CN112396476A

CN112396476A - CPM system and method for controlling ecpc value

Info

Publication number: CN112396476A
Application number: CN202011543192.2A
Authority: CN
Inventors: 陈冰
Original assignee: Shanghai Cangtai Information Technology Co ltd
Current assignee: Shanghai Cangtai Information Technology Co ltd
Priority date: 2020-12-23
Filing date: 2020-12-23
Publication date: 2021-02-23

Abstract

A CPM system for controlling an ecc value comprises a receiving module, a sending module and a processing module, wherein the receiving module is used for receiving a DSP advertisement request and acquiring feature data of a target user; the screening module is used for screening out corresponding advertisements to be launched from the advertisements to be launched according to the characteristic data; the machine learning module is used for learning historical data of advertisement delivery to obtain a predicted CTR; the historical data is stored in a release record database; the control module is used for calculating the CPM according to the ecpc and the prediction CTR, comparing the CPM with the set bid price, controlling the final bid price and sequencing the advertisements to be launched according to the final bid price; the releasing module is used for bidding according to the highest person in the final bidding determined by the control module and sending the advertising content according to the feedback condition; and the recording module is used for recording the advertisement putting condition and updating the advertisement putting record database. The present invention makes it possible to secure an ecpc also in the CPM mode.

Description

CPM system and method for controlling ecpc value

Technical Field

The invention relates to the technical field of internet advertisements, in particular to a CPM system and a method for controlling an ecpc value.

Background

In the field of internet advertising, the CPM system is one of the most common settlement methods. The network platform in the CPM advertising mode provides rich media and video presentation advertisements, which are more flexible than conventional banner advertisements or panel advertisements. Meanwhile, animation, video and dynamic interactive advertisements bring higher-level visual contact to the browser. For traditional brands where brand awareness and performance is a primary concern, the CPM model is more popular with advertisers. Even in collaboration with agents, they will have maximum control over their advertising content, trying to maximize brand impact. CPM advertisers are more concerned than their web site visits with maximizing brand exposure awareness, and thus CPM models are naturally the best choice for this advertising line. For example, by taking advantage of the interactivity of rich media advertisements, a brand of advertisement may require a viewer to select their favorite tastes or colors, and the advertiser may create a series of products by analyzing the user's favorite tastes or colors based on the data reflected by the advertisement. For an advertiser, the advertisement needs to be displayed with a certain click rate, so as to avoid the situation that a large number of advertisements are displayed and only a small number of clicks are made. ecpc (earning by Cost Per click) refers to the revenue Per click. Since ecpc is revenue for the advertising facilitator and expense for the advertiser, ecpc is also of great significance to the advertiser. The ecpc is an index for optimizing the amount of purchases made on click. CTR (Click-Through-Rate), which is a term commonly used for internet advertisements, refers to the Click Through Rate of a web advertisement (picture advertisement/text advertisement/keyword advertisement/ranked advertisement/video advertisement, etc.), i.e., the actual number of clicks of the advertisement (strictly speaking, the number of pages to reach the target page) divided by the advertisement presentation amount (Show content). Therefore, the eCPC value needs to be considered for the CPM mode to improve the advertisement presentation effect.

The invention provides a bidding method and a bidding device. The model establishing module establishes at least one model of the CTR estimation model, the CTR calibration model, the media quality evaluation model, the CVR estimation model and the probability black and white list model; the model fusion module fuses the established models; the flow value estimation module acquires historical data and real-time data of advertisement putting for a given KPI parameter, and estimates the flow value of the advertisement to be put based on the fused model; and the bidding module determines a proper bidding strategy to bid based on the estimated flow value. The invention provides a plurality of modes, and selects a proper model according to the evaluation of the media quality, thereby realizing the targeted advertisement delivery to the media, but the technical scheme can not meet the diversified requirements of the advertiser under the condition of the cost settlement mode of the advertiser.

Disclosure of Invention

The technical problem to be solved by the present invention is to provide a CPM system and method for controlling the value of an ecc, which can reduce the potential ecc of a bid by controlling the bid of advertisement delivery, so as to ensure the ecc even in a CPM mode.

