Push information screening method and device, computer device, and storage medium

Abstract

The application relates to a push information screening method and device, computer equipment and a storage medium. The method comprises the following steps: screening coarse selection push information matched with a user identifier based on user information corresponding to the user identifier; predicting a predicted conversion rate of the user identifier for the coarse selection push information; obtaining a crowd optimization queue corresponding to the coarse selection push information; the crowd optimization queue stores historical predicted conversion rates corresponding to the coarse selection push information in a first-in-first-out manner; determining a sorting result of the predicted conversion rate in the crowd optimization queue; and screening target push information from the coarse selection push information based on the sorting result and the predicted conversion rate corresponding to the coarse selection push information. The method can obtain more accurate push information optimization results.

Description

TECHNICAL FIELD

[0001] The present application relates to the network technical field, in particular to a push information screening method and device, computer equipment and storage medium. BACKGROUND

[0002] An advertisement is information delivered to the public through a certain form of media. Previously, advertisers usually deliver their commercial information to the public with the help of traditional media (such as newspapers, magazines, radio, television, etc.). In recent years, with the rapid development of the Internet, Internet advertising has become an important part of modern marketing strategies because of its fast speed and good effect.

[0003] In the related art, an advertiser (traffic demand party) realizes recommendation of information to a terminal user through a media owner (traffic supply party). The realization manner is that the advertiser puts information to the media owner, predicts a recommendation index of each information based on a prediction algorithm of the media owner, realizes screening of the information based on the recommendation index, and pushes the screened optimal information to a corresponding user.

[0004] However, the traditional prediction algorithm refers to less dimensions when modeling, and usually only considers real-time related data of the information, which leads to a large deviation between the prediction result and the actual situation, and further leads to inaccurate results of information optimization. SUMMARY

[0005] Therefore, it is necessary to provide a push information screening method, device, computer equipment and storage medium capable of improving the accuracy of push information optimization in view of the above technical problems.

[0006] A push information screening method, the method comprising:

[0007] Screening coarse selection push information matched with the user identifier based on user information corresponding to the user identifier;

[0008] Predicting a predicted conversion rate of the user identifier for the coarse selection push information;

[0009] Obtaining a crowd optimization queue corresponding to the coarse selection push information; the crowd optimization queue stores historical predicted conversion rates corresponding to the coarse selection push information in a first-in-first-out manner;

[0010] Determining a ranking result of the predicted conversion rate in the crowd optimization queue;

[0011] Screening target push information from the coarse selection push information based on the ranking result and the predicted conversion rate corresponding to the coarse selection push information.

[0012] A push information screening device, the device comprising:

[0013] The coarse selection push information screening module is configured to screen coarse selection push information matched with the user identifier based on user information corresponding to the user identifier.

[0014] The conversion rate prediction module is configured to predict a predicted conversion rate of the user identifier for the coarse selection push information.

[0015] The crowd selection queue obtaining module is configured to obtain a crowd selection queue corresponding to the coarse selection push information. The crowd selection queue stores historical predicted conversion rates corresponding to the coarse selection push information in a first-in-first-out manner.

[0016] The sorting result determining module is configured to determine a sorting result of the predicted conversion rate in the crowd selection queue.

[0017] The target push information screening module is configured to screen target push information from the coarse selection push information based on the sorting result and the predicted conversion rate corresponding to the coarse selection push information.

[0018] In an embodiment, the apparatus further includes a crowd selection queue maintaining module configured to: after the predicted conversion rate is predicted, add the predicted conversion rate to the tail of the crowd selection queue corresponding to the coarse selection push information; and remove a historical predicted conversion rate at the head of the crowd selection queue.

[0019] In an embodiment, the sorting result determining module is further configured to: obtain a bucket sorting result after the crowd selection queue is divided into buckets; each bucket in the bucket sorting result is sorted according to a corresponding conversion rate interval; and determine the sorting result of the predicted conversion rate in the crowd selection queue based on a target position of a target bucket corresponding to a conversion rate interval matched with the predicted conversion rate in the bucket sorting result.

[0020] In an embodiment, the sorting result determining module is further configured to: determine a number of buckets and a conversion rate interval corresponding to each bucket for a bucketing operation based on historical predicted conversion rates stored in the crowd selection queue; divide the crowd selection queue into the buckets according to conversion rate intervals to which the historical predicted conversion rates and the predicted conversion rate respectively belong, to determine the buckets respectively corresponding to the historical predicted conversion rates and the predicted conversion rate; and sort each bucket according to a size of the conversion rate interval corresponding to each bucket, to obtain a bucket sorting result corresponding to the crowd selection queue.

[0021] In an embodiment, the sorting result determining module is further configured to: determine a sorting position of the predicted conversion rate in the crowd selection queue as a target position of a target bucket corresponding to a conversion rate interval matched with the predicted conversion rate in the bucket sorting result, to obtain the sorting result of the predicted conversion rate in the crowd selection queue.

[0022] In an embodiment, the sorting result determining module is further configured to: determine a target bucket corresponding to the predicted conversion rate matching conversion rate interval in the bucket sorting result, a target position; perform in-bucket sorting on conversion rate values falling into the target bucket to obtain an in-bucket sorting result; and determine the sorting result of the predicted conversion rate in the population preference queue based on the target position and the in-bucket sorting result.

[0023] In an embodiment, the target push information screening module is further configured to: calculate a push score of the rough selection push information based on the sorting result and the predicted conversion rate corresponding to the rough selection push information, respectively; and screen target push information from the rough selection push information based on the push score.

[0024] In an embodiment, the target push information screening module is further configured to: calculate an expected value of the rough selection push information based on the predicted conversion rate of the rough selection push information, respectively; calculate a population quality score of the rough selection push information based on the sorting result and the expected value of the rough selection push information, respectively; obtain an information quality score of the rough selection push information; and calculate a push score of the rough selection push information based on the expected value, the population quality score, and the information quality score of the rough selection push information, respectively.

[0025] In an embodiment, the target push information screening module is further configured to: determine a push mode currently adopted; and calculate an expected value of the rough selection push information based on the predicted conversion rate of the rough selection push information according to a determination manner of the expected value corresponding to the push mode.

[0026] In an embodiment, the target push information screening module is further configured to: obtain a basic resource value corresponding to the rough selection push information; predict a predicted click rate of the user identifier for the rough selection push information; and calculate an expected value of the rough selection push information based on the basic resource value, the predicted conversion rate, and the predicted click rate of the rough selection push information, respectively.

[0027] In an embodiment, the target push information screening module is further configured to: calculate a population quality coefficient of the rough selection push information based on the sorting result of the rough selection push information, respectively; and calculate a population quality score of the rough selection push information based on the population quality coefficient and the expected value of the rough selection push information, respectively.

[0028] In an embodiment, the target push information screening module is further configured to: acquire mapping parameters corresponding to the data amount of the historical predicted conversion rate stored in the crowd selection queue; and calculate a crowd quality coefficient of the rough selection push information based on the sorting result of the rough selection push information and the mapping parameters.

[0029] A computer device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the following steps when executing the computer program:

[0030] screening rough selection push information matching the user identifier based on user information corresponding to the user identifier;

[0031] predicting a predicted conversion rate of the user identifier for the rough selection push information;

[0032] acquiring a crowd selection queue corresponding to the rough selection push information; the crowd selection queue stores historical predicted conversion rates corresponding to the rough selection push information in a first-in-first-out manner;

[0033] determining a sorting result of the predicted conversion rate in the crowd selection queue;

[0034] screening target push information from the rough selection push information based on the sorting result and the predicted conversion rate corresponding to the rough selection push information.

[0035] A computer readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the following steps:

[0036] screening rough selection push information matching the user identifier based on user information corresponding to the user identifier;

[0037] predicting a predicted conversion rate of the user identifier for the rough selection push information;

[0038] acquiring a crowd selection queue corresponding to the rough selection push information; the crowd selection queue stores historical predicted conversion rates corresponding to the rough selection push information in a first-in-first-out manner;

[0039] determining a sorting result of the predicted conversion rate in the crowd selection queue;

[0040] screening target push information from the rough selection push information based on the sorting result and the predicted conversion rate corresponding to the rough selection push information.

[0041] A computer program, comprising computer instructions stored in a computer readable storage medium, a processor of a computer device reads the computer instructions from the computer readable storage medium, the processor executes the computer instructions, so that the computer device executes the following steps:

[0042] Filtering the rough selection push information matched with the user identifier based on the user information corresponding to the user identifier;

[0043] Predicting the predicted conversion rate of the user identifier for the rough selection push information;

[0044] Obtaining a crowd selection queue corresponding to the rough selection push information; the crowd selection queue stores the historical predicted conversion rate corresponding to the rough selection push information in a first-in-first-out manner;

[0045] Determining the ranking result of the predicted conversion rate in the crowd selection queue;

[0046] Filtering the target push information from the rough selection push information based on the ranking result and the predicted conversion rate corresponding to the rough selection push information.

