CN107529654A - A kind of dynamic bid method towards displaying advertisement bidding jettison system - Google Patents

A kind of dynamic bid method towards displaying advertisement bidding jettison system Download PDF

Info

Publication number
CN107529654A
CN107529654A CN201710671569.4A CN201710671569A CN107529654A CN 107529654 A CN107529654 A CN 107529654A CN 201710671569 A CN201710671569 A CN 201710671569A CN 107529654 A CN107529654 A CN 107529654A
Authority
CN
China
Prior art keywords
advertising
price
rate
request
advertising campaign
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
Application number
CN201710671569.4A
Other languages
Chinese (zh)
Inventor
刘梦娟
曾贵川
吴书漫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN201710671569.4A priority Critical patent/CN107529654A/en
Publication of CN107529654A publication Critical patent/CN107529654A/en
Pending legal-status Critical Current

Links

Abstract

The present invention discloses a kind of dynamic bid method towards displaying advertisement bidding jettison system, it is therefore an objective to which bidding price can be calculated by launching the advertising display machine of rule to meeting for each advertising campaign.Bid methodologies are divided into offline and on-line stage：In off-line phase, it is necessary to establish the win mark rate distribution table of each period according to the historical data of advertising campaign, prediction clicking rate wins each budget of period in mark rate mapping equation, and distribution next advertisement putting cycle；The stage is launched online in advertisement, for the advertising display chance for meeting dispensing rule received, user's clicking rate of advertising display chance is predicted first, it is then based on predicting that clicking rate maps out advertising display window and hopes the win mark rate reached, the bidding price of advertising display chance is determined finally by the win mark rate distribution table for searching the corresponding period.Mobile state adjustment is entered according to the adjustable strategies actually spent, budget and bidding price that can be to each period present invention additionally comprises one.

