CN113807897A

CN113807897A - Participation control method, system and storage medium

Info

Publication number: CN113807897A
Application number: CN202111127951.1A
Authority: CN
Inventors: 苏毓敏; 潘泽宇; 张彬彬; 朱鑫; 支荣
Original assignee: Beijing Wodong Tianjun Information Technology Co Ltd
Current assignee: Beijing Wodong Tianjun Information Technology Co Ltd
Priority date: 2021-09-26
Filing date: 2021-09-26
Publication date: 2021-12-17

Abstract

The disclosure provides a participation control method, a participation control system and a storage medium, and relates to the technical field of computers. The disclosed competition participation control method comprises: acquiring real-time pre-estimated conversion parameters of the advertisement through a pre-estimated model; determining a flow quality layer to which real-time consumption at the moment corresponding to the real-time pre-estimated conversion parameter belongs according to the real-time pre-estimated conversion parameter and the pre-estimated conversion parameter layering threshold; aiming at each flow quality layer, adjusting the competition probability of each flow quality layer according to the difference between the actual consumption of the previous time window and the target consumption of the next time window in the two adjacent time windows; and carrying out advertisement participation according to the adjusted participation probability. By the method, the generation of burst flow can be avoided, and the stability of the participated flow is improved.

Description

Participation control method, system and storage medium

Technical Field

The present disclosure relates to the field of computer technologies, and in particular, to a participation control method, system, and storage medium.

Background

The online advertising industry is an emerging industry that has been vigorously developed in recent years, and has a wide market of billions of dollars worldwide. In the online advertising system, the RTB (Real Time Bidding) advertisement occupies a large part of market share so far, and the technology in this field is also developing and innovating continuously, however, because the number of service advertisers is large and the demand is diversified, how to make the RTB advertisement is still a big problem in the industry.

RTB advertisers pursue advertisement delivery effects and delivery experiences on the basis of a certain budget.

Disclosure of Invention

One object of the present disclosure is how to improve the smoothness of the participating flows.

According to an aspect of some embodiments of the present disclosure, there is provided a method of participating in competition control, including: acquiring real-time pre-estimated conversion parameters of the advertisement through a pre-estimated model; determining a flow quality layer to which real-time consumption at the moment corresponding to the real-time pre-estimated conversion parameter belongs according to the real-time pre-estimated conversion parameter and the pre-estimated conversion parameter layering threshold; aiming at each flow quality layer, adjusting the competition probability of each flow quality layer according to the difference between the actual consumption of the previous time window and the target consumption of the next time window in the two adjacent time windows; and carrying out advertisement participation according to the adjusted participation probability.

In some embodiments, adjusting the competing probabilities for each traffic quality layer comprises: acquiring consumption errors of actual consumption of a previous time window and target consumption of a subsequent time window in two adjacent time windows; under the condition that the consumption error belongs to a preset first interval, the competition participation probability is not adjusted; and under the condition that the consumption error does not belong to a preset first interval, sequentially adjusting the competition probability of each flow quality layer according to a preset adjusting strategy according to the magnitude sequence of the pre-estimated conversion interval threshold corresponding to the flow quality layer.

In some embodiments, obtaining the consumption error of the actual consumption of the previous time window and the target consumption of the next time window in the two adjacent time windows comprises: according to the time length ratio of the later time window to the prior time window, the actual consumption of the prior time window is adjusted, and the proportional change consumption is obtained; the difference between the target consumption and the proportional change consumption of the later time window is obtained as the consumption error.

In some embodiments, when the consumption error does not belong to the predetermined first interval, sequentially adjusting the bidding probability of each flow quality layer according to the predetermined adjustment strategy according to the magnitude sequence of the pre-estimated conversion interval threshold corresponding to the flow quality layer includes: under the condition that the consumption error is smaller than a low threshold value of a preset first interval, sequentially reducing the competition probability of each flow quality layer according to the sequence from low to high of the threshold value of the pre-estimated transformation interval corresponding to the flow quality layer; and under the condition that the consumption error is larger than a high threshold value of a preset first interval, sequentially improving the competition probability of each flow quality layer according to the sequence from high to low of the pre-estimated conversion interval threshold values corresponding to the flow quality layers.

In some embodiments, adjusting the competing probabilities for each traffic quality layer according to a predetermined adjustment strategy comprises: for each flow quality layer, obtaining the sum of the consumption error and the actual consumption of the previous time window of the corresponding flow quality layer, and determining the adjustment consumption; and adjusting the competition participation probability of the previous time window according to the proportion of the adjustment consumption to the actual consumption of the previous time window of the corresponding flow quality layer, and acquiring the competition participation probability of the later time window.

In some embodiments, the participation control method further comprises: and under the condition that the actual consumption of the traffic quality layers is smaller than a preset first consumption threshold, raising the advertisement bid according to a preset percentage, and updating the participation probability of each traffic quality layer until the actual consumption of each traffic quality layer is not smaller than the preset first consumption threshold.

In some embodiments, the participation control method further comprises: before determining and adjusting the competitive probability of each flow quality layer, determining whether data flow blockage occurs according to the variable quantity of the accumulated value of actual consumption; determining that data flow blocking occurs when the difference between the latest accumulated value of actual consumption and the previous accumulated value of actual consumption of a predetermined number is less than or equal to a blocking threshold; otherwise, executing an operation of adjusting the competition probability of each flow quality layer according to the difference between the actual consumption of the previous time window and the target consumption of the next time window in two adjacent time windows aiming at each flow quality layer.

In some embodiments, the participation control method further comprises: the target consumption for the later time window is determined based on the expected consumption value, the actual cumulative consumption fraction for the previous time window and the expected cumulative consumption fraction for the later time window, or based on a distribution of expected cumulative consumption, expected single ad costs and actual single ad costs.

In some embodiments, determining the target consumption for the later time window based on the expected consumption value, the ratio of actual cumulative consumption for the previous time window and the ratio of expected cumulative consumption for the later time window, or based on a distribution of expected cumulative consumption, expected single ad cost and actual single ad cost comprises: in the case where the actual conversion value is greater than the predetermined first conversion value, if the actual single-ad cost at the present time of the day is not lower than the expected single-ad cost, a target consumption in the later time window is determined according to the distribution of the expected cumulative consumption, the expected single-ad cost, and the actual single-ad cost.

In some embodiments, determining the target consumption for the later time window based on the expected consumption value, the ratio of actual cumulative consumption for the previous time window and the ratio of expected cumulative consumption for the later time window, or based on a distribution of expected cumulative consumption, expected single ad cost and actual single ad cost further comprises:

in the event that the actual conversion value is greater than the predetermined first conversion value, if the actual single-bar cost is less than the expected single-bar cost at the current date, then

Determining an expected consumption value based on the expected single-ad cost and the actual conversion value, and determining a first consumption in a subsequent time window based on the expected consumption value, an actual cumulative consumption fraction in the previous time window, and an expected cumulative consumption fraction in the subsequent time window;

determining a second consumption for the later time window based on the expected consumption profile, the expected single ad cost, and the actual single ad cost;

determining a larger one of the first consumption and the second consumption as a target consumption in a later time window.

In some embodiments, determining the target consumption for the later time window based on the expected consumption value, the ratio of actual cumulative consumption for the previous time window and the ratio of expected cumulative consumption for the later time window, or based on a distribution of expected cumulative consumption, expected single ad cost and actual single ad cost further comprises: in the case where the actual conversion value is not greater than the predetermined first conversion value, if there is actual consumption data for the history date, taking a weighted sum of the actual cumulative consumption for the previous time window of the present day and the actual cumulative consumption at the corresponding time of the previous day as an expected consumption value; a target consumption for a later time window is determined based on the expected consumption value, the actual cumulative consumption duty for the current day, and the expected cumulative consumption duty for the later time window.

In some embodiments, the participation control method further comprises: in the case where the actual conversion value is not greater than the predetermined first conversion value, if there is no actual consumption data for the history date, the product of the actual cumulative consumption duty currently on the day and the expected all-day consumption based on the expected single advertisement cost is taken as the target consumption.

In some embodiments, the participation control method further comprises: and determining the expected accumulated consumption of the later time window according to the distribution of the actual consumption of the historical date, the actual accumulated consumption proportion of the current date and the target total consumption value of the current date.

In some embodiments, the participation control method further comprises: and determining a target total consumption value of the current date according to the actual total consumption value of the historical date, the expected single advertisement cost and the actual single advertisement cost.