[0047] The push information filtering method, device, computer device and storage medium, filter the rough selection push information matched with the user identifier based on the user information corresponding to the user identifier; predict the predicted conversion rate of the user identifier for the rough selection push information; obtain a crowd selection queue corresponding to the rough selection push information; the crowd selection queue stores the historical predicted conversion rate corresponding to the rough selection push information in a first-in-first-out manner; determine the ranking result of the predicted conversion rate in the crowd selection queue; filter the target push information from the rough selection push information based on the ranking result and the predicted conversion rate corresponding to the rough selection push information, so as to model based on the real-time related data and the historical related data of the dynamic change of information and other multi-dimensions, and then finally obtain more accurate push information optimization result. Since the ranking result of the predicted conversion rate in the crowd selection queue represents the ranking of the predicted conversion rate of the current user in the historical crowd, the target push information is filtered out by considering the predicted conversion rate and the ranking of the current user in the crowd, so that the target push information filtered out is the information with the highest push value, especially when the push information is an advertisement, it can bring better benefits to the advertiser. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 The application environment diagram of the push information filtering method in one embodiment;

[0049] Figure 2 The flowchart of the push information filtering method in one embodiment;

[0050] Figure 3 A schematic diagram of bucket sorting of a group selection queue in one embodiment;

[0051] Figure 4 A flowchart of a push information screening method in another embodiment;

[0052] Figure 5 A schematic diagram of an advertisement push system in one embodiment;

[0053] Figure 6 A schematic diagram of total advertisement revenue in one embodiment;

[0054] Figure 7 A schematic diagram of advertisement revenue in the gaming industry in one embodiment;

[0055] Figure 8 A schematic diagram of advertisement revenue in the education industry in one embodiment;

[0056] Figure 9 A block diagram of a push information screening device in one embodiment;

[0057] Figure 10 An internal structure diagram of a computer device in one embodiment;

[0058] Figure 11 An internal structure diagram of a computer device in one embodiment. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0060] The push information screening method provided by the present application can be implemented based on cloud technology. Cloud technology refers to a series of resources such as hardware, software, network, etc. unified in a wide area network or local area network to realize data calculation, storage, processing and sharing. Cloud technology is a kind of hosting technology based on network technology, information technology, integration technology, management platform technology, application technology, etc. applied in cloud computing business model, which can form a resource pool, and be used on demand, flexibly and conveniently. Cloud computing technology will become an important support. The background service of a technical network system needs a large amount of computing and storage resources, such as video websites, picture websites and more portal websites. With the high development and application of the Internet industry, in the future, every item may have its own identification mark and needs to be transmitted to the background system for logical processing. Different levels of data will be processed separately, and various industry data needs strong system support, which can only be realized through cloud computing.

[0061] Further, the push information screening method provided by the application can be implemented based on big data processing in cloud computing. Big data refers to a data set that cannot be captured, managed and processed within a certain time range by conventional software tools, and is a massive, high-growth and diversified information asset that requires new processing modes to have stronger decision-making, insight discovery and process optimization capabilities. With the advent of the cloud era, big data has attracted more and more attention. Big data requires special technology to effectively process large amounts of data over time. Technologies suitable for big data include large-scale parallel processing databases, data mining, distributed file systems, distributed databases, cloud computing platforms, the Internet and scalable storage systems.

[0062] The push information screening method provided by the application can be applied to an application environment as shown in Figure 1 . The terminal 102 communicates with the server 104 through a network. The push information screening method provided by the embodiments of the application can be executed by the terminal 102 and the conference server 104 in cooperation, and can also be executed by the terminal 102 or the conference server 104 alone. Taking the execution on the server 104 alone as an example, the server 104 screens the rough selection push information matched with the user identifier based on the user information corresponding to the user identifier; predicts the predicted conversion rate of the user identifier for the rough selection push information; obtains a people selection queue corresponding to the rough selection push information; the people selection queue stores the historical predicted conversion rate corresponding to the rough selection push information in a first-in-first-out manner; determines the sorting result of the predicted conversion rate in the people selection queue; and based on the sorting result and the predicted conversion rate corresponding to the rough selection push information, the target push information is screened from the rough selection push information. The terminal 102 can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers and portable wearable devices. The server 104 can be a standalone physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs, and basic cloud computing services such as big data and artificial intelligence platforms. The terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The terminal and the conference server can be directly or indirectly connected through wired or wireless communication, which is not limited in the application.

[0063] In one embodiment, as shown in Figure 2 , a push information screening method is provided. Taking the server 104 in Figure 1 as an example, the method includes the following steps:

[0064] Step S202, based on the user information corresponding to the user identifier, screening the rough selection push information matched with the user identifier.

[0065] The user identifier is used to uniquely identify a user, and the user identifier can be a username for logging into an application or a terminal hardware identifier or an IP address. The user information is information related to the user corresponding to the user identifier, and specifically includes the user's search information, the user's personal basic information, the user's interests, the user's terminal information, etc. The user's search information can be the user's current search keyword information. The user's personal basic information can be the user's gender, age, region, work, etc. The user's interests can be at least one of the information based on the user's browsing log, the user's shopping record, or the content followed by the user's social account. The extracted interest label for marking the user's interests, and the user's terminal information can be the user terminal's operating system information, the user's current network connection mode, etc. The rough selection push information is the information that can be pushed and is preliminarily screened based on the user information. The information that can be pushed includes media information such as videos, news, articles, and advertisements.

[0066] Specifically, the server has pre-stored push information, and each push information has a corresponding information label. The information label is used to describe the category or the push range of the push information. For example, when the push information is an advertisement, the information label can be the category, price, and target population of the commodity. When the push information is a video, the information label can be the name of the video and the keyword of the video content. The user triggers an information push request through the terminal used by the user. The server receives the information push request, obtains the corresponding user information according to the user identifier carried in the push request, and screens the rough selection push information matched with the user identifier based on the matching degree of the user information and the pre-stored push information. Specifically, the matching degree of the user information and the information label can be calculated, the push information can be sorted according to the matching degree, and the push information that meets the selection condition after sorting can be determined as the rough selection push information matched with the user identifier. The selection condition can be to select a certain number of push information with high rankings.

[0067] Taking the information that can be pushed as an advertisement as an example, the server's advertisement pool has pre-stored a large number of advertisements belonging to different advertisers. Each advertisement has a corresponding advertisement label, which is used to describe the category or the delivery range of the advertisement. When the user needs to open a web page with an advertisement position, the terminal will send an information push request to the server. The server receives the information push request, obtains the user information according to the user identifier carried in the information push request, calculates the matching degree of the user information and the advertisement label, sorts the advertisements according to the matching degree, and determines the advertisements that meet the selection condition after sorting as the rough selection advertisements matched with the user identifier. For example, the top 500 advertisements are determined as the rough selection advertisements.

[0068] In an embodiment, after obtaining the user information corresponding to the user identifier, the server inputs the user information and the information tag of the pre-stored push information into a coarse ranking model, respectively extracts features of the user information and the pre-stored push information through a convolution layer in the coarse ranking model, obtains a user description feature vector and a push information feature vector, calculates a coarse ranking score of each push information based on the user description feature vector and the push information feature vector, ranks the push information based on the coarse ranking score, and obtains a coarse ranking result, and further screens the coarse selection push information matched with the user identifier based on the coarse ranking result.

[0069] The coarse ranking model can be a pre-trained machine learning model, and the feature extraction part of the coarse ranking model is a double-tower structure, which respectively takes the user information and the pre-stored push information as input sources to extract the user description feature vector and the push information feature vector.

[0070] Specifically, after extracting the user description feature vector and the push information feature vector, the coarse ranking model performs inner product calculation on the user description feature vector and the push information feature vector to obtain the coarse ranking score of each push information.

[0071] In an embodiment, the information push request received by the server also carries information display position information, and the server can also screen the coarse selection push information matched with the user identifier and the information display position from the pre-stored push information according to the user information and the information display position information. The information display position is used to display the target push information pushed by the server, and the information display position information includes the position and size in the page. When the push information is an advertisement, the information display position can also be called an advertisement position.

[0072] Step S204: predicting a predicted conversion rate of the user identifier for the coarse selection push information.

[0073] Conversion Rate (CVR) is the ratio of the number of times of completing a conversion target to the total number of clicks of the promotion information in a statistical period. The promotion information is the target push information that is finally pushed and displayed to the user after the rough selection push information is finely screened. The conversion target is also called a conversion target page or a target page, which refers to the task that the user completes after the target push information is displayed by the promoter, such as registration, order placement, payment, and the like. Conversion refers to the completion of the conversion target expected by the promoter by the user, for example, the user stays on the website for a certain period of time, registers or submits an order on the website, the user leaves a message through the website or uses the online instant messaging tool of the website for consultation, the user actually makes a payment, and the like. The predicted conversion rate (pCVR) is an estimated value of the conversion rate of each pushable information before the information is pushed, which is used to estimate the probability of generating a conversion behavior corresponding to the conversion target after the push information is pushed to the user and clicked by the user.