Description

A kind of dynamic bid method towards displaying advertisement bidding jettison system
Technical field
The invention belongs to Internet technical field, more particularly to a kind of dynamic towards displaying advertisement bidding jettison system goes out Valency method.
Background technology
Online advertisement in recent years achieves huge success in industrial quarters, and only not various internet sites have found scale Change the means cashed, and provide a kind of brand-new advertisement marketing channel using accurate contact target user as methodology.At present The form of online advertisement is broadly divided into search advertisements and the displaying major class of advertisement two.Wherein search advertisements are closed according to the search of user Keyword, corresponding advertisement link is shown simultaneously with result of page searching, such as the search advertisements bid ranking of Baidu and Google Mechanism；Displaying advertisement is illustrated in the form of image or video on the advertisement position of website offer, suitable for various interconnections Website is netted, Fig. 1 is the schematic diagram of search advertisements and Fig. 2 is the schematic diagram for showing advertisement.Present invention is generally directed to show advertisement Launch in real time, i.e., advertisement position is embedded on webpage by website, when a user opens the page, handed over by website or advertisement Easy platform launches advertisement in real time to respective advertisement position, if user is interested in advertisement, it will usually clicks on and opens corresponding advertisement The page or video.Therefore displaying advertisement can targetedly orient the user for being delivered to and meeting advertising objective client characteristics, from And the income of advertisement marketing can be greatly improved, and website can also obtain income by launching displaying advertisement from advertiser.
The dispensing of displaying advertisement at present is broadly divided into Design by Contract Pattern and real time bid pattern.Wherein Design by Contract Pattern refers to by wide Accuse business and the contract for ensureing advertising display number is signed with website, the contract of this guarantee formula will ensure in advertisement putting cycle Intranet Stand according to rule is launched as defined in advertiser, complete the advertising display of predetermined number of times.And real time bid pattern is then that website general is wide Position trustship is accused to the vendor platforms of real time bid system (Real Time Biding, RTB) to be managed, and is carried out therewith Benefit share, and advertiser is then registered by party in request's platform in real time bid system, sets the dispensing of oneself advertisement Regular and daily budget, by party in request's platform according to regular and daily budget is launched, the advertisement exhibition of rule is launched to meeting Show that chance carries out real time bid, and the display machine to win can pay.In real time bid system, advertiser is put down in party in request The advertisement launched on platform to needs is registered, and sets the process for launching rule and budget daily, is referred to as initiated one wide Announcement activity.
Fig. 3 is the schematic diagram that displaying advertisement is launched by real time bid pattern.The basic process bidded is as follows：(1) when one When the female user of individual 30 years old browses the webpage of mother and baby website, the scripted code of embedded advertisement position can be to real-time on webpage Provider's platform in Ask-Bid System initiates the request of an advertising display chance；(2) provider's platform receives this advertisement exhibition When showing the request of chance, this can be directed to and ask to initiate the competing of advertising display chance to the trade center of real time bid system Valency is asked, and the request carries the cookie of the user and the contextual information of webpage；(3) trade center receives bid request Afterwards, the bid request is issued to the party in request's platform for being connected into real time bid system, also carries user's in bid request Cookie and webpage contextual information；(4) after party in request's platform receives the bid request of advertising display chance, pass through first The data management platform of cookie mapping techniques and inquiry correlation, obtain the feature tag of user, such as sex, age, purchase Thing interest etc.；(5) feature tag of the user is returned to party in request's platform by data management platform；(6) party in request's platform is according to every The dispensing rule of individual advertising campaign, retrieves this advertising campaign that advertising display chance meets；(7) party in request's platform is according to each The budget of advertising campaign, each to meet the advertising campaign calculating bidding price for launching rule；(8) party in request's platform is first flat Carry out a wheel inside platform to bid, the auction of platform interior is won in the advertising campaign of highest bidding price；(9) highest bidding price will Bidding price as party in request's platform returns to trade center；(10) trade center returns to all party in request's platforms received Bidding price, auctioned according to broad sense the second high price mechanism, i.e., display machine won by the advertising campaign of highest bidding price Meeting, but realized price is determined by the second high bidding price in this auction；(11) trade center notice triumph side, And deducted fees according to realized price；(12) link of the advertising campaign of triumph is returned to provider and put down by trade center Platform；(13) advertisement link is returned into the page that user browses by provider's platform；(14) generally real time bid system can track User will click on to the respondent behavior of the advertisement of dispensing or conversion behavior is to party in request's platform report, in order to which party in request puts down Platform further optimizes bid methodologies.
In above-mentioned bid process, the bid methodologies that party in request's platform uses are the keys for showing advertisement bidding jettison system One of technology.For advertiser, the bid methodologies of optimization can cause advertiser to be obtained in the case where giving daily budget Maximized income, for example, obtain most numbers of clicks or minimum each click spend (cost-per-click, CPC).Here it is a statistical value to click on every time and spend CPC, i.e., after the completion of the advertisement putting on the same day, by total cost divided by acquisition Number of clicks.Platform widely used bid methodologies in current needs side's include：Fixed bid, bid at random, based on advertisement exhibition Show the bid of chance quality.Wherein fixed bid refers to that for meeting that regular each display machine is launched in advertising campaign fixation can be gone out Price competitive bidding；Random bid refers to randomly generate a bidding price according to a price range；Based on advertising display chance matter The bid of amount refers to that bidding price is related to the quality of the advertising display chance, and quality is higher, and bid is higher, and system is launched real-time In system, weighed usually using user's clicking rate after advertisement putting or conversion ratio, therefore in this kind of bid methodologies, bid Function generally by the increasing function that simple designs are clicking rate or conversion ratio, that is, predicts advertisement putting to after corresponding to advertisement position, The possibility that user clicks on the advertisement is higher, and bidding price is higher.Wherein performance is most preferably a kind of is based on advertising display chance The linear bid methodologies of quality, the i.e. bidding price of advertising display chance and the clicking rate of display machine meeting are linearly incremented by relation. But in the displaying advertisement bidding jettison system of reality, due to the diversity of advertising display chance in itself, different periods user The otherness of advertisement behavior, and the various advertising campaigns of different periods are caused competing to the otherness of display machine meeting degree of contention Marked price lattice and the relation of advertising display chance quality become extremely complex.
