Cboth case pushing method and device, electronic equipment and storage medium
Through the copy push method based on real-time click-through rate and exposure order, the problem of cold start of new copy and slow response at the end of the copy life cycle is solved, and the accuracy and response speed of copy push is improved.
Patent Information
- Application Number
- CN202411941301.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-26
- Publication Date
- 2025-04-18
AI Technical Summary
In the prior art, the accuracy of copywriting push is low, mainly due to the difficulty of cold start of new copywriting and the inability to respond quickly at the end of the copywriting life cycle.
By obtaining the first real-time click-through rate and the second real-time click-through rate of the copy to be pushed, calculating the estimated click-through rate and recommended scores, combining exposure magnitude and group analysis, optimizing the copy push strategy to improve the accuracy and response speed of copy exposure magnitude allocation.
The accuracy of copywriting push is improved, and through effective exposure magnitude allocation and fast response mechanism, the exposure magnitude of new copywriting and the overall effect of copywriting push is improved.
Smart Images
Figure CN119848345A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of content recommendation, and more specifically, to a method, device, electronic device, and storage medium for pushing content. Background Art
[0002] The intelligent push system analyzes user behavior patterns and preferences, and uses data mining, machine learning and other technologies to conduct in-depth analysis and mining of users' historical data, thereby predicting products or services that users may be interested in and proactively pushing these contents to users.
[0003] At present, copywriting is generally generated by operations personnel or algorithm models, and a copywriting pool is formed after review by operations personnel. The intelligent push system selects the best copywriting from the copywriting pool and pushes it to users. Due to the problems of cold start difficulty and short life cycle of copywriting, the accuracy of copywriting push is low. Summary of the Invention
[0004] The purpose of the embodiments of the present application is to provide a copy pushing method, device, electronic device and storage medium to improve the accuracy of copy pushing.
[0005] In a first aspect, an embodiment of the present application provides a copy pushing method, the method comprising: obtaining a copy to be pushed; based on a first real-time click-through rate and a second real-time click-through rate of the copy to be pushed, obtaining an estimated click-through rate and a first score of the copy to be pushed; wherein, the first real-time click-through rate is used to characterize the overall real-time click-through rate of the copy to be pushed; the second real-time click-through rate is used to characterize the real-time click-through rate of the copy to be pushed in each group; the first score is positively correlated with the estimated click-through rate; based on the exposure level of the copy to be pushed in each group, obtaining a second score of the copy to be pushed; wherein, the second score is negatively correlated with the exposure level; based on the first score and the second score, obtaining a recommendation score of the copy to be pushed; and based on the recommendation score, pushing the copy to be pushed.
[0006] In the process of implementing the above scheme, on the one hand, the above scheme can consider the exposure level of the copy in different groups to support the new copy, which is conducive to improving the problem of unreasonable distribution of the exposure level of the new copy, and thus improving the accuracy of the copy push; on the other hand, the above scheme can consider the real-time click-through rate of the copy to adjust the recommendation score of the copy, which can achieve a rapid response to the copy push based on the real-time click-through rate of the copy, which is conducive to further improving the accuracy of the copy push.
[0007] In an implementation of the first aspect, obtaining the estimated click-through rate of the copy to be pushed based on the first real-time click-through rate and the second real-time click-through rate of the copy to be pushed includes: using the first real-time click-through rate and the second real-time click-through rate of the copy to be pushed as inputs of a click-through rate prediction model, and obtaining the estimated click-through rate of the copy to be pushed output by the click-through rate prediction model.
[0008] In the implementation of the above solution, the click-through rate prediction model is used to comprehensively consider the real-time click-through rate of the copy to be pushed to obtain the estimated click-through rate of the copy to be pushed. This can accurately capture the real-time performance of the copy, and then achieve timely response to user behavior, which is conducive to improving the accuracy of copy push.
[0009] In an implementation of the first aspect, obtaining the estimated click-through rate of the copy to be pushed based on the first real-time click-through rate and the second real-time click-through rate of the copy to be pushed includes: using the basic material characteristics, copy characteristics and user characteristics as inputs of a click-through rate prediction model, and obtaining the estimated click-through rate of the copy to be pushed output by the click-through rate prediction model; wherein the copy characteristics include characteristics constructed based on the first real-time click-through rate and the second real-time click-through rate of the copy to be pushed.
[0010] In the process of implementing the above solution, comprehensively considering the basic characteristics of the material, the characteristics of the copy and the characteristics of the user to obtain the estimated click-through rate is conducive to improving the estimation accuracy of the estimated click-through rate, and then improving the copy push accuracy of the above copy push method.
[0011] In an implementation of the first aspect, the method of using the basic material characteristics, copy characteristics and user characteristics as inputs of the click-through rate prediction model to obtain the estimated click-through rate of the copy to be pushed output by the click-through rate prediction model includes: splicing the basic material characteristics, copy characteristics and user characteristics of the copy to be pushed to obtain splicing characteristics; inputting the splicing characteristics into the trained click-through rate prediction model to obtain the estimated click-through rate of the copy to be pushed output by the click-through rate prediction model.
[0012] In the process of implementing the above solution, the basic material characteristics, copy characteristics and user characteristics are spliced together and then input into the click-through rate prediction model, which is conducive to providing more comprehensive and detailed information to the click-through rate prediction model, and is conducive to improving the estimation accuracy of the estimated click-through rate, thereby improving the copy push accuracy of the above copy push method.
[0013] In an implementation of the first aspect, obtaining the second score of the copy to be pushed based on the exposure level of the copy to be pushed in each group includes: calculating the second score of the copy to be pushed based on the exposure level in each group and the exposure level of the copy to be pushed in each group using a group confidence interval upper bound algorithm.
[0014] In the implementation process of the above solution, the second score of the copy to be pushed is calculated using the cluster confidence interval upper bound algorithm. On the one hand, it is conducive to effectively optimizing the copy selection strategy in the copy cold start phase, thereby improving the copy push accuracy of the above copy push method; on the other hand, it enables the above copy push method to be applicable to more application scenarios, which is conducive to improving the adaptability of the above copy push method.
[0015] In an implementation of the first aspect, obtaining the recommended score of the copy to be pushed based on the first score and the second score includes: obtaining the recommended score of the copy to be pushed based on the first score, the second score and the exploration coefficient; wherein the exploration coefficient is used to adjust the push priority between the first category of copy and the second category of copy; the first category of copy is a copy with a lower exposure level; the second category of copy is a copy with a higher estimated click-through rate.
[0016] In the implementation process of the above solution, the exploration coefficient can effectively balance the relationship between exploring new copywriting and utilizing deterministic copywriting, which is conducive to improving the decision-making quality of the above copywriting push method, and further improving the copywriting push accuracy of the above copywriting push method.
[0017] In an implementation of the first aspect, the method further includes: utilizing a real-time message queue to perform real-time statistics on the overall exposure of the copy, the overall click volume of the copy, the exposure of each group, the click volume of each group, the exposure of the copy in each group, and the click volume of the copy in each group.
[0018] During the implementation of the above solution, various data can be quickly obtained in real time through the real-time message queue, so that the above copy push method can quickly respond to user behavior and update the copy push strategy in time, which is conducive to improving the copy push accuracy of the above copy push method.
[0019] In a second aspect, an embodiment of the present application provides a document pushing device, comprising:
[0020] The module for obtaining the copy to be pushed is used to obtain the copy to be pushed;
[0021] A first score acquisition module is configured to acquire an estimated click-through rate and a first score of the document to be pushed based on a first real-time click-through rate and a second real-time click-through rate of the document to be pushed; wherein the first real-time click-through rate is used to represent the overall real-time click-through rate of the document to be pushed; the second real-time click-through rate is used to represent the real-time click-through rate of the document to be pushed in each group; and the first score is positively correlated with the estimated click-through rate;
[0022] A second score acquisition module is configured to acquire a second score of the document to be pushed based on the exposure level of the document to be pushed in each group; wherein the second score is negatively correlated with the exposure level;
[0023] A recommendation score acquisition module, configured to acquire a recommendation score for the document to be pushed based on the first score and the second score;
[0024] The copy pushing module is used to push the copy to be pushed based on the recommendation score.
[0025] In a third aspect, an embodiment of the present application provides an electronic device comprising: a processor, a memory, and a communication bus, wherein the processor and the memory communicate with each other through the communication bus; the memory stores computer program instructions that can be executed by the processor, and when the computer program instructions are read and run by the processor, the method provided in the first aspect or any possible implementation of the first aspect is executed.
[0026] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which computer program instructions are stored. When the computer program instructions are read and run by a processor, the method provided by the first aspect or any possible implementation of the first aspect is executed.
[0027] Other features and advantages of the present application will be described in the following description and, in part, will become apparent from the description or be understood by practicing the embodiments of the present application. The objectives and other advantages of the present application can be achieved and obtained through the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.
[0029] Figure 1 A flowchart of the method for pushing a copy provided in an embodiment of the present application;
[0030] Figure 2 A schematic diagram of the grouping dimensions and grouping results in a certain scenario provided in an embodiment of the present application;
[0031] Figure 3 A flowchart of a data acquisition solution in a certain scenario provided in an embodiment of the present application;
[0032] Figure 4 A schematic diagram of the training process of a click-through rate prediction model in a certain scenario provided in an embodiment of the present application;
[0033] Figure 5 A schematic diagram of the structure of the document pushing device provided in an embodiment of the present application;
[0034] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0035] The following will describe the technical solutions in the embodiments of the present application in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present application and are therefore only examples and cannot be used to limit the scope of protection of the present application.
[0036] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned figure descriptions are intended to cover non-exclusive inclusions.
[0037] In the description of the embodiments of this application, the technical terms "first" and "second" are used only to distinguish different objects and should not be understood to indicate or imply relative importance or implicitly specify the quantity, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, the meaning of "plurality" is more than two, unless otherwise clearly and specifically defined.
[0038] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.
[0039] In the description of the embodiments of this application, the term "and / or" is simply a description of the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent the following three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.
[0040] Currently, intelligent push notification systems generally select the best copy from a pool of copy and push it to users. However, the accuracy of copy push is low due to the following problems:
[0041] (1) Difficulty in cold start of new copywriting:
[0042] The above-mentioned new copy refers to: copy that has never been exposed since its creation;
[0043] When cold-starting a new copy, the relevant technology generally directly provides a certain amount of random traffic support to the new copy, then estimates the click-through rate of the copy after random traffic support, and finally selects the copy with a higher estimated click-through rate for push.
[0044] The expected exposure level of a new copy is generally reversely estimated based on indicators such as the click-through rate of the intelligent push system. For example, if the business believes that a copy has received 10 click-through pv (PageView) and has passed the support, and the average click-through rate of the copy is 1%, then the support exposure level of the copy can be set to 1000 exposure pv.
[0045] The aforementioned random traffic support scheme for new content suffers from an irrational distribution of impressions, which can lead to low accuracy in content push notifications. For example, if an app has 90% male users and 10% female users, when randomly distributing 1,000 impression PVs, male users will receive 900, potentially generating a high click-through rate, while female users will receive only 100, resulting in a far lower click-through rate. Consequently, the accuracy of new content placement is low.
[0046] The above-mentioned unreasonable distribution is even more obvious in some vertical applications. For example, in an application specializing in selling clothing categories, 90% of the supporting traffic may be taken away by the clothing category, while some niche categories have not been fully explored.
[0047] (2) Inability to respond quickly at the end of the copywriting lifecycle:
[0048] Since copywriting has a life cycle, when the copywriting is at the end of its life cycle, the click-through rate of the copywriting may continue to decline, and the copywriting needs to be replaced and traffic needs to be reduced in a timely manner. The copywriting push solutions in related technologies cannot respond in time, thereby reducing the accuracy of the copywriting push.
[0049] Based on this, an embodiment of the present application provides a method for pushing a copy, which calculates a first score and a second score based on the real-time click-through rate and exposure level of the copy to be pushed, respectively, and obtains a recommended score for the copy to be pushed based on the first score and the second score, and finally pushes the copy to be pushed based on the recommended score that comprehensively considers the real-time click-through rate and exposure level of the copy. On the one hand, the above scheme can consider the exposure level of the copy under different groups to support the new copy, which is conducive to improving the problem of unreasonable distribution of the exposure level of the new copy, and thus improving the accuracy of the copy push; on the other hand, the above scheme can consider the real-time click-through rate of the copy to adjust the recommended score of the copy, which can achieve a rapid response to the copy push based on the real-time click-through rate of the copy, which is conducive to further improving the accuracy of the copy push.
[0050] The following is a detailed description of the above copywriting push method. Figure 1 , an embodiment of the present application provides a copywriting push method, the method comprising:
[0051] Step S110: Obtain the document to be pushed;
[0052] Step S120: Based on the first real-time click-through rate and the second real-time click-through rate of the copy to be pushed, obtaining an estimated click-through rate and a first score of the copy to be pushed; wherein the first real-time click-through rate is used to represent the overall real-time click-through rate of the copy to be pushed; the second real-time click-through rate is used to represent the real-time click-through rate of the copy to be pushed in each group; the first score is positively correlated with the estimated click-through rate;
[0053] Step S130: Based on the exposure level of the copy to be pushed in each group, obtain a second score of the copy to be pushed; wherein the second score is negatively correlated with the exposure level;
[0054] Step S140: Obtaining a recommendation score for the document to be pushed based on the first score and the second score;
[0055] Step S150: Push the content to be pushed based on the recommendation score.
[0056] For example, the content to be pushed obtained in step S110 may include new content that has never been exposed since its creation and / or content that has been exposed in the intelligent push system. The new content may be generated by an operator or an algorithm model.
[0057] The following describes the grouping in step S120:
[0058] The purpose of grouping users is to obtain the performance of the copy in different groups to determine whether the current exposure level is credible. The general method of group analysis is to analyze the exposure level, click-through rate and other indicators of the copy in different groups. In general, attributes with high discrimination can be used as grouping dimensions. For example, in a certain scenario, the click-through rate of male users is 1% and the click-through rate of female users is 3%. In this case, gender can be used as a grouping dimension. If the exposure and / or click-through rate of different genders are basically the same, grouping can be performed without gender. Grouping dimensions generally include:
[0059] (1) Crowd attributes: gender, age, device, user experience, user activity, etc.
[0060] (2) Material group attributes: product category, brand, etc.
[0061] The following describes an optional implementation of the grouping solution in the embodiment of the present application:
[0062] The clustering dimensions include:
[0063] (1) Gender: male and female;
[0064] (2) Age: young, middle-aged, and elderly;
[0065] (3) Brand: BrandA and BrandB.
[0066] Perform a Cartesian product operation on the three clustering dimensions above to obtain the final clustering. For the clustering results, see Figure 2 .
[0067] Before describing the specific implementation of obtaining the estimated click rate and the first score of the to-be-pushed copy in step S120 and the specific implementation of obtaining the second score of the to-be-pushed copy in step S130, the above-mentioned solution for obtaining various real-time data is first introduced:
[0068] As an optional implementation of the above-mentioned copy push method, the above-mentioned copy push method further includes: using a real-time message queue to calculate in real time the overall exposure of the copy, the overall click volume of the copy, the exposure of each group, the click volume of each group, the exposure of the copy in each group, and the click volume of the copy in each group. For example, this implementation method:
[0069] See Figure 3 By embedding points in the real-time message queue of exposure click operations, the following data can be obtained:
[0070] Data A: The overall exposure and click data of the copy, including the overall exposure and click volume of the copy;
[0071] The data format can be: copy ID_behavior type: data amount, for example:
[0072] Copy 1 exposure: 1000, Copy 1 clicks: 2, Copy 2 exposure: 1000, Copy 2 clicks: 5;
[0073] Data B: Exposure and click data for different groups, including exposure and click volume for each group;
[0074] The data format can be: group ID_behavior type: data amount, for example:
[0075] Group A exposure: 10,000, Group A clicks: 20, Group B exposure: 10,000, Group B clicks: 50;
[0076] Data C: The impression and click-through data of the copy in different groups, including the impression volume and click-through volume of the copy in each group;
[0077] The data format can be: copy ID_group ID_behavior type: data amount, for example:
[0078] Copywriting 1_Group A_Exposure: 100, Copywriting 1_Group A_Clicks: 0, Copywriting 2_Group B_Exposure: 100, Copywriting 2_Group B_Clicks: 1.
[0079] Exemplarily, the above data may be stored in a cache so that the above copy pushing method can quickly obtain the required data.
[0080] The above solution can quickly obtain various data in real time through the real-time message queue, so that the above copy push method can quickly respond to user behavior and update the copy push strategy in time, which is conducive to improving the copy push accuracy of the above copy push method.
[0081] The following describes the solution for obtaining the estimated click rate of the copy to be pushed in step S120:
[0082] As an optional implementation of the above-mentioned copy pushing method, the above-mentioned step S120 obtains the estimated click-through rate of the copy to be pushed based on the first real-time click-through rate and the second real-time click-through rate of the copy to be pushed, including: using the first real-time click-through rate and the second real-time click-through rate of the copy to be pushed as inputs of the click-through rate prediction model, and obtaining the estimated click-through rate of the copy to be pushed output by the click-through rate prediction model.
[0083] For example, the first real-time click-through rate can be calculated by dividing the total click volume of the copy by the total exposure volume of the copy. The second real-time click-through rate can be calculated by dividing the click volume of the copy in each group by the click volume of the copy in the corresponding group.
[0084] The above-mentioned click-through rate prediction model can adopt models such as deep learning models, logistic regression models, gradient boosting decision trees, and random forests.
[0085] The above solution uses a click-through rate prediction model and comprehensively considers the real-time click-through rate of the copy to be pushed to obtain the estimated click-through rate of the copy to be pushed. It can accurately capture the real-time performance of the copy, and then achieve timely response to user behavior, which is conducive to improving the accuracy of copy push.
[0086] As an optional implementation of the above-mentioned copy push method, the above-mentioned step S120 obtains the estimated click-through rate of the copy to be pushed based on the first real-time click-through rate and the second real-time click-through rate of the copy to be pushed, including: using the basic material characteristics, copy characteristics and user characteristics as inputs of the click-through rate estimation model, and obtaining the estimated click-through rate of the copy to be pushed output by the click-through rate estimation model; wherein the copy characteristics include features constructed based on the first real-time click-through rate and the second real-time click-through rate of the copy to be pushed. This implementation example:
[0087] Obtain basic material characteristics, document characteristics, and user characteristics, including:
[0088] Basic material features, including material categories, material title content segmentation, etc.
[0089] Copywriting features, mainly including: copywriting content segmentation, copywriting source, key attribute information, etc.
[0090] User characteristics mainly include: sociological statistical characteristics, user behavior characteristics, long-term and short-term category preference characteristics, browsing behavior sequence characteristics, etc. Among them, browsing behavior sequence characteristics mainly highlight copywriting and its attribute information.
[0091] See Figure 4 The following provides a training plan for the above-mentioned click-through rate prediction model in a certain scenario. In this scenario, the click-through rate prediction model uses a DNN model (Deep Neural Network, deep learning model). After feature engineering the above-mentioned basic material features, copywriting features, and user features, it is connected to a 256-128-64-1 DNN fully connected layer.
[0092] It is understandable that before inputting features into the CTR prediction model, different types of input features can be processed as follows:
[0093] (1) For int statistical value features, such as copy exposure PV, click PV, etc., the processing method is to log them;
[0094] (2) Double statistical value features, such as copywriting click-through rate, can be directly used as input features;
[0095] (3) For single-value ID features, such as copy ID, the ID can be Embedded, and the Embed dimension can be 64 dimensions;
[0096] (4) For multi-valued ID features, such as text segmentation, the processing method is to embed the ID and then perform pooling. The Emb dimension is set to 64 dimensions, and the pooling method can be average pooling (avg pooling).
[0097] It is understandable that when training the above click-through rate prediction model, the cross entropy loss function can be used to characterize the gap between the estimated click-through rate (y_predict) output by the model and the actual click (y_true). The expression of the loss function can be:
[0098] Loss=-y true *log(y predict )
[0099] The optimizer during training can use Adam (Adaptive Moment Estimation) as the optimizer to minimize the loss function.
[0100] In addition, it is understandable that the newly generated data every day can also be used to perform daily incremental training on the above-mentioned click rate prediction model so that the model can be updated in a timely manner.
[0101] Before launching the above-mentioned click-through rate prediction model, you can use the training set from the day before yesterday and before to train the model, and use the training set from yesterday to evaluate the model. The model evaluation metric can be the AUC metric.
[0102] In addition, it is understandable that in addition to using basic material features, copy features and user features as model input features, features such as context features and date can also be used as input features to obtain the estimated click-through rate of the copy.
[0103] The above scheme comprehensively considers the basic characteristics of the material, the characteristics of the copy and the characteristics of the user to obtain the estimated click-through rate, which is conducive to improving the estimation accuracy of the estimated click-through rate, and then improving the copy push accuracy of the above copy push method.
[0104] As an optional implementation method of the above-mentioned copy pushing method, the above-mentioned method uses the basic material characteristics, copy characteristics and user characteristics as the input of the click-through rate prediction model to obtain the estimated click-through rate of the copy to be pushed output by the click-through rate prediction model, including: splicing the basic material characteristics, copy characteristics and user characteristics of the copy to be pushed to obtain the splicing features; inputting the splicing features into the trained click-through rate prediction model to obtain the estimated click-through rate of the copy to be pushed output by the click-through rate prediction model.
[0105] It is understandable that the splicing methods of the above-mentioned basic material features, copywriting features and user features can adopt direct splicing, splicing based on attention mechanism and other splicing methods.
[0106] The above solution combines the basic material characteristics, copy characteristics and user characteristics and then inputs them into the click-through rate prediction model, which is conducive to providing more comprehensive and detailed information to the click-through rate prediction model, and is conducive to improving the estimation accuracy of the estimated click-through rate, thereby improving the copy push accuracy of the above copy push method.
[0107] The following describes the solution for obtaining the second score of the document to be pushed in step S130:
[0108] As an optional implementation of the above-mentioned copywriting push method, the above-mentioned step S130 includes: based on the exposure amount of each group and the exposure amount of the copywriting to be pushed in each group, using the group confidence interval upper bound (UpperConfidenceBound) algorithm, calculating the second score of the copywriting to be pushed. This implementation method is, for example:
[0109] Use the following formula to calculate the second score of the submitted copy (also known as the cold start score):
[0110]
[0111] Among them, n is the exposure under the group; n j It is the exposure of the copy in the corresponding group.
[0112] It is understandable that the efficiency performance of the copywriting in each group can be fully explored based on the above second score.
[0113] The above scheme uses the cluster confidence interval upper bound algorithm to calculate the second score of the copy to be pushed. On the one hand, it is conducive to effectively optimizing the copy selection strategy in the copy cold start phase, thereby improving the copy push accuracy of the above copy push method; on the other hand, it enables the above copy push method to be applicable to more application scenarios, which is conducive to improving the adaptability of the above copy push method.
[0114] The following describes the solution for obtaining the recommendation score in step S140:
[0115] As an optional implementation of the above-mentioned copywriting push method, the above-mentioned step S140 includes: obtaining a recommendation score of the copywriting to be pushed based on the first score, the second score, and the exploration coefficient; wherein the exploration coefficient is used to adjust the push priority between the first category of copywriting and the second category of copywriting; the first category of copywriting is copywriting with a lower exposure level; the second category of copywriting is copywriting with a higher estimated click-through rate. For example, this implementation method:
[0116] The recommended score of the copywriting is calculated according to the following formula:
[0117] score=ctr_score+α*coldstart_score
[0118] Among them, score is the recommendation score of the copy to be pushed; ctr_score is the first score of the copy to be pushed; coldstart_score is the second score; α is the exploration coefficient.
[0119] It is understandable that the above ctr_score can be directly set to the estimated click-through rate of the copy to be pushed.
[0120] It can be understood that the larger the exploration coefficient α is, the more the above-mentioned copy push method tends to explore copy with less exposure, which can help the intelligent push system fully explore new copy; the smaller the exploration coefficient α is, the more it tends to select deterministic copy with a higher estimated click-through rate, and the push efficiency is higher.
[0121] The exploration coefficient in the above scheme can effectively balance the relationship between exploring new copywriting and utilizing deterministic copywriting, which is conducive to improving the decision-making quality of the above copywriting push method, and further improving the copywriting push accuracy of the above copywriting push method.
[0122] The following describes the solution for pushing the copywriting in step S150:
[0123] For example, after obtaining the recommendation score of each copy, the copy can be sorted based on the recommendation score, and the top-ranked copy, that is, the one or several higher-ranked copies with the highest recommendation score, can be pushed.
[0124] See Figure 5 Based on the same inventive concept, the present application also provides a document pushing device 200, which includes:
[0125] The module 210 for acquiring the document to be pushed is used to acquire the document to be pushed;
[0126] The first score acquisition module 220 is configured to acquire an estimated click-through rate and a first score of the document to be pushed based on the first real-time click-through rate and the second real-time click-through rate of the document to be pushed; wherein the first real-time click-through rate is used to represent the overall real-time click-through rate of the document to be pushed; the second real-time click-through rate is used to represent the real-time click-through rate of the document to be pushed in each group; and the first score is positively correlated with the estimated click-through rate;
[0127] A second score acquisition module 230 is configured to acquire a second score of the document to be pushed based on the exposure level of the document to be pushed in each group; wherein the second score is negatively correlated with the exposure level;
[0128] A recommendation score acquisition module 240 is configured to acquire a recommendation score for the document to be pushed based on the first score and the second score;
[0129] The copy pushing module 250 is configured to push the copy to be pushed based on the recommendation score.
[0130] As an optional implementation of the above-mentioned copy pushing device, the above-mentioned first score acquisition module 220 is specifically used to: use the first real-time click rate and the second real-time click rate of the copy to be pushed as inputs of the click rate prediction model, and obtain the estimated click rate of the copy to be pushed output by the click rate prediction model.
[0131] As an optional implementation of the above-mentioned copy pushing device, the above-mentioned first score acquisition module 220 is specifically used to: use the basic material characteristics, copy characteristics and user characteristics as inputs of the click-through rate prediction model, and obtain the estimated click-through rate of the copy to be pushed output by the click-through rate prediction model; wherein, the copy characteristics include characteristics constructed based on the first real-time click-through rate and the second real-time click-through rate of the copy to be pushed.
[0132] As an optional implementation of the above-mentioned copy pushing device, the above-mentioned first score acquisition module 220 uses the basic material characteristics, copy characteristics and user characteristics as the input of the click-through rate prediction model to obtain the estimated click-through rate of the copy to be pushed output by the click-through rate prediction model, including: splicing the basic material characteristics, copy characteristics and user characteristics of the copy to be pushed to obtain splicing characteristics; inputting the splicing characteristics into the trained click-through rate prediction model to obtain the estimated click-through rate of the copy to be pushed output by the click-through rate prediction model.
[0133] As an optional implementation of the above-mentioned copy pushing device, the above-mentioned second score acquisition module 230 is specifically used to: calculate the second score of the copy to be pushed based on the exposure amount under each group and the exposure amount of the copy to be pushed under each group, using the group confidence interval upper bound algorithm.
[0134] As an optional implementation of the above-mentioned copy pushing device, the above-mentioned recommendation score acquisition module 240 is specifically used to: obtain the recommendation score of the copy to be pushed based on the first score, the second score and the exploration coefficient; wherein the exploration coefficient is used to adjust the push priority between the first category of copy and the second category of copy; the first category of copy is a copy with a lower exposure level; the second category of copy is a copy with a higher estimated click-through rate.
[0135] As an optional implementation of the above-mentioned document pushing device, the document pushing device 200 further includes:
[0136] The data acquisition module is used to use the real-time message queue to collect real-time statistics on the overall exposure of the copy, the overall click volume of the copy, the exposure of each group, the click volume of each group, the exposure of the copy in each group, and the click volume of the copy in each group.
[0137] Figure 6 This is a schematic diagram of an electronic device provided in an embodiment of the present application. Figure 6 The electronic device 300 includes a processor 310, a memory 320, and a communication interface 330. These components are interconnected and communicate with each other via a communication bus 340 and / or other forms of connection mechanisms (not shown).
[0138] The memory 320 includes one or more (only one is shown in the figure), which may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc. The processor 310 and other possible components can access the memory 320 and read and / or write data therein.
[0139] The processor 310 includes one or more (only one is shown in the figure), which can be an integrated circuit chip with signal processing capabilities. The above-mentioned processor 310 can be a general-purpose processor, including a central processing unit (CPU), a microcontroller unit (MCU), a network processor (NP), or other conventional processors; it can also be a special-purpose processor, including a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0140] Communication interface 330 includes one or more (only one is shown in the figure) interfaces that can be used to communicate directly or indirectly with other devices to exchange data. For example, communication interface 330 can be an Ethernet interface; a mobile communication network interface, such as a 3G, 4G, or 5G network interface; or other types of interfaces that have data transmission and reception capabilities.
[0141] One or more computer program instructions may be stored in the memory 320 , and the processor 310 may read and execute these computer program instructions to implement the document push method and other desired functions provided in the embodiments of the present application.
[0142] I understand. Figure 6 The structure shown is for illustration only. The electronic device 300 may also include Figure 6 More or fewer components than shown, or with Figure 6 Different configurations shown. Figure 6 Each component shown in the figure can be implemented using hardware, software, or a combination thereof. For example, the electronic device 300 can be a single server (or other device with computing processing capabilities), a combination of multiple servers, a cluster of a large number of servers, etc., and can be both a physical device and a virtual device.
[0143] The present application also provides a computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are read and executed by a computer processor, the method for pushing the document provided in the present application is executed. For example, the computer-readable storage medium can be implemented as Figure 6 The memory 320 in the electronic device 300.
[0144] In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interface, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
[0145] In addition, the units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0146] Furthermore, the functional modules in each embodiment of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.
[0147] The above description is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, various modifications and variations of the present application are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.
Claims
1. A copywriting push method, characterized in that: The method comprises: Get the copy to be pushed; Based on the first real-time click rate and the second real-time click rate of the copy to be pushed, an estimated click rate and a first score of the copy to be pushed are obtained; wherein the first real-time click rate is used to represent the overall real-time click rate of the copy to be pushed; the second real-time click rate is used to represent the real-time click rate of the copy to be pushed in each group; the first score is positively correlated with the estimated click rate; Based on the exposure level of the to-be-pushed document in each group, obtaining a second score of the to-be-pushed document; wherein the second score is negatively correlated with the exposure level; Based on the first score and the second score, obtaining a recommendation score for the copy to be pushed; Based on the recommendation score, the to-be-pushed copy is pushed.
2. The method for pushing a copywriting according to claim 1, characterized in that: The obtaining the estimated click rate of the copy to be pushed based on the first real-time click rate and the second real-time click rate of the copy to be pushed includes: The first real-time click rate and the second real-time click rate of the text to be pushed are used as inputs of a click rate prediction model to obtain an estimated click rate of the text to be pushed output by the click rate prediction model.
3. The method for pushing a copywriting according to claim 1, characterized in that: The obtaining the estimated click rate of the copy to be pushed based on the first real-time click rate and the second real-time click rate of the copy to be pushed includes: Using basic material characteristics, text characteristics and user characteristics as inputs of a click-through rate prediction model, the estimated click-through rate of the text to be pushed output by the click-through rate prediction model is obtained; wherein the text characteristics include characteristics constructed based on a first real-time click-through rate and a second real-time click-through rate of the text to be pushed.
4. The method for pushing a copy of the document according to claim 3, characterized in that: The method of taking the basic material features, the copy features and the user features as inputs of the click rate prediction model and obtaining the estimated click rate of the copy to be pushed output by the click rate prediction model includes: Splicing the basic material features, the copy features and the user features of the copy to be pushed to obtain splicing features; The splicing features are input into the trained click-through rate prediction model to obtain the estimated click-through rate of the to-be-pushed copy output by the click-through rate prediction model.
5. The method for pushing a copywriting according to any one of claims 1 to 4, characterized in that: The acquiring a second score of the to-be-pushed document based on the exposure level of the to-be-pushed document in each group includes: Based on the exposure amount in each group and the exposure amount of the to-be-pushed copy in each group, a second score of the to-be-pushed copy is calculated using a grouping confidence interval upper bound algorithm.
6. The method for pushing a copywriting according to any one of claims 1 to 4, characterized in that: The obtaining, based on the first score and the second score, a recommendation score of the to-be-pushed document includes: Based on the first score, the second score and the exploration coefficient, the recommendation score of the to-be-pushed copy is obtained; wherein the exploration coefficient is used to adjust the push priority between the first category of copy and the second category of copy; the first category of copy is copy with a lower exposure level; the second category of copy is copy with a higher estimated click-through rate.
7. The method for pushing a copywriting according to any one of claims 1 to 4, characterized in that: The method further comprises: Utilize the real-time message queue to collect statistics on the overall exposure of the copy, the overall click volume of the copy, the exposure of each group, the click volume of each group, the exposure of the copy in each group, and the click volume of the copy in each group in real time.
8. A copywriting recommendation device, characterized in that: include: A module for obtaining documents to be pushed, used to obtain documents to be pushed; A first score acquisition module is used to acquire an estimated click rate and a first score of the copy to be pushed based on a first real-time click rate and a second real-time click rate of the copy to be pushed; wherein the first real-time click rate is used to represent the overall real-time click rate of the copy to be pushed; the second real-time click rate is used to represent the real-time click rate of the copy to be pushed in each group; and the first score is positively correlated with the estimated click rate; A second score acquisition module, configured to acquire a second score of the to-be-pushed document based on the exposure level of the to-be-pushed document in each group; wherein the second score is negatively correlated with the exposure level; A recommendation score acquisition module, configured to acquire a recommendation score of the to-be-pushed document based on the first score and the second score; The copy pushing module is used to push the copy to be pushed based on the recommendation score.
9. An electronic device, characterized in that: include: A processor, a memory and a communication bus, wherein the processor and the memory communicate with each other via the communication bus; The memory stores program instructions executable by the processor, and the processor can execute the method according to any one of claims 1 to 7 by calling the program instructions.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a computer, the computer is caused to perform the method according to any one of claims 1 to 7.