CN109377280B - Advertisement sequencing mechanism generation method and generation system - Google Patents
Advertisement sequencing mechanism generation method and generation system Download PDFInfo
- Publication number
- CN109377280B CN109377280B CN201811260738.6A CN201811260738A CN109377280B CN 109377280 B CN109377280 B CN 109377280B CN 201811260738 A CN201811260738 A CN 201811260738A CN 109377280 B CN109377280 B CN 109377280B
- Authority
- CN
- China
- Prior art keywords
- service
- increment
- display
- services
- class
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0242—Determining effectiveness of advertisements
- G06Q30/0244—Optimization
Abstract
The invention discloses a method and a system for generating an advertisement sequencing mechanism, relates to the technical field of advertisement sequencing mechanism calculation, and aims to solve the problem that the existing advertisement sequencing mechanism cannot support a scene with 0 advertisement display positions. The key points of the technical scheme are as follows: step S1, configuring a class-I characteristic parameter interval, and filtering services of the class-I characteristic parameters outside the class-I characteristic parameter interval; s2, constructing a prediction sample, and obtaining the purchase rate of the user to the service and the conversion rate to the main demand through a classification decision algorithm and the prediction sample; and step S3, generating a list of non-display occupation ratios of the display times accounting for the total display times with the number of the advertisement display positions being 0, calculating corresponding service revenue increment according to the non-display occupation ratios, and generating a sequencing index calculation formula of the main demand conversion single increment/the service revenue increment reaching a set value. The technical scheme of the application has the effect of being capable of adapting to the scene with the number of the advertisement display positions being 0.
Description
Technical Field
The invention relates to the technical field of advertisement sorting mechanism calculation, in particular to an advertisement sorting mechanism generation method and an advertisement sorting mechanism generation system.
Background
At present, more sorting mechanisms taking platform revenue as an advertisement sorting index are taken out to be used as a Strategy module for independent research, for example, the OCPC sorting mechanism is designed from two sorting indexes of advertiser revenue + α× platform revenue and platform revenue, so that candidate services are sorted, and the purposes of improving advertiser revenue and platform revenue are achieved.
Based on the above-mentioned purpose that promotes advertiser revenue and platform revenue, some internet platforms and internet enterprises have also promoted different online advertisement putting schemes, for example:
d1: the Chinese patent with the application number of '201510947002', which is applied by Beijing Qihu technology Limited company on 12, 16.2015, discloses a prompt window advertisement space release control method and device, which releases a schedule for a plurality of sub-window advertisement spaces and different types of sub-window advertisement spaces, so that a prompt window popped up by a browser client performs advertisement display on the plurality of sub-window advertisement spaces, namely, the display of the plurality of sub-window advertisement spaces can be supported, the displayed advertisement content is richer, the advertisement data of different sub-window advertisement spaces and different types of sub-window advertisement spaces can be conveniently adjusted, and the display efficiency is effectively improved.
D2: the Chinese patent with application number '201611242538' applied by Xinlang net technology (China) Co., Ltd in 2016, 12, 29 discloses an on-line advertisement display method and device, the method comprises: selecting all advertisements meeting current business requirements as a first candidate advertisement set for displaying a page browser on a current advertisement position; acquiring a click rate threshold value of each advertisement in the first candidate advertisement set, and estimating an estimated click rate of each advertisement in the first candidate advertisement set; judging whether each advertisement selects a current page browser or not according to the estimated click rate and the click rate threshold value of each advertisement in the first candidate advertisement set, and taking all advertisements selecting the current page browser as a second candidate advertisement set; selecting the displayed advertisement for the current page viewer according to the putting tension coefficient and the estimated click rate of each advertisement in the second candidate advertisement set; and displaying the selected displayed advertisement to a page browser, thereby improving the click rate of the advertisement.
As can be seen from the above documents, there are already perfect technical solutions for improving advertiser revenue and platform revenue based on improving advertisement display efficiency and advertisement click rate, but in the prior art, the number of advertisement display slots is a static value N (N ≧ 1), and there is no consideration for a scene where the number of advertisement display slots may be 0, for example, the number of advertisement display slots may be a positive number N greater than 0 or may also be 0, and in a service scene of a sale, the number of advertisement display slots is often 0, but the prior art does not support, and therefore, a new solution is proposed in the present application.
Disclosure of Invention
The invention aims to provide an advertisement sequencing mechanism generation method and an advertisement sequencing mechanism generation system, which have the effect of being capable of adapting to a scene with 0 number of advertisement display positions.
The above object of the present invention is achieved by the following technical solutions:
an advertisement ranking mechanism generating method comprises the following steps:
step S1, configuring a class of characteristic parameter interval allowing the display of the service, filtering the service of which the class of characteristic parameter is outside the class of characteristic parameter interval, and putting the rest service into a recall pool;
step S2, constructing a prediction sample according to historical data information, and obtaining the purchase rate of a user to service and the conversion rate to main demand through a classification decision algorithm and the prediction sample;
and S3, generating a list of non-display occupation ratios of the display times accounting for the total display times with the number of the advertisement display positions being 0 according to the historical data information, calculating corresponding main demand conversion unit quantity increment and service revenue increment according to the non-display occupation ratios in the list of the non-display occupation ratios and the purchase rate of the user for the service and the conversion rate of the main demand in the step S2, and generating a sequencing index calculation formula of dividing the main demand conversion unit quantity increment by the service revenue increment to reach a set value.
By adopting the technical scheme, the dynamic property of the number of the advertisement display positions can be captured by not displaying the proportion, and the method is suitable for the scene that the number of the advertisement display positions can be 0. On the other hand, the user experience can be directly fed back through the main demand conversion single quantity, the relation between the user experience and the platform revenue can be directly fed back through the ROI-like index of the main demand conversion single quantity increment divided by the service revenue increment, and the optimization of the user experience is more direct.