Therefore, the present invention proposes a kind of dynamic bid method towards displaying advertisement bidding jettison system, this method includes One budget allocation strategy simply based on history conclusion of the business daily record and click logs, and a win mark based on each period The dynamic bid method of rate distribution table and available budget.Experiment proves that dynamic bid method proposed by the present invention can be given pre- Most user click frequencies is obtained in the case of calculation.
The content of the invention
It is an object of the invention to provide a kind of dynamic bid method towards displaying advertisement bidding jettison system, this method energy Each advertising campaign is enough helped to obtain most user click frequencies under given budget.For achieving the above object, this hair The dynamic bid method of bright offer, it is characterised in that comprise the following steps：
Step 1：A complete advertisement putting cycle is averagely divided into T period first, is designated as { 1,2,3 ... T }, Historical datas of the advertising campaign a L advertisement putting cycle is then based on, the win mark rate of each period is built for advertising campaign a Distribution table, comprise the following steps that：
Step 1.1：For advertising campaign a, all advertisement exhibitions won within L advertisement putting cycle in period t are calculated Show the prediction clicking rate of chance, the prediction clicking rate of each advertising display chance is designated as pCTR；
Step 1.2：For advertising campaign a, the prediction clicking rate of advertising display chance is divided into m grade, grade i model Enclose for [pCTRi,pCTRi+1), all prediction clicking rates are more than or equal to pCTRiAnd it is less than pCTRi+1Advertising display chance division To grade i, highest ranking m scope is [pCTRm, 1], all prediction clicking rates are more than or equal to pCTRmIt is and wide less than or equal to 1 Grade m can be divided into by accusing display machine；
Step 1.3：For it is each prediction clicking rate grade in advertising display chance, count first its period t into Price Range is handed over, concluded price is then divided into n scale of price { price1,price2,…,pricen, priceiRepresent Scale of price i, refer in all advertising display chances won, concluded price is less than or equal to priceiAdvertising display chance, pricenScale of price n is represented, is the highest concluded price in all advertising display chances won；
Step 1.4：In period t win mark rate distribution table, wijRepresent for advertising campaign a, advertising display chance it is pre- Survey clicking rate and belong to grade i, when the grade of bidding price is j, corresponding win mark rate, wij=nij/ni, wherein nijRepresent prediction Clicking rate belongs to grade i and concluded price is less than or equal to pricejAdvertising display chance number, niRepresent that prediction clicking rate belongs to Level i and concluded price be less than or equal to pricenAdvertising display chance number, that is, predict that clicking rate belongs to all of grade i and won The number of advertising display chance；
Repeat the above steps 1.1 to 1.4, mark rate distribution table can be won for advertising campaign a structure of each period；Similarly, Each advertising campaign that above-mentioned steps can be used to register on party in request's platform calculates the win mark rate distribution table of each period；Table 1 It is win mark rate distribution tables of the advertising campaign a in period t；
Win mark rate distribution tables of the advertising campaign a of table 1 in period t
Step 2：Historical data based on advertising campaign a L advertisement putting cycle, determine advertising campaign a future position The rate of hitting-win mark rate mapping equation, such as shown in (1)：
Wherein win_rate (a, pCTR) represents the advertising display machine that advertising campaign a is pCTR for a prediction clicking rate The win mark rate reached can it is expected；CTR_Low (a) is advertising campaign a prediction clicking rate lower threshold, when an advertising display machine When the prediction clicking rate of meeting is less than or equal to CTR_Low (a), wins mark rate and be mapped as 0, represent that advertising campaign a can be abandoned to this Advertising display chance is bidded；CTR_High (a) is advertising campaign a prediction clicking rate upper limit threshold, when an advertising display When the prediction clicking rate of chance is more than or equal to CTR_High (a), wins mark rate and be mapped as 1；It is pre- when advertising display chance When surveying clicking rate more than CTR_Low (a) and being less than CTR_High (a), advertising display is determined according to the linear relationship in formula (1) Chance it is expected the win mark rate reached, here α (a), and β (a) is the mapping parameters set by party in request's platform for advertising campaign a；This In advertising campaign a prediction clicking rate lower threshold CTR_Low (a) be defined as follows：
Wherein num_clks (pCTR≤CTR_Low (a)) represents that the prediction clicking rate that advertising campaign a is won is less than or equal to The number of click behavior occurs in CTR_Low (a) advertising display chance, num_imps (pCTR≤CTR_High (a)), represents The number of advertising display chance of the prediction clicking rate that advertising campaign a is won less than or equal to CTR_Low (a)；θLowIt is by party in request The clicking rate lower limit parameter that platform is set；
Advertising campaign a prediction clicking rate higher limit CTR_High (a) is defined as follows：
The prediction clicking rate that wherein num_clks (pCTR >=CTR_High (a)) represents to win in advertising campaign a is more than etc. The number of click behavior occurs in CTR_High (a) advertising display chance, num_imps (pCTR >=CTR_High (a) number for advertising display chance of the prediction clicking rate more than or equal to CTR_High (a) that advertising campaign a is won), is represented； θHighIt is the clicking rate upper limit parameter set by party in request's platform；
Step 3：Historical data based on advertising campaign a L advertisement putting cycle, it is every to next advertisement putting cycle The budget b (a, t) of individual period is allocated, shown in computational methods such as formula (2)：
Master budgets of the advertising campaign a that wherein B expressions advertiser is set a complete Ad dispensing cycle, reqs (a, D, t) represent that the satisfaction that period ts of the advertising campaign a d-th of advertisement putting cycle is received launches regular advertising display chance Number, L are a time windows set by party in request's platform, L >=1, for limiting the L advertisement putting cycle occurred recently Data are as prediction and the basis of statistics；
Step 4：After next advertisement putting cycle is entered, for advertising campaign a, when party in request's platform is received in period t To one meet its advertisement putting rule advertising display chance request when, first determine whether advertising campaign a period t whether Also available budget, if without available budget, abandon bidding to request；If also available budget, continue to hold Row step 5；
Step 5：For advertising campaign a, the bidding price for advertising display chance request is calculated, specific method is such as Under：
Step 5.1：For the feature of the request advertising display chances carried, predicted using advertising campaign a clicking rate Model, calculate advertising display chance request prediction clicking rate pCTR (request, a)；