In order to solve the technical problem, the invention provides a CPM system for controlling an ecpc value, which comprises a receiving module, a screening module, a machine learning module, a control module, a releasing module and a recording module, wherein the receiving module is used for receiving a CPM value;

the receiving module is used for receiving the DSP advertisement request and acquiring the characteristic data of the target user;

the screening module is used for screening out corresponding advertisements to be launched from the advertisements capable of being launched according to the characteristic data; the advertisements to be released all have corresponding set bids and target ecpc;

the machine learning module is used for learning historical data of advertisement delivery to obtain a predicted CTR; the historical data is stored in a release record database;

the control module is used for calculating the CPM according to the ecpc and the prediction CTR, comparing the CPM with the set bid price, controlling the final bid price and sequencing the plurality of advertisements to be released according to the final bid price;

the releasing module is used for bidding according to the highest person in the final bidding determined by the control module and sending the advertising content according to the feedback condition;

and the recording module is used for recording the advertisement putting condition and updating the advertisement putting record database.

As an improvement of the scheme, the DSP advertisement request can come from a channel provider and media.

In the technical scheme, the bid request of the channel provider and the bid request of the media are received, so that the application range of the system is ensured. As long as the advertisement request is in a DSP mode, the system can provide corresponding display, and the system is mainly controlled internally and has better adaptability.

As an improvement of the scheme, the control module ranks the advertisements to be placed from high to low according to the final bid.

In the technical scheme, because the DSP mode is carried out in a bidding mode, the advertisement to be released with higher bidding price is arranged in front, and the showing opportunity is more easily obtained after the bidding price. Compared with the method of sequencing by taking the ecpc or CTR as an index, the scheme ensures the competitiveness of the advertisements selected by the system and has higher display success rate.

As an improvement of the scheme, the machine learning module only learns the historical data in the past year.

In the above technical solution, since the module needs to process each advertisement request, a huge amount of data processing requirements may be caused. And only the data in one year is processed, a complete period of people's activities in the society is covered, and the existing sufficient data processing capacity makes the processing effect more accurate, also makes the processing capacity of advertisement control in certain scope, makes the system process the data faster, also makes the resource consumption of system less, promotes the speed of system operation.

As an improvement of the above scheme, the CPM calculation formula is CPM = ecpc CTR 1000.

In the technical scheme, the value of the CPM is calculated according to the value of the ecpc set by the advertiser, and the two indexes set by the advertiser can be compared. Wherein CTR is predicted CTR. Through the conversion of the indexes, the consideration of the two indexes is realized.

As an improvement of the above, the control module takes the lower of the CPM and the set bid as the final bid.

In the above technical solution, when the CPM value is low, the ecpc corresponding to the CPM is the ecpc value expected by the advertiser, and since the CPM is in a direct proportion to the ecpc and the set bid is greater than the CPM, the ecpc corresponding to the set bid is greater than the expectation of the advertiser, so that it is possible to ensure that the ecpc value is more reasonable with the CPM as the final bid. And when the CPM value is higher, the ecpc value corresponding to the CPM is higher, the ecpc value of the advertisement with the set bid is lower, and the set bid is used as the final bid, so that the advertisement consumption of an advertiser can be reduced, and the benefit of the advertiser is guaranteed.

As an improvement of the scheme, if the final bids of the advertisements to be placed are the same, the advertisements are ranked from low to high by taking the ecpc as a secondary index.

In the above technical solution, when the final bid is the same, having a lower ecpc means that the advertisement needs to face a weaker competition, i.e. the winning price is relatively lower, and there are more display opportunities when bidding with the final bid. The technical scheme enables the advertisement to obtain the maximum display opportunity and enables the display times of the advertisement to be more.

Correspondingly, the invention also provides a CPM method for controlling the ecpc value, which comprises the following steps.

A. And receiving the DSP advertisement request by using the receiving module and acquiring the characteristic data of the target user.

In the step, data is provided for the next step of matching the advertisements to be delivered by receiving the DSP advertisement request, so as to obtain a set of the advertisements which meet the requirements of advertisers.

B. Screening out corresponding advertisements to be launched from the advertisements capable of being launched by using the screening module according to the characteristic data; and the advertisements to be launched all have corresponding setting bids and target ecpc.