In some embodiments, determining the expected cumulative consumption for the later time window comprises: determining a value to be consumed of the current date according to the actual accumulated consumption ratio of the current date and the target total consumption value of the current date; and according to the distribution of the actual consumption of the historical date, distributing the value to be consumed of the current date in the remaining period of the current date, and determining the expected accumulated consumption of the later time window.

In some embodiments, determining the target total consumption value comprises: under the condition that the actual conversion value of the historical date is larger than or equal to the preset second conversion value, adjusting the actual total consumption value corresponding to the historical date according to the proportion of the difference between the expected single advertisement cost and the actual single advertisement cost to the expected single advertisement cost, and obtaining a first total consumption value; under the condition that the actual conversion value corresponding to the historical date is smaller than a preset second conversion value, determining the actual total consumption value corresponding to the historical date as a first total consumption value; multiplying the first total consumption value and the expected single advertisement cost respectively with the corresponding hyperparameters, comparing the hyperparameters with a preset minimum target, and determining a second total consumption value of which the largest item is the current date; in the case where the number of history dates is greater than 1, a weighted sum of second total consumption values for different history dates is acquired as a target total consumption value for the current date.

In some embodiments, determining the target total consumption value based on the historical date total actual consumption value, the expected single ad cost, and the actual single ad cost further comprises: in the case where the number of the history dates is equal to 1, the second total consumption value is set as the target total consumption value for the current date.

In some embodiments, the participation control method further comprises: according to the distribution of the estimated transformation parameters of the historical dates of the preset days and the preset layering number, dividing the estimated transformation parameters into equal parts of the preset layering number, and acquiring the quantile point parameters for dividing the flow quality layers as the layering threshold of the estimated transformation parameters.

In some embodiments, the participation control method further comprises at least one of: initializing the competition participation probability of all the flow quality layers of all the advertisement units to be a preset first initial value at the starting moment of each day; initializing the competition participation probability of all the flow quality layers of the newly added advertisement unit to be a preset second initial value under the condition that the newly added advertisement unit exists; under the condition that the target consumption of a later time window is less than or equal to a preset second consumption threshold, setting the competition probability of the flow quality layer with the highest threshold value of the pre-estimated conversion interval as a preset first probability, and setting the competition probabilities of other flow quality layers as being less than the preset first probability; or after adjusting the competition participation probability of each flow quality layer each time, adjusting the competition participation probability of the layer with the highest pre-estimated conversion interval threshold value in each flow quality layer with the competition participation probability of 0 to be the preset second probability.

In some embodiments, the participation control method further comprises: and under the condition that the actual consumption of the prior time window is less than or equal to the preset second consumption threshold, if the consumption error does not belong to the preset first interval, modifying the actual consumption of the prior time window into preset third consumption.

In some embodiments, the upper limit of the adjustment of the competition participation probability is 1, and the lower limit of the adjustment is 0.

According to an aspect of some embodiments of the present disclosure, there is provided a competition participation control system including: the estimation unit is configured to obtain real-time estimation transformation parameters of the advertisements through the estimation model; the layering unit is configured to determine a flow quality layer to which real-time consumption at a moment corresponding to the real-time pre-estimated conversion parameter belongs according to the real-time pre-estimated conversion parameter and a pre-estimated conversion parameter layering threshold; the probability adjusting unit is configured to adjust the competitive probability of each flow quality layer according to the difference between the actual consumption of the previous time window and the target consumption of the next time window in two adjacent time windows aiming at each flow quality layer; and the competition participating unit is configured to participate in the advertisement competition according to the adjusted competition participating probability.

In some embodiments, the participation control system further comprises a congestion determination unit configured to: before determining and adjusting the competitive probability of each flow quality layer, determining whether data flow blockage occurs according to the variable quantity of the accumulated value of actual consumption; determining that data flow blocking occurs when the difference between the latest accumulated value of actual consumption and the previous accumulated value of actual consumption of a predetermined number is less than or equal to a blocking threshold; otherwise, the probability adjustment unit is activated.

In some embodiments, the participation control system further comprises: a window target consumption determination unit configured to determine a target consumption for a later time window based on the expected consumption value, the actual cumulative consumption fraction for the previous time window and the expected cumulative consumption fraction for the later time window, or based on a distribution of the expected cumulative consumption, the expected single advertisement cost and the actual single advertisement cost.

In some embodiments, the participation control system further comprises: an expected cumulative consumption determination unit configured to determine an expected cumulative consumption for a later time window based on the distribution of the actual consumption for the history date, the actual cumulative consumption proportion for the current date, and the target total consumption value for the current date.

In some embodiments, the participation control system further comprises: and a target total consumption value determining unit configured to determine a target total consumption value of the current date according to the actual total consumption value of the historical date, the expected single advertisement cost and the actual single advertisement cost.

In some embodiments, the participation control system further comprises: and the layering threshold determining unit is configured to divide the pre-estimated transformation parameters into equal parts of preset layering quantity according to the distribution of the pre-estimated transformation parameters of the historical dates of preset days and the preset layering quantity, and acquire the quantile point parameters for dividing the flow quality layers as the layering thresholds of the pre-estimated transformation parameters.

In some embodiments, the participation control system further comprises a first processing unit, further configured to perform at least one of: initializing the competition participation probability of all the flow quality layers of all the advertisement units to be a preset first initial value at the starting moment of each day; initializing the competition participation probability of all the flow quality layers of the newly added advertisement unit to be a preset second initial value under the condition that the newly added advertisement unit exists; under the condition that the target consumption of a later time window is less than or equal to a preset second consumption threshold, setting the competition probability of the flow quality layer with the highest threshold value of the pre-estimated conversion interval as a preset first probability, and setting the competition probabilities of other flow quality layers as being less than the preset first probability; or after adjusting the competition participation probability of each flow quality layer each time, adjusting the competition participation probability of the layer with the highest pre-estimated conversion interval threshold value in each flow quality layer with the competition participation probability of 0 to be the preset second probability.

According to an aspect of some embodiments of the present disclosure, there is provided a competition participation control system including: a memory; and a processor coupled to the memory, the processor configured to perform any of the above participating control methods based on instructions stored in the memory.

According to an aspect of some embodiments of the present disclosure, a computer-readable storage medium is proposed, on which computer program instructions are stored, which instructions, when executed by a processor, implement the steps of any of the above participating control methods.

Drawings

The accompanying drawings, which are included to provide a further understanding of the disclosure and are incorporated in and constitute a part of this disclosure, illustrate embodiments of the disclosure and together with the description serve to explain the disclosure and not to limit the disclosure. In the drawings:

FIG. 1 is a flow chart of some embodiments of a participation control method of the present disclosure.

FIG. 2 is a flow diagram of some embodiments of participant probability adjustment in a participant control method of the present disclosure.

FIG. 3 is a flow chart of another embodiment of a participation control method of the present disclosure.

FIG. 4 is a flow diagram of some embodiments of target consumption determination in a participation control method of the present disclosure.

FIG. 5 is a flow chart of some embodiments of expected cumulative consumption determinations in the participation control method of the present disclosure.

Fig. 6 is a flow chart of some embodiments of target total consumption value determination in the participation control method of the present disclosure.

FIG. 7 is a schematic diagram of some embodiments of a participation control system of the present disclosure.

FIG. 8 is a schematic diagram of further embodiments of a participant competition control system of the present disclosure.

FIG. 9 is a schematic diagram of still other embodiments of a race participation control system of the present disclosure.

Detailed Description

The technical solution of the present disclosure is further described in detail by the accompanying drawings and examples.

In the related art, it is often ensured that budget consumption of the current day is completed by accelerating play at night, or budget consumption within a period of time is allocated by dividing the budget into time intervals, or an advertisement effect is ensured by using a PID (proportional-derivative-Integral) feedback adjustment algorithm originally introduced from the field of ship autopilot.

For the technology of guaranteeing the completion of budget consumption of the current day by accelerating the playing at night, the technology may concentrate on participating in competition at night and getting some low-quality traffic at high price at night, so that the advertising effect cannot reach the expectation of an advertiser, and the network pressure may increase rapidly; in addition, the probability of traffic blocking is greatly increased, great pressure is caused to an advertisement platform or an advertiser platform, and meanwhile, for a user, the condition of reducing the access speed is caused, and the user experience is reduced. The method for dividing the budget set by the advertiser into the time periods can guarantee the uniform consumption of the budget all day long, but needs a large amount of manual operation and experience accumulation, has a certain time delay in effectiveness, and has the risk of being unable to track the traffic environment change. Although the PID feedback regulation algorithm can adjust the advertisement putting effect in time, the extreme conditions such as short-time explosion amount of the advertisement cannot be avoided.