Step 5.2：For advertising campaign a, the win for it is expected to reach based on formula (1) calculating advertising display chance request Mark rate win_rate (request, a, t)；If win_rate (request, a, t)=0, then abandon participating in the advertising display Chance is bidded；Win_rate if (request, a, t) ＞ 0, then continue executing with step 5.3；
Step 5.3：For the win mark rate win_rate (request, a, t) for it is expected to reach, party in request's platform searches advertisement Win mark rate distribution tables of the movable a in period t, it is first determined pCTR (request, prediction clicking rate grade a) belonged to, Ran Hou The prediction clicking rate grade i belonged to, minimum bidding price according to corresponding to win_rate (request, a, t) determines it pricej, meet win_rate (request, a, t)≤wij；
Step 5.4：Advertising campaign a is calculated for bidding price final advertising display chance request, such as formula (3) It is shown：
Bid (a, t, request)=γ (a, t) pricej (3)
Here γ (a, t) is bidding price regulatory factors of the advertising campaign a in period t, is before in the period, t starts, by Available budget and period t allocation budget after period t adjustment determine, shown in computational methods such as formula (4)：
Here b (a, t) represents the budget distributed in step 3, b*(a, t) represents to start it in next advertisement putting cycle Afterwards, before period t starts, the budget after being adjusted according to actual cost situation dynamic, γ (a, t) >=1；
Step 6：For advertising campaign a, judge whether its final bidding price bid (a, t, request) works as more than it The available budget of preceding period, if it does, then abandoning bidding to display machine meeting request, if be no more than, continue to participate in Display machine meeting request's bids；
Step 7：If advertising campaign a bidding price is the highest bidding price in real-time transaction system, hand in real time Easy center will notify advertising campaign a to win advertising display chance request, and according to the second high-priced auction mechanism, with competitive bidding valency The second high price in lattice is deducted fees as concluded price；
Step 8：After the period, t terminated, it is updated, is updated according to budget of the actual cost to the t+1 periods in period t Formula is such as shown in (5)：
Here cost (i) represents to have completed the period i of advertisement putting actual cost；In period t+1 available budget After renewal, formula (4) calculation interval t+1 price adjustment factor gamma (a, t+1) can be utilized based on the budget after renewal；
When the advertisement putting for completing new all periods in advertisement putting cycle, the history in this advertisement putting cycle can be obtained Data, then based on the historical data in L advertisement putting cycle of newest generation, return to step 1, restart to perform, be The bidding price of next advertisement putting computation of Period advertising display chance.
Can be that each advertising campaign accords with given budget for each by dynamic bid method provided by the invention Bidding price can be calculated by closing the advertising display machine of dispensing rule, so as to obtain most user click frequencies.
Brief description of the drawings
Fig. 1 is search advertisements schematic diagram provided by the invention
Fig. 2 is displaying advertisement schematic diagram provided by the invention
Fig. 3 is the basic process schematic diagram provided by the invention that displaying advertisement is launched by real time bid pattern
Fig. 4 is the flow chart of the off-line phase of dynamic bid method provided by the invention
Fig. 5 is the flow chart in the online dispensing stage of dynamic bid method provided by the invention
Fig. 6 is the flow chart of calculating bidding price provided by the invention
Fig. 7 is the frame diagram of dynamic bid method provided by the invention
Embodiment
The embodiment of the present invention is described below in conjunction with the accompanying drawings, so as to those skilled in the art preferably Understand the present invention.Requiring particular attention is that in the following description, when known function and the detailed description of design perhaps When can desalinate the main contents of the present invention, these descriptions will be ignored herein.
The historical data that the initial preceding L advertisement putting cycle is collected in advertising campaign for a new registration is described first Specific implementation method：For the advertising campaign a of new registration, dynamic bid method provided by the invention is first its setting Regular price is designated as bid_fixed_price (a), this price is of a relatively high, so that in the competitive bidding of reality as bidding price During, advertising campaign a can win most of advertising display chance for meeting its dispensing rule；Then advertising campaign a is with fixation The mode of bid participates in the competitive bidding of the advertising display chance in L advertisement putting cycle, and advertising campaign here is only to meeting its dispensing The advertising display chance of rule is submitted a tender, and L is a time window set by party in request's platform, and L >=1 is next predicting The various key parameter indexs in individual advertisement putting cycle, establish clicking rate forecast model, establish the win mark rate distribution of each period During table, based on the daily record data of bid methodologies of the invention using the L advertisement putting periodic recording occurred recently；
Party in request's platform is needed to advertising campaign a situation of bidding, advertisement putting situation, point in each advertisement putting cycle The situation of hitting is recorded, record be by every meet launch rule advertising display chance in units of, including but not limited to Lower information：(1) numbering of advertising display chance, the operating system that (2) user uses, the browser that (3) user uses, (4) are used Geographical position where family, the IP address that (5) user uses, the URL that (6) user accesses, the preference label of (7) user, (8) are wide Accuse length, the width of position, the floor valency of (9) advertising display chance, the bidding price that (10) party in request platform calculates；(11) it is wide Whether announcement activity wins display machine meeting, (12) concluded price, click behavior etc. whether occurs after (13) advertisement putting；
Wherein (1)~(9) are the relevant informations of advertising display chance, including the relevant information of user, the correlation of advertisement position Information, the floor valency of (9) display machine meeting refer to the rserve price set by provider's platform for advertising display chance, i.e., only When having the highest bidding price to be higher than floor valency, the auction of advertising display chance is just set up；(10)~(12) are and bidded and advertisement Related information is launched, if advertising campaign a wins this advertising display chance, corresponding concluded price is actual conclusion of the business Price, it is otherwise 0；(13) it is record to the respondent behavior of user after advertisement putting, can be represented with 1 bit, if with Click behavior occurs for family, bit 1, is otherwise 0；After the historical data for completing L advertisement putting cycle is collected, ability Advertising campaign a is helped to carry out dynamic bid using dynamic bid method provided by the invention.
Dynamic bid method provided by the invention mainly include off-line phase and it is online launch the stage, wherein off-line phase Basic step is as shown in figure 4, be specially：
Step 1：Historical data based on advertising campaign a L advertisement putting cycle, when building each for advertising campaign a The win mark rate distribution table of section；
