Advertisement pushing method and device, computer equipment, storage medium and computer program product

Abstract

The invention relates to an advertisement pushing method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: acquiring an object information feature corresponding to a to-be-pushed object identifier; advertisement features of all candidate advertisements corresponding to the to-be-pushed object identifier are obtained, all the candidate advertisements comprise unconverted advertisements and converted advertisements, the advertisement features of the converted advertisements comprise converted advertisement information features and converted features, and the converted features are obtained by conducting feature extraction on conversion behavior information of the converted advertisements; carrying out advertisement conversion possibility calculation on the basis of the object information features and the advertisement features to obtain the conversion possibility of each candidate advertisement; and screening each candidate advertisement based on the conversion possibility to obtain a to-be-pushed advertisement, and pushing the to-be-pushed advertisement to a terminal corresponding to the to-be-pushed object identifier. By adopting the method, the accuracy of advertisement pushing can be improved.

Description

Technical Field

[0001] This application relates to the field of information push technology, and in particular to an advertising push method, apparatus, computer equipment, storage medium and computer program product. Background Technology

[0002] With the development of ad push technology, in ad push services, when an ad recipient converts an ad and requests another ad push, the already converted ad is often expected to have a higher conversion rate, leading to duplicate pushes. However, some ads are unsuitable for repeated pushes, such as application registration ads. Pushing a registration ad for the same application again after a user has already registered would be inappropriate. Pushing a converted ad again would waste resources and reduce the accuracy of ad pushes. Summary of the Invention

[0003] Therefore, it is necessary to provide an advertising push method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can save push resources and improve the accuracy of advertising push, in order to address the above-mentioned technical problems.

[0004] Firstly, this application provides an advertising push method. The method includes:

[0005] Obtain the object information features corresponding to the identifier of the object to be pushed;

[0006] Obtain the advertising features of each candidate ad corresponding to the identifier of the target to be pushed. Each candidate ad includes unconverted ads and converted ads. The advertising features of converted ads include converted ad information features and converted features. The converted features are obtained by feature extraction of the conversion behavior information of converted ads. The advertising features of unconverted ads include unconverted ad information features.

[0007] The conversion probability of an ad is calculated based on the object information features and ad features to obtain the conversion probability of each candidate ad. The conversion probability is used to characterize the likelihood that the target object will convert the candidate ad.

[0008] Based on the conversion probability, each candidate ad is filtered to obtain the ad to be pushed, and the ad to be pushed is pushed to the terminal corresponding to the target audience identifier.

[0009] Secondly, this application also provides an advertising push device. The device includes:

[0010] The object feature acquisition module is used to acquire the object information features corresponding to the identifier of the object to be pushed;

[0011] The advertising feature acquisition module is used to acquire the advertising features of each candidate advertisement corresponding to the identifier of the target to be pushed. Each candidate advertisement includes unconverted advertisements and converted advertisements. The advertising features of converted advertisements include converted advertisement information features and converted features. The converted features are obtained by feature extraction of the conversion behavior information of converted advertisements. The advertising features of unconverted advertisements include unconverted advertisement information features.

[0012] The conversion calculation module is used to calculate the conversion probability of advertisements based on object information features and advertisement features, and obtain the conversion probability of each candidate advertisement. The conversion probability is used to characterize the likelihood that the target object will convert the candidate advertisement.

[0013] The push module is used to filter each candidate advertisement based on the conversion probability, obtain the advertisement to be pushed, and push the advertisement to be pushed to the terminal corresponding to the target object identifier.

[0014] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0015] Obtain the object information features corresponding to the identifier of the object to be pushed;

[0016] Obtain the advertising features of each candidate ad corresponding to the identifier of the target to be pushed. Each candidate ad includes unconverted ads and converted ads. The advertising features of converted ads include converted ad information features and converted features. The converted features are obtained by feature extraction of the conversion behavior information of converted ads. The advertising features of unconverted ads include unconverted ad information features.

[0017] The conversion probability of an ad is calculated based on the object information features and ad features to obtain the conversion probability of each candidate ad. The conversion probability is used to characterize the likelihood that the target object will convert the candidate ad.

[0018] Based on the conversion probability, each candidate ad is filtered to obtain the ad to be pushed, and the ad to be pushed is pushed to the terminal corresponding to the target audience identifier.

