WO2021233015A1 - 一种信息处理方法、装置及计算机可读存储介质 - Google Patents

一种信息处理方法、装置及计算机可读存储介质 Download PDF

Info

Publication number
WO2021233015A1
WO2021233015A1 PCT/CN2021/086778 CN2021086778W WO2021233015A1 WO 2021233015 A1 WO2021233015 A1 WO 2021233015A1 CN 2021086778 W CN2021086778 W CN 2021086778W WO 2021233015 A1 WO2021233015 A1 WO 2021233015A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
target
rate
push
data
Prior art date
Application number
PCT/CN2021/086778
Other languages
English (en)
French (fr)
Inventor
严超
Original Assignee
腾讯科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 腾讯科技（深圳）有限公司 filed Critical 腾讯科技（深圳）有限公司
Publication of WO2021233015A1 publication Critical patent/WO2021233015A1/zh
Priority to US17/727,168 priority Critical patent/US20220245495A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/31—Indexing; Data structures therefor; Storage structures

Definitions

  • This application relates to the field of information processing technology, in particular to an information processing method, device, and computer-readable storage medium.
  • the push owner who pays for the push information can pay the publisher (push platform) to place his push information through web pages, search engines, browsers, or other online media, so as to promote their products well.
  • the push master can create multiple dynamic push messages under one push message to form different ideas, and simultaneously push multiple dynamic push messages online randomly, according to each push message.
  • the performance of the dynamic push information on the client select the dynamic push information with the best performance, and realize the creative optimization.
  • the embodiments of the present application provide an information processing method, device, and computer-readable storage medium, which can improve the accuracy of information processing.
  • the embodiment of the application provides an information processing method, including:
  • An information processing device includes:
  • a statistics unit used to count feedback data of historical push information, where the feedback data includes at least exposure data and click data;
  • a generating unit configured to generate a first probability distribution corresponding to the click rate of each push information in the historical push information based on the exposure data and click data;
  • the first coarse sorting unit is configured to determine the first predicted click rate of each push information to be pushed according to the first probability distribution, and select a preset from the push information to be pushed according to the first predicted click rate Number of first push messages;
  • the fine ranking unit is configured to obtain the target predicted click-through rate of each of the first push information, and select the target dynamic push information from the first push information for pushing according to the target predicted click-through rate.
  • a computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for loading by a processor to execute the steps in the above-mentioned information processing method.
  • This embodiment of the application collects statistics on the feedback data of historical push information; generates a first probability distribution corresponding to the click rate of each historical push information based on the exposure data and click data in the feedback data; selects a preset number of first probability distributions according to the first probability distribution 1.
  • Push information Obtain the target predicted click rate of each first push message, and select the target dynamic push information from the first push information according to the target predicted click rate to push, thereby greatly improving the accuracy of information processing.
  • FIG. 1 is a schematic diagram of a scene of an information processing system provided by an embodiment of the present application
  • Fig. 2a is a schematic flowchart of an information processing method provided by an embodiment of the present application.
  • Figure 2b is a schematic diagram of a product of an information processing method provided by an embodiment of the application.
  • Fig. 2c is a schematic diagram of another product of the information processing method provided by an embodiment of the application.
  • Fig. 2d is a schematic diagram of another product of the information processing method provided by an embodiment of the application.
  • Fig. 2e is a schematic diagram of another product of the information processing method provided by an embodiment of the application.
  • FIG. 3 is another schematic flowchart of an information processing method provided by an embodiment of the present application.
  • FIG. 4a is a schematic diagram of a framework of an information processing method provided by an embodiment of the present application.
  • FIG. 4b is another schematic diagram of the framework of the information processing method provided by an embodiment of the present application.
  • FIG. 5 is a schematic structural diagram of an information processing device provided by an embodiment of the present application.
  • Fig. 6 is a schematic structural diagram of a server provided by an embodiment of the present application.
  • the embodiments of the present application provide an information processing method, device, and computer-readable storage medium.
  • FIG. 1 is a schematic diagram of a scenario of an information processing system provided by an embodiment of the application, including: terminal A, and a server (the information processing system may also include other terminals except terminal A, the specific number of terminals It is not limited here), the terminal A and the server can be connected through a communication network.
  • the communication network can include a wireless network and a wired network.
  • the wireless network includes a wireless wide area network, a wireless local area network, a wireless metropolitan area network, and a wireless personal network. A combination of one or more of them.
  • the network includes network entities such as routers and gateways, which are not shown in the figure.
  • Terminal A can interact with the server through the communication network.
  • terminal A when terminal A is running applications containing various types of push information, such as video, short video, Weibo, and shopping applications, terminal A can detect the user's operation of pushing information Information (that is, feedback data), the operation information includes at least exposure data and click data, and the operation information is sent to the server.
  • pushing information Information that is, feedback data
  • the operation information includes at least exposure data and click data
  • the information processing system of the embodiment of the present application may include an information processing device, and the information processing device may be specifically integrated in a server.
  • the server can receive the operation information uploaded by terminal A, and count the operation information of historical push information.
  • the operation information includes at least exposure data and click data, and generates clicks for each historical push information based on the exposure data and click data.
  • the first probability distribution corresponding to the first probability distribution the first predicted click rate of each push information to be pushed is determined according to the first probability distribution, and a preset number is selected from the push information to be pushed according to the first predicted click rate Get the target predicted click rate of each first push message, and select the target push information to push according to the target predicted click rate.
  • the target dynamic push information is more in line with the user’s preferences and can achieve better creative optimization effects .
  • the solutions of the embodiments are also applicable to the dynamic push of information, and some embodiments below take dynamic push of information as an example for description.
  • the information processing system may also include terminal A, which can install various applications required by users, such as video, short video, microblog, and shopping applications. For example, when terminal A is running video applications, terminal A can Display the push information, detect the user's operation information on the push information, the operation information includes at least exposure data and click data, and send the operation information to the server.
  • terminal A can install various applications required by users, such as video, short video, microblog, and shopping applications.
  • terminal A can Display the push information, detect the user's operation information on the push information, the operation information includes at least exposure data and click data, and send the operation information to the server.
  • FIG. 1 the schematic diagram of the scenario of the information processing system shown in FIG. 1 is only an example.
  • the information processing system and scenario described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute As for the limitation of the technical solutions provided by the embodiments of the present application, those of ordinary skill in the art will know that with the evolution of information processing systems and the emergence of new business scenarios, the technical solutions provided by the embodiments of the present application are equally applicable to similar technical problems.
  • the information processing device may be specifically integrated in a server with a storage unit and a microprocessor with computing capability.
  • the server may be an independent physical server. It can also be a server cluster or distributed system composed of multiple physical servers, or it can provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, and security services. , CDN, and cloud servers for basic cloud computing services such as big data and artificial intelligence platforms.
  • FIG. 2a is a schematic flowchart of an information processing method provided by an embodiment of the present application.
  • the information processing method includes:
  • step 101 the operation information of each dynamic push information is counted.
  • the push information is for the push owner to pay for the push platform, and to promote the relevant introduction information of their products through webpages, search engines, browsers or online media.
  • the push information can be advertisements, etc., the dynamic creativity
  • the push information can be a dynamic creative advertisement (Dynamic Creative, DC), which is a creative display form of push information, which can display different advertising ideas to users, and users can create different dynamic push messages for a push message.
  • DC Dynamic Creative
  • Figure 2b is a product schematic diagram of the information processing method provided by an embodiment of the application.
  • the user can click the dynamic creative advertisement control 11 on the client to create a dynamic push message of push information.
  • the settings such as placement, scheduling, and bidding settings, are no different from ordinary ads.
  • FIG. 2c is a schematic diagram of another product of the information processing method provided by the embodiment of the application.
  • the user can add multiple pictures through the picture adding control 13 in the creative picture area 12, and through the creative copy area 14
  • the copy adding control 15 adds multiple creative copy, and the user can also add multiple titles.
  • Figure 2d is a schematic diagram of another product of the information processing method provided by an embodiment of the application.
  • the client can combine multiple pictures, multiple copywriting, and multiple titles to form multiple push messages
  • a dynamic push message i.e. creative will be selected from among the push messages to users.
  • Figure 2e is a schematic diagram of another product of the information processing method provided by the embodiment of the application.
  • the feedback data of each creative can be monitored through the client page shown in Figure 2e (I.e. the above-mentioned operational information)
  • the dimensions can include all aspects of the pusher who wants to know about the dynamic push information, for example, the exposure data, click data, conversion data, click-through rate, virtual cost data (i.e. cost) of each creative, etc. , To realize the intuitive understanding of the pros and cons of the overall creativity, and to help the push owner to select the outstanding dynamic push information.
  • Cloud technology is a general term for network technology, information technology, integration technology, management platform technology, application technology, etc., based on the application of cloud computing business models. It can form a resource pool, which can be used as needed, which is flexible and convenient. Cloud computing technology will become an important support.
  • the background service of the technical network system requires a large amount of computing and storage resources, such as video websites, image websites and more portal websites.
  • each item may have its own identification mark, which needs to be transmitted to the back-end system for logical processing. Data of different levels will be processed separately, and all types of industry data need to be powerful The backing of the system can only be achieved through cloud computing.
  • this application can use cloud technology to count the operation information of each dynamic push information in the same push information in real time.
  • the operation information can include at least exposure data and click data.
  • the client exposes the dynamic push information users can click on the dynamic push information according to their own interests and jump to the corresponding push page, such as the application download page, recharge page, etc. If the user is not interested, they can also close the dynamic Pushing information does not generate click data, which can reflect the user's degree of interest in the dynamic push of information.
  • a first probability distribution (for example, a first target beta distribution) corresponding to the click rate of each dynamic push information is generated based on the exposure data and the click data.
  • AI Artificial Intelligence
  • digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results.
  • artificial intelligence is a comprehensive technology of computer science, which attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a similar way to human intelligence.
  • Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that the machines have the functions of perception, reasoning and decision-making.
  • Artificial intelligence technology is a comprehensive discipline, covering a wide range of fields, including both hardware-level technology and software-level technology.
  • Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, and mechatronics.
  • Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning/deep learning.
  • Machine Learning is a multi-field interdisciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other subjects. Specializing in the study of how computers simulate or realize human learning behaviors in order to acquire new knowledge or skills, and reorganize the existing knowledge structure to continuously improve its own performance.
  • Machine learning is the core of artificial intelligence, the fundamental way to make computers intelligent, and its applications cover all fields of artificial intelligence.
  • Machine learning and deep learning usually include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and style teaching learning.
  • the server when the server pushes the dynamic push information, it can adopt random selection and the previous click-through rate selection strategy based on creativity.
  • Random selection means that every time it is selected, any creative idea is randomly selected with equal probability from all the dynamic push information for exposure.
  • An obvious disadvantage of this method is that it will waste exposure opportunities on inferior ideas, and there is no The ability to iteratively evolve and optimize based on creative feedback data.
  • Thompson Sampling The performance of each dynamic push message does not have a clear push recommendation strength. This Thompson sampling can solve the problem of how to obtain the recommendation strength of each dynamic push message.
  • the core of the Thompson sampling is Beta distribution.
  • the Beta distribution is a continuous probability distribution defined in the interval [0,1].
  • the prior distribution information and posterior distribution information of the Beta distribution have a unified form, assuming that the prior distribution is Beta( ⁇ , ⁇ ), after s successes and f failures, the posterior distribution information can be Beta(s+ ⁇ , f+ ⁇ ), which realizes that the posterior distribution information is added to the prior distribution information, so that the entire data varies with Update with real changes.
  • the posterior distribution information Add the prior distribution information CTR ⁇ beta( ⁇ , ⁇ ) to generate the first target beta distribution corresponding to the click rate of each dynamic push message, and the first target beta distribution represents the actual clicks of each dynamic push message in practice Rate situation.
  • the step of generating the first target beta distribution corresponding to the click rate of each dynamic push message based on the exposure data and the click data includes:
  • the click-through rate of all historical dynamic push information has a rough range prediction, that is, it conforms to a rough range prediction, so the click-through rate of the historical dynamic push information can be obtained to generate the corresponding prior distribution information CTR ⁇ beta( ⁇ , ⁇ ), and generate the first beta distribution corresponding to the click-through rate of each dynamic push information according to the prior distribution information.
  • the click data generated during the exposure of each dynamic push message and the corresponding non-click data that the user did not click during the exposure are counted.
  • the non-click data can be generated by subtracting the click data from the exposure data, so that according to the click data Generate posterior distribution information corresponding to click-through rate with non-click data
  • the first beta distribution is adjusted according to the posterior distribution information corresponding to the click-through rate, and the first target beta distribution corresponding to the click-through rate of each dynamic push information is obtained.
  • step 103 a preset number of dynamic push messages are selected according to the first target beta distribution.
  • Step 103 Determine a first predicted click rate of each push information to be pushed according to the first probability distribution, and select a preset number of first push information from each push information to be pushed according to the first predicted click rate .
  • the first target beta distribution is continuously combined with the user's click habits during actual use, the first target beta distribution will fit more and more closely to the real click usage with continuous use and learning, and has an excellent click-through rate performance.
  • the peak of the beta distribution of the first target of the dynamic push information is getting higher and higher, and the peak of the first target beta distribution of the dynamic push information with poor click-through rate performance is getting lower and lower, and then the different dynamic push information is continuously adjusted according to Differentiating the actual use can better express the click-through rate of the dynamic push information.
  • a preset number of dynamic push messages with the best click-through rate performance can be selected according to the first target beta distribution corresponding to each dynamic push message, and the preset number can be 5, 10, and so on.
  • the step of selecting a preset number of dynamic push information according to the first target beta distribution may include:
  • the sampling value of the first target beta distribution is the predicted click rate of the dynamic push information. Therefore, the sampling value of the first target beta distribution corresponding to each dynamic push message can be obtained separately to obtain each dynamic push message.
  • the predicted click-through rate the higher the predicted click-through rate, the more the corresponding dynamic recommendation information is liked by users, the lower the predicted click-through rate, the less the corresponding dynamic recommendation information is liked by users, according to the predicted click-through rate Select the best from high to low, and select the best preset number of targets to dynamically push information.
  • step 104 the target predicted click-through rate of the preset number of dynamic push information is acquired, and the target dynamic push information is selected for pushing according to the target predicted click-through rate.
  • Step 104 Use the preset target click-through rate prediction model to obtain the target predicted click-through rate of each first push information in the first push information, and select target push information from the first push information according to the target predicted click-through rate Push it.
  • the embodiment of the present application may also use an advertisement click-through rate prediction (Predict Click-Through Rate, Pctr) model to refine the preset number of dynamic push information.
  • Pctr Predict Click-Through Rate
  • the Pctr model It is a mature advertising prediction model that can predict the precise target predicted click-through rate of dynamic push information. In this way, the Pctr model can be used to obtain a preset number of target predicted click-through rates of dynamic push information.
  • the higher the target predicted click-through rate the The greater the click probability of the dynamic push information, the lower the target predicted click rate, which indicates that the click probability of the dynamic push information is lower, and an accurate secondary prediction is realized.
  • the preset number of dynamic push information can be refined according to the target predicted click-through rate, and the first or second-ranked target dynamic push information can be selected for push to achieve secondary screening, so the target dynamic push information is the best Creative, the target push information can be pushed to users first.
  • the step of obtaining the target predicted click-through rate of the preset number of dynamic push information, and selecting the target dynamic push information according to the target predicted click-through rate to perform the push step may include:
  • the target click rate prediction model (Pctr) is used to perform fine scheduling prediction on the preset number of dynamic push information, and obtain the target predicted click rate of the preset number of dynamic push information.
  • the exposure data of each dynamic push information is summed to obtain the total target exposure data, which reflects the exposure of the dynamic push information.
  • the target exposure data is less than a certain amount in the early stage of exposure, the exposure of each creative idea of dynamic push information is limited. Therefore, the corresponding target dynamic can be selected from the preset number of dynamic push information based on the idea of random sampling.
  • Push information the higher the target predicted click-through rate, the higher the probability of being selected, the lower the target predicted click-through rate, the lower the probability of being selected, so that in the initial stage of exposure, the preset number of dynamic push messages are pushed Chance. If in the later stage of the exposure, that is, the target exposure data is greater than a certain amount, the accuracy of the Pctr model estimation reaches a certain level, and the target dynamic push information with the highest target prediction click rate can be directly selected for push.
  • the embodiment of this application calculates the operation information of each dynamic push information; generates the first target beta distribution corresponding to the click rate of each dynamic push information based on the exposure data and click data; selects the prediction based on the first target beta distribution.
  • Set the number of dynamic push information obtain the target predicted click rate of the preset number of dynamic push information, and select the target dynamic push information for push according to the target predicted click rate.
  • the operation information of each dynamic push information is counted in real time, and the first target beta distribution corresponding to the click-through rate of each dynamic push information is generated based on the Thompson sampling idea, and a preset number of dynamic push information is selected according to the first target beta distribution.
  • FIG. 3 is a schematic flowchart of another information processing method provided by an embodiment of the application.
  • the method flow can include:
  • step 201 the server counts the operation information of each dynamic push information.
  • FIG 4a is a schematic diagram of the information processing method provided by an embodiment of this application.
  • each dynamic creative advertisement includes multiple pieces of exclusive dynamic push information, which is It can be understood as advertisement creativity.
  • the data between different dynamic creative advertisements is isolated.
  • the server obtains the operation information of each dynamic push information in each dynamic creative advertisement through the real-time data stream module 21 in real time.
  • the operation information includes at least exposure data, Click data, conversion data, virtual spending data, etc.
  • the operation information of each dynamic push information is counted through the summary data stream, and the duration of the statistics may be the same day, the last three days, or all the summary data, and so on.
  • step 202 the server collects update information of the dynamic push information according to a preset period, and performs an update operation on the dynamic push information according to the update information.
  • the server collects and pushes the update information of the dynamic push information of the main pair of dynamic push information every 1 minute through the real-time subscription module 23 according to the preset period.
  • the update information performs a real-time replacement operation on the push information stored in the server, ensuring that the push master's operation of the dynamic push information can be fed back to the background in time.
  • step 203 the server calculates the virtual cost data of each dynamic push information based on the virtual cost data and the conversion data, and freezes the dynamic push information corresponding to the virtual cost data greater than the preset virtual data.
  • the pusher sets the expected cost price target cpa for each dynamic push message
  • the server is responsible for controlling the actual exposure bid, and the cost of achieving each conversion behavior is within 1.2 times the expected cost price of the pusher.
  • the cost control module 24 is responsible for suppressing the dynamic push information with poor cost performance in the virtual cost data of the dynamic push information, so as to achieve the goal of controlling the expected cost price of the push main.
  • the cost control module 24 will calculate the virtual cost data of each dynamic push information according to the ratio of the virtual cost data (ie, the cost) to the conversion data after the conversion volume of the dynamic push information exceeds a certain amount, for example, calculate The formula is as follows:
  • the cpa is the virtual cost data
  • the sun (cost) is the virtual cost data
  • the sum (conversion) is the conversion data.
  • the ratio of the virtual cost data to the conversion data can be used to calculate the value of each dynamic push information.
  • the preset virtual data can be 1.2 times the expected cost price. When the virtual cost data is greater than the expected cost price, it means that the cost is running short, and the exposure opportunities of dynamic push information that cost running short should be reduced, namely The dynamic push information corresponding to the virtual cost data greater than the preset virtual data can be frozen without participating in subsequent exposure, thereby protecting the interests of the push owner.
  • the information filtered by the cost control module 24 can be quickly transmitted to the data stream 25 Rough arrangement module 26.
  • step 204 the server obtains the prior distribution information corresponding to the click rate of the historical dynamic push information, and generates the first beta distribution corresponding to the click rate of each dynamic push information according to the prior distribution information.
  • the dynamic push information can be predicted with an average probability, so that the click-through rate of historical dynamic push information can be obtained
  • CTR ⁇ beta( ⁇ , ⁇ ) corresponds to the prior distribution information
  • a first beta distribution corresponding to the click rate of each dynamic push information is generated, and the first beta distribution is an average click rate distribution.
  • step 205 the server counts the click data and non-click data generated when each dynamic push information is exposed, and generates posterior distribution information corresponding to the click rate.
  • the server counts the click data generated during the exposure of each dynamic push message, and subtracts the click data from the exposure data to obtain the non-click data, and generates the corresponding posterior distribution information of the click rate based on the click data and the non-click data
  • step 206 the server adjusts the first beta distribution according to the posterior distribution information corresponding to the click-through rate, and obtains the first target beta distribution corresponding to the click-through rate of each dynamic push information.
  • the posterior distribution information of server click-through rate The curve of the first beta distribution is adjusted to obtain the first target beta distribution corresponding to the click-through rate of each dynamic push.
  • the first target beta distribution is continuously combined with the user's click habits during actual use, so the first target beta The distribution will be more and more suitable for real usage.
  • step 207 the server obtains the target conversion data of each dynamic push information.
  • the conversion data is the behavior data of the main expected event such as application download or virtual recharge after the user clicks on the dynamic push information.
  • the conversion effect is the core benefit of the pusher, so corresponding improvements are needed.
  • the step of obtaining the target conversion data of each dynamic push message may include summing the conversion data of each dynamic push message to obtain the target conversion data.
  • the server can sum the conversion data of each dynamic push message to obtain the total target conversion data.
  • the step of obtaining the target conversion data of each dynamic push information may further include counting the average conversion data of the dynamic push information, and determining the average conversion data as the target conversion data.
  • the premise of considering the conversion data at the push level is that the average of each dynamic push information
  • the number of conversions reaches a certain number. For example, when the average number of conversions for each dynamic push message in the same dynamic advertisement is greater than 1, the condition is met.
  • the server can count the sum of the conversion data of all the dynamically pushed information, calculate the ratio of the sum to the number of dynamically pushed information (ie, the number of ideas), obtain the average conversion data, and determine the average conversion data as the target conversion data.
  • step 208 the server detects whether the target conversion data is less than a first preset threshold.
  • the first preset threshold is a critical value that defines whether the target conversion data reaches a certain number, and the number of the first preset threshold may be the number of dynamically pushed information.
  • step 209 the server obtains the predicted click rate of each dynamic push information according to the first target beta distribution, and selects a preset number of target dynamic push information in descending order of the predicted click rate.
  • the server when the server detects that the target conversion data is less than the first preset threshold, it indicates that the conversion data does not meet the requirements. In this way, the server can separately obtain the sampling value of the first target beta distribution corresponding to each dynamic push message to obtain each A predicted click-through rate of the dynamic push information is selected in the order of the predicted click-through rate from high to low, and a preset number of target dynamic push information is selected to achieve a rough selection method of the coarse sorting module 26.
  • step 210 the server counts the conversion data and non-conversion data generated when each dynamic push message is clicked, and generates posterior distribution information corresponding to the conversion rate.
  • the server when the server detects that the target conversion data is not less than the first preset threshold, it indicates that the conversion data meets the requirements and can start to estimate the conversion rate of each dynamic push message. Based on the Thompson sampling principle, count the clicks of each dynamic push message.
  • the conversion data and non-conversion data generated at the time the conversion data is the data of the number of times the user completes the expected operation after clicking the dynamic push message, and the non-conversion data is the data of the number of times the user does not complete the expected operation after clicking the dynamic push message. Therefore, the posterior distribution information CVR_post ⁇ beta (conversion, click-conversion) corresponding to the conversion rate can be generated directly according to the conversion data and non-conversion data.
  • the conversion is the conversion data
  • the click is the click data.
  • step 211 the server generates a second target beta distribution corresponding to the conversion rate of each dynamic push information according to the posterior distribution information corresponding to the conversion rate.
  • the second target beta distribution corresponding to the conversion rate of each dynamic push information is generated according to the posterior distribution information of the conversion rate CVR_post ⁇ beta (conversion, click-conversion), and the second target beta distribution can be continuously combined with the actual use process of the user Therefore, the second target beta distribution will be more and more suitable for real conversion usage with continuous use and learning.
  • the server obtains the predicted click rate of each dynamic push message according to the first target beta distribution, obtains the predicted conversion rate of each dynamic push message according to the second target beta distribution, and calculates the predicted click rate of each dynamic push message. Combine with the predicted conversion rate to obtain the combination rate, and select a preset number of dynamic push information in the order of the combination rate from high to low.
  • the server obtains the sampling value of the first target beta distribution corresponding to each dynamic push message, obtains the predicted click rate of each dynamic push message, and obtains the sampling value of the second target beta distribution corresponding to each dynamic push message.
  • the predicted conversion rate of each dynamic push information is combined with the predicted click rate and predicted conversion rate of each dynamic push information. For example, the predicted click rate and the predicted conversion rate are multiplied to obtain the combination rate.
  • step 213 the server predicts the preset number of dynamic push information through the target click-through rate prediction model, obtains the target predicted click-through rate of the preset number of dynamic push information, and obtains the target exposure data of each dynamic push information.
  • the coarse layout module 26 includes a lightweight click-through rate prediction model (Litectr) and a lightweight conversion prediction model (Litecvr).
  • the lightweight click-through rate prediction model can Obtain the predicted click-through rate of each dynamic push message, and the lightweight conversion estimation model can obtain the predicted conversion rate of each dynamic push message. Then obtain the effective cost-per-thousand impressions advertising revenue (ecpm) indicator of the advertisement. According to the ecpm, all advertisements are sorted from high to low, and the top N advertisements are selected.
  • ecpm effective cost-per-thousand impressions advertising revenue
  • the ocpa_bid is the advertising bid
  • the Litectr is the predicted click-through rate
  • the Litecvr is the predicted conversion rate.
  • the optimization method of advertisement creative refers to the step of selecting a preset number of target dynamic push information. After selecting a preset number of target dynamic push information, it is sent to the fine ranking module 27.
  • the fine ranking module 27 includes a target click-through rate prediction model (Pctr).
  • target conversion rate prediction model (Pcvr) through the target click rate prediction model to predict the preset number of dynamic push information, get the target predicted click rate of the preset number of dynamic push information, and use the target conversion rate prediction model to predict the preset number
  • the number of dynamic push information is predicted to obtain the target predicted conversion rate of the preset number of dynamic push information. Since the fine sorting module 27 inputs more features of Pctr and Pcvr, the prediction accuracy of the Pctr and Pcvr will be higher.
  • the step of obtaining the target exposure data of each dynamic push information may include summing the exposure data of each dynamic push information to obtain the target exposure data.
  • the server may sum the exposure data of each dynamic push information to obtain the total target exposure data.
  • the step of obtaining the target conversion data of each dynamic push information may further include counting the average exposure data of the dynamic push information, and determining the average exposure data as the target exposure data.
  • the server can count the sum of the exposure data of all dynamically pushed information, calculate the ratio of the sum to the number of dynamically pushed information (that is, the number of creatives), obtain the average exposure data, and determine the average exposure data as the target exposure data.
  • step 214 the server detects whether the target exposure data is less than a second preset threshold.
  • the target click-through rate prediction model because the learning of the target click-through rate prediction model requires certain data support, in the early stage of exposure, that is, when the target exposure data is less, the target click-through rate prediction model may not be accurate, and in the post-exposure stage, the target exposure data In many cases, the target click-through rate prediction model is accurate, so it is necessary to set a second preset threshold to distinguish between the current early stage and the late stage of exposure.
  • the second preset threshold can be 5000 or 300, etc., when the server detects When the target exposure data is less than the second preset threshold, step 215 is executed, and when the server detects that the target exposure data is not less than the second preset threshold, step 216 is executed.