The invention is further configured to: the step S1 includes the following sub-steps:
s11, configuring a class of characteristic upper limit parameters allowing the display of the service, filtering the service of which the class of characteristic upper limit parameters is greater than or equal to the class of characteristic upper limit parameters, and putting the rest of the service into a recall pool;
and S12, detecting the number of the services in the recall pool, and if the number of the services in the recall pool is 0, adding the service with the minimum characteristic parameter in the filtered services and putting the services into the recall pool again.
By adopting the technical scheme, the data can be effectively screened, and the method is suitable for the scene that the number of the advertisement display positions can be 0. On the other hand, the user experience is also increased through the setting of the upper limit parameters of the features of the same type.
The invention is further configured to: the step S2 includes the following sub-steps:
s21, acquiring historical data and assembling the historical data into a feature data set, and learning a service purchase rate predictor and a main demand conversion rate predictor on the feature data set by using a classification decision algorithm;
s22, polling the service in the recall pool, and splicing the service with the characteristic in the step S21 as a prediction sample;
s23, taking the prediction sample as a reference, inputting the reference into a service purchase rate predictor, and obtaining the purchase rates of different users for services in different scenes and when different services are bought; and taking the prediction sample as a reference, inputting the reference into a main demand conversion rate predictor, and obtaining the conversion rate of different users to the main demand in different scenes and different service selling processes.
By adopting the technical scheme, the purchase rate of the service and the conversion rate of the main demand during pre-estimation (different users, different services sold in different scenes) are facilitated, and the reliability of the data is improved.
The invention is further configured to: the step S21 includes the following sub-steps:
s211, acquiring historical data needing to be estimated, and assembling a user tag of a user id by a system according to the user id in the historical data;
s212, assembling a feature data set comprising practice class features according to the time class data in the user tags;
and S213, learning a service purchase rate predictor and a main demand conversion rate predictor on the characteristic data set by using a classification decision algorithm.
By adopting the technical scheme, the data accuracy is higher through the practice type characteristics of the user id, the user experience is better optimized, and the advertising owner revenue and the platform revenue are improved.
The invention is further configured to: the step S3 includes the following sub-steps:
s31, acquiring historical data, generating a list of non-display ratios of 0 display times to the total display times according to m% of one step length, and generating a list of coefficients α in a ranking index calculation formula according to n step lengths;
s32, according to the purchase rate of the user to the service and the conversion rate of the main demand in the step S23, calculating the main demand conversion unit increment and the service revenue increment on the binary group which does not show the occupation ratio and the coefficient α, and obtaining a mapping table α which enables the main demand conversion unit increment to be divided by the service revenue increment to be maximum under the condition that the occupation ratio is not shown and the occupation ratio is not shown;
s33, randomly selecting one α from the mapping table in the step S32 to generate an initial ranking index calculation formula;
s34, calculating the ranking index value of each advertisement display request on the line according to the initial ranking index calculation formula and forming a ranking index sequence according to the ranking index value;
s35, obtaining quantiles of the sequencing index sequence according to q step lengths, calculating corresponding service revenue increment according to the quantiles, and outputting a mapping table of the quantiles and the service revenue increment under the quantiles;
s36, according to the set service revenue increment, obtaining the corresponding quantile in the mapping table in the step S35, and taking the quantile as the corresponding non-display proportion to obtain the corresponding α in the mapping table in the step S32;
s37, updating α of the ranking index calculation formula according to α of the step S36, and updating the mapping table of the step S35 according to the updated ranking index calculation formula.
By adopting the technical scheme, the method can be used for adjusting the ROI-like index of the main demand conversion single increment divided by the service revenue increment in a linkage manner to reach an optimal sequencing mechanism according to the expected service revenue increment, so that the control on the relation between the user experience and the service revenue is enhanced, and the user experience is better.
The second aim of the invention is realized by the following technical scheme:
an advertisement ranking mechanism generation system comprising:
the service limiting module is used for configuring a class of characteristic parameter intervals allowing the services to be displayed, filtering the services with class of characteristic parameters outside the class of characteristic parameter intervals and putting the rest services into a recall pool;
the prediction module is used for constructing a prediction sample according to historical data information and obtaining the purchase rate of a user to service and the conversion rate of the user to main demand through a classification decision algorithm and the prediction sample;
the sorting mechanism generating module is used for generating a list of non-display ratios of display times of which the number of the advertisement display positions is 0 to the total display times according to the historical data information;
the sorting mechanism generating module is further used for calculating corresponding main demand conversion single quantity increment and service revenue increment according to the non-display proportion in the list without the display proportion, the purchase rate of the user to the service and the conversion rate of the main demand, and generating a sorting index calculating formula of dividing the main demand conversion single quantity increment by the service revenue increment to reach a set value.
By adopting the technical scheme, the sequencing mechanism generation module can capture the dynamic property of the number of the advertisement display positions and is suitable for the scene that the number of the advertisement display positions can be 0. On the other hand, the user experience can be directly fed back through the main demand conversion single quantity, the relation between the user experience and the platform revenue can be directly fed back through the ROI-like index of the main demand conversion single quantity increment divided by the service revenue increment, and the optimization of the user experience is more direct.
The invention is further configured to: the service restriction module includes:
the upper limit limiting unit is used for configuring a class of characteristic upper limit parameters allowing the services to be displayed, filtering the services of which the class of characteristic upper limit parameters is greater than or equal to the class of characteristic upper limit parameters, and putting the rest services into the recall pool;
and the lower limit limiting unit is used for detecting the number of the services in the recall pool and adding the service with the minimum characteristic parameter in the filtered services to the recall pool again when the number of the services in the recall pool is 0.