[0074] Specifically, the server obtains the historical conversion rate of each rough selection push information and the user information of the historical user corresponding to the historical conversion rate, and determines the association relationship between the user information and the conversion rate based on the user information of the historical user, the information label of the push information, and the historical conversion rate of the push information. After obtaining the user information corresponding to the user identifier and the rough selection push information matched with the user identifier, the predicted conversion rate of each rough selection push information is predicted based on the association relationship between the user information and the conversion rate. The historical conversion rate is the historical conversion rate calculated based on the collected conversion data after the current predicted rough selection push information is pushed and displayed as the target push information in the historical period. For example, the current predicted rough selection push information is pushed and displayed 50,000 times in the past month, of which the number of times of conversion is 5,000, and the historical conversion rate of the rough selection push information is 10%.

[0075] In one embodiment, the server can also obtain the historical conversion rate of each rough selection push information, the historical display position information, and the user information of the historical user corresponding to the historical conversion rate, and determine the association relationship between the user information and the conversion rate based on the user information of the historical user, the information label of the push information, the historical display position information of the push information, and the historical conversion rate of the push information. After obtaining the user information corresponding to the user identifier and the rough selection push information matched with the user identifier, the predicted conversion rate of each rough selection push information is predicted based on the association relationship between the user information and the conversion rate.

[0076] In an embodiment, the server can predict the predicted conversion rate of the user identifier for the rough selection push information through a pre-trained conversion rate prediction model. The conversion rate prediction model can be obtained by training historical user information, information tags of the push information, and historical conversion rates of the push information, and the like. The conversion rate prediction model is a fully connected neural network, including a feature extraction layer, a feature fusion layer, an input layer, a hidden layer, and an output layer. In actual implementation, the server inputs the obtained user information and information tags of the rough selection push information into the feature extraction layer to obtain user description features and push features, and fuses the user description features and the push features through the feature fusion layer to obtain fused features. The fused features are input into the input layer to be transmitted to the hidden layer through the input layer. Through the hidden layer, the activation function is called to obtain the hidden layer features corresponding to the fused features. Through the output layer, the hidden layer features are predicted to obtain the prediction result.

[0077] In an embodiment, the information push request received by the server further carries context information of a web page for displaying the push information. The server trains a conversion rate prediction model based on historical context information of the web page, historical user information of the user, information tags of the push information, and historical conversion rates of the push information. After obtaining the user information corresponding to the user identifier, the context information of the web page, and the rough selection push information matched with the user identifier, the server inputs the user information corresponding to the user identifier, the context information of the web page, and the rough selection push information matched with the user identifier into the conversion rate prediction model as inputs of the conversion rate prediction model, so as to output the predicted conversion rate through the conversion rate prediction model.

[0078] In step S206, a crowd selection queue corresponding to the rough selection push information is obtained.

[0079] The crowd selection queue stores historical predicted conversion rates corresponding to the rough selection push information in a first-in-first-out manner.

[0080] Specifically, the historical predicted conversion rate is a predicted conversion rate of the push information predicted by the server based on historical user information in a historical time period. Each piece of push information stored by the server maintains a corresponding crowd selection queue. The crowd selection queue of each piece of push information stores the latest preset amount of historical predicted conversion rates. For example, the crowd selection queue of each piece of push information stores the latest 10,000 historical predicted conversion rates.

[0081] In one embodiment, the process of maintaining the crowd selection queue for the push information by the server is as follows: after the predicted conversion rate is predicted, the predicted conversion rate is added to the tail of the crowd selection queue corresponding to the push information in the rough selection; the historical predicted conversion rate at the head of the crowd selection queue is removed, so as to ensure that the crowd selection queue maintained for the push information always maintains a preset data amount, and the crowd selection queue always stores the historical predicted conversion rate closest to the current time.

[0082] The historical predicted conversion rates in the crowd selection queue are sorted in the queue according to the prediction time, the historical predicted conversion rate at the head of the crowd selection queue is generated at the time farthest from the current time, and the historical predicted conversion rate at the tail of the crowd selection queue is generated at the time closest to the current time.

[0083] Step S208, determine the sorting result of the predicted conversion rate in the crowd selection queue.

[0084] Specifically, after the server obtains the crowd selection queue corresponding to the rough selection push information, for each rough selection push information, the historical predicted conversion rate and the latest added current predicted conversion rate in the crowd selection queue are sorted respectively, so as to determine the sorting result of the predicted conversion rate in the crowd selection queue. The sorting is sorting according to the numerical value of the historical predicted conversion rate and the predicted conversion rate, such as sorting according to the numerical value from large to small or sorting according to the numerical value from small to large. The sorting of the historical predicted conversion rate and the predicted conversion rate can be realized by using a classical sorting algorithm, which can be bubble sort, insertion sort, quick sort or bucket sort.

[0085] In one embodiment, step S208 includes the following steps: obtaining the bucket sorting result after the crowd selection queue is divided into buckets; each bucket in the bucket sorting result is sorted according to the corresponding conversion rate interval; based on the target position of the target bucket corresponding to the matched conversion rate interval of the predicted conversion rate in the bucket sorting result, the sorting result of the predicted conversion rate in the crowd selection queue is determined.

[0086] The bucketing is an operation of sorting the historical predicted conversion rate and the predicted conversion rate in the crowd preferential queue, and is used for dividing the historical predicted conversion rate and the predicted conversion rate into different buckets. Each divided bucket corresponds to a different conversion rate interval. For example, the conversion rate interval [0, 0.25) corresponds to a bucket numbered 1, the conversion rate interval [0.25, 0.5) corresponds to a bucket numbered 2, the conversion rate interval [0.5, 0.75) corresponds to a bucket numbered 3, and the conversion rate interval [0.75, 1] corresponds to a bucket numbered 4. The bucket sorting result is the sorting result between different buckets, which can also be referred to as an inter-bucket sorting result. For example, the buckets are sorted in ascending order of the conversion rate interval, and the bucket sorting result is “number 1, number 2, number 3, number 4”. The position of the bucket numbered 3 in the bucket sorting result is the third position.

[0087] In step S210, the target push information is filtered from the rough selection push information based on the sorting result corresponding to the rough selection push information and the predicted conversion rate.

[0088] The sorting result corresponding to the rough selection push information is the sorting result of the corresponding predicted conversion rate in the crowd preferential queue. That is, the sorting result represents the sorting of the predicted conversion rate of the current user in the historical crowd. Therefore, the target push information is filtered by comprehensively considering the predicted conversion rate and the sorting of the current user in the crowd, so that the target push information filtered is the information with the highest push value. Especially when the push information is an advertisement, it can bring better benefits to the advertiser.

[0089] Specifically, the server sorts the rough selection push information based on the predicted conversion rate corresponding to the rough selection push information and the sorting result of the predicted conversion rate in the crowd preferential queue, obtains a push sorting result, and filters the target push information from the rough selection push information based on the push sorting result. For example, the rough selection push information with the first sorting position in the push sorting result is determined as the target push information; or when the push information display position is multiple, the multiple rough selection push information with the front sorting position in the push sorting result and matching the number of push information display positions is determined as the target push information.

[0090] In one embodiment, step S210 specifically includes the following steps: calculating the push score of the rough selection push information according to the sorting result corresponding to the rough selection push information and the predicted conversion rate; and filtering the target push information from the rough selection push information based on the push score. The push score is used to sort the rough selection push information.

[0091] Specifically, after obtaining the predicted conversion rate corresponding to the rough selection push information and the ranking result of the predicted conversion rate in the crowd selection queue, the server calculates the push score of each rough selection push information according to a preset rough selection push score calculation manner, ranks each rough selection push information according to the push score corresponding to each rough selection push information, obtains a push ranking result, and screens target push information from the rough selection push information based on the push ranking result. Wherein, after obtaining the push score of each rough selection push information, the server can determine the target push information by using the following formula:

[0092]

[0093] Wherein, M is the final screened target push information, ranking score i is the push score of the i-th rough selection push information.

[0094] In one embodiment, after obtaining the predicted conversion rate corresponding to the rough selection push information and the ranking result of the predicted conversion rate in the crowd selection queue, the server obtains the information quality score corresponding to each rough selection push information, and based on the ranking result, the predicted conversion rate and the information quality score corresponding to the rough selection push information, respectively calculates the push score of each rough selection push information, ranks each rough selection push information according to the push score corresponding to each rough selection push information, obtains a push ranking result, and screens target push information from the rough selection push information based on the push ranking result.