  • step 215 the server normalizes the target predicted click-through rate to obtain target prediction vector information of a preset number of dimensions, divides the probability interval based on the target prediction vector information, and randomly accesses the probability interval.
  • the push information is determined to be the target dynamic push information for push.
  • the server detects that the target exposure data is less than the second preset threshold, it means that the fine-ranking module 27 is in the early stage of exposure, the exposure of the preset number of dynamic push information is limited, and the Pctr model is inaccurate, and it cannot be completely at this time.
  • the target prediction click rate needs to be normalized by the softmax function to obtain the target prediction vector information of the preset number of dimensions, and the probability range of the vector element of each dimension is between (0, 1) , And the sum of the vector elements of all dimensions is 1. The higher the predicted click-through rate of the target, the larger the probability range after conversion.
  • the softmax function is as follows:
  • the Win_ratei is the target prediction vector information
  • the e is a constant
  • the pctri is the target predicted click-through rate
  • the num of creatives is a value minus 1 from the preset number.
  • the probability interval of (0, 1) is divided according to the probability of each element in the target prediction vector information.
  • target dynamic promotion information represents the dynamic creative advertisement participating in the subsequent bidding.
  • the target predicted click-through rate is low, but the dynamic push information with high conversion rate can also be creatively recommended, and the push host is fully considered
  • the benefits of the target’s dynamic push information is diversified.
  • step 216 the server determines the dynamic push information with the highest target predicted click rate as the target dynamic push information for push.
  • the server detects that the target exposure data is not less than the second preset threshold, it means that the fine-ranking module 27 is in the post-exposure phase, and the preset number of dynamic push information exposure meets the requirements.
  • the Pctr model predicts accurately and can directly predict the target.
  • the dynamic push information with the highest click-through rate is determined as the target push information for push, that is, the target dynamic push information represents the dynamic creative advertisement to participate in the subsequent bidding.
  • the fine-ranking module 27 determines the optimal target dynamic push information of each dynamic creative advertisement as a representative, the more accurate target ecpm indicators of the first N advertisements are further calculated, and selected according to the target ecpm indicators The best 1 to 2 advertisements are pushed to users. For example, please refer to the following formula:
  • the ocpa_bid is the advertising bid
  • the Pctr is the target predicted click-through rate
  • the Litecvr is the target predicted conversion rate.
  • the advertisement bid, target predicted click rate and target predicted conversion rate can be calculated by the above formula to calculate the refined ecpm index of each advertisement. , And select the best 1 to 2 ads according to the refined ecpm index and push them to users.
  • the fine-ranking module 27 predicts the preset number of dynamic push information through the target click-through rate prediction model, and after obtaining the target predicted click-through rate of the preset number of dynamic push information, it may further include:
  • the server detects whether the target exposure data is less than the second preset threshold.
  • step (3) when the server detects that the target exposure data is less than the second preset threshold, step (3) is executed, and when the server detects that the target exposure data is not less than the second preset threshold, step (4) is executed.
  • the server normalizes the target combination rate to obtain the combined prediction vector information of the preset number dimensions, divides the probability interval based on the combined prediction vector information, and randomly accesses the probability interval, and determines the dynamic push information corresponding to the access probability interval Push for the target dynamic push information.
  • the server determines the dynamic push information with the highest target combination rate as the target dynamic push information for push.
  • the fine ranking module 27 further introduces the selection and recommendation of the target dynamic push information at the conversion level, so that the target dynamic push information is more accurate.
  • the embodiment of this application calculates the operation information of each dynamic push information; generates the first target beta distribution corresponding to the click rate of each dynamic push information based on the exposure data and click data; selects the prediction based on the first target beta distribution.
  • Set the number of dynamic push information obtain the target predicted click rate of the preset number of dynamic push information, and select the target dynamic push information for push according to the target predicted click rate.
  • the operation information of each dynamic push information is counted in real time, and the first target beta distribution corresponding to the click-through rate of each dynamic push information is generated based on the Thompson sampling idea, and a preset number of dynamic push information is selected according to the first target beta distribution.
  • the conversion data level is introduced in the selection process to filter the dynamic push information, the push of the target dynamic push information is more in line with the requirements of the push owner, the effect of creative optimization is improved, and the accuracy of information processing is further improved.
  • the embodiment of the present application also provides an apparatus based on the foregoing information processing method.
  • the meanings of the nouns are the same as those in the foregoing information processing method.
  • FIG. 5 is a schematic structural diagram of an information processing device provided by an embodiment of the application.
  • the information processing device may include a statistical unit 301, a generating unit 302, a first coarse sorting unit 303, and a fine sorting unit 304.
  • the statistics unit 301 is configured to count feedback data of historical push information, and the feedback data includes at least exposure data and click data.
  • the generating unit 302 is configured to generate a first probability distribution corresponding to the click rate of each historical dynamic push information based on the exposure data and the click data.
  • the generating unit 302 is configured to: obtain the prior distribution information corresponding to the click-through rate of historical dynamic push information, and generate the first beta distribution corresponding to the click-through rate of each dynamic push information according to the prior distribution information; Count the click data and non-click data generated during the exposure of each dynamic push information, and generate the posterior distribution information corresponding to the click rate; adjust the first beta distribution according to the posterior distribution information corresponding to the click rate to obtain each The click-through rate of the dynamic push information corresponds to the first target beta distribution.
  • the first coarse sorting unit 303 is configured to determine the first predicted click rate of each push information to be pushed according to the first probability distribution, and select a preset from the push information to be pushed according to the first predicted click rate Number of first push messages.
  • the first coarse sorting unit 303 is configured to: obtain the predicted click rate of each dynamic push information according to the first probability distribution; select a preset number of click rates in descending order of the predicted click rate The target pushes information dynamically.
  • the fine ranking unit 304 is configured to obtain the target predicted click rate of each first push information by using a preset target click rate prediction model, and select the target dynamic push information for pushing according to the target predicted click rate.
  • the fine sorting unit 304 includes:
  • the prediction subunit is used to predict the preset number of dynamic push information through the target click-through rate prediction model to obtain the target predicted click-through rate of the preset number of dynamic push information;
  • the exposure sub-unit is used to obtain the target exposure data of each dynamic push information
  • the combination subunit is used to combine the target's predicted click-through rate and target exposure data to select target dynamic push information for push.
  • the combining subunit is configured to: when it is detected that the target exposure data is less than a second preset threshold, normalize the target predicted click-through rate to obtain a target prediction vector of a preset number of dimensions Information; divide the probability interval based on the target prediction vector information, randomly visit the probability interval, and determine the dynamic push information corresponding to the visited probability interval as the target dynamic push information for push; when it is detected that the target exposure data is not less than the second preset When the threshold is set, the dynamic push information with the highest predicted click rate of the target is determined as the target dynamic push information for push.
  • the operation information further includes conversion data
  • the device further includes:
  • the conversion unit is used to obtain the target conversion data of each dynamic push information
  • the first coarse sorting unit is configured to perform the step of selecting a preset number of dynamic push information according to the first target beta distribution when it is detected that the target conversion data is less than a first preset threshold;
  • the second rough sorting unit is used to count the conversion data and non-conversion data generated when each dynamic push message is clicked when it is detected that the target conversion data is not less than the first preset threshold, and generate a posterior distribution corresponding to the conversion rate Information; according to the posterior distribution information corresponding to the conversion rate, generate a second target beta distribution corresponding to the conversion rate of each dynamic push message; combine the first target beta distribution and the second target beta distribution to select a preset number of dynamic push messages .
  • the second coarse sorting unit is further used to: when it is detected that the target conversion data is not less than the first preset threshold, count the conversion data and non-conversion data generated when each dynamic push message is clicked. Generate the posterior distribution information corresponding to the conversion rate; generate the second target beta distribution corresponding to the conversion rate of each dynamic push message according to the posterior distribution information corresponding to the conversion rate; obtain each dynamic push message according to the first target beta distribution According to the second target beta distribution, obtain the predicted conversion rate of each dynamic push information; combine the predicted click rate and the predicted conversion rate of each dynamic push information to obtain the combination rate; according to the combination rate from high Select a preset number of dynamic push messages in the lowest order.
  • the operation information further includes virtual expense data
  • the device further includes:
  • the cost control unit is used to calculate the virtual cost data of each dynamic push information according to the virtual cost data and the conversion data; and freeze the dynamic push information corresponding to the virtual cost data greater than the preset virtual data.
  • the device further includes:
  • the update unit is configured to collect update information of the dynamic push information according to a preset period; and perform an update operation on the dynamic push information according to the update information.
  • the embodiment of the present application uses the statistical unit 301 to count the operation information of each dynamic push message; the generating unit 302 generates the first target beta distribution corresponding to the click rate of each dynamic push message based on the exposure data and click data; first The coarse sorting unit 303 selects a preset number of dynamic push information according to the first target beta distribution; the fine sorting unit 304 obtains the target predicted click rate of the preset number of dynamic push information, and selects the target dynamic push information for pushing according to the target predicted click rate.
  • the operation information of each dynamic push information is counted in real time, and the first target beta distribution corresponding to the click-through rate of each dynamic push information is generated based on the Thompson sampling idea, and a preset number of dynamic push information is selected according to the first target beta distribution.
  • the embodiment of the present application also provides a server, as shown in FIG. 6, which shows a schematic structural diagram of the server involved in the embodiment of the present application, specifically:
  • the server may include one or more processing core processors 401, one or more computer-readable storage medium memory 402, power supply 403, input unit 404 and other components.
  • processing core processors 401 one or more computer-readable storage medium memory 402, power supply 403, input unit 404 and other components.
  • FIG. 6 does not constitute a limitation on the server, and may include more or fewer components than shown in the figure, or a combination of certain components, or different component arrangements. in:
  • the processor 401 is the control center of the server. It uses various interfaces and lines to connect various parts of the entire server, and by running or executing software programs and/or modules stored in the memory 402, and calling data stored in the memory 402, Perform various functions of the server and process data to monitor the server as a whole.
  • the processor 401 may include one or more processing cores; preferably, the processor 401 may integrate an application processor and a modem processor, where the application processor mainly processes the operating system, user interface, application programs, etc. , The modem processor mainly deals with wireless communication. It can be understood that the foregoing modem processor may not be integrated into the processor 401.
  • the memory 402 may be used to store software programs and modules.
  • the processor 401 executes various functional applications and data processing by running the software programs and modules stored in the memory 402.
  • the memory 402 may mainly include a program storage area and a data storage area.
  • the program storage area may store an operating system, an application program required by at least one function (such as a sound playback function, an image playback function, etc.), etc.; Data created by the use of the server, etc.
  • the memory 402 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.
  • the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.
  • the server also includes a power supply 403 for supplying power to various components.
  • the power supply 403 may be logically connected to the processor 401 through a power management system, so that functions such as charging, discharging, and power consumption management can be managed through the power management system.
  • the power supply 403 may also include any components such as one or more DC or AC power supplies, a recharging system, a power failure detection circuit, a power converter or inverter, and a power status indicator.
  • the server may further include an input unit 404, which can be used to receive input digital or character information, and generate keyboard, mouse, joystick, optical or trackball signal input related to user settings and function control.
  • an input unit 404 which can be used to receive input digital or character information, and generate keyboard, mouse, joystick, optical or trackball signal input related to user settings and function control.
  • the server may also include a display unit, etc., which will not be repeated here.
  • the processor 401 in the server loads the executable file corresponding to the process of one or more application programs into the memory 402 according to the following instructions, and the processor 401 runs the executable file stored in the memory.
  • the application program in 402 realizes various functions, as follows:
  • the operation information includes at least exposure data and click data; generate the first target beta distribution corresponding to the click rate of each dynamic push information based on the exposure data and click data; according to the first target Beta distribution selects a preset number of dynamic push information; obtains the target predicted click-through rate of the preset number of dynamic push information, and selects the target dynamic push information for push according to the target predicted click-through rate.
  • the server of the embodiment of the present application can calculate the operation information of each dynamic push information; generate the first target beta distribution corresponding to the click rate of each dynamic push information based on the exposure data and click data; according to the first target beta A preset number of dynamic push information is selected for distribution; the target predicted click rate of the preset number of dynamic push information is obtained, and the target dynamic push information is selected for push according to the target predicted click rate.
  • the operation information of each dynamic push information is counted in real time, and the first target beta distribution corresponding to the click-through rate of each dynamic push information is generated based on the Thompson sampling idea, and a preset number of dynamic push information is selected according to the first target beta distribution.
  • an embodiment of the present application provides a computer-readable storage medium in which multiple instructions are stored, and the instructions can be loaded by a processor to execute the steps in any information processing method provided in the embodiments of the present application.
  • the instruction can perform the following steps:
  • the operation information includes at least exposure data and click data; generate the first target beta distribution corresponding to the click rate of each dynamic push information based on the exposure data and click data; according to the first target Beta distribution selects a preset number of dynamic push information; obtains the target predicted click-through rate of the preset number of dynamic push information, and selects the target dynamic push information for push according to the target predicted click-through rate.
  • the computer-readable storage medium may include: read only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disks or optical disks, etc.
  • the instructions stored in the computer-readable storage medium can execute the steps in any information processing method provided in the embodiments of the present application, it can implement the steps in any information processing method provided in the embodiments of the present application.
  • the beneficial effects that can be achieved refer to the previous embodiment for details, and will not be repeated here.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Computational Mathematics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Algebra (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Information Transfer Between Computers (AREA)

