CN116976982A

CN116976982A - Information delivery method, device, electronic equipment, storage medium and program product

Info

Publication number: CN116976982A
Application number: CN202310567322.3A
Authority: CN
Inventors: 胡乐
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2023-05-18
Filing date: 2023-05-18
Publication date: 2023-10-31

Abstract

The application discloses an information delivery method and a related product, wherein, the estimated conversion rate of candidate recommended information and the current expected conversion cost thereof are obtained; determining a target cost set in which the current expected conversion cost is located; determining a first target correction coefficient corresponding to the target cost set according to a preset corresponding relation between the cost set and the correction coefficient; correcting the estimated conversion rate according to the first target correction coefficient to obtain corrected estimated conversion rate; if the candidate recommendation information meets the release condition according to the corrected estimated conversion rate, releasing the candidate recommendation information; the correction coefficient corresponding to the cost set is determined according to the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the cost set, and is obtained after adjustment according to the overall change trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs. The application can improve the release efficiency of the recommended information.

Description

Information delivery method, device, electronic equipment, storage medium and program product

Technical Field

The present application relates to the field of information recommendation technologies, and in particular, to an information delivery method, an information delivery device, an electronic device, a storage medium, and a program product.

Background

At present, with the development of computer network technology, people use terminal devices to process various matters in life, work and study through a computer network more and more. Accordingly, computer network-based information recommendation techniques have evolved. Based on the information recommendation technology, more and more organizations and individuals start to realize the release of recommended information through a computer network, and attract objects to execute corresponding transformation behaviors to obtain benefits. The conversion behavior may be customized by the presenter of the recommendation information according to actual needs, for example, for an entity commodity recommendation object, the conversion behavior may be a purchase behavior of the entity commodity, for an application program recommendation object, the conversion behavior may be a registration behavior corresponding to the application program, and so on.

Whether the recommendation information is effectively put in can be measured according to whether corresponding conversion behaviors occur after the recommendation information is put in, for example, the recommendation information is converted after the recommendation information is put in and can be considered as one-time effective put in. However, in the related art, the recommended information is difficult to obtain conversion after delivery due to the influence of various factors, resulting in low delivery efficiency of the recommended information.

Disclosure of Invention

The embodiment of the application provides an information delivery method, an information delivery device, electronic equipment, a computer readable storage medium and a computer program product, which can improve the delivery efficiency of recommended information.

In a first aspect, the information delivery method provided by the present application includes:

obtaining the estimated conversion rate of the candidate recommendation information and the current expected conversion cost of the candidate recommendation information;

determining a target cost set in which the current expected conversion cost is located, wherein the cost set is obtained by dividing the current expected conversion cost according to the first historical expected conversion cost of the candidate recommendation information in order of magnitude;

determining a first target correction coefficient corresponding to the target cost set according to a preset corresponding relation between the cost set and the correction coefficient;

correcting the estimated conversion rate according to the first target correction coefficient to obtain corrected estimated conversion rate;

if the candidate recommendation information meets the release condition according to the corrected estimated conversion rate, releasing the candidate recommendation information;

the correction coefficient corresponding to the cost set is determined according to the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the cost set, and is obtained after adjustment according to the overall change trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs.

In a second aspect, the present application provides an information delivery apparatus, including:

the conversion rate acquisition module is used for acquiring the estimated conversion rate of the candidate recommendation information and the current expected conversion cost of the candidate recommendation information;

the set determining module is used for determining a target cost set in which the current expected conversion cost is located, wherein the cost set is obtained by dividing the current expected conversion cost according to the first historical expected conversion cost of the candidate recommendation information in order of magnitude;

the coefficient determining module is used for determining a first target correction coefficient corresponding to the target cost set according to a preset corresponding relation between the cost set and the correction coefficient;

the conversion rate correction module is used for correcting the estimated conversion rate according to the first target correction coefficient to obtain the corrected estimated conversion rate;

the information delivery module is used for delivering the candidate recommendation information if the candidate recommendation information meets the delivery condition according to the corrected estimated conversion rate;

Optionally, in an embodiment, the information delivery device provided by the present application further includes a coefficient generating module, configured to: for the ith cost set obtained by dividing, obtaining conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set, wherein i epsilon [1, N ] and N represent the number of the cost sets obtained by dividing; acquiring a historical exposure result corresponding to a first historical expected conversion cost in the ith cost set, wherein the exposure result comprises non-clicked, clicked and converted; and determining an initial correction coefficient corresponding to the ith cost set according to the historical exposure result and the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set.

Optionally, in an embodiment, the coefficient generating module is configured to determine, according to a historical exposure result, whether the candidate recommendation information is clicked in a historical exposure process corresponding to the ith cost set; if the candidate recommendation information is not clicked in the historical exposure process corresponding to the ith cost set, calculating a first average value of conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set; and determining the calculated first average value as an initial correction coefficient of the ith cost set.

Optionally, in an embodiment, the coefficient generating module is further configured to, if it is determined that the candidate recommendation information is clicked in a historical exposure process corresponding to the ith cost set, allocate a weighting weight to a conversion rate estimated deviation corresponding to a first historical expected conversion cost in the ith cost set according to a historical exposure result; according to the distributed weighting weights, weighting operation is carried out on conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set, and a weighting operation result is obtained; and determining the weighted operation result as an initial correction coefficient of the ith cost set; the historical exposure result is that the weight corresponding to the clicked is greater than the weight corresponding to the historical exposure result which is not clicked, and the historical exposure result is that the weight corresponding to the converted is greater than the weight corresponding to the historical exposure result which is clicked.

Optionally, in an embodiment, the coefficient generating module is further configured to determine a multiplicative adjustment coefficient of the initial correction coefficient corresponding to the ith cost set according to an overall variation trend and/or an overall stability degree of the second historical expected conversion cost corresponding to the recommendation policy to which the candidate recommendation information belongs; and performing multiplication adjustment on the initial correction coefficient of the ith cost set according to the multiplication adjustment coefficient to obtain the correction coefficient of the ith cost set.

Optionally, in an embodiment, the coefficient generating module is configured to obtain a change rate of the overall change trend if the overall change trend is rising and the overall stability is greater than or equal to a degree threshold; and determining a multiplicative adjustment coefficient of the initial correction coefficient corresponding to the ith cost set according to the change rate, wherein the multiplicative adjustment coefficient is inversely related to the change rate.

Optionally, in an embodiment, the coefficient generating module is further configured to obtain a second average value of all current expected conversion costs corresponding to a recommendation policy to which the candidate recommendation information belongs, if the overall variation trend is rising and the overall stability is less than the degree threshold; and determining a multiplication adjustment coefficient of the initial correction coefficient corresponding to the ith cost set according to the second average value, the current expected conversion cost and the initial correction coefficient, wherein the multiplication adjustment coefficient is positively correlated with the second average value and the initial correction coefficient and negatively correlated with the current expected conversion cost.

Optionally, in an embodiment, the coefficient generating module is further configured to determine, according to the initial correction coefficient of the ith cost set, a multiplicative adjustment coefficient of the initial correction coefficient corresponding to the ith cost set if the overall variation trend is decreasing, where the multiplicative adjustment coefficient is positively correlated with the initial correction coefficient of the ith cost set.

Optionally, in an embodiment, the conversion rate obtaining module is configured to obtain an information attribute feature and a context environmental feature of the candidate recommendation information, and obtain a first object attribute feature of a candidate audience object of the candidate recommendation information; inputting the information attribute characteristics, the context environment characteristics and the first object attribute characteristics into a conversion rate estimation model to estimate the conversion rate, so as to obtain the initial estimated conversion rate of the candidate recommended information; acquiring industry characteristics of industries to which the candidate recommendation information belongs and second object attribute characteristics of expected audience objects of the candidate recommendation information; acquiring recommendation strategy characteristics of recommendation strategies to which candidate recommendation information belongs and information type characteristics of the candidate recommendation information; inputting the information attribute characteristics, the context environment characteristics, the first object attribute characteristics, the second object attribute characteristics, the industry characteristics, the recommended strategy characteristics and the information type characteristics into a correction coefficient prediction model to predict the correction coefficient, and obtaining a second target correction coefficient corresponding to the initial estimated conversion rate; and correcting the initial estimated conversion rate according to the second target correction coefficient to obtain the estimated conversion rate.

Optionally, in an embodiment, the conversion rate obtaining module is configured to obtain a policy identifier of a recommendation policy to which the candidate recommendation information belongs, and a cost setting feature of a second historical expected conversion cost corresponding to the recommendation policy to which the candidate recommendation information belongs in a previous preset period; and taking the strategy identification and the cost setting characteristic as recommended strategy characteristics.

Optionally, in an embodiment, the information delivery device provided by the present application further includes a model training module, configured to obtain a sample information attribute feature, a sample context environmental feature, and a transformation tag of the sample recommendation information, where the transformation tag is used to indicate whether the sample recommendation information is transformed after exposure; inputting sample information attribute characteristics, sample context environment characteristics and first sample object attribute characteristics into a conversion rate estimation model to estimate conversion rate so as to obtain sample initial estimated conversion rate of sample recommended information; acquiring sample industry characteristics of industries to which the sample recommendation information belongs and second sample object attribute characteristics of expected audience objects of the sample recommendation information; acquiring recommendation policy characteristics of a recommendation policy to which sample recommendation information belongs and sample information type characteristics of the sample recommendation information; inputting the sample information attribute characteristics, the sample context environment characteristics, the first sample object attribute characteristics, the sample industry characteristics, the second sample object attribute characteristics, the recommended strategy characteristics and the sample information type characteristics into a correction coefficient prediction model to predict correction coefficients, and obtaining sample target correction coefficients corresponding to initial estimated conversion rates of samples; correcting the initial estimated conversion rate of the sample according to the sample target correction coefficient to obtain the estimated conversion rate of the sample recommended information; determining the prediction loss of the correction coefficient prediction model according to the sample estimated conversion rate and the conversion label of the sample recommended information; and updating the network parameters of the correction coefficient prediction model according to the prediction loss until a preset updating stop condition is met.

Optionally, in an embodiment, the model training module is configured to determine an exposure time of the sample recommendation information, and assign a loss weight to the sample recommendation information according to the exposure time; and determining the prediction loss of the correction coefficient prediction model according to the sample estimated conversion rate, the conversion label and the loss weight of the sample recommended information.

Optionally, in an embodiment, the model training module is configured to, for sample recommendation information belonging to a recommendation policy, allocate a loss weight to sample recommendation information belonging to the recommendation policy with a constraint that a loss weight corresponding to a previous change period of an exposure time is smaller than a loss weight corresponding to a subsequent change period of the exposure time, and a loss weight corresponding to the same change period of the exposure time is the same, if there is a periodic change in a third historical expected conversion cost corresponding to the recommendation policy over time; if the third historical expected conversion cost corresponding to the recommendation strategy does not have periodic variation along with time, the loss weight corresponding to the front exposure time is smaller than the loss weight corresponding to the rear exposure time, and the loss weight is distributed to sample recommendation information subordinate to the recommendation strategy.

Optionally, in an embodiment, the model training module is further configured to discard, for a recommended policy that the third history expects that the conversion cost changes periodically with time, sample recommendation information that exposure time under the recommended policy is before a preset change period.

Optionally, in an embodiment, the information delivery device further includes an information determining module, configured to determine, in response to a delivery request for a delivery location, RTA analogies information matching the delivery location, non-RTA analogies information matching the delivery location and having a number of changes of the expected conversion cost greater than a preset number of times, as the candidate recommendation information.