By adopting the technical scheme, the upper limit limiting unit and the lower limit limiting unit can effectively screen the data, the method is suitable for the scene that the number of the advertisement display positions can be 0, and due to the setting of the upper limit parameters of the first class of characteristics, the service which does not meet the conditions is filtered, and the user experience is increased.
The invention is further configured to: the estimation module comprises:
the predictor training unit is used for acquiring historical data and assembling the historical data into a characteristic data set, and is also used for learning a service purchase rate predictor and a main demand conversion rate predictor on the characteristic data set by using a classification decision algorithm;
a sample construction unit for polling services in the recall pool and stitching features in the service and feature data sets into predicted samples;
the prediction unit is used for inputting the prediction sample as an input parameter into a service purchase rate predictor to obtain the purchase rates of different users for services in different scenes and when different services are bought by sale; the prediction sample is used as an input parameter and is input into a main demand conversion rate predictor, and the conversion rates of different users to main demands in different scenes and different services are obtained.
By adopting the technical scheme, the purchase rate of the service and the conversion rate of the main demand during pre-estimation (different users, different services sold in different scenes) are facilitated, the conversion single increment of the main demand and the service revenue increment are conveniently calculated, and the reliability of the data is increased.
The invention is further configured to: the predictor training unit is also used for assembling a user label of the user id according to the user id in the historical data, and the features in the feature data set comprise practice class features assembled according to the time class data in the user label.
By adopting the technical scheme, the practice type characteristics of the user id are fully captured, so that the data accuracy is higher, and the optimization of user experience is facilitated.
The invention is further configured to: the ranking mechanism generation module comprises:
the non-display occupation and coefficient α mapping table generating module is used for generating a non-display occupation list of which the number of display times of the advertisement display positions is 0 in the total display times according to m% of one step after acquiring the historical data, generating a list of coefficients α in a sorting index calculation formula according to n one step, calculating a main demand conversion unit increment and a service operation increment on a binary group of the non-display occupation and coefficients α according to the purchase rate of the service and the conversion rate of the main demand of the user, and obtaining a first mapping table α which enables the main demand conversion unit increment to be divided by the service operation increment to be maximum under the non-display occupation and the non-display occupation;
the non-display occupation and service revenue increment mapping table generation module is used for randomly selecting α from the mapping table I to generate an initial sequencing index calculation formula, calculating the sequencing index value of each advertisement display request on the line according to the initial sequencing index calculation formula and forming a sequencing index sequence according to the sequencing index value, and is also used for obtaining the quantile of the sequencing index sequence according to q one step length, calculating the corresponding service revenue increment according to the quantile and outputting a mapping table II of the quantile and the service revenue increment under the quantile;
and the sequencing calculation formula updating module is used for obtaining a corresponding quantile in the second mapping table according to the set service revenue increment, taking the quantile as a corresponding non-display duty ratio and obtaining a corresponding α in the first mapping table, and is also used for updating α in the sequencing index calculation formula according to the corresponding α and updating the second mapping table according to the updated sequencing index calculation formula.
By adopting the technical scheme, α in the sequencing index calculation formula can be updated in a linkage manner according to the expected service revenue increment, so that a sequencing mechanism which enables the ROI-like index of the main demand conversion single increment divided by the service revenue increment to reach the optimum is generated according to the updated sequencing index calculation formula, the control on the relation between the user experience and the service revenue is enhanced, and the user experience is better.
In conclusion, the beneficial technical effects of the invention are as follows:
1. through the setting of non-display proportion, the dynamic property of the number of the advertisement display positions can be captured, and the method is suitable for the scene that the number of the advertisement display positions is 0;
2. by converting the setting of the ROI-like index of the single increment divided by the service revenue increment, the relation between the user experience and the platform revenue can be directly fed back, and the user experience is conveniently optimized;
3. through the setting of two mapping tables, the control on the relation between the user experience and the service revenue is enhanced, and the improvement of the user experience and the service revenue is facilitated.
Drawings
FIG. 1 is a flow chart illustrating a method for generating an advertisement ranking mechanism according to an embodiment of the invention;
fig. 2 is a flowchart of step S1 according to the second embodiment of the present invention;
FIG. 3 is a flow chart illustrating a service id filtering process according to a second embodiment of the present invention;
fig. 4 is a flowchart of step S2 according to the second embodiment of the present invention;
fig. 5 is a flowchart of step S21 according to the second embodiment of the present invention;
fig. 6 is a flowchart of step S3 according to the second embodiment of the present invention;
fig. 7 is a schematic structural diagram of an advertisement ranking mechanism generation system according to a third embodiment of the present invention.
In the figure, the device comprises a service limiting module 10, an upper limit limiting unit 11, a lower limit limiting unit 12, a pre-estimating module 20, a pre-estimating training unit 21, a pre-estimating training unit 22, a sample constructing unit 23, a pre-estimating unit 30, a sorting mechanism generating module 31, a mapping table generating module for not displaying the occupation ratio and the coefficient α, a mapping table generating module for not displaying the occupation ratio and the service revenue increment 32, and a sorting calculation formula updating module 33.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
Example one
Referring to fig. 1, a method for generating an advertisement ranking mechanism disclosed by the present invention includes the following steps:
step S1, configuring a class of characteristic parameter interval allowing the display of the service, filtering the service with the class of characteristic parameter outside the class of characteristic parameter interval, and putting the rest service into a recall pool. Specifically, in this embodiment, the first-class characteristic parameter interval is a price interval of the service, and the service in the price interval is placed in the recall pool to ensure user experience.