Step 2：Historical data based on advertising campaign a L advertisement putting cycle, determine advertising campaign a future position The rate of hitting-win mark rate mapping equation, as shown in formula (1)；
Step 3：Historical data based on advertising campaign a L advertisement putting cycle, using formula (2) to next wide The budget b (a, t) for accusing dispensing each period in cycle is allocated；
Step 1~3 are completed before off-line phase i.e. next advertisement putting cycle starts, wherein needing in step 1 A key technology being used is to establish clicking rate forecast model, the method due to establishing clicking rate forecast model be one into Ripe method, therefore be not included in scope of the presently claimed invention, but bid methodologies provided by the invention need to use a little Rate forecast model is hit, the method for building up that simply will click on rate forecast model here is described as follows：Lived for a specific advertisement Dynamic a, the information for all advertising display chances that advertising campaign is won within the L nearest advertisement putting cycle, and its correspondingly User click on behavior be used as training set, using machine learning algorithm training clicking rate forecast model, wherein there is click behavior As positive sample the information of advertising display chance of click behavior does not occur for the information of advertising display chance as negative sample； Current most widely used clicking rate training pattern is Logic Regression Models；
Fig. 5 is the basic step that the stage is launched in advertisement online, is specially：
Step 4：After next advertisement putting cycle is entered, for advertising campaign a, when party in request's platform is received in period t To one meet its advertisement putting rule advertising display chance request when, first determine whether advertising campaign a period t whether Also available budget, if without available budget, abandon bidding to request；If also available budget, continue to hold Row step 5；
Step 5：For advertising campaign a, the bidding price for advertising display chance request is calculated, specific steps are such as Shown in Fig. 6：
Step 5.1：For the feature of the request advertising display chances carried, predicted using advertising campaign a clicking rate Model, calculate advertising display chance request prediction clicking rate pCTR (request, a)；
Step 5.2：For advertising campaign a, the win for it is expected to reach based on formula (1) calculating advertising display chance request Mark rate win_rate (request, a, t)；If win_rate (request, a, t)=0, then abandon participating in the advertising display Chance is bidded；Win_rate if (request, a, t) ＞ 0, then continue executing with step 5.3；
Step 5.3：For the win mark rate win_rate (request, a, t) for it is expected to reach, party in request's platform searches advertisement Win mark rate distribution tables of the movable a in period t, it is first determined pCTR (request, prediction clicking rate grade a) belonged to, Ran Hou The prediction clicking rate grade i belonged to, minimum bidding price according to corresponding to win_rate (request, a, t) determines it pricej, meet win_rate (request, a, t)≤wij；
Step 5.4：Advertising campaign a is calculated for bidding price final advertising display chance request, such as formula (3) It is shown：
Bid (a, t, request)=γ (a, t) pricej (3)
Here γ (a, t) is bidding price regulatory factors of the advertising campaign a in period t, is before in the period, t starts, by Available budget and period t allocation budget after period t adjustment determine, shown in computational methods such as formula (4)：
Here b (a, t) represents the budget distributed in step 3, b*(a, t) represents to start it in next advertisement putting cycle Afterwards, before period t starts, the budget after being adjusted according to actual cost situation dynamic, γ (a, t) >=1；
Step 6：For advertising campaign a, judge whether its final bidding price bid (a, t, request) works as more than it The available budget of preceding period, if it does, then abandoning bidding to display machine meeting request, if be no more than, continue to participate in Display machine meeting request's bids；
Step 7：If advertising campaign a bidding price is the highest bidding price in real-time transaction system, hand in real time Easy center will notify advertising campaign a to win advertising display chance request, and according to the second high-priced auction mechanism, with competitive bidding valency The second high price in lattice is deducted fees as concluded price；
Step 8：After the period, t terminated, it is updated, is updated according to budget of the actual cost to the t+1 periods in period t Formula is such as shown in (5)：
Here cost (i) represents to have completed the period i of advertisement putting actual cost；In period t+1 available budget After renewal, formula (4) calculation interval t+1 price adjustment factor gamma (a, t+1) can be utilized based on the budget after renewal.
Embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with embodiment to technology therein Scheme is clearly and completely described.
In this example, it is assumed that advertiser have registered a new advertising campaign A on party in request's platform, and by advertisement The dispensing cycle is arranged to one day, and daily budget is 10000 yuan, while sets advertising campaign A dispensing rule；For new registration Advertisement, because no history launches daily record as analysis foundation, therefore dynamic bid method provided by the invention is wide first Announcement activity A determines 0.30 yuan of a regular price, then launches the advertising display machine of rule to meeting according to this regular price It can be bidded, assume L=2 in the present embodiment, be i.e. all predictions to new one day are all based only on the historical data of nearest 2 days Carry out；
In the present embodiment, party in request's platform needs situation of bidding, the advertisement putting situation to nearest 2 days advertising campaign A And click on situation and recorded, record is in units of the advertising display chance of the every dispensing for meeting advertising campaign A rule , including but not limited to data below：(1) numbering of advertising display chance, the operating system that (2) user uses, (3) user make Browser, the geographical position where (4) user, the IP address that (5) user uses, the URL that (6) user accesses, (7) are used The preference label at family, the length of (8) advertisement position, width, the floor valency of (9) advertising display chance, (10) party in request platform calculate Bidding price；(11) whether advertising campaign wins display machine meeting, (12) concluded price, occur a little after (13) advertisement putting Hit behavior etc.；
After the advertisement real time bid completed 2 days is launched, dynamic bid method provided by the invention is first with this 2 days The information for all advertising display chances that advertising campaign A is won, and its corresponding user click on behavior as training set, wherein having The information of the advertising display chance of click behavior as positive sample, make by the information that the advertising display chance of click behavior does not occur For negative sample.Logic-based regression model training advertising campaign A clicking rate forecast model, then according to the history of nearest 2 days Data perform the step of off-line phase：
Step 1：Assuming that the time is divided into 12 periods, i.e., calculated since 0 point, every 2 hours are a period, Based on history log, the win mark rate distribution table of each period is established；
