CN108681915B - Click rate estimation method and device and electronic equipment - Google Patents
Click rate estimation method and device and electronic equipment Download PDFInfo
- Publication number
- CN108681915B CN108681915B CN201810349459.0A CN201810349459A CN108681915B CN 108681915 B CN108681915 B CN 108681915B CN 201810349459 A CN201810349459 A CN 201810349459A CN 108681915 B CN108681915 B CN 108681915B
- Authority
- CN
- China
- Prior art keywords
- click rate
- estimated
- model
- advertisement
- average
- 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
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Economics (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the invention provides a click rate estimation method, a click rate estimation device and electronic equipment, wherein the method comprises the following steps: the method comprises the steps of taking the characteristics of an advertisement to be estimated as the input of a pre-trained estimation model, and taking the output result of the estimation model as the initial estimation click rate of the advertisement to be estimated, wherein the estimation model is obtained by training according to the characteristics of the advertisement to be estimated and the actual click rate of the advertisement to be estimated, wherein the characteristics of the advertisement to be estimated are included in historical data of the advertisement to be estimated; and taking the initial estimated click rate as the input of a pre-trained calibration model, and taking the output result of the calibration model as the final estimated click rate of the advertisement to be estimated, wherein the calibration model is obtained by training according to the actual click rate of the advertisement to be estimated and the initial estimated click rate of the advertisement to be estimated, which are included in the historical data of the advertisement to be estimated. Through the technical scheme of the embodiment of the invention, the deviation between the estimated click rate obtained by the estimation model and the actual click rate can be reduced, and the estimation accuracy is improved.
Description
Technical Field
The invention relates to the technical field of internet, in particular to a click rate estimation method and device and electronic equipment.
Background
With the development of electronic commerce, network advertisements have become a new medium advertisement to enter people's lives. Typically, before placing an advertisement, an advertiser wants to know the click-through rate of the placed advertisement at an ad spot on a website and make a decision to book the ad spot based on the known click-through rate. In addition, the click rate of the advertisement put on a certain advertisement position can be estimated for the reference of the advertiser for providing the basis of the preset advertisement position decision for the advertiser.
The method generally adopted in the prior art for estimating the advertisement click rate is as follows: training a prediction model by using historical data of the advertisement to be predicted, wherein the historical data comprises the characteristics and the actual click rate of the advertisement to be predicted; and taking the characteristics of the advertisement to be estimated as the input of the estimation model, and taking the output result of the estimation model as the estimated click rate of the advertisement to be estimated. The characteristics of the advertisement can be templates of the advertisement, characteristics of people targeted by the advertisement, types of products corresponding to the advertisement, platforms for displaying the advertisement, and the like. The predictive model may be, for example, a logistic regression model, a factorizer model, or the like.
However, the inventor finds that the prior art has at least the following problems in the process of implementing the invention:
In the prior art, the estimated click rate of the advertisement obtained only through the characteristics of the advertisement when the click rate is estimated has larger deviation with the actual click rate and lower estimation accuracy.
Disclosure of Invention
The embodiment of the invention aims to provide a click rate estimation method, a click rate estimation device and electronic equipment, so as to reduce the deviation between the estimated click rate and the actual click rate of an advertisement and improve the estimation accuracy. The specific technical scheme is as follows:
in one aspect of the present invention, a click rate estimation method is provided, including:
the method comprises the steps of taking the characteristics of an advertisement to be estimated as the input of a pre-trained estimation model, and taking the output result of the estimation model as the initial estimation click rate of the advertisement to be estimated, wherein the estimation model is obtained by training according to the characteristics of the advertisement to be estimated and the actual click rate of the advertisement to be estimated, wherein the characteristics are included in historical data of the advertisement to be estimated;
and taking the initial estimated click rate as the input of a pre-trained calibration model, and taking the output result of the calibration model as the final estimated click rate of the advertisement to be estimated, wherein the calibration model is obtained by training according to the actual click rate of the advertisement to be estimated and the initial estimated click rate of the advertisement to be estimated, which are included in the historical data of the advertisement to be estimated.
Optionally, the training process of the calibration model includes:
acquiring historical data of the advertisement to be estimated in a plurality of time periods, wherein each historical data comprises an initial estimated click rate of the advertisement to be estimated and an actual click rate corresponding to the initial estimated click rate;
and training a preset model by using the historical data, and taking the trained preset model as the calibration model.
Optionally, the preset model is any one of a logistic regression model, a linear regression model, a factorization model, and a order-preserving regression model.
Optionally, when the preset model is an order preserving regression model, the training of the preset model by using the historical data includes:
dividing the historical data into N sections of order-preserving data, wherein N is an integer greater than 1; each section of order preserving data comprises the same amount of historical data;
taking the average value of the initial estimated click rate in each section of order-preserving data as the average initial estimated click rate of the section of order-preserving data; taking the average value of the actual click rate in each section of order preserving data as the average actual click rate of the section of order preserving data;
and sorting the N sections of order-preserving data from small to large according to the average initial estimated click rate, fitting the average initial estimated click rate and the average actual click rate by using an order-preserving regression algorithm to obtain a calibration value of the average initial estimated click rate, and training the preset model by using the calibration value of the average initial estimated click rate and the average actual click rate.
Optionally, the calibration model is:
wherein d isiA calibration value d of the average estimated click rate of the ith segment of order-preserving datai+1The calibration value p of the average initial estimated click rate of the i +1 th section of order-preserving dataiAverage initial estimated click rate, p, of the ith segment of order-preserving datai+1The average initial estimated click rate of the i +1 th section of order-preserving data, x is the initial estimated click rate to be calibrated, and f (x) is the final estimated click rate after calibration.