Abstract

一种信息处理方法、装置及计算机可读存储介质。所述方法包括：统计历史推送信息的反馈数据，其中包括曝光数据和点击数据（101）；基于曝光数据和点击数据生成每一历史推送信息的点击率相应的第一概率分布（102）；根据第一概率分布确定待推送的各推送信息的第一预测点击率，根据第一预测点击率从待推送的各推送信息中选取预设数量的第一推送信息（103）；利用预设的目标点击率预测模型获取各第一推送信息的目标预测点击率，根据目标预测点击率选取目标动态推送信息进行推送（104）。

Description

一种信息处理方法、装置及计算机可读存储介质
本申请要求于2020年05月19日提交中国专利局、申请号为2020104210251.9、发明名称为“一种信息处理方法、装置及计算机可读存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及信息处理技术领域，具体涉及一种信息处理方法、装置及计算机可读存储介质。
发明背景
随着网络的发展和计算机的广泛应用，线上信息推送市场迅速扩展。出钱投放推送信息的推送主可以向发布者(推送平台)付费，以通过网页、搜索引擎、浏览器或其它在线媒体来投放自己的推送信息，从而很好地推广自己的产品。
现有技术中，更好实现更好的创意挑选，推送主可以在一个推送信息下建立多个动态推送信息，形成不同的创意，并将多个动态推送信息同时进行随机线上推送，根据各个动态推送信息的在客户端上的表现，挑选出表现最佳的动态推送信息，实现创意优选。
发明内容
本申请实施例提供一种信息处理方法、装置及计算机可读存储介质，可以提升信息处理的准确率。
本申请实施例提供一种信息处理方法，包括：
统计历史推送信息的反馈数据，所述反馈数据至少包括曝光数据和点击数据；
基于所述曝光数据和点击数据生成所述历史推送信息中每一推送信息的点击率相应的第一概率分布；
根据所述第一概率分布确定待推送的各推送信息的第一预测点击率，根据所述第一预测点击率从所述待推送的各推送信息中选取预设数量的第一推送信息；
获取各所述第一推送信息的目标预测点击率，根据所述目标预测点击率从所述第一推送信息中选取目标动态推送信息进行推送。
一种信息处理装置，包括：
统计单元，用于统计历史推送信息的反馈数据，所述反馈数据至少包括曝光数据和点击数据；
生成单元，用于基于所述曝光数据和点击数据生成所述历史推送信息中每一推送信息的点击率相应的第一概率分布；
第一粗排单元，用于根据所述第一概率分布确定待推送的各推送信息的第一 预测点击率，根据所述第一预测点击率从所述待推送的各推送信息中选取预设数量的第一推送信息；
精排单元，用于获取各所述第一推送信息的目标预测点击率，根据所述目标预测点击率从所述第一推送信息中选取目标动态推送信息进行推送。
一种计算机可读存储介质，所述计算机可读存储介质存储有多条指令，所述指令适于处理器进行加载，以执行上述信息处理方法中的步骤。
本申请实施例通过统计历史推送信息的反馈数据；基于反馈数据中的曝光数据和点击数据生成每一历史推送信息的点击率相应的第一概率分布；根据第一概率分布选取预设数量的第一推送信息；获取各第一推送信息的目标预测点击率，根据目标预测点击率从第一推送信息中选取目标动态推送信息进行推送，从而极大的提升了信息处理的准确率。
附图简要说明
为了更清楚地说明本申请实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1是本申请实施例提供的信息处理系统的场景示意图；
图2a是本申请实施例提供的信息处理方法的流程示意图；
图2b为本申请实施例提供的信息处理方法的产品示意图；
图2c为本申请实施例提供的信息处理方法的另一产品示意图；
图2d为本申请实施例提供的信息处理方法的另一产品示意图；
图2e为本申请实施例提供的信息处理方法的另一产品示意图；
图3是本申请实施例提供的信息处理方法的另一流程示意图；
图4a是本申请实施例提供的信息处理方法的框架示意图；
图4b是本申请实施例提供的信息处理方法的另一框架示意图；
图5是本申请实施例提供的信息处理装置的结构示意图；
图6是本申请实施例提供的服务器的结构示意图。
实施本发明的方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
本申请实施例提供一种信息处理方法、装置、及计算机可读存储介质。
请参阅图1，图1为本申请实施例所提供的信息处理系统的场景示意图，包 括：终端A、和服务器(该信息处理系统还可以包括除终端A之外的其他终端，终端具体个数在此处不作限定)，终端A与服务器之间可以通过通信网络连接，该通信网络，可以包括无线网络以及有线网络，其中无线网络包括无线广域网、无线局域网、无线城域网、以及无线个人网中的一种或多种的组合。网络中包括路由器、网关等等网络实体，图中并未示意出。终端A可以通过通信网络与服务器进行信息交互，比如终端A在运行包含各类推送信息的应用，例如视频、短视频、微博和购物各类应用时，终端A可以检测用户对推送信息的操作信息(也即反馈数据)，该操作信息至少包括曝光数据和点击数据，并将该操作信息发送至服务器。
在对现有技术的研究和实践过程中，本申请的发明人发现，现有技术中，由于随机推送的方式，会导致在劣质的动态推送信息上浪费曝光的机会，影响创意优选的结果，导致信息处理的准确率较低。
本申请实施例的信息处理系统可以包括信息处理装置，该信息处理装置具体可以集成在服务器中。在图1中，该服务器可以接收终端A上传的操作信息，统计历史推送信息的操作信息，该操作信息至少包括曝光数据和点击数据，基于该曝光数据和点击数据生成每一历史推送信息的点击率相应的第一概率分布，根据该第一概率分布确定待推送的各推送信息的第一预测点击率，根据所述第一预测点击率从所述待推送的各推送信息中选取预设数量的第一推送信息，获取各第一推送信息的目标预测点击率，根据该目标预测点击率选取目标推送信息进行推送，该目标动态推送信息更符合用户的喜好，可以实现更好的创意优选效果。各实施例的方案也适用于动态推送信息，后文的一些实施例以动态推送信息为例进行说明。
该信息处理系统还可以包括终端A，该终端A可以安装各种用户需要的应用，例如视频、短视频、微博和购物等各类应用，例如，终端A在运行视频应用时，终端A可以显示推送信息，检测用户对推送信息的操作信息，该操作信息至少包括曝光数据和点击数据，并将该操作信息发送至服务器。
需要说明的是，图1所示的信息处理系统的场景示意图仅仅是一个示例，本申请实施例描述的信息处理系统以及场景是为了更加清楚的说明本申请实施例的技术方案，并不构成对于本申请实施例提供的技术方案的限定，本领域普通技术人员可知，随着信息处理系统的演变和新业务场景的出现，本申请实施例提供的技术方案对于类似的技术问题，同样适用。
以下分别进行详细说明。需说明的是，以下实施例的序号不作为对实施例优选顺序的限定。
实施例一、
在本实施例中，将从信息处理装置的角度进行描述，该信息处理装置具体可以集成在具备储存单元并安装有微处理器而具有运算能力的服务器中，该服务器可以是独立的物理服务器，也可以是多个物理服务器构成的服务器集群或者分布式系统，还可以是提供云服务、云数据库、云计算、云函数、云存储、网络服务、云通信、中间件服务、域名服务、安全服务、CDN、以及大数据和人工智能平台等基础 云计算服务的云服务器。
请参阅图2a，图2a是本申请实施例提供的信息处理方法的流程示意图。该信息处理方法包括：
在步骤101中，统计每一动态推送信息的操作信息。
需要说明的是，推送信息为推送主出钱向推送平台付费，通过网页、搜索引擎、浏览器或者在线媒体来推广自己的产品的相关介绍信息，该推送信息可以为广告等等，该动态创意推送信息可以为动态创意广告(Dynamic Creative，DC)，该动态推送信息是一种推送信息的创意展现形式，可以对用户展示不同的广告创意，用户可以为一推送信息建立不同的动态推送信息，为了更好的说明本申请实施例的动态推送信息，可以参照以下说明：
请一并参阅图2b所示，图2b为本申请实施例提供的信息处理方法的产品示意图，用户可以在客户端上点击动态创意广告控件11来创建推送信息的动态推送信息，该动态推送信息的设置，如版位、排期、出价方面的设置和普通广告无区别。
请继续参阅图2c所示，图2c为本申请实施例提供的信息处理方法的另一产品示意图，用户可以通过创意图片区域12的图片添加控件13增加多个图片，并通过创意文案区域14的文案添加控件15增加多个创意文案，用户还可以增加多个标题。
请继续参阅图2d所示，图2d为本申请实施例提供的信息处理方法的另一产品示意图，客户端可以将该多个图片、多个文案和多个标题进行组合，形成推送信息的多个动态推送信息的创意，如图2d所示，由4个图片、3个标题和4个创意文案(即描述)，可以组成4*3*4＝48个动态推送信息，在后续的推送信息的推送中会从中择优选出一个动态推送信息(即创意)给用户，为了实现广告主的资源最大化利用，需要挑选多个动态推送信息中创意最佳的动态推送信息进行展示，例如，请继续参阅图2e所示，图2e为本申请实施例提供的信息处理方法的另一产品示意图，在该动态推送信息投放之后，可以通过图2e所示的客户端页面监控每个创意的反馈数据(即上述操作信息)，维度可以包括推送主想要了解关于动态推送信息的各个方面，例如，每个创意的曝光数据、点击数据、转化数据、点击率、虚拟成本数据(即花费)等等，实现对于整体创意的优劣的直观认识，帮助推送主挑选出其中优秀的动态推送信息。
云技术(Cloud technology)基于云计算商业模式应用的网络技术、信息技术、整合技术、管理平台技术、应用技术等的总称，可以组成资源池，按需所用，灵活便利。云计算技术将变成重要支撑。技术网络系统的后台服务需要大量的计算、存储资源，如视频网站、图片类网站和更多的门户网站。伴随着互联网行业的高度发展和应用，将来每个物品都有可能存在自己的识别标志，都需要传输到后台系统进行逻辑处理，不同程度级别的数据将会分开处理，各类行业数据皆需要强大的系统后盾支撑，只能通过云计算来实现。
其中，本申请可以通过云技术实时统计同一推送信息中每一动态推送信息的 操作信息，该操作信息至少可以包括曝光数据和点击数据，该曝光数据即为每一动态推送信息在曝光(即展示)时产生的数量，曝光数据越大，说明动态推送信息展示的次数越多，该曝光数据越小，说明动态推送信息展示的次数越少，该点击数据为动态推送信息在曝光时进行点击产生的数量，客户端在曝光动态推送信息时，用户根据自身兴趣可以点击该动态推送信息，跳转至相应的推送页面，如应用下载页面、充值页面等，用户如果不感兴趣，也可以关闭该动态推送信息，不产生点击数据，该点击数据可以反映用户对动态推送信息的感兴趣程度。
在步骤102中，基于曝光数据和点击数据生成每一动态推送信息的点击率相应的第一概率分布(例如第一目标贝塔分布)。
人工智能(Artificial Intelligence，AI)是利用数字计算机或者数字计算机控制的机器模拟、延伸和扩展人的智能，感知环境、获取知识并使用知识获得最佳结果的理论、方法、技术及应用系统。换句话说，人工智能是计算机科学的一个综合技术，它企图了解智能的实质，并生产出一种新的能以人类智能相似的方式做出反应的智能机器。人工智能也就是研究各种智能机器的设计原理与实现方法，使机器具有感知、推理与决策的功能。
人工智能技术是一门综合学科，涉及领域广泛，既有硬件层面的技术也有软件层面的技术。人工智能基础技术一般包括如传感器、专用人工智能芯片、云计算、分布式存储、大数据处理技术、操作/交互系统、机电一体化等技术。人工智能软件技术主要包括计算机视觉技术、语音处理技术、自然语言处理技术以及机器学习/深度学习等几大方向。