Although the application scenario of the present application is described by taking price as a class of characteristic parameters as an example in the present application, those skilled in the art can understand that the technical solution of the present application is also applicable to various parameter scenarios, such as advertisement delivery ratio, advertisement delivery time, and the like, and the present application is not limited to this specifically.
And step S2, constructing a prediction sample according to the historical data information, and obtaining the purchase rate of the user to the service and the conversion rate to the main demand through a classification decision algorithm and the prediction sample. In the embodiment, the XGBoost learner is used in the classification decision algorithm.
And S3, generating a list of non-display occupation ratios of the display times accounting for the total display times with the number of the advertisement display positions being 0 according to the historical data information, calculating corresponding main demand conversion unit quantity increment and service revenue increment according to the non-display occupation ratios in the list of the non-display occupation ratios and the purchase rate of the user for the service and the conversion rate of the main demand in the step S2, and generating a sort index calculation formula of the type ROI index of the main demand conversion unit quantity increment divided by the service revenue increment to reach a set value.
It should be noted that the dynamics of capturing the number of advertisement display positions can be captured by not showing the proportion, the user experience can be directly fed back through the main demand conversion unit increment, and the relationship between the user experience and the platform revenue can be directly fed back through the ROI-like index of dividing the main demand conversion unit increment by the service revenue increment, wherein the main demand conversion unit increment = (the main demand conversion rate when a certain service is shown-the main demand conversion rate when no task service is shown) × 1, the service revenue increment = (the service purchase rate when a certain service is shown-the service purchase rate when no task service is shown) × a certain service price, the ranking index calculates the main demand conversion unit increment standardized by (1- α) × + the service revenue increment standardized by α×, and α is a real number between 0 and 1.
The implementation principle of the above embodiment is as follows:
through not showing the dynamic of the number of the proportion capture advertisement display positions, through the direct feedback user experience of main demand conversion single quantity increment, through the direct feedback user experience of the similar ROI index of main demand conversion single quantity increment divided by service revenue increment and platform revenue. Through the technical scheme, the method and the device can be used for adjusting the order mechanism for enabling the similar ROI index to reach the optimal in a linkage manner according to the expected service revenue increment, not only can be suitable for the scene that the number of the advertisement display positions is 0, but also strengthens the control on the relation between user experience and service revenue.
Example two
Referring to fig. 2, based on the first embodiment, the difference between the present embodiment and the first embodiment is that step S1 includes the following sub-steps:
s11, configuring a class of characteristic upper limit parameters allowing to display the service, filtering the service with the class of characteristic upper limit parameters being larger than or equal to the class of characteristic upper limit parameters, and putting the rest service into a recall pool.
And S12, detecting the number of the services in the recall pool, and if the number of the services in the recall pool is 0, adding the service with the minimum characteristic parameter in the filtered services and putting the services into the recall pool again.
Specifically, referring to fig. 3, after the service id is input, it is determined whether the service price divided by the main demand price is equal to or greater than a set upper limit parameter. And if the service price of the part of services is divided by the main demand price and is more than or equal to the set upper limit parameter, filtering the service id of the part of services, and putting the rest service id into the recall pool. And if the service price of the service divided by the main demand price is smaller than the set upper limit parameter, all the service ids are put into the recall pool. Before outputting the service id in the recall pool, whether the number of the service id in the recall pool is equal to 0 or not is detected. If the price is equal to 0, adding the service id with the lowest price in all the services and putting the service id into a recall pool; and if not, the recall pool does not process and outputs the service id in the recall pool.
Referring to fig. 4, step S2 includes the following sub-steps:
and S21, acquiring historical data, assembling the historical data into a feature data set, and learning a service purchase rate predictor and a main demand conversion rate predictor on the feature data set by using an XGboost learner.
S22, polling the service in the recall pool, splicing the service and the characteristics in the step S21 into prediction samples.
S23, taking the prediction sample as a reference, inputting the reference into a service purchase rate predictor, and obtaining the purchase rates of different users for the service in different scenes and when different services are bought; and taking the prediction sample as an input parameter, inputting the prediction sample into a main demand conversion rate predictor, and obtaining the conversion rate of different users to the main demand in different scenes and different service selling processes.
Referring to fig. 5, step S21 includes the following sub-steps:
s211, acquiring historical data needing to be estimated, and assembling a user tag of a user id by the system according to the user id in the historical data. In particular, the user tags include near real-time tags and history class tags.
And S212, assembling a feature data set comprising practice class features according to the time class data in the user tags. Specifically, the practice-like features include year, month, week, day of week, day of month, day of holiday, etc.
And S213, learning a service purchase rate predictor and a main demand conversion rate predictor on the characteristic data set by using the XGboost learner.
Referring to fig. 6, step S3 includes the following sub-steps:
s31, obtaining historical data, generating a list of non-display ratios of display times of 0 advertisement display positions in m% of one step length to the total display times, and generating a list of coefficients α in a calculation formula of a ranking index in n step lengths.
And S32, calculating (not displaying the proportion, α) the conversion unit increment of the main demand and the service revenue increment on the duplet according to the purchase rate of the user to the service and the conversion rate of the main demand in the step S23, and outputting (not displaying the proportion, not displaying α which enables the ROI index of the class to be maximum under the proportion) a mapping table.
S33, in the mapping table in step S32, one α is randomly selected, and an initial ranking index calculation formula is generated.
S34, calculating the ranking index value of each advertisement display request on the line according to the initial ranking index calculation formula, and forming a ranking index sequence according to the ranking index values.
