CN116976974A

CN116976974A - Information promotion method, information promotion device, computer equipment and storage medium

Info

Publication number: CN116976974A
Application number: CN202310260662.1A
Authority: CN
Inventors: 田凯; 吴晔; 姜磊; 姚伶伶
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2023-03-10
Filing date: 2023-03-10
Publication date: 2023-10-31

Abstract

The application relates to an information promotion method, an information promotion device, computer equipment, a storage medium and a computer program product. According to the method, the information characteristics of information to be promoted and the multidimensional object characteristics of a target promotion receiving object are obtained, the information characteristics and the multidimensional object characteristics are processed by utilizing a pre-trained pre-estimation model, initial promotion values are respectively predicted by each predictor of the pre-estimation model, the probability of each predictor can be output according to the multidimensional object characteristics and the information characteristics, as the predictors are trained for fitting the data distribution of the logarithmic unimodal intervals of the promotion values according to the multidimensional object characteristics and the information characteristics, for the promotion values of the logarithmic peak distribution characteristics, the pre-estimation model can fit the logarithmic multimodal distribution of the promotion values and is consistent with the actual conditions of the logarithmic multimodal distribution of the promotion values, so that the accuracy of the pre-estimation of the promotion values can be improved by utilizing the pre-estimation model, and the promotion efficiency is improved.

Description

Information promotion method, information promotion device, computer equipment and storage medium

Technical Field

The present application relates to the field of computer information processing technology, and in particular, to an information promotion method, an information promotion device, a computer device, a storage medium, and a computer program product.

Background

The online information promotion refers to a mode of information promotion through an internet network. With the development of the wide-range interconnection technology, the specific gravity of online information popularization is continuously increased.

Different from the limited popularization channel of off-line information popularization, the channel and the popularization form of on-line information popularization all present diversified characteristics. Therefore, how to determine the information promotion strategy, so that the information to be promoted is promoted in a proper form, the information promotion efficiency can be improved, and the method becomes a problem to be solved urgently.

Disclosure of Invention

In view of the foregoing, it is desirable to provide an information promotion method, apparatus, computer device, computer-readable storage medium, and computer program product that can improve promotion efficiency.

In a first aspect, the present application provides an information promotion method. The method comprises the following steps:

acquiring information characteristics of information to be promoted;

acquiring multidimensional object characteristics of the target popularization receiving object of the information to be promoted;

inputting the multi-dimensional object features and the information features into a pre-trained pre-estimation model, wherein the pre-estimation model comprises a gate network and a plurality of predictors, the logarithm of the predicted popularization value is in multimodal distribution and comprises a plurality of logarithmic unimodal intervals of the popularization value, and each predictor is trained for fitting the data distribution of the logarithmic unimodal intervals of the corresponding popularization value according to the multi-dimensional object features and the information features respectively;

Processing the multi-dimensional object features and the information features through each predictor of the prediction model, predicting initial popularization values according to the extracted features by each predictor, performing feature extraction processing on the multi-dimensional object features and the information features through a gate network of the prediction model, and outputting the probability of each predictor according to the extracted features by the gate network;

determining the predicted popularization value of the target popularization receiving object according to the probability of each predictor and the initial popularization value;

and determining the popularization strategy of the information to be promoted according to the predicted popularization value of the target popularization receiving object, and promoting the information to be promoted according to the popularization strategy.

In a second aspect, the application further provides an information promotion device. The device comprises:

the feature acquisition module is used for acquiring information features of information to be promoted and acquiring multidimensional object features of target promotion receiving objects of the information to be promoted;

the input module is used for inputting the multi-dimensional object characteristics and the information characteristics into a pre-trained pre-estimation model, the pre-estimation model comprises a gate network and a plurality of predictors, the logarithm of the predicted popularization value is in multimodal distribution and comprises a plurality of logarithm unimodal intervals of the popularization value, and each predictor is trained for fitting the data distribution of the logarithm unimodal interval corresponding to the popularization value according to the multi-dimensional object characteristics and the information characteristics;

The prediction module is used for extracting and processing the characteristics of the multi-dimensional object and the information through each predictor of the prediction model, predicting initial popularization value according to the extracted characteristics by each predictor, processing the characteristics of the multi-dimensional object and the information through a gate network of the prediction model, and outputting the probability of each predictor according to the extracted characteristics by the gate network;

the fusion module is used for determining the predicted popularization value of the target popularization receiving object according to the probability of each predictor and the initial popularization value;

and the promotion module is used for determining the promotion strategy of the information to be promoted according to the predicted promotion value of the target promotion receiving object and promoting the information to be promoted according to the promotion strategy.

In a third aspect, the present application also provides a computer device. The computer device comprises a memory storing a computer program and a processor which when executing the computer program performs the steps of:

acquiring information characteristics of information to be promoted;

In a fourth aspect, the present application also provides a computer-readable storage medium. The computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:

acquiring information characteristics of information to be promoted;

In a fifth aspect, the present application also provides a computer program product. The computer program product comprises a computer program which, when executed by a processor, implements the steps of:

acquiring information characteristics of information to be promoted;

According to the information promotion method, the information promotion device, the computer equipment, the storage medium and the computer program product, the information characteristics of information to be promoted and the multidimensional object characteristics of a target promotion receiving object are obtained, the information characteristics and the multidimensional object characteristics are processed by utilizing the pre-trained pre-estimation model, the initial promotion value is respectively predicted by each predictor of the pre-estimation model, the probability of each predictor can be output according to the multidimensional object characteristics and the information characteristics by a gate network, as the predictors are trained for fitting the data distribution of the logarithmic unimodal interval of the promotion value according to the multidimensional object characteristics and the information characteristics, for the promotion value of the logarithmic peak distribution characteristics, the pre-estimation model can fit the logarithmic multimodal distribution of the promotion value and is consistent with the actual situation of the logarithmic multimodal distribution of the promotion value, therefore, the accuracy of the pre-estimation of the promotion value can be improved by utilizing the pre-estimation model, the promotion strategy of the information to be promoted is determined by utilizing the pre-estimation promotion value, and the promotion efficiency can be improved according to the promotion strategy.