A budget control strategy based on a bidding rate is provided in the related technology, a Key Performance Indicator (KPI) is introduced for optimization, the interval estimation of the KPI quantiles exposed in the current period is statistically completed, the KPI bids according to the bidding rate in the interval, the KPI bids when the KPI is higher than the upper bound and does not bid when the KPI is lower than the lower bound. However, there are many parameters to be estimated, and the fluctuation of the paging _ rate is too large and unstable, and the fluctuation of the paging _ rate directly affects the fluctuation of the traffic, resulting in unstable traffic and unstable network.

A flow diagram of some embodiments of the participation control method of the present disclosure is shown in fig. 1.

In step 130, real-time pre-estimated conversion parameters of the advertisement are obtained through the pre-estimation model. In some embodiments, a predictive model may be constructed to predict in real time the advertising effectiveness of the current request as a real-time predictive conversion parameter. The real-time pre-estimated conversion parameter may be an indicator concerned by an advertiser, such as a PCVR (pre-estimated conversion Rate) or a PCTR (pre-estimated Click Through Rate).

In some embodiments, the pre-estimated model structure may vary with the actual conditions of training data magnitude, actual business accuracy requirements, time-consuming requirements, and the like.

In some embodiments, the pre-estimation operation is triggered to be executed each time actual consumption (i.e., participating in a bidding for ads) is generated.

In step 140, a flow quality layer to which real-time consumption at a time corresponding to the real-time estimated transformation parameter belongs is determined according to the real-time estimated transformation parameter and the estimated transformation parameter layering threshold.

In some embodiments, a real-time pre-estimated transformation parameter may be determined after the actual consumption occurs, and a flow quality layer is determined according to the real-time pre-estimated transformation parameter, so that the actual consumption belongs to the determined flow quality layer.

In step 160, for each of the qos layers, the contention probability of each of the qos layers is adjusted according to a difference between an actual consumption of a previous time window and a target consumption of a next time window in two adjacent time windows.

In step 170, advertisement participation is performed according to the adjusted participation probability.

By the method, the actual consumption can be processed in a layered mode according to real-time pre-estimated conversion parameters, the participatory competition probability of each flow quality layer is adjusted by taking a time window as a time processing unit according to the consumption conditions and target consumption conditions of different layers, and therefore the differentiated processing capacity of different quality flows is improved.

In some embodiments, as shown in fig. 1, between step 140 and step 160, step 150 may be further included: it is determined whether a data flow blockage has occurred. If no data traffic is blocked, go to step 160; if the data traffic is blocked, step 151 is executed.

In some embodiments, whether data flow blocking occurs may be determined according to the amount of change in the accumulated value of actual consumption; and under the condition that the variation of the latest accumulated value of actual consumption and the next latest accumulated value of actual consumption is less than or equal to the blocking threshold, determining that data flow blocking occurs, otherwise, determining that no blocking currently occurs.

In some embodiments, whether a blockage occurs may be determined by:

a. maintaining a one-dimensional array, [(s)₁,tm₁)，(s₂,tm₂)，…，(s_n,tm_n)]And n is a hyperparameter and represents the maximum number of elements of the recorded array. s_nTm being the sum of the 1 st to nth consumptions in the time period_nTime stamp for nth consumption;

b. calculating the current latest consumption sum S in the time period of all the advertisement units, and recording the current time stamp T_M(ii) a The advertisement unit refers to the minimum unit of advertisement putting, and can count the exposure click consumption number and the brought order number to measure the putting effect.

c. Update array, [(s)₂,tm₂)，…，(s_n,tm_n)，(S,TM)]And sorting according to the time stamp tm;

d. starting from the last digit of the array and determining if (S-S) is present_i)>interv；

Where i ═ n, n-1, …, 1, interv is a hyperparameter, indicating the occlusion threshold. If the inequality is established, judging that the data is not blocked, and normally executing the step of subsequently updating the participation probability; if the inequality still does not hold after traversing the array, judging that the data flow is blocked.

In step 151, a flow blocking alarm is issued to automatically or manually perform emergency measures. In some embodiments, the probability of participating in a race can be kept unchanged, and the race is directly quitted and an alarm is sent.

By the method, the situation of data flow blockage can be timely found and processed, the running reliability of the participation of the advertisement in competition is improved, the error regulation of the probability of the participation in competition caused by the data flow blockage is also avoided, and the reliability of the participation in competition control is improved.

In some embodiments, to facilitate data management, a hierarchical data structure may be maintained for each ad unit in each ad unit hierarchical data structure statistics module, each layer corresponding to an ad request group. Each advertisement request group may include a real-time budget allocation value, a conversion rate pcvr distribution interval, consumption in a latest time period, and data blocking judgment, where the real-time budget allocation value, i.e., a calculation result generated by the small-scale consumption control, obtains a time-period budget initial value of an advertisement unit granularity.

In the step of counting the real consumption of the time window, the actual consumption of the previous time window, the time span of the previous time window, the target consumption of the next time window and the time span of the next time window are counted. Taking 0 point as the initialization time of each day as an example, the actual consumption of the previous window at the initialization time of 0 point of each day refers to the consumption from 0 point to the current time of the day, and the consumption from the timestamp updated last time to the current time in normal operation refers to the consumption from the last time to the current time. The last window time span refers to the time span between the time stamp of the latest data that can be fetched at the time of initialization of 0 point and the initialization time (0 point that has passed) every day, and refers to the time span between the time stamp of the latest data that can be fetched at the time of normal operation and the time stamp of the end of the last program operation. The next window target consumption refers to a difference between the accumulated target consumption to the end time of the next window and the current actual accumulated consumption. The next window time span is the length of time between the current time and the next nearest window cutoff (the next hour, if it is in hours).

In some embodiments, since the estimated value of the model is affected by various factors, and the hundred percent accuracy cannot be guaranteed, in step 130, after the real-time preliminary estimated transformation parameters are obtained directly based on the estimated model, the current real-time preliminary estimated transformation parameters may be corrected based on the difference between the historical actual transformation parameters and the estimated transformation parameters corresponding to the historical time period. Taking the PCVR as an example, the current PCVR is corrected by using the difference between the actual CVR (conversion Rate, conversion probability) based on the history and the estimated PCVR corresponding to the historical time period, the statistical frequency may be set according to the traffic condition (for example, in hour), and the formula may be as follows:

pcvr_use＝pcvr_predict*(cvr_hour/pcvr_hour)

wherein, pcvr _ predict is an output value of model prediction, i.e. a real-time preliminary prediction transformation parameter, cvr _ hour is a credible cvr value in a history period of time, pcvr _ hour is a model prediction mean value in a corresponding history period of time, and pcvr _ use is a corrected current request prediction value, i.e. the real-time prediction transformation parameter determined in step 130, and is used for executing subsequent processing.

By the method, the deviation of the pre-estimated model can be corrected, the accuracy of actual consumption layering is improved, the requirement of an advertiser on the conversion effect can be met while the flow stability is improved, and the method is favorable for popularization and application of the bidding control scheme for improving the flow stability.

In some embodiments, the calculation cvr _ hour may be performed at ad unit granularity and the threshold number of clicks may be set. If the data volume of the advertisement unit granularity is not enough, the data can be returned in sequence, for example, the data can be returned according to the sequence of an advertiser + a third class → a whole granularity, so that the click data volume can reach a credible and convincing threshold value, and the credibility of the data is ensured.

In some embodiments, pcvr _ hour may be calculated at ad unit granularity and a threshold number of clicks may be set. If the data of the last 1 hour of the granularity of the advertisement unit is insufficient, the data is preferentially returned to the data of 2 hours until the last moment before the model is updated yesterday (the estimated value of the same model is guaranteed to be corrected, and the error caused by the model updating is avoided), and then the logic backspacing according to the sequence of the advertiser + tertiary category → whole granularity is used.

By the method, the data volume to be processed is ensured to have universal significance, influence caused by accidental data is avoided, and the processing reliability is improved.

In some embodiments, the pre-estimated conversion parameter stratification threshold may be determined based on an analysis of historical data. In some embodiments, the participation control method further comprises: according to the distribution of the estimated transformation parameters of the historical dates of the preset days and the preset layering number, dividing the estimated transformation parameters into equal parts of the preset layering number, and acquiring the quantile point parameters for dividing the flow quality layers as the layering threshold of the estimated transformation parameters. By the method, the layering of the flow quality layer can be matched with the actual conversion condition of the advertisement unit, and the accuracy and the reaction efficiency of flow stability control are improved.