机器学习(Machine Learning,ML)是一门多领域交叉学科，涉及概率论、统计学、逼近论、凸分析、算法复杂度理论等多门学科。专门研究计算机怎样模拟或实现人类的学习行为，以获取新的知识或技能，重新组织已有的知识结构使之不断改善自身的性能。机器学习是人工智能的核心，是使计算机具有智能的根本途径，其应用遍及人工智能的各个领域。机器学习和深度学习通常包括人工神经网络、置信网络、强化学习、迁移学习、归纳学习、式教学习等技术。
本申请实施例提供的方案涉及人工智能的深度学习等技术，具体通过如下实施例进行说明：
在相关技术中，服务器对动态推送信息进行推送时，可以采用随机优选和基于创意以往的点击率优选策略。
(a)随机优选即在每次优选时，从所有动态推送信息中随机等概率挑选其中的任意个创意进行曝光，这种方法的一个明显缺点是会在劣质的创意上浪费曝光机会，而且没有根据创意的反馈数据进行迭代演进优选的能力。
(b)基于创意以往的点击率优选策略是根据每一动态推送信息的历史点击率进行打分排序，选取其中历史点击率最高的任意个创意进行推荐，但是这种策略的缺点在于动态推送信息前期数据较低时，会因为播放的偶然因素导致将不好的动态推送信息进行集中播放，造成优选结果不好。
其中，汤普森采样(Thompson Sampling)：每一动态推送信息的表现并没有一个明确的推送推荐力度，该汤普森采样可以解决如何获取每一动态推送信息的推荐力度的问题，该汤普森采样的核心即为贝塔(Beta)分布，该贝塔分布为一个定义在[0，1]区间上的连续概率分布，贝塔分布的先验分布信息和后验分布信息具有统一的形式，假设先验分布为Beta(α，β)，经过s次成功和f次失败后，其后验分布信息可以为Beta(s+α，f+β)，实现将后验分布信息加入到先验分布信息中，使得整个数据随着真实变化而更新。
进一步的，通过该后验分布信息
Figure PCTCN2021086778-appb-000001
Figure PCTCN2021086778-appb-000002
加入先验分布信息CTR～beta(α,β)，生成每一动态推送信息的点击率相应的第一目标贝塔分布，该第一目标贝塔分布代表了每一动态推送信息在实际中真实的点击率情况。
在一些实施方式中，基于该曝光数据和点击数据生成每一动态推送信息的点击率相应的第一目标贝塔分布的步骤，包括：
(1)获取历史动态推送信息的点击率相应的先验分布信息，根据该先验分布信息生成每一动态推送信息的点击率相应的第一贝塔分布；
(2)统计每一动态推送信息在曝光时生成的点击数据和非点击数据，生成点击率相应的后验分布信息；
(3)根据该点击率相应的后验分布信息对该第一贝塔分布进行调整，得到每一动态推送信息的点击率相应的第一目标贝塔分布。
其中，由于历史的所有的动态推送信息的点击率有一个大概范围预测，即符合一个大概范围的预测，所以可以获取历史动态推送信息的点击率生成相应的先验分布信息CTR～beta(α,β)，并根据该先验分布信息生成每一动态推送信息的点击率相应的第一贝塔分布。
进一步的，统计每一动态推送信息在曝光时生成的点击数据和在曝光时用户未点击相应的未点击数据，该未点击数据可以通过曝光数据减去点击数据生成，以此，根据该点击数据和非点击数据生成点击率相应的后验分布信息
Figure PCTCN2021086778-appb-000003
根据该点击率相应的后验分布信息对该第一贝塔分布进行调整，得到每一动态推送信息的点击率相应的第一目标贝塔分布。
在步骤103中，根据第一目标贝塔分布选取预设数量的动态推送信息。
步骤103根据所述第一概率分布确定待推送的各推送信息的第一预测点击率，根据所述第一预测点击率从所述待推送的各推送信息中选取预设数量的第一推送信息。其中，由于该第一目标贝塔分布不断结合用户实际使用过程中的点击习惯，因此该第一目标贝塔分布会随着不断的使用学习越来贴合于真实的点击使用情况，对于点击率表现优秀的动态推送信息的第一目标贝塔分布的峰值越来越高，对于点击率表现不佳的动态推送信息的第一目标贝塔分布的峰值越来越低，进而不断的将不同的动态推送信息按照实际使用进行区分化，可以更好对动态推送信息的点击率 好坏进行准确的表达。基于此，可以根据每一动态推送信息相应的第一目标贝塔分布选取点击率表现最佳的预设数量的动态推送信息，该预设数量可以为5个、10个等等。
在一实施方式中，该根据第一目标贝塔分布选取预设数量的动态推送信息的步骤，可以包括：
(1)根据该第一目标贝塔分布获取每一动态推送信息的预测点击率；
(2)按照该预测点击率由高至低的顺序选取预设数量的目标动态推送信息。
其中，该第一目标贝塔分布的采样值即为该动态推送信息的预测点击率，因此，可以分别获取每一动态推送信息相应的第一目标贝塔分布的采样值，以得到每一动态推送信息的预测点击率，该预测点击率越高，说明相应的动态推荐信息的越受用户的喜爱，该预测点击率越低，说明相应的动态推荐信息越不受用户的喜爱，按照该预测点击率由高至低的顺序进行择优，选取最佳的预设数量的目标动态推送信息。
在步骤104中，获取预设数量的动态推送信息的目标预测点击率，根据目标预测点击率选取目标动态推送信息进行推送。
步骤104利用预设的目标点击率预测模型获取所述第一推送信息中每个第一推送信息的目标预测点击率，根据所述目标预测点击率从所述第一推送信息中选取目标推送信息进行推送。其中，在获取预设数量的动态推送信息时，本申请实施例还可以通过广告点击率预测(Predict Click-Through Rate，Pctr)模型对该预设数量的动态推送信息进行精排，该Pctr模型为成熟的广告预测模型，可以预测动态推送信息精确的目标预测点击率，以此，可以通过该Pctr模型获取预设数量的动态推送信息的目标预测点击率，该目标预测点击率越高，说明该动态推送信息的点击概率越大，该目标预测点击率越低，说明该动态推送信息的点击概率越低，实现精确的二次预测。
进一步，可以根据该目标预测点击率对该预设数量的动态推送信息进行精排，选取其中排名一或者二的目标动态推送信息进行推送，实现二次筛选，所以该目标动态推送信息为最佳创意，可以将该目标推送信息优先推送至用户。
在一实施方式中，该获取预设数量的动态推送信息的目标预测点击率，根据目标预测点击率选取目标动态推送信息进行推送步骤，可以包括：
(1)通过目标点击率预测模型对该预设数量的动态推送信息进行预测，得到该预设数量的动态推送信息的目标预测点击率；
(2)将每一动态推送信息的曝光数据进行求和，得到目标曝光数据；
(3)结合该目标预测点击率和目标曝光数据选取目标动态推送信息进行推送。
其中，通过目标点击率预测模型(Pctr)对预设数量的动态推送信息进行精排预测，得到预设数量的动态推送信息的目标预测点击率。将每一动态推送信息的曝光数据进行求和，得到总的目标曝光数据，该目标曝光数据反映了动态推送信息 的曝光情况。
进一步的，如果在曝光初期，即目标曝光数据小于一定数量，每个动态推送信息的创意的曝光有限，以此，可以基于随机采样的思想从预设数量的动态推送信息中选取相应的目标动态推送信息，该目标预测点击率越高，被选取的概率越高，该目标预测点击率越低，被选取的概率越低，实现在曝光初期，预设数量的动态推送信息均有被推送的机会。如果在曝光后期，即目标曝光数据大于一定数量，该Pctr模型估计的准确度达到一定水平，可以直接选取目标预测点击率最高的目标动态推送信息进行推送。
由上述可知，本申请实施例通过统计每一动态推送信息的操作信息；基于曝光数据和点击数据生成每一动态推送信息的点击率相应的第一目标贝塔分布；根据第一目标贝塔分布选取预设数量的动态推送信息；获取预设数量的动态推送信息的目标预测点击率，根据目标预测点击率选取目标动态推送信息进行推送。以此，实时统计每一动态推送信息的操作信息，基于汤普森采样思想生成每一动态推送信息的点击率相应的第一目标贝塔分布，根据第一目标贝塔分布选取预设数量的动态推送信息并获取相应的目标点击率，根据目标预测点击率选取精确的目标动态推送信息进行推送，极大的提升了信息处理的准确率。
实施例二、
根据实施例一所描述的方法，以下将举例作进一步详细说明。
在本实施例中，将以该信息处理装置具体集成在服务器中为例进行说明。
请参阅图3，图3为本申请实施例提供的信息处理方法的另一流程示意图。该方法流程可以包括：
在步骤201中，服务器统计每一动态推送信息的操作信息。
其中，请一并参阅图4a，图4a为本申请实施例提供的信息处理方法的框架示意图，需要说明的是，每一动态创意广告中包括多条专属的动态推送信息，该动态推送信息即可以理解为广告创意，不同动态创意广告之间的数据是隔离的，服务器通过实时数据流模块21实时获取每一动态创意广告中每一动态推送信息的操作信息，该操作信息至少包括曝光数据、点击数据、转化数据和虚拟花销数据等等。
进一步的，通过汇总数据流统计每一动态推送信息的操作信息，该统计的时长可以为当天、最近三天或者所有汇总数据等等。
在步骤202中，服务器按照预设周期采集动态推送信息的更新信息，根据更新信息对动态推送信息进行更新操作。
其中，请继续参阅图4a，服务器通过实时订阅模块23按照预设周期，如每隔1分钟采集推送主对动态推送信息的更新信息，该更新信息包括增、删、查、和改信息，根据该更新信息实时对服务器中存储的推送信息进行更换操作，保证推送主对动态推送信息的操作可以及时反馈到后台。
在步骤203中，服务器根据虚拟花销数据和转化数据计算每一动态推送信息的虚拟成本数据，将虚拟成本数据大于预设虚拟数据相应的动态推送信息进行冻结。
其中，推送主为每一动态推送信息设定预期成本价target cpa，由服务器负责控制实际曝光时的出价，达成每个转化行为平摊的成本在推送主预期成本价的1.2倍以内。请继续阅图4a，该成本控制模块24负责对动态推送信息的虚拟成本数据中成本表现不好的动态推送信息进行打压，实现推送主预期成本价控制的目标。
进一步的，该成本控制模块24会在动态推送信息的转化量超过一定数量之后，根据虚拟花销数据(即花销)与转化数据的比值计算出每一动态推送信息的虚拟成本数据，例如计算公式如下：
Figure PCTCN2021086778-appb-000004
该cpa即为虚拟成本数据，该sun(cost)即为虚拟花销数据，该sum(conversion)即为转化数据，通过该虚拟花销数据和转化数据的比值可以计算出每一动态推送信息的虚拟成本数据cpa，该预设虚拟数据可以为预期成本价的1.2倍，当该虚拟成本数据大于预期成本价时，说明成本跑飞，应该减少该成本跑飞的动态推送信息的曝光机会，即可以将该虚拟成本数据大于预设虚拟数据相应的动态推送信息进行冻结，不参与后续的曝光，从而保护推送主的利益，经过该成本控制模块24筛选过的信息可以通过数据流25快速传输至粗排模块26中。
在步骤204中，服务器获取历史动态推送信息的点击率相应的先验分布信息，根据先验分布信息生成每一动态推送信息的点击率相应的第一贝塔分布。
其中，由于历史动态推送信息的点击率具有一个大概范围，因此，在进行动态推送信息的推送之前，对该动态推送信息即可以有平均概率预测，以此，可以获取历史动态推送信息的点击率相应的先验分布信息CTR～beta(α,β)，根据该先验分布信息生成每一动态推送信息的点击率相应的第一贝塔分布，该第一贝塔分布为平均点击率分布。
在步骤205中，服务器统计每一动态推送信息在曝光时生成的点击数据和非点击数据，生成点击率相应的后验分布信息。