Drawings

FIG. 1 is an application environment diagram of an information promotion method in one embodiment;

FIG. 2 is a flow diagram of a method of information promotion in one embodiment;

FIG. 3 is a schematic diagram of a pre-estimated model in one embodiment;

FIG. 4 is a flow chart of a training process of a predictive model in one embodiment;

FIG. 5 is a schematic diagram of a pre-estimated model according to another embodiment;

FIG. 6 is a diagram of an application interface of a method of information promotion in one embodiment;

FIG. 7 is a block diagram of an information dissemination device in one embodiment;

fig. 8 is an internal structural diagram of a computer device in one embodiment.

Detailed Description

The present application will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.

Artificial intelligence (Artificial Intelligence, AI) is the 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 the 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. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and other directions.

Machine Learning (ML) is a multi-domain interdisciplinary, involving multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, etc. It is specially studied how a computer simulates or implements learning behavior of a human to acquire new knowledge or skills, and reorganizes existing knowledge structures to continuously improve own performance. Machine learning is the core of artificial intelligence, a fundamental approach to letting computers have intelligence, which is applied throughout various areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, confidence networks, reinforcement learning, transfer learning, induction learning, teaching learning, and the like.

With research and advancement of artificial intelligence technology, research and application of artificial intelligence technology is being developed in various fields, such as common smart home, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned, automatic driving, unmanned aerial vehicles, robots, smart medical treatment, smart customer service, etc., and it is believed that with the development of technology, artificial intelligence technology will be applied in more fields and with increasing importance value.

The scheme provided by the embodiment of the application relates to the technology of artificial intelligence such as machine learning, and the like, and is specifically described by the following embodiments:

the information promotion method provided by the embodiment of the application can be applied to an application environment shown in fig. 1. Wherein the terminal 102 communicates with the server 104 via a network, and the information facilitator terminal 106 communicates with the server via the network. The data storage system may store data that the server 104 needs to process. The data storage system may be integrated on the server 104 or may be located on the cloud or other servers. The information popularization party sets popularization parameters through the information popularization party terminal 106, and the server 104 acquires information characteristics of information to be popularized; acquiring multidimensional object characteristics of a target popularization receiving object of information to be promoted; inputting the multi-dimensional object features and the information features into a pre-trained pre-estimation model, wherein the pre-estimation model comprises a gate network and a plurality of predictors, the logarithm of the predicted popularization value is in multimodal distribution, the pre-estimation model comprises a plurality of logarithm unimodal intervals of the popularization value, and each predictor is trained for fitting the data distribution of the logarithm unimodal interval corresponding to the popularization value according to the multi-dimensional object features and the information features; processing the multidimensional object features and the information features through each predictor of the prediction model, predicting initial popularization value according to the extracted features, extracting the multidimensional object features and the information features through a gate network of the prediction model, and outputting the probability of each predictor according to the extracted features by the gate network; according to the probability of each predictor and the initial popularization value, determining the predicted popularization value of the target popularization receiving object; and determining a promotion policy of the information to be promoted according to the predicted promotion value of the target promotion receiving object, and promoting the information to be promoted according to the promotion policy. The terminal 102 may be, but not limited to, various desktop computers, notebook computers, smart phones, tablet computers, internet of things devices, and portable wearable devices, where the internet of things devices may be smart speakers, smart televisions, smart air conditioners, smart vehicle devices, and the like. The portable wearable device may be a smart watch, smart bracelet, headset, or the like. The server 104 may be implemented as a stand-alone server or as a server cluster of multiple servers.

In one embodiment, as shown in fig. 2, an information promotion method is provided, and the method is applied to the server in fig. 1 for illustration, and includes the following steps:

step 202, obtaining information characteristics of information to be promoted.

The information to be promoted is content which is produced by an information promoting party for promotion and comprises promotion products, and the promotion information can comprise data in different forms, including but not limited to texts, images, videos, audios, links and the like. The promotional information can also be a combination of different forms of data, such as a combination including text, video, and audio. Wherein, the promotion purpose can be advertising, marketing or propaganda. Taking promotion purpose as an advertisement as an example, the information to be promoted can be the advertisement, and the information promoting party can be the advertiser.

The information features are content features of the information to be promoted, wherein the information features can be from multiple dimensions including product attributes, product audience attributes and the like. The product attribute may be a product name, a commodity category to which the product belongs, and the like. The product subject attribute may be a target user group of the product, or the like.

Step 204, obtaining the multidimensional object characteristics of the target popularization receiving object of the information to be promoted.

The target promotion reception object is a receiver of promotion information. The target promotion receiving object is related to a promotion channel, and is usually an application party of the channel. Taking a popularization channel as an application example, a target popularization receiving object is taken as a user of the target application, and information popularization is realized by receiving popularization information in the process that the target popularization receiving object uses the target application in a mode of throwing popularization information in the target application. The target application may be a social application, a news application, a video application, a search application, or the like. The target popularization receiving object can be a user of the target application, and can also be a set of partial or all users of the target application.

In one embodiment, the method further comprises: and matching the information characteristics of the information to be promoted with the multidimensional object characteristics of the target application user, and determining the target promotion receiving object of the information to be promoted from the target application user. Through the mode of feature matching, the target popularization receiving object can be accurately positioned for the information to be promoted.

The multidimensional object feature can further comprise basic information and tag information of the target popularization receiving object. Basic information of the target popularization receiving object can be obtained from a database of the target application, and the basic information can comprise age, region and the like. And obtaining the label information of each target popularization receiving object according to the multidimensional object data of the target popularization receiving object in a label classification mode.

Step 206, inputting the multi-dimensional object features and the information features into a pre-trained pre-estimation model, wherein the pre-estimation model is used for predicting the popularization value, the pre-estimation model comprises a gate network and a plurality of predictors, the logarithm of the predicted popularization value is in multi-peak distribution, the predicted popularization value comprises a plurality of logarithm unimodal intervals of the popularization value, and each predictor is trained for fitting the data distribution of the logarithm unimodal interval corresponding to the popularization value according to the multi-dimensional object features and the information features.