S35, obtaining quantiles of the sequencing index sequence according to q step lengths, calculating corresponding service revenue increment according to the quantiles, and outputting a mapping table (the quantiles and the service revenue increment). Specifically, when the ranking index value requested on line is smaller than the quantile value, the advertisement is not shown, and under the operation, the quantile of the ranking index sequence is an approximate estimation of the non-shown proportion, that is, the mapping table is approximate to the mapping table (the non-shown proportion, the service revenue increment).
S36, according to the set service revenue increment, the corresponding quantile is obtained in the mapping table in the step S35, and the corresponding α is obtained in the mapping table in the step S32 by taking the quantile as the corresponding non-display proportion.
S37, updating α of the calculation formula of the ranking index according to α of the step S36, and updating the mapping table (not showing the occupation ratio, service revenue increment) of the step S35 according to the updated calculation formula of the ranking index.
The implementation principle of the above embodiment is as follows:
through not showing the dynamic of the number of the proportion capture advertisement display positions, through the direct feedback user experience of main demand conversion single quantity increment, through the direct feedback user experience of the similar ROI index of main demand conversion single quantity increment divided by service revenue increment and platform revenue. By the technical scheme, the mapping table can be updated in real time (the proportion is not shown, and the service revenue increment is not shown), the sequencing index calculation formula can be updated in a linkage manner according to the expected service revenue increment, and a sequencing mechanism enabling the similar ROI index to reach the optimal is generated. The technical scheme of the application not only can adapt to the scene that the number of advertisement display positions is 0, but also strengthens the control of the relationship between user experience and service revenue.
EXAMPLE III
Referring to fig. 7, the advertisement ranking mechanism generating system disclosed in the present invention includes a service limiting module 10, a pre-estimating module 20, and a ranking mechanism generating module 30.
Referring to fig. 7, the service restriction module 10 is configured to configure a class of feature parameter intervals that allow the services to be exposed, filter the services having class of feature parameters outside the class of feature parameter intervals, and place the remaining services in the recall pool. Specifically, in this embodiment, the first-class characteristic parameter interval is a price interval of the service, and the service in the price interval is placed in the recall pool to ensure user experience.
Although the application scenario of the present application is described by taking price as a class of characteristic parameters as an example in the present application, those skilled in the art can understand that the technical solution of the present application is also applicable to various parameter scenarios, such as advertisement delivery ratio, advertisement delivery time, and the like, and the present application is not limited to this specifically.
Referring to fig. 7, the prediction module 20 is configured to construct a prediction sample according to the historical data information, and obtain a purchase rate of the user for the service and a conversion rate of the main demand through a classification decision algorithm and the prediction sample. In the embodiment, the XGBoost learner is used in the classification decision algorithm.
Referring to fig. 7, the ranking mechanism generating module 30 is configured to generate a list of advertisement display slots with a number of 0 showing times to a non-showing ratio of the total showing times according to the history data information. The sorting mechanism generating module 30 is further configured to calculate a corresponding main demand conversion single increment and a corresponding service revenue increment according to the non-display proportion in the list of the non-display proportion, the purchase rate of the user for the service and the conversion rate of the main demand, and generate a sorting index calculation formula in which the ROI-like index of the main demand conversion single increment divided by the service revenue increment reaches a set value.
The method comprises the steps of calculating a main demand conversion unit increment (= main demand conversion rate when a certain service is shown-main demand conversion rate when no task service is shown) × 1, calculating a service revenue increment (= service purchase rate when a certain service is shown-service purchase rate when no task service is shown) × certain service price, calculating a ranking index according to a formula (1- α) × standardized main demand conversion unit increment + α× standardized service revenue increment, and calculating α a real number between 0 and 1.
The implementation principle of the above embodiment is as follows:
the dynamic property of the number of the advertisement display positions can be captured by not displaying the occupation ratio, the user experience can be directly fed back through the main demand conversion single quantity increment, and the relation between the user experience and the platform revenue can be directly fed back through the similar ROI index of the main demand conversion single quantity increment divided by the service revenue increment. Through the technical scheme, the method and the device can be used for adjusting the order mechanism for enabling the similar ROI index to reach the optimal in a linkage manner according to the expected service revenue increment, not only can be suitable for the scene that the number of the advertisement display positions is 0, but also strengthens the control on the relation between user experience and service revenue.
Example four
Referring to fig. 7, based on the first embodiment, the present embodiment is different from the first embodiment in that the service restriction module 10 includes an upper limit restriction unit 11 and a lower limit restriction unit 12. The upper limit limiting unit 11 is used for configuring a class of feature upper limit parameters allowing the services to be exposed, filtering the services of which the class of feature upper limit parameters is greater than or equal to the class of feature upper limit parameters, and placing the rest services into a recall pool. The lower limit limiting unit 12 is configured to detect the number of services in the recall pool, and to add a service with the smallest characteristic parameter of the filtered services to the recall pool when the number of services in the recall pool is 0.
Specifically, the upper limit limiting unit 11 is configured to determine whether the service price divided by the main demand price is greater than or equal to a set characteristic upper limit parameter of a type. And if the service price of the part of services is divided by the main demand price and is more than or equal to the set upper limit parameter, filtering the service id of the part of services, and putting the rest service id into the recall pool. And if the service price of the service divided by the main demand price is smaller than the set upper limit parameter, all the service ids are put into the recall pool. The lower limit limiting unit 12 is configured to detect whether the number of service ids in the recall pool is equal to 0, and if the number of service ids in the recall pool is equal to 0, add a service id with a lowest price in all services to the recall pool; and if not, the recall pool does not process and outputs the service id in the recall pool.
Referring to fig. 7, the estimation module 20 includes an estimator training unit 21, a sample construction unit 22, and an estimation unit 23.
Referring to FIG. 7, predictor training unit 21 is used to obtain and assemble historical data into a feature data set, and is also used to learn a service purchase rate predictor and a primary demand conversion rate predictor on the feature data set using an XGboost learner.