In some embodiments, a flow chart of some embodiments of the participation control method in step 160 is shown in fig. 2.

In step 261, a consumption error between the actual consumption of the preceding time window and the target consumption of the following time window in the two adjacent time windows is obtained.

In some embodiments, the difference between the actual consumption of the previous time window and the target consumption of the next time window in the two adjacent time windows can be directly calculated as the consumption error.

In some embodiments, considering that the lengths of the preceding and following time windows may not be consistent, the window length is normalized, and the consumption error is calculated according to the following formula:

residual＝c_next–c_last*(tm_next/tm_last)

where residual is the consumption error, c_nextFor target consumption in a later time window, c_lastTm for the actual consumption of the preceding time window_nextFor the length of the following time window, tm_lastIs the length of the preceding time window.

In step 262, it is determined that the consumption error belongs to a predetermined first interval. In some embodiments, the predetermined first interval may be a numerical interval close to the value 0, for example the predetermined first interval is 0 itself, or [ -0.01,0.01], or the like. The predetermined first interval may be adjusted as desired.

If the consumption error belongs to the predetermined first interval, it indicates that the actual consumption is close to the expected consumption, go to step 263; otherwise, step 264 is performed.

In step 263, the participant race probabilities are not adjusted.

In step 264, it is determined that the consumption error is less than the low threshold of the predetermined first interval. If the consumption error is less than the low threshold of the predetermined first interval, it indicates that the actual consumption is faster than expected, go to step 265; otherwise, indicating that the actual consumption is slower than expected, step 266 is performed.

In step 265, the competition probability of each flow quality layer is sequentially reduced according to the sequence from low to high of the pre-estimated transformation interval threshold corresponding to the flow quality layer. In some embodiments, the lower limit of the adjustment of the contention probability is 0, and if the contention probability of the traffic quality layer being adjusted is already 0, the layer is skipped.

In step 266, the competitive probability of each flow quality layer is sequentially increased according to the sequence from high to low of the pre-estimated transformation interval threshold corresponding to the flow quality layer. In some embodiments, the upper limit of the adjustment of the contention probability is 1, and if the contention probability of the traffic quality layer being adjusted is already 1, the layer is skipped.

In some embodiments, for the determination of the new contention probabilities in

steps

265 and 266 described above, the sum of the consumption error and the actual consumption of the previous time window of the corresponding traffic quality layer may be obtained for each traffic quality layer to determine the adjusted consumption. And further adjusting the competition participation probability of the previous time window according to the proportion of the adjustment consumption to the actual consumption of the previous time window of the corresponding flow quality layer, and acquiring the competition participation probability of the later time window.

In some embodiments, the adjustment may be based on the formula:

p_next(l)＝p_last(l)*(c_last(l)+residual)/c_last(l)

in the formula p_next(l) Is the participant probability, p, of the ith traffic quality layer of the subsequent time window_last(l) Is the probability of participation in the first floor of the preceding time window, residual is the consumption error, c_last(l) And l is the actual consumption of the l layer of the previous time window, and is a flow quality layer identifier, a positive integer and a value range within the layer number range of the flow quality layer.

By the method, the flow can be adjusted in real time according to the difference between consumption and expected consumption, so that sudden increase of flow caused by over-fast consumption and subsequent increase of flow caused by over-slow consumption are avoided, and the flow stability is improved; the method has the advantages that the competition probability of the low-quality flow quality layer is preferentially reduced, the competition probability of the high-quality flow quality layer is preferentially improved, the influence of adjustment on the service of an advertiser can be reduced, and the influence efficiency of adjustment on the flow stability can also be improved.

In some embodiments, the number of the qos layer layers may be initialized to L ═ n, and each layer shares a global contention probability of the super-parameter G. Each layer of the first time window generates certain consumption which can be used as the basis for the adjustment of the probability of the later participation competition.

And under the condition that the actual consumption of the traffic quality layers is smaller than a preset first consumption threshold, raising the advertisement bid according to a preset percentage, and updating the participation probability of each traffic quality layer until the actual consumption of each traffic quality layer is not smaller than the preset first consumption threshold. For example, if an ad unit bid is low, resulting in some layers not being consumed, the ad unit bid may be increased by a percentage until all layers are consumed and are no longer increased. Each adjustment will save the competition probability that the consumption of each layer is not 0 last time, and can be used as the basis for carrying out competition probability adjustment in the mode shown in fig. 2 later.

By the method, the situation that the competition participation probability is initially 0, so that subsequent actual consumption is avoided and the competition participation control is trapped in endless loop can be avoided, and the robustness of the competition participation control is improved.

In some embodiments, corresponding processing may be performed for some special cases.

In some embodiments, the bidding probability of all traffic quality layers of all advertisement units initialized at the starting time of each day is a predetermined first initial value, for example, when the advertising unit is initialized at 0 o' clock each day, the bidding probability of all advertisement units is initialized to the over-parameter G1, so that data deviation caused by daily processing is avoided, and the stability of operation is ensured.

In some embodiments, in the presence of a newly added ad unit, the participation probability of all traffic quality layers of the newly added ad unit is initialized to a predetermined second initial value, for example, the participation probability of each layer of the newly added ad unit in the current day is initialized to a super-parameter G2, and the subsequent first adjustment is adjusted on the basis of G2, so that automatic operation in the case of the newly added ad unit can be realized.

In some embodiments, in the case that the target consumption in the later time window is less than or equal to the predetermined second consumption threshold, the competition probability of the traffic quality layer with the highest threshold value of the pre-estimated transition interval is set as the predetermined first probability, and the competition probabilities of the other traffic quality layers are set to be less than the predetermined first probability. For example, if the target consumption of the next window is less than or equal to 0, the participation probability is set to 0.01 except for the highest layer, and the other layers are set to 0. By the method, overbooking can be reasonably avoided, and the situation that the system falls into a state incapable of being automatically recovered due to the fact that the participation probabilities of all layers are 0 is avoided, and the robustness of the system is improved.

In some embodiments, after adjusting the competition probability of each of the traffic quality layers each time, the competition probability of the layer with the highest predicted transformation interval threshold in each of the traffic quality layers with the competition probability of 0 is adjusted to a predetermined second probability, for example, the highest pcvr layer with a bidding rate of 0 is set to 0.1 for accumulating data.

In some embodiments, in case the actual consumption of the preceding time window is less than or equal to the predetermined second consumption threshold, the actual consumption of the preceding time window is modified to the predetermined third consumption if the consumption error does not belong to the predetermined first interval. For example, if the last window consumption of a certain layer is 0 and the competition probability of the layer needs to be adjusted (the consumption error is not within the predetermined first interval), the last window consumption is set as the default hyper-parameter F. The method can avoid the situation that the subsequent competition-participating probability changes too much under the condition that the actual consumption of the previous window is too small, and further improve the stability of the flow.

In some embodiments, the target consumption for the later time window may be a specified value based on experience or statistics, thereby reducing system stress and increasing processing speed.

In other embodiments, the target consumption for the later time window is a ratio based on the expected consumption value, the actual cumulative consumption ratio for the previous time window, and the expected cumulative consumption for the later time window, or a distribution of expected cumulative consumption, expected single ad costs, and actual single ad costs.

By the method, the target consumption of the later time window can be dynamically determined according to the historical data and the real-time actual consumption condition, the reliability of the determined target consumption of the later time window is improved, and the robustness of the effect of controlling the flow stability is further improved.

In some embodiments, the expected cumulative consumption that needs to be used in determining the target consumption for a later time window may be a specified value based on experience or statistics, thereby reducing system stress and increasing processing speed.

In other embodiments, the expected cumulative consumption for the later time window may be determined based on the distribution of actual consumption on the historical date, the actual cumulative consumption fraction on the current date, and the target total consumption value on the current date. By the method, the expected accumulated consumption can be dynamically determined according to historical data and actual consumption conditions, the reliability of the expected accumulated consumption is improved, and the robustness of the effect of controlling the flow stability is further improved.

In some embodiments, the target total consumption value for the current date required in determining the expected cumulative consumption process may be a specified value based on experience or statistics, thereby reducing system stress and increasing processing speed.

In other embodiments, a target total consumption value for the current date may be determined based on the actual total consumption value for the historical date, the expected single ad cost, and the actual single ad cost. By the method, the target consumption total value of the current date can be dynamically determined, the reliability of the target consumption total value of the current date is improved, and the robustness of the effect of controlling the flow stability is further improved.