The prediction model is trained in advance according to multidimensional object characteristics of the popularization and receiving objects, popularization information of object interaction and actual popularization value generated by the interaction. The actual popularization value is annotation data, and the training target of the prediction model predicts the popularization value of the target popularization receiving object by inputting the multidimensional object characteristics of the target popularization receiving object and the information characteristics of the information to be promoted. When the target promotion object is one, predicting the promotion value of the target promotion receiving object. And when the target popularization object is part or all of the user set of the target application, predicting the total popularization value of the user set.

The interaction is specifically a behavior process and a behavior result implemented by the object on the popularization information. The interactive behavior of the target popularization receiving object on the popularization information can embody the popularization value. The popularization value is a quantization index of the popularization effect and is used for evaluating the interaction effect of the target popularization receiving object and the popularization information.

Specifically, the better the interaction effect is, the higher the popularization value is, and the better the popularization effect is achieved. The interaction effect may be an interaction result. Taking the example that the interaction result comprises purchasing advertisement commodity, the amount of the order of the target popularization receiving object can be taken as the popularization value. The interactive effect can also be from the interactive process, and the longer the interactive process is, the higher the popularization value is, and the better the popularization effect is achieved. The interaction process may include a viewing duration and/or a viewing completion. The watching duration represents the duration of the target popularization receiver watching the information to be promoted. The viewing completion represents the ratio of the viewing duration of the target popularization receiver to the total duration of the information to be promoted.

By analyzing the actual business, it can be found that no matter what form of interaction effect the promotion value is, such as the following amount, the viewing duration or the viewing completion, the promotion value is usually a real value. Taking the following bill amount as an example, the popularization value can be 50 yuan. Taking the watching duration as an example, the popularization value can be 50 seconds, and taking the watching completion as an example, the popularization value can be 50%.

In order to achieve an ideal popularization effect, the popularization value can be predicted by training a pre-estimated model in advance. However, if the distribution properties of the promotion values are not distinguished, it is difficult to accurately predict the promotion values, and the promotion accuracy and promotion efficiency are reduced.

It is generally assumed that the promotional value follows a Log-Normal (LN) distribution, and a predictive model is trained based thereon. However, in actual business application, for users of the application platform, the value of the commodity promoted by the information to be promoted is different, and the content information amount of the information to be promoted is different (such as different video duration), so that the promotion value may be distributed differently on different information to be promoted. When the actual service is analyzed and the logarithmic data of the actual popularization value is distributed in multiple modes (also called multiple modes), the model may be overestimated for low-value users and underestimated for high-value users when the model is used for predicting the popularization value, so that the popularization value is difficult to be accurately predicted. By analyzing the actual traffic, the logarithmic data distribution of the actual promotional value is a multi-modal (also referred to as multimodal) distribution. In one embodiment, the logarithmic data distribution of the promotional value includes a plurality of peaks, and the inflection points on both sides of each peak enclose a logarithmic unimodal interval of the promotional value, and the logarithmic data distribution of the promotional value includes a logarithmic unimodal interval corresponding to the plurality of peaks.

Based on the method, a pre-estimation model is trained in advance, the pre-estimation model comprises a plurality of pre-estimators, each pre-estimation model is trained to correspond to a logarithmic unimodal interval of popularization value, and data distribution of the logarithmic unimodal interval corresponding to the popularization value is fitted according to multi-dimensional object characteristics and information characteristics. Thus, each predictor can predict the initial popularization value according to the multidimensional object characteristics and the information characteristics.

And step 208, performing feature extraction processing on the multi-dimensional object features and the information features through each predictor of the prediction model, predicting initial popularization value according to the extracted features by each predictor, performing feature extraction on the multi-dimensional object features and the information features through a gate network of the prediction model, and outputting the probability of each predictor according to the extracted features by the gate network.

The architecture of the predictive model of one embodiment is shown in FIG. 3 and includes an input layer, a feature extraction layer, a plurality of predictors, a gate network, and an output layer. The information features of the popularization information and the multidimensional object features of the target popularization receiving object are converted into vector expressions, the vector expressions are used as input of a pre-estimated model, and feature extraction processing is carried out on the multidimensional object features and the information features through a feature extraction layer, so that extracted features are obtained. The feature extraction layer may be a multi-layer sensor (also called a multi-layer neural network), and features are extracted by using a multi-layer network structure of the multi-layer sensor, and it is understood that the more the number of layers of the neural network is, the finer the extracted features are.

The extracted features are input into each predictor and gate network respectively. And each predictor performs feature extraction processing on the multi-dimensional object features and the information features to acquire initial popularization values output by each predictor, wherein the predictor can adopt a dense neural network (Dense Neural Networks).

The input of the gate network is consistent with the input of the predictor, the output of the gate network is a probability sequence, and the number of probabilities in the probability sequence is consistent with the number of the predictor. The probability in the probability sequence may be used to represent the probability that the logarithm of the predicted promotional value belongs to the logarithmic unimodal interval of the corresponding promotional value, and may also represent the weight of the initial promotional value of the corresponding predictor in the predicted promotional value. Wherein z is _k Is the kth output value, q of the gate network _k (x) Representing the probability value of the kth predictor.

Compared with a mode that a single predictor is used for predicting the popularization value, the prediction model comprises a plurality of predictors, each predictor is used for fitting data distribution of a logarithmic unimodal interval corresponding to the popularization value according to multidimensional object characteristics and information characteristics, probability of each predictor is output through a gate network, accordingly, the prediction model can fit logarithmic unimodal distribution of the popularization value, the estimated model accords with practical conditions that logarithms of the popularization value are in multimodal distribution, and further accuracy of predicting the popularization value is improved.

Step 210, determining the predicted popularization value of the target popularization receiving object according to the probability of each predictor and the initial popularization value.

The method for determining the predicted popularization value of the target popularization receiving object according to the probability of the predictor and the initial popularization value can comprise two modes. One way can determine the predicted popularization value of the target popularization receiving object according to the predictor with the highest probability and the initial popularization value. The method takes the initial popularization value output by the predictor with highest probability as the predicted popularization value of the target popularization receiving object. In one mode, the predicted popularization value of the target popularization receiving object can be obtained according to the weighted sum of the probability of the predictor and the initial popularization value.