First, calculated using advertising campaign A clicking rate forecast model in the advertisement putting of nearest 2 days, it is all to win The prediction clicking rate of advertising display chance；Then according to the distribution situation of prediction clicking rate, prediction clicking rate is divided into some Individual grade, assume to be divided into 10 grades in the present embodiment, the scope of each prediction clicking rate grade is as follows：[0,0.1), [0.1,0.2), [0.2,0.3), [0.3,0.4), [0.4,0.5), [0.5,0.6), [0.6,0.7), [0.7,0.8), [0.8, 0.9), [0.9,1.0]；On this basis, recorded according to the conclusion of the business of nearest 2 days, divide the advertisement of each prediction clicking rate grade The concluded price distribution of display machine meeting, assumes 6 scale of price of division in the present embodiment, be respectively 0.05,0.10,0.15, 0.20,0.25,0.30 } (member), 0.05 yuan of implication here is concluded price less than or equal to 0.05 yuan it is all win it is wide 0.05 yuan of this scale of price can be belonged to by accusing display machine；Here 0.30 yuan represents in all advertising display chances won, Highest concluded price is 0.30 yuan, and 0.30 yuan of scale of price includes all advertising display chances won；
In this example, it is assumed that nearest 2 days in the period 1 advertising campaign A won altogether 10000 prediction clicking rates Advertising display chance less than 0.1, wherein the concluded price for having 4280 advertising display chances is less than or equal to 0.05 yuan, therefore, The period 1 for prediction clicking rate grade [0,0.1), 0.05 yuan of scale of price, it is corresponding win mark rate be w11=4280/10000 =0.428, other prediction clicking rate grades can be similarly obtained, and mark rate is won corresponding to other scale of price, as shown in table 2.Weight Multiple above-mentioned steps, can be calculated win mark rate distribution tables of the advertising campaign A in 12 periods；
Win mark rate distribution tables of the advertising campaign A of table 2 in the period 1
Step 2：Historical data based on advertising campaign A at nearest 2 days, determine advertising campaign A prediction clicking rate-win mark Rate mapping equation, in this example, it is assumed that the θ that advertising campaign A is setLow=0.0001, θHigh=0.9, α (A)=3.54, β (A)=0.646, therefore prediction clicking rate-win mark rate mapping equation is：
, can according to the definition of formula (6) and formula (7) wherein it is required to determine that CTR_Low (A) and CTR_High (A) Obtain needing the condition met as follows：
In the present embodiment, it is assumed that advertising campaign A is in the historical data of nearest 2 days, and the prediction clicking rate that wins is less than etc. In 0.001 advertising display chance sum be 10002, wherein only 1 advertising display there occurs click on behavior, meet condition：
Therefore settable CTR_Low (A)=0.001；Similarly, it is assumed that advertising campaign A in the historical data of nearest 2 days, Advertising display chance number of the prediction clicking rate won more than or equal to 0.1 is 100, wherein there is 94 advertising displays to be clicked on Behavior, meet condition：
Therefore settable CTR_High (A)=0.1；
According to formula (1), following result can be obtained：It is less than or equal to 0.001 advertising display chance for prediction clicking rate, its It is 0 to map obtained expectation and win mark rate, and advertising campaign A abandons the competitive bidding to this advertising display chance；For predicting clicking rate Advertising display chance more than or equal to 0.1, it is 1 that it, which maps obtained expectation and wins mark rate, is belonged to for prediction clicking rate The advertising display chance of (0.001,0.1) scope, it is expected to win mark rate to map using a linear formula, it is assumed that an advertisement exhibition The prediction clicking rate for showing chance is 0.05, then according to formula (1), it is expected that winning mark rate is：
Win_rate (A, 0.05)=3.54 × 0.05+0.646=0.823
Step 3：Historical data based on advertising campaign A at nearest 2 days, to next each period in advertisement putting cycle Budget b (A, t) is allocated, in this example, it is assumed that daily budget was 10000 yuan, in the real time bid of nearest 2 days In, meet that advertising campaign A launches the advertising display chance number distribution of rule as shown in table 3：
Table meets that advertising campaign A launches distributed number situation of the advertising display chance in 12 periods of rule for 3 nearest 2 days
Period 1 2 3 4 5 6 7 8 9 10 11 12
1st day 1000 1099 4324 4338 7311 3826 4797 4644 3674 5594 3887 2099
2nd day 1200 1030 4300 4000 7050 3800 4700 4600 3600 5500 3800 2000
Therefore, can obtain：
Similarly, the budget allocation of each period can be obtained, as shown in table 4：
The advertising campaign A of table 4 is to the new budget allocation scheme of one day
Period 1 2 3 4 5 6
The budget of distribution 238.68 230.97 935.63 904.60 1558.04 827.35
Period 7 8 9 10 11 12
The budget of distribution 1030.34 1002.896 789.168 1203.61 833.97 444.71
Step 4：Launched into the new advertisement bidding of one day, in this example, it is assumed that master budget is 10000 yuan, it is false The period 1 is located at, has one to meet the advertising display chance R arrival that advertising campaign A launches rule, first determines whether that advertising campaign A exists Whether the period 1 also has available budget, if also available budget, performs step 5, otherwise abandons to the competing of advertising display chance Mark；
Step 5：Calculate bidding prices of the advertising campaign A to advertising display chance R；
First, the clicking rate forecast model of training, calculating advertising display chance R prediction clicking rate, it is assumed that be are utilized 0.0025；Then, according to prediction clicking rate-win mark rate mapping equation, the win mark rate that advertising display chance R it is expected to reach is calculated：
Win_rate (A, 0.0025)=3.54 × 0.0025+0.646=0.655
3rd, according to the win mark rate distribution table won mark rate and search the period 1 for it is expected to reach, because prediction clicking rate is 0.0025 belong to [0,0.1) scope, meet 0.655 ＜ 0.798, therefore minimum bidding price is 0.10 yuan；Due to being in the period 1, new budget does not change with original budget alloments, therefore can obtain：
Bidding price final to advertising display chance R advertising campaign A is 0.10 yuan；
Step 6：In this example, it is assumed that to be not above advertising campaign A available pre- in the period 1 for final bidding price Calculate, then continue executing with and advertising display chance R is bidded；
Step 7：If 0.10 yuan is that advertising display chance R highest competitive bidding valency, advertisement are lived in real-time transaction system Dynamic A wins advertising display chance R, and system deducts concluded price according to the second high price from system, in this example, it is assumed that Second high bidding price is 0.008 yuan, then available budgets of the advertising campaign A in the period 1 will subtract 0.008 yuan；
Step 8：After the period 1 terminates, it is assumed that the period only takes 200 yuan of budget, then needs residue is available pre- Calculation is re-assigned in following sessions, according to formula (5), the new budget of calculation interval 2：
The bidding price dynamic Dynamic gene of calculation interval 2：
Assuming that in the period 2, there is one to meet that advertising campaign A launches the advertising display chance r of rule, by searching the period 2 Win mark rate distribution table, obtain being 0.20 yuan to advertising display chance r minimum bidding price, then final bidding price is：
Bid (A, 2, r)=γ (A, 2) pricej=1.003 × 0.20=0.2006 members
Although the illustrative embodiment of the present invention is described above, in order to the technology of the art Personnel understand the present invention, it should be apparent that the invention is not restricted to the scope of embodiment, to the common skill of the art For art personnel, if various change in the spirit and scope of the present invention that appended claim limits and determines, these Change is it will be apparent that all utilize the innovation and creation of present inventive concept in the row of protection.