Optionally, when the preset model is a linear regression model, the training of the preset model using the historical data includes:
taking the difference between the initial estimated click rate and the actual click rate in each historical data as the error value of each historical data;
taking the average value of the error values corresponding to all the historical data as an average error value;
training the calibration model with the average error value;
the calibration model is as follows:
(x) x-b; wherein b is the average error value, x is the initial estimated click rate to be calibrated, and f (x) is the final estimated click rate after calibration.
In another aspect of the present invention, there is also provided a click rate estimation device, including:
The estimation unit is used for inputting the characteristics of the advertisement to be estimated as an estimation model trained in advance and using the output result of the estimation model as the initial estimation click rate of the advertisement to be estimated, wherein the estimation model is obtained by training according to the characteristics of the advertisement to be estimated and the actual click rate of the advertisement to be estimated, wherein the characteristics of the advertisement to be estimated are included in the historical data of the advertisement to be estimated;
and the calibration unit is used for inputting the initial estimated click rate as a pre-trained calibration model and taking an output result of the calibration model as a final estimated click rate of the advertisement to be estimated, wherein the calibration model is obtained by training according to the actual click rate of the advertisement to be estimated and the initial estimated click rate of the advertisement to be estimated, which are included in the historical data of the advertisement to be estimated.
Optionally, the apparatus further comprises:
the data acquisition unit is used for acquiring historical data of the advertisement to be estimated in a plurality of time periods, wherein each historical data comprises an initial estimated click rate of the advertisement to be estimated and an actual click rate corresponding to the initial estimated click rate;
and the model determining unit is used for training a preset model by using the historical data, and taking the trained preset model as the calibration model.
Optionally, the preset model is any one of a logistic regression model, a linear regression model, a factorization model, and a order-preserving regression model.
Optionally, when the preset model is a order-preserving regression model, the model determining unit includes:
the data dividing subunit is used for dividing the historical data into N sections of order-preserving data, wherein N is an integer greater than 1; each section of order preserving data comprises the same amount of historical data;
the click rate calculation subunit is used for taking the average value of the initial estimated click rate in each section of order-preserving data as the average initial estimated click rate of the section of order-preserving data; taking the average value of the actual click rate in each section of order preserving data as the average actual click rate of the section of order preserving data;
and the calibration model determining subunit is used for sequencing the N sections of order-preserving data from small to large according to the average initial estimated click rate, fitting the average initial estimated click rate and the average actual click rate by using an order-preserving regression algorithm to obtain a calibration value of the average initial estimated click rate, and training the preset model by using the calibration value of the average initial estimated click rate and the average actual click rate.
Optionally, the calibration model is:
wherein d isiA calibration value d of the average initial estimated click rate of the ith segment of order-preserving datai+1The calibration value p of the average initial estimated click rate of the i +1 th section of order-preserving dataiAverage initial estimated click rate, p, of the ith segment of order-preserving datai+1The average initial estimated click rate of the i +1 th section of order-preserving data, x is the initial estimated click rate to be calibrated, and f (x) is the final estimated click rate after calibration.
Optionally, when the preset model is a linear regression model, the model determining unit includes:
the error calculation subunit is used for taking the difference between the initial estimated click rate and the actual click rate in each historical data as the error value of each historical data;
the average error calculation subunit is used for taking the average value of the error values corresponding to all the historical data as an average error value;
a calibration model deriving subunit configured to train the calibration model using the average error value;
the calibration model is as follows:
(x) x-b; wherein b is the average error value, x is the initial estimated click rate to be calibrated, and f (x) is the final estimated click rate after calibration.
In another aspect of the present invention, there is also provided an electronic device, including a processor, a communication interface, a memory and a communication bus, where the processor, the communication interface, and the memory complete communication with each other through the communication bus;
A memory for storing a computer program;
and the processor is used for realizing any click rate estimation method when executing the program stored in the memory.
In yet another aspect of the present invention, there is also provided a computer-readable storage medium, having stored therein instructions, which when run on a computer, cause the computer to execute any one of the click rate estimation methods described above.
In yet another aspect of the present invention, there is also provided a computer program product containing instructions which, when run on a computer, cause the computer to perform any of the above click rate estimation methods.
According to the click rate estimation method, the click rate estimation device and the electronic equipment provided by the embodiment of the invention, when the click rate estimation is carried out, the characteristics of the advertisement to be estimated are used as the input of the estimation model, and the output result of the estimation model is used as the initial estimation click rate of the advertisement to be estimated; and taking the initial estimated click rate as the input of a calibration model, and taking the output result of the calibration model as the final estimated click rate of the advertisement to be estimated.
According to the invention, the initial estimated click rate output by the estimation model is calibrated again, so that the deviation between the estimated click rate obtained by the estimation model and the actual click rate can be reduced, and the estimation accuracy is improved. Of course, not all of the advantages described above need to be achieved at the same time in the practice of any one product or method of the invention.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below.
Fig. 1 is a schematic flow chart of a click rate estimation method according to an embodiment of the present invention;
FIG. 2 is a schematic flow chart illustrating training of a calibration model according to an embodiment of the present invention;
FIG. 3 is a schematic flow chart of a method for obtaining a calibration model according to an embodiment of the present invention;
FIG. 4 is a schematic flow chart illustrating a method for deriving a calibration model according to an embodiment of the present invention;
FIG. 5 is a schematic flow chart illustrating a click rate estimation method according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of a click rate estimation device according to an embodiment of the present invention;
fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be described below with reference to the drawings in the embodiments of the present invention.
In order to reduce the deviation between the estimated click rate and the actual click rate of the advertisement and improve the estimated accuracy, the embodiment of the invention provides the click rate estimation method, the device and the electronic equipment.