The predicted popularization value can be an interaction result of the predicted target popularization receiving object to the popularization information, or an interaction effect of the predicted target popularization receiving object to the popularization information, and the predicted popularization value can be an order amount of the predicted target popularization receiving object to the popularization information. Taking the interaction effect as an example, the predicted popularization value can be the watching duration or watching completion degree of the predicted target popularization receiving object to the popularization information.

And step 212, determining a promotion policy of the information to be promoted according to the predicted promotion value of the target promotion receiving object, and promoting the information to be promoted according to the promotion policy.

The predicted popularization value of the predicted target popularization receiving object can be an interaction effect of information to be promoted or an interaction result. Therefore, the promotion effect which can be achieved by the promotion information can be evaluated based on the predicted promotion value of the target promotion receiving object, and the promotion strategy is further determined.

The target promotion object can be determined according to the predicted promotion value, and the target promotion object is promoted at fixed points. The target promotion object may be the first N users with highest predicted promotion value for the target promotion receiving object.

The promotion effect can be determined according to the predicted promotion value, and the promotion strategy can be determined according to the promotion effect.

According to the information promotion method, the information characteristics of the information to be promoted and the multidimensional object characteristics of the target promotion receiving object are obtained, the information characteristics and the multidimensional object characteristics are processed by utilizing the pre-trained pre-estimation model, the pre-estimation models are used for respectively predicting the initial promotion value, the door network can output the probability of each pre-estimation device according to the multidimensional object characteristics and the information characteristics, as the pre-estimation devices are trained for fitting the data distribution of the logarithmic unimodal intervals of the promotion values according to the multidimensional object characteristics and the information characteristics, for the promotion values of the logarithmic peak distribution characteristics, the pre-estimation models can fit the logarithmic multimodal distribution of the promotion values and are consistent with the actual situation of the logarithmic multimodal distribution of the promotion values, therefore, the accuracy of the promotion value pre-estimation can be improved by utilizing the pre-estimation models, the promotion strategy of the information to be promoted is determined by utilizing the predictive promotion values, and the promotion efficiency can be improved according to the promotion strategy.

In another embodiment, determining the predicted promotional value of the target promotional receiving object according to the probability of each predictor and the initial promotional value comprises: and determining the predicted popularization value of the target popularization receiving object according to the initial popularization value corresponding to the predictor with the highest probability, wherein the probability of the predictor is used for representing the probability that the logarithm of the predicted popularization value belongs to the corresponding logarithm unimodal interval.

Specifically, the probability of the predictor is used for representing the probability that the logarithm of the predicted popularization value belongs to the corresponding logarithm unimodal interval, then the probability that the logarithm of the predicted popularization value belongs to the corresponding logarithm unimodal interval is the maximum value in the probability sequence output by the gate network, the probability that the logarithm of the predicted popularization value belongs to the logarithm unimodal interval is the maximum value, and the initial popularization value of the corresponding predictor is determined as the predicted popularization value of the target popularization receiving object.

According to the method, the estimated popularization value of the target popularization receiving object is determined by the probability that the logarithm of the estimated popularization value belongs to the corresponding logarithm unimodal interval and output by each gate network, wherein the initial popularization value corresponds to the predictor with the highest probability, the estimated popularization value of the data distribution of the logarithm unimodal interval of the estimated popularization value can be estimated according to the predictor to which the logarithm of the estimated popularization value belongs, the characteristic that the logarithm of the estimated popularization value is in multimodal distribution is reflected, and the prediction precision is improved.

In another embodiment, the probabilities are used to represent weights of the initial promotional value of the corresponding predictor in predicting promotional value. The probability of the predictor is used for representing the probability that the predicted popularization value belongs to the corresponding logarithmic unimodal interval. That is, the greater the probability, the greater the weight of the predictor's initial promotional value in predicting promotional value.

According to the probability of each predictor and the initial popularization value, determining the predicted popularization value of the target popularization receiving object comprises the following steps: and determining the weighted sum of the initial popularization value of each predictor according to the weight of each predictor, and obtaining the predicted popularization value of the target popularization receiving object.

Specifically, the initial popularization value of each predictor is fused through a gate network, the gate network outputs the weight of each predictor, and accordingly the predicted popularization value of the target popularization receiving object is obtained, and the specific formula is as follows:

wherein p represents a predicted popularization value, q _k (x) Representing the probability value of the kth predictor,μ _k representing the mean value, sigma, of the kth predictor predictions _k Represents the standard deviation of the kth predictor prediction, N (log (y|μ) _k ,σ _k ) And (5) representing the initial popularization value obtained according to the average value and standard deviation predicted by the kth predictor.

According to the method, the weight of the initial promotion value of the corresponding predictor in the predicted promotion value is output through each gate network, the weighted sum of the initial promotion values of the predictors is determined according to the weight of each predictor, the predicted promotion value of the target promotion receiving object is obtained, the initial promotion value of the predictor can be weighted according to the probability of the corresponding promotion value log-unimodal interval of the predictor to which the probability of the predicted promotion value belongs, the characteristic that the logarithm of the predicted promotion value is in multimodal distribution is reflected, and the prediction precision is improved.

In another embodiment, the method further comprises: and acquiring the popularization parameters of the information to be promoted. Determining a promotion policy of the information to be promoted according to the predicted promotion value of the target promotion receiving object, promoting the information to be promoted according to the promotion policy, and comprising: determining estimated popularization effects of information to be popularized according to the estimated popularization value and the popularization parameters; and determining a promotion policy of the information to be promoted according to the estimated promotion effect, and promoting the information to be promoted according to the promotion policy.

In the promotion service, to characterize the promotion effect, some quantization parameters may be used, for example, the estimated promotion effect may be ECPM. ECPM is a desired benefit of thousands of presentations, with higher ECPM indicating better promotional effect.

In the field of ROI advertisement popularization, the popularization value can be the amount of the order, and the predicted popularization value is the predicted possible amount of the order of each target popularization receiving user. And obtaining the estimated delivery value of the information to be promoted according to the possible amount of the target promotion receiving user.