Optionally, in an embodiment, the information delivery device provided by the application further includes a correction setting module, configured to display a delivery interface to a candidate recommended information dispenser, where the delivery interface includes a correction control that indicates correction of the estimated conversion rate of the candidate recommended information;

the conversion rate acquisition module is used for responding to the selection operation of the correction control by the user, and acquiring the estimated conversion rate of the candidate recommendation information and the current expected conversion cost of the candidate recommendation information.

In a third aspect, the present application provides an electronic device, including a memory and a processor, where the memory stores a computer program, and the processor is configured to execute the computer program in the memory, to implement the steps in the information delivery method provided by the present application.

In a fourth aspect, the present application provides a computer readable storage medium storing a plurality of instructions adapted to be loaded by a processor, for implementing the steps in the information delivery method provided by the present application.

In a fifth aspect, the present application provides a computer program product comprising a computer program or instructions which, when executed by a processor, implement the steps in the information delivery method provided by the present application.

According to the application, a plurality of cost sets are divided according to the first historical expected conversion cost of candidate recommendation information and according to the size sequence, and corresponding correction coefficients are configured for each cost set according to the following constraint: and determining a correction coefficient corresponding to the cost set according to the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the cost set, and adjusting according to the overall change trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs. The correction coefficient is preliminarily determined in a cost set dividing mode, so that estimated deviation of conversion rates caused by different audiences under different expected conversion costs is eliminated, and the final correction coefficient of the cost set is obtained by combining the overall change trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs, and correcting the preliminarily determined correction coefficient according to the acceptance degree of the estimated deviation by the dispenser of the reflected candidate recommendation information. Therefore, when information is put in, the current expected conversion cost of the candidate recommended information can be obtained, a target cost set where the current expected conversion cost is located is determined, a first target correction coefficient corresponding to the target cost set is determined according to the preset corresponding relation between the cost set and the correction coefficient, and the estimated conversion rate is corrected according to the first target correction coefficient, so that the corrected estimated conversion rate is obtained. Therefore, within the receiving degree of the estimated deviation of the user, the estimated deviation of the conversion rate caused by different audiences under different expected conversion costs is eliminated, the conversion rate of the candidate recommended information is accurately estimated, if the candidate recommended information is determined to meet the release condition according to the corrected estimated conversion rate, the candidate recommended information is likely to be converted after release, and the candidate recommended information is released at the moment, so that the candidate recommended information is ensured to be converted as much as possible, and the release efficiency is improved.

Drawings

In order to more clearly illustrate the technical solutions of the present embodiments, the drawings that are needed in the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1a is a schematic view of a scenario of an information delivery system according to an embodiment of the present application;

FIG. 1b is a schematic flow chart of an information delivery method according to an embodiment of the present application;

FIG. 1c is a diagram showing a placement of recommended information at a placement location according to the present embodiment;

FIG. 1d is a schematic diagram of obtaining a predicted conversion rate through a conversion rate prediction model and a correction coefficient prediction model in the present embodiment;

FIG. 1e is an exemplary diagram of a correction control provided in this embodiment;

fig. 2 is another flow chart of the information delivery method provided by the embodiment of the application;

fig. 3 is a schematic structural diagram of an information delivery device according to an embodiment of the present application;

fig. 4 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

Detailed Description

It should be noted that the principles of the present application are illustrated as implemented in a suitable computing environment. The following description is based on illustrative embodiments of the application and should not be taken as limiting other embodiments of the application not described in detail herein.

In the following description of the present application reference is made to "some embodiments" which describe a subset of all possible embodiments, but it is to be understood that "some embodiments" can be the same subset or a different subset of all possible embodiments and can be combined with each other without conflict.

In the following description of the present application, the terms "first", "second", "third" and "third" are merely used to distinguish similar objects from each other, and do not represent a particular ordering of the objects, it being understood that the "first", "second", "third" may be interchanged with a particular order or sequence, as permitted, to enable embodiments of the present application described herein to be practiced otherwise than as illustrated or described herein.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of the application only and is not intended to be limiting of the application.

In order to improve accuracy of estimated conversion rate of recommended information, embodiments of the present application provide an information delivery method, an information delivery device, an electronic device, a computer readable storage medium, and a computer program product. The information delivery method can be executed by the information delivery device or by the electronic equipment integrated with the information delivery device.

The technical solutions of the present embodiment will be clearly and completely described below with reference to the drawings in the present embodiment, and it is apparent that the embodiments described below are only some embodiments of the present application, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to fall within the scope of the application.

Referring to fig. 1a, the present application further provides an information delivery system, as shown in fig. 1a, where the information delivery system includes an electronic device 100, and the information delivery apparatus provided by the present application is integrated in the electronic device 100. For example, the electronic device 100 may first obtain the estimated conversion rate of the candidate recommendation information and obtain the current expected conversion cost of the candidate recommendation information. Then, the electronic device further determines a target cost set in which the current expected conversion cost is located, wherein the cost set is obtained by dividing the cost set according to the first historical expected conversion cost of the candidate recommendation information in order of magnitude. Then, the electronic equipment determines a first target correction coefficient corresponding to the target cost set according to a preset corresponding relation between the cost set and the correction coefficient. And then correcting the obtained estimated conversion rate according to the determined first target correction coefficient to obtain the corrected estimated conversion rate. And finally, if the candidate recommendation information is determined to meet the release condition according to the corrected estimated conversion rate, releasing the candidate recommendation information. And determining a correction coefficient corresponding to the cost set according to the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the cost set, and adjusting according to the overall change trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs.

The electronic device 100 may be any device with a processor configured to have a processing capability, such as a mobile electronic device with a processor, such as a smart phone, a tablet computer, a palm computer, a notebook computer, a virtual reality device, an augmented reality device, or a mixed reality device, or a stationary electronic device with a processor, such as a desktop computer, a television, a server, or an industrial device.

In addition, as shown in fig. 1a, the information delivery system may further include a memory 200, configured to store raw data, intermediate data, and result data in the information delivery process, for example, the electronic device 100 stores the estimated conversion rate and the current expected conversion cost (raw data) of the obtained candidate recommended information, a first target correction coefficient (intermediate data) determined according to a preset correspondence between the cost set and the correction coefficient, and a corrected estimated conversion rate (result data) obtained by correcting the estimated conversion rate in the memory 200.

It should be noted that, the schematic view of the scenario of the information delivery system shown in fig. 1a is only an example, and the information delivery system and scenario described in the embodiment of the present application are for more clearly describing the technical solution of the embodiment of the present application, and do not constitute a limitation on the technical solution provided by the embodiment of the present application, and those skilled in the art can know that, with the evolution of the information delivery system and the appearance of a new service scenario, the technical solution provided by the embodiment of the present application is equally applicable to similar technical problems.

The following will describe in detail. The numbers of the following examples are not intended to limit the preferred order of the examples.

Referring to fig. 1b, fig. 1b is a schematic flow chart of an information delivery method according to an embodiment of the present application, and as shown in fig. 1b, the flow chart of the information delivery method according to the present application is as follows:

at 110, a predicted conversion rate of the candidate recommendation information and a current expected conversion cost of the candidate recommendation information are obtained.

In the following disclosure, a presenter refers to a party paying a cost to present recommended information, a recommendation platform refers to a party providing a recommended information presentation service by using a network platform or technology thereof, and an audience object refers to an object to which the recommended information is presented, such as a user of a terminal device, a viewer of an advertisement screen, and the like.

The recommendation information refers to information for recommending one or more recommendation objects (may be an entity object or a non-entity object), for example, announcement information, statement information, etc. for no purpose, advertisement information for purpose, etc. for a purpose of profit, etc. The recommendation information may be delivered by a presenter (which may be an organization or a person) via one or more recommendation platforms. After a recommendation message is put in through the recommendation platform, the following process is performed: firstly, an audience object sees the recommended information through a terminal device, and the process is called exposure; after exposure, the audience object of interest may click on the recommendation information, and the terminal device may present a new page or window to display the recommendation object, such as a commodity purchase page, an application download page, a video play page, a display page of announcement information, etc., which is referred to as clicking; after jumping to the presentation page of the recommended object by clicking, the audience object may operate the terminal device to purchase the merchandise object, download the application object, view the bulletin information, and so on, which is referred to as conversion. And for recommended information, determining whether the recommended information is converted after exposure according to conversion information returned by the dispenser.

The estimated conversion rate refers to the probability that the corresponding conversion behavior is executed after the audience object clicks the recommended information, and the probability is larger, the probability indicating that the conversion behavior is generated is larger.

The expected conversion cost refers to the expected cost of the single conversion behavior of the recommended information set by the presenter of the recommended information, in other words, the amount of resources of the recommended platform that the presenter expects for the single conversion behavior of the recommended information should interact with. For example, taking recommendation information as advertisement information, the expected conversion cost is the price expected by the advertisement dispenser for a single conversion action of the advertisement information.

Candidate recommendation information refers to recommendation information that may be delivered at a delivery location. For example, a network page is provided with a delivery position of recommended information, on one hand, when the terminal device browses the network page provided with the delivery position, a delivery request of the recommended information is triggered, on the other hand, an electronic device (belonging to a recommendation platform) for executing the information delivery method responds to the delivery request, and the recommended information matched with the delivery position is screened out from a preset recommended information set and used as candidate recommended information of the delivery position.

In this embodiment, for a candidate recommendation information, a predicted conversion rate of the candidate recommendation information is obtained. For example, the estimated conversion rate obtained by estimating the conversion rate of the candidate recommended information can be determined by the electronic equipment executing the information delivery method or obtained from other electronic equipment.

For example, a machine-learning-based conversion rate estimation model, or a non-machine-learning conventional algorithm may be employed to obtain the estimated conversion rate of the candidate recommendation information. For example, attribute features of candidate audience objects (i.e., objects to which a terminal device triggering a delivery request belongs) may be obtained (such as some related description information of the candidate audience objects), attribute features of candidate recommendation information (such as features of candidate recommendation information constituent elements, such as image elements, audio elements, text elements, etc.), and context environment features (some related description information of a delivery location, such as related information of display content of a page where the delivery location is located, etc.), and the obtained features are input into a conversion rate estimation model to perform conversion rate estimation, so as to obtain estimated conversion rate output by the conversion rate estimation model.

In addition, the expected conversion cost of the dispenser for the recommended information is usually dynamically changed, and at different moments, the dispenser can consider various factors to adjust the set expected conversion cost in real time, for example, when the dispenser is satisfied with the dispensing effect, the expected conversion cost is usually increased, and when the dispenser is not satisfied with the dispensing effect, the expected conversion cost is usually reduced. Correspondingly, the embodiment also obtains the expected conversion cost set by the user at the current moment of executing information release, and records the expected conversion cost for the candidate recommended information at the current moment as the current expected conversion cost.

In 120, a set of target costs at which the current expected conversion cost is located is determined, wherein the set of costs is partitioned in order of magnitude according to a first historical expected conversion cost of the candidate recommendation information.

In this embodiment, the historical expected conversion cost of the candidate recommended information is recorded as the first historical expected conversion cost, and the historical expected conversion cost of the candidate recommended information refers to the expected conversion cost set by the presenter at different times in a period of time before the current time for the candidate recommended information. For the selection of the time period, the selection is not particularly limited, and may be performed by a person skilled in the art according to actual needs, for example, the embodiment uses 7 continuous natural days as a period, and for the candidate recommendation information, selects the expected conversion cost set by the user at different times in 4 periods before the current time as the first historical expected conversion cost.