First, the click rate estimation method provided by the embodiment of the present invention is described below.
It should be noted that the click rate method provided by the embodiment of the invention can be applied to the internet advertisement industry and is suitable for online advertisements in internet advertisements.
As shown in fig. 1, the click rate estimation method provided in the embodiment of the present invention may include the following steps:
s110: the method comprises the steps of taking the characteristics of an advertisement to be estimated as the input of a pre-trained estimation model, and taking the output result of the estimation model as the initial estimation click rate of the advertisement to be estimated, wherein the estimation model is obtained by training according to the characteristics of the advertisement to be estimated and the actual click rate of the advertisement to be estimated, wherein the characteristics are included in the historical data of the advertisement to be estimated, and the actual click rate of the advertisement to be estimated is obtained.
In practical application, the characteristics of the advertisement and the click rate of the advertisement have certain relation, and the click rate of the advertisement can be estimated according to the characteristics of the advertisement. Therefore, in the embodiment of the present invention, the server may use the characteristics of the advertisement to be estimated as the input of the pre-trained estimation model, and use the output result of the estimation model as the initial estimation click rate of the advertisement to be estimated.
Because the actual click rate of the advertisement in the historical data is known data, the machine learning algorithm trains a model according to the historical data of the advertisement to be estimated, and the relation between the actual click rate of the advertisement to be estimated and the characteristics of the advertisement can be accurately reflected. Therefore, in the embodiment of the present invention, the pre-trained estimation model may be obtained by training according to the characteristics of the to-be-estimated advertisement included in the historical data of the to-be-estimated advertisement and the actual click rate of the to-be-estimated advertisement.
The prediction model may be any one of a factorization model, a logistic regression model, a linear regression model, and the like, and a person skilled in the art may select the prediction model according to actual needs, which is not specifically limited in the present application. Those skilled in the art can train the factorization machine model, the logistic regression model, and the linear regression model through known data by using a method commonly used in the prior art, so as to obtain the pre-estimation model, which is further described in the present application.
The characteristics of the advertisement may include user characteristics, characteristics of the advertisement itself, ad spot context characteristics, and the like. The user characteristics refer to information of a user clicking the advertisement, and the advertisement space context characteristics refer to characteristics of an environment where the advertisement is located. For example, user characteristics may include the user's age, gender, interests, historical liveness, and so forth; the characteristics of the ad itself may include ad material, ad categories, advertisers, bid terms, etc., and the ad spot context characteristics may include ad spot identification, contextual content of the location of the ad, the networking mode the user clicked on the ad, the type and/or model of device the user used to click on the ad, etc.
S120: and taking the initial estimated click rate as the input of a pre-trained calibration model, and taking the output result of the calibration model as the final estimated click rate of the advertisement to be estimated, wherein the calibration model is obtained by training according to the actual click rate of the advertisement to be estimated and the initial estimated click rate of the advertisement to be estimated, which are included in the historical data of the advertisement to be estimated.
Step S110 is to obtain an initial estimated click rate of the advertisement according to the characteristics of the advertisement, and there is a large error between the initial estimated click rate and the actual click rate of the advertisement. It can be understood that, in the historical data of the advertisement, the initial estimated click rate of the advertisement is also known data, in the historical data of the advertisement, by analyzing the relationship between the actual click rate and the initial estimated click rate of the advertisement, the law followed between the initial estimated click rate and the actual click rate of the advertisement, that is, the law followed by the error of the initial estimated click rate can be obtained, and according to the law, the initial estimated click rate is calibrated, so that the accuracy of the initial estimated click rate can be improved.
Therefore, in the embodiment of the present invention, the server may use the initial estimated click rate as an input of a pre-trained calibration model, and use an output result of the calibration model as a final estimated click rate of the to-be-estimated advertisement, wherein the calibration model is obtained by training according to an actual click rate of the to-be-estimated advertisement and the initial estimated click rate of the to-be-estimated advertisement included in the historical data of the to-be-estimated advertisement.
The historical data may be data of the advertisement to be predicted in a preset historical time period, for example, data of the advertisement to be predicted in the previous day, data of the advertisement in the previous week, and the like.
According to the click rate estimation method provided by the embodiment of the invention, the initial estimated click rate output by the estimation model is calibrated again, so that the deviation between the estimated click rate obtained by the estimation model and the actual click rate can be reduced, and the estimation accuracy is improved.
It can be understood that a certain estimated error exists between the initial estimated click rate of the advertisement obtained according to the characteristics of the advertisement and the actual click rate of the advertisement, and the rule followed by the estimated error can be obtained by analyzing the relationship between the initial estimated click rate and the actual click rate. In an implementation manner of the embodiment of the present invention, as shown in fig. 2, the training process of the calibration model in step S120 may include the following steps:
s121: acquiring historical data of the advertisement to be estimated in a plurality of time periods, wherein each historical data comprises an initial estimated click rate of the advertisement to be estimated and an actual click rate corresponding to the initial estimated click rate;
S122: and training a preset model by using the historical data, and taking the trained preset model as the calibration model.
The historical data in the multiple time periods may be historical data of the advertisement obtained by the server in the multiple time periods in the past week, or may be historical data in the multiple time periods in the past month, for example, the historical data of the advertisement in the past week, in each day, 09: 00 to 12: historical data over a period of 00 hours. Those skilled in the art can select appropriate historical data according to actual needs, and the invention is not limited to the number of historical data and the time period.
In practical application, the server may fit the initial estimated click rate and the actual click rate included in the history data to obtain a calibration value of the initial estimated click rate, and train the preset model using the calibration value of the initial estimated click rate and the actual click rate.
