A CPC advertising system and a parameter optimization method in the bidding process

Abstract

This invention relates to a CPC advertising system and a parameter optimization method in the bidding process. The system includes an ad recall module, a filtering module, a bid adjustment module, and a ranking module. The ad recall module is configured to reduce high-priced ads and / or supplement low-priced ads to obtain a first ad recall set when the number / proportion of low-priced ads recalled according to preset conditions is lower than a threshold. The filtering module obtains the estimated CTR of the ads in the first ad recall set and filters all or part of the ads whose estimated CTR is lower than the CTR threshold to obtain a second ad recall set. The bid adjustment module is configured to adjust the bids of the ads in the second ad recall set so that the estimated ROI is not less than the historical statistical ROI within a preset historical period. The ranking module is configured to rank the ads in the second ad recall set after bid adjustment according to eCPM. This invention can effectively control the average cost per click while improving ad CTR.

Description

Technical Field

[0001] This invention relates to internet advertising systems, and more particularly to a CPC (Cost Per Click) advertising system and a parameter optimization method in the bidding process. Background Technology

[0002] In modern society, advertising is an effective tool for increasing product and service awareness and promoting sales and transactions. There are various forms of advertising, among which CPC (Cost Per Click) advertising is a type of advertising where payment is made per click. Whether competing for ad space or recommending ads to users, the advertising system on an advertising platform mainly involves two stages when selecting an ad: ad recall and ranking of the recalled ads. The final selected ads should maximize the platform's revenue while satisfying the advertiser's interests. For CPC bidding ads, the advertiser's bid (BID) and the ad's click-through rate (CTR) are the two main factors determining the ranking. Therefore, most advertising systems rank ads using eCPM = CTR * BID, where eCPM (Effective Cost per Mille) refers to the advertising revenue earned per thousand impressions.

[0003] From the advertiser's perspective, the higher the CTR of an ad, the more likely the advertiser is to profit from the ad. However, experiments have shown that when the optimization goal is CTR, the ACP (Average Click Price) metric cannot be effectively controlled. When CTR increases, ACP also rises sharply, leading to a decrease in the advertiser's ROI (Return on Investment). Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention proposes a CPC advertising system and a parameter optimization method in the bidding process, which can effectively control ACP while improving the CTR of advertisements.

[0005] To address the technical problems in the prior art, according to one aspect of the present invention, a CPC advertising system is provided, comprising an ad recall module, a filtering module, a bid adjustment module, and a ranking module. The ad recall module is configured to reduce high-priced ads and / or supplement low-priced ads to obtain a first ad recall set when the number / proportion of low-priced ads recalled according to preset conditions is lower than a threshold. The filtering module is connected to the ad recall module and configured to obtain the estimated CTR of the ads in the first ad recall set, and filter all or part of the ads whose estimated CTR is less than the CTR threshold to obtain a second ad recall set. The bid adjustment module is connected to the filtering module and configured to adjust the bids of the ads in the second ad recall set according to the principle that the estimated return on investment is not less than the historical statistical return on investment within a preset historical period. The ranking module, along with the filtering module and the bid adjustment module, is configured to rank the ads in the second ad recall set after bid adjustment according to eCPM.

[0006] According to another aspect of the present invention, the present invention also provides a parameter optimization method in the CPC advertising bidding process, comprising the following steps: during the advertising recall process, when the number / proportion of low-priced ads in the recalled ads is lower than a threshold, reducing high-priced ads and / or supplementing low-priced ads to obtain a first advertising recall set; filtering all or part of the ads whose CTR estimates are less than the CTR threshold from the first advertising recall set based on the CTR estimate to obtain a second advertising recall set; adjusting the ad bids in the second advertising recall set according to the principle that the estimated return on investment is not less than the historical statistical return on investment within a historical preset time period; and sorting the ads in the second advertising recall set after the bid adjustment according to eCPM.