[0095] Wherein, the information quality score of the rough selection push information is determined based on the feedback information of the historical user, the historical user is the user object to which the rough selection push information is successfully pushed in the historical period, the feedback information is positive feedback information and negative feedback information, the positive feedback information refers to the information returned after the user clicks the push information, and the positive feedback information includes the number of positive clicks, the user identifier clicking the push information, the page where the push information is clicked, the information label of the push information clicked, etc. The negative feedback information refers to the negative emotion related information fed back by the user, such as clicking the button reflecting negative emotion, such as not interested in push information button, shielding button or irrelevant to me button, etc. The negative feedback information includes the number of negative clicks, the user identifier clicking the negative feedback button, the page where the negative feedback button is clicked, the information label of the push information to which the negative feedback button belongs, etc.

[0096] The push information screening method, the server screens the rough selection push information matched with the user identifier based on the user information corresponding to the user identifier, predicts the prediction conversion rate of the user identifier for the rough selection push information, obtains a crowd optimization queue corresponding to the rough selection push information, the crowd optimization queue stores the historical prediction conversion rate corresponding to the rough selection push information in a first-in-first-out manner, determines the sorting result of the prediction conversion rate in the crowd optimization queue, and screens the target push information from the rough selection push information based on the sorting result and the prediction conversion rate corresponding to the rough selection push information, so that the modeling can be performed based on the real-time related data and the historical related data of the dynamic change of the information and the like, and then a more accurate push information optimization result is finally obtained. The sorting result of the prediction conversion rate in the crowd optimization queue represents the sorting of the prediction conversion rate of the current user in the historical crowd, so that the target push information is screened out by comprehensively considering the prediction conversion rate and the sorting of the current user in the crowd, and the target push information screened out is the information with the highest push value, and especially when the push information is an advertisement, better benefits can be brought to the advertiser.

[0097] In one embodiment, the server obtaining the bucket sorting result of the crowd optimization queue after the bucketing comprises: determining the number of buckets for the bucketing operation and the conversion rate interval corresponding to each bucket based on the historical prediction conversion rate stored in the crowd optimization queue; performing the bucketing on the crowd optimization queue according to the conversion rate interval to which the historical prediction conversion rate and the prediction conversion rate belong, to determine the buckets corresponding to the historical prediction conversion rate and the prediction conversion rate respectively; and sorting the buckets according to the size of the conversion rate interval corresponding to each bucket, to obtain the bucket sorting result corresponding to the crowd optimization queue.

[0098] Specifically, after obtaining the crowd optimization queue, the server determines the data amount of the historical prediction conversion rate and the prediction conversion rate in the crowd optimization queue, and determines the maximum value and the minimum value of the historical prediction conversion rate and the prediction conversion rate in the crowd optimization queue, determines the number of user buckets in the bucket sorting based on the data amount, and determines the conversion rate interval corresponding to the crowd optimization queue based on the maximum value and the minimum value, and then determines the conversion rate interval corresponding to each bucket based on the number of buckets and the conversion rate interval corresponding to the crowd optimization queue. After a certain number of buckets with different conversion rate intervals are determined, the historical prediction conversion rate and the prediction conversion rate in the crowd optimization queue are traversed, the crowd optimization queue is bucketed based on the conversion rate interval to which the historical prediction conversion rate and the prediction conversion rate belong, to determine the buckets corresponding to the historical prediction conversion rate and the prediction conversion rate respectively, and after the bucketing operation is completed, the buckets are sorted according to the size of the conversion rate interval corresponding to each bucket, to obtain the sorting result of the crowd optimization queue.

[0099] For example, Figure 3As an embodiment of the crowd selection queue bucket sorting diagram, the server maintains a crowd selection queue with a data volume of 10000 in a first-in first-out manner for rough selection push information A. After obtaining the current predicted conversion rate, the server adds the current predicted conversion rate to the tail of the crowd selection queue, and removes the historical predicted conversion rate at the head of the crowd selection queue. The maximum value of the conversion rate in the current crowd selection queue is 0.9000, and the minimum value is 0.4000, that is, the conversion rate interval corresponding to the crowd selection queue is [0.4000, 0.9000]. In order to facilitate quick sorting, the number of buckets can be set to 1000. Correspondingly, 1000 conversion rate subintervals can be divided from the conversion rate interval [0.4000, 0.9000] corresponding to the crowd selection queue, and each conversion rate subinterval is determined as a conversion rate interval corresponding to a bucket. The conversion rate intervals of the 1000 buckets obtained are respectively “[0.4000, 0.4005), [0.4005, 0.4010), …, [0.8990, 0.8995), [0.8995, 0.9000]”. After obtaining the buckets with different conversion rate intervals, the crowd selection queue is divided into buckets according to the conversion rate intervals to which the historical predicted conversion rate and the predicted conversion rate belong, to determine the buckets corresponding to the historical predicted conversion rate and the predicted conversion rate respectively. During the bucket division, the buckets are sorted according to the conversion rate interval corresponding to each bucket after the bucket division is completed, to obtain the sorting result of the crowd selection queue. While maintaining the selection queue, the buckets can also be maintained. Specifically, the historical predicted conversion rate falling into each bucket can be counted. When the historical predicted conversion rate at the head is removed, the count of the bucket corresponding to the historical predicted conversion rate is reduced by one. When the current predicted conversion rate is added to the tail, the count of the bucket corresponding to the current predicted conversion rate is increased by one.

[0100] In the above embodiment, the server obtains the bucket sorting result corresponding to the crowd selection queue by sorting the crowd selection queue into buckets, so as to quickly determine the sorting result of the predicted conversion rate for each rough selection push information, thereby improving the efficiency of screening the target push information.

[0101] In an embodiment, the server determines the sorting result of the predicted conversion rate in the crowd selection queue based on the target position of the target bucket corresponding to the conversion rate interval matched by the predicted conversion rate in the bucket sorting result. The step includes: determining the sorting position of the predicted conversion rate in the crowd selection queue as the target position of the target bucket corresponding to the conversion rate interval matched by the predicted conversion rate in the bucket sorting result, to obtain the sorting result of the predicted conversion rate in the crowd selection queue. This avoids sorting all data in the crowd selection queue, reduces the amount of sorted data, improves the speed of determining the sorting result of the predicted conversion rate, and further improves the efficiency of screening the target push information.

[0102] In one embodiment, the step of determining the ranking result of the predicted conversion rate in the crowd selection queue based on the target position of the target bucket corresponding to the conversion rate interval matching the predicted conversion rate in the bucket ranking result includes: determining the target position of the target bucket corresponding to the conversion rate interval matching the predicted conversion rate in the bucket ranking result; performing in-bucket ranking on the conversion rate values falling into the target bucket to obtain an in-bucket ranking result; and determining the ranking result of the predicted conversion rate in the crowd selection queue based on the target position and the in-bucket ranking result.

[0103] wherein the conversion rate value is a value of the historical predicted conversion rate or the predicted conversion rate, falling into the target bucket means that the historical predicted conversion rate and the value of the predicted conversion rate belong to the conversion rate interval corresponding to the target bucket, and the in-bucket ranking is ranking the historical predicted conversion rate and the predicted conversion rate falling into the target bucket, and the in-bucket ranking can use the quicksort method.

[0104] Specifically, after the server performs in-bucket ranking on the conversion rate values falling into the target bucket to obtain an in-bucket ranking result, the server determines the ranking position of the predicted conversion rate in the in-bucket ranking result, then determines other buckets before the target position in the bucket ranking result, and counts the data amount of the historical predicted conversion rate falling into the other buckets before the target position, and determines the ranking result of the predicted conversion rate in the crowd selection queue based on the data amount of the historical predicted conversion rate falling into the other buckets and the ranking position of the predicted conversion rate in the in-bucket ranking result.

[0105] For example, after the server performs in-bucket ranking on the conversion rate values falling into the target bucket to obtain an in-bucket ranking result, the server determines that the ranking position of the predicted conversion rate in the in-bucket ranking result is the third, there are two other buckets before the target position in the bucket ranking result, and 10 and 12 historical predicted conversion rates fall into the two other buckets respectively, and then the server determines that the ranking position of the predicted conversion rate in the crowd selection queue is the 25th.

[0106] In the above embodiment, the server determines the target position of the target bucket corresponding to the conversion rate interval matching the predicted conversion rate in the bucket ranking result, and performs in-bucket ranking on the conversion rate values falling into the target bucket to obtain an in-bucket ranking result, so that the ranking result of the predicted conversion rate in the crowd selection queue can be determined based on the target position and the in-bucket ranking result, avoiding sorting all data in the crowd selection queue, reducing the amount of sorted data, improving the speed of determining the ranking result of the predicted conversion rate, and further improving the efficiency of screening target push information.

[0107] In an embodiment, the server calculates the push score of the rough selection push information according to the ranking result and the predicted conversion rate of the rough selection push information, and the steps include the following steps: calculating the expected value of the rough selection push information according to the predicted conversion rate of the rough selection push information; calculating the population quality score of the rough selection push information based on the ranking result and the expected value of the rough selection push information; obtaining the information quality score of the rough selection push information; and calculating the push score of the rough selection push information based on the expected value, the population quality score and the information quality score of the rough selection push information.

