distribution of winning price approach the sample
The distribution of middle KPI and the distribution of winning price, if approached, model construction is completed, otherwise, return to step S2, and change sample
This interval quantity.
In Fig. 2, abscissa represents the value of CTR, and ordinate represents the probability density (pdf) of CTR, and dotted line is represented in sample
The probability distribution of CTR, solid line represents the probability distribution of the CTR that sample data is obtained according to model profile function.In Fig. 3, horizontal seat
Mark represents the value of CTR, and ordinate represents winning price, line band "." symbol represent sample in winning price distribution, line
The winning price distribution that the expression sample data of band " * " symbol is obtained according to pattern function.From figure 2 it can be seen that two songs
The line goodness of fit is relatively good, illustrates that the effect for mixing beta fittings of distribution is relatively good.In Fig. 3, although the CTR of prediction is in 0.05-0.1
Between real price in the pattern function winning price and the sample that calculate have a deviation, but the overall trend of data and sample
Data are identical.
Using the online competitive price probabilistic model, can be to emulating and simulating online competitive price process, convenient discovery and analysis are asked
Topic.Such as：Some ctr are higher, but can not but bid successfully, and that can just adjust bidding strategy, meaningless so as to avoid
Go to obtain some high cost flows.When needing adjustment bidding strategy, first can be simulated using this online competitive price probabilistic model,
Analog result meet it is expected after, then on line specifically carry out implementation operation, during simulating, business and unaffected on line.
Before online competitive price, tested repeatedly using this online competitive price probabilistic model, matched somebody with somebody so as to explore best service
Put.And can be according to this online competitive price Probability Analysis and the bidding strategy of simulation rival, so as to preferably formulate
The strategy of oneself.
Above-mentioned implementation method is preferred case of the invention, is not intended to limit protection scope of the present invention.
Claims (4)
1. a kind of online competitive price probabilistic model generation method, it is characterised in that including：
Step 1, input is bidded and successfully ask record as sample；
Step 2, the KPI that every request record is predicted based on pre-defined rule, by the sample according to the prediction
KPI point it is interval for several, count the average of the KPI of the statistics in each interval, the variance of KPI, the average of winning price and
The variance of winning price, and count the distribution of each interval KPI and the distribution of winning price；
Step 3, using mixing beta fittings of distribution, each interval sample data obtains model profile, and obtains each interval
The average of the KPI of calculating and the variance of KPI；
Step 4, is fitted each interval KPI and winning price obtains pattern function；
Step 5, regenerates each interval one group and meets the new samples of the model profile, and unite according to the model profile
The average and variance of the KPI of the new samples, the winning price that the new samples are obtained according to the pattern function are counted, and is united
Meter obtains the average and variance of winning price；
Step 6, judge the average and variance of the KPI of new samples in step 5, the average of winning price and variance whether approximation step
The variance of the average of the KPI of statistics, the variance of KPI, the average of winning price and winning price, if approached, enters described in 2
Enter next step, if it is not, then return to step 3, and change the quantity of beta distributions in the mixing beta distributions；
Step 7, judge the KPI of new samples in step 5 distribution and winning price distribution whether sample described in approximation step 2
The distribution of middle KPI and the distribution of winning price, if approached, model construction is completed, otherwise, return to step 2, and change described
The interval quantity of sample.
2. online competitive price probabilistic model generation method according to claim 1, it is characterised in that：
Wherein, the request record includes click behavior, bid, knock-down price, the point before natural quality, medium property, user
Hit and convert.
3. online competitive price probabilistic model generation method according to claim 1, it is characterised in that：
Wherein, the KPI is included：Click on cost, click volume, conversion cost, inversion quantity, CTR, CVR.