In the ROI advertising promotion, the information promoting party configures promotion parameters including a target ROI (investment return ratio), and according to the ratio of the predicted promotion value and the target ROI, a current BID price (BID) can be obtained, and then the BID price calculates ECPM.

Wherein the promotion parameter may be a target return on investment Ratio (ROI). The target investment return ratio may be preset by the promoting party.

Wherein, confirm the estimated promotion effect of waiting to promote the information according to predicting promotion value and promotion parameter, include: according to the information characteristics of the information to be promoted and the information characteristics of the information to be promoted, the predicted click rate and the predicted conversion rate of the information to be promoted are determined, the target input of the information to be promoted is determined according to the predicted promotion value and the target input return ratio, and the estimated promotion effect of the information to be promoted is determined according to the target input, the predicted click rate and the predicted conversion rate.

The estimated click rate and the estimated conversion rate may be predicted based on big data learning. For example, learning is performed according to popularization information and multidimensional object characteristics, and click rate and conversion rate are predicted.

The target return on investment Ratio (ROI) may be a promotion parameter set by the information promoter. At the time of advertisement bidding, a bid (bid) of the current bid is calculated based on the predicted popularization value and the target investment-return ratio:

wherein pLTV is the predicted popularization value, targetROI is the target delivery return ratio.

And determining the estimated popularization effect of the information to be promoted according to the target investment, the predicted click rate and the predicted conversion rate. Taking the popularization efficiency as ECPM as an example, the ECPM is mainly related to the bid willingness of an advertiser and the advertisement quality (namely the possibility of clicking and converting behaviors generated by a user) in statistical theory, and the calculation formula is as follows

Ecpm=bid (Pbid) ×estimated click rate (pCTR) ×estimated conversion rate (pCVR) ×1000

By using ECPM as a quantized representation of the estimated promotional effect, the estimated promotional effect can be quantized. And determining the popularization strategy of the information to be promoted according to the ECPM of the information to be promoted, for example, the higher the ECPM is, the better advertisement position can be obtained. In this embodiment, by determining the estimated promotion effect of the promotion information according to the estimated promotion value and the promotion parameter, the promotion policy is determined according to the promotion effect, and promotion efficiency can be improved.

In specific popularization and application, determining a popularization strategy of the information to be popularized according to the estimated popularization effect, and popularizing the information to be popularized according to the popularization strategy can be as follows: determining a target promotion position of information to be promoted according to the estimated promotion effect; and promoting the information to be promoted through the target promotion position.

In this embodiment, the quantized estimated promotion effect is utilized, and the promotion information with good promotion effect is preferentially placed on the high-quality promotion position, so that the promotion information has a better promotion effect. This approach may be applied to a competition scenario of promotion sites. Taking estimated popularization effect as ECPM as an example, from the perspective of advertisers, ECPM is estimated cost of thousands of times of display. The higher the ECPM, the more competitive the advertisement and the greater the traffic value. In the business scenario of 'promotion position competition', the advertising system decides who can obtain better promotion position according to the height of ECPM, and promotion efficiency can be improved.

In another embodiment, determining a promotion policy of the information to be promoted according to the estimated promotion effect, and promoting the information to be promoted according to the promotion policy may be: determining target popularization information of a target popularization position from a plurality of pieces of information to be promoted according to the estimated popularization effect; and promoting the target promotion information through the target promotion position.

In this embodiment, by using quantized estimated promotion efficiency, when determining promotion information for displaying a target promotion position, the target promotion information of the target promotion position may be determined from a plurality of pieces of information to be promoted according to the estimated promotion effect, and the target promotion information may be promoted through the target promotion position. The method can be applied to a business scenario with only one promotion site competition. For example, in an auction with only one promotion location, the promotion information with the best promotion effect is placed at the promotion location, thereby improving the promotion efficiency of the promotion location.

In another embodiment, the training method of the pre-estimation model is shown in fig. 4, and includes the following steps:

step 402, acquiring a training data set, wherein the training data set comprises multidimensional object characteristics of an object, information characteristics of promotion information and actual promotion value generated by interaction between the object and the promotion information.

Specifically, the training data set may be related data of an object of the application platform, including multidimensional object features of the object and information features of promotion information, and takes actual promotion values generated by interaction between the object and the promotion information as data labels. Taking the popularization value as a long-term value of a user as an example, the training data set comprises multidimensional object characteristics of the object and information characteristics of popularization information, and takes the actual long-term value (amount of the order) of the user generated by interaction of the object and the popularization information as data annotation.

Step 404, inputting the multidimensional object features and the information features into each predictor of the prediction model to be trained and a gate network for iterative training, wherein each predictor processes the multidimensional object features and the information features, predicts standard deviation and average value of logarithms of predicted popularization values of logarithmic unimodal intervals of popularization values corresponding to the predictors, and the gate network processes the multidimensional object features and the information features and outputs probability of each predictor.

Wherein each training is an iteration. In one embodiment, the network structure of the predictive model is shown in fig. 5, and includes an input layer, a feature extraction layer, a plurality of predictors, a gate network, and an output layer. The information features of the popularization information and the multidimensional object features of the target popularization receiving object are converted into vector expressions, the vector expressions are used as input of a pre-estimated model, and the multidimensional object features and the information features are processed through a feature extraction layer to obtain extracted features. The feature extraction layer may be a multi-layer sensor (also called a multi-layer neural network), and features are extracted by using a multi-layer network structure of the multi-layer sensor, and it is understood that the more the number of layers of the neural network is, the finer the extracted features are.

The extracted features are input into each predictor and gate network respectively. And processing the multidimensional object characteristics and the information characteristics by each predictor to obtain the initial popularization value output by each predictor, wherein the predictor can adopt a dense neural network (Dense Neural Networks). The underlying structures used by different predictors are shared and can be split independently, and each predictor has an independent underlying structure.

The input of the gate network is consistent with the input of the predictor, the output of the gate network is a probability sequence, and the number of probabilities in the probability sequence is consistent with the number of the predictor. The probability in the probability sequence may be used to represent the probability that the predicted promotional value belongs to a logarithmic unimodal interval of the corresponding promotional value, and may also represent the weight of the initial promotional value of the corresponding predictor in the predicted promotional value.