[0007] This invention optimizes ad bids that influence ACP at each stage of the ad bidding process. This effectively solves the problem of excessive ACP increases and decreased ROI for advertisers when improving CTR. It also effectively prevents exposure from focusing on high-bid products, thereby mitigating the Matthew effect, improving the matching efficiency between traffic and the platform, and ensuring a healthy and orderly ecosystem for commercial advertising on the platform, which is conducive to forming a virtuous cycle. This invention protects advertisers' revenue and improves their user experience, thereby attracting and expanding the number of advertisers and their budgets, and increasing the platform's overall commercial revenue. Attached Figure Description

[0008] The preferred embodiments of the present invention will now be described in further detail with reference to the accompanying drawings, wherein:

[0009] Figure 1 This is a block diagram illustrating the principle of a CPC advertising system according to an embodiment of the present invention;

[0010] Figure 2This is a block diagram illustrating the principle of an advertising recall module according to an embodiment of the present invention;

[0011] Figure 3 This is a schematic diagram of a process for recalling compliant advertisements according to an embodiment of the present invention;

[0012] Figure 4 This is a schematic diagram of a process for recalling compliant advertisements according to another embodiment of the present invention;

[0013] Figure 5 This is a block diagram of an advertising recall module according to another embodiment of the present invention;

[0014] Figure 6 This is a block diagram of the advertising recall module according to another embodiment of the present invention;

[0015] Figure 7 This is a schematic block diagram of a filtering module according to an embodiment of the present invention; and

[0016] Figure 8 This is a schematic diagram of a bid adjustment module according to an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] In the following detailed description, reference can be made to the accompanying drawings, which form part of this application and illustrate specific embodiments of the present application. In the drawings, similar reference numerals describe substantially similar components in different figures. Specific embodiments of the present application are described in sufficient detail below to enable those skilled in the art to implement the technical solutions of the present application. It should be understood that other embodiments may also be utilized, or structural, logical, or electrical changes may be made to the embodiments of the present application.

[0019] Figure 1This is a block diagram of a CPC advertising system according to an embodiment of the present invention. The system includes an ad recall module 1, a filtering module 2, a bid adjustment module 3, and a sorting module 4. The ad recall module 1 recalls ads from a pool of tens of thousands of ad materials (or recall sources) based on preset conditions. When the number / proportion of low-priced ads in the recalled ads is lower than a threshold, high-priced ads are reduced and / or low-priced ads are added to obtain a first ad recall set of hundreds. The filtering module 2 obtains the estimated CTR of the ads in the first ad recall set and filters out at least the ads that do not meet the CTR requirements to obtain a second ad recall set of several numbers. The bid adjustment module 3 adjusts the bids of the ads in the second ad recall set according to the principle that the estimated return on investment is not less than the historical statistical return on investment within a preset historical period. The sorting module 4 sorts the ads in the second ad recall set of several numbers after the bid adjustment according to eCPM. The number of advertisements in the first and second advertising recall sets can vary depending on the application scenario, such as a homepage or a single page; it can also vary depending on the number of advertisements that can be displayed on the terminal screen.

[0020] In one embodiment, such as Figure 2 As shown, the ad recall module 1 includes a CTR statistics unit 11, a bid monitoring unit 12, a high-bid ad filtering unit 13, and a recall unit 14. The CTR statistics unit 11 is used to calculate the CTR of ads in the recall source (ad material pool) over a pre-defined historical period, such as calculating the historical CTR of all ads within 90 days, and calculating the average CTR for the same scenario. The bid monitoring unit 12 obtains the bid BID of ads in the recall source and calculates the average bid. The high-priced ad filtering unit 13 filters high-priced ads according to a preset price filtering strategy. The recall unit 14 is connected to the CTR statistics unit 11, the bid monitoring unit 12, and the high-priced ad filtering unit 13, respectively, and recalls a first ad recall set from the recall source that meets the CTR requirements and whose number / proportion of low-priced ads is not less than the number / proportion threshold according to the recall strategy.