其中，服务器统计每一动态推送信息在曝光时生成的点击数据，通过曝光数据减去点击数据得到非点击数据，根据该点击数据和非点击数据生成点击率相应的后验分布信息
Figure PCTCN2021086778-appb-000005
在步骤206中，服务器根据点击率相应的后验分布信息对第一贝塔分布进行调整，得到每一动态推送信息的点击率相应的第一目标贝塔分布。
其中，服务器点击率的后验分布信息
Figure PCTCN2021086778-appb-000006
Figure PCTCN2021086778-appb-000007
对第一贝塔分布的曲线进行调整，得到每一动态推送的点击率相应的第一目标贝塔分布，该第一目标贝塔分布在不断结合用户实际使用过程中的点击习惯，因而该第一目标贝塔分布的会越来越贴合真实的使用情况。
在步骤207中，服务器获取每一动态推送信息的目标转化数据。
其中，在对动态推送信息进行推送的相关技术中，均未考虑到转化数据层面的效果，转化数据为用户点击动态推送信息后实现应用下载或者虚拟充值等推送主预期事件的行为数据，而该转化效果为推送主的核心利益，因此，需要进行相应的 改进。
在一实施方式中，该获取每一动态推送信息的目标转化数据的步骤，可以包括将每一动态推送信息的转化数据进行求和，得到目标转化数据。
在本申请实施例中，由于转化数据积累的速度比较慢，前期可能长时间没有转化数据出现或者转化数据较少，因此，在将转化数据考虑到推送层面的前提是目标转化数据达到一定的数量，以此，服务器可以将每一动态推送信息的转化数据进行求和，得到总的目标转化数据。
在另一实施方式中，该获取每一动态推送信息的目标转化数据的步骤，还可以包括统计动态推送信息的平均转化数据，将所述平均转化数据确定为目标转化数据。
在本申请实施例中，由于转化数据积累的速度比较慢，前期可能长时间没有转化数据出现或者转化数据较少，因此，在将转化数据考虑到推送层面的前提是每一动态推送信息的平均转化数达到一定的数量，例如，当同一动态广告中每一动态推送信息的平均转化数大于1时，满足条件。以此，服务器可以统计所有动态推送信息的转化数据的总和，计算该总和与动态推送信息数量(即创意数量)的比值，得到平均转化数据，将该平均转化数据确定为目标转化数据。
在步骤208中，服务器检测目标转化数据是否小于第一预设阈值。
其中，该第一预设阈值即为界定该目标转化数据是否达到一定数量的临界值，该第一预设阈值的数量可以为动态推送信息的数量，当服务器检测到目标转化数据小于第一预设阈值时，执行步骤209。当服务器检测到目标转化数据不小于第一预设阈值时，执行步骤210。
在步骤209中，服务器根据第一目标贝塔分布获取每一动态推送信息的预测点击率，按照预测点击率由高至低的顺序选取预设数量的目标动态推送信息。
其中，当服务器检测到目标转化数据小于第一预设阈值时，说明转化数据未符合要求，以此，服务器可以分别获取每一动态推送信息相应的第一目标贝塔分布的采样值，以得到每一动态推送信息的预测点击率，按照预测点击率由高至低的顺序择优选取预设数量的目标动态推送信息，实现粗排模块26的一种粗选方式。
在步骤210中，服务器统计每一动态推送信息在点击时生成的转化数据和非转化数据，生成转化率相应的后验分布信息。
其中，当服务器检测到目标转化数据不小于第一预设阈值时，说明转化数据符合要求，可以开始预估每一动态推送信息的转化率，基于汤普森采样原理，统计每一动态推送信息在点击时生成的转化数据和非转化数据，该转化数据为用户点击动态推送信息后完成预期操作的次数数据，该非转化数据为用户点击动态推送信息后未完成预期操作的次数数据，由于数据具有一定的规模，所以可以直接根据该转化数据和非转化数据，生成转化率相应的后验分布信息CVR_post～beta(conversion，click-conversion)，该conversion为转化数据，该click为点击数据。
在步骤211中，服务器根据转化率相应的后验分布信息生成每一动态推送信息的转化率相应的第二目标贝塔分布。
其中，根据转化率的后验分布信息CVR_post～beta(conversion，click-conversion)生成每一动态推送信息的转化率相应的第二目标贝塔分布，该第二目标贝塔分布可以不断结合用户实际使用过程中的转化习惯，因此，该第二目标贝塔分布会随着不断的使用学习越来越贴合于真实的转化使用情况。
在步骤212中，服务器根据第一目标贝塔分布获取每一动态推送信息的预测点击率，根据第二目标贝塔分布获取每一动态推送信息的预测转化率，将每一动态推送信息的预测点击率和预测转化率进行结合，得到结合率，按照结合率由高至低的顺序选取预设数量的动态推送信息。
其中，服务器分别获取每一动态推送信息相应的第一目标贝塔分布的采样值，得到每一动态推送信息的预测点击率，分别获取每一动态推送信息相应的第二目标贝塔分布的采样值获取每一动态推送信息的预测转化率，将该每一动态推送信息的预测点击率和预测转化率进行结合，如将预测点击率和预测转化率相乘结合，得到结合率，按照结合率由高至低的顺序选取预设数量的动态推送信息，实现该预设数量的动态推送信息结合转化数据层面的效果，综合点击率维度和转化率维度进行动态推送信息的选择，实现粗排模块26的另一种粗选方式。
在步骤213中，服务器通过目标点击率预测模型对预设数量的动态推送信息进行预测，得到预设数量的动态推送信息的目标预测点击率，获取每一动态推送信息的目标曝光数据。
其中，请一并参阅图4a和图4b，粗排模块26包括轻量级点击率预估模型(Litectr)和轻量级转化预估模型(Litecvr)，该轻量级点击率预估模型可以获得每一动态推送信息的预测点击率，该轻量级转化预估模型可以获得每一动态推送信息的预测转化率。进而得到广告有效的每千次展示费用广告收益(ecpm)指标，根据该ecpm对所有的广告按照由高至低的顺序进行排序，选出前N个广告，例如，请参照如下公式：
ecpm＝ocpa_bid*Litectr*Litecvr
该ocpa_bid为广告出价，该Litectr为预测点击率，该Litecvr为预测转化率，通过上述公式计算该广告出价、预测点击率和预测转化率的乘积，可以算出每一广告的粗排ecpm指标。
进一步的，由于该前N个广告中包括动态创意广告，该动态创意广告中包含多个创意(即动态推送信息)，因此需要对前N个广告中每一动态创意广告进行广告创意优选，具体广告创意优选的方式参照上述选取预设数量的目标动态推送信息的步骤，在选取预设数量的目标动态推送信息之后送入精排模块27，精排模块27包括目标点击率预测模型(Pctr)和目标转化率预测模型(Pcvr)，通过目标点击率预测模型对预设数量的动态推送信息进行预测，得到预设数量的动态推送信息的目 标预测点击率，通过目标转化率预测模型对预设数量的动态推送信息进行预测，得到预设数量的动态推送信息的目标预测转化率，由于该精排模块27输入Pctr和Pcvr的特征更多，因而该Pctr和Pcvr的预测精度会更高。
在一实施方式中，该获取每一动态推送信息的目标曝光数据的步骤，可以包括将每一动态推送信息的曝光数据进行求和，得到目标曝光数据。
在本申请实施例中，服务器可以将每一动态推送信息的曝光数据进行求和，得到总的目标曝光数据。
在另一实施方式中，该获取每一动态推送信息的目标转化数据的步骤，还可以包括统计动态推送信息的平均曝光数据，将所述平均曝光数据确定为目标曝光数据。
在本申请实施例中，服务器可以统计所有动态推送信息的曝光数据的总和，计算该总和与动态推送信息数量(即创意数量)的比值，得到平均曝光数据，将该平均曝光数据确定为目标曝光数据。
在步骤214中，服务器检测目标曝光数据是否小于第二预设阈值。
其中，由于该目标点击率预测模型的学习需要一定的数据支撑，在曝光前期，即目标曝光数据较少的情况下，该目标点击率预测模型不一定精准，而在曝光后期，即目标曝光数据较多的情况下，该目标点击率预测模型精准，所以需要设定第二预设阈值用于区分当前处于曝光前期还是后期，该第二预设阈值可以为5000或者300等等，当服务器检测到该目标曝光数据小于第二预设阈值时，执行步骤215，当服务器检测到该目标曝光数据不小于第二预设阈值时，执行步骤216。
在步骤215中，服务器将目标预测点击率进行归一化处理，得到预设数量维度的目标预测向量信息，基于目标预测向量信息划分概率区间，随机访问概率区间，将访问的概率区间相应的动态推送信息确定为目标动态推送信息进行推送。
其中，当服务器检测到该目标曝光数据小于第二预设阈值时，说明精排模块27处于曝光前期，预设数量的动态推送信息的曝光有限，Pctr模型存在不精确的地方，此时不能完全根据Pctr模型进行优选，需要通过softmax函数将目标预测点击率进行归一化处理，得到预设数量维度的目标预测向量信息，每一维度的向量元素的概率范围都在(0，1)之间，且所有维度的向量元素的总和为1，该目标预测点击率越大，转化后的概率范围越大，该softmax函数如下：
Figure PCTCN2021086778-appb-000008
该Win_ratei为目标预测向量信息，该e为常数，该pctri为目标预测点击率，该num of creatives为预设数量减1的值。
进一步的，根据该目标预测向量信息中的每一元素的概率将(0，1)的概率区间进行划分，概率越高，分配的概率区间越大，概率越低，分配的概率越小，随机访问该划分后的概率区间，需要说明是，概率区间越大，被访问的几率越大，概率区间越小，被访问的几率越小，将访问的概率区间相应的动态推送信息确定为目标动态推送信息进行推送，即将目标动态推动信息代表该动态创意广告参与后续的 竞价，以此，目标预测点击率低，但是转化率高的动态推送信息同样也可以得到创意推荐，充分的考虑了推送主的利益，使得目标动态推送信息的推送多元化。
在步骤216中，服务器将目标预测点击率最高的动态推送信息确定为目标动态推送信息进行推送。
其中，当服务器检测到该目标曝光数据不小于第二预设阈值时，说明精排模块27处于曝光后期，预设数量的动态推送信息的曝光达到要求，Pctr模型预测精确，可以直接将目标预测点击率最高的动态推送信息确定为目标推送信息进行推送，即将目标动态推动信息代表该动态创意广告参与后续的竞价。
请继续参阅图4b，在精排模块27确定出每一动态创意广告的最优目标动态推送信息为代表后，进一步计算前N个广告的更为准确的目标ecpm指标，根据该目标ecpm指标选出最佳的1至2个广告推送给用户，例如，请参照如下公式：
ecpm＝ocpa_bid*Pctr*Pcvr
该ocpa_bid为广告出价，该Pctr为目标预测点击率，该Litecvr为目标预测转化率，通过上述公式计算该广告出价、目标预测点击率和目标预测转化率，可以算出每一广告的精排ecpm指标，并根据该精排ecpm指标选出最佳的1至2个广告推送给用户。
在一些实施方式中，精排模块27通过目标点击率预测模型对预设数量的动态推送信息进行预测，得到预设数量的动态推送信息的目标预测点击率之后，还可以包括：
(1)通过目标转化率(Pcvr)模型对预设数量的动态推送信息进行预测，得到预设数量的动态推送信息的目标预测转化率，将该目标预测点击率和目标预测转化率进行相乘，得到目标结合率，将每一动态推送信息的曝光数据进行求和，得到目标曝光数据。
(2)服务器检测目标曝光数据是否小于第二预设阈值。
其中，当服务器检测到该目标曝光数据小于第二预设阈值时，执行步骤(3)，当服务器检测到该目标曝光数据不小于第二预设阈值时，执行步骤(4)。