[0019] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0020] Obtain the object information features corresponding to the identifier of the object to be pushed;

[0021] Obtain the advertising features of each candidate ad corresponding to the identifier of the target to be pushed. Each candidate ad includes unconverted ads and converted ads. The advertising features of converted ads include converted ad information features and converted features. The converted features are obtained by feature extraction of the conversion behavior information of converted ads. The advertising features of unconverted ads include unconverted ad information features.

[0022] The conversion probability of an ad is calculated based on the object information features and ad features to obtain the conversion probability of each candidate ad. The conversion probability is used to characterize the likelihood that the target object will convert the candidate ad.

[0023] Based on the conversion probability, each candidate ad is filtered to obtain the ad to be pushed, and the ad to be pushed is pushed to the terminal corresponding to the target audience identifier.

[0024] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps:

[0025] Obtain the object information features corresponding to the identifier of the object to be pushed;

[0026] Obtain the advertising features of each candidate ad corresponding to the identifier of the target to be pushed. Each candidate ad includes unconverted ads and converted ads. The advertising features of converted ads include converted ad information features and converted features. The converted features are obtained by feature extraction of the conversion behavior information of converted ads. The advertising features of unconverted ads include unconverted ad information features.

[0027] The conversion probability of an ad is calculated based on the object information features and ad features to obtain the conversion probability of each candidate ad. The conversion probability is used to characterize the likelihood that the target object will convert the candidate ad.

[0028] Based on the conversion probability, each candidate ad is filtered to obtain the ad to be pushed, and the ad to be pushed is pushed to the terminal corresponding to the target audience identifier.

[0029] The aforementioned advertising push method, apparatus, computer equipment, storage medium, and computer program product acquire object information features corresponding to the identifier of the object to be pushed; acquire advertising features of each candidate advertisement corresponding to the identifier of the object to be pushed, wherein each candidate advertisement includes unconverted advertisements and converted advertisements, the advertising features of converted advertisements include converted advertisement information features and converted features, the converted features are obtained by feature extraction of the conversion behavior information of converted advertisements, and the advertising features of unconverted advertisements include unconverted advertisement information features; and calculate the conversion probability of each candidate advertisement based on the object information features and advertising features, thereby obtaining the conversion probability of each candidate advertisement, the conversion probability being used to characterize the likelihood of the identifier of the object to be pushed converting the candidate advertisement. This method calculates ad conversion probability by using both converted ad information features and conversion features of converted ads. This avoids the problem of overestimating conversions when using only ad information features. The addition of conversion features allows for a more granular representation of converted ads, thus improving the accuracy of the conversion probability of the obtained converted ads. This, in turn, improves the accuracy of the conversion probability of each candidate ad. Then, each candidate ad is filtered according to its conversion probability to obtain the ads to be pushed. These ads are then pushed to the terminals corresponding to the target audience identifiers. This reduces the duplication of converted ads, saves push resources, and improves the accuracy of ad push. Attached Figure Description

[0030] Figure 1 This is an application environment diagram of the advertising push method in one embodiment;

[0031] Figure 2 This is a flowchart illustrating an advertising push method in one embodiment;

[0032] Figure 3 This is a schematic diagram of the ad filtering process in one embodiment;

[0033] Figure 4 This is a schematic diagram illustrating the principle of calculating the conversion probability in one embodiment;

[0034] Figure 5 This is a schematic diagram of the architecture of a conversion probability prediction model in one embodiment;

[0035] Figure 6 This is a schematic diagram illustrating the process of obtaining the advertisement to be pushed in one embodiment;

[0036] Figure 7 This is a schematic diagram illustrating the process of obtaining the advertisement to be pushed in another embodiment;

[0037] Figure 8 This is a schematic diagram of the ad push process in a specific embodiment;

[0038] Figure 9 This is a schematic diagram of the architecture of an ad push method in a specific embodiment;

[0039] Figure 10 This is a schematic diagram of a page displaying an advertisement in a specific embodiment;

[0040] Figure 11 This is a schematic diagram of a video advertisement in a specific embodiment;

[0041] Figure 12 This is a structural block diagram of an advertising push device in one embodiment;

[0042] Figure 13 This is an internal structural diagram of a computer device in one embodiment;

[0043] Figure 14 This is a diagram of the internal structure of a computer device in another embodiment. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0045] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0046] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, pre-trained model technology, operating / interactive systems, and mechatronics. Pre-trained models, also known as large-scale models or foundational models, can be widely applied to downstream tasks across various AI fields after fine-tuning. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0047] Natural Language Processing (NLP) is an important field within computer science and artificial intelligence. It studies the theories and methods for enabling effective communication between humans and computers using natural language. NLP involves natural language—the language people use in daily life—and is closely related to linguistics; it also involves computer science and mathematics, making it a science that integrates these three disciplines. Therefore, research in this field involves natural language, i.e., the language people use in daily life, and thus has a close connection with linguistics. Pre-trained models, a crucial technique for model training in artificial intelligence, evolved from large language models in NLP. After fine-tuning, large language models can be widely applied to downstream tasks. NLP techniques typically include text processing, semantic understanding, machine translation, question answering, and knowledge graphs.

[0048] The solutions provided in this application involve technologies such as semantic understanding in artificial intelligence, which are specifically illustrated through the following embodiments:

[0049] The advertising push method provided in this application embodiment can be applied to, for example, Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on another server. Server 104 can receive advertising push requests sent by terminal 102 corresponding to the identifier of the object to be pushed, and in response to the advertising push request, server 104 retrieves the object information features corresponding to the identifier of the object to be pushed from the data storage system. Server 104 can obtain the advertising features of each candidate advertisement corresponding to the target object identifier from the data storage system. Each candidate advertisement includes unconverted and converted advertisements. The advertising features of converted advertisements include converted advertisement information features and converted features. The converted features are obtained by extracting features from the conversion behavior information of converted advertisements. The advertising features of unconverted advertisements include unconverted advertisement information features. Server 104 calculates the conversion probability of each candidate advertisement based on the target information features and advertising features. The conversion probability is used to characterize the likelihood that the target object identifier will convert to the candidate advertisement. Server 104 filters each candidate advertisement based on the conversion probability to obtain the advertisement to be pushed, and pushes the advertisement to be pushed to the terminal 102 corresponding to the target object identifier. The terminal can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. Portable wearable devices can be smartwatches, smart bracelets, head-mounted devices, etc. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The terminal and server 106 can be connected directly or indirectly through wired or wireless communication, which is not limited herein.

[0050] In one embodiment, such as Figure 2 As shown, an ad push method is provided, which is applied to... Figure 1 Taking a server as an example, it can be understood that this method can also be applied to a terminal, and also to a system that includes both a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0051] S202, Obtain the object information features corresponding to the identifier of the object to be pushed.

[0052] The target object identifier uniquely identifies the object to which the advertisement will be pushed. This object can be a real object, such as a real person, or a virtual object, such as a virtual robot, virtual idol, or virtual person. Object information features are obtained by extracting features from information representing the object. This information can include the object's attribute information, behavioral information, etc. The object's attribute information represents its specific attributes; for example, if the object is a person, the attribute information could be basic human attributes. The object's behavioral information can be obtained based on the object's historical interactions with advertisements. For example, the object's behavioral information can be information about its past interactions with advertisements, such as the types of advertisements the object interacted with in the past, and the types of behaviors the object interacted with in the past.

[0053] Specifically, the server can periodically push advertisements, for example, at a fixed time each day. When the server detects that the current time corresponds to the scheduled push time, it retrieves the identifiers of the objects to be pushed to from the database. These identifiers can include multiple identifiers, and the server can then search the database for the corresponding object information features. Alternatively, the server can push advertisements when it receives an advertisement push request from a terminal corresponding to an object identifier. In other words, the server can receive advertisement push requests in real time, and these requests can carry the identifiers of the objects to be pushed. The server can simultaneously receive a large number of advertisement push requests and respond to each one concurrently. It then searches the database for the corresponding object information features based on the identifier of the object to be pushed in the advertisement push request.

[0054] In one embodiment, the server can obtain object information corresponding to the identifier of the object to be pushed, and then perform feature extraction on the object information to obtain object information features. The server can use a neural network for feature extraction to extract features from the object information. Alternatively, the server can use a feature extraction algorithm to extract features from the object information; for example, it can use a vectorization algorithm to vectorize the object information to obtain object information features. In another embodiment, the server can also obtain object information features corresponding to the identifier of the object to be pushed uploaded by the terminal. This terminal can be the terminal corresponding to the identifier of the object to be pushed, or it can be the management terminal for advertising push.

[0055] S204, obtain the advertising features of each candidate advertisement corresponding to the identifier of the target to be pushed. Each candidate advertisement includes unconverted advertisements and converted advertisements. The advertising features of converted advertisements include converted advertisement information features and converted features. The converted features are obtained by feature extraction of the conversion behavior information of converted advertisements. The advertising features of unconverted advertisements include unconverted advertisement information features.

[0056] Candidate ads refer to ads that are shortlisted for consideration before being pushed to the target audience. These are online ads, meaning they are disseminated through the internet, including various websites, web pages, and internet applications. Candidate ads can be of different types, such as ads for product transactions, app downloads, promotions, services, or events. They can also appear in different scenarios, such as search engines, social media platforms, applications, or email. Furthermore, they can be in various media formats, including images, text, video, and audio. Finally, they can be displayed as banners, pop-ups, or links.

[0057] A converted ad is an ad where a conversion has already occurred between the target audience and the ad. This conversion refers to the specific behavior the advertiser expects the ad to achieve, such as a click, download, registration, activation, transaction, or following. Converted ad information features are obtained by extracting features from the ad information of converted ads. Ad information represents the specific content of the ad; for example, product ad information might be product descriptions, while game ad information might include game descriptions and download links. Conversion features are obtained by extracting features from the conversion behavior information of converted ads. Conversion behavior information refers to information obtained based on conversion behavior; for example, conversion behavior information can typically be represented as (ad identifier, target identifier, conversion time point, conversion behavior). The ad identifier is used to uniquely identify the corresponding ad. Conversion behavior information can determine the conversion behavior and time of the target audience's interaction with the ad. A non-converted ad is an ad where no conversion has occurred between the target audience and the ad; that is, the target audience has not performed any conversion action on the ad before the current time point. A non-converted ad can be any ad other than a converted ad. The features of unconverted ad information are obtained by extracting features from the ad information of unconverted ads.

[0058] Specifically, the server can pre-extract the ad features of unconverted and converted ads corresponding to the object identifier, and save all ad features associating them with their corresponding object identifiers in the database. Then, when an ad needs to be pushed, the server searches the database for a matching object identifier based on the object identifier to be pushed, and then searches the database for the corresponding unconverted and converted ad features based on the matching object identifier, thus obtaining the ad features of each candidate ad corresponding to the object identifier to be pushed.

[0059] In one embodiment, the server can look up the corresponding converted ad identifiers and non-converted ad identifiers in the database based on the identifier of the target object. The server can consider any ad in the database that has not yet generated a conversion for the target object as a non-converted ad. Then, based on the identifier of the converted ad, the server looks up the ad information and conversion behavior information of the converted ad in the database. Simultaneously, based on the ad identifier of the non-converted ad, the server looks up the ad information of the non-converted ad. Then, the server extracts features from the ad information and conversion behavior information of the converted ad to obtain converted ad information features and converted features. Simultaneously, the server extracts features from the ad information of the non-converted ad to obtain non-converted ad information features. At this point, the server obtains the ad features of each candidate ad corresponding to the target object identifier, and these candidate ads include both converted and non-converted ads.

[0060] In one embodiment, the server can obtain the ad features of multiple unconverted ads and multiple converted ads based on the identifier of the target object, thus obtaining the ad features of each candidate ad. These multiple features can be at least two. The server can obtain all historical converted ads stored in the database for the target object, and it can also obtain all converted ads within a long historical period, such as all converted ads within the past six months, all converted ads within the past three months, etc.

[0061] S206. Based on object information features and advertising features, calculate the conversion probability of each candidate advertisement to obtain the conversion probability of each candidate advertisement. The conversion probability is used to characterize the likelihood that the target object will convert the candidate advertisement.

[0062] Here, conversion probability refers to the likelihood that a target user will convert their view after a candidate ad is pushed to a terminal corresponding to the conversion probability. The higher the conversion probability, the more likely the target user is to convert their view after the ad is pushed. The conversion probability of each candidate ad includes the conversion probability of converted ads and the conversion probability of non-converted ads. The conversion probability of non-converted ads represents the likelihood that the target user will initially convert their view on a candidate ad. The conversion probability of converted ads represents the likelihood that the target user will convert their view on a converted ad again.

[0063] Specifically, the server uses the object information features corresponding to the target object identifier and the advertising features of each candidate ad to calculate the ad conversion probability. The server can perform this calculation sequentially, that is, using the object information features and the advertising features of each candidate ad one by one to calculate the ad conversion probability. Alternatively, the server can perform the calculation in parallel, using the object information features and the advertising features of each candidate ad simultaneously to calculate the ad conversion probability for each candidate ad. The server can also use a linear regression algorithm to calculate the conversion probability using both object information features and ad features, or it can use a neural network to calculate the conversion probability using both object information features and ad features.

[0064] S208: Based on the conversion probability, each candidate advertisement is filtered to obtain the advertisement to be pushed, and the advertisement to be pushed is pushed to the terminal corresponding to the target object identifier.

[0065] Among them, the advertisement to be pushed refers to the candidate advertisement that needs to be pushed to the terminal corresponding to the target identifier. The advertisement to be pushed can be an unconverted advertisement or a converted advertisement.

[0066] Specifically, the server can compare the conversion probabilities of each candidate ad and select the one with the highest conversion probability as the ad to be pushed. Alternatively, the server can rank the candidate ads based on their conversion probabilities and select the top-ranked ads as the ads to be pushed; for example, the top-ranked candidate ad. Or, the server can select a preset number of top-ranked candidate ads as the ads to be pushed. This preset number can be determined based on the settings of the target audience; for example, if the target audience is set to push three ads, the server can select the top three candidate ads as the ads to be pushed. Finally, the server pushes the ads to the terminals corresponding to the target audience identifiers. The terminals receive the ads pushed by the server and display them.

[0067] The aforementioned ad push method involves: obtaining object information features corresponding to the object identifier to be pushed; obtaining ad features for each candidate ad corresponding to the object identifier to be pushed, where each candidate ad includes unconverted ads and converted ads; the ad features for converted ads include converted ad information features and converted features, where converted features are obtained by feature extraction of the conversion behavior information of converted ads; and the ad features for unconverted ads include unconverted ad information features. Based on the object information features and ad features, the ad conversion probability is calculated to obtain the conversion probability of each candidate ad. The conversion probability is used to characterize the likelihood that the object identifier to be pushed will convert the candidate ad. This method calculates ad conversion probability by using both converted ad information features and conversion features of converted ads. This avoids the problem of overestimating conversions when using only ad information features. The addition of conversion features allows for a more granular representation of converted ads, thus improving the accuracy of the conversion probability of the obtained converted ads. This, in turn, improves the accuracy of the conversion probability of each candidate ad. Then, each candidate ad is filtered according to its conversion probability to obtain the ads to be pushed. These ads are then pushed to the terminals corresponding to the target audience identifiers. This reduces the duplication of converted ads, saves push resources, and improves the accuracy of ad push.

[0068] In one embodiment, S204, namely obtaining the advertising features of each candidate advertisement corresponding to the identifier of the object to be pushed, includes:

[0069] Obtain the advertising information of unconverted ads from each candidate ad group, extract features from the advertising information of unconverted ads to obtain the features of unconverted ads; obtain the advertising information and conversion behavior information of converted ads from each candidate ad group; extract features from the advertising information of converted ads to obtain the features of converted ads, and extract features from the conversion behavior information of converted ads to obtain the features of converted ads.

[0070] In this embodiment, the server can obtain the advertising information of unconverted ads corresponding to the target object identifier in real time from the database, and then use a feature extraction algorithm to extract features from the unconverted ad information to obtain the unconverted ad information features. Simultaneously, the server can obtain the advertising information of converted ads and the conversion behavior information of converted ads corresponding to the target object identifier in real time from the database, and then use a feature extraction algorithm to extract features from the converted ad information and the conversion behavior information of converted ads respectively to obtain converted ad information features and converted features. The feature extraction algorithm can be obtained using a vectorization algorithm or a neural network algorithm. In one embodiment, when the advertising information is text, a text vectorization algorithm can be used for feature extraction, such as a hot coding algorithm, a bag-of-words model algorithm, a machine learning algorithm, etc. In one embodiment, when the advertising information is an image, image features can be extracted using techniques such as color histograms, texture analysis, and edge detection, or image features can be extracted using a neural network algorithm, such as a convolutional neural network, a feedforward neural network, a recurrent neural network, etc.

[0071] In the above embodiments, advertising information of unconverted ads, advertising information of converted ads, and conversion behavior information of converted ads are acquired. Then, feature extraction is performed on the acquired advertising information and conversion behavior information respectively to obtain advertising information features and conversion features. That is, feature extraction using real-time acquired advertising information and conversion behavior information improves the accuracy of the obtained features.

[0072] In one embodiment, converted ads include at least two;

[0073] like Figure 3 As shown, the process involves obtaining advertising information and conversion behavior information of converted ads from each candidate ad group, including:

[0074] S302, obtain at least two converted ads corresponding to the identifier of the target to be pushed and the conversion time points corresponding to the at least two converted ads respectively.

[0075] S304: Filter at least two converted ads according to the conversion time point and the preset conversion filtering time period to obtain the target converted ads.

[0076] S306, Obtain advertising information and conversion behavior information of the target converted ads.

[0077] The conversion time point refers to the point in time when a conversion occurs. Different converted ads can have the same conversion time point or different conversion time points. The conversion behavior corresponding to the same ad can be the same or different. For example, the conversion behavior corresponding to a product transaction ad can be set as a click behavior or as a transaction behavior. Usually, an ad is set to only one conversion behavior, and the conversion behavior corresponding to a product transaction ad is usually set to a transaction behavior.

[0078] The preset conversion filtering time period refers to a pre-set time frame during which converted ads are filtered out. When the conversion time of a converted ad falls within this preset conversion filtering time period, it needs to be filtered out from all candidate ads. In other words, converted ads whose conversion time falls within the preset conversion filtering time period are not considered as candidate ads. Typically, this preset conversion filtering time period is set to a period prior to the current time. For example, setting it to 24 hours prior to the current time means that ads that have already converted within those 24 hours do not need to be pushed again and should be filtered out when pushing ads at the current time. Target converted ads refer to converted ads whose conversion time does not fall within the preset conversion filtering time period.

[0079] Specifically, the server retrieves the identifiers of each converted ad corresponding to the identifier of the target object from the database, and also retrieves the conversion time point corresponding to the identifier of each converted ad. Then, it compares each conversion time point with a preset conversion filtering time period to determine if the conversion time point is within the preset conversion filtering time period. If the conversion time point is within the preset conversion filtering time period, the converted ad corresponding to that conversion time point is filtered out from all candidate ads. If the conversion time point is not within the preset conversion filtering time period, the converted ad corresponding to that conversion time point is taken as the target converted ad. Then, based on the identifier of the target converted ad, the server retrieves the ad information and conversion behavior information of the target converted ad from the database.

[0080] In one embodiment, the server can filter at least two converted ads based on conversion time, preset conversion behavior, and preset conversion filtering time period to obtain the target converted ad. The preset conversion behavior refers to the pre-set conversion behavior expected to occur for the ad to be pushed. Specifically: the server obtains the conversion time and conversion behavior of each converted ad, then compares the conversion time with the preset conversion filtering time period and the conversion behavior with the preset conversion behavior. When the conversion time is within the preset conversion filtering time period and the conversion behavior is the same as the preset conversion behavior, the converted ad corresponding to that conversion time is filtered out from the candidate ads. When the conversion time is not within the preset conversion filtering time period or the conversion behavior is not the same as the preset conversion behavior, the converted ad corresponding to that conversion time is taken as the target converted ad. That is, by filtering converted ads that are within the preset conversion filtering time period and have the same preset conversion behavior, the server can avoid filtering converted ads with different conversion behaviors, thereby improving the accuracy of converted ad filtering. This also prevents the repeated pushing of converted ads with the same conversion behavior within a short period, improving the accuracy of ad push.

[0081] In the above embodiments, at least two converted ads are filtered according to the conversion time point and a preset conversion filtering time period to obtain the target converted ad. The target converted ad is then used as a candidate ad, thereby filtering out converted ads within the preset conversion filtering time period and avoiding repeated pushes of converted ads within that time period, thus improving the accuracy of ad pushes.

[0082] In one embodiment, S206, the conversion probability of an advertisement is calculated based on object information features and advertisement features to obtain the conversion probability of each candidate advertisement, including the following steps:

[0083] The conversion probability of ads is calculated based on the characteristics of the object information and the characteristics of the unconverted ads, and the conversion probability of the unconverted ads is obtained. The conversion probability of ads is calculated based on the characteristics of the object information, the characteristics of the converted ads, and the converted features, and the conversion probability of the converted ads is obtained.

[0084] In this embodiment, when calculating the conversion probability of unconverted ads, the server can use object information features and unconverted ad information features to calculate the probability that the target audience will initially convert to the unconverted ad, i.e., the conversion probability of the unconverted ad. When calculating the conversion probability of converted ads, the server uses object information features, converted ad information features, and converted features to calculate the probability that the target audience will convert to the converted ad again, i.e., the conversion probability of the converted ad. A general neural network model for predicting ad conversion probability can be used to calculate the ad conversion probability. This neural network model can be a conversion rate prediction model, which can be built using a convolutional neural network. If the conversion behavior is a click behavior, the neural network model for predicting ad conversion probability can also be a click-through rate prediction model, which can also be built using a convolutional neural network.

[0085] In one embodiment, statistical characteristics of the target audience and the advertisements can also be obtained. These statistical characteristics include the number of advertisements the target audience has historically clicked or converted from. Then, using the target audience information characteristics, unconverted advertisement information characteristics, and statistical characteristics, the conversion probability of the advertisements is calculated to obtain the conversion probability of the unconverted advertisements. Simultaneously, using the target audience information characteristics, converted advertisement information characteristics, statistical characteristics, and converted characteristics, the conversion probability of the advertisements is calculated to obtain the conversion probability of the converted advertisements. By adding statistical characteristics to the calculation of the conversion probability of advertisements, the feature information is enhanced, thereby improving the accuracy of the obtained conversion probability.

[0086] In the above embodiments, by calculating the conversion probability of unconverted ads using object information features and unconverted ad information features, and by calculating the conversion probability of converted ads using object information features, converted ad information features, and converted features, the problem of overestimating the conversion probability of converted ads can be reduced while ensuring the accuracy of unconverted ad calculations, thus improving the accuracy of conversion probability calculations. Then, ads are pushed according to the conversion probability of each candidate ad, further improving the accuracy of ad information delivery.

[0087] In one embodiment, S206, the conversion probability of an advertisement is calculated based on object information features and advertisement features to obtain the conversion probability of each candidate advertisement, including the following steps:

[0088] The object information features and the advertisement features are reduced in dimensionality to obtain the target object features and the target advertisement features. The target object features and the target advertisement features are concatenated to obtain the concatenated features. Based on the concatenated features, feature semantic interaction is performed to obtain the interaction features. Based on the interaction features, a full connection operation is performed to obtain the conversion probability of each candidate advertisement.

[0089] In this embodiment, target object features refer to the features of the object information obtained after dimensionality reduction. Target advertisement features refer to the features of the advertisement information obtained after dimensionality reduction. Concatenation features refer to the features obtained by concatenating the target object features and target advertisement features end-to-end. Interaction features refer to the features obtained by performing feature semantic interaction between the target object features and target advertisement features. For example... Figure 4 The diagram illustrates the principle of calculating conversion probability. Specifically, the server acquires object information features and advertisement features, and then performs dimensionality reduction on both. Dimensionality reduction can be achieved using a pooling neural network, which increases the receptive field, allowing the convolutional layer to see more information. Next, the server concatenates the target object features and target advertisement features. This can be done by using the target object features as the first part and the target advertisement features as the last part, or vice versa, resulting in concatenated features. Finally, the server performs feature semantic interaction based on the concatenated features. This can be achieved using a neural network that performs explicit high-order feature crosses, resulting in interactive features. Finally, a fully connected neural network is used to perform fully connected operations on the interactive features. This fully connected neural network performs regression through feature extraction, typically using the sigmoid function as the output, to obtain the conversion probability.

[0090] In the above embodiments, by reducing the dimensionality of object information features and advertising features respectively, target object features and target advertising features are obtained. These features are then concatenated to obtain concatenated features. Feature semantic interaction is performed based on these concatenated features to obtain interactive features. Finally, a fully connected operation is performed based on these interactive features to obtain the conversion probability of each candidate advertisement. In other words, through feature semantic interaction, deeper semantic information can be learned, improving the accuracy of the obtained interactive features. Then, the conversion probability is calculated based on these interactive features, further improving the accuracy of the obtained conversion probability.

[0091] In one embodiment, S206, the conversion probability of an advertisement is calculated based on object information features and advertisement features to obtain the conversion probability of each candidate advertisement, including the following steps:

[0092] The object information features and advertising features are input into the conversion probability prediction model. The object information features and advertising features are reduced in dimensionality by the dimensionality reduction network in the conversion probability prediction model to obtain the target object features and target advertising features. The target object features and target advertising features are concatenated by the feature semantic interaction network in the conversion probability prediction model to obtain the concatenated features. Based on the concatenated features, feature semantic interaction is performed to obtain the interaction features. The interaction features are then fully connected to perform a fully connected operation on the interaction features in the conversion probability prediction model to obtain the conversion probability of each candidate advertisement.

[0093] The conversion probability prediction model is a neural network model used to predict conversion probability. This conversion probability prediction model can be obtained by training the neural network using the converted training ads of the training object and the corresponding training labels of the converted training ads. The training labels corresponding to the converted training ads are determined based on information about whether the training object will convert again on the converted training ads.

[0094] Specifically, the server can pre-train the neural network using converted training ads and their corresponding training labels. This neural network can be a pre-trained network, for example, it can be obtained by pre-training an initialized neural network using unconverted training ads and their corresponding training labels. The training labels for the converted training ads are determined based on information about whether the training object made an initial conversion to the unconverted training ads. Alternatively, the neural network can be a directly initialized neural network. In this case, the server can simultaneously train the neural network using both converted training ads and their corresponding training labels, as well as unconverted training ads and their corresponding training labels. Once training is complete, a conversion probability prediction model is obtained. The server then deploys and uses this conversion probability prediction model.

[0095] When the server needs to use the conversion probability prediction model, it invokes the model and inputs object information features and ad features into it. In this model, a dimensionality reduction network reduces the dimensionality of the object information features and ad features respectively, obtaining target object features and target ad features. Then, these target object features and target ad features are input into a feature semantic interaction network, which concatenates them to obtain concatenated features. Based on these concatenated features, feature semantic interaction is performed to obtain interactive features. Finally, the interactive features are input into a fully connected network to perform fully connected operations, yielding the output conversion probability. Specifically, when the input is object information features and the ad features of a converted ad, the output is the conversion probability of the converted ad; when the input is object information features and the ad features of an unconverted ad, the output is the conversion probability of the unconverted ad. Alternatively, the server can input object information features and the ad features of all candidate ads as a sequence into the conversion probability prediction model to obtain the conversion probability of each candidate ad output by the model.

[0096] In a specific embodiment, such as Figure 5 The diagram illustrates the architecture of a conversion probability prediction model, which includes a dimensionality reduction network (Pooling), a feature semantic interaction network (DCN, Deep Cross Network), and a fully connected network (FC). Specifically, the server obtains object information features, advertising information features of converted ads, statistical features, and conversion features of converted ads. These features are then fed into the dimensionality reduction network (Pooling) for dimensionality reduction, resulting in concatenated features. These concatenated features are then input into the feature semantic interaction network (DCN) to obtain the output interaction features. Finally, the interaction features are input into the fully connected network (FC) for regression calculation using a regression function, yielding the output prediction result, i.e., the conversion probability of the converted ads. This conversion probability prediction model improves prediction efficiency.

[0097] In the above embodiments, by inputting object information features and advertising features into the conversion probability prediction model, the conversion probability of each candidate advertisement is predicted using the conversion probability prediction model. That is, by directly using the trained neural network model for prediction, the efficiency of obtaining conversion probability can be improved.

[0098] In one embodiment, such as Figure 6 As shown, training the conversion probability prediction model includes the following steps:

[0099] S602, Obtain the training object information features corresponding to the training object identifier;

[0100] S604, obtain the training ad features and training tags of the converted training ads corresponding to the training object identifier. The training tags are determined based on the conversion result information corresponding to the converted training ads. The conversion result information is obtained after pushing the converted training ads to the terminal corresponding to the training object identifier. The training ad features include the converted training ad information features and the converted training features. The converted training features are obtained by feature extraction of the conversion behavior information of the converted training ads.

[0101] Among them, the training object identifier refers to the object identifier used during training to uniquely identify the training object. The training object information features are obtained by extracting features from the information representing the training object. This information can be the training object's attribute information, behavioral information, etc. Converted training ads refer to converted ads used during training, meaning the conversion time of the converted training ad is before the training time. Conversion result information is obtained after pushing the converted training ad to the terminal corresponding to the training object identifier. This conversion result information includes whether the training object converted again after seeing the converted training ad and whether the training object did not convert again after seeing the converted training ad. The determined training tags include conversion tags and non-conversion tags. The conversion tag indicates that the training object converted again after seeing the converted training ad, and the non-conversion tag indicates that the training object did not convert again after seeing the converted training ad. The converted training ad information features are obtained by extracting features from the advertising information of the converted training ad. The converted training features are obtained by extracting features from the conversion behavior information of the converted training ad. The conversion behavior information of the converted training ad refers to the information about the training object's initial conversion behavior towards the converted training ad.

[0102] Specifically, the server can obtain the identifiers of each training object and then use each identifier sequentially for training. During training, the server can directly look up the training object information features in the database based on the training object identifier, or it can look up the training object information based on the training object identifier and then extract the features of the training object information to obtain the training object information features. Simultaneously, the server can retrieve the identifiers of the converted training ads corresponding to the training object identifiers from the database, and then look up the corresponding training ad features and training tags in the database based on the identifiers of the converted training ads. The server pre-obtains the conversion result information obtained after pushing the converted training ads to the terminals corresponding to the training object identifiers, then determines the training tags corresponding to the converted training ads based on the conversion result information, and then associates and saves the converted training ads and their corresponding training tags in the database.

[0103] In one embodiment, the server can also obtain training object information features