The preset model trained by the actual click rate and the initial estimated click rate in the historical data of the advertisement is used as a calibration model, the rule followed by the error between the initial estimated click rate and the actual click rate obtained according to the characteristics of the advertisement can be well reflected, the click rate estimated by the estimated model is calibrated according to the rule, namely the calibration model, the error between the initial estimated click rate and the actual click rate obtained by the estimated model can be reduced, and therefore the accuracy of the initial estimated click rate is improved.
In an implementation manner of the embodiment of the present invention, the preset model may be one of a logistic regression model, a linear regression model, a factorization model, and a order-preserving regression model. The predetermined model may be the same as the estimated model or different from the estimated model.
It can be understood that the order-preserving regression model is a regression model determined by an order-preserving regression algorithm, and the order-preserving regression model is determined by sorting the numbers in a number sequence and modifying the numbers in the number sequence to minimize the sum of squares of differences between the modified numbers and a preset value. In a specific application, the sequence-preserving regression model can be used for calibrating each estimated value, so that each calibrated estimated value is closer to an actual value.
Therefore, in an implementation manner of the embodiment of the present invention, when the preset model is a order-preserving regression model, as shown in fig. 3, the specific steps of the training process of the preset model in step S122 may be:
s1221: dividing the historical data into N sections of order-preserving data, wherein N is an integer greater than 1; wherein, each segment of order preserving data comprises the same amount of historical data.
It can be understood that the historical data obtained by the server may include a large amount of data, and if the server uses each data to train the preset model, the calculation amount may be very large, resulting in slow calculation or even wrong calculation of the server. Therefore, in an implementation manner of the embodiment of the present invention, the history data may be divided into N segments of order-preserving data, where N is an integer greater than 1; wherein, each segment of order preserving data comprises the same amount of historical data.
For example, if 1000000 pieces of data are included in the historical data, the 1000000 pieces of historical data may be divided into 1000 segments, and each segment of data includes 1000 pieces of data. The technical personnel in the field can determine the value of the N according to the actual quantity of the historical data, the calculation capacity of the server and the requirement on the estimation precision, wherein the larger the N is, the higher the precision is, but the higher the calculation complexity is correspondingly. The value of N is not particularly limited.
S1222: taking the average value of the initial estimated click rate in each section of the order-preserving data as the average initial estimated click rate of the section of the order-preserving data; and taking the average value of the actual click rate in each section of the order-preserving data as the average actual click rate of the section of the order-preserving data.
After the server divides the historical data into N sections of order preserving data, each section of order preserving data comprises a plurality of groups of data, and only one group of data corresponding to each section of order preserving data is needed to train the order preserving regression model. Therefore, in an implementation manner of the embodiment of the present invention, the server may use an average value of the initial estimated click rate in each segment of the order-preserving data as an average initial estimated click rate of the segment of order-preserving data; and taking the average value of the actual click rate in each section of the order-preserving data as the average actual click rate of the section of the order-preserving data, so that each section of the order-preserving data corresponds to one group of data, thereby obtaining the data for training the order-preserving regression model.
S1223: and sequencing the N sections of order-preserving data from small to large according to the average initial estimated click rate, fitting the average initial estimated click rate and the average actual click rate by using an order-preserving regression algorithm to obtain a calibration value of the average initial estimated click rate, and training the preset model by using the calibration value of the average initial estimated click rate and the average actual click rate.
In an embodiment of the present invention, the server may sort the N pieces of order preserving data from small to large according to the average initial estimated click rate, so that N pieces of average initial estimated click rates to be calibrated corresponding to the N pieces of order preserving data form data in non-decreasing arrangement, and thus, the order preserving regression model may be trained using the order preserving data including the average initial estimated click rate.
It can be understood that to obtain the preset model, the average initial estimated click rate of each segment of order preserving data needs to be known first, the calibration value obtained by calibrating the average initial estimated click rate by using an order preserving regression algorithm is used, and the order preserving regression model is trained by using the calibration value of the average initial estimated click rate and the average actual click rate. Therefore, in an implementation manner of the embodiment of the present invention, the server may use a order-preserving regression algorithm to fit the average initial estimated click rate and the average actual click rate to obtain a calibration value of each average initial estimated click rate, and train the preset model by using the calibration value of the average initial estimated click rate and the average actual click rate. And the calibration value corresponding to the sorted average initial estimated click rate needs to meet the requirement of non-decreasing.
The calibration value after the average initial estimated click rates are calibrated by using a order-preserving regression algorithm needs to satisfy the following formula (1):
wherein d isiTo calibrate the average initial estimated click rate of the ith segment of order-preserving data, omegaiWeight coefficient for i-th segment of order-preserving data, ciAnd (5) the average actual click rate of the ith segment of order-preserving data.
For example, N pieces of order-preserving data are sorted from small to large according to the average initial estimated click rate to 0.5, 0.6, 0.6, 0.8 and 0.9, where N is 5, the average actual click rate corresponding to the average initial estimated click rate is 0.4, 0.3, 0.3, 0.5 and 0.7, and the calibration value after 5 average initial estimated click rates are calibrated by using the formula (1) of the order-preserving regression algorithm is 0.4, 0.4, 0.4, 0.5 and 0.7; and training a sequence preserving regression model according to sequences 0.4, 0.3, 0.3, 0.5 and 0.7 consisting of the average actual click rate and sequences 0.4, 0.4, 0.4, 0.5 and 0.7 consisting of the calibration value after the average initial estimated click rate is calibrated, and taking the trained sequence preserving regression model as the calibration model.