4. online competitive price probabilistic model generation method according to claim 1, it is characterised in that：
In step 2, also need to count each interval sample size, if the quantity of single interval sample is less than threshold value,
Needs merge interval with other intervals, and update the average of KPI of the interval statistics after merging, the variance of KPI,
The average of winning price and the variance of winning price, and update the distribution of each interval KPI and dividing for winning price after merging
Cloth.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710041394.9A CN106803194A (en) | 2017-01-20 | 2017-01-20 | Online competitive price probabilistic model generation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710041394.9A CN106803194A (en) | 2017-01-20 | 2017-01-20 | Online competitive price probabilistic model generation method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106803194A true CN106803194A (en) | 2017-06-06 |
Family
ID=58986972
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710041394.9A Pending CN106803194A (en) | 2017-01-20 | 2017-01-20 | Online competitive price probabilistic model generation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106803194A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108053101A (en) * | 2017-11-28 | 2018-05-18 | 晶赞广告（上海）有限公司 | For budget control method and device, storage medium, the terminal of party in request's platform |
| CN108985911A (en) * | 2018-08-15 | 2018-12-11 | 武汉万般上品信息技术有限公司 | A kind of aviation based on reversed price-bidding model is super to sell price competing method |
| CN109102321A (en) * | 2018-07-09 | 2018-12-28 | 北京奇艺世纪科技有限公司 | A kind of budget is breasted the tape prediction technique and device |
-
2017
- 2017-01-20 CN CN201710041394.9A patent/CN106803194A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108053101A (en) * | 2017-11-28 | 2018-05-18 | 晶赞广告（上海）有限公司 | For budget control method and device, storage medium, the terminal of party in request's platform |
| CN109102321A (en) * | 2018-07-09 | 2018-12-28 | 北京奇艺世纪科技有限公司 | A kind of budget is breasted the tape prediction technique and device |
| CN108985911A (en) * | 2018-08-15 | 2018-12-11 | 武汉万般上品信息技术有限公司 | A kind of aviation based on reversed price-bidding model is super to sell price competing method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Simonovska et al. | Trade models, trade elasticities, and the gains from trade | |
| TWI564829B (en) | Method and apparatus for generating traffic quality comparison parameters, advertising charging method | |
| De Menezes et al. | Review of guidelines for the use of combined forecasts | |
| Li et al. | A MIDAS modelling framework for Chinese inflation index forecast incorporating Google search data | |
| CN104077693B (en) | Commodity control methods, server, client and e-commerce system | |
| Kim et al. | Measuring the performance of large-scale combinatorial auctions: A structural estimation approach | |
| CN104035926B (en) | A kind of dispensing of internet information and system | |
| CN105205696A (en) | Internet advertisement real-time bidding multi-advertiser multi-factor flow distribution method | |
| CN110880127B (en) | Consumption level prediction method and device, electronic equipment and storage medium | |
| CN103309894B (en) | Based on search implementation method and the system of user property | |
| CN106296247A (en) | The online sort method of network information resource and device | |
| JP2012088994A (en) | Advertising effect analysis system and method of the same | |
| CN106803194A (en) | Online competitive price probabilistic model generation method | |
| CN106021341A (en) | A control method and device for sequencing presentation information in a user terminal | |
| CN106296287A (en) | For the method and apparatus adjusting the bid of promotion message | |
| Glynn et al. | Economic impacts of future changes in the energy system—global perspectives | |
| CN110717666A (en) | Market competitiveness evaluation method for enterprise products | |
| Song et al. | Evaluation model of click rate of electronic commerce advertising based on fuzzy genetic algorithm | |
| CN110866779A (en) | Data processing method, device, equipment and storage medium based on different channel advertisement delivery | |
| CN115587238A (en) | Product brand marketing big data service method | |
| KR20130054691A (en) | Method and server for advertisement by evaluation of user group | |
| CN112950279A (en) | Accurate marketing strategy model construction method and device based on machine learning | |
| Tian et al. | The Value of Information: Game Model of Supply Chain Based on C2M | |
| Yu et al. | Research on Real Estate Marketing Innovation System in the Era of Big Data | |
| Wu et al. | Efficient Ad-level Impression Forecasting based on Monotonicity and Sampling |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170606 |