Each predictor processes the multidimensional object characteristics and the information characteristics and predicts the popularization corresponding to the predictorThe standard deviation and the average value of the logarithm of the predicted popularization value of the logarithm single-peak interval of the value can be obtained according to the standard deviation and the average value of the logarithm of the predicted popularization value, which isμ represents the mean of the logarithm of the predicted promotional value, σ represents the standard deviation of the logarithm of the predicted promotional value.

And step 406, calculating a loss value according to the probability of each predictor, the standard deviation and the mean value of the logarithm of the predicted popularization value during each iterative training, and adjusting the parameters of the prediction model according to the loss value.

Specifically, the estimation of the promotional value is a regression problem. A commonly used loss function for regression models is the mean square error loss function (Mean Square Error). And the popularization value may have different distributions on different commodities. Taking the popularization value as the order amount as an example, the order amount of advertisements in the order scene of the electronic commerce is different from 10 yuan to thousands of yuan, the range is larger, and the difference between different industries and commodities is larger. The MSE loss function is simply used, so that the model has larger punishment to samples with large popularization value, and has small punishment to samples with small popularization value, so that the model has large estimated deviation. Although log-normal distributions can alleviate such problems to some extent, model predictive bias remains large when there is a multimodal distribution of training data.

In this embodiment, considering that there may be different distributions of popularization values on different commodities, a new structure and a loss function for estimating multimodal distribution based on a plurality of predictors are provided, so that a model can better fit training data distribution.

And a loss function of the mixed log-ethernet distribution is set. The loss function uses maximum likelihood estimation, targeting minimizing loss, the loss function is as follows:

wherein N represents the number of samples, Z _i Representing the class of the predictor corresponding to the sample i, wherein the class comprises [1, K ]],y _i The label of sample i, i.e. the actual promotional value,representation predictor Z _i The predicted samples i of (1) predict the mean of the logarithm of the promotional value,/->Representation predictor Z _i The standard deviation of the logarithm of the predicted popularization value of the predicted sample i; p (k) represents the probability of the kth predictor;

represents p (Z) _i |x _i ) For sample i to belong to predictor class Z _i Is a posterior probability of (c).

The loss function adopts maximum likelihood estimation, aims at minimizing loss, and counter propagates the estimated model according to the loss value, and adjusts parameters of the estimated model.

And step 408, obtaining a trained pre-estimated model when the model iteration training stopping condition is met.

In the embodiment, the logarithmic multimodal distribution of popularization value is modeled by adjusting the structure and the loss function of the predictor, so that the model is better fitted with training data distribution, and the prediction accuracy is improved.

In order to improve the training efficiency of the model, the number of predictors can be adjusted in advance according to practical application. For example, heuristic clustering may be performed on the data labels to obtain a number of predictors with stronger suitability.

Specifically, the logarithm of the actual popularization value can be clustered, and the data distribution peak value of the logarithm of the actual popularization value is determined according to the clustering result; and determining the number of predictors of the prediction model according to the number of the data distribution peaks of the logarithm of the actual popularization value.

In the embodiment, the predictor pre-determines the data of the evaluator in advance according to the data distribution condition of the sample label, so that training time can be saved, and model training efficiency can be improved.

In practical business applications, taking a Long Time Value (LTV) as an example, the Long Time Value of a user is related to a benefit generated after the user receives promotion information, for example, an order amount based on the promotion information. If the user has shopping behaviors based on the popularization information, the user has high long-term value, and the popularization value of the user is high.

The ROI bid is taken as a bid product of an e-commerce advertisement, and aims to estimate the user's or long-term value for different users and advertisements according to the ROI set by an advertiser, and takes the ROI as a bid basis of the advertisement bid. Wherein the long-term value of the user can be represented by the payment behavior of the user and the placing of the bill. Here, the following list amount represents a long-term value of the user as an example.

The existing Long Time Value (LTV) prediction algorithm is mostly based on a neural network, and generally performs sparse representation mapping on object features and advertisement features, then inputs the object features and the advertisement features into the neural network, and outputs a real Value to represent the amount of money that can be rendered on the user after the advertisement is converted. The existing technical scheme generally assumes that payment amount distribution is subject to Log-Normal (LN) distribution, but in different commodity promotion scenarios, due to the nature of its unimodal distribution, the model may be overestimated for low value users and underestimated for high value users.

The existing long-term value estimation algorithm of the user is based on a neural network model, the input of the neural network is some object characteristics and advertisement characteristics, and the output value represents the expected amount which the user can realize. Where existing models generally assume that the user value follows a logarithmic positive-going distribution, such models are subject to large deviations in prediction when the true data distribution is a multi-modal (also referred to as multimodal) distribution. How to use modeling multimode distribution is the focus of the study of the present application.

The algorithm core of the technical scheme is to provide a new structure and a loss function for estimating multimodal distribution based on a plurality of predictors, so that a model can be better fitted with training data distribution in consideration of the fact that different distributions of user values possibly exist on different commodities.

Specifically, the structure of the pre-estimated model is shown in fig. 5, the model takes object characteristics and advertisement characteristics as input, and the long-term value predicted value pLTV of the user is finally given by the output layer through a plurality of full-connection layers. The application constructs a plurality of predictors. Meanwhile, in order to obtain the weight of each predictor in final prediction, a gate network is introduced. The input of the gate network is shared with the predictors, and the output of the gate network is K (consistent with the number of the predictors), and each output value represents the weight of one predictor in the final predicted value. The output value of the gate network is converted into probability distribution through a softmax layer, and finally the prediction results of a plurality of predictors are fused through the gate network.

In the online service, the new model structure simultaneously starts a plurality of predictors to predict the user value, and finally weights of the predictors are weighted and fused through a gate network to serve as a model final user long-term value predicted value. In advertisement bidding, the bid for the current bid (bid) will be calculated online based on the user's long-term value pLTV and the target ROI (targetROI) of the advertisement:

the application method of the method is that in the service scene of the ROI advertisement, as shown in fig. 6, in the advertisement putting stage, when an advertiser puts the ROI advertisement, the advertiser needs to select depth conversion optimization on a corresponding interface, set a target ROI, and call a pre-trained estimated model to implement the popularization method of the application.