According to the embodiment of the invention, a calibration value after calibration is carried out on the average initial estimated click rate of each section of order-preserving data according to an order-preserving regression algorithm, and then a calibration model is trained according to the calibration value and the average actual click rate corresponding to the calibration value. Since the average initial estimated click rate to be calibrated is sequenced and then calibrated in the embodiment of the invention, no matter what distribution state the average initial estimated click rate of the advertisement to be estimated is, such as normal distribution or discrete distribution, the calibration model obtained in the embodiment of the invention can calibrate the average initial estimated click rate, and the calibration accuracy is higher.
In an implementation manner of the embodiment of the present invention, the calibration model in step S1223 may be:
wherein d isiA calibration value d of the average initial estimated click rate of the ith segment of order-preserving datai+1The calibration value p of the average initial estimated click rate of the i +1 th section of order-preserving dataiAverage initial estimated click rate, p, of the ith segment of order-preserving datai+1The average initial estimated click rate of the i +1 th section of order-preserving data, x is the initial estimated click rate to be calibrated, and f (x) is the final estimated click rate after calibration.
When the formula (2) is used, the initial estimated click rate estimated by the estimation model is judged to correspond to which section of the N sections of order-preserving data in the historical data, and then the initial estimated click rate is directly substituted into the formula (2), so that the final estimated click rate corresponding to the initial estimated click rate can be obtained, the operation process is simple, and the calculation speed of the final estimated click rate is high.
It can be understood that the error between the estimated click rate and the actual click rate may be a certain rule in general, for example, the estimated click rate may be larger than the actual click rate by a certain value, and at this time, the linear regression model may be selected as the preset model. Therefore, in an implementation manner of the embodiment of the present invention, when the preset model is a linear regression model, as shown in fig. 4, the specific steps of the training process of the preset model in step S122 may be:
S1224: taking the difference between the initial estimated click rate and the actual click rate in each historical data as the error value of each historical data;
s1225: taking the average value of the error values corresponding to all the historical data as an average error value;
s1226: training the calibration model with the average error value;
the calibration model obtained in step S1226 may be:
f(x)＝x-b (3)
wherein b is the average error value, x is the initial estimated click rate, and f (x) is the final estimated click rate.
The difference between the initial estimated click rate and the actual click rate in the historical data can be obtained through simple operation, so that the server can obtain the final estimated click rate with higher accuracy more quickly according to the average value of the difference between the initial estimated click rate and the actual click rate.
In another embodiment of the present invention, a click-through rate estimation method is applied to a server, as shown in fig. 5, the method may include the following steps:
s510: training a calibration model on line;
s520: and estimating the click rate of the advertisement to be estimated on line.
The step S510 may be implemented as the following steps:
s511: a command to begin creating a calibration model is received.
S512: acquiring historical data of an advertisement to be estimated in a plurality of historical time periods; each historical data comprises an initial estimated click rate of the advertisement to be estimated and an actual click rate corresponding to the initial estimated click rate.
S513: dividing the historical data in the plurality of historical time periods into N sections of order-preserving data, wherein N is an integer greater than 1, and each section of order-preserving data comprises the same number of historical data; and taking the average value of the initial estimated click rate in each section of order-preserving data as the average initial estimated click rate of the section of order-preserving data, and taking the average value of the actual click rate in each section of order-preserving data as the average actual click rate of the section of order-preserving data.
S514: sorting the N sections of order-preserving data from small to large according to the average initial estimated click rate, fitting the average initial estimated click rate and the average actual click rate by using an order-preserving regression algorithm to obtain a calibration value of the average initial estimated click rate, training a preset model by using the calibration value of the average initial estimated click rate and the average actual click rate, and taking the trained preset model as a calibration model.
The step S520 may be implemented as the following steps:
s521: receiving a prediction request of the advertisement to be predicted, and judging whether the advertisement to be predicted is cold start flow or not; the advertisement to be estimated is on line for the first time, and no historical data exists.
S522: and when the advertisement to be estimated is not the cold start flow, estimating the advertisement to be estimated by using a factorization model to obtain an initial estimated click rate. The factorization machine model is a commonly used estimation model for estimating the advertisement click rate, and the factorization machine model is not specifically explained in the invention.
S523: the initial estimated click rate obtained in step S522 is calibrated using the calibration model.
S524: the result of calibrating the initial estimated click rate in step S523 is used as the final estimated click rate.
S525: when the advertisement to be estimated is the cold start traffic, the advertisement to be estimated is estimated by using the logistic regression model, and the final estimated click rate in step S524 is obtained. For example, the feature information of the advertisement to be estimated can be used as the input of the logistic regression model, and the output of the logistic regression model can be used as the final estimated click rate. It can be understood that, performing click rate estimation on the advertisement to be estimated by using the logistic regression model is a common method in the field, and the present invention is not described in detail herein.
The click rate estimating device provided by the embodiment of the invention is introduced below.
As shown in fig. 6, the click rate estimating apparatus provided in the embodiment of the present invention may include:
the pre-estimation unit 610 is configured to use characteristics of an advertisement to be pre-estimated as input of a pre-trained pre-estimation model, and use an output result of the pre-estimation model as an initial pre-estimated click rate of the advertisement to be pre-estimated, where the pre-estimation model is obtained by training according to the characteristics of the advertisement to be pre-estimated included in history data of the advertisement to be pre-estimated and an actual click rate of the advertisement to be pre-estimated;
The calibration unit 620 is configured to use the initial estimated click rate as an input of a pre-trained calibration model, and use an output result of the calibration model as a final estimated click rate of the to-be-estimated advertisement, where the calibration model is obtained by training according to an actual click rate of the to-be-estimated advertisement and an estimated click rate of the to-be-estimated advertisement included in historical data of the to-be-estimated advertisement.
In an implementation manner of the embodiment of the present invention, the apparatus may further include:
the data acquisition unit is used for acquiring historical data of the advertisement to be estimated in a plurality of time periods, and each historical data comprises an estimated click rate of the advertisement to be estimated and an actual click rate corresponding to the estimated click rate;
and the model determining unit is used for training a preset model by using the historical data, and taking the trained preset model as the calibration model.