In the step, the advertisements which can be put in are screened according to the characteristic data in the previous step, and the advertisements which meet the requirements and are to be put in are screened out. Corresponding to each advertisement to be released, each advertisement to be released has a unique bid and a target ecpc; where bids are placed to participate in bidding with other ads and the target ecpc is the value of the ecpc set by the advertiser that needs to be guaranteed at the time of presentation, these two data are critical data to the present system.

C. Learning historical data of advertisement delivery by using the machine learning module to obtain a predicted CTR; the historical data is stored in a release record database.

In this step, since the CTR of each advertisement request can be obtained after the advertisement is delivered, the machine learning module needs to learn the historical data of advertisement delivery, so that the current CTR value can be predicted more accurately. In the CPM mode, because the advertisement is displayed for a plurality of times, the predicted CTR can obtain a more accurate value in the long term, and the value of the ecpc can be effectively controlled through calculation, so that the value of the ecpc in the advertisement delivery can meet the requirement of an advertiser.

D. And calculating the CPM by using the control module according to the ecpc and the prediction CTR, comparing the CPM with the set bid to control the final bid, and sequencing the advertisements to be delivered according to the final bid.

In this step, the ecpc and the bid price are balanced in consideration of both the ecpc and the bid price. When the bid price is too high, taking the CPM corresponding to the ecpc as a final bid price; when the bid price is low, the set bid price is taken as the final bid price, the dynamic balance of two indexes set by the advertiser can be realized, and the value of the ecpc index is also ensured while the bid price opportunity of the advertiser is ensured.

E. And the putting module is used for bidding according to the highest person in the final bids determined by the control module and sending the advertising content according to the feedback condition.

In this step, a bid is placed based on the advertisement determined in the previous step. The step is to offer the advertisement and send the subsequent advertisement content, so as to realize the advertisement putting process.

F. And recording the advertisement putting condition by using the recording module, and updating the advertisement putting record database.

In the step, the advertisement putting condition is recorded, and the advertisement putting record database is updated, so that the data in the step C is kept in the latest state, and the machine learning result in the step C is ensured.

The invention has the following beneficial effects.

The invention meets the diversified requirements of advertisers by adding the ecpc index in the CPM mode. According to the invention, the predicted CTR is obtained through learning of historical data, and the actual value of the CTR after delivery can be effectively evaluated. According to the invention, the predicted CTR is closer to the actual CTR through a large amount of advertisements, so that the CTR value set by an advertiser can be realized. The invention realizes the comparison between the ecpc and the set bid through the conversion of the ecpc, and realizes the dynamic balance of the bid and the ecpc. The invention processes the advertisement requests in the order of the advertisement sequencing, thereby not only ensuring that the advertisements meet the requirements of advertisers, but also ensuring that each advertisement request is fed back and ensuring the response of the system. The invention updates the release record database in real time, so that the prediction of the CTR is more accurate.

Drawings

FIG. 1 is a schematic diagram showing the structure of a CPM system for controlling an ecc value according to a first embodiment of the present invention.

Fig. 2 is an advertisement sequence according to the present invention.

FIG. 3 is a flowchart of a CPM method of controlling an ecc value according to a first embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings.

In a first embodiment of the present invention, as shown in fig. 1, a CPM system for controlling an ecpc value is provided, which includes a receiving module 100, a screening module 200, a machine learning module 300, a control module 400, a delivering module 500, and a recording module 600.

And the receiving module 100 is configured to receive the DSP advertisement request and obtain feature data of the target user.

Specifically, the DSP advertisement request may come from both the distributor and the media. The receiving module 100 interfaces with multiple channels or media, but only one of them. The receiving module 100 processes the received advertisement requests individually. The receiving module 100 needs to identify its characteristic data for the received advertisement request in order to compare with the targeting conditions set by the advertiser. The characteristic data includes an IP address, country/region, operating system, and the like. The feature data here has a one-to-one correspondence with the targeting conditions set by the advertiser. If the channel provider or media does not have the characterizing data attached to it when requesting a bid, the receiving module 100 sends a request to the channel to obtain the characterizing data.

The screening module 200 is configured to screen out a corresponding advertisement to be delivered from the advertisements that can be delivered according to the feature data; and the advertisements to be launched all have corresponding setting bids and target ecpc.