[0021] like Figure 3 The diagram shows the process of the ad recall module 1 recalling eligible ads from the recall source according to a recall strategy.

[0022] Step S11a: Calculate the CTR of the ads in the recall source over a preset historical period (e.g., 90 days) and sort the ads in the recall source in descending order of historical CTR.

[0023] Step S12a: Extract ads with historical CTR greater than or equal to the average CTR as the original ad recall set.

[0024] Step S13a: Traverse the ads in the original ad recall set, compare their bids (BID) with the average bid to determine the number of ads with a bid (BID) lower than the average . In the present invention, ads with a bid (BID) lower than the average are called low - price ads, and ads with a bid (BID) higher than the average are called high - price ads.

[0025] Step S14a: Calculate the proportion pro1 of low - price ads in the total number of ads in the original ad recall set.

[0026] Step S15a: Determine whether the proportion pro1 of low - price ads in the total number of ads is lower than the threshold v pro , such as 20%. If the proportion pro1 of low - price ads in the total number of ads is lower than v pro , execute Step S16a. If the proportion pro1 of low - price ads in the total number of ads is not lower than v pro , then use the current ad recall set as the first ad recall set in Step S22a.

[0027] Step S16a: Sort the high - price ads in the original ad recall set according to CTR * BID.

[0028] Step S17a: Calculate the number of ads n1 to be excluded and the total number of ads n in the current original ad recall set after excluding high - price ads t ;

[0029] Step S18a: Determine whether the total number of ads n in the current original ad recall set after excluding high - price ads t is greater than or equal to the minimum recall quantity Nmin. If n t ≥Nmin, execute Step S19a. If n t <Nmin, execute Step S20a.

[0030] Step S19: Delete the lowest - ranked n1 ads. And execute Step S22a.

[0031] Step S20a: Determine the number of low - price ads n2 to be supplemented.

[0032] Step S21a: Recall the n2 low - price ads with the highest CTR ranking from the original recall source.

[0033] Step S22a: Use the current ad set as the ad recall set as the first ad recall set.

[0034] In this recall strategy, the CTR recall is the main, and the ratio of high - price ads to low - price ads in the ad recall set that meets the recall quantity is within a reasonable range.

[0035] In another embodiment, such as Figure 4 As shown, another recall strategy is used to recall eligible advertisements.

[0036] Step 11b: Based on the current recall source type, filter out high-priced ads with a preset proportion. For example, filter real-time recall sources that acquire data in real time and hourly recall sources that acquire data hourly. For ads with bids higher than the first threshold BID1 (e.g., a bid of 50 yuan), filter them with a probability of 0.5, while ads with bids within the […] Ads between [BID1] are filtered with a probability of 0.3, and bids lower than the average bid are considered. Ads are not filtered. When filtering recall sources that obtain data on a daily basis, ads with bids higher than the first threshold BID1 are filtered with a probability of 0.3, while ads with bids within the […] range are filtered. Ads between [BID1] are filtered with a probability of 0.1, which is less than the average bid. Ads will not be filtered. The probabilities of 0.5, 0.3, and 0.1 are examples and can be adjusted according to the actual number of ads.

[0037] Step 12b: Calculate the CTR of ads in the recall source after high-price filtering for a historical period, and sort the ads in the recall source in descending order of historical CTR.

[0038] Step 13b: Extract ads with historical CTR greater than or equal to the average CTR as the second original ad recall set.

[0039] Step S14b: Traverse the ads in the second original ad recall set and compare their bid BID with the average bid. The size of the bid BID determines whether it is below the average. Number of advertisements: n3.

[0040] Step S15b: Compare whether the number of low-priced ads n3 is greater than or equal to the threshold v. n If the value is greater than 1, then in step S18b, the current ad recall set is used as the first ad recall set; if the value is not greater than 1, then step S16b is executed.

[0041] Step S16b: Calculate the number n4 of low-priced ads that need to be added.