In an implementation manner of the embodiment of the present invention, the preset model is any one of a logistic regression model, a linear regression model, a factorization model, and a order-preserving regression model.
In an implementation manner of the embodiment of the present invention, when the preset model is a order-preserving regression model, the model determining unit may include:
The data dividing subunit is used for dividing the historical data into N sections of order-preserving data, wherein N is an integer greater than 1; each section of order preserving data comprises the same amount of historical data;
the click rate calculation subunit is used for taking the average value of the estimated click rate in each section of order-preserving data as the average estimated click rate of the section of order-preserving data; taking the average value of the actual click rate in each section of order preserving data as the average actual click rate of the section of order preserving data;
and the calibration model determining subunit is used for sequencing the N sections of order-preserving data from small to large according to the average estimated click rate, fitting the average initial estimated click rate and the average actual click rate by using an order-preserving regression algorithm to obtain a calibration value of the average initial estimated click rate, and training the preset model by using the calibration value of the average initial estimated click rate and the average actual click rate.
In an implementation manner of the embodiment of the present invention, the calibration model may be:
wherein, ciAverage actual click rate of ith segment of order-preserving data, ci+1Average actual click rate, p, for segment i +1 order-preserving dataiAverage estimated click rate, p, of the ith segment of order-preserving data i+1The average estimated click rate of the i +1 th section of order-preserving data, x is the initial estimated click rate, and f (x) is the final estimated click rate.
In an implementation manner of the embodiment of the present invention, when the preset model is a linear regression model, the model determining unit may include:
the error calculation subunit is used for taking the difference between the estimated click rate and the actual click rate in each historical data as the error value of each historical data;
the average error calculation subunit is used for taking the average value of the error values corresponding to all the historical data as an average error value;
a calibration model deriving subunit configured to train the calibration model using the average error value;
the calibration model is as follows:
(x) x-b; wherein b is the average error value, x is the initial estimated click rate, and f (x) is the final estimated click rate.
According to the click rate estimation device provided by the embodiment of the invention, the initial estimated click rate output by the estimation model is calibrated again, so that the deviation between the estimated click rate obtained by the estimation model and the actual click rate can be reduced, and the estimation accuracy is improved.
The electronic device provided by the embodiment of the invention is described below.
The electronic device provided in the embodiment of the present invention, as shown in fig. 7, includes a processor 701, a communication interface 702, a memory 703 and a communication bus 704, where the processor 701, the communication interface 702, and the memory 703 complete mutual communication through the communication bus 704;
a memory 703 for storing a computer program;
the processor 701 is configured to implement the following steps when executing the program stored in the memory 703:
the method comprises the steps of taking the characteristics of an advertisement to be estimated as the input of a pre-trained estimation model, and taking the output result of the estimation model as the initial estimation click rate of the advertisement to be estimated, wherein the estimation model is obtained by training according to the characteristics of the advertisement to be estimated and the actual click rate of the advertisement to be estimated, wherein the characteristics are included in historical data of the advertisement to be estimated;
and taking the initial estimated click rate as the input of a pre-trained calibration model, and taking the output result of the calibration model as the final estimated click rate of the advertisement to be estimated, wherein the calibration model is obtained by training according to the actual click rate of the advertisement to be estimated and the initial estimated click rate of the advertisement to be estimated, which are included in the historical data of the advertisement to be estimated.
According to the electronic equipment provided by the embodiment of the invention, the initial estimated click rate output by the estimated model is calibrated again, so that the deviation between the estimated click rate obtained by the estimated model and the actual click rate can be reduced, and the estimation accuracy is improved.
The communication bus mentioned in the electronic device may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The communication bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is shown, but this is not intended to represent only one bus or type of bus.
The communication interface is used for communication between the electronic equipment and other equipment.
The Memory may include a Random Access Memory (RAM) or a Non-Volatile Memory (NVM), such as at least one disk Memory. Optionally, the memory may also be at least one memory device located remotely from the processor.
The Processor may be a general-purpose Processor, including a Central Processing Unit (CPU), a Network Processor (NP), and the like; but also Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs) or other Programmable logic devices, discrete Gate or transistor logic devices, discrete hardware components.
In another embodiment of the present invention, a computer-readable storage medium is further provided, where instructions are stored in the computer-readable storage medium, and when the instructions are executed on a computer, the computer is caused to execute the click rate estimation method in any one of the above embodiments.
According to the computer-readable storage medium provided by the embodiment of the invention, the initial estimated click rate output by the estimated model is calibrated again, so that the deviation between the estimated click rate obtained by the estimated model and the actual click rate can be reduced, and the estimation accuracy is improved.
In yet another embodiment of the present invention, there is also provided a computer program product comprising instructions which, when run on a computer, cause the computer to perform the click rate estimation method of any of the above embodiments.
According to the computer program product provided by the embodiment of the invention, the initial estimated click rate output by the estimated model is calibrated again, so that the deviation between the estimated click rate obtained by the estimated model and the actual click rate can be reduced, and the estimation accuracy is improved.
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When implemented in software, it may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, cause the processes or functions described in accordance with the embodiments of the invention to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, from one website site, computer, server, or data center to another website site, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that incorporates one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in the process, method, article, or apparatus that comprises the element.
All the embodiments in the present specification are described in a related manner, and the same and similar parts among the embodiments may be referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the apparatus/device/medium/program product embodiments, since they are substantially similar to the method embodiments, the description is relatively simple, and reference may be made to some descriptions of the method embodiments for relevant points.
The above description is only for the preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention shall fall within the protection scope of the present invention.