(3)服务器将目标结合率进行归一化处理，得到预设数量维度的结合预测向量信息，基于结合预测向量信息划分概率区间，随机访问概率区间，将访问的概率区间相应的动态推送信息确定为目标动态推送信息进行推送。
(4)服务器将目标结合率最高的动态推送信息确定为目标动态推送信息进行推送。
由于引入目标预测转化率与目标预测点击率进行结合，在精排模块27进一步引入转化层面的目标动态推送信息的择优推荐，使得目标动态推送信息越准确，相同描述部分参照上述，此处不作具体赘述。
由上述可知，本申请实施例通过统计每一动态推送信息的操作信息；基于曝光数据和点击数据生成每一动态推送信息的点击率相应的第一目标贝塔分布；根据第一目标贝塔分布选取预设数量的动态推送信息；获取预设数量的动态推送信息的 目标预测点击率，根据目标预测点击率选取目标动态推送信息进行推送。以此，实时统计每一动态推送信息的操作信息，基于汤普森采样思想生成每一动态推送信息的点击率相应的第一目标贝塔分布，根据第一目标贝塔分布选取预设数量的动态推送信息并获取相应的目标点击率，根据目标预测点击率选取精确的目标动态推送信息进行推送，极大的提升了信息处理的准确率。
进一步的，由于在选取过程中引入转化数据层面对动态推送信息进行筛选，使得目标动态推送信息的推送更符合推送主的要求，提升创意优选的效果，进一步提升了信息处理的准确率。
实施例三、
为便于更好的实施本申请实施例提供的信息处理方法，本申请实施例还提供一种基于上述信息处理方法的装置。其中名词的含义与上述信息处理方法中相同，具体实现细节可以参考方法实施例中的说明。
请参阅图5，图5为本申请实施例提供的信息处理装置的结构示意图，其中该信息处理装置可以包括统计单元301、生成单元302、第一粗排单元303、及精排单元304等。
统计单元301，用于统计历史推送信息的反馈数据，该反馈数据至少包括曝光数据和点击数据。
生成单元302，用于基于该曝光数据和点击数据生成每一历史动态推送信息的点击率相应的第一概率分布。
在一些实施方式中，生成单元302，用于：获取历史动态推送信息的点击率相应的先验分布信息，根据该先验分布信息生成每一动态推送信息的点击率相应的第一贝塔分布；统计每一动态推送信息在曝光时生成的点击数据和非点击数据，生成点击率相应的后验分布信息；根据该点击率相应的后验分布信息对该第一贝塔分布进行调整，得到每一动态推送信息的点击率相应的第一目标贝塔分布。
第一粗排单元303，用于根据该第一概率分布确定待推送的各推送信息的第一预测点击率，根据所述第一预测点击率从所述待推送的各推送信息中选取预设数量的第一推送信息。
在一些实施例中，该第一粗排单元303，用于：根据该第一概率分布获取每一动态推送信息的预测点击率；按照该预测点击率由高至低的顺序选取预设数量的目标动态推送信息。
精排单元304，用于利用预设的目标点击率预测模型获取各第一推送信息的目标预测点击率，根据该目标预测点击率选取目标动态推送信息进行推送。
在一些实施例中，该精排单元304，包括：
预测子单元，用于通过目标点击率预测模型对该预设数量的动态推送信息进行预测，得到该预设数量的动态推送信息的目标预测点击率；
曝光子单元，用于获取每一动态推送信息的目标曝光数据；
结合子单元，用于结合该目标预测点击率和目标曝光数据选取目标动态推送 信息进行推送。
在一些实施例中，该结合子单元，用于：当检测到该目标曝光数据小于第二预设阈值时，将该目标预测点击率进行归一化处理，得到预设数量维度的目标预测向量信息；基于该目标预测向量信息划分概率区间，随机访问该概率区间，将访问的概率区间相应的动态推送信息确定为目标动态推送信息进行推送；当检测到该目标曝光数据不小于第二预设阈值时，将该目标预测点击率最高的动态推送信息确定为目标动态推送信息进行推送。
在一些实施例中，该操作信息还包括转化数据，该装置还包括：
转化单元，用于获取每一动态推送信息的目标转化数据；
该第一粗排单元，用于当检测到该目标转化数据小于第一预设阈值时，执行根据该第一目标贝塔分布选取预设数量的动态推送信息的步骤；
第二粗排单元，用于当检测到该目标转化数据不小于第一预设阈值时，统计每一动态推送信息在点击时生成的转化数据和非转化数据，生成转化率相应的后验分布信息；根据该转化率相应的后验分布信息生成每一动态推送信息的转化率相应的第二目标贝塔分布；结合该第一目标贝塔分布和第二目标贝塔分布选取预设数量的动态推送信息。
在一些实施例中，该第二粗排单元，还用于：检测到该目标转化数据不小于第一预设阈值时，统计每一动态推送信息在点击时生成的转化数据和非转化数据，生成转化率相应的后验分布信息；根据该转化率相应的后验分布信息生成每一动态推送信息的转化率相应的第二目标贝塔分布；根据该第一目标贝塔分布获取每一动态推送信息的预测点击率；根据该第二目标贝塔分布获取每一动态推送信息的预测转化率；将每一动态推送信息的预测点击率和预测转化率进行结合，得到结合率；按照该结合率由高至低的顺序选取预设数量的动态推送信息。
在一些实施例中，该操作信息还包括虚拟花销数据，该装置还包括：
成本控制单元，用于根据该虚拟花销数据和该转化数据计算每一动态推送信息的虚拟成本数据；将该虚拟成本数据大于预设虚拟数据相应的动态推送信息进行冻结。
在一些实施例中，该装置还包括：
更新单元，用于按照预设周期采集动态推送信息的更新信息；根据该更新信息对动态推送信息进行更新操作。
以上各个单元的具体实施可参见前面的实施例，在此不再赘述。
由上述可知，本申请实施例通过统计单元301统计每一动态推送信息的操作信息；生成单元302基于曝光数据和点击数据生成每一动态推送信息的点击率相应的第一目标贝塔分布；第一粗排单元303根据第一目标贝塔分布选取预设数量的动态推送信息；精排单元304获取预设数量的动态推送信息的目标预测点击率，根据目标预测点击率选取目标动态推送信息进行推送。以此，实时统计每一动态推送信息的操作信息，基于汤普森采样思想生成每一动态推送信息的点击率相应的第一目 标贝塔分布，根据第一目标贝塔分布选取预设数量的动态推送信息并获取相应的目标点击率，根据目标预测点击率选取精确的目标动态推送信息进行推送，极大的提升了信息处理的准确率。
实施例四、
本申请实施例还提供一种服务器，如图6所示，其示出了本申请实施例所涉及的服务器的结构示意图，具体来讲：
该服务器可以包括一个或者一个以上处理核心的处理器401、一个或一个以上计算机可读存储介质的存储器402、电源403和输入单元404等部件。本领域技术人员可以理解，图6中示出的服务器结构并不构成对服务器的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。其中：
处理器401是该服务器的控制中心，利用各种接口和线路连接整个服务器的各个部分，通过运行或执行存储在存储器402内的软件程序和/或模块，以及调用存储在存储器402内的数据，执行服务器的各种功能和处理数据，从而对服务器进行整体监控。可选的，处理器401可包括一个或多个处理核心；优选的，处理器401可集成应用处理器和调制解调处理器，其中，应用处理器主要处理操作系统、用户界面和应用程序等，调制解调处理器主要处理无线通信。可以理解的是，上述调制解调处理器也可以不集成到处理器401中。
存储器402可用于存储软件程序以及模块，处理器401通过运行存储在存储器402的软件程序以及模块，从而执行各种功能应用以及数据处理。存储器402可主要包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需的应用程序(比如声音播放功能、图像播放功能等)等；存储数据区可存储根据服务器的使用所创建的数据等。此外，存储器402可以包括高速随机存取存储器，还可以包括非易失性存储器，例如至少一个磁盘存储器件、闪存器件、或其他易失性固态存储器件。相应地，存储器402还可以包括存储器控制器，以提供处理器401对存储器402的访问。
服务器还包括给各个部件供电的电源403，优选的，电源403可以通过电源管理系统与处理器401逻辑相连，从而通过电源管理系统实现管理充电、放电、以及功耗管理等功能。电源403还可以包括一个或一个以上的直流或交流电源、再充电系统、电源故障检测电路、电源转换器或者逆变器、电源状态指示器等任意组件。
该服务器还可包括输入单元404，该输入单元404可用于接收输入的数字或字符信息，以及产生与用户设置以及功能控制有关的键盘、鼠标、操作杆、光学或者轨迹球信号输入。
尽管未示出，服务器还可以包括显示单元等，在此不再赘述。具体在本实施例中，服务器中的处理器401会按照如下的指令，将一个或一个以上的应用程序的进程对应的可执行文件加载到存储器402中，并由处理器401来运行存储在存储器402中的应用程序，从而实现各种功能，如下：
统计每一动态推送信息的操作信息，该操作信息至少包括曝光数据和点击数 据；基于该曝光数据和点击数据生成每一动态推送信息的点击率相应的第一目标贝塔分布；根据该第一目标贝塔分布选取预设数量的动态推送信息；获取该预设数量的动态推送信息的目标预测点击率，根据该目标预测点击率选取目标动态推送信息进行推送。
在上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述的部分，可以参见上文针对信息处理方法的详细描述，此处不再赘述。
由上述可知，本申请实施例的服务器可以通过统计每一动态推送信息的操作信息；基于曝光数据和点击数据生成每一动态推送信息的点击率相应的第一目标贝塔分布；根据第一目标贝塔分布选取预设数量的动态推送信息；获取预设数量的动态推送信息的目标预测点击率，根据目标预测点击率选取目标动态推送信息进行推送。以此，实时统计每一动态推送信息的操作信息，基于汤普森采样思想生成每一动态推送信息的点击率相应的第一目标贝塔分布，根据第一目标贝塔分布选取预设数量的动态推送信息并获取相应的目标点击率，根据目标预测点击率选取精确的目标动态推送信息进行推送，极大的提升了信息处理的准确率。
实施例五、
本领域普通技术人员可以理解，上述实施例的各种方法中的全部或部分步骤可以通过指令来完成，或通过指令控制相关的硬件来完成，该指令可以存储于一计算机可读存储介质中，并由处理器进行加载和执行。
为此，本申请实施例提供一种计算机可读存储介质，其中存储有多条指令，该指令能够被处理器进行加载，以执行本申请实施例所提供的任一种信息处理方法中的步骤。例如，该指令可以执行如下步骤：
统计每一动态推送信息的操作信息，该操作信息至少包括曝光数据和点击数据；基于该曝光数据和点击数据生成每一动态推送信息的点击率相应的第一目标贝塔分布；根据该第一目标贝塔分布选取预设数量的动态推送信息；获取该预设数量的动态推送信息的目标预测点击率，根据该目标预测点击率选取目标动态推送信息进行推送。
其中，该计算机可读存储介质可以包括：只读存储器(ROM，Read Only Memory)、随机存取记忆体(RAM，Random Access Memory)、磁盘或光盘等。
由于该计算机可读存储介质中所存储的指令，可以执行本申请实施例所提供的任一种信息处理方法中的步骤，因此，可以实现本申请实施例所提供的任一种信息处理方法所能实现的有益效果，详见前面的实施例，在此不再赘述。
以上对本申请实施例所提供的一种信息处理方法、装置及计算机可读存储介质进行了详细介绍，本文中应用了具体个例对本申请的原理及实施方式进行了阐述，以上实施例的说明只是用于帮助理解本申请的方法及其核心思想；同时，对于本领域的技术人员，依据本申请的思想，在具体实施方式及应用范围上均会有改变之处，综上所述，本说明书内容不应理解为对本申请的限制。。