Claims (3)

  1. A kind of 1. dynamic bid method towards displaying advertisement bidding jettison system, it is characterised in that comprise the following steps：
    Step 1：Historical data based on advertising campaign a L advertisement putting cycle, each period is built for advertising campaign a Mark rate distribution table is won, a complete advertisement putting cycle is averagely divided into T period here, is designated as { 1,2,3 ... T }；
    Step 2：Historical data based on advertising campaign a L advertisement putting cycle, determine advertising campaign a prediction clicking rate- Mark rate mapping equation is won, such as shown in (1)：
    Wherein win_rate (a, pCTR) represents the advertising display window that advertising campaign a is pCTR for a prediction clicking rate Hope the win mark rate reached；CTR_Low (a) is advertising campaign a prediction clicking rate lower threshold, when advertising display chance When prediction clicking rate is less than or equal to CTR_Low (a), it is expected that winning mark rate is mapped as 0, represents that advertising campaign a can be abandoned to this Advertising display chance is bidded；CTR_High (a) is advertising campaign a prediction clicking rate upper limit threshold, when an advertising display When the prediction clicking rate of chance is more than or equal to CTR_High (a), it is expected that winning mark rate is mapped as 1；When an advertising display chance Prediction clicking rate be more than CTR_Low (a) and when being less than CTR_High (a), determine advertisement according to the linear relationship in formula (1) Displaying window hopes the win mark rate reached, here α (a), and β (a) is joined by the mapping that party in request's platform is advertising campaign a settings Number；
    Step 3：Historical data based on advertising campaign a L advertisement putting cycle, when each to next advertisement putting cycle The budget b (a, t) of section is allocated, shown in computational methods such as formula (2)：
    Wherein B represents that the advertising campaign a that advertiser is set launches the master budget in cycle, reqs (a, d, t) in a complete Ad Represent that the satisfaction that period ts of the advertising campaign a d-th of advertisement putting cycle is received launches regular advertising display chance number, L is One time window set by party in request's platform, L >=1, the data for limiting the L advertisement putting cycle occurred recently are made For the basis predicted and counted；
    Step 4：Into after next advertisement putting cycle, for advertising campaign a, when party in request's platform receives one in period t When meeting the advertising display chance request of its advertisement putting rule, first determine whether that advertising campaign a can whether period t also has With budget, if without available budget, abandon bidding to request；If also available budget, continue executing with step 5；
    Step 5：For advertising campaign a, the bidding price for advertising display chance request is calculated, specific method is as follows：
    Step 5.1：For the feature of the request advertising display chances carried, mould is predicted using advertising campaign a clicking rate Type, calculate advertising display chance request prediction clicking rate pCTR (request, a)；
    Step 5.2：For advertising campaign a, the win mark rate for it is expected to reach based on formula (1) calculating advertising display chance request win_rate(request,a,t)；If win_rate (request, a, t)=0, then abandon participating in the advertising display chance Bid；Win_rate if (request, a, t) ＞ 0, then continue executing with step 5.3；
    Step 5.3：For the win mark rate win_rate (request, a, t) for it is expected to reach, party in request's platform searches advertising campaign a In period t win mark rate distribution table, it is first determined (request, prediction clicking rate grade a) belonged to, is then belonging to pCTR Prediction clicking rate grade i, minimum bidding price price according to corresponding to win_rate (request, a, t) determines itj, it is full Sufficient win_rate (request, a, t)≤wij；
    Step 5.4：Advertising campaign a is calculated for bidding price final advertising display chance request, as shown in formula (3)：
    Bid (a, t, request)=γ (a, t) pricej (3)
    Here γ (a, t) is bidding price regulatory factors of the advertising campaign a in period t, is before in the period, t starts, by period t Available budget and period t allocation budget after adjustment determine, shown in computational methods such as formula (4)：
    Here b (a, t) represents the budget distributed in step 3, b*(a, t) is represented after next advertisement putting cycle starts, when Before section t starts, the budget after being adjusted according to actual cost situation dynamic, γ (a, t) >=1；
    Step 6：For advertising campaign a, judge its final bidding price bid (a, t, request) whether exceed its it is current when The available budget of section, if it does, then abandoning bidding to display machine meeting request, if be no more than, continue to participate in displaying Chance request's bids；
    Step 7：If advertising campaign a bidding price is the highest bidding price in real-time transaction system, in real-time deal The heart will notify advertising campaign a to win advertising display chance request, and according to the second high-priced auction mechanism, with bidding price The second high price deducted fees as concluded price；
    Step 8：After the period, t terminated, it is updated according to budget of the actual cost to the t+1 periods in period t, more new formula As shown in (5)：
    Here cost (i) represents to have completed the period i of advertisement putting actual cost；Updated in period t+1 available budget Afterwards, formula (4) calculation interval t+1 price adjustment factor gamma (a, t+1) can be utilized based on the budget after renewal；
    When the advertisement putting for completing new all periods in advertisement putting cycle, the history in a new advertisement putting cycle can be obtained Data, then based on the historical data in L newest advertisement putting cycle, return to step 1, restart to perform.
  2. 2. the method as described in claim 1, it is characterised in that the win mark rate distribution table of advertising campaign a each periods Construction method, including：
    Step 1.1：For advertising campaign a, all advertising display machines won within L advertisement putting cycle in period t are calculated The prediction clicking rate of meeting；
    Step 1.2：For advertising campaign a, the prediction clicking rate of advertising display chance is divided into m grade, grade i scope is [pCTRi,pCTRi+1), all prediction clicking rates are more than or equal to pCTRiAnd it is less than pCTRi+1Advertising display chance be divided into grade I, grade m scope are [pCTRm, 1], all prediction clicking rates are more than or equal to pCTRmAnd the advertising display chance less than or equal to 1 It is divided into grade m；
    Step 1.3：For the advertising display chance in each prediction clicking rate grade, its knock-down price in period t is counted first Lattice scope, concluded price is then divided into n scale of price { price1,price2,…,pricen, priceiRepresent price Grade i, refer in all advertising display chances won, concluded price is less than or equal to priceiAdvertising display chance, pricen Scale of price n is represented, is the highest concluded price in all advertising display chances won；
    Step 1.4：In period t win mark rate distribution table, wijRepresent for advertising campaign a, the future position of advertising display chance The rate of hitting belongs to grade i, when the grade of bidding price is j, corresponding win mark rate, and wij=nij/ni, wherein nijRepresent that prediction is clicked on Rate belongs to grade i and concluded price is less than or equal to pricejAdvertising display chance number, niRepresent that prediction clicking rate belongs to grade i And concluded price is less than or equal to pricenAdvertising display chance number, that is, predict clicking rate belong to grade i it is all win it is wide Accuse the number of display machine meeting；
    Repeat the above steps 1.1 to 1.4, mark rate distribution table can be won for advertising campaign a structure of each period；Similarly, can adopt Each advertising campaign to be registered on party in request's platform calculates the win mark rate distribution table of each period in aforementioned manners；Table 1 is wide Win mark rate distribution tables of the announcement activity a in period t.
    Win mark rate distribution tables of the advertising campaign a of table 1 in period t
  3. 3. the method as described in claim 1, it is characterised in that the prediction clicking rate lower threshold of the advertising campaign a and upper The definition method of threshold value is limited, it is specific as follows：
    Step 2.1：Advertising campaign a prediction clicking rate lower threshold CTR_Low (a) definition is as shown in formula (6)：
    Wherein num_clks (pCTR≤CTR_Low (a)) represents that the prediction clicking rate that advertising campaign a is won is less than or equal to CTR_ The number of click behavior occurs in Low (a) advertising display chance, num_imps (pCTR≤CTR_Low (a)) represents that advertisement is lived The number of advertising display chance of the prediction clicking rate that dynamic a is won less than or equal to CTR_Low (a)；θLowIt is to be set by party in request's platform The clicking rate lower limit parameter put；
    Step 2.2：Advertising campaign a prediction clicking rate higher limit CTR_High (a) definition is as shown in formula (7)：
    Wherein num_clks (pCTR >=CTR_High (a)) represents to be more than or equal in the advertising campaign a prediction clicking rates won The number of click behavior occurs in CTR_High (a) advertising display chance, num_imps (pCTR >=CTR_High (a)) is represented The number of advertising display chance of the prediction clicking rate that advertising campaign a is won more than or equal to CTR_High (a)；θHighIt is by demand The clicking rate upper limit parameter that Fang Pingtai is set.
CN201710671569.4A 2017-08-08 2017-08-08 A kind of dynamic bid method towards displaying advertisement bidding jettison system Pending CN107529654A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710671569.4A CN107529654A (en) 2017-08-08 2017-08-08 A kind of dynamic bid method towards displaying advertisement bidding jettison system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710671569.4A CN107529654A (en) 2017-08-08 2017-08-08 A kind of dynamic bid method towards displaying advertisement bidding jettison system