[0042] Step S17b: Supplement the original recall source with n4 low-priced ads ranked first by CTR.

[0043] Step S18b: Use the current ad set as the first ad recall set.

[0044] In this embodiment, the recall strategy first filters out high-priced ads from the ad source, then recalls them according to CTR, while also taking into account the number of low-priced ads in the recall set.

[0045] The two embodiments above are merely examples of recalling advertisements with different bids during the recall stage of this invention. Under the premise that the number / proportion of low-priced advertisements in the advertisements is not less than the threshold, there may be other control processes depending on the application scenario, recall source, and other different situations, which will not be elaborated here.

[0046] This invention ensures a certain recall volume for low-priced ads during the recall phase, preventing the focus of product exposure on high-bid ads and avoiding the Matthew effect where higher-priced ads have higher click-through rates and lower-priced ads have lower click-through rates. By increasing the number of low-priced ads with high CTR, the matching efficiency between traffic and the platform can be effectively improved, enhancing user experience. This not only protects the interests of advertisers but also promotes the healthy development of the advertising system.

[0047] like Figure 5 As shown, in another specific embodiment, the ad recall module 1 further includes a CVR statistics unit 15, used to count the CVR of ads in the recall source over a historical period when recommending ads on the homepage and single-product pages. In these two application scenarios, the recall unit sorts the ads according to the CTR*CVR value and the corresponding threshold. Figure 3 The recall source or Figure 4 A certain number of advertisements were extracted from the recall source after it had been filtered at a high price.

[0048] exist Figure 6 In another embodiment shown, the ad recall module 1 further includes a search unit 16, corresponding to the application scenario where a user enters a search term on a page to perform a retrieval. At this time, while performing text matching between the user's search term and the ads in the recall source, the search unit 16 obtains the matched ad bids from the bid monitoring unit 12, filtering out ads with bids lower than a second threshold, thereby obtaining the original ad recall set; correspondingly, the recall unit 14 retrieves the original ad recall set according to... Figure 3 or Figure 4 The recall strategy in the middle recalls ads to obtain the first ad recall set.

[0049] Figure 7This is a block diagram illustrating the principle of a filtering module according to an embodiment of the present invention. In this embodiment, the filtering module 2 includes a CTR prediction unit 21 and a filtering unit 22. The CTR prediction unit 21 obtains the CTR prediction value of the advertisements in the first ad recall set according to the CTR model. To prevent the influence of too many high-priced samples on the CTR, the sampling weight of high-priced samples is limited and the sampling weight of low-priced samples is increased during the training of the CTR model. For example, the sampling weight of high-priced samples is controlled within the range of 1.5-2, and the sampling weight of low-priced samples is controlled within the range of 0.5-0.8, thereby effectively balancing the proportion of samples with different bids. The filtering unit 22 is connected to the CTR prediction unit 21 and filters the advertisements in the first ad recall set according to a threshold based on the CTR prediction value to obtain a second ad recall set.

[0050] Better still, in another embodiment, the filtering module further includes a CVR prediction unit 23, which obtains the CVR prediction value of the advertisements in the first ad recall set according to the CVR model. The filtering unit 22 is connected to the CVR prediction unit 23, and filters the advertisements in the first ad recall set according to a corresponding threshold based on the CTR prediction value and the CVR prediction value to obtain a second ad recall set. For example, the filtering unit 22 calculates the product of the CTR prediction value and the CVR prediction value of each advertisement, compares it with the threshold, and selects advertisements whose product of the CTR prediction value and the CVR prediction value is greater than or equal to the threshold as the second ad recall set.

[0051] like Figure 8 As shown, the bid adjustment module 3 includes an ROI statistics unit 31, an ROI estimation unit 32, and a bid adjustment unit 33. Here, the conversion rate p(c|u, a) is defined, representing the probability that user u completes (orders or payments) after clicking on advertisement a, v a For a given ad, the consumer's single purchase cost is the advertiser's revenue. Therefore, the GMV (Gross Merchandise Volume, or total transaction value) for one click is: GMV = p(c|u, a)·v a For CPC advertising, the cost is per click, meaning the ad's bid is calculated based on the number of clicks. a .