[0108] The expected value (expected Cost-Per-Impression, eCPM) is used to represent the cost that the information promoter may consume if the rough selection push information is pushed at present, and the calculation method of the expected value is related to the billing mode when the information is pushed. The billing mode when the information is pushed includes an oCPM (optimized Cost Per Mille, oCPM) mode and a CPA (Cost Per Action, CPA) mode. The oCPM mode is an optimized CPM (Cost Per Mille, CPM) mode, and the CPM mode refers to a transaction per thousand times of pushing, that is, the fee that the pusher needs to pay for every thousand times of pushing the rough selection push information. The oCPM mode is based on the bid objective of the pusher to calculate the fee that the pusher needs to pay for every thousand times of pushing the rough selection push information, wherein the bid objective can also be referred to as an optimization objective. The CPA mode refers to a transaction per conversion, that is, the pusher pays the fee according to the number of conversion behaviors. The population quality score (Population quality) can also be referred to as a deep quality score (Deep quality). The population quality score is used to represent the ranking of the predicted conversion rate of the corresponding rough selection push information in the population selection queue, that is, the quality of the rough selection push information in the population corresponding to the population selection queue.

[0109] Specifically, the server calculates the push score of the rough selection push information based on the expected value, the population quality score and the information quality score of the rough selection push information, and the calculation can be performed by the following formula:

[0110] Score Ranking = eCPM + Score Information quality + Score Deep quality

[0111] Score Ranking is the push score of the rough selection push information, eCPM is the expected value of the rough selection push information, Score Information quality is the information quality score of the rough selection push information, and Score Deep qualityThe crowd quality score of the rough selection push information is calculated.

[0112] In the above embodiment, the server calculates the expected value of the rough selection push information according to the predicted conversion rate of the rough selection push information, respectively; the crowd quality score of the rough selection push information is calculated based on the ranking result and the expected value of the rough selection push information; the information quality score of the rough selection push information is obtained, and the push score of the rough selection push information is calculated based on the expected value, the crowd quality score and the information quality score of the rough selection push information, that is, the target push information is selected from the dimensions of the predicted conversion rate of the information, the ranking of the current user in the crowd, and the quality of the rough selection push information itself, so that the selected target push information is the information with the highest push value, especially when the push information is an advertisement, the better revenue can be brought to the advertiser.

[0113] In one embodiment, the step of calculating the expected value of the rough selection push information according to the predicted conversion rate of the rough selection push information includes the following steps: determining the currently adopted push mode; and calculating the expected value of the rough selection push information based on the predicted conversion rate of the rough selection push information according to the expected value determination manner corresponding to the push mode. The push mode is the charging mode when the information is pushed, including the oCPM mode and the CPA mode.

[0114] Specifically, if the current push mode is the CPA mode, the expected value of the rough selection push information is calculated based on the predicted conversion rate of the rough selection push information according to the expected value determination manner corresponding to the CPA mode; if the current push mode is the oCPM mode, the expected value of the rough selection push information is calculated based on the predicted conversion rate of the rough selection push information according to the expected value determination manner corresponding to the oCPM mode.

[0115] In the above embodiment, the server calculates the expected value of the rough selection push information based on the predicted conversion rate of the rough selection push information according to the expected value determination manner corresponding to the currently adopted push mode, so as to ensure that the finally selected target push information is the optimal push information under the current push mode.

[0116] In one embodiment, the step of calculating the expected value of the rough selection push information according to the predicted conversion rate of the rough selection push information includes the following steps: obtaining a conversion frequency target, and calculating the expected value of the rough selection push information based on the conversion frequency target and the predicted conversion rate of the rough selection push information.

[0117] Specifically, if the current push mode is the CPA mode, the server obtains the conversion frequency target pre-provided by the promoter, and calculates the expected value of the rough selection push information according to the conversion frequency target and the predicted conversion rate after obtaining the predicted conversion rate of the rough selection push information. The expected value of the rough selection push information can be calculated by the following formula under the CPA mode:

[0118] eCPM = pCVR x CPA

[0119] Wherein, eCPM is the expected value of the rough selected push information, pCVR is the predicted conversion rate of the rough selected push information, and CPA is the given conversion times target.

[0120] In the above embodiment, the server obtains the conversion times target, and calculates the expected value of the rough selected push information based on the conversion times target and the predicted conversion rate of the rough selected push information, so as to ensure that the finally selected target push information is the optimal push information under the CPA mode.

[0121] In another embodiment, the step of calculating the expected value of the rough selected push information according to the predicted conversion rate of the rough selected push information comprises the following steps: obtaining the basic resource value corresponding to the rough selected push information; predicting the predicted click rate of the user identifier for the rough selected push information; and calculating the expected value of the rough selected push information based on the basic resource value, the predicted conversion rate and the predicted click rate of the rough selected push information.

[0122] Wherein, the basic resource value is the basic bid of the bid objective given by the push party under the oCPM mode. The click rate (Click Through Rate, CTR) is the ratio of the number of clicks to the number of push displays, which reflects the degree of attention of the push information. The predicted click rate (predict Click Through Rate, pCTR) is the estimated value of the click rate of each pushable information before the information push, which is used to estimate the probability of the corresponding push information being clicked by the user after being pushed and displayed.

[0123] Specifically, if the current push mode is the oCPM mode, the server needs to predict the predicted click rate of the user identifier for the rough selected push information after filtering the rough selected push information matched with the user identifier based on the user information corresponding to the user identifier. For example, a pre-trained click rate prediction model is used to predict the predicted click rate of the rough selected information, and the expected value of the rough selected push information is calculated based on the basic resource value, the predicted conversion rate and the predicted click rate of the rough selected push information by the following formula:

[0124] eCPM = bid x pCVR x pCTR

[0125] Wherein, eCPM is the expected value of the rough selected push information, bid is the basic resource value corresponding to the rough selected push information, pCVR is the predicted conversion rate of the rough selected push information, and pCTR is the predicted click rate corresponding to the rough selected push information.

[0126] In the above embodiments, the server predicts the predicted click rate of the user identifier for the rough selection push information by obtaining the basic resource value corresponding to the rough selection push information, calculates the expected value of the rough selection push information based on the basic resource value, the predicted conversion rate and the predicted click rate of the rough selection push information, so as to ensure that the finally screened target push information is the optimal push information in the eCPM mode.

[0127] In one embodiment, the pusher sets multiple conversion targets for one rough selection push information, and gives different basic bids for different conversion targets. Such rough selection push information can also be referred to as multi-bid push information. Thus, for the same rough selection push information, the server needs to estimate the predicted conversion rates corresponding to different conversion targets, and then calculate the expected values corresponding to different conversion targets, and then calculate the push scores corresponding to different conversion targets based on the expected values. Among them, the different conversion targets given by the promoter have a time or logical progressive relationship, such as two conversion targets of a game APP push information, the first conversion target is a shallow target of downloading, and the second conversion target is a deep target of registration. The conversion behaviors corresponding to the first conversion target and the second conversion target have a chronological relationship, that is, the downloading needs to be completed first and then the registration can be performed.

[0128] Specifically, for the multi-bid rough selection push information, the server estimates the predicted conversion rates pCVR i Then, the expected values eCPM i of different conversion targets are calculated respectively by the following formula:

[0129] eCPM i bid i ×pCVR i ×pCTR

[0130] Wherein, eCPM i is the expected value of the i th conversion target of the rough selection push information, bid i is the basic resource value corresponding to the i th conversion target of the rough selection push information, pCVR i is the predicted conversion rate corresponding to the i th conversion target of the rough selection push information. pCTR is the predicted click rate corresponding to the rough selection push information. Since the realization of different conversion targets is the occurrence of the push information after clicking, the predicted click rates corresponding to different conversion targets of the rough selection push information are the same.

[0131] In the above embodiments, the server calculates the push scores corresponding to different conversion targets after estimating the predicted conversion rates corresponding to different conversion targets, determines whether the multi-bid rough selection push information can be pushed from the dimensions of multiple conversion targets, and when the push score corresponding to one of the conversion targets meets the push, pushes the multi-bid rough selection push information with the bid of the corresponding conversion target.

[0132] In one embodiment, the server calculates the crowd quality scores of the rough selection push information based on the ranking results of the rough selection push information and the expected values, and the steps include the following steps: calculating the crowd quality coefficients of the rough selection push information according to the ranking results of the rough selection push information; and calculating the crowd quality scores of the rough selection push information based on the crowd quality coefficients of the rough selection push information and the expected values.

[0133] The crowd quality coefficient has a value range of 0 to 1, and is used to represent the ranking of the predicted conversion rate of the rough selection push information of the rough selection push in the crowd selection queue, i.e., the ranking of the user corresponding to the rough selection push information in the crowd corresponding to the crowd selection queue.