According to the application, when the expected conversion cost set by a user is different for the same recommended information, the exposure obtained by the recommended information is different from the audience object, when the expected conversion cost is set to be higher, the exposure of the recommended information is large, the magnitude of the audience object is also high, so that the magnitude of the candidate audience object facing the conversion rate estimation is also high, and the accuracy of the conversion rate estimation is affected; when the expected conversion cost is set to be low, the exposure of the recommended information is small, the audience object level is low, the candidate audience object level facing the conversion rate estimation is also low, and the influence on the conversion rate estimation accuracy is small. In addition, the trend of the expected conversion cost reflects the cost consideration behind the dispenser, and when the trend of the change is increased, the dispenser is satisfied with the expected conversion cost in the past period of time, and when the trend of the change is reduced, the dispenser is not satisfied with the expected conversion cost in the past period of time.

Based on this, in this embodiment, the first historical expected conversion cost of the candidate recommended information is divided into a plurality of sets in advance according to the order of magnitude, and is recorded as a cost set, and for each cost set, a corresponding correction coefficient is configured, where the correction coefficient can reflect the degree to which the estimated conversion rate is underestimated or overestimated, and is used to correct the estimated conversion rate. For example, when the correction coefficient value is negative, it reflects that the estimated conversion rate is underestimated, the value of the estimated conversion rate needs to be increased, and when the correction coefficient value is positive, it reflects that the estimated conversion rate is overestimated, the value of the estimated conversion rate needs to be reduced. The cost set dividing method is not particularly limited, for example, in this embodiment, the number of divided cost sets is positively correlated with the overall number of the first historical expected conversion costs, and the number of the first historical expected conversion costs included in each cost set is the same as a constraint, so as to divide the cost sets.

Wherein, the correction coefficients corresponding to different cost sets can be configured according to the following constraint.

And determining a correction coefficient corresponding to the cost set according to the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the cost set, and according to the overall change trend (the change trend reflected by all the second historical expected conversion costs corresponding to the recommendation strategies to which the candidate recommendation information belongs and the overall change trend) and/or the overall stability (the stability degree reflected by all the second historical expected conversion costs corresponding to the recommendation strategies to which the candidate recommendation information belongs and the overall stability degree) of the second historical expected conversion cost corresponding to the recommendation strategies to which the candidate recommendation information belongs, and obtaining the correction coefficient after adjustment.

It should be noted that, the recommendation policy is also called a delivery policy, and is used to instruct the presenter to set the logic, the assessment mode, the characteristics of the desired conversion cost, the desired audience object of the recommended information delivery (depending on the presenter, the desire of the presenter to deliver the recommended information), the place (may be a certain geographic area, or a specific type of terminal device of the delivery, for example, an advertisement device on a public transportation means, an advertisement device in an elevator, etc.), and so on. It should be noted that, for a presenter, a plurality of recommendation strategies are generally configured, and each recommendation strategy may belong to a plurality of recommendation information. The estimated conversion deviation is used to characterize the difference between the estimated conversion and the actual conversion, and may be measured as a percentage.

In this embodiment, the historical expected conversion cost of all the recommendation information belonging to the recommendation policy to which the candidate recommendation information belongs is recorded as the second historical expected conversion cost, and for a recommendation information belonging to the recommendation policy to which the candidate recommendation information belongs, the historical expected conversion cost refers to the expected conversion cost set by the presenter at different times in a period of time before the current time for the recommendation information. For the selection of the time period, the selection is not particularly limited in this embodiment, and a person skilled in the art may select the time period according to actual needs, for example, in this embodiment, the time set of configuring the second historical expected conversion cost span is smaller than the time set of configuring the first historical expected conversion cost span, and for the recommendation policy to which the candidate recommendation information belongs, the expected conversion cost of the recommendation information under the recommendation policy set by the presenter at different times in 2 natural days before the current time is selected as the second historical expected conversion cost.

When the correction coefficient is set, the correction coefficient is preliminarily determined in a cost set dividing mode, so that estimated deviation of conversion rates caused by different audiences under different expected conversion costs is eliminated, and the final correction coefficient of the cost set is obtained by combining the overall change trend and/or the overall stability degree of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs, and the reflected acceptance degree of the estimated deviation by the candidate recommendation information dispenser.

Correspondingly, after the current expected conversion cost of the candidate recommendation information is obtained, a cost set in which the current expected conversion cost is located is further determined and is recorded as a target cost set, so that a correction coefficient matched with the current expected conversion cost of the candidate recommendation information is selected based on the target cost set to be used for correcting the estimated conversion rate of the candidate recommendation information.

Optionally, in an embodiment, an optional configuration of the correction coefficient is provided, including:

for the ith cost set obtained by dividing, obtaining conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set, wherein i epsilon [1, N ] and N represent the number of the cost sets obtained by dividing;

Acquiring a historical exposure result corresponding to a first historical expected conversion cost in an ith cost set, wherein the exposure result comprises non-clicked, clicked and converted;

and determining an initial correction coefficient corresponding to the ith cost set according to the historical exposure result and the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set.

For a cost set, according to the difference of the historical exposure results corresponding to the first historical expected conversion cost in the cost set, the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the cost set is utilized in different modes to more accurately determine the initial correction coefficient corresponding to the cost set.

Optionally, in an embodiment, determining the initial correction coefficient corresponding to the ith cost set according to the historical exposure result and the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set includes:

determining whether the candidate recommendation information is clicked in the historical exposure process corresponding to the ith cost set according to the historical exposure result;

if the candidate recommendation information is determined not to be clicked in the historical exposure process corresponding to the ith cost set, calculating a first average value of conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set;

And determining the calculated first average value as an initial correction coefficient of the ith cost set.

The historical exposure process corresponding to the ith cost set of the candidate recommendation information, namely the exposure process when the candidate recommendation information is set as the first historical expected conversion cost in the ith cost set before the current moment.

If the candidate recommended information is determined not to be clicked in the historical exposure process corresponding to the ith cost set according to the historical exposure result, the candidate recommended information is not converted in the historical exposure process, the fact that the estimated conversion rate deviation corresponding to each first historical expected conversion cost in the ith cost set has the same contribution to the correction coefficient is judged, at the moment, the average value of the estimated conversion rate deviations corresponding to each first historical expected conversion cost in the ith cost set is calculated and recorded as a first average value, and the first average value is determined as the initial correction coefficient of the ith cost set.

Optionally, in an embodiment, after determining, according to the historical exposure result, whether the candidate recommendation information is clicked in the historical exposure process corresponding to the ith cost set, the method further includes:

If the candidate recommendation information is determined to be clicked in the historical exposure process corresponding to the ith cost set, distributing a weighting weight for the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set according to the historical exposure result;

according to the distributed weighting weights, weighting operation is carried out on conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set, and a weighting operation result is obtained;

and determining the weighted operation result as an initial correction coefficient of the ith cost set.

If it is determined that the candidate recommended information is clicked in the historical exposure process corresponding to the ith cost set according to the historical exposure result, it is indicated that the candidate recommended information may be converted in the historical exposure process, and it is considered that the contribution of the conversion rate estimated deviation corresponding to each first historical expected conversion cost in the ith cost set to the correction coefficient is different according to the different historical exposure results.

Correspondingly, taking the historical exposure result as the weight corresponding to the clicked weight greater than the historical exposure result as the weight corresponding to the non-clicked weight, taking the historical exposure result as the weight corresponding to the converted weight greater than the historical exposure result as the weight corresponding to the clicked weight as the constraint, distributing the weight for the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set according to the historical exposure result, carrying out the weight operation on the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set according to the distributed weight, obtaining the weight operation result, and determining the weight operation result as the initial correction coefficient of the ith cost set.

Optionally, in an embodiment, after determining the initial correction coefficient corresponding to the ith cost set according to the historical exposure result and the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set, the method further includes:

determining a multiplication adjustment coefficient of an initial correction coefficient corresponding to the ith cost set according to the overall change trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs;

and performing multiplication adjustment on the initial correction coefficient of the ith cost set according to the multiplication adjustment coefficient to obtain the correction coefficient of the ith cost set.

In this embodiment, further, according to the overall change trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation policy to which the candidate recommendation information belongs, the reflected receiving degree of the candidate recommendation information on the estimated deviation by the dispenser adopts a mode of performing multiplication adjustment on the initial correction coefficient to increase or decrease the correction degree of the correction coefficient on the estimated deviation, so that on the premise of improving the accuracy of the estimated conversion rate, the corrected estimated conversion rate can be ensured to be received by the dispenser.

Different recommendation information subordinate to the recommendation strategy required by the candidate recommendation information can be calculated, the cost average value of the second historical expected conversion cost at each historical moment is calculated, and the overall change trend of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs is measured according to the difference value of the cost average values of the two adjacent historical moments.

The method includes the steps of calculating the ratio between the average value of the costs and the time difference between two adjacent historical moments, if the calculated ratios are positive, considering that the overall change trend of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs is rising, and if the calculated ratios are negative, considering that the overall change trend of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs is falling; or if the ratio of the positive ratio to the total ratio calculated reaches the preset ratio (which can be set by a person skilled in the art according to actual needs), the overall change trend of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs is considered to be ascending, and if the ratio of the negative ratio calculated to the total ratio reaches the preset ratio, the overall change trend of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs is considered to be descending.

In addition, the overall stability of the second historical expected conversion cost corresponding to the recommendation policy to which the candidate recommendation information belongs may be measured by using statistics such as a mean value, a variance, or a standard deviation, for example, in this embodiment, the standard deviation of the second historical expected conversion cost corresponding to the recommendation policy to which the candidate recommendation information belongs is calculated, and the standard deviation is used to measure the overall stability of the second historical expected conversion cost corresponding to the recommendation policy to which the candidate recommendation information belongs. The overall stability reflects the variation range of the expected conversion cost set by the dispenser, and the greater the overall stability, the smaller the variation range of the expected conversion cost set by the dispenser, and the greater the variation range of the expected conversion cost set by the dispenser.

Optionally, in an embodiment, determining the multiplicative adjustment coefficient of the initial correction coefficient corresponding to the ith cost set according to the overall variation trend and/or the overall stability degree of the second historical expected conversion cost corresponding to the recommendation policy to which the candidate recommendation information belongs includes:

If the overall change trend is rising and the overall stability degree is greater than or equal to the degree threshold, acquiring the change rate of the overall change trend;

and determining a multiplication adjustment coefficient of the initial correction coefficient corresponding to the ith cost set according to the change rate of the overall change trend, wherein the multiplication adjustment coefficient is inversely related to the change rate.

In this embodiment, when the overall change trend of the second historical expected conversion cost corresponding to the recommendation policy to which the candidate recommendation information belongs is an increase, and the overall stability is greater than or equal to a degree threshold (used to measure whether the change range of the expected conversion cost set by the dispenser is stable or not, and can be configured by a person skilled in the art according to actual needs), it is determined that the dispenser is satisfied with the historical dispensing effect, the receiving degree of the estimated deviation is higher, the negative correlation between the multiplicative adjustment coefficient and the change rate of the overall change trend is a constraint, and a smaller multiplicative adjustment coefficient corresponding to the ith cost set is determined according to the change rate, so that the correction degree of the correction coefficient on the estimated conversion rate is reduced.