Specifically, the predictor training unit 21 is further configured to assemble a user tag of the user id according to the user id in the historical data. The user tags comprise near real-time tags and history class tags, the features in the feature data set comprise practice class features, and the practice class features comprise year, month, week, day of the week, day of the month, whether holiday, day of holiday and the like which are assembled according to the time class data in the user tags.
Referring to fig. 7, the sample construction unit 22 is used to poll the services in the recall pool and concatenate the services with the features in the feature data set into predicted samples.
Referring to fig. 7, the prediction unit 23 is configured to input the prediction samples as input parameters into a service purchase rate predictor, so as to obtain purchase rates of different users for services in different scenarios and for different services for sale. The prediction unit 23 is further configured to input the prediction sample as an input parameter into a main demand conversion rate predictor, so as to obtain conversion rates of different users to the main demand in different scenes and for different services.
Referring to fig. 7, the sorting mechanism generating module 30 includes a non-exposure duty-to-coefficient α mapping table generating module 31, a non-exposure duty-to-service revenue increment mapping table generating module 32, and a sorting calculation formula updating module 33.
Referring to fig. 7, the non-display duty-to-coefficient α mapping table generating module 31 is configured to generate a list of non-display duty ratios, in which the number of display times of the advertisement display slots is 0 accounts for the total display times, in m% of one step, and generate a list of coefficients α in the ranking index calculation formula in n one step, the non-display duty-to-coefficient α mapping table generating module 31 is further configured to calculate (non-display duty, α) a main demand conversion unit increment and a service revenue increment on the binary set according to a purchase rate of a service and a conversion rate of a main demand by a user, and obtain a first mapping table of α, that is, (α) a mapping table of maximizing the category index ROI without displaying the duty ratio and without displaying the main demand conversion unit increment by the service revenue increment.
Referring to fig. 7, the non-display duty and service revenue increment mapping table generating module 32 is configured to randomly select α from a mapping table (α where the ROI-like index is maximized without displaying the duty) to generate an initial ranking index calculation formula, calculate a ranking index value of each advertisement display request on the line according to the initial ranking index calculation formula, and form a ranking index sequence according to the ranking index value.
Specifically, when the ranking index value requested on line is smaller than the quantile value, the advertisement is not shown, and under the operation, the quantile of the ranking index sequence is an approximate estimation of the non-shown proportion, that is, the mapping table is approximate to the mapping table (the non-shown proportion, the service revenue increment).
Referring to fig. 7, the sorting calculation formula update module 33 is configured to obtain a corresponding quantile in the (quantile, service revenue increment) mapping table according to the set service revenue increment, and obtain a corresponding α in the (α where the ROI-like index is maximized without showing the quantile) mapping table by using the quantile as a corresponding non-shown proportion, the sorting calculation formula update module 33 is further configured to update α in the sorting calculation formula according to the corresponding α, and update (without showing the quantile, service revenue increment) the mapping table according to the updated sorting calculation formula.
The implementation principle of the above embodiment is as follows:
through not showing the dynamic of the number of the proportion capture advertisement display positions, through the direct feedback user experience of main demand conversion single quantity increment, through the direct feedback user experience of the similar ROI index of main demand conversion single quantity increment divided by service revenue increment and platform revenue. By the technical scheme, the mapping table can be updated in real time (the proportion is not shown, and the service revenue increment is not shown), and the sequencing index calculation formula can be updated in a linkage manner according to the expected service revenue increment, so that a sequencing mechanism enabling the similar ROI index to reach the optimal is generated. The method not only can adapt to the scene that the number of the advertisement display positions is 0, but also strengthens the control of the relationship between the user experience and the service revenue.
The embodiments of the present invention are preferred embodiments of the present invention, and the scope of the present invention is not limited by these embodiments, so: all equivalent changes made according to the structure, shape and principle of the invention are covered by the protection scope of the invention.
Claims (10)
1. An advertisement ranking mechanism generating method is characterized by comprising the following steps:
step S1, configuring a class of characteristic parameter interval allowing the display of the service, filtering the service of which the class of characteristic parameter is outside the class of characteristic parameter interval, and putting the rest service into a recall pool;
step S2, constructing a prediction sample according to historical data information, and obtaining the purchase rate of a user to service and the conversion rate to main demand through a classification decision algorithm and the prediction sample;
and S3, generating a list of non-display occupation ratios of the display times accounting for the total display times with the number of the advertisement display positions being 0 according to the historical data information, calculating corresponding main demand conversion unit quantity increment and service revenue increment according to the non-display occupation ratios in the list of the non-display occupation ratios and the purchase rate of the user for the service and the conversion rate of the main demand in the step S2, and generating a sequencing index calculation formula of dividing the main demand conversion unit quantity increment by the service revenue increment to reach a set value.
2. The method of generating an advertisement ranking mechanism according to claim 1 wherein the step S1 includes the following sub-steps:
s11, configuring a class of characteristic upper limit parameters allowing the display of the service, filtering the service of which the class of characteristic upper limit parameters is greater than or equal to the class of characteristic upper limit parameters, and putting the rest of the service into a recall pool;
and S12, detecting the number of the services in the recall pool, and if the number of the services in the recall pool is 0, adding the service with the minimum characteristic parameter in the filtered services and putting the services into the recall pool again.