Specifically, the screening module 200 first compares feature data of all advertisement that can be delivered in the system according to the feature data of the user, and screens out all advertisements whose targeting conditions are consistent with the feature data, that is, the advertisements to be delivered. It should be noted that the advertisement that can be placed is an advertisement in all the placements, that is, the user does not pause the placement of the advertisement, and the system does not pause the placement of the advertisement due to the end of budget or the like. Each advertisement has a bid and a target ecpc set by the advertiser.

A machine learning module 300, configured to learn historical data of advertisement delivery to obtain a predicted CTR; the historical data is stored in a release record database.

Specifically, the machine learning module 300 performs learning prediction on data in a supervised learning manner, and only learns historical data in the past year. One year is a complete cycle of life for people in society. In one year, people experience spring, summer, autumn and winter, and meanwhile, arrangement of work, study and life is carried out in a time unit taking the year as a main time unit, so that great difference exists in one year, certain similarity exists among different years, and accurate results can be obtained by carrying out study prediction in a complete period. The machine learning module 300 reads the CTR data from the delivery record database, and then learns to predict CTR after learning. When machine learning is carried out, all the throwing parameters are regarded as a variable, a coefficient is distributed to the variable, and a vector is formed. Through learning training, the value of each coefficient is calculated. The vector is then converted to a function of the quantity between 0 and 1. And substituting the function into the putting data to obtain the predicted CTR. The release record database uses MariaDB instead of MySQL because MariaDB supports hash join. During connection, the optimizer establishes a HASH table in a memory by using a connection KEY (JOIN KEY) by using a smaller table (usually the smaller table or data source) of the two tables, stores column data into the HASH list, scans a larger table, performs HASH on the JOIN KEY in the same way, and then detects the HASH table to find out a row matched with the HASH table. It should be noted that: if the HASH table is too large to be constructed in the memory at one time, the HASH table is divided into several partitions, and the partitions are written into the temporal segment of the disk, which results in one more write, and thus the efficiency is reduced. This approach is suitable for cases where the smaller tables can be placed in memory at all, so that the total cost is the sum of the costs of accessing both tables. However, when the table is large, the memory cannot be completely put into the table, the optimizer divides the table into a plurality of different partitions, and writes the partition into a temporary section of the disk when the part which cannot be put into the memory, and a large temporary time period is needed so as to improve the performance of the I/O as much as possible.

And the control module 400 is used for calculating the CPM according to the ecpc and the prediction CTR, comparing the CPM with the set bid, controlling the final bid and sequencing the advertisements to be released according to the final bid.

Specifically, the CPM is calculated by CPM = ecpc CTR 1000. The control module 400 takes the lower of the CPM and set bids as the final bid. The control module 400 sorts the advertisements to be delivered according to the final bid from high to low. If the final bids of the advertisements to be delivered are the same, the advertisements are ranked from low to high with ecpc as a secondary index. For example, there are 5 ads as shown in FIG. 2, with the final bids being ranked from high to low B, (A, E), C, D, where A and E bid the same. Since the target ecpc of A is larger than that of E, then A is ranked before E to obtain the final ranks B, E, A, C, D.

And the releasing module 500 is used for bidding according to the highest person in the final bidding determined by the control module and sending the advertisement content according to the feedback condition.

Specifically, the highest in the final bid refers to the highest bidder for the advertisement that is not bidding. If the channel or media requests an advertisement and the highest bid fails, the highest bid of the remaining uncanded advertisements is used as the highest bidder. Still as illustrated in the above example, if advertisement B to be delivered is obtained, the delivery module 500 bids according to bid 1.4 of B. If the bidding is successful, the advertisement content request sent by the channel provider or the media is received, and the delivery module 500 sends the corresponding advertisement content to the channel provider or the media to complete the advertisement delivery. If the bidding fails, the advertisement content request sent by the channel provider or the media can not be received.

The recording module 600 is configured to record the advertisement delivery status and update the advertisement delivery record database.

Specifically, the recording module 600 is configured to write the data of the placement situations into the advertisement placement record database in a structured storage manner. The advertisement putting record database is stored in a structured storage mode, so that the reading and writing speed of data can be improved, and the running and response speed of the system can be improved. The recording module 600 updates the advertisement delivery record database in real time, including not only the advertisement that has successfully bid, but also the advertisement that has failed to bid. The recording module 600 also deletes the records for more than one year, so as to control the size of the advertisement delivery record database, shorten the reading and writing time, and improve the response speed of the system.