The change rate of the overall change trend can be measured according to the ratio between the difference between the cost average values of the two adjacent historical moments and the time difference value, which are calculated in the above embodiment. For example, the average value of all the ratios may be calculated as the change rate of the overall change trend.

if the overall change trend is rising and the overall stability is smaller than the degree threshold, acquiring a second average value of all current expected conversion costs corresponding to the recommendation strategy to which the candidate recommendation information belongs;

and determining a multiplication adjustment coefficient of the initial correction coefficient corresponding to the ith cost set according to the second average value, the current expected conversion cost of the candidate recommendation information and the initial correction coefficient, wherein the multiplication adjustment coefficient is positively correlated with the second average value and the initial correction coefficient, and is negatively correlated with the current expected conversion cost of the candidate recommendation information.

In this embodiment, when the overall change trend of the second historical expected conversion cost corresponding to the recommendation policy to which the candidate recommendation information belongs is an increase and the overall stability is less than the degree threshold, it is determined that the user is satisfied with the historical delivery effect, but is not willing to set a higher expected conversion cost for the low-value audience object. Specifically, when the expected conversion cost set by the presenter for a certain audience object is high, the presenter determines that the audience object is high in value and needs to strive for, and the correction logic for determining the calibration coefficient is as follows: strengthening the effect of underestimation while weakening the effect of overestimation; when the expected conversion cost set by the presenter to the audience object is low, the presenter considers the audience object to be low in value, and the correction logic for determining the calibration coefficient is as follows: the influence of underestimation is weakened, and the influence of overestimation is strengthened, so that the estimation is accurate, and the conversion effect is ensured. Correspondingly, a second average value of all current expected conversion costs corresponding to the recommendation strategy to which the candidate recommendation information belongs is obtained, and the multiplication adjustment coefficient matched with the current expected conversion cost of the candidate recommendation information is dynamically determined according to the current expected conversion cost of the candidate recommendation information set by a user by taking the positive correlation of the multiplication adjustment coefficient, the second average value and the initial correction coefficient as a constraint.

Optionally, in an embodiment, determining the multiplicative adjustment coefficient of the initial correction coefficient corresponding to the ith cost set according to the overall variation trend and/or the overall stability degree of the second historical expected conversion cost corresponding to the recommendation policy to which the candidate recommendation information belongs, further includes:

if the overall change trend is declining, determining a multiplication adjustment coefficient of the initial correction coefficient corresponding to the ith cost set according to the initial correction coefficient of the ith cost set, wherein the multiplication adjustment coefficient is positively correlated with the initial correction coefficient of the ith cost set.

In this embodiment, when the overall change trend of the second historical expected conversion cost corresponding to the recommendation policy to which the candidate recommendation information belongs is a decrease, it is determined that the user is not satisfied with the historical delivery effect, and at this time, the correction logic for determining the calibration coefficient is: underestimation is possible and overestimation is avoided. Correspondingly, the multiplication adjustment coefficient corresponding to the initial correction coefficient of the ith cost set is determined by taking the positive correlation of the multiplication adjustment coefficient and the initial correction coefficient of the ith cost set as a constraint.

The above determining, according to the overall variation trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation policy to which the candidate recommendation information belongs, the relevant content of the multiplicative adjustment coefficient of the initial correction coefficient corresponding to the ith cost set may be expressed as:

Wherein b _i A is a multiplicative adjustment coefficient representing an initial correction coefficient corresponding to the ith cost set _i ' represents the initial correction coefficient of the ith cost set, k represents the base parameter, and can be calculated by those skilled in the art according to actual needs at [0.1, 10]The internal value j represents the change rate of the overall change trend of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs, sigma represents the overall stability degree of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs, thresh represents the degree threshold, cpa represents the current expected conversion cost of the candidate recommendation information,and the second average value of all the current expected conversion costs corresponding to the recommendation strategies to which the candidate recommendation information belongs is represented.

According to the multiplication adjustment coefficient, the initial correction coefficient of the ith cost set is subjected to multiplication adjustment, so that the correction coefficient of the ith cost set is obtained, and the correction coefficient can be expressed as:

a _i ＝a _i ’×b；

in 130, a first target correction coefficient corresponding to the target cost set is determined according to a preset correspondence between the cost set and the correction coefficient.

As described above, the present embodiment configures corresponding correction coefficients for different cost sets in advance. Correspondingly, after determining a target cost set in which the current expected conversion cost of the candidate recommendation information is located, further determining a correction coefficient corresponding to the target cost set according to a preset corresponding relation between the cost set and the correction coefficient, and recording the correction coefficient as a first target correction coefficient.

In 140, the estimated conversion rate is corrected according to the first target correction coefficient, so as to obtain the corrected estimated conversion rate.

As described above, after determining the first target correction coefficient, that is, correcting the estimated conversion rate of the candidate recommended information according to the first target correction coefficient, it may be expressed as:

wherein a represents a first target correction factor, pCVR _original Representing predicted conversion rate, pCVR, of candidate recommendation information _final Indicating the estimated conversion after correction.

For example, when a= -20%, pCVR is indicated _original Is underestimated by 20%, at which time the pCVR of the current request should be used _original Multiplying by 1/(1-20%) =1.25 to obtain the final corrected pCVR _final 。

In 150, if it is determined that the candidate recommendation information meets the delivery condition according to the corrected estimated conversion rate, the candidate recommendation information is delivered.

In this embodiment, a delivery condition is configured, and the delivery condition is used to measure whether to deliver candidate recommendation information. The release condition can be configured only by taking the estimated conversion rate as a constraint, for example, the release condition is configured to release candidate recommendation information with the maximum estimated conversion rate; the launch conditions can also be constrained by both the estimated conversion and other launch-related parameters.

Exemplary, the configuration conditions of this embodiment are: and putting the candidate recommendation information with the maximum product of the current expected conversion cost, the estimated conversion rate (corrected) and the estimated click rate (the probability of clicking the audience object after representing the candidate recommendation information is exposed at the putting position). Referring to fig. 1c, a placement position in a page is shown, for each candidate recommendation information, products of a current expected conversion cost, a predicted click rate and a corrected predicted conversion rate of the candidate recommendation information can be calculated, and the candidate recommendation information with the largest corresponding product is placed in the placement position for exposure according to the products of the candidate recommendation information.

Optionally, in an embodiment, an artificial intelligence based method for obtaining the estimated conversion rate is provided. Artificial intelligence (Artificial Intelligence, AI) is a theory, method, technique, and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend, and extend human intelligence, sense the environment, acquire knowledge, and use knowledge to obtain optimal results. In other words, artificial intelligence is an integrated technology of computer science that attempts to understand the essence of intelligence and to produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence, i.e. research on design principles and implementation methods of various intelligent machines, enables the machines to have functions of sensing, reasoning and decision.

The artificial intelligence technology is a comprehensive subject, and relates to the technology with wide fields, namely the technology with a hardware level and the technology with a software level. Artificial intelligence infrastructure technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. Artificial intelligence software technology mainly includes Machine Learning (ML) technology, wherein Deep Learning (DL) is a new research direction in Machine Learning, which is introduced into Machine Learning to make it closer to an original target, i.e., artificial intelligence. At present, deep learning is mainly applied to the fields of machine vision, natural language processing and the like.

Deep learning is the inherent regularity and presentation hierarchy of learning sample data, and information obtained during such learning processes greatly aids in interpretation of data such as text, image and sound. By means of deep learning technology and corresponding training sets, network models realizing different functions can be obtained through training, for example, a conversion rate estimation model for estimating conversion rate of recommended information can be obtained through training based on one training set, a correction coefficient prediction model for predicting conversion rate of the corresponding conversion rate estimation model can be obtained through training based on another training set, the correction coefficient is used for correcting estimated conversion rate obtained through estimating of the conversion rate estimation model, and accuracy of the correction coefficient prediction model is improved. In this embodiment, obtaining the estimated conversion rate of the candidate recommendation information includes:

Acquiring information attribute characteristics and context environment characteristics of candidate recommendation information, and acquiring first object attribute characteristics of candidate audience objects of the candidate recommendation information;

inputting the information attribute characteristics, the context environment characteristics and the first object attribute characteristics into a conversion rate estimation model to estimate the conversion rate, so as to obtain the initial estimated conversion rate of the candidate recommended information;

acquiring industry characteristics of industries to which the candidate recommendation information belongs and second object attribute characteristics of expected audience objects of the candidate recommendation information;

acquiring recommendation strategy characteristics of recommendation strategies to which candidate recommendation information belongs and information type characteristics of the candidate recommendation information;

inputting the information attribute characteristics, the context environment characteristics, the first object attribute characteristics, the second object attribute characteristics, the industry characteristics, the recommended strategy characteristics and the information type characteristics into a correction coefficient prediction model to predict the correction coefficient, so as to obtain a second target correction coefficient corresponding to the initial estimated conversion rate;

and correcting the initial estimated conversion rate according to the second target correction coefficient to obtain the estimated conversion rate of the candidate recommended information.

In this embodiment, a conversion rate estimation model is pre-trained, where the conversion rate estimation model is configured to take as input information attribute characteristics of recommended information (such as characteristics of recommended information elements, such as image elements, audio elements, text elements, and the like), context environment characteristics (such as some related descriptive information triggering a delivery position of a delivery request, such as related information of display content of a page where the delivery position is located, and the like), and object attribute characteristics of candidate audience objects (such as some related descriptive information of the candidate audience objects), and take as output conversion rate of the recommended information after exposure to the candidate audience objects. It should be noted that, in this embodiment, the model architecture and the training mode of the conversion rate estimation model are not limited, and those skilled in the art may construct the model architecture of the conversion rate estimation model according to actual needs and select a suitable training mode. For example, the conversion rate estimation model can be trained by adopting a gradient descent mode based on the basic models of different architectures such as a depth cross model, a commodity-based neural network, a factorization machine and the like.

In addition, a correction coefficient prediction model is also trained in advance, the correction coefficient prediction model is configured to take information attribute characteristics, context environment characteristics, information type characteristics (used for indicating whether the recommended information is RTA analogized information or not), industry characteristics of industries to which the recommended information belongs (such as some related description information of the industries), object attribute characteristics of candidate audience objects, object attribute characteristics of expected audience objects and recommendation strategy characteristics of recommendation strategies to which the recommended information belongs as inputs, and takes correction coefficients as outputs, wherein the correction coefficients are used for correcting conversion rate output by the conversion rate prediction model and improving accuracy of the conversion rate. It should be noted that, in this embodiment, the model architecture and the training mode of the conversion rate estimation model are not limited, and those skilled in the art may construct the model architecture of the conversion rate estimation model according to actual needs and select a suitable training mode. For example, the embodiment adopts the same model architecture as the conversion rate estimation model to train the correction coefficient prediction model, and is different in that for the correction coefficient prediction model, recommended information samples of different industries are added, meanwhile, input features of the differences are added, the correction coefficient prediction model is trained by taking the correction coefficient as output, and the conversion rate estimation model is trained by taking the conversion rate as output.

RTA (Realtime API) is a technical means for meeting real-time and personalized delivery requirements of a dispenser, wherein real-time refers to real-time calling of an API interface, providing the dispenser with option, enabling the dispenser to select a recommendation strategy, setting expected conversion cost and the like before delivering recommendation information.

The RTA analogies information refers to recommendation information for realizing release based on RTA technology, and the non-RTA analogies information refers to recommendation information for realizing release not based on RTA technology.

Accordingly, referring to fig. 1d, when obtaining the estimated conversion rate of the candidate recommendation information, firstly, obtaining the information attribute feature and the context environmental feature of the candidate recommendation information, and obtaining the object attribute feature of the candidate audience object of the candidate recommendation information, and marking the obtained information attribute feature, the context environmental feature and the first object attribute feature as the first object attribute feature, then inputting the obtained information attribute feature, the obtained context environmental feature and the obtained first object attribute feature into a conversion rate estimation model for conversion rate estimation, and marking the conversion rate output by the conversion rate estimation model at the moment as the initial estimated conversion rate of the candidate recommendation information.