[0052] Therefore, the return on investment (ROI) of an ad (a) for user u to click once can be calculated using formula 1-3:

[0053]

[0054] The conversion rate of a particular advertisement for the same exposure in history can be obtained through statistical analysis, as shown in Equation 1-1 below:

[0055]

[0056] Where, n u This represents the total number of clicks made by users on the same ad over a given period of time. Therefore, E u [p(c|u, a)] represents the conversion rate per click for a single ad impression in historical data. At this point, b... a The bid for the advertisement.

[0057] ROI statistical unit 31 calculates the single statistical return on investment (ROI) of the first advertisement in the second advertising recall set within the historical preset time period according to formula 1-1. a .

[0058] ROI estimation unit 32 calculates the estimated return on investment (ROI) for each advertisement in the second advertising recall set according to formula 1-2. (u，a) .

[0059]

[0060] p(c|u,a) is the single-cycle estimated conversion rate, v a The cost per purchase by a consumer is known. The pre-bid price for the advertisement.

[0061] This invention indirectly influences ACP by limiting the estimated ROI. When the estimated ROI is greater than the historical ROI, it can prevent ACP from fluctuating excessively upwards. Therefore, parameters need to be adjusted to ensure that the ROI... a ≤roi (u，a) That is, should be It is established, and through derivation, it should be made Established. The following is a categorized discussion:

[0062] when This indicates that the advertisement is a high-quality traffic advertisement, and therefore can be adjusted upwards. This increases the current bid relative to the advertiser's original bid, with an adjustment range of, for example,

[0063] when This indicates that the advertisement is a low-quality traffic ad and can be adjusted downwards. This reduces the current bid relative to the advertiser's original bid, with the adjustment range being, for example, [missing information]. It can save advertisers money while still allowing low-quality traffic ads to meet ROI requirements.

[0064] Therefore, the bid adjustment unit 33 adjusts the bid when the estimated conversion rate p(c|u,a) is greater than or equal to the statistical conversion rate E. uIncrease the pre-bid price when [p(c|u,a)] Make the pre-bid price Greater than or equal to the original bid b a When the single estimated conversion rate p(c|u,a) is less than the single statistical conversion rate E u Reduce the pre-bid price when [p(c|u,a)] Make the pre-bid price Less than the original bid for the advertisement b a .

[0065] The sorting module 4 sorts the ads in the second ad recall set after bid adjustment according to eCPM = CTR * BID. The click-through rate (CTR) has been obtained from the CTR prediction unit 21 of the filtering module 2, and the bid BID has been obtained from the bid adjustment module 3. The sorting module 4 calculates the product of the two, sorts the results, and provides the top n ads to the corresponding other systems as needed.

[0066] This invention effectively solves the problems of decreased ROI and excessively high ACP increases for advertisers when increasing CTR by optimizing the ad bids that affect ACP at each stage of the ad bidding process.

[0067] The above embodiments are for illustrative purposes only and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the scope of the invention. Therefore, all equivalent technical solutions should also fall within the scope of the invention.

Claims

1. A CPC advertising system, comprising: an ad recall module configured to reduce high-priced ads and / or supplement low-priced ads to obtain a first ad recall set when the number and / or proportion of low-priced ads in the ads recalled according to preset conditions is lower than a threshold value; a filtering module connected to the ad recall module and configured to obtain CTR estimates of the ads in the first ad recall set, and filter all or part of the ads with CTR estimates less than a CTR threshold value to obtain a second ad recall set; a bid adjustment module connected to the filtering module and configured to adjust the bids of the ads in the second ad recall set so that the estimated return on investment is not less than the historical statistical return on investment in a historical preset period; and a ranking module connected to the filtering module and the bid adjustment module and configured to rank the ads in the second ad recall set with adjusted bids according to eCPM. The ad recall module comprises: a CTR statistics unit configured to statistically obtain the CTR of the ads in a recall source in a preset historical period; a bid monitoring unit configured to obtain the bids of the ads in the recall source and the average bid, and the ads with bids lower than the average bid are low-priced ads, and the ads with bids higher than the average bid are high-priced ads; a high-priced ad filtering unit configured to filter high-priced ads according to a preset price filtering strategy; and a recall unit connected to the CTR statistics unit, the bid monitoring unit and the high-priced ad filtering unit respectively and configured to recall a first ad recall set with a number / proportion of low-priced ads not less than a threshold value from the recall source according to CTR reverse ranking. The estimated return on investment is the ratio between the product of the single estimated conversion rate and the single purchase cost of a consumer and the pre-bid of the ads in the second ad recall set, and the historical statistical return on investment is the ratio between the product of the single statistical conversion rate and the single purchase cost of a consumer and the original bid of the ads in a historical period. 2.The system of claim 1, wherein the ad recall module further comprises: a CVR statistics unit configured to statistically obtain the CVR of the ads in a recall source in a preset historical period; and correspondingly, the recall unit recalls a first ad recall set with a number / proportion of low-priced ads not less than a threshold value from the recall source according to CTR and CVR product reverse ranking. 3.The system of claim 1 or 2, wherein the ad recall module further comprises: a search unit configured to obtain ads with bids not less than a second bid threshold value from the recall source based on text matching between the search words of a user and the ads to obtain an original ad recall set; and correspondingly, the recall unit recalls ads from the original ad recall set to obtain the first ad recall set. 4.The system of claim 1, wherein the filtering module comprises: a CTR estimation unit configured to obtain CTR estimates of the ads in the first ad recall set according to a CTR model; wherein, in training the CTR model, the sampling weight of high-priced samples ranges from 1.5 to 2, and the sampling weight of low-priced samples ranges from 0.5 to 0.8; and ​ ​ a filtering unit configured to filter the ads in the first ad recall set based on the CTR estimates to obtain a second ad recall set.

5. The system of claim 4, wherein the filtering module further comprises: a CVR estimation unit configured to obtain CVR estimates of the ads in the first ad recall set according to a CVR model; correspondingly, the filtering unit filters the ads in the first ad recall set whose product of the CTR estimate and the CVR estimate is less than a second threshold to obtain the second ad recall set.

6. The system of claim 1, wherein the bid adjustment module comprises: a ROI statistics unit configured to calculate a single-time return on investment (ROI) for each advertisement in the second advertisement recall set over a historical preset time period a ; wherein, E u [p(c|u,a)] is the statistically derived single conversion rate of the ad in a historical time period, v a is the consumer's single purchase expenditure, b a is the ad's original bid; a ROI estimation unit configured to calculate a single-estimate return on investment (ROI) for each advertisement in the second advertisement recall set (u,a) ; wherein, p(c|u, a) is a single-estimate conversion rate, v a is the consumer's single purchase spend, is the ad's pre-bid; and the bid adjustment unit is configured to increase the pre-bid u when the single-estimate conversion rate p(c|u,a) is greater than or equal to the single-statistic conversion rate E the pre-bid to be greater than or equal to the ad original bid b a ; and decrease the pre-bid u when the single-estimate conversion rate p(c|u,a) is less than the single-statistic conversion rate E the pre-bid to be less than the ad original bid b a .