A flow chart of further embodiments of the participation control method of the present disclosure is shown in fig. 3.

In step 311, a target total consumption value for the current date is determined based on the actual total consumption value for the historical date, the expected single ad cost, and the actual single ad cost.

In step 312, the expected cumulative consumption for the later time window is determined based on the distribution of actual consumption on the historical date, the actual cumulative consumption percentage on the current date, and the target total consumption value on the current date.

In step 320, a target consumption for the later time window is determined based on the expected consumption value, the actual cumulative consumption fraction of the previous time window and the expected cumulative consumption fraction of the later time window, or based on a distribution of the expected cumulative consumption, the expected single ad cost and the actual single ad cost.

By the method, the total target consumption value and the expected accumulated consumption of the current date and the target consumption of the later time window can be determined step by step, so that the adjustment of the participation probability is in accordance with the real-time condition, and the robustness of the effect of controlling the flow stability is further improved.

In some embodiments, a flow chart of some embodiments of step 320 described above is shown in fig. 4.

In step 4201, the actual conversion value for the current day is obtained. The actual conversion value for the current day refers to the actual conversion value for the current date from the start time to the current time.

In step 4202, it is determined that the actual conversion value is greater than the predetermined first conversion value. In some embodiments, the setting of the predetermined first conversion value may be a confidence threshold value, and if the setting is greater than the confidence threshold value, the current actual conversion value data is trusted data; otherwise, the data is considered to be too small and is an untrusted value. In some embodiments, the predetermined first conversion value may be

If the actual conversion value is greater than the predetermined first conversion value, go to step 4203; otherwise, step 4208 is executed.

In step 4203, a determination is made as to whether the actual individual advertisement cost at the current day is less than the expected individual advertisement cost. If the expected single ad cost is lower, then go to step 4204; otherwise, step 4207 is executed.

In some embodiments, the actual individual ad cost current day may be calculated as the ratio of the actual cumulative consumption current day to the total number of clicks per ad.

In step 4204, an expected consumption value is determined based on the expected single ad cost and the actual conversion value, and a first consumption in a subsequent time window is determined based on the expected consumption value, an actual cumulative consumption fraction in the previous time window, and an expected cumulative consumption fraction in the subsequent time window.

In some embodiments, the expected consumption value resumable _ cost is tcpa order, where tcpa is the expected single ad cost, i.e., the advertiser expected order line cost, and order is the actual conversion value. In some embodiments, since the price of an ad unit may be calculated in units of number of clicks, i.e., a unit price charged per click, the expected single ad cost is the cost of an expected ad being clicked once.

In some embodiments, the first consumption of the later time window may be determined according to the following formula:

target_cost_1＝reasonable_cost/ratio_1*ratio_2，

wherein, target _ cost _1 is the first consumption of the following time window, ratio _1 is the actual accumulated consumption proportion of the preceding time window, and ratio _2 is the expected accumulated consumption proportion of the following time window.

In step 4205, a second consumption for the later time window is determined based on the expected consumption distribution, the expected single ad cost, and the actual single ad cost.

In some embodiments, the second consumption for the later time window may be determined according to the following formula:

target_cost_2＝old_target_cost*(1-UPDATE_RATE*cpa_diff),

cpa_diff＝(tcpa-cpa)/cpa

wherein target _ cost _2 is the second consumption in the following time window; old _ target _ cost is the cumulative target consumption proportion of the expected cut-off point of the previous time window, and the cumulative target consumption proportion refers to the cumulative value of the target consumption proportion from the starting time of the corresponding date to the target time. cpa _ diff is the difference value between the expected single advertisement cost and the actual single advertisement cost, tcpa is the expected single advertisement cost, cpa is the actual single advertisement cost, and UPDATE _ RATE is a hyperparameter representing the UPDATE step.

In step 4206, the larger one of the first consumption and the second consumption is determined as the target consumption in the following time window, i.e., target _ cost is max (target _ cost _1, target _ cost _2), where target _ cost is the target consumption in the following time window.

In step 4207, the target consumption in the later time window is determined according to the distribution of expected cumulative consumption, expected cost of the single ad and actual cost of the single ad, and in some embodiments, the specific operation may be the same as in step 4205, that is, the operation is the same as that in step 4205

target_cost＝target_cost_2。

In step 4208, it is determined whether there is actual consumption data for the historical date, i.e., whether the ad unit is a new ad unit for the current day. If not, the advertisement unit is provided with actual consumption data of historical date. If there is actual consumption data for the historical date, go to step 4209; otherwise, step 4211 is executed.

In step 4209, the weighted sum of the actual cumulative consumption of the previous time window of the day and the actual cumulative consumption at the corresponding time of the previous day is set as the expected consumption value, and step 4210 is further executed. In some embodiments, the expected consumption value may be obtained by weighting and calculating the accumulated actual consumption of the previous day at the time corresponding to the previous time window and the accumulated actual consumption of the current day at the time corresponding to the previous time window, for example, calculating the expected consumption value according to the following formula:

reasonable_cost＝cost_base*A+cost*B

the reasonable _ cost is an expected consumption value, a is a super parameter and can be set according to an actual scene, a + B is 1, cost _ base is the accumulated actual consumption of the previous day at the time corresponding to the previous time window, and cost is the accumulated actual consumption of the current day at the time corresponding to the previous time window.

In step 4210, a target consumption for the later time window is determined based on the expected consumption value, the actual cumulative consumption duty for the current day, and the expected cumulative consumption duty for the later time window.

In some embodiments, the target consumption for the later time window may be calculated according to the following formula:

target_cost＝reasonable_cost/ratio_3*ratio_4

wherein, ratio _1 is the ratio of the actual cumulative consumption of the preceding time window, and ratio _2 is the ratio of the expected cumulative consumption of the following time window.

In step 4211, the target consumption is the product of the current day's actual cumulative consumption duty and the expected all day consumption based on the expected single ad cost.

target_cost＝A1*tcpa*ratio_5

wherein, ratio _5 is the actual accumulated consumption ratio of the previous time window, and a1 is a hyper-parameter and can be set according to the actual scene.

By the method, the target consumption of the later time window can be dynamically determined to serve as an explosion-proof quantity consumption threshold value, and when the unit consumption reaches or exceeds the explosion-proof quantity consumption threshold value, the unit is stopped to continue playing, so that the explosion quantity problem is prevented, and the pressure on a network and a platform is reduced; as input data for controlling the advertisement unit layered competition participating probability, the method can help to produce the competition participating probability, so that stable consumption meeting real-time change can be realized when the advertisement unit is normally played, and the flow stability is improved.

In some embodiments, after calculating the target _ cost according to the actual condition of the advertisement unit, the calculation result is scaled to adjust the severity of consumption control, and the target consumption in the later time window is modified according to the following formula:

target_cost_final＝target_cost*A2

wherein, target _ cost _ final represents the target consumption of the correction in the later time window, and a2 is a hyperparameter representing the relaxation coefficient, and can be set according to the actual scene.

In some embodiments, the time window length may be set to be a fixed one hour, and after obtaining the target _ cost _ final at a certain time point (hour +1), the target consumption of all hours from the (hour +1) point to 23 points on the day is updated circularly, so that,

target_cost_hour[j]＝target_cost_final/ratio_h1*ratio_h[j]

where j belongs to [ hour +2,23], target _ cost _ hour [ j ] represents the updated target consumption at the jth point, ratio _ h1 is the expected (hour +1) th hour consumption fraction, and ratio _ h [ j ] is the expected jth hour consumption fraction.

At this point, after the hour-level consumption control value is updated, the hour granularity can be further refined according to the actual situation, and the hour granularity is shortened to be updated in minutes until the hour granularity is updated in real time.

By the mode, the target consumption of each time period can be determined step by step, so that abundant basic data are provided for subsequent participation adjustment, and the flexibility of the subsequent participation adjustment is improved.

In some embodiments, a flow chart of some embodiments of the expected cumulative consumption determination at step 312 described above is shown in FIG. 5.

In step 5121, the value to be consumed on the current date is determined according to the actual cumulative consumption percentage on the current date and the target total consumption value on the current date.

In step 5122, the expected cumulative consumption for the later time window is determined by distributing the value to be consumed for the current date over the remaining period of the current date based on the distribution of actual consumption for the historical date. In some embodiments, the small-scale consumption distribution proportion of y days of each granularity history can be counted, wherein y is a hyper-parameter representing the number of days of the statistic history, and can be set according to an actual scene.