Further, industry characteristics of industries to which the candidate recommendation information belongs and object attribute characteristics of expected audience objects of the candidate recommendation information are obtained and recorded as second object attribute characteristics. In addition, the recommendation policy characteristics of the recommendation policies to which the candidate recommendation information belongs are also obtained, including:

Acquiring a strategy identification of a recommendation strategy to which the candidate recommendation information belongs and a cost setting feature of a second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs in a preset period;

and taking the strategy identification and the cost setting characteristic as recommended strategy characteristics.

The strategy identification is used for uniquely identifying the recommendation strategy, and can characterize information such as logic, assessment mode, characteristics, expected audience objects and places of recommendation information delivery, and the like of the user set expected conversion cost indicated by the identified recommendation strategy.

The preset period may be configured by those skilled in the art according to need, for example, the preset period is configured to be 7 consecutive natural days in this embodiment. For example, for the recommendation policy to which the candidate recommendation information belongs, the cost average value and standard deviation of the second historical expected conversion cost in each of seven natures before the current moment of the recommendation policy can be obtained, so that the cost setting feature in a multi-value form is obtained.

In this embodiment, an information type feature of the candidate recommendation information is also obtained, where the information type feature is used to indicate whether the candidate recommendation information is RTA analogized information.

After the above information attribute feature, the context environment feature, the first object attribute feature, the second object attribute feature, the industry feature, the recommended strategy feature and the information type feature are obtained, the obtained features are input into a correction coefficient prediction model to predict correction coefficients, and the correction coefficients output by the correction coefficient prediction model are recorded as second target correction coefficients for correcting the initial estimated conversion rate. And correcting the initial estimated conversion rate according to the second target correction coefficient to obtain the estimated conversion rate of the candidate recommended information.

The correction of the initial estimated conversion rate according to the second target correction coefficient can be expressed as:

pCVR _original ＝pCVR_bias×pCVR′ _original ；

wherein pCVR' _original Representing the initial estimated conversion, pcvr_bias represents the second target correction factor.

Optionally, in an embodiment, a training scheme of the correction coefficient prediction model is further provided, including:

acquiring sample information attribute characteristics, sample context environment characteristics and conversion labels of sample recommendation information, and acquiring first sample object attribute characteristics of candidate audience objects of the sample recommendation information, wherein the conversion labels are used for indicating whether the sample recommendation information is converted after exposure;

Inputting sample information attribute characteristics, sample context environment characteristics and first sample object attribute characteristics into a conversion rate estimation model to estimate conversion rate so as to obtain sample initial estimated conversion rate of sample recommended information;

acquiring sample industry characteristics of industries to which sample recommendation information belongs and second sample object attribute characteristics of expected audience objects of the sample recommendation information;

acquiring recommendation policy characteristics of a recommendation policy to which sample recommendation information belongs and sample information type characteristics of the sample recommendation information;

inputting the sample information attribute characteristics, the sample context environment characteristics, the first sample object attribute characteristics, the sample industry characteristics, the second sample object attribute characteristics, the recommended strategy characteristics and the sample information type characteristics into a correction coefficient prediction model to predict correction coefficients, and obtaining sample target correction coefficients corresponding to initial estimated conversion rates of samples;

correcting the initial estimated conversion rate of the sample according to the sample target correction coefficient to obtain the estimated conversion rate of the sample recommended information;

according to the sample estimated conversion rate and the conversion label of the sample recommended information, determining the prediction loss of the correction coefficient prediction model;

And updating network parameters of the correction coefficient prediction model according to the prediction loss until a preset updating stop condition is met.

The RTA analoging information can be selected as sample recommendation information, and the RTA analoging information can also be selected as sample recommendation information.

For a sample recommendation message, information attribute features of the sample recommendation message are obtained and marked as sample attribute features, context environment features of the sample recommendation message are marked as sample context environment features, and conversion labels of the sample recommendation message are used for indicating whether the sample recommendation message is converted in a historical exposure process. In addition, object attribute characteristics of candidate audience objects of the sample recommendation information are obtained and recorded as first sample object attribute characteristics. After the characteristics are obtained, the obtained characteristics are input into a conversion rate estimation model to estimate the conversion rate, and the conversion rate output by the conversion rate estimation model at the moment is recorded as the initial estimated conversion rate of the sample recommended information.

Further, the industry characteristic of the industry to which the sample recommendation information belongs is also obtained and is marked as a sample industry characteristic, and the object attribute characteristic of the expected audience object of the sample recommendation information is marked as a second sample object attribute characteristic. In addition, a recommended policy feature of the recommended policy to which the sample recommended information belongs (which may be implemented correspondingly by referring to the recommended policy feature of the recommended policy to which the candidate recommended information belongs in the above embodiment, including a policy identifier of the recommended policy and a cost setting feature of a third historical expected conversion cost corresponding to the recommended policy in a previous preset period) and a sample information type feature of the sample recommended information are also obtained. And after the characteristics are obtained, inputting the obtained characteristics into a correction coefficient prediction model to predict the correction coefficient, and recording the correction coefficient output by the correction coefficient prediction model at the moment as a sample target correction coefficient corresponding to the initial estimated conversion rate of the sample.

Further, for the sample recommended information, correcting the initial estimated conversion rate of the sample according to the corresponding sample target correction coefficient to obtain the estimated conversion rate of the sample recommended information. In the same manner as above, the sample estimated conversion rate of all the sample recommendation information can be obtained.

And finally, determining the prediction loss of the correction coefficient prediction model according to the sample estimated conversion rate and the conversion label of the sample recommended information, and updating the network parameters of the correction coefficient prediction model according to the prediction loss until the preset updating stop condition is met. The configuration of the preset update stop condition is not particularly limited, and may be configured by a person skilled in the art according to actual needs, for example, the preset update stop condition may be configured as follows: the predicted loss converges, and a preset update stop condition may be configured as follows: and the update times of the network parameters of the correction coefficient prediction model reach the preset times.

Wherein, the prediction loss can be expressed as:

wherein Loss represents the prediction Loss of the correction coefficient prediction model, n represents the number of sample recommendation information used for training the correction coefficient prediction model, y is a conversion label of the sample recommendation information (if the sample recommendation information is converted in the corresponding historical exposure process, y takes a value of 1, otherwise takes a value of 0), And representing the sample estimated conversion rate corresponding to the sample recommended information.

Optionally, in an embodiment, a time-dependent loss weight is introduced to enhance an adaptability of the correction coefficient prediction model to a change of a desired conversion cost with time, wherein determining a prediction loss of the correction coefficient prediction model according to a sample estimated conversion rate and a conversion label of sample recommendation information includes:

determining exposure time of the sample recommended information, and distributing loss weight for the sample recommended information according to the exposure time;

and determining the prediction loss of the correction coefficient prediction model according to the sample estimated conversion rate, the conversion label and the loss weight of the sample recommended information.

In this embodiment, for a sample recommendation information, a loss weight is assigned to the sample recommendation information according to the exposure time of the sample recommendation information. Illustratively, the sample recommendation information may be assigned a penalty weight according to the sample recommendation information exposure time, subject to the following constraints:

for sample recommendation information belonging to a recommendation strategy, if the third historical expected conversion cost corresponding to the recommendation strategy has periodic variation along with time, distributing loss weight for sample recommendation information belonging to the recommendation strategy by taking the loss weight corresponding to the previous variation period of the exposure time as smaller than the loss weight corresponding to the subsequent variation period, and the loss weights corresponding to the same variation period of the exposure time as constraints;

If the third historical expected conversion cost corresponding to the recommendation strategy does not have periodic variation along with time, the loss weight corresponding to the front exposure time is smaller than the loss weight corresponding to the rear exposure time, and the loss weight is distributed to sample recommendation information subordinate to the recommendation strategy.

The third expected conversion cost corresponding to the recommendation strategy refers to the expected conversion cost of the recommendation information subordinate to the recommendation strategy, and is set by the user of the recommendation strategy at different moments in a previous period of time. The periodic change of the third historical expected conversion cost corresponding to the recommended strategy with time refers to the periodic increase or the periodic decrease of the third historical expected conversion cost corresponding to the recommended strategy with time.

The exposure time preceding means that the exposure time is farther from the current time, and the exposure time following means that the exposure time is closer to the current time. Similarly, a change period is preceded by a corresponding time being farther from the current time, and a change period is followed by a corresponding time being closer to the current time.

When assigning the loss weight, assigning the same loss weight to the change period where the exposure time is located in the last of the recommended strategies with periodic change and the exposure time is in the last of the recommended strategies without periodic change, and assigning the loss weight to other sample recommendation information in the recommended strategies to which each belongs according to the constraint.

Illustratively, taking the sample recommendation information pertaining to a recommendation policy with periodic variation as an example, the loss weight may be assigned according to the following formula:

wherein weight represents a loss weight of the sample recommendation information, t represents an exposure time of the sample recommendation information, and the window function is used for determining a change period corresponding to the exposure time, for example, assuming that the determined change period is 7 natural days, window (1 day) =1, represents a first change period, window (3 days) =1, window (9 days) =2, and represents a second change period.

In this embodiment, according to the sample estimated conversion rate, the conversion label and the loss weight of the sample recommendation information, the prediction loss of the correction coefficient prediction model is determined, which may be expressed as:

optionally, in an embodiment, the information delivery method provided by the present application further includes:

and discarding sample recommendation information of which the exposure time is in front of a preset change period, wherein the sample recommendation information belongs to a recommendation strategy of which the third historical expected conversion cost changes periodically with time.

The preset change period refers to a change period far from the current moment, and can be configured by a person skilled in the art according to actual needs. In the embodiment, the contribution of the sample recommendation information before the exposure time in the preset change period to the prediction loss is determined to be smaller, the reference meaning is not possessed, and the sample recommendation information is discarded. For example, in this embodiment, for a recommended policy in which the third history expected conversion cost varies periodically with time, sample recommendation information, which is subordinate to the recommended policy and whose exposure time is 5 variation periods ago, is discarded.

Optionally, in an embodiment, before obtaining the estimated conversion rate of the candidate recommendation information and the current expected conversion cost of the candidate recommendation information, the method further includes:

and responding to the throwing request aiming at the throwing position, determining RTA analogies information matched with the throwing position, and non-RTA analogies information which is matched with the throwing position and has the expected conversion cost with the changing times larger than the preset times as candidate recommendation information.

It should be noted that, the information delivery method provided by the application is suitable for correcting the estimated conversion rate of the non-RTA analogies besides correcting the estimated conversion rate of the RTA analogies. The difference is that, for the non-RTA analogies, when determining the correction coefficients of the different cost sets, the overall variation trend and the overall stability of the second historical expected conversion cost corresponding to the recommendation strategy are replaced by the overall variation trend and the overall stability of the expected conversion cost set by the user of the non-RTA analogies (the obtaining manner can refer to the obtaining manner of the overall variation trend and the overall stability of the second historical expected conversion cost corresponding to the recommendation strategy in the above embodiment, which is not repeated here).

In addition, when the estimated conversion rate of the non-RTA analogy information is obtained, zero filling is performed on the recommendation strategy feature of the non-RTA analogy information, and the cost setting feature of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs in the previous preset period is replaced by the cost setting feature of the user of the non-RTA analogy information in the previous preset period (the obtaining manner can refer to the obtaining manner of the cost setting feature of the second historical expected conversion cost corresponding to the recommendation strategy in the above embodiment, and is not repeated here).