[0134] Specifically, after determining the crowd quality coefficient, the server calculates the crowd quality scores of the rough selection push information based on the crowd quality coefficient of the rough selection push information and the expected value by using the following formula:

[0135] Score Deep quality = Q deep × eCPM × delta

[0136] Wherein, Score Deep quality is the crowd quality score of the rough selection push information, Q deep is the crowd quality coefficient of the rough selection push information, eCPM is the expected value of the rough selection push information, and delta is a hyperparameter. The hyperparameter is related to the scene involved in the current push.

[0137] In the above embodiments, the server calculates the crowd quality coefficients of the rough selection push information according to the ranking results of the rough selection push information, and calculates the crowd quality scores of the rough selection push information based on the crowd quality coefficients of the rough selection push information and the expected values, so as to quantify the influence of the ranking of the current user in the historical crowd on the screening of the push information, and further improve the accuracy of the screening of the push information.

[0138] In an embodiment, the server calculates the crowd quality coefficient of the rough selection push information according to the ranking result of the rough selection push information, and the calculation includes the following steps: obtaining a mapping parameter corresponding to a data amount of a historical predicted conversion rate stored in the crowd selection queue; and calculating the crowd quality coefficient of the rough selection push information based on the ranking result of the rough selection push information and the mapping parameter.

[0139] Specifically, if the ranking result of the rough selection push information is an accurate ranking position of the predicted conversion rate in the crowd selection queue, the crowd quality coefficient of the rough selection push information is calculated based on a mapping parameter corresponding to a data amount of a historical predicted conversion rate stored in the crowd selection queue, and the calculation uses the following formula:

[0140]

[0141] wherein Q deep is the crowd quality coefficient of the rough selection push information, ruank numi is the ranking position of the rough selection push information in the crowd selection queue, and m is the mapping parameter. numi When the ranking result of the rough selection push information is an accurate ranking position of the predicted conversion rate in the crowd selection queue, m is a data amount of a historical predicted conversion rate stored in the crowd selection queue. numi When the ranking result of the rough selection push information is a target position of the predicted conversion rate in a target bucket of the crowd selection queue, m is a number of buckets obtained after a bucketing operation on the historical predicted conversion rate stored in the crowd selection queue.

[0142] In the above embodiment, the server obtains a mapping parameter corresponding to a data amount of a historical predicted conversion rate stored in the crowd selection queue, and calculates the crowd quality coefficient of the rough selection push information based on the ranking result of the rough selection push information and the mapping parameter, so as to quantify the ranking of the current user in the historical crowd, and further quantify the influence of the ranking of the current user in the historical crowd on the screening of the push information, thereby improving the accuracy of the screening of the push information.

[0143] In an embodiment, as shown in Figure 4 , a push information screening method is also provided, and the method is applied to the server 104 in Figure 1 for example, and includes the following steps:

[0144] In step S402, the rough selection push information matched with the user identifier is screened based on the user information corresponding to the user identifier.

[0145] In step S404, the predicted conversion rate of the user identifier for the rough selection push information is predicted.

[0146] Step S406, a crowd selection queue corresponding to the coarse selection push information is obtained; the crowd selection queue stores historical predicted conversion rates corresponding to the coarse selection push information in a first-in-first-out manner.

[0147] Step S408, based on the historical predicted conversion rates stored in the crowd selection queue, the number of buckets for the bucketing operation and the conversion rate interval corresponding to each bucket are determined.

[0148] Step S410, the crowd selection queue is bucketed according to the historical predicted conversion rate and the conversion rate interval to which the predicted conversion rate belongs, and the bucket corresponding to the historical predicted conversion rate and the predicted conversion rate is determined respectively.

[0149] Step S412, each bucket is sorted according to the size of the conversion rate interval corresponding to each bucket, and a bucket sorting result corresponding to the crowd selection queue is obtained.

[0150] Step S414, the target position of the target bucket corresponding to the conversion rate interval matched with the predicted conversion rate in the bucket sorting result is determined as the sorting position of the predicted conversion rate in the crowd selection queue, and a sorting result of the predicted conversion rate in the crowd selection queue is obtained.

[0151] Step S416, the basic resource value corresponding to the coarse selection push information is obtained.

[0152] Step S418, the predicted click rate of the user identifier for the coarse selection push information is predicted.

[0153] Step S420, the expected value of the coarse selection push information is calculated based on the basic resource value, the predicted conversion rate and the predicted click rate of the coarse selection push information respectively.

[0154] Step S422, a mapping parameter corresponding to the data volume of the historical predicted conversion rate stored in the crowd selection queue is obtained.

[0155] Step S424, the crowd quality coefficient of the coarse selection push information is calculated based on the sorting result of the coarse selection push information and the mapping parameter respectively.

[0156] Step S426, the crowd quality score of the coarse selection push information is calculated based on the crowd quality coefficient and the expected value of the coarse selection push information respectively.

[0157] Step S428, the information quality score of the coarse selection push information is obtained.

[0158] Step S430, the push score of the coarse selection push information is calculated based on the expected value, the crowd quality score and the information quality score of the coarse selection push information respectively.

[0159] Step S432, the target push information is selected from the coarse selection push information based on the push score.

[0160] The application also provides an application scenario, which is an advertisement pushing scenario, and the application scenario applies the pushing information screening method. Specifically, the pushing information screening method is applied as follows in the application scenario:

[0161] Figure 5 For an embodiment, an advertisement pushing system schematic diagram is provided, which is an advertisement pushing system on the side of an advertisement owner platform. The system includes a request receiving module, an advertisement sorting module, a display module, and a data backflow module. The request receiving module is configured to receive a pushing information display request and transmit the information display request to the advertisement sorting module. The advertisement sorting module is configured to screen out a target advertisement, i.e., a winning advertisement, based on the received information display request by using the pushing information screening method of the application. The display module is configured to display the winning advertisement. The data backflow module is configured to collect exposure, click, conversion, and other data of the winning advertisement after the winning advertisement, and transmit the collected data back to the advertisement sorting module, so that the advertisement sorting module optimizes the advertisement sorting algorithm based on the transmitted data. Specifically, the advertisement sorting module includes a rough selection sorting module and a fine selection sorting module. After rough selection advertisements are screened out through the rough selection sorting module, the fine selection sorting module is used to predict the predicted conversion rate of the rough selection advertisements, and the rough selection advertisements are placed in their corresponding people group selection optimization queues, so as to determine the people group quality scores of the rough selection advertisements, and sort the rough selection advertisements based on the people group quality scores and the predicted conversion rates of the rough selection advertisements, so as to screen out the target advertisement, i.e., the winning advertisement. The screening of the target advertisement is completed based on the real-time related data and historical people group data of the rough selection advertisements, which improves the accuracy of the selection and sorting of the advertisements, so that a better advertisement pushing effect can be achieved, and better benefits can be brought to the advertisement owners.

[0162] Figure 6 For an embodiment, an advertisement total revenue schematic diagram is provided, from which it can be seen that the total revenue of the advertisement pushing system is increasing from 2019 to 2020. Figure 6 From the above, it can be seen that the deep double bid consumption of the big board increased from 360W / day on average in the second half of 2019 to 1700W / day in April 2020. Figure 7 For an embodiment, a game industry advertisement revenue schematic diagram is provided, from which it can be seen that the deep target consumption proportion of the game industry increased from 40% (daily consumption of 375W) at the beginning of 2020 to 75% in April 2020. Figure 7 For an embodiment, a game industry advertisement revenue schematic diagram is provided, from which it can be seen that the deep target consumption proportion of the game industry increased from 40% (daily consumption of 375W) at the beginning of 2020 to 75% in April 2020. Figure 8 For an embodiment, an education industry advertisement revenue schematic diagram is provided, from which it can be seen that the deep target consumption proportion of the education industry increased from 10% (daily consumption of 150W) at the beginning of 2020 to 80% (daily consumption of 650W) in April 2020. Figure 8 For an embodiment, an education industry advertisement revenue schematic diagram is provided, from which it can be seen that the deep target consumption proportion of the education industry increased from 10% (daily consumption of 150W) at the beginning of 2020 to 80% (daily consumption of 650W) in April 2020.

[0163] It should be understood that, althoughFigure 2 , 4 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 2 , 4 At least some of the steps in the process may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the steps or stages in other steps.

[0164] In one embodiment, such as Figure 9 As shown, a push notification filtering device is provided. This device can be a software module, a hardware module, or a combination of both integrated into a computer device. Specifically, the device includes: a coarse-selection push notification filtering module 902, a conversion rate prediction module 904, a target audience optimization queue acquisition module 906, a ranking result determination module 908, and a target push notification filtering module 910, wherein:

[0165] The coarse selection push information filtering module 902 is used to filter coarse selection push information that matches the user identifier based on the user information corresponding to the user identifier.

[0166] The conversion rate prediction module 904 is used to predict the conversion rate of user identifiers for the initial push information.

[0167] The audience selection queue acquisition module 906 is used to acquire the audience selection queue corresponding to the coarse selection push information; the audience selection queue stores the historical predicted conversion rate corresponding to the coarse selection push information in a first-in-first-out manner.