By the method, the consumption distribution in an ideal state can be determined based on the actual consumption distribution in the historical data, and further, on the basis, the consumption distribution of a subsequent event window is dynamically adjusted according to the actual consumption situation generated in the current day, so that the adaptive degree of the expected accumulated consumption change is improved.

In some embodiments, a flowchart of some embodiments of the target total consumption value determination in step 311 described above is shown in fig. 6.

In step 6101, it is determined whether or not the number of history dates on which the current advertisement unit exists is greater than 1, that is, whether or not it is not an advertisement unit newly on the previous day at the initial time. If the number is larger than 1, the advertisement unit on the previous day is not new, then execute step 6102; otherwise, step 6109 is performed.

In step 6102, a history date to be processed is selected for subsequent processing. In some embodiments, a smaller number may be selected as the number of history dates to be processed from among the number of history dates and the predetermined maximum number, and data of the history dates may be sequentially taken forward from the previous day of the current date.

In step 6103, it is determined whether or not the actual conversion value of the history date of the current process is equal to or greater than a predetermined second conversion value (for example, the predetermined second conversion value is 3). If the value is greater than or equal to the predetermined second conversion value, go to step 6105; otherwise, go to step 6104.

In step 6104, the actual total consumed value cost corresponding to the history date is determined as the first total consumed value target _ cost 1.

In step 6105, the actual total consumption value corresponding to the historical date is adjusted according to the ratio of the difference between the expected single advertisement cost and the actual single advertisement cost to the expected single advertisement cost, and a first total consumption value is obtained.

In some embodiments, the first total consumption value may be calculated according to the following formula:

target_cost1＝cost/(1+w1*math.tanh(cpa_diff))

where, target _ cost1 is a first total consumption value, w1 is a first weight, and math.tanh (cpa _ diff) represents that hyperbolic tangent function operation is performed on cpa _ diff, so as to realize nonlinear estimation on target _ cost 1.

Because the hyperbolic tangent function has the characteristics that the calculation form is simple, the calculation speed is high, the origin is symmetrical, and the gradient gradually becomes small from the origin, the part to the left and right gentle is the state of gradient saturation, can make cpa _ diff adjust the amplitude unanimously under the positive and negative condition, can not have preference to positive or negative, and make the denominator big when cpa _ diff absolute value is less, the adjustment amplitude is little, the denominator is little when cpa _ diff absolute value is great, the adjustment amplitude is big, and then make the target of whole adjustment process unified: i.e. towards cpa _ diff ═ 0.

In step 6106, the first total consumed value and the expected single ad cost are multiplied by the corresponding hyperparameters, respectively, and compared to a predetermined minimum target to determine a second total consumed value in which the largest term is the current date, e.g., according to the following formula

target_cost2＝max(target_cost1*w2,w3*tcpa,w4)

Where target _ cost2 is the second total consumed value for the current date, w2, w3 are the weights of the first total consumed value and the expected single ad cost, respectively, and w4 is the predetermined minimum target.

In step 6107, it is determined whether there is a history date to be processed. If yes, go to step 6102; otherwise, go to step 6108.

In step 6108, a weighted sum of the second total consumed values for different historical dates is obtained as the target total consumed value target _ cost0 for the current date. For example, the historical dates that are closer to the current date may be weighted more heavily, or weighted down or up for a particular date, to adjust the reference value for different dates. In some embodiments, the weights of different historical dates are hyper-parameters, and can be set and adjusted according to actual scenes.

In step 6109, it is judged whether or not the actual conversion value of the history date of the current process is equal to or greater than a predetermined second conversion value. The current history date of processing is the only history date that exists. If the value is greater than or equal to the predetermined second conversion value, go to step 6111; otherwise, step 6110 is performed.

In step 6110, the actual total consumed value corresponding to the history date is determined to be the first total consumed value, i.e., target _ cost1 is equal to cost.

In step 6111, the actual total consumption value corresponding to the historical date is adjusted according to the ratio of the difference between the expected advertisement cost and the actual advertisement cost to the expected advertisement cost, and a first total consumption value, that is, the first total consumption value is obtained

target_cost1＝cost/(1+w5*math.tanh(cpa_diff))

W5 is a hyperparameter, and can be the same as w 1.

In step 6112, the first total consumption value and the expected single advertisement cost are multiplied by the corresponding hyperparameters, respectively, and compared with the predetermined minimum target, and the second total consumption value, the largest one of which is the current date, is determined

target_cost2＝max(target_cost1*w6,w7*tcpa,w8)

Where w6, w7 are the first total consumed value and the weight of the expected single ad cost, respectively, w8 is the predetermined minimum target, w6 may be the same as w2, w7 may be the same as w3, and w8 may be the same as w 4.

In step 6113, the second total consumption value is set as the target total consumption value of the current date, i.e., target _ cost0 is set to target _ cost 2.

By the method, the consumption in the level of days can be controlled based on historical data, the total target consumption value can be dynamically determined, the self-adaption degree is improved, and the maintenance effect of competitive adjustment on the flow stability is facilitated.

A schematic diagram of some embodiments of the participation control system of the present disclosure is shown in fig. 7.

The estimation unit 731 can obtain real-time estimation transformation parameters of the advertisement through the estimation model. In some embodiments, a predictive model may be constructed to predict in real time the advertising effectiveness of the current request as a real-time predictive conversion parameter.

The layering unit 732 can determine a flow quality layer to which real-time consumption at a time corresponding to the real-time estimated transformation parameter belongs according to the real-time estimated transformation parameter and the estimated transformation parameter layering threshold.

The probability adjustment unit 733 can adjust, for each of the traffic quality layers, the competing probability of each of the traffic quality layers based on a difference between the actual consumption of the preceding time window and the target consumption of the succeeding time window in the two adjacent time windows.

The participation unit 734 can participate in the advertisement participation according to the adjusted participation probability.

The participatory bidding control system can carry out layered processing on actual consumption according to real-time pre-estimated conversion parameters, adjust the participatory bidding probability of each flow quality layer by taking a time window as a time processing unit according to consumption conditions and target consumption conditions of different layers, thereby improving the differentiated processing capacity of different quality flows, and can intervene in time on the probability of subsequent consumption, avoid the generation of burst flow, improve the stability of participatory bidding flow on the basis of not reducing or even improving the advertising effect required by an advertiser, and simultaneously contribute to ensuring the access speed and user experience of a user.

In some embodiments, as shown in fig. 7, a block determination unit 735 may be further included, which is capable of determining whether a data flow block occurs. If the data traffic is not blocked, the probability adjustment unit 733 is activated to perform its function; and if the data flow is blocked, sending a flow blocking alarm so as to automatically or manually execute emergency measures. In some embodiments, the probability of participating in a race can be kept unchanged, and the race is directly quitted and an alarm is sent.

The system can timely find and process the data flow blocking condition, improve the running reliability of the advertisement participation competition, avoid the error regulation of the participation competition probability caused by the data flow blocking, and improve the reliability of the participation competition control.

In some embodiments, as shown in fig. 7, a first processing unit 736 may be further included, which can perform corresponding processing for some special cases.

In some embodiments, the first processing unit 736 can initialize the bidding probability of all traffic quality layers of all advertisement units to a predetermined first initial value at the starting time of each day, for example, when initializing at 0 o' clock each day, initialize the bidding probability of all advertisement units to the over-parameter G1, thereby avoiding data deviation caused by processing day by day and ensuring the stability of operation.

In some embodiments, the first processing unit 736 can initialize the competition probability of all traffic quality layers of the newly added ad unit to a predetermined second initial value in the presence of the newly added ad unit, for example, initialize each layer competition probability of the newly added ad unit to a super parameter G2, and adjust the subsequent first adjustment based on G2, thereby implementing automatic operation in the presence of the newly added ad unit.

In some embodiments, the first processing unit 736 can set the candidate probability of the traffic quality layer with the highest threshold of the pre-estimated transition interval to a predetermined first probability and set the candidate probabilities of other traffic quality layers to be less than the predetermined first probability in case the target consumption of the later time window is less than or equal to the predetermined second consumption threshold. For example, if the target consumption of the next window is less than or equal to 0, the participation probability is set to 0.01 except for the highest layer, and the other layers are set to 0. By the method, overbooking can be reasonably avoided, and the situation that the system falls into a state incapable of being automatically recovered due to the fact that the participation probabilities of all layers are 0 is avoided, and the robustness of the system is improved.

In some embodiments, the first processing unit 736 may adjust, after performing the adjustment of the competition probability of each qos layer each time, the competition probability of the layer with the highest pre-estimated transformation interval threshold in each qos layer with the competition probability of 0 to be a predetermined second probability, for example, setting the highest pcvr layer with a bidding rate of 0 to be 0.1, for accumulating data.

In some embodiments, the first processing unit 736 is capable of modifying the actual consumption of the previous time window to a predetermined third consumption if the consumption error does not belong to the predetermined first interval, in case the actual consumption of the previous time window is less than or equal to the predetermined second consumption threshold. For example, if the last window consumption of a certain layer is 0 and the competition probability of the layer needs to be adjusted (the consumption error is not within the predetermined first interval), the last window consumption is set as the default hyper-parameter F. The method can avoid the situation that the subsequent competition-participating probability changes too much under the condition that the actual consumption of the previous window is too small, and further improve the stability of the flow.

In some embodiments, as shown in fig. 7, a stratification threshold determining unit 701 may further be included, configured to divide the predicted transformation parameters into equal parts of a predetermined number of stratification according to the distribution of the predicted transformation parameters on the historical dates of the predetermined number of days and the predetermined number of stratification, and obtain a quantile point parameter for dividing the flow quality layer as the predicted transformation parameter stratification threshold.

The system can ensure that the layering of the flow quality layer can be matched with the actual conversion condition of the advertisement unit, and improve the accuracy and the reaction efficiency of flow stability control.

In some embodiments, as shown in fig. 7, a window target consumption determination unit 721 capable of determining the target consumption in the later time window and serving as the operation base data of the probability adjustment unit 733 may be further included. In some embodiments, the window target consumption determination unit 721 can calculate the target consumption for the following time window based on the expected consumption value, the actual cumulative consumption fraction for the preceding time window and the expected cumulative consumption fraction for the following time window, or based on the distribution of the expected cumulative consumption, the expected single ad cost and the actual single ad cost

The system can dynamically determine the target consumption of the later time window according to the historical data and the real-time actual consumption condition, improve the reliability of the determined target consumption of the later time window, and further improve the robustness of the effect of controlling the flow stability.

In some embodiments, as shown in fig. 7, an expected cumulative consumption determination unit 712 may be further included, which can determine the expected cumulative consumption in the later time window according to the distribution of the actual consumption of the history date, the actual cumulative consumption percentage of the current date, and the target total consumption value of the current date, as a data basis for the operation of the window target consumption determination unit 721.

The system can dynamically determine expected accumulated consumption according to historical data and actual consumption conditions, improve the reliability of the expected accumulated consumption and further improve the robustness of the effect of flow stability control.

In some embodiments, as shown in fig. 7, a target consumption total value determination unit 711 may be further included, which can determine a target consumption total value of the current date according to the actual consumption total value of the historical date, the expected advertisement unit cost and the actual advertisement unit cost, so as to serve as a data basis for the operation of the expected cumulative consumption determination unit 712.

The system can dynamically determine the target total consumption value of the current date, improve the reliability of the target total consumption value of the current date and further improve the robustness of the effect of controlling the flow stability.

A schematic structural diagram of an embodiment of the competition control system of the present disclosure is shown in fig. 8. The participant control system includes a memory 801 and a processor 802. Wherein: the memory 801 may be a magnetic disk, flash memory, or any other non-volatile storage medium. The memory is for storing instructions in the corresponding embodiments of the participant control method above. Coupled to the memory 801, the processor 802 may be implemented as one or more integrated circuits, such as a microprocessor or microcontroller. The processor 802 is configured to execute instructions stored in a memory to improve the smoothness of the participating traffic flows.

In one embodiment, as also shown in fig. 9, the participating control system 900 includes a memory 901 and a processor 902. The processor 902 is coupled to the memory 901 via a BUS 903. The participation control system 900 may also be coupled to an external storage device 905 for invoking external data via storage interface 904, and may also be coupled to a network or another computer system (not shown) via network interface 906. And will not be described in detail herein.

In this embodiment, the data instructions are stored in the memory and then processed by the processor, so as to improve the stability of the participating traffic.

In another embodiment, a computer-readable storage medium has stored thereon computer program instructions which, when executed by a processor, implement the steps of the method in a corresponding embodiment of the participation control method. As will be appreciated by one skilled in the art, embodiments of the present disclosure may be provided as a method, apparatus, or computer program product. Accordingly, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present disclosure is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the disclosure. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

Thus far, the present disclosure has been described in detail. Some details that are well known in the art have not been described in order to avoid obscuring the concepts of the present disclosure. It will be fully apparent to those skilled in the art from the foregoing description how to practice the presently disclosed embodiments.

The methods and apparatus of the present disclosure may be implemented in a number of ways. For example, the methods and apparatus of the present disclosure may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order for the steps of the method is for illustration only, and the steps of the method of the present disclosure are not limited to the order specifically described above unless specifically stated otherwise. Further, in some embodiments, the present disclosure may also be embodied as programs recorded in a recording medium, the programs including machine-readable instructions for implementing the methods according to the present disclosure. Thus, the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.

Finally, it should be noted that: the above examples are intended only to illustrate the technical solutions of the present disclosure and not to limit them; although the present disclosure has been described in detail with reference to preferred embodiments, those of ordinary skill in the art will understand that: modifications to the specific embodiments of the disclosure or equivalent substitutions for parts of the technical features may still be made; all such modifications are intended to be included within the scope of the claims of this disclosure without departing from the spirit thereof.

Claims

1. A participation control method comprises the following steps:

acquiring real-time pre-estimated conversion parameters of the advertisement through a pre-estimated model;

determining a flow quality layer to which real-time consumption at the moment corresponding to the real-time pre-estimated transformation parameters belongs according to the real-time pre-estimated transformation parameters and the pre-estimated transformation parameter layering threshold;

aiming at each flow quality layer, adjusting the competition probability of each flow quality layer according to the difference between the actual consumption of the previous time window and the target consumption of the next time window in the two adjacent time windows;

and carrying out advertisement participation according to the adjusted participation probability.

2. The method of claim 1, wherein,

the adjusting the competition participation probability of each flow quality layer comprises the following steps:

acquiring consumption errors of actual consumption of a previous time window and target consumption of a subsequent time window in two adjacent time windows;

in the case that the consumption error belongs to a predetermined first interval, not adjusting the participation probability;

and under the condition that the consumption error does not belong to the preset first interval, sequentially adjusting the competition probability of each flow quality layer according to a preset adjusting strategy according to the magnitude sequence of the pre-estimated conversion interval threshold corresponding to the flow quality layer.

3. The method of claim 2, wherein the obtaining of the consumption error of the actual consumption of the previous time window and the target consumption of the subsequent time window in the two adjacent time windows comprises:

according to the time length ratio of the subsequent time window to the previous time window, adjusting the actual consumption of the previous time window to obtain the proportional change consumption;

and acquiring the difference value between the target consumption and the proportional change consumption of the later time window as the consumption error.

4. The method of claim 2, wherein,

under the condition that the consumption error does not belong to the preset first interval, sequentially adjusting the competition participation probability of each flow quality layer according to a preset adjustment strategy and the size sequence of the pre-estimated conversion interval threshold corresponding to the flow quality layer comprises the following steps:

under the condition that the consumption error is smaller than a low threshold value of a preset first interval, sequentially reducing the competition probability of each flow quality layer according to the sequence from low to high of the threshold value of the pre-estimated transformation interval corresponding to the flow quality layer; and

and under the condition that the consumption error is larger than a high threshold value of a preset first interval, sequentially improving the competition participation probability of each flow quality layer according to the sequence from high to low of the pre-estimated conversion interval threshold values corresponding to the flow quality layers.

5. The method according to any one of claims 2 to 4, wherein the adjusting the contention probability of each traffic quality layer according to a predetermined adjustment strategy comprises:

for each of the flow mass layers,

obtaining a sum of the consumption error and the actual consumption of the preceding time window of the corresponding flow mass layer, and determining an adjustment consumption;

and adjusting the competition participation probability of the previous time window according to the proportion of the adjustment consumption to the actual consumption of the previous time window of the corresponding flow quality layer, and acquiring the competition participation probability of the later time window.

6. The method of claim 1, further comprising:

and under the condition that the actual consumption of the traffic quality layers is smaller than a preset first consumption threshold, raising the advertising bid according to a preset percentage, and updating the participation probability of each traffic quality layer until the actual consumption of each traffic quality layer is not smaller than the preset first consumption threshold.