In addition, when the non-RTA analogies information is used as the sample recommendation information for training the correction coefficient prediction model, zero filling is performed on the recommendation strategy feature, the cost setting feature of the third historical expected conversion cost corresponding to the recommendation strategy to which the sample recommendation information belongs in the previous preset period is replaced by the cost setting feature of the presenter of the non-RTA analogies information in the previous preset period (the acquiring manner can refer to the acquiring manner of the cost setting feature of the second historical expected conversion cost corresponding to the recommendation strategy in the above embodiment, and the description is omitted here).

In general, for RTA referral information, more elaborate cost setting logic of its contributors to a particular recommendation strategy is employed for correction, while for non-RTA referral information, cost setting logic of its contributors overall is employed for correction. In addition to the above differences, other embodiments may be implemented correspondingly with reference to the above embodiments, and will not be described herein.

Optionally, in an embodiment, before providing the option of whether to perform conversion rate correction to the presenter and obtaining the estimated conversion rate of the candidate recommendation information and the current expected conversion cost of the candidate recommendation information, the method further includes:

displaying a delivery interface to a delivery person of the candidate recommendation information, wherein the delivery interface comprises a correction control for indicating to correct the estimated conversion rate of the candidate recommendation information;

obtaining the estimated conversion rate of the candidate recommendation information and the current expected conversion cost of the candidate recommendation information comprises the following steps:

and responding to the selection operation of the user on the correction control, and acquiring the estimated conversion rate of the candidate recommendation information and the current expected conversion cost of the candidate recommendation information.

It should be noted that, the delivery interface provided in this embodiment is an interface for configuring delivery of the recommended information, and in addition to normal delivery options (although not described in detail, it should be understood by those skilled in the art that the configurable delivery options generally include a desired audience object, a delivery location, etc. of the recommended information), a correction control is provided, where the correction control is used to instruct correction of the estimated conversion rate of the recommended information. The presentation form of the correction control is not particularly limited here, and may be configured by those skilled in the art according to actual needs. In specific implementation, a correction control may be used to instruct conversion rate correction to all recommended information of the presenter, and may also be used to instruct conversion rate correction to recommended information subordinate to a presenter's partial recommendation strategy.

For example, referring to fig. 1e, a correction control in the form of a selection box is shown, and prompt information indicating the function of the correction control indicates whether to perform conversion correction on the under-name recommended information, and when the correction control is selected, conversion correction is performed on all recommended information under the name of the presenter.

In this embodiment, when the correction control corresponding to the candidate recommendation information is selected, the estimated conversion rate of the candidate recommendation information and the current expected conversion cost of the candidate recommendation information are obtained, and the estimated conversion rate is corrected based on the current expected conversion cost of the candidate recommendation information, which is specifically referred to the related description in the above embodiment and will not be repeated herein.

As can be seen from the above, the present application divides a plurality of cost sets according to the size sequence according to the first historical expected conversion cost of the candidate recommendation information, and configures corresponding correction coefficients for each cost set according to the following constraint: and determining a correction coefficient corresponding to the cost set according to the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the cost set, and adjusting according to the overall change trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs. The correction coefficient is preliminarily determined in a cost set dividing mode, so that estimated deviation of conversion rates caused by different audiences under different expected conversion costs is eliminated, and the final correction coefficient of the cost set is obtained by combining the overall change trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs, and correcting the preliminarily determined correction coefficient according to the acceptance degree of the estimated deviation by the dispenser of the reflected candidate recommendation information. Therefore, when information is put in, the current expected conversion cost of the candidate recommended information can be obtained, a target cost set where the current expected conversion cost is located is determined, a first target correction coefficient corresponding to the target cost set is determined according to the preset corresponding relation between the cost set and the correction coefficient, and the estimated conversion rate is corrected according to the first target correction coefficient, so that the corrected estimated conversion rate is obtained. Therefore, within the receiving degree of the estimated deviation of the user, the estimated deviation of the conversion rate caused by different audiences under different expected conversion costs is eliminated, the conversion rate of the candidate recommended information is accurately estimated, if the candidate recommended information is determined to meet the release condition according to the corrected estimated conversion rate, the candidate recommended information is likely to be converted after release, and the candidate recommended information is released at the moment, so that the candidate recommended information is ensured to be converted as much as possible, and the release efficiency is improved.

According to the information delivery method provided by the above embodiment, the recommended information is used as RTA advertisement information, and the information delivery device is integrated in the electronic device for further detailed description.

Referring to fig. 2, the flow of the information delivery method may further be as follows:

at 210, the electronic device obtains information attribute features and contextual environmental features of the candidate RTA advertisement information, and obtains first object attribute features of candidate audience objects of the candidate RTA advertisement information.

When the terminal equipment browses a network page provided with a delivery position, a delivery request of RTA advertisement information is triggered, and on the other hand, the electronic equipment responds to the delivery request and screens out RTA advertisement information matched with the delivery position from a preset RTA advertisement information set to serve as candidate RTA advertisement information of the delivery position.

It should be noted that, in this embodiment, a conversion rate estimation model is pre-trained, where the conversion rate estimation model is configured to take as input information attribute features of advertisement information (such as features of advertisement information constituent elements, such as image elements, audio elements, text elements, and the like), context environment features (such as some related description information triggering a delivery location of a delivery request, such as related information of display content of a page where the delivery location is located, and the like), and object attribute features of candidate audience objects (such as some related description information of candidate audience objects) of the candidate audience objects (objects corresponding to terminal devices triggering the delivery request), and take as output conversion rate of the advertisement information after exposure to the candidate audience objects. It should be noted that, in this embodiment, the model architecture and the training mode of the conversion rate estimation model are not limited, and those skilled in the art may construct the model architecture of the conversion rate estimation model according to actual needs and select a suitable training mode. For example, the conversion rate estimation model can be trained by adopting a gradient descent mode based on the basic models of different architectures such as a depth cross model, a commodity-based neural network, a factorization machine and the like.

In this embodiment, the electronic device first obtains the information attribute feature and the context environment feature of the candidate RTA advertisement information, and obtains the object attribute feature of the candidate audience object of the candidate RTA advertisement information, and records the object attribute feature as the first object attribute feature.

In 220, the electronic device inputs the obtained information attribute feature, the context environmental feature and the first object attribute feature into a conversion rate estimation model to perform conversion rate estimation, so as to obtain an initial estimated conversion rate of the candidate RTA advertisement information.

After the characteristics are obtained, the electronic equipment inputs the obtained characteristics into a conversion rate estimation model to estimate the conversion rate, and the conversion rate output by the conversion rate estimation model at the moment is recorded as the initial estimated conversion rate of the candidate RTA advertisement information.

At 230, the electronic device obtains industry characteristics of the industry to which the candidate RTA advertisement information pertains, and second object attribute characteristics of the desired audience object of the candidate RTA advertisement information.

In 240, the electronic device obtains recommendation policy characteristics of a recommendation policy to which the candidate RTA advertisement information pertains, and information type characteristics of the candidate RTA advertisement information.

In 250, the electronic device inputs the obtained information attribute feature, the context environmental feature, the first object attribute feature, the second object attribute feature, the industry feature, the recommended policy feature and the information type feature into a correction coefficient prediction model to predict the correction coefficient, and obtains a second target correction coefficient corresponding to the initial estimated conversion rate.

It should be noted that, in this embodiment, a correction coefficient prediction model is further trained in advance, where the correction coefficient prediction model is configured to take as input an information attribute feature, a context environmental feature, an information type feature (for indicating whether the advertisement information is RTA advertisement information), an industry feature of the industry to which the advertisement information belongs (such as some related description information of the industry), an object attribute feature of a candidate audience object, an object attribute feature of an expected audience object, and a recommendation policy feature of a recommendation policy to which the advertisement information belongs, and take as output a correction coefficient, where the correction coefficient is used to correct a conversion rate output by the conversion rate prediction model, so as to improve accuracy of the conversion rate prediction model. It should be noted that, in this embodiment, the model architecture and the training mode of the conversion rate estimation model are not limited, and those skilled in the art may construct the model architecture of the conversion rate estimation model according to actual needs and select a suitable training mode. For example, the embodiment adopts the same model architecture as the conversion rate estimation model to train the correction coefficient prediction model, and is different in that advertisement information samples of different industries are added for the correction coefficient prediction model, meanwhile, the input characteristics of the differences are added, the correction coefficient prediction model is trained by taking the correction coefficient as output, and the conversion rate estimation model is trained by taking the conversion rate as output.

Further, the electronic device also obtains industry characteristics of the industry to which the candidate RTA advertisement information belongs and object attribute characteristics of the expected audience object of the candidate RTA advertisement information, and records the object attribute characteristics as second object attribute characteristics. In addition, the recommendation strategy characteristics of the recommendation strategy to which the candidate RTA advertisement information belongs are obtained, wherein the recommendation strategy characteristics comprise strategy identification, and cost setting characteristics of second historical expected conversion cost corresponding to the recommendation strategy to which the candidate RTA advertisement information belongs in a preset period.

The strategy identification is used for uniquely identifying the recommendation strategy, and can characterize information such as logic, assessment mode, characteristics, expected audience objects and places of advertisement information delivery, and the like of the user set expected conversion cost indicated by the identified recommendation strategy.

The preset period may be configured by those skilled in the art according to need, for example, the preset period is configured to be 7 consecutive natural days in this embodiment. For example, for a recommended policy to which the candidate RTA advertisement information belongs, a cost average value and a standard deviation of a second historical expected conversion cost in each of seven natures before the current moment of the recommended policy can be obtained, so that a multi-value cost setting feature is obtained.

In addition, the electronic device also obtains information type characteristics of the candidate RTA advertisement information, wherein the information type characteristics are used for indicating whether the candidate RTA advertisement information is RTA advertisement information.

After the above information attribute feature, the context environment feature, the first object attribute feature, the second object attribute feature, the industry feature, the recommended strategy feature and the information type feature are obtained, the electronic device inputs the obtained features into a correction coefficient prediction model to predict the correction coefficient, and the correction coefficient output by the correction coefficient prediction model is recorded as a second target correction coefficient for correcting the initial estimated conversion rate.

In 260, the electronic device corrects the initial estimated conversion rate of the candidate RTA advertisement information according to the second target correction coefficient, to obtain the estimated conversion rate of the candidate RTA advertisement information.

pCVR _original ＝pCVR_bias×pCVR′ _original ；

In 270, the electronic device obtains a pre-expected conversion cost of the candidate RTA advertisement information and determines a set of target costs at which the current expected conversion cost is located, wherein the set of costs is partitioned in order of magnitude according to a first historical expected conversion cost of the candidate RTA advertisement information.

It should be noted that, the expected conversion cost of the advertisement information by the presenter is usually dynamically changed, and at different moments, the presenter can consider various factors to adjust the set expected conversion cost in real time, for example, when the presenter is satisfied with the delivery effect, the presenter generally increases the expected conversion cost, and when the presenter is not satisfied with the delivery effect, the presenter generally decreases the expected conversion cost. Correspondingly, the embodiment also obtains the expected conversion cost set by the user at the current moment of executing information delivery, and records the expected conversion cost for the candidate RTA advertisement information at the current moment as the current expected conversion cost.