Claims (19)

  1. 一种信息处理方法，应用于服务器，其特征在于，包括：
    统计历史推送信息的反馈数据，所述反馈数据至少包括曝光数据和点击数据；
    基于所述曝光数据和点击数据生成所述历史推送信息中每一推送信息的点击率的第一概率分布；
    根据所述第一概率分布确定待推送的各推送信息的第一预测点击率，根据所述第一预测点击率从所述待推送的各推送信息中选取预设数量的第一推送信息；
    利用预设的目标点击率预测模型获取所述第一推送信息中每个第一推送信息的目标预测点击率；
    根据所述目标预测点击率从所述第一推送信息中选取目标推送信息进行推送。
  2. 根据权利要求1所述的信息处理方法，其特征在于，所述基于所述曝光数据和点击数据生成所述历史推送信息中每一推送信息的点击率相应的第一概率分布的步骤，包括：
    获取历史推送信息的点击率相应的先验分布信息，根据所述先验分布信息生成所述每一推送信息的点击率相应的第一概率分布；
    统计所述每一推送信息在曝光时生成的点击数据和非点击数据，生成点击率相应的后验分布信息；
    根据所述点击率相应的后验分布信息对所述第一概率分布进行调整，得到所述每一推送信息的点击率相应的第一概率分布。
  3. 根据权利要求1所述的信息处理方法，其特征在于，所述根据所述第一概率分布选取预设数量的第一推送信息的步骤，包括：
    按照所述第一预测点击率由高至低的顺序选取预设数量的第一推送信息。
  4. 根据权利要求1所述的信息处理方法，其特征在于，所述反馈信息还包括转化数据，所述方法还包括：
    所述根据所述第一概率分布选取预设数量的第一推送信息之前，获取所述每一推送信息的目标转化数据；
    当检测到所述目标转化数据小于第一预设阈值时，执行根据所述第一概率分布选取预设数量的第一推送信息的步骤；
    当检测到所述目标转化数据不小于第一预设阈值时，统计所述每一推送信息在点击时生成的转化数据和非转化数据，生成转化率相应的后验分布信息；
    根据所述转化率相应的后验分布信息生成所述每一推送信息的转化率相应的第二概率分布；
    结合所述第一概率分布和第二概率分布选取预设数量的第一推送信息。
  5. 根据权利要求4所述的信息处理方法，其特征在于，所述结合所述第一概率分布和第二概率分布选取预设数量的第一推送信息的步骤，包括：
    根据所述第一概率分布获取所述每一推送信息的预测点击率；
    根据所述第二概率分布获取所述每一推送信息的预测转化率；
    将所述每一推送信息的预测点击率和预测转化率进行结合，得到结合率；
    按照所述结合率由高至低的顺序选取预设数量的第一推送信息。
  6. 根据权利要求4所述的信息处理方法，其特征在于，所述反馈数据还包括虚拟花销数据，所述方法还包括：
    根据所述虚拟花销数据和所述转化数据计算所述每一推送信息的虚拟成本数据；
    将所述虚拟成本数据大于预设虚拟数据相应的推送信息进行冻结。
  7. 根据权利要求1至6任一项所述的信息处理方法，其特征在于，根据所述目标预测点击率从所述第一推送信息中选取目标推送信息进行推送的步骤，包括：
    获取第一推送信息中每一第一推送信息的目标曝光数据；
    结合所述目标预测点击率和目标曝光数据选取目标推送信息进行推送。
  8. 根据权利要求7所述的信息处理方法，其特征在于，所述结合所述目标预测点击率和目标曝光数据选取目标推送信息进行推送的步骤，包括：
    当检测到所述目标曝光数据小于第二预设阈值时，将所述目标预测点击率进行归一化处理，得到预设数量维度的目标预测向量信息；
    基于所述目标预测向量信息划分概率区间，随机访问所述概率区间，将访问的概率区间相应的推送信息确定为目标推送信息进行推送；
    当检测到所述目标曝光数据不小于第二预设阈值时，将所述目标预测点击率最高的推送信息确定为目标推送信息进行推送。
  9. 根据权利要求1至6任一项所述的信息处理方法，其特征在于，所述方法还包括：
    按照预设周期采集推送信息的更新信息；
    根据所述更新信息对推送信息进行更新操作。
  10. 一种信息处理装置，其特征在于，包括：
    统计单元，用于统计历史推送信息的反馈数据，所述反馈数据至少包括曝光数据和点击数据；
    生成单元，用于基于所述曝光数据和点击数据生成所述历史推送信息中每一推送信息的点击率相应的第一概率分布；
    第一粗排单元，用于根据所述第一概率分布确定待推送的各推送信息的第一预测点击率，根据所述第一预测点击率从所述待推送的各推送信息中选取预设数量的第一推送信息，利用预设的目标点击率预测模型获取所述第选取预设数量的第一推送信息；
    精排单元，用于利用预设的目标点击率预测模型获取所述第一推送信息中每个第一推送信息的目标预测点击率，根据所述目标预测点击率从所述第一推送信息中选取目标推送信息进行推送。
  11. 根据权利要求10所述的信息处理装置，其特征在于，所述生成单元，用于：
    获取历史推送信息的点击率相应的先验分布信息，根据所述先验分布信息生成 所述每一推送信息的点击率相应的第一概率分布；
    统计所述每一推送信息在曝光时生成的点击数据和非点击数据，生成点击率相应的后验分布信息；
    根据所述点击率相应的后验分布信息对所述第一概率分布进行调整，得到所述每一推送信息的点击率相应的第一概率分布。
  12. 根据权利要求10所述的信息处理装置，其特征在于，所述第一粗排单元，用于：
    按照所述第一预测点击率由高至低的顺序选取预设数量的第一推送信息。
  13. 根据权利要求10所述的信息处理装置，其特征在于，所述反馈数据还包括转化数据，所述装置还包括：
    转化单元，用于获取所述每一推送信息的目标转化数据；
    所述第一粗排单元，用于当检测到所述目标转化数据小于第一预设阈值时，执行根据所述第一概率分布选取预设数量的第一推送信息的步骤；
    第二粗排单元，用于当检测到所述目标转化数据不小于第一预设阈值时，统计所述每一推送信息在点击时生成的转化数据和非转化数据，生成转化率相应的后验分布信息；
    根据所述转化率相应的后验分布信息生成所述每一推送信息的转化率相应的第二概率分布；
    结合所述第一概率分布和第二概率分布选取预设数量的第一推送信息。
  14. 根据权利要求13所述的信息处理装置，其特征在于，所述第二粗排单元，还用于：
    检测到所述目标转化数据不小于第一预设阈值时，统计每一推送信息在点击时生成的转化数据和非转化数据，生成转化率相应的后验分布信息；
    根据所述转化率相应的后验分布信息生成每一推送信息的转化率相应的第二概率分布；
    根据所述第一概率分布获取所述每一推送信息的预测点击率；
    根据所述第二概率分布获取所述每一推送信息的预测转化率；
    将每一推送信息的预测点击率和预测转化率进行结合，得到结合率；
    按照所述结合率由高至低的顺序选取预设数量的第一推送信息。
  15. 根据权利要求10至14任一项所述的信息处理装置，其特征在于，所述精排单元包括：
    曝光子单元，用于获取第一推送信息中每一第一推送信息的目标曝光数据；
    结合子单元，用于结合所述目标预测点击率和目标曝光数据选取目标推送信息进行推送。
  16. 根据权利要求15所述的信息处理装置，其特征在于，所述结合子单元用于：
    当检测到所述目标曝光数据小于第二预设阈值时，将所述目标预测点击率进行归一化处理，得到预设数量维度的目标预测向量信息；
    基于所述目标预测向量信息划分概率区间，随机访问所述概率区间，将访问的概率区间相应的推送信息确定为目标推送信息进行推送；
    当检测到所述目标曝光数据不小于第二预设阈值时，将所述目标预测点击率最高的推送信息确定为目标推送信息进行推送。
  17. 根据权利要求10所述的信息处理装置，其特征在于，所述操作信息还包括虚拟花销数据，所述装置还包括：
    成本控制单元，用于根据所述虚拟花销数据和所述转化数据计算每一动态推送信息的虚拟成本数据；
    将所述虚拟成本数据大于预设虚拟数据相应的动态推送信息进行冻结。
  18. 根据权利要求10所述的信息处理装置，其特征在于，进一步包括：
    更新单元，用于按照预设周期采集动态推送信息的更新信息；
    根据所述更新信息对动态推送信息进行更新操作。
  19. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质存储有多条指令，所述指令适于处理器进行加载，以执行权利要求1至9任一项所述的信息处理方法中的步骤。
PCT/CN2021/086778 2020-05-19 2021-04-13 一种信息处理方法、装置及计算机可读存储介质 WO2021233015A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/727,168 US20220245495A1 (en) 2020-05-19 2022-04-22 Information processing method and apparatus, and computer-readable storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010427162.9 2020-05-19
CN202010427162.9A CN113688305A (zh) 2020-05-19 2020-05-19 一种信息处理方法、装置及计算机可读存储介质

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/727,168 Continuation US20220245495A1 (en) 2020-05-19 2022-04-22 Information processing method and apparatus, and computer-readable storage medium

Publications (1)

Publication Number Publication Date
WO2021233015A1 true WO2021233015A1 (zh) 2021-11-25

Family

ID=78576364

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/086778 WO2021233015A1 (zh) 2020-05-19 2021-04-13 一种信息处理方法、装置及计算机可读存储介质

Country Status (3)

Country Link
US (1) US20220245495A1 (zh)
CN (1) CN113688305A (zh)
WO (1) WO2021233015A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114430504A (zh) * 2022-01-28 2022-05-03 腾讯科技（深圳）有限公司 一种媒体内容的推荐方法以及相关装置
CN114707097A (zh) * 2022-05-31 2022-07-05 每日互动股份有限公司 一种获取目标消息流量的数据处理系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106294457A (zh) * 2015-05-29 2017-01-04 腾讯科技（深圳）有限公司 网络信息推送方法及装置
CN110008399A (zh) * 2019-01-30 2019-07-12 阿里巴巴集团控股有限公司 一种推荐模型的训练方法及装置、一种推荐方法及装置
CN110347781A (zh) * 2019-07-18 2019-10-18 腾讯科技（深圳）有限公司 文章倒排方法、文章推荐方法、装置、设备及存储介质
CN110378434A (zh) * 2019-07-24 2019-10-25 腾讯科技（深圳）有限公司 点击率预测模型的训练方法、推荐方法、装置及电子设备
US20200142944A1 (en) * 2014-02-07 2020-05-07 Google Llc Systems and methods for automatically creating content modification scheme

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200142944A1 (en) * 2014-02-07 2020-05-07 Google Llc Systems and methods for automatically creating content modification scheme
CN106294457A (zh) * 2015-05-29 2017-01-04 腾讯科技（深圳）有限公司 网络信息推送方法及装置
CN110008399A (zh) * 2019-01-30 2019-07-12 阿里巴巴集团控股有限公司 一种推荐模型的训练方法及装置、一种推荐方法及装置
CN110347781A (zh) * 2019-07-18 2019-10-18 腾讯科技（深圳）有限公司 文章倒排方法、文章推荐方法、装置、设备及存储介质
CN110378434A (zh) * 2019-07-24 2019-10-25 腾讯科技（深圳）有限公司 点击率预测模型的训练方法、推荐方法、装置及电子设备

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114430504A (zh) * 2022-01-28 2022-05-03 腾讯科技（深圳）有限公司 一种媒体内容的推荐方法以及相关装置
CN114430504B (zh) * 2022-01-28 2023-03-10 腾讯科技（深圳）有限公司 一种媒体内容的推荐方法以及相关装置
CN114707097A (zh) * 2022-05-31 2022-07-05 每日互动股份有限公司 一种获取目标消息流量的数据处理系统

Also Published As

Publication number Publication date
CN113688305A (zh) 2021-11-23
US20220245495A1 (en) 2022-08-04

Similar Documents

Publication Publication Date Title
CN106372249B (zh) 一种点击率预估方法、装置及电子设备
US8893012B1 (en) Visual indicator based on relative rating of content item
WO2021233015A1 (zh) 一种信息处理方法、装置及计算机可读存储介质
CN105677881A (zh) 一种信息推荐方法、装置及服务器
Lommatzsch et al. Real-time recommendations for user-item streams
CN111523939B (zh) 一种推广内容的投放方法、装置、存储介质和电子设备
WO2016174678A1 (en) Auto-expanding campaign optimization
CN111582955A (zh) 推广信息展示方法、装置、电子设备以及存储介质
Zhou et al. Statistical study of view preferences for online videos with cross-platform information
CN110213209B (zh) 一种推送信息点击的作弊检测方法、装置及存储介质
US20240107125A1 (en) System and method for modelling access requests to multi-channel content sharing platforms
CN111569412B (zh) 一种云游戏资源调度的方法和装置
CN112905879A (zh) 推荐方法、装置、服务器以及存储介质
WO2022247671A1 (zh) 用户召回方法、装置、计算机设备和存储介质
CN110245969B (zh) 推广信息处理方法、装置、电子设备及计算机可读介质
CN113572802B (zh) 对象流量的控制方法、装置及介质
Rokicki et al. Just in time: Controlling temporal performance in crowdsourcing competitions
CN116069959A (zh) 确定展示数据的方法和装置
CN113886732A (zh) 一种信息推荐方法
CN111461188A (zh) 一种目标业务控制方法、装置、计算设备及存储介质
CN116011893B (zh) 基于垂直领域的市区融媒评价方法、装置及电子设备
CN111917809B (zh) 多媒体数据推送方法及其装置
CN113726722B (zh) 多媒体信息推荐方法、装置、服务器及存储介质
CN113516494A (zh) 信息推送方法、装置、电子设备及计算机可读介质
Kheradmand et al. Price and QoS competition in cloud market by using cellular learning automata

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21808511

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 180423)

122 Ep: pct application non-entry in european phase

Ref document number: 21808511

Country of ref document: EP

Kind code of ref document: A1