[0168] The sorting result determination module 908 is used to determine the sorting result of the predicted conversion rate in the population optimization queue.

[0169] The target push information filtering module 910 is used to filter target push information from the coarsely selected push information based on the sorting results and predicted conversion rate of the coarsely selected push information.

[0170] In one embodiment, the apparatus further includes: a crowd selection queue maintenance module, configured to: after the predicted conversion rate is obtained, add the predicted conversion rate to the tail of the crowd selection queue corresponding to the coarse selection push information; and remove the historical predicted conversion rate from the head of the crowd selection queue.

[0171] In an embodiment, the sorting result determination module 908 is further configured to: obtain a bucket sorting result of the bucketing of the crowd selection queue; each bucket in the bucket sorting result is sorted according to a corresponding conversion rate interval; and determine the sorting result of the predicted conversion rate in the crowd selection queue based on a target position of a target bucket corresponding to a conversion rate interval matched by the predicted conversion rate in the bucket sorting result.

[0172] In an embodiment, the target push information screening module 910 is further configured to: calculate a push score of the rough selection push information according to the sorting result corresponding to the rough selection push information and the predicted conversion rate; and screen the target push information from the rough selection push information based on the push score.

[0173] In the above embodiments, the rough selection push information matching the user identifier is screened based on the user information corresponding to the user identifier, the predicted conversion rate of the user identifier for the rough selection push information is predicted, the crowd selection queue corresponding to the rough selection push information is obtained, the crowd selection queue stores the historical predicted conversion rates corresponding to the rough selection push information in a first-in first-out manner, the sorting result of the predicted conversion rate in the crowd selection queue is determined, and the target push information is screened from the rough selection push information based on the sorting result corresponding to the rough selection push information and the predicted conversion rate, so that the modeling can be performed based on multi-dimensional real-time related data and historical related data and the like that dynamically change with information, and a more accurate push information optimization result can be finally obtained. Since the sorting result of the predicted conversion rate in the crowd selection queue represents the sorting of the predicted conversion rate of the current user in the historical crowd, the target push information is screened by comprehensively considering the predicted conversion rate and the sorting of the current user in the crowd, so that the target push information screened is the information with the highest push value, and especially when the push information is an advertisement, better benefits can be brought to the advertiser.

[0174] In an embodiment, the sorting result determination module 908 is further configured to: determine the number of buckets for the bucketing operation and the conversion rate interval corresponding to each bucket based on the historical predicted conversion rates stored in the crowd selection queue; perform the bucketing on the crowd selection queue according to the conversion rate intervals to which the historical predicted conversion rates and the predicted conversion rate respectively belong, to determine the buckets corresponding to the historical predicted conversion rates and the predicted conversion rate respectively; and sort each bucket according to the size of the conversion rate interval corresponding to each bucket, to obtain the bucket sorting result corresponding to the crowd selection queue.

[0175] In the above embodiments, the bucket sorting result corresponding to the crowd selection queue is obtained by performing the bucketing and sorting on the crowd selection queue, so that the sorting result of the predicted conversion rate can be quickly determined for each rough selection push information, and the efficiency of screening the target push information is improved.

[0176] In an embodiment, the sorting result determination module 908 is further configured to: determine a target position of the target bucket corresponding to the conversion rate interval matching the predicted conversion rate in the bucket sorting result as the sorting position of the predicted conversion rate in the crowd selection queue, to obtain the sorting result of the predicted conversion rate in the crowd selection queue.

[0177] In the above embodiment, by determining the target position of the target bucket corresponding to the conversion rate interval matching the predicted conversion rate in the bucket sorting result as the sorting position of the predicted conversion rate in the crowd selection queue, the sorting result of the predicted conversion rate in the crowd selection queue is obtained, which avoids sorting all data in the crowd selection queue, reduces the amount of sorted data, improves the speed of determining the sorting result of the predicted conversion rate, and further improves the efficiency of screening the target push information.

[0178] In an embodiment, the sorting result determination module 908 is further configured to: determine a target position of the target bucket corresponding to the conversion rate interval matching the predicted conversion rate in the bucket sorting result; perform in-bucket sorting on the conversion rate values falling into the target bucket to obtain an in-bucket sorting result; and determine the sorting result of the predicted conversion rate in the crowd selection queue based on the target position and the in-bucket sorting result.

[0179] In the above embodiment, by determining the target position of the target bucket corresponding to the conversion rate interval matching the predicted conversion rate in the bucket sorting result, and performing in-bucket sorting on the conversion rate values falling into the target bucket to obtain an in-bucket sorting result, the sorting result of the predicted conversion rate in the crowd selection queue can be determined based on the target position and the in-bucket sorting result, which avoids sorting all data in the crowd selection queue, reduces the amount of sorted data, improves the speed of determining the sorting result of the predicted conversion rate, and further improves the efficiency of screening the target push information.

[0180] In an embodiment, the target push information screening module 910 is further configured to: calculate the expected value of the rough selection push information according to the predicted conversion rate of the rough selection push information; calculate the crowd quality score of the rough selection push information based on the sorting result and the expected value of the rough selection push information; obtain the information quality score of the rough selection push information; and calculate the push score of the rough selection push information based on the expected value, the crowd quality score, and the information quality score of the rough selection push information.

[0181] In the above embodiments, the expected values of the rough selection push information are respectively calculated according to the predicted conversion rates of the rough selection push information; the crowd quality scores of the rough selection push information are respectively calculated based on the ranking results and the expected values of the rough selection push information; the information quality scores of the rough selection push information are obtained, and the push scores of the rough selection push information are calculated based on the expected values, the crowd quality scores and the information quality scores of the rough selection push information, that is, the target push information is selected from the dimensions of the predicted conversion rate of the information, the ranking of the current user in the crowd and the quality of the rough selection push information itself, so that the selected target push information is the information with the highest push value, and especially when the push information is an advertisement, better benefits can be brought to the advertiser.

[0182] In one embodiment, the target push information screening module 910 is further configured to determine a currently adopted push mode; and calculate the expected values of the rough selection push information based on the predicted conversion rates of the rough selection push information according to an expected value determination manner corresponding to the push mode.

[0183] In the above embodiments, the expected values of the rough selection push information are calculated based on the predicted conversion rates of the rough selection push information according to the expected value determination manner corresponding to the currently adopted push mode, so that the finally selected target push information is the optimal push information under the current push mode.

[0184] In one embodiment, the target push information screening module 910 is further configured to obtain a basic resource value corresponding to the rough selection push information; predict a predicted click rate of the user identifier for the rough selection push information; and calculate the expected values of the rough selection push information based on the basic resource value, the predicted conversion rate and the predicted click rate of the rough selection push information.

[0185] In the above embodiments, the expected values of the rough selection push information are calculated based on the basic resource value, the predicted conversion rate and the predicted click rate of the rough selection push information by obtaining the basic resource value corresponding to the rough selection push information and predicting the predicted click rate of the user identifier for the rough selection push information, so that the finally selected target push information is the optimal push information under the eCPM mode.

[0186] In one embodiment, the target push information screening module 910 is further configured to calculate the crowd quality coefficients of the rough selection push information according to the ranking results of the rough selection push information; and calculate the crowd quality scores of the rough selection push information based on the crowd quality coefficients and the expected values of the rough selection push information.

[0187] In the above embodiment, the crowd quality coefficients of the rough selection push information are calculated according to the sorting results of the rough selection push information, and the crowd quality scores of the rough selection push information are calculated based on the crowd quality coefficients of the rough selection push information and the expected values, so that the influence of the sorting of the current user in the historical crowd on the screening of the push information can be quantified, and the accuracy of the screening of the push information is improved.

[0188] In one embodiment, the target push information screening module 910 is further configured to: obtain mapping parameters corresponding to the data quantity of the historical predicted conversion rate stored in the crowd selection queue; and calculate the crowd quality coefficients of the rough selection push information based on the sorting results of the rough selection push information and the mapping parameters.

[0189] In the above embodiment, the mapping parameters corresponding to the data quantity of the historical predicted conversion rate stored in the crowd selection queue are obtained, and the crowd quality coefficients of the rough selection push information are calculated based on the sorting results of the rough selection push information and the mapping parameters, so that the sorting of the current user in the historical crowd can be quantified, the influence of the sorting of the current user in the historical crowd on the screening of the push information can be quantified, and the accuracy of the screening of the push information is improved.

[0190] The specific limitations of the push information screening device can be referred to the limitations of the push information screening method described above, and will not be repeated here. Each module in the above push information screening device can be realized by software, hardware and their combination in whole or in part. The above modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0191] In one embodiment, a computer device is provided, which can be a server, and the internal structure diagram thereof can be as shown in Figure 10 The computer device includes a processor, a memory and a network interface connected by a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store related data of the push information. The network interface of the computer device is used to communicate with the external terminal through the network connection. The computer program is executed by the processor to implement a push information screening method.