In order to distinguish, in this embodiment, the historical expected conversion cost of the candidate RTA advertisement information is recorded as the first historical expected conversion cost, where the historical expected conversion cost of the candidate RTA advertisement information refers to the expected conversion cost set by the presenter at different moments in a period of time before the current moment. The selection of the period of time is not particularly limited, and may be selected by a person skilled in the art according to actual needs, for example, the embodiment uses 7 consecutive natural days as a period, and selects, as the first historical expected conversion cost, the expected conversion cost set by the user at different times in 4 periods before the current time for the candidate RTA advertisement information.

According to the application, when the expected conversion cost set by a user is different for the same advertisement information, the exposure amount obtained by the advertisement information is different from the audience object, when the expected conversion cost is set to be higher, the exposure amount of the advertisement information is large, the magnitude of the audience object is also high, so that the magnitude of the candidate audience object facing when the conversion rate is estimated is also high, and the accuracy of conversion rate estimation is influenced; when the expected conversion cost is set to be low, the exposure of advertisement information is small, the audience object level is low, the candidate audience object level facing during conversion rate estimation is also low, and the influence on conversion rate estimation accuracy is small. In addition, the trend of the expected conversion cost reflects the cost consideration behind the dispenser, and when the trend of the change is increased, the dispenser is satisfied with the expected conversion cost in the past period of time, and when the trend of the change is reduced, the dispenser is not satisfied with the expected conversion cost in the past period of time.

Based on this, in this embodiment, the first historical expected conversion cost of the candidate RTA advertisement information is divided into a plurality of sets in advance according to the order of magnitude, and is recorded as a cost set, and for each cost set, a corresponding correction coefficient is configured, where the correction coefficient can reflect the degree to which the estimated conversion rate is underestimated or overestimated, and is used to correct the estimated conversion rate. For example, when the correction coefficient value is negative, it reflects that the estimated conversion rate is underestimated, the value of the estimated conversion rate needs to be increased, and when the correction coefficient value is positive, it reflects that the estimated conversion rate is overestimated, the value of the estimated conversion rate needs to be reduced. The cost set dividing method is not particularly limited, for example, in this embodiment, the number of divided cost sets is positively correlated with the overall number of the first historical expected conversion costs, and the number of the first historical expected conversion costs included in each cost set is the same as a constraint, so as to divide the cost sets.

Wherein, the correction coefficients corresponding to different cost sets can be configured according to the following constraint:

and determining a correction coefficient corresponding to the cost set according to the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the cost set, and according to the overall change trend (the change trend reflected by all the second historical expected conversion costs corresponding to the recommendation strategies to which the candidate RTA advertisement information belongs) and/or the overall stability (the stability degree reflected by all the second historical expected conversion costs corresponding to the recommendation strategies to which the candidate RTA advertisement information belongs) of the second historical expected conversion cost corresponding to the recommendation strategies to which the candidate RTA advertisement information belongs, and marking the overall stability degree) of the second historical expected conversion cost corresponding to the recommendation strategies to which the candidate RTA advertisement information belongs.

It should be noted that, the recommendation policy is also called a delivery policy, and is used to instruct the presenter to set the logic, the assessment mode, the characteristics of the desired conversion cost, the desired audience object of the advertisement information delivery (depending on the presenter, the desire of the presenter to deliver the advertisement information), the place (may be a certain geographic area, or a specific type of terminal device of the delivery, for example, an advertisement device on a public transportation means, an advertisement device in an elevator, etc.), and so on. It should be noted that, for a presenter, a plurality of recommendation strategies are generally configured, and each recommendation strategy may belong to a plurality of advertisement information. The estimated conversion deviation is used to characterize the difference between the estimated conversion and the actual conversion, and may be measured as a percentage.

In this embodiment, the historical expected conversion cost of all advertisement information under the recommendation policy to which the candidate RTA advertisement information belongs is recorded as the second historical expected conversion cost, and for an advertisement information under the recommendation policy to which the candidate RTA advertisement information belongs, the historical expected conversion cost refers to the expected conversion cost set by the presenter at different times in a period of time before the current time for the advertisement information. For the selection of the period of time, the selection is not particularly limited in this embodiment, and a person skilled in the art may select the selection according to actual needs, for example, in this embodiment, the time set of configuring the second historical expected conversion cost span is smaller than the time set of configuring the first historical expected conversion cost span, and for the recommendation policy to which the candidate RTA advertisement information belongs, the expected conversion cost of the advertisement information under the recommendation policy set by the presenter at different times in 2 natural days before the current time is selected as the second historical expected conversion cost.

When the correction coefficient is set, the correction coefficient is preliminarily determined in a cost set dividing mode, so that estimated deviation of conversion rates caused by different audiences under different expected conversion costs is eliminated, and the final correction coefficient of the cost set is obtained by combining the overall change trend and/or the overall stability degree of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate RTA advertisement information belongs, and the reflected acceptance degree of the estimated deviation of the candidate RTA advertisement information by a user is corrected.

Accordingly, after the current expected conversion cost of the candidate RTA advertisement information is obtained, a cost set in which the current expected conversion cost is located is further determined and is recorded as a target cost set, so that a correction coefficient matched with the current expected conversion cost of the candidate RTA advertisement information is selected based on the target cost set to be used for correcting the estimated conversion rate of the candidate RTA advertisement information.

In 280, the electronic device determines a first target correction coefficient corresponding to the target cost set according to a preset correspondence between the cost set and the correction coefficient, and corrects the estimated conversion rate according to the first target correction coefficient to obtain the corrected estimated conversion rate.

As described above, the present embodiment configures corresponding correction coefficients for different cost sets in advance. Correspondingly, after determining the target cost set in which the current expected conversion cost of the candidate RTA advertisement information is located, the electronic device further determines the correction coefficient corresponding to the target cost set according to the preset corresponding relation between the cost set and the correction coefficient, and records the correction coefficient as a first target correction coefficient.

As above, after determining the first target correction coefficient, that is, correcting the estimated conversion rate of the candidate RTA advertisement information according to the first target correction coefficient, it may be expressed as:

Wherein a represents a first target correction factor, pCVR _original Representing predicted conversion rate, pCVR, of candidate RTA advertisement information _final Indicating the estimated conversion after correction.

In 290, if it is determined that the candidate RTA advertisement information meets the delivery condition according to the corrected estimated conversion rate, the electronic device delivers the candidate RTA advertisement information.

In this embodiment, a placement condition is configured, where the placement condition is used to measure whether to place the candidate RTA advertisement information. The placement condition can be configured only by taking the estimated conversion rate as a constraint, for example, the placement condition is configured to place the candidate RTA advertisement information with the maximum estimated conversion rate; the launch conditions can also be constrained by both the estimated conversion and other launch-related parameters.

Exemplary, the configuration conditions of this embodiment are: and putting the candidate recommendation information with the maximum product of the current expected conversion cost, the estimated conversion rate (corrected) and the estimated click rate (the probability of clicking the audience object after representing the candidate recommendation information is exposed at the putting position).

In order to facilitate better implementation of the information delivery method provided by the embodiment of the application, the embodiment of the application also provides an information delivery device based on the information delivery method. The meaning of the nouns is the same as that of the information delivery method, and specific implementation details refer to the description in the embodiment of the method.

Referring to fig. 3, fig. 3 is a schematic structural diagram of an information delivery device according to an embodiment of the present application, where the information delivery device may include a conversion rate obtaining module 310, a set determining module 320, a coefficient determining module 330, a conversion rate correcting module 340, and an information delivery module 350,

a conversion rate obtaining module 310, configured to obtain an estimated conversion rate of the candidate recommendation information and a current expected conversion cost of the candidate recommendation information;

the set determining module 320 is configured to determine a target cost set in which the current expected conversion cost is located, where the cost set is obtained by dividing the cost set according to a first historical expected conversion cost of the candidate recommendation information in order of magnitude;

the coefficient determining module 330 is configured to determine a first target correction coefficient corresponding to the target cost set according to a preset correspondence between the cost set and the correction coefficient;

the conversion rate correction module 340 is configured to correct the estimated conversion rate according to the first target correction coefficient, so as to obtain a corrected estimated conversion rate;

the information delivery module 350 is configured to deliver the candidate recommendation information if it is determined that the candidate recommendation information meets the delivery condition according to the corrected estimated conversion rate;

Optionally, in an embodiment, the conversion rate obtaining module 310 is configured to obtain an information attribute feature and a context environmental feature of the candidate recommendation information, and obtain a first object attribute feature of the candidate audience object of the candidate recommendation information; inputting the information attribute characteristics, the context environment characteristics and the first object attribute characteristics into a conversion rate estimation model to estimate the conversion rate, so as to obtain the initial estimated conversion rate of the candidate recommended information; acquiring industry characteristics of industries to which the candidate recommendation information belongs and second object attribute characteristics of expected audience objects of the candidate recommendation information; acquiring recommendation strategy characteristics of recommendation strategies to which candidate recommendation information belongs and information type characteristics of the candidate recommendation information; inputting the information attribute characteristics, the context environment characteristics, the first object attribute characteristics, the second object attribute characteristics, the industry characteristics, the recommended strategy characteristics and the information type characteristics into a correction coefficient prediction model to predict the correction coefficient, and obtaining a second target correction coefficient corresponding to the initial estimated conversion rate; and correcting the initial estimated conversion rate according to the second target correction coefficient to obtain the estimated conversion rate.

the conversion rate obtaining module 310 is configured to obtain, in response to a selection operation of the correction control by the presenter, an estimated conversion rate of the candidate recommendation information and a current expected conversion cost of the candidate recommendation information.

The specific implementation of each module can be referred to the previous embodiments, and will not be repeated here.

In this embodiment, according to the first historical expected conversion cost of the candidate recommendation information, a plurality of cost sets are divided according to the size sequence, and for each cost set, a corresponding correction coefficient is configured according to the following constraint: and determining a correction coefficient corresponding to the cost set according to the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the cost set, and adjusting according to the overall change trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs. The correction coefficient is preliminarily determined in a cost set dividing mode, so that estimated deviation of conversion rates caused by different audiences under different expected conversion costs is eliminated, and the final correction coefficient of the cost set is obtained by combining the overall change trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs, and correcting the preliminarily determined correction coefficient according to the acceptance degree of the estimated deviation by the dispenser of the reflected candidate recommendation information. In this way, when information is put in, the conversion rate obtaining module 310 may obtain the current expected conversion cost of the candidate recommended information, the set determining module 320 determines the target cost set where the current expected conversion cost is located, the coefficient determining module 330 determines the first target correction coefficient corresponding to the target cost set according to the preset correspondence between the cost set and the correction coefficient, and the conversion rate correcting module 340 corrects the estimated conversion rate according to the first target correction coefficient to obtain the corrected estimated conversion rate. Therefore, within the receiving degree of the estimated deviation of the user, the estimated deviation of the conversion rate caused by different audiences under different expected conversion costs is eliminated, the conversion rate of the candidate recommended information is accurately estimated, if the candidate recommended information is determined to meet the release condition according to the corrected estimated conversion rate, the candidate recommended information is likely to be converted after release, and the candidate recommended information is released at the moment, so that the candidate recommended information is ensured to be converted as much as possible, and the release efficiency is improved.

The embodiment of the application also provides electronic equipment, which comprises a memory and a processor, wherein the processor is used for executing the steps in the information delivery method provided by the embodiment by calling the computer program stored in the memory.

Referring to fig. 4, fig. 4 is a schematic structural diagram of an electronic device according to an embodiment of the application.