The final offline experiment shows that the model offline effect is improved by 1-3% under the condition of no information loss and no error. After the experiment was brought online, a 10% + consumption boost was achieved on the e-commerce ROI advertisement through a stringent AB test.

The popularization effect obtained based on the popularization model is shown in the following table, and the popularization scene of the electronic commerce is mainly divided into two types: commodity popularization and merchant direct casting are separately modeled in the model, so the evaluation is also separately performed. The method and the device have the advantages that the model is compared with the existing model from the two angles of the model estimated deviation and the ordering stability, and meanwhile, the technical scheme has landed on WeChat advertisements, and good benefits are obtained through online AB tests. The estimated deviation of the evaluation model is calculated through the RMSE index and the GINI coefficient, the smaller the RMSE index is, the smaller the estimated deviation of the model is, the larger the GINI coefficient is, the better the order-preserving performance of the estimated result of the model is, and the more stable the online result is.

The RMSE and MAPE evaluation models estimate the deviation with smaller values. The larger the value of the order-preserving property estimated by the GINI coefficient estimation model is, the better the value is. By comparing the tables, the Multi-Modal Log-Normal loss function is better than the Log-Normal in a plurality of indexes in two application scenes. And because the underlying structures of the two models are the same, the benefits of offline indicators can be considered to be brought by the upgrade of the models in the scheme.

In addition, the new model was tested for online flow and subjected to rigorous AB testing. Online brings a significant improvement of over 10% to the revenue and GMV of the e-commerce ROI advertisement.

In conclusion, the algorithm based on the multi-mode regression not only ensures that the performance of the algorithm is not deteriorated, but also can improve the model effect. And the system has flexible pluggable performance and can be quickly migrated to other similar regression tasks.

It should be understood that, although the steps in the flowcharts related to the embodiments described above are sequentially shown as indicated by arrows, these steps are not necessarily sequentially performed in the order indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in the flowcharts described in the above embodiments may include a plurality of steps or a plurality of stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of the steps or stages is not necessarily performed sequentially, but may be performed alternately or alternately with at least some of the other steps or stages.

Based on the same inventive concept, the embodiment of the application also provides an information promotion device for realizing the information promotion method. The implementation scheme of the solution to the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitation in one or more embodiments of the information promotion device provided below may refer to the limitation of the information promotion method hereinabove, and will not be repeated here.

In one embodiment, as shown in fig. 7, there is provided an information promotion apparatus including:

the feature acquisition module 702 is configured to acquire information features of information to be promoted and obtain multidimensional object features of a target promotion receiving object of the information to be promoted.

The input module 704 is configured to input the multidimensional object feature and the information feature into a pre-trained prediction model, where the prediction model includes a gate network and a plurality of predictors, and the logarithm of the predicted popularization value is in a multimodal distribution, and includes a plurality of logarithm unimodal intervals of the popularization value, and each predictor is trained to fit the data distribution of the logarithm unimodal interval of the corresponding popularization value according to the multidimensional object feature and the information feature, respectively.

The prediction module 706 is configured to process the multidimensional object feature and the information feature through each predictor of the prediction model, predict an initial popularization value according to the extracted feature, and perform feature extraction on the multidimensional object feature and the information feature through a gate network of the prediction model, and output a probability of each predictor according to the extracted feature.

And the fusion module 708 is configured to determine a predicted popularization value of the target popularization receiving object according to the probability of each predictor and the initial popularization value.

And the promotion module 710 is configured to determine a promotion policy of the information to be promoted according to the predicted promotion value of the target promotion receiving object, and promote the information to be promoted according to the promotion policy.

In one embodiment, the fusion module is configured to determine a predicted popularization value of the target popularization receiving object according to an initial popularization value corresponding to a predictor with a highest probability, where the probability of the predictor is used to represent a probability that a logarithm of the predicted popularization value belongs to a corresponding logarithm unimodal interval.

In one embodiment, the probability is used to represent the weight of the initial promotional value of the corresponding predictor in the predicted promotional value; and the fusion module is used for determining the weighted sum of the initial popularization values of the predictors according to the weights of the predictors to obtain the predicted popularization value of the target popularization receiving object.

In another embodiment, the feature acquisition module is further configured to acquire a promotion parameter of the information to be promoted, and the promotion module is configured to determine an estimated promotion effect of the information to be promoted according to the predicted promotion value and the promotion parameter; and determining a promotion policy of the information to be promoted according to the estimated promotion effect, and promoting the information to be promoted according to the promotion policy.

The promotion module is also used for determining a target promotion position of the information to be promoted according to the estimated promotion effect; and promoting the information to be promoted through the target promotion position.

The promotion module is further used for determining target promotion information of target promotion positions from the plurality of information to be promoted according to the estimated promotion effect; and promoting the target promotion information through the target promotion position.

In another embodiment, the information promotion device further includes:

the data set acquisition module is used for acquiring a training data set, wherein the training data set comprises multidimensional object characteristics of objects, information characteristics of popularization information and actual popularization value generated by interaction between the training data set and the popularization information.

The training module is used for inputting the multi-dimensional object characteristics and the information characteristics into each predictor of the prediction model to be trained and the gate network to carry out iterative training, each predictor processes the multi-dimensional object characteristics and the information characteristics, and predicts standard deviation and mean value of logarithm of predicted popularization value of the corresponding logarithmic single-peak interval of the predictor; the gate network processes the multidimensional object features and the information features and outputs the probability of each predictor.

And the adjusting module is used for adjusting parameters of the pre-estimated model according to the probability of the pre-estimator, the standard deviation and the mean value of the logarithm of the predicted popularization value and the loss value during each iterative training.

And the control module is used for obtaining a trained estimated model when the model iteration training stopping condition is met.

In another embodiment, the information promotion device further includes: the predictor processing module is used for clustering the logarithm of the actual popularization value and determining a data distribution peak value of the logarithm of the actual popularization value according to a clustering result; and determining the number of predictors of the prediction model according to the number of data distribution peaks of the actual popularization value.