3. The method of claim 2, wherein the step S2 includes the following sub-steps:
s21, acquiring historical data and assembling the historical data into a feature data set, and learning a service purchase rate predictor and a main demand conversion rate predictor on the feature data set by using a classification decision algorithm;
s22, polling the service in the recall pool, and splicing the service with the characteristic in the step S21 as a prediction sample;
s23, taking the prediction sample as a reference, inputting the reference into a service purchase rate predictor, and obtaining the purchase rates of different users for services in different scenes and when different services are bought; and taking the prediction sample as a reference, inputting the reference into a main demand conversion rate predictor, and obtaining the conversion rate of different users to the main demand in different scenes and different service selling processes.
4. The method of claim 3, wherein the step S21 includes the following sub-steps:
s211, acquiring historical data needing to be estimated, and assembling a user tag of a user id by a system according to the user id in the historical data;
s212, assembling a feature data set comprising practice class features according to the time class data in the user tags;
and S213, learning a service purchase rate predictor and a main demand conversion rate predictor on the characteristic data set by using a classification decision algorithm.
5. The method for generating an advertisement ranking mechanism according to claim 3 or 4 wherein the step S3 includes the following sub-steps:
s31, acquiring historical data, generating a list of non-display ratios of 0 display times to the total display times according to m% of one step length, and generating a list of coefficients α in a ranking index calculation formula according to n step lengths;
s32, according to the purchase rate of the user to the service and the conversion rate of the main demand in the step S23, calculating the main demand conversion unit increment and the service revenue increment on the binary group which does not show the occupation ratio and the coefficient α, and obtaining a mapping table α which enables the main demand conversion unit increment to be divided by the service revenue increment to be maximum under the condition that the occupation ratio is not shown and the occupation ratio is not shown;
s33, randomly selecting one α from the mapping table in the step S32 to generate an initial ranking index calculation formula;
s34, calculating the ranking index value of each advertisement display request on the line according to the initial ranking index calculation formula and forming a ranking index sequence according to the ranking index value;
s35, obtaining quantiles of the sequencing index sequence according to q step lengths, calculating corresponding service revenue increment according to the quantiles, and outputting a mapping table of the quantiles and the service revenue increment under the quantiles;
s36, according to the set service revenue increment, obtaining the corresponding quantile in the mapping table in the step S35, and taking the quantile as the corresponding non-display proportion to obtain the corresponding α in the mapping table in the step S32;
s37, updating α of the ranking index calculation formula according to α of the step S36, and updating the mapping table of the step S35 according to the updated ranking index calculation formula.
6. An advertisement ranking mechanism generation system, comprising:
a service restriction module (10) for configuring a class of feature parameter intervals allowing to expose services, filtering services having a class of feature parameters outside the class of feature parameter intervals, and placing the remaining services in a recall pool;
the prediction module (20) is used for constructing a prediction sample according to historical data information and obtaining the purchase rate of a user to a service and the conversion rate of a main demand through a classification decision algorithm and the prediction sample;
a sorting mechanism generating module (30) for generating a list of non-display ratios of display times of 0 to total display times according to the historical data information;
the sorting mechanism generating module (30) is further configured to calculate corresponding main demand conversion unit increment and service revenue increment according to the non-display proportion in the list of the non-display proportion, the purchase rate of the user for the service and the conversion rate of the main demand, and generate a sorting index calculation formula in which the main demand conversion unit increment is divided by the service revenue increment to reach a set value.
7. The advertisement ranking mechanism generation system according to claim 6 wherein the service restriction module (10) includes:
an upper limit limiting unit (11) for configuring a class of feature upper limit parameters allowing to show services, filtering services of which the class of feature upper limit parameters is greater than or equal to the class of feature upper limit parameters, and placing the rest services into the recall pool;
and a lower limit limiting unit (12) for detecting the number of services in the recall pool, and for adding a service with the minimum characteristic parameter in the filtered services to the recall pool again when the number of services in the recall pool is 0.
8. The advertisement ranking mechanism generation system of claim 7 wherein the forecast module (20) includes:
the predictor training unit (21) is used for acquiring historical data and assembling the historical data into a characteristic data set, and is also used for learning a service purchase rate predictor and a main demand conversion rate predictor on the characteristic data set by using a classification decision algorithm;
a sample construction unit (22) for polling services in the recall pool and stitching features in the service and feature data sets into predicted samples;
the prediction unit (23) is used for inputting the prediction sample as an input parameter into a service purchase rate predictor to obtain the purchase rates of different users for services in different scenes and when different services are bought; the prediction sample is used as an input parameter and is input into a main demand conversion rate predictor, and the conversion rates of different users to main demands in different scenes and different services are obtained.
9. The system of claim 8, wherein the predictor training unit (21) is further configured to assemble a user label of the user id according to the user id in the historical data, and the features in the feature data set comprise practice-class features assembled according to the time-class data in the user label.