7. The method of claim 1, further comprising:

before determining and adjusting the competitive probability of each flow quality layer, determining whether data flow blockage occurs according to the variable quantity of the actually consumed accumulated value;

determining that data flow blocking occurs when the difference between the latest accumulated value of actual consumption and the previous accumulated value of actual consumption of a predetermined number is less than or equal to a blocking threshold;

otherwise, executing the operation of adjusting the competition probability of each flow quality layer according to the difference between the actual consumption of the previous time window and the target consumption of the next time window in the two adjacent time windows aiming at each flow quality layer.

8. The method of claim 1, further comprising:

the target consumption for the later time window is determined based on the expected consumption value, the actual cumulative consumption duty for the earlier time window and the expected cumulative consumption duty for the later time window, or based on a distribution of expected cumulative consumption, expected single ad costs and actual single ad costs.

9. The method of claim 8, wherein the determining the target consumption for the later time window based on the expected consumption value, the actual cumulative consumption duty for a prior time window and the expected cumulative consumption duty for the later time window, or based on a distribution of expected cumulative consumption, expected single ad costs and actual single ad costs comprises:

and under the condition that the actual conversion value is larger than the preset first conversion value, if the current actual single advertisement cost is not lower than the expected single advertisement cost on the current day, determining the target consumption of the later time window according to the distribution of the expected accumulated consumption, the expected single advertisement cost and the actual single advertisement cost.

10. The method of claim 9, wherein the determining the target consumption for the later time window based on the expected consumption value, the actual cumulative consumption duty for a prior time window and the expected cumulative consumption duty for the later time window, or based on a distribution of expected cumulative consumption, expected single ad costs and actual single ad costs further comprises:

Determining the expected consumption value based on the expected single-ad cost and the actual conversion value, and determining a first consumption for the following time window based on the expected consumption value, an actual cumulative consumption duty for the preceding time window, and an expected cumulative consumption duty for the following time window;

determining a second consumption of the later time window based on the expected consumption profile, expected individual ad costs, and actual individual ad costs;

determining a larger one of the first consumption and the second consumption as a target consumption of the later time window.

11. The method of claim 8, wherein the determining a target consumption for the later time window based on an expected consumption value, an actual cumulative consumption duty for a prior time window, and an expected cumulative consumption duty for the later time window, or based on a distribution of the expected cumulative consumption, an expected single ad cost, and an actual single ad cost, further comprises:

in the case where the actual conversion value is not greater than the predetermined first conversion value, if there is actual consumption data of the history date, then

Taking a weighted sum of the actual cumulative consumption of the previous time window of the day and the actual cumulative consumption at the corresponding time of the previous day as the expected consumption value;

determining a target consumption for the later time window based on the expected consumption value, the actual accumulated consumption duty for the current day, and the expected accumulated consumption duty for the later time window.

12. The method of claim 8, further comprising:

in the case where the actual conversion value is not greater than the predetermined first conversion value, if there is no actual consumption data of the history date, the target consumption is taken as the product of the actual cumulative consumption proportion currently on the day and the expected all-day consumption based on the expected single advertisement cost.

13. The method of claim 8, further comprising: and determining the expected accumulated consumption of the later time window according to the distribution of the actual consumption of the historical date, the actual accumulated consumption ratio of the current date and the target total consumption value of the current date.

14. The method of claim 13, further comprising:

and determining a target total consumption value of the current date according to the actual total consumption value of the historical date, the expected single advertisement cost and the actual single advertisement cost.

15. The method of claim 13, wherein the determining the expected cumulative consumption for the later time window comprises:

determining a value to be consumed of the current date according to the actual accumulated consumption ratio of the current date and the target total consumption value of the current date;

and according to the distribution of the actual consumption of the historical date, distributing the value to be consumed of the current date in the remaining period of the current date, and determining the expected accumulated consumption of the later time window.

16. The method of claim 14, wherein the determining the target total consumption value comprises:

under the condition that the actual conversion value of the historical date is larger than or equal to a preset second conversion value, adjusting the actual total consumption value corresponding to the historical date according to the proportion of the difference between the expected single advertisement cost and the actual single advertisement cost to the expected single advertisement cost, and obtaining a first total consumption value; determining the actual total consumption value corresponding to the historical date as the first total consumption value under the condition that the actual conversion value corresponding to the historical date is smaller than a preset second conversion value;

multiplying the first total consumption value and the expected single advertisement cost respectively by corresponding hyperparameters, comparing the hyperparameters with a preset minimum target, and determining a second total consumption value of which the largest item is the current date;

and in the case that the number of the historical dates is more than 1, acquiring the weighted sum of the second total consumption values of different historical dates as the target total consumption value of the current date.

17. The method of claim 16, the determining the target total consumption value based on historical date total actual consumption values, expected individual ad costs, and actual individual ad costs further comprising:

and in the case that the number of the historical dates is equal to 1, taking the second total consumption value as a target total consumption value of the current date.

18. The method of claim 1, further comprising: dividing the estimated transformation parameters into equal parts of preset layering quantity according to the distribution of the estimated transformation parameters of the historical dates of preset days and the preset layering quantity, and acquiring the quantile point parameters for dividing the flow quality layers as the layering threshold of the estimated transformation parameters.

19. The method of claim 1, further comprising at least one of:

initializing the competition participation probability of all the flow quality layers of all the advertisement units to be a preset first initial value at the starting moment of each day;

initializing the competition participation probability of all the flow quality layers of the newly added advertisement unit to be a preset second initial value under the condition that the newly added advertisement unit exists;

under the condition that the target consumption of the later time window is less than or equal to a preset second consumption threshold, setting the competition probability of the flow quality layer with the highest threshold value of the pre-estimated conversion interval as a preset first probability, and setting the competition probabilities of other flow quality layers as being less than the preset first probability; or

After the adjustment of the competition participation probability of each flow quality layer is performed each time, the competition participation probability of the layer with the highest pre-estimated conversion interval threshold value in each flow quality layer with the competition participation probability of 0 is adjusted to be a preset second probability.

20. The method of claim 2, further comprising:

and under the condition that the actual consumption of the prior time window is less than or equal to a preset second consumption threshold, if the consumption error does not belong to the preset first interval, modifying the actual consumption of the prior time window into a preset third consumption.

21. The method of claim 1, wherein the bidding probability has an upper adjustment limit of 1 and a lower adjustment limit of 0.

22. A participant competition control system comprising:

the estimation unit is configured to obtain real-time estimation transformation parameters of the advertisements through the estimation model;

the layering unit is configured to determine a flow quality layer to which real-time consumption at a moment corresponding to the real-time pre-estimated conversion parameter belongs according to the real-time pre-estimated conversion parameter and a pre-estimated conversion parameter layering threshold;

the probability adjusting unit is configured to adjust the competition probability of each flow quality layer according to the difference between the actual consumption of the previous time window and the target consumption of the next time window in two adjacent time windows aiming at each flow quality layer;

and the competition participating unit is configured to participate in the advertisement competition according to the adjusted competition participating probability.

23. The system of claim 22, further comprising an occlusion determination unit configured to:

otherwise, the probability adjustment unit is activated.

24. The system of claim 22, further comprising:

a window target consumption determination unit configured to determine a target consumption for a following time window based on an expected consumption value, an actual accumulated consumption proportion for a preceding time window and an expected accumulated consumption proportion for the following time window, or based on a distribution of expected accumulated consumption, an expected single advertisement cost and an actual single advertisement cost.

25. The system of claim 22, further comprising:

an expected cumulative consumption determination unit configured to determine the expected cumulative consumption of the later time window according to the distribution of the actual consumption of the history date, the actual cumulative consumption proportion of the current date, and the target total consumption value of the current date.

26. The system of claim 25, further comprising:

and a target total consumption value determining unit configured to determine a target total consumption value of the current date according to the actual total consumption value of the historical date, the expected single advertisement cost and the actual single advertisement cost.

27. The system of claim 22, further comprising:

and the layering threshold determining unit is configured to divide the estimated transformation parameters into equal parts of preset layering quantities according to the distribution of the estimated transformation parameters of the historical dates of preset days and preset layering quantities, and acquire the quantile point parameters for dividing the flow quality layers as the layering threshold of the estimated transformation parameters.

28. The system of claim 22, further comprising a first processing unit further configured to perform at least one of:

29. A participant competition control system comprising:

a memory; and

a processor coupled to the memory, the processor configured to perform the method of any of claims 1-21 based on instructions stored in the memory.

30. A computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the steps of the method of any one of claims 1 to 21.