The modules in the information promotion device can be realized in whole or in part by software, hardware and a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.

In one embodiment, a computer device is provided, which may be a server, and the internal structure of which may be as shown in fig. 8. The computer device includes a processor, a memory, an Input/Output interface (I/O) and a communication interface. The processor, the memory and the input/output interface are connected through a system bus, and the communication interface is connected to the system bus through the input/output interface. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer device is for storing training data. The input/output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement an information promotion method.

It will be appreciated by those skilled in the art that the structure shown in FIG. 8 is merely a block diagram of some of the structures associated with the present inventive arrangements and is not limiting of the computer device to which the present inventive arrangements may be applied, and that a particular computer device may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.

In one embodiment, a computer device is provided, including a memory and a processor, where the memory stores a computer program, and the processor implements the information promotion method of each of the above embodiments when executing the computer program.

In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which when executed by a processor implements the information promotion method of the above embodiments.

In one embodiment, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the information promotion method of the above embodiments.

It should be noted that, the user information (including but not limited to user equipment information, user personal information, etc.) and the data (including but not limited to data for analysis, stored data, presented data, etc.) related to the present application are information and data authorized by the user or sufficiently authorized by each party, and the collection, use and processing of the related data need to comply with the related laws and regulations and standards of the related country and region.

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, database, or other medium used in embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high density embedded nonvolatile Memory, resistive random access Memory (ReRAM), magnetic random access Memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric Memory (Ferroelectric Random Access Memory, FRAM), phase change Memory (Phase Change Memory, PCM), graphene Memory, and the like. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, and the like. By way of illustration, and not limitation, RAM can be in the form of a variety of forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), and the like. The databases referred to in the embodiments provided herein may include at least one of a relational database and a non-relational database. The non-relational database may include, but is not limited to, a blockchain-based distributed database, and the like. The processor referred to in the embodiments provided in the present application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, or the like, but is not limited thereto.

The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.

The foregoing examples illustrate only a few embodiments of the application and are described in detail herein without thereby limiting the scope of the application. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the application, which are all within the scope of the application. Accordingly, the scope of the application should be assessed as that of the appended claims.

Claims

1. An information promotion method, which is characterized by comprising the following steps:

acquiring information characteristics of information to be promoted;

inputting the multi-dimensional object features and the information features into a pre-trained pre-estimation model, wherein the pre-estimation model is used for predicting popularization values, the pre-estimation model comprises a gate network and a plurality of predictors, the logarithm of the predicted popularization values is in multimodal distribution and comprises a plurality of logarithm unimodal intervals of the popularization values, and each predictor is trained for fitting the data distribution of the logarithm unimodal intervals of the corresponding popularization values according to the multi-dimensional object features and the information features;

Performing feature extraction processing on the multi-dimensional object features and the information features through each predictor of the prediction model, predicting initial popularization values according to the extracted features by each predictor, performing feature extraction on the multi-dimensional object features and the information features through a gate network of the prediction model, and outputting the probability of each predictor according to the extracted features by the gate network;

2. The method of claim 1, wherein determining the predicted promotional value of the target promotional receiving object based on the probability of each predictor and the initial promotional value comprises:

and determining the predicted popularization value of the target popularization receiving object according to the initial popularization value corresponding to the predictor with the highest probability, wherein the probability of the predictor is used for representing the probability that the logarithm of the predicted popularization value belongs to the corresponding logarithm unimodal interval.

3. The method of claim 1, wherein the probability is used to represent a weight of the initial promotional value of a corresponding predictor in the predicted promotional value;

the determining the predicted popularization value of the target popularization receiving object according to the probability of each predictor and the initial popularization value comprises the following steps:

and determining the weighted sum of the initial popularization value of each predictor according to the weight of each predictor, and obtaining the predicted popularization value of the target popularization receiving object.

4. The method according to claim 1, wherein the method further comprises:

acquiring popularization parameters of the information to be promoted;

the step of determining the promotion policy of the information to be promoted according to the predicted promotion value of the target promotion receiving object, and promoting the information to be promoted according to the promotion policy, includes:

determining the estimated promotion effect of the information to be promoted according to the estimated promotion value and the promotion parameters;

and determining the promotion strategy of the information to be promoted according to the estimated promotion effect, and promoting the information to be promoted according to the promotion strategy.

5. The method of claim 4, wherein the determining the promotion policy of the information to be promoted according to the estimated promotion effect, and promoting the information to be promoted according to the promotion policy, comprises at least one of:

first kind:

determining a target promotion position of the information to be promoted according to the estimated promotion effect;

the information to be promoted is promoted through the target promotion position;

second kind:

determining target popularization information of a target popularization position from a plurality of pieces of information to be promoted according to the estimated popularization effect;

and promoting the target promotion information through the target promotion position.

6. The method according to any one of claims 1 to 5, wherein the training method of the predictive model comprises:

acquiring a training data set, wherein the training data set comprises multidimensional object characteristics of an object, information characteristics of popularization information and actual popularization value generated by interaction between the object and the popularization information;

inputting the multi-dimensional object features and the information features into each predictor of a prediction model to be trained and a gate network for iterative training, processing the multi-dimensional object features and the information features by each predictor, and predicting standard deviation and mean value of the logarithm of predicted popularization value of a logarithmic unimodal interval of the popularization value corresponding to the predictor; the gate network processes the multi-dimensional object features and the information features and outputs the probability of each predictor;

Calculating a loss value according to the probability of each predictor and the standard deviation and the mean value of the logarithm of the predicted popularization value in each iterative training, and adjusting the parameters of the prediction model according to the loss value;

and when the model iteration training stopping condition is met, obtaining a trained estimated model.

7. The method of claim 6, wherein the method further comprises:

clustering the logarithm of the actual popularization value, and determining a data distribution peak value of the logarithm of the actual popularization value according to a clustering result;

and determining the number of predictors of the prediction model according to the number of the data distribution peaks of the logarithm of the actual popularization value.

8. An information promotion device, characterized in that the device comprises:

9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 7 when the computer program is executed.

10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 7.