7. A method for parameter optimization in a CPC ad auction process, comprising: in an ad recall process, when the number and / or proportion of low-price ads in the recalled ads is lower than a threshold, reducing the high-price ads and / or supplementing the low-price ads to obtain a first ad recall set; filtering all or part of the ads whose CTR estimate is less than a CTR threshold from the first ad recall set to obtain a second ad recall set; adjusting the bids of the ads in the second ad recall set so that the estimated return on investment is not less than the historical statistical return on investment in a preset historical period; and ranking the ads in the second ad recall set with adjusted bids according to eCPM; the ad recall process further comprises: counting the CTR of the ads in the recall source in a preset historical period; obtaining the bids of the ads in the recall source and the bid mean value, the ads with bids lower than the mean value are low-price ads, and the ads with bids higher than the mean value are high-price ads; filtering the high-price ads according to a preset price filtering strategy; and recalling a first ad recall set with a number / proportion of low-price ads not lower than a threshold from the recall source according to CTR reverse ranking; the estimated return on investment is the ratio between the product of the single estimated conversion rate and the consumer single purchase cost and the pre-bid of the ads in the second ad recall set; and the historical statistical return on investment is the ratio between the product of the single statistical conversion rate and the consumer single purchase cost and the original bid of the ads in a historical period.

8. The method of claim 7, further comprising: counting the CTR of the ads in the recall source in a preset historical period and ranking the ads in the recall source in descending order of CTR; obtaining the bids of the ads in the recall source and the bid mean value; filtering the high-price ads according to a preset price filtering strategy; and recalling a first ad recall set with a number / proportion of low-price ads not lower than a threshold from the recall source according to CTR reverse ranking.

9. The method of claim 8, further comprising: counting the CVR of the ads in the recall source in a preset historical period and recalling a first ad recall set with a number / proportion of low-price ads not lower than a threshold from the recall source according to reverse ranking of the product of CTR and CVR.

10. The method of claim 8 or 9, further comprising: performing text matching between the search term of a user and the ads in the recall source; obtaining, from the ads matching the search query text of the user, ads with an ad bid no less than a second bid threshold to obtain an original ad recall set; and recalling ads from the original ad recall set to obtain the first ad recall set.

11. The method of claim 8, wherein the price filtering policy comprises: The corresponding filtering bid threshold and filtering ratio are set according to the type of the recall source.

12. The method of claim 7, further comprising: training a CTR model, wherein the sampling weight of a high-price sample in the training set ranges from 1.5 to 2, and the sampling weight of a low-price sample ranges from 0.5 to 0.8; and obtaining a CTR estimate of an ad in the first ad recall set based on the CTR model.

13. The method of claim 12, further comprising: training a CVR model; obtaining a CVR estimate of an ad in the first ad recall set based on the CVR model; and filtering the ads in the first ad recall set based on the product of the CTR estimate and the CVR estimate to obtain a second ad recall set.

14. The method of claim 7, further comprising: computing a single statistical return on investment (roi) for each advertisement in the second advertisement recall set over a historical preset time period a ; wherein, E u [p(c|u,a)] is a statistically derived single conversion rate of the ad over a historical time period, v a is the consumer's single purchase expenditure, b a is the original bid for the ad; computing a single-estimate return on investment (roi) for each advertisement in the second advertisement recall set (u,a) ; wherein, p(c|u, a) is a single-estimate conversion rate, v a is the consumer's single purchase spend, is the ad's pre-bid. Comparing the single-estimate conversion rate p(c|u,a) and the single-statistical conversion rate E u [p(c|u,a)]; and increasing the pre-bid when the single-estimate conversion rate p(c|u,a) is greater than or equal to the single-statistical conversion rate E u [p(c|u,a)] when the single-estimate conversion rate p(c|u,a) is greater than or equal to the single-statistical conversion rate E decreasing the pre-bid when the single-estimate conversion rate p(c|u,a) is less than the single-statistical conversion rate E making the pre-bid less than the ad original bid b a ; and decreasing the pre-bid when the single-estimate conversion rate p(c|u,a) is less than the single-statistical conversion rate E u [p(c|u,a)] when the single-estimate conversion rate p(c|u,a) is less than the single-statistical conversion rate E making the pre-bid less than the ad original bid b making the pre-bid less than the ad original bid b a。 15. The method of claim 14, wherein the pre-bid is increased in the range [1, The pre-bid is decreased in the range

Images