Publications (1)

Publication Number Publication Date
CN107529654A true CN107529654A (en) 2018-01-02

Family

ID=60766239

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710671569.4A Pending CN107529654A (en) 2017-08-08 2017-08-08 A kind of dynamic bid method towards displaying advertisement bidding jettison system

Country Status (1)

Country Link
CN (1) CN107529654A (en)

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108985829A (en) * 2018-07-04 2018-12-11 厦门美柚信息科技有限公司 A kind of method that the web advertisement is quickly launched, system, electronic equipment and its medium
CN109272356A (en) * 2018-09-19 2019-01-25 电子科技大学 Optimal Bidding Strategies method based on AdWords
CN109377257A (en) * 2018-09-07 2019-02-22 上海晶赞融宣科技有限公司 Method and device that the online budget of advertisement smoothly distributes, readable storage medium storing program for executing
CN109697635A (en) * 2018-12-27 2019-04-30 拉扎斯网络科技（上海）有限公司 One kind is bidded acquisition methods, device, electronic equipment and medium
CN110070378A (en) * 2018-01-23 2019-07-30 优估（上海）信息科技有限公司 A kind of DSP advertisement dispensing price competing method, apparatus and system
CN110992114A (en) * 2019-12-24 2020-04-10 斑马网络技术有限公司 Advertisement transaction method, platform, electronic device and computer readable storage medium
CN111091400A (en) * 2018-10-23 2020-05-01 第四范式（北京）技术有限公司 Method and device for generating advertisement conversion prediction model and delivering advertisement
CN111401943A (en) * 2020-03-10 2020-07-10 厦门美图之家科技有限公司 Multi-source advertisement bidding system and method
CN111598677A (en) * 2020-07-24 2020-08-28 北京淇瑀信息科技有限公司 Resource quota determining method and device and electronic equipment
CN113222656A (en) * 2021-04-29 2021-08-06 西安点告网络科技有限公司 Programmed advertisement putting method, system, device, equipment and storage medium
CN113674013A (en) * 2021-07-08 2021-11-19 上海百秋电子商务有限公司 Advertisement bidding adjustment method and system based on merchant self-defined rules

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110070378A (en) * 2018-01-23 2019-07-30 优估（上海）信息科技有限公司 A kind of DSP advertisement dispensing price competing method, apparatus and system
CN108985829A (en) * 2018-07-04 2018-12-11 厦门美柚信息科技有限公司 A kind of method that the web advertisement is quickly launched, system, electronic equipment and its medium
CN109377257A (en) * 2018-09-07 2019-02-22 上海晶赞融宣科技有限公司 Method and device that the online budget of advertisement smoothly distributes, readable storage medium storing program for executing
CN109377257B (en) * 2018-09-07 2023-08-18 上海晶赞融宣科技有限公司 Advertisement online budget smooth allocation method and device and readable storage medium
CN109272356A (en) * 2018-09-19 2019-01-25 电子科技大学 Optimal Bidding Strategies method based on AdWords
CN111091400A (en) * 2018-10-23 2020-05-01 第四范式（北京）技术有限公司 Method and device for generating advertisement conversion prediction model and delivering advertisement
CN109697635A (en) * 2018-12-27 2019-04-30 拉扎斯网络科技（上海）有限公司 One kind is bidded acquisition methods, device, electronic equipment and medium
CN110992114A (en) * 2019-12-24 2020-04-10 斑马网络技术有限公司 Advertisement transaction method, platform, electronic device and computer readable storage medium
CN111401943A (en) * 2020-03-10 2020-07-10 厦门美图之家科技有限公司 Multi-source advertisement bidding system and method
CN111598677A (en) * 2020-07-24 2020-08-28 北京淇瑀信息科技有限公司 Resource quota determining method and device and electronic equipment
CN113222656A (en) * 2021-04-29 2021-08-06 西安点告网络科技有限公司 Programmed advertisement putting method, system, device, equipment and storage medium
CN113222656B (en) * 2021-04-29 2023-11-24 西安点告网络科技有限公司 Programmed advertisement putting method, system, device, equipment and storage medium
CN113674013A (en) * 2021-07-08 2021-11-19 上海百秋电子商务有限公司 Advertisement bidding adjustment method and system based on merchant self-defined rules
CN113674013B (en) * 2021-07-08 2024-04-30 上海百秋新网商数字科技有限公司 Advertisement bidding adjustment method and system based on merchant custom rules

Similar Documents

Publication Publication Date Title
CN107529654A (en) A kind of dynamic bid method towards displaying advertisement bidding jettison system
US9129313B1 (en) System and method for optimizing real-time bidding on online advertisement placements utilizing mixed probability methods
US7908238B1 (en) Prediction engines using probability tree and computing node probabilities for the probability tree
Chan et al. Consumer search activities and the value of ad positions in sponsored search advertising
Miralles-Pechuán et al. A novel methodology for optimizing display advertising campaigns using genetic algorithms
CN105741133A (en) Stepping management method for online advertisement delivery
US20160210657A1 (en) Real-time marketing campaign stimuli selection based on user response predictions
US20120253927A1 (en) Machine learning approach for determining quality scores
EP3371762A1 (en) Systems and methods for performance driven dynamic geo-fence based targeting
US11093977B2 (en) Ad ranking system and method utilizing bids and adjustment factors based on the causal contribution of advertisements on outcomes
CN107067274A (en) One DSP real time bid ad system based on blended learning model
CN101268483A (en) Framework for selecting and delivering advertisements over a network based on user behavioral interests
US8296176B1 (en) Matching visitors as leads to lead buyers
US20160210656A1 (en) System for marketing touchpoint attribution bias correction
WO2013010114A1 (en) Analyzing effects of advertising
US11367102B2 (en) Using on-line and off-line projections to control information delivery to mobile devices
CN111489186A (en) Time-interval budget management method oriented to automatic advertisement display putting device
US20140058793A1 (en) Forecasting a number of impressions of a prospective advertisement listing
CN109615442A (en) RTB real time bid method based on excitation video ads
CN112163886B (en) Real-time bidding advertisement resource allocation method based on reinforcement learning
CN106296287A (en) For the method and apparatus adjusting the bid of promotion message
US10181130B2 (en) Real-time updates to digital marketing forecast models
EP3682403A1 (en) A method and system for intelligent adaptive bidding in an automated online exchange network
WO2020104806A1 (en) Real-time bidding
US8712844B2 (en) Use of natural query events to improve online advertising campaigns

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
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20180102

WD01 Invention patent application deemed withdrawn after publication