Accordingly, as shown in fig. 3, there is provided a CPM method of controlling an ecc value, including the following steps.

S1, receiving a DSP advertisement request by using the receiving module, and acquiring characteristic data of a target user.

In this step, the server receives the DSP advertisement request and obtains the feature data of the target user. If the feature data of the target user is missing, the missing information needs to be requested from the channel trader or media.

S2, screening out corresponding advertisements to be launched from the advertisements capable of being launched by using the screening module according to the characteristic data; and the advertisements to be launched all have corresponding setting bids and target ecpc.

In the step, the advertisement to be delivered which meets the requirement is identified by comparison according to the characteristic data in the previous step and the directional condition set by the advertiser in the system. Wherein, the advertisements to be released are all released advertisements. In this step, the bid and the target ecpc corresponding to the advertisement to be delivered are also obtained.

S3, learning historical data of advertisement putting by using the machine learning module to obtain a predicted CTR; the historical data is stored in a release record database.

In this step, the data of advertisement delivery is learned by machine learning to obtain a predicted CTR for evaluating the advertisement request. The specific implementation of this step is described in the above example. This step may be performed after step S2, or may be performed before step S2.

And S4, calculating the CPM by using the control module according to the ecpc and the prediction CTR, comparing the CPM with the set bid, controlling the final bid, and sequencing the plurality of advertisements to be released according to the final bid.

In this step, by converting the ecpc into the CPM, comparison between the ecpc and the setting bid can be realized, so that the ecpc and the setting bid can be considered, and the effect of controlling the ecpc in the CPM mode can be realized. This step is performed after the above steps are finished.

And S5, using the putting module to bid according to the highest person in the final bids determined by the control module, and sending the advertising content according to the feedback condition.

In this step, the advertisement selected in the previous step is subjected to a bidding process, and a subsequent advertisement content request of a channel provider or a media is responded to complete the final display of the advertisement.

And S6, recording the advertisement putting condition by using the recording module, and updating the advertisement putting record database.

In this step, the recording module updates the advertisement delivery record database in real time, which is the last one executed in all the steps.

In this embodiment, repeated parts are not described again in the previous embodiment, and parts not illustrated refer to the previous embodiment.

While the invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not to be limited to the disclosed embodiment, but on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims

1. A CPM system that controls an ecpc value, characterized in that: the system comprises a receiving module, a screening module, a machine learning module, a control module, a releasing module and a recording module;

2. The CPM system of claim 1, wherein the DSP advertisement request is from both a distributor and a media.

3. The CPM system for controlling ecc values of claim 1, wherein the control module ranks the ads to be placed from high to low final bids.

4. The CPM system for controlling ecpc values of claim 1, wherein the machine learning module learns only historical data over a past year.

5. A CPM system for controlling ecc values according to claim 1, wherein the CPM is calculated as CPM = ecc CTR 1000.

6. The CPM system of claim 1, wherein the control module is further configured to set the final bid to be the lower of the CPM and the set bid.

7. A CPM system for controlling the value of ecpc according to claim 3, wherein ecpc is used as a secondary indicator in order from low to high if the final bid for the ad to be served is the same.

8. A CPM method of controlling an ecc value, comprising:

A. receiving a DSP advertisement request by using a receiving module as claimed in any one of claims 1-7 and obtaining feature data of a target user;

B. screening out corresponding advertisements to be delivered from the advertisements capable of being delivered according to the characteristic data by using a screening module according to any one of claims 1-7; the advertisements to be released all have corresponding set bids and target ecpc;

C. learning historical data of ad impressions using a machine learning module of any of claims 1-7 to obtain a predicted CTR; the historical data is stored in a release record database;

D. calculating CPM from ecpc and predictive CTR using a control module according to any of claims 1-7, comparing CPM with a set bid to control a final bid, and ordering a plurality of said advertisements to be delivered according to the final bid;

E. using the putting module according to any one of claims 1-7 to bid according to the highest one of the final bids determined by the control module and to transmit advertising content according to feedback;

F. recording the advertisement placement using a logging module according to any of claims 1-7 and updating the advertisement placement record database.