Claims (9)
1. A click through rate estimation method is characterized by comprising the following steps:
the method comprises the steps of taking the characteristics of an advertisement to be estimated as the input of a pre-trained estimation model, and taking the output result of the estimation model as the initial estimation click rate of the advertisement to be estimated, wherein the estimation model is obtained by training according to the characteristics of the advertisement to be estimated and the actual click rate of the advertisement to be estimated, wherein the characteristics are included in historical data of the advertisement to be estimated;
taking the initial estimated click rate as the input of a pre-trained calibration model, and taking the output result of the calibration model as the final estimated click rate of the advertisement to be estimated, wherein the calibration model is obtained by training according to the actual click rate of the advertisement to be estimated and the initial estimated click rate of the advertisement to be estimated, which are included in the historical data of the advertisement to be estimated;
the training process of the calibration model comprises the following steps:
Acquiring historical data of the advertisement to be estimated in a plurality of time periods, wherein each historical data comprises an initial estimated click rate of the advertisement to be estimated and an actual click rate corresponding to the initial estimated click rate;
training a preset model by using the historical data, and taking the trained preset model as the calibration model;
when the preset model is a order-preserving regression model, training the preset model by using the historical data comprises:
dividing the historical data into N sections of order-preserving data, wherein N is an integer greater than 1; each section of order preserving data comprises the same amount of historical data;
taking the average value of the initial estimated click rate in each section of order-preserving data as the average initial estimated click rate of the section of order-preserving data; taking the average value of the actual click rate in each section of order preserving data as the average actual click rate of the section of order preserving data;
and sorting the N sections of order-preserving data from small to large according to the average initial estimated click rate, fitting the average initial estimated click rate and the average actual click rate by using an order-preserving regression algorithm to obtain a calibration value of the average initial estimated click rate, and training the preset model by using the calibration value of the average initial estimated click rate and the average actual click rate.
2. The method according to claim 1, wherein the preset model further comprises any one of a logistic regression model, a linear regression model, and a factorization model.
3. The method of claim 1, wherein the calibration model is:
wherein d isiThe calibration value of the average estimated click rate of the ith segment of order-preserving data, di+1The calibration value p of the average initial estimated click rate of the i +1 th section of order-preserving dataiAverage initial estimated click rate, p, of the ith segment of order-preserving datai+1The average initial estimated click rate of the i +1 th section of order-preserving data, x is the initial estimated click rate to be calibrated, and f (x) is the final estimated click rate after calibration.
4. The method of claim 2, wherein when the predetermined model is a linear regression model, the training the predetermined model using the historical data comprises:
taking the difference between the initial estimated click rate and the actual click rate in each historical data as the error value of each historical data;
taking the average value of the error values corresponding to all the historical data as an average error value;
training the calibration model with the average error value;
the calibration model is as follows:
(x) x-b; wherein b is the average error value, x is the initial estimated click rate to be calibrated, and f (x) is the final estimated click rate after calibration.
5. A click rate estimation device, comprising:
the estimation unit is used for inputting the characteristics of the advertisement to be estimated as an estimation model trained in advance and using the output result of the estimation model as the initial estimation click rate of the advertisement to be estimated, wherein the estimation model is obtained by training according to the characteristics of the advertisement to be estimated and the actual click rate of the advertisement to be estimated, wherein the characteristics of the advertisement to be estimated are included in the historical data of the advertisement to be estimated;
the calibration unit is used for inputting the initial estimated click rate as a pre-trained calibration model and taking an output result of the calibration model as a final estimated click rate of the advertisement to be estimated, wherein the calibration model is obtained by training according to the actual click rate of the advertisement to be estimated and the initial estimated click rate of the advertisement to be estimated, which are included in the historical data of the advertisement to be estimated;
the data acquisition unit is used for acquiring historical data of the advertisement to be estimated in a plurality of time periods, wherein each historical data comprises an initial estimated click rate of the advertisement to be estimated and an actual click rate corresponding to the initial estimated click rate;
The model determining unit is used for training a preset model by using the historical data, and taking the trained preset model as the calibration model;
when the preset model is an order preserving regression model, the model determining unit includes:
the data dividing subunit is used for dividing the historical data into N sections of order-preserving data, wherein N is an integer greater than 1; each section of order preserving data comprises the same amount of historical data;
the click rate calculation subunit is used for taking the average value of the initial estimated click rate in each section of order-preserving data as the average initial estimated click rate of the section of order-preserving data; taking the average value of the actual click rate in each section of order preserving data as the average actual click rate of the section of order preserving data;
and the calibration model determining subunit is used for sequencing the N sections of order-preserving data from small to large according to the average initial estimated click rate, fitting the average initial estimated click rate and the average actual click rate by using an order-preserving regression algorithm to obtain a calibration value of the average initial estimated click rate, and training the preset model by using the calibration value of the average initial estimated click rate and the average actual click rate.
6. The apparatus of claim 5, wherein the predetermined model further comprises any one of a logistic regression model, a linear regression model, and a factorization machine model.
7. The apparatus of claim 5, wherein the calibration model is:
wherein, diA calibration value d of the average estimated click rate of the ith segment of order-preserving datai+1The calibration value p of the average initial estimated click rate of the i +1 th section of order-preserving dataiAverage initial estimated click rate, p, of the ith segment of order-preserving datai+1The average initial estimated click rate of the i +1 th section of order-preserving data, x is the initial estimated click rate to be calibrated, and f (x) is the final estimated click rate after calibration.