The electronic device may include one or more processing cores 'processors 101, one or more computer-readable storage media's memory 102, power supply 103, and input unit 104, among other components. Those skilled in the art will appreciate that the electronic device structure shown in fig. 4 is not limiting of the electronic device and may include more or fewer components than shown, or may combine certain components, or may be arranged in different components. Wherein:

the processor 101 is a control center of the electronic device, connects various parts of the entire electronic device using various interfaces and lines, performs various functions of the electronic device and processes data by running or executing software programs and/or modules stored in the memory 102, and invoking data stored in the memory 102. Optionally, processor 101 may include one or more processing cores; alternatively, the processor 101 may integrate an application processor that primarily handles operating systems, object interfaces, applications, etc., with a modem processor that primarily handles wireless communications. It will be appreciated that the modem processor described above may not be integrated into the processor 101.

The memory 102 may be used to store software programs and modules, and the processor 101 executes various functional applications and data processing by executing the software programs and modules stored in the memory 102. The memory 102 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program (such as a sound playing function, an image playing function, etc.) required for at least one function, and the like; the storage data area may store data created according to the use of the electronic device, etc. In addition, memory 102 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 102 may also include a memory controller to provide access to the memory 102 by the processor 101.

The electronic device further comprises a power supply 103 for powering the various components, optionally, the power supply 103 may be logically connected to the processor 101 by a power management system, whereby the functions of managing charging, discharging, and power consumption are performed by the power management system. The power supply 103 may also include one or more of any of a direct current or alternating current power supply, a recharging system, a power failure detection circuit, a power converter or inverter, a power status indicator, and the like.

The electronic device may further comprise an input unit 104, which input unit 104 may be used for receiving input digital or character information and generating keyboard, mouse, joystick, optical or trackball signal inputs in connection with object settings and function control.

Although not shown, the electronic device may further include a display unit, an image acquisition component, and the like, which are not described herein. In particular, in this embodiment, the processor 101 in the electronic device loads executable codes corresponding to one or more computer programs into the memory 102 according to the following instructions, and the steps in the information delivery method provided by the present application are executed by the processor 101, for example:

It should be noted that, the electronic device provided in the embodiment of the present application and the information delivery method in the above embodiment belong to the same concept, and detailed implementation processes of the electronic device are shown in the above related embodiments, which are not repeated herein.

The present application also provides a computer-readable storage medium, on which a computer program is stored, which when executed on a processor of an electronic device provided by an embodiment of the present application, causes the processor of the electronic device to execute the steps in the information delivery method provided by the present application. The storage medium may be a magnetic disk, an optical disk, a Read Only Memory (ROM), a random access Memory (Random Access Memory, RAM), or the like.

The present application also provides a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions to cause the computer device to perform various alternative implementations of the information delivery method described above.

The foregoing has described in detail the method, apparatus, electronic device, storage medium and program product for information delivery provided by the present application, and specific examples have been applied to illustrate the principles and embodiments of the present application, and the description of the foregoing examples is only used to help understand the method and core idea of the present application; meanwhile, as those skilled in the art will have variations in the specific embodiments and application scope in light of the ideas of the present application, the present description should not be construed as limiting the present application.

It should be noted that when the above embodiments of the present application are applied to specific products or technologies, related data concerning users are required to obtain user approval or consent, and the collection, use and processing of the related data are required to comply with related laws and regulations and standards of related countries and regions.

Claims

1. An information delivery method is characterized by comprising the following steps:

obtaining the estimated conversion rate of candidate recommendation information and the current expected conversion cost of the candidate recommendation information;

and determining a correction coefficient corresponding to the cost set according to the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the cost set, and adjusting according to the overall change trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs.

2. The information delivery method according to claim 1, wherein before the obtaining the estimated conversion rate of the candidate recommended information and the current expected conversion cost of the candidate recommended information, further comprises:

for the ith cost set obtained by dividing, obtaining conversion rate estimated deviation corresponding to a first historical expected conversion cost in the ith cost set, wherein i epsilon [1, N ] and N represent the number of the cost sets obtained by dividing;

acquiring a historical exposure result corresponding to a first historical expected conversion cost in the ith cost set, wherein the exposure result comprises non-clicked, clicked and converted;

and determining an initial correction coefficient corresponding to the ith cost set according to a historical exposure result and a conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set.

3. The information delivery method according to claim 2, wherein the determining an initial correction coefficient corresponding to the ith cost set according to the historical exposure result and the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set includes:

determining whether the candidate recommended information is clicked in the historical exposure process corresponding to the ith cost set according to the historical exposure result;

If the candidate recommendation information is determined not to be clicked in the historical exposure process corresponding to the ith cost set, calculating a first average value of conversion rate estimated deviation corresponding to a first historical expected conversion cost in the ith cost set;

4. The information delivery method according to claim 3, wherein the determining, according to the historical exposure result, whether the candidate recommendation information is clicked in the historical exposure process corresponding to the ith cost set further includes:

determining the weighted operation result as an initial correction coefficient of the ith cost set;

The historical exposure result is that the weight corresponding to the clicked is greater than the weight corresponding to the historical exposure result which is not clicked, and the historical exposure result is that the weight corresponding to the converted is greater than the weight corresponding to the historical exposure result which is clicked.

5. The information delivery method according to claim 2, wherein after determining the initial correction coefficient corresponding to the ith cost set according to the historical exposure result and the conversion rate estimated deviation corresponding to the first historical expected conversion cost in the ith cost set, the method further comprises:

6. The information delivery method according to claim 5, wherein the determining the multiplicative adjustment coefficient of the initial correction coefficient corresponding to the ith cost set according to the overall variation trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation policy to which the candidate recommendation information belongs includes:

If the overall change trend is rising and the overall stability degree is greater than or equal to a degree threshold, acquiring the change rate of the overall change trend;

and determining a multiplication adjustment coefficient corresponding to the initial correction coefficient of the ith cost set according to the change rate, wherein the multiplication adjustment coefficient is inversely related to the change rate.

7. The information delivery method according to claim 6, wherein the determining the multiplicative adjustment coefficient of the initial correction coefficient corresponding to the ith cost set according to the overall variation trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation policy to which the candidate recommendation information belongs further comprises:

and determining a multiplication adjustment coefficient of an initial correction coefficient corresponding to the ith cost set according to the second average value, the current expected conversion cost and the initial correction coefficient, wherein the multiplication adjustment coefficient is positively correlated with the second average value and the initial correction coefficient and negatively correlated with the current expected conversion cost.

8. The information delivery method according to claim 5, wherein the determining the multiplicative adjustment coefficient of the initial correction coefficient corresponding to the ith cost set according to the overall variation trend and/or the overall stability of the second historical expected conversion cost corresponding to the recommendation policy to which the candidate recommendation information belongs further comprises:

if the overall change trend is declining, determining a multiplication adjustment coefficient corresponding to the initial correction coefficient of the ith cost set according to the initial correction coefficient of the ith cost set, wherein the multiplication adjustment coefficient is positively correlated with the initial correction coefficient of the ith cost set.

9. The information delivery method according to claim 1, wherein the obtaining the estimated conversion rate of the candidate recommended information includes:

acquiring information attribute characteristics and context environmental characteristics of the candidate recommendation information, and acquiring first object attribute characteristics of candidate audience objects of the candidate recommendation information;

inputting the information attribute characteristics, the context environment characteristics and the first object attribute characteristics into a conversion rate estimation model to estimate conversion rate so as to obtain initial estimated conversion rate of the candidate recommended information;

acquiring recommendation strategy characteristics of recommendation strategies to which the candidate recommendation information belongs and information type characteristics of the candidate recommendation information;

inputting the information attribute characteristics, the context environment characteristics, the first object attribute characteristics, the second object attribute characteristics, the industry characteristics, the recommended strategy characteristics and the information type characteristics into a correction coefficient prediction model to predict correction coefficients, so as to obtain a second target correction coefficient corresponding to the initial estimated conversion rate;

and correcting the initial estimated conversion rate according to the second target correction coefficient to obtain the estimated conversion rate.

10. The information delivery method according to claim 9, wherein the obtaining the recommendation policy feature of the recommendation policy to which the candidate recommendation information belongs includes:

acquiring a strategy identifier of a recommendation strategy to which the candidate recommendation information belongs and a cost setting feature of a second historical expected conversion cost corresponding to the recommendation strategy to which the candidate recommendation information belongs in a preset period;

And taking the strategy identification and the cost setting characteristic as the recommended strategy characteristic.

11. The information delivery method according to claim 9, wherein before the obtaining the estimated conversion rate of the candidate recommended information and the current expected conversion cost of the candidate recommended information, further comprises:

inputting the sample information attribute characteristics, the sample context environmental characteristics and the first sample object attribute characteristics into a conversion rate estimation model to estimate conversion rate so as to obtain sample initial estimated conversion rate of the sample recommended information;

acquiring sample industry characteristics of the industry to which the sample recommendation information belongs and second sample object attribute characteristics of an expected audience object of the sample recommendation information;

acquiring recommendation policy characteristics of a recommendation policy to which the sample recommendation information belongs and sample information type characteristics of the sample recommendation information;

Inputting the sample information attribute characteristics, the sample context environment characteristics, the first sample object attribute characteristics, the sample industry characteristics, the second sample object attribute characteristics, the recommended strategy characteristics and the sample information type characteristics into a correction coefficient prediction model to predict correction coefficients, so as to obtain a sample target correction coefficient corresponding to the initial estimated conversion rate of the sample;

determining the prediction loss of the correction coefficient prediction model according to the sample estimated conversion rate and the conversion label of the sample recommended information;

and updating the network parameters of the correction coefficient prediction model according to the prediction loss until a preset updating stop condition is met.

12. The information delivery method according to claim 11, wherein the determining the prediction loss of the correction coefficient prediction model according to the sample estimated conversion rate and the conversion label of the sample recommended information includes:

determining the exposure time of the sample recommended information, and distributing loss weight for the sample recommended information according to the exposure time;

13. The information delivery method according to claim 12, wherein the assigning a loss weight to the sample recommended information according to the exposure time includes:

14. The information delivery method of claim 13, further comprising:

15. The information delivery method according to any one of claims 1 to 14, wherein before obtaining the estimated conversion rate of the candidate recommended information and the current expected conversion cost of the candidate recommended information, further comprises:

responding to a delivery request aiming at a delivery position, determining RTA analogies information matched with the delivery position, non-RTA analogies information matched with the delivery position and with expected conversion cost changing times larger than preset times as the candidate recommendation information.

16. The information delivery method according to any one of claims 1 to 14, wherein before obtaining the estimated conversion rate of the candidate recommended information and the current expected conversion cost of the candidate recommended information, further comprises:

displaying a release interface to a release person of the candidate recommendation information, wherein the release interface comprises a correction control for indicating to correct the estimated conversion rate of the candidate recommendation information;

the obtaining the estimated conversion rate of the candidate recommended information and the current expected conversion cost of the candidate recommended information comprises the following steps:

17. An information delivery apparatus, comprising:

the conversion rate correction module is used for correcting the estimated conversion rate according to the first target correction coefficient to obtain corrected estimated conversion rate;

the information delivery module is used for delivering the candidate recommendation information if the candidate recommendation information is determined to meet the delivery conditions according to the corrected estimated conversion rate;

18. An electronic device comprising a memory storing a computer program and a processor for running the computer program in the memory to perform the steps of the information delivery method of any of claims 1 to 16.

19. A computer readable storage medium storing a plurality of instructions adapted to be loaded by a processor to perform the steps in the information delivery method of any of claims 1 to 16.

20. A computer program product comprising a computer program or instructions which, when executed by a processor, carries out the steps of the information delivery method of any one of claims 1 to 16.