[0192] In one embodiment, a computer device is provided, which can be a terminal, and the internal structure diagram thereof can be as shown in Figure 11As shown in the figure. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is used for wired or wireless communication with external terminals. Wireless mode can be achieved through WIFI, operator network, NFC (near field communication) or other technologies. The computer program is executed by the processor to implement a push information filtering method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0193] Those skilled in the art can understand that, Figure 10 Or Figure 11 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0194] In one embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the above method embodiments.

[0195] In one embodiment, a computer readable storage medium is provided, storing a computer program, which is executed by a processor to implement the steps in the above method embodiments.

[0196] In one embodiment, a computer program product or computer program is provided, which includes computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to make the computer device execute the steps in the above method embodiments.

[0197] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0198] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, but as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0199] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that for those skilled in the art, without departing from the concept of the present application, some modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A push information filtering method characterized by, The method comprises: When receiving a user terminal triggered information push request, filtering coarse selection push information matched with the user identifier based on the user information corresponding to the user identifier in the information push request; Predicting a predicted conversion rate of the user identifier for the coarse selection push information; Obtaining a crowd selection queue corresponding to the coarse selection push information; the crowd selection queue stores historical predicted conversion rates corresponding to the coarse selection push information in a first-in-first-out manner; Obtaining a bucket sorting result after the crowd selection queue is divided into buckets; each bucket in the bucket sorting result is sorted according to a corresponding conversion rate interval; Determining a sorting position of the predicted conversion rate in the crowd selection queue based on a target position of a target bucket corresponding to a conversion rate interval matched with the predicted conversion rate in the bucket sorting result; Respectively calculating an expected value of the coarse selection push information according to the predicted conversion rate of the coarse selection push information; Obtaining a data amount of the historical predicted conversion rates stored in the crowd selection queue as a mapping parameter; Determining a crowd quality coefficient as a ratio of the sorting position of the coarse selection push information and the mapping parameter; Respectively calculating a crowd quality score of the coarse selection push information based on the crowd quality coefficient and the expected value of the coarse selection push information; Obtaining an information quality score of the coarse selection push information; Calculating a push score of the coarse selection push information based on the expected value, the crowd quality score and the information quality score of the coarse selection push information; Filtering target push information from the coarse selection push information based on the push score.

2. The method of claim 1, wherein, The method further comprises: After the predicted conversion rate is predicted, adding the predicted conversion rate to the tail of the crowd selection queue corresponding to the coarse selection push information; Removing a historical predicted conversion rate at the head of the crowd selection queue.

3. The method of claim 1, wherein, The obtaining of the bucket sorting result after the crowd selection queue is divided into buckets comprises: Determining a number of buckets used for the bucket division operation and a conversion rate interval corresponding to each bucket based on the historical predicted conversion rates stored in the crowd selection queue; Dividing the crowd selection queue into buckets according to conversion rate intervals to which the historical predicted conversion rates and the predicted conversion rate respectively belong, to determine the buckets respectively corresponding to the historical predicted conversion rates and the predicted conversion rate; Sorting each bucket according to a size of the conversion rate interval corresponding to each bucket, to obtain the bucket sorting result corresponding to the crowd selection queue.

4. The method of claim 1, wherein, The determining of the sorting position of the predicted conversion rate in the crowd selection queue based on the target position of the target bucket corresponding to the conversion rate interval matched with the predicted conversion rate in the bucket sorting result comprises: Determining the target position of the target bucket corresponding to the conversion rate interval matched with the predicted conversion rate in the bucket sorting result as the sorting position of the predicted conversion rate in the crowd selection queue.

5. The method of claim 1, wherein, The determining of the sorting position of the predicted conversion rate in the crowd selection queue based on the target position of the target bucket corresponding to the conversion rate interval matched with the predicted conversion rate in the bucket sorting result comprises: determining a target bucket corresponding to the predicted conversion rate matching a conversion rate interval, a target position of the bucket sorting result; performing in-bucket sorting on the conversion rate values falling into the target bucket to obtain an in-bucket sorting result; based on the target position and the in-bucket sorting result, determining a sorting position of the predicted conversion rate in the population preference queue.

6. The method of claim 1, wherein, The expected value of the rough selection push information is calculated respectively based on the predicted conversion rate of the rough selection push information, including: determining a currently adopted push mode; determining the expected value of the rough selection push information based on the predicted conversion rate of the rough selection push information according to the expected value determination manner corresponding to the push mode.

7. The method of claim 1, wherein, The expected value of the rough selection push information is calculated respectively based on the predicted conversion rate of the rough selection push information, including: obtaining a basic resource value corresponding to the rough selection push information; predicting a predicted click rate of the user identifier for the rough selection push information; based on the basic resource value, the predicted conversion rate and the predicted click rate of the rough selection push information, calculating the expected value of the rough selection push information respectively.

8. A push information filtering apparatus characterized by comprising: The device includes: a rough selection push information screening module configured to, when receiving an information push request triggered by a user terminal, screen rough selection push information matched with a user identifier based on user information corresponding to the user identifier in the information push request; a conversion rate prediction module configured to predict a predicted conversion rate of the user identifier for the rough selection push information; a population preference queue acquisition module configured to acquire a population preference queue corresponding to the rough selection push information; the population preference queue stores historical predicted conversion rates corresponding to the rough selection push information in a first-in-first-out manner; a sorting result determination module configured to acquire a bucket sorting result after the population preference queue is binned; each bucket in the bucket sorting result is sorted according to a corresponding conversion rate interval; based on a target position of a target bucket corresponding to the predicted conversion rate matching a conversion rate interval in the bucket sorting result, a sorting position of the predicted conversion rate in the population preference queue is determined; a target push information screening module configured to calculate an expected value of the rough selection push information respectively based on the predicted conversion rate of the rough selection push information; acquire a data amount of the historical predicted conversion rates stored in the population preference queue as a mapping parameter; determine a population quality coefficient as a ratio of the sorting position of the rough selection push information and the mapping parameter; calculate a population quality score of the rough selection push information based on the population quality coefficient of the rough selection push information and the expected value; acquire an information quality score of the rough selection push information; calculate a push score of the rough selection push information based on the expected value, the population quality score and the information quality score of the rough selection push information; and screen target push information from the rough selection push information based on the push score.

9. The push information filtering apparatus according to claim 8, characterized by The device further comprises a crowd preferential queue maintaining module, configured to, after obtaining the predicted conversion rate, add the predicted conversion rate to the tail of a crowd preferential queue corresponding to the rough selection push information; and remove a historical predicted conversion rate at the head of the crowd preferential queue.

10. The push information filtering apparatus according to claim 8, wherein The sorting result determining module is further configured to determine, based on the historical predicted conversion rates stored in the crowd preferential queue, a number of buckets for a bucketing operation and a conversion rate interval corresponding to each bucket; perform bucketing on the crowd preferential queue according to the conversion rate intervals to which the historical predicted conversion rates and the predicted conversion rate respectively belong, to determine the buckets respectively corresponding to the historical predicted conversion rates and the predicted conversion rate; and sort each bucket according to the size of the conversion rate interval corresponding to each bucket, to obtain a bucket sorting result corresponding to the crowd preferential queue.

11. The push information filtering apparatus according to claim 8, wherein The sorting result determining module is further configured to determine, as the sorting position of the predicted conversion rate in the crowd preferential queue, a target position of a target bucket corresponding to a conversion rate interval to which the predicted conversion rate belongs in the bucket sorting result.

12. The push information filtering apparatus according to claim 8, characterized by The sorting result determining module is further configured to determine, as the sorting position of the predicted conversion rate in the crowd preferential queue, a target position of a target bucket corresponding to a conversion rate interval to which the predicted conversion rate belongs in the bucket sorting result; perform in-bucket sorting on conversion rate values falling into the target bucket, to obtain an in-bucket sorting result; and determine the sorting position of the predicted conversion rate in the crowd preferential queue based on the target position and the in-bucket sorting result.

13. The push information filtering apparatus according to claim 8, characterized by The target push information screening module is further configured to determine a push mode currently adopted; and calculate, according to a determination manner of the expected value corresponding to the push mode, an expected value of the rough selection push information based on the predicted conversion rate of the rough selection push information.

14. The push information filtering apparatus according to claim 8, characterized by The target push information screening module is further configured to obtain a basic resource value corresponding to the rough selection push information; predict a predicted click rate of the user identifier for the rough selection push information; and calculate, based on the basic resource value, the predicted conversion rate and the predicted click rate of the rough selection push information, an expected value of the rough selection push information.

15. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the method in any one of claims 1 to 7.

16. A computer readable storage medium storing a computer program, wherein the computer program comprises program instructions configured to cause a processor to perform the method according to any one of claims 1 to 15. The computer program is executed by the processor to implement the method in any one of claims 1 to 7.

17. A computer program product comprising computer instructions, characterized in that, The computer program is executed by the processor to implement the method in any one of claims 1 to 7.

Images