8. The apparatus of claim 6, wherein when the predetermined model is a linear regression model, the model determining unit comprises:
the error calculation subunit is used for taking the difference between the initial estimated click rate and the actual click rate in each historical data as the error value of each historical data;
the average error calculation subunit is used for taking the average value of the error values corresponding to all the historical data as an average error value;
a calibration model deriving subunit configured to train the calibration model using the average error value;
The calibration model is as follows:
(x) x-b; wherein b is the average error value, x is the initial estimated click rate to be calibrated, and f (x) is the final estimated click rate after calibration.
9. An electronic device is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor and the communication interface are used for realizing mutual communication by the memory through the communication bus;
a memory for storing a computer program;
a processor for implementing the method steps of any of claims 1 to 4 when executing a program stored in the memory.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810349459.0A CN108681915B (en) | 2018-04-18 | 2018-04-18 | Click rate estimation method and device and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810349459.0A CN108681915B (en) | 2018-04-18 | 2018-04-18 | Click rate estimation method and device and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108681915A CN108681915A (en) | 2018-10-19 |
| CN108681915B true CN108681915B (en) | 2022-06-03 |
Family
ID=63801112
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810349459.0A Active CN108681915B (en) | 2018-04-18 | 2018-04-18 | Click rate estimation method and device and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108681915B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110933499B (en) * | 2018-09-19 | 2021-12-24 | 飞狐信息技术（天津）有限公司 | Video click rate estimation method and device |
| CN109522506A (en) * | 2018-10-30 | 2019-03-26 | 广东原昇信息科技有限公司 | The dynamic prediction method of visitor's behavioral data conversion ratio |
| CN109670632B (en) * | 2018-11-26 | 2021-01-29 | 北京达佳互联信息技术有限公司 | Advertisement click rate estimation method, advertisement click rate estimation device, electronic device and storage medium |
| CN109978606B (en) * | 2019-03-04 | 2022-09-30 | 北京达佳互联信息技术有限公司 | Method and device for processing advertisement click rate data and computer readable storage medium |
| CN110263982A (en) * | 2019-05-30 | 2019-09-20 | 百度在线网络技术（北京）有限公司 | The optimization method and device of ad click rate prediction model |
| CN110399979B (en) * | 2019-06-17 | 2022-05-13 | 深圳大学 | Click rate pre-estimation system and method based on field programmable gate array |
| CN110866776B (en) * | 2019-10-12 | 2023-11-24 | 上海掌门科技有限公司 | Data calibration method for popularization resources, electronic equipment and readable storage medium |
| CN111028005B (en) * | 2019-11-29 | 2021-11-05 | 腾讯科技（深圳）有限公司 | Advertisement display method and device |
| CN111768251A (en) * | 2020-09-03 | 2020-10-13 | 北京悠易网际科技发展有限公司 | Advertisement putting method and device based on traffic information evaluation and electronic equipment |
| CN112270571B (en) * | 2020-11-03 | 2023-06-27 | 中国科学院计算技术研究所 | Meta-model training method for cold-start advertisement click rate estimation model |
| CN113222656B (en) * | 2021-04-29 | 2023-11-24 | 西安点告网络科技有限公司 | Programmed advertisement putting method, system, device, equipment and storage medium |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105046532A (en) * | 2015-08-07 | 2015-11-11 | 北京品友互动信息技术有限公司 | Bidding method and device |
| CN106296286A (en) * | 2016-08-09 | 2017-01-04 | 北京奇虎科技有限公司 | The predictor method of ad click rate and estimating device |
-
2018
- 2018-04-18 CN CN201810349459.0A patent/CN108681915B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN108681915A (en) | 2018-10-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108681915B (en) | Click rate estimation method and device and electronic equipment | |
| WO2020088007A1 (en) | Method and device for determining consumer financial default risk | |
| CN108985830B (en) | Recommendation scoring method and device based on heterogeneous information network | |
| US8812362B2 (en) | Method and system for quantifying user interactions with web advertisements | |
| JP6854748B2 (en) | Information providing equipment, information providing method, and program | |
| CN110544155A (en) | User credit score acquisition method, acquisition device, server and storage medium | |
| WO2014169776A1 (en) | Cluster method and apparatus based on user interest | |
| ITMI20061897A1 (en) | SYSTEMS AND METHODS FOR CLASSIFYING RESULTS OF SEARCH ENGINES | |
| WO2018214503A1 (en) | Method and device for setting sample weight, and electronic apparatus | |
| CN109165691B (en) | Training method and device for model for identifying cheating users and electronic equipment | |
| CN111798280B (en) | Multimedia information recommendation method, device and equipment and storage medium | |
| WO2007070205A1 (en) | Data independent relevance evaluation utilizing cognitive concept relationship | |
| WO2023000491A1 (en) | Application recommendation method, apparatus and device, and computer-readable storage medium | |
| CN111210258A (en) | Advertisement putting method and device, electronic equipment and readable storage medium | |
| US20180240037A1 (en) | Training and estimation of selection behavior of target | |
| CN103761266A (en) | Click rate predicting method and system based on multistage logistic regression | |
| CN114331521A (en) | Business data monitoring, analyzing and managing method, system, equipment and storage medium | |
| CN113590945B (en) | Book recommendation method and device based on user borrowing behavior-interest prediction | |
| CN108805332B (en) | Feature evaluation method and device | |
| CN111881007B (en) | Operation behavior judgment method, device, equipment and computer readable storage medium | |
| JP6866069B2 (en) | Information processing equipment, information processing methods and information processing programs | |
| CN107527128B (en) | Resource parameter determination method and equipment for advertisement platform | |
| CN112199500A (en) | Emotional tendency identification method and device for comments and electronic equipment | |
| CN113836412A (en) | Information recommendation method and device, electronic equipment and storage medium | |
| US20110208738A1 (en) | Method for Determining an Enhanced Value to Keywords Having Sparse Data |
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 |