10. The advertisement ranking mechanism generating system according to claim 8 or 9 wherein the ranking mechanism generating module (30) comprises:
the non-display proportion and coefficient α mapping table generating module (31) is used for generating a list of non-display proportions of display times of which the number of advertisement display positions is 0 in the total display times according to m% of one step after acquiring historical data, generating a list of coefficients α in a sorting index calculation formula according to n one step, calculating a main demand conversion unit increment and a service operation increment on a binary group of the non-display proportions and the coefficients α according to the purchase rate of a user to service and the conversion rate of main demand, and obtaining a first mapping table of α which enables the main demand conversion unit increment to be divided by the service operation increment to be maximum under the non-display proportions and the non-display proportions;
a non-display proportion and service revenue increment mapping table generation module (32) which is used for randomly selecting α to generate an initial sequencing index calculation formula in the mapping table I, calculating the sequencing index value of each advertisement display request on the line according to the initial sequencing index calculation formula and forming a sequencing index sequence according to the sequencing index value, and is also used for obtaining the quantile of the sequencing index sequence according to q one step length, calculating the corresponding service revenue increment according to the quantile and outputting a mapping table II of the quantile and the service revenue increment under the quantile;
and the sequencing calculation formula updating module (33) is used for obtaining a corresponding quantile in the second mapping table according to the set service revenue increment, taking the quantile as a corresponding non-display duty ratio and obtaining corresponding α in the first mapping table, and is also used for updating α in the sequencing index calculation formula according to the corresponding α and updating the second mapping table according to the updated sequencing index calculation formula.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811260738.6A CN109377280B (en) | 2018-10-26 | 2018-10-26 | Advertisement sequencing mechanism generation method and generation system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811260738.6A CN109377280B (en) | 2018-10-26 | 2018-10-26 | Advertisement sequencing mechanism generation method and generation system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109377280A CN109377280A (en) | 2019-02-22 |
| CN109377280B true CN109377280B (en) | 2020-07-17 |
Family
ID=65389955
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811260738.6A Active CN109377280B (en) | 2018-10-26 | 2018-10-26 | Advertisement sequencing mechanism generation method and generation system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109377280B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111324800B (en) * | 2020-02-12 | 2023-04-21 | 腾讯科技（深圳）有限公司 | Business item display method, device and computer readable storage medium |
| CN112232857A (en) * | 2020-09-25 | 2021-01-15 | 上海淇毓信息科技有限公司 | Automatic advertisement sequencing method and device and electronic equipment |
| CN112200610A (en) * | 2020-10-10 | 2021-01-08 | 苏州创旅天下信息技术有限公司 | Marketing information delivery method, system and storage medium |
| CN113032445B (en) * | 2021-05-24 | 2021-08-17 | 武汉卓尔数字传媒科技有限公司 | Data conversion sorting method and device and electronic equipment |
| CN113516519A (en) * | 2021-07-28 | 2021-10-19 | 北京字节跳动网络技术有限公司 | Model training method, advertisement putting method, device, equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103177371A (en) * | 2011-12-21 | 2013-06-26 | 阿里巴巴集团控股有限公司 | Method and device for exhibiting information |
| CN106485529A (en) * | 2015-09-02 | 2017-03-08 | 北京国双科技有限公司 | The sort method of advertisement position and device |
| CN108335137A (en) * | 2018-01-31 | 2018-07-27 | 北京三快在线科技有限公司 | Sort method and device, electronic equipment, computer-readable medium |
| CN108510309A (en) * | 2018-02-27 | 2018-09-07 | 阿里巴巴集团控股有限公司 | The method and device that advertisement is recalled |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10528986B2 (en) * | 2015-01-15 | 2020-01-07 | Xandr Inc. | Modifying bid price for online advertising auction based on user impression frequency |
| US10650403B2 (en) * | 2016-09-13 | 2020-05-12 | Adobe Inc. | Distributing online ads by targeting online ad requests |
-
2018
- 2018-10-26 CN CN201811260738.6A patent/CN109377280B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103177371A (en) * | 2011-12-21 | 2013-06-26 | 阿里巴巴集团控股有限公司 | Method and device for exhibiting information |
| CN106485529A (en) * | 2015-09-02 | 2017-03-08 | 北京国双科技有限公司 | The sort method of advertisement position and device |
| CN108335137A (en) * | 2018-01-31 | 2018-07-27 | 北京三快在线科技有限公司 | Sort method and device, electronic equipment, computer-readable medium |
| CN108510309A (en) * | 2018-02-27 | 2018-09-07 | 阿里巴巴集团控股有限公司 | The method and device that advertisement is recalled |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109377280A (en) | 2019-02-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109377280B (en) | Advertisement sequencing mechanism generation method and generation system | |
| US8417558B2 (en) | Systems and methods for identifying offered incentives that will achieve an objective | |
| US8001004B2 (en) | System and method for brokering the sale of internet advertisement inventory | |
| US8478658B2 (en) | Auction-based selection and presentation of polls to users | |
| US8775322B2 (en) | System for matching buyers and sellers based on buyer seller preferences | |
| US8082295B2 (en) | Reporting to a website owner one or more appearances of a specified word in one or more page-specific open-ended comments concerning one or more particular web pages of a website | |
| US8341047B1 (en) | Systems and methods for optimizing an electronic advertising campaign based on organic content | |
| CN101802856A (en) | Measuring a location based advertising campaign | |
| EP0994430A2 (en) | Contents management method | |
| US20140081774A1 (en) | Self service advertising method and system | |
| US20050198175A1 (en) | Systems and methods for optimizing dynamic mailings | |
| US20140279016A1 (en) | Behavioral tracking system and method in support of high-engagement communications | |
| US8719081B1 (en) | Bid adjustment scheduling for electronic advertising | |
| CN101809604A (en) | Content item pricing | |
| WO2008024997A2 (en) | System and method for modeling value of an on-line advertisement campaign | |
| US20120303418A1 (en) | Dynamic pricing of access to content where pricing varies with user behavior over time to optimize total revenue and users are matched to specific content of interest | |
| CN112150219A (en) | Online advertisement delivery system and use method thereof | |
| Molnar et al. | High-speed Internet access and housing values | |
| CN113283930A (en) | Marketing activity operation method and device | |
| CN114119111A (en) | Power transaction user management method, device, equipment and medium based on big data | |
| US20110231254A1 (en) | Method and system for providing advertisements, and computer-readable recording medium | |
| US20140289054A1 (en) | Behavioral tracking system and method in support of high-engagement communications | |
| US7991661B1 (en) | Method and system for providing market analysis for wireless voice markets | |
| EP2034712A1 (en) | System and method for optimising the choice of a telephone contract by automatically studying electronic telephone bills | |
| CN114971716A (en) | Service interface quality evaluation method and device, equipment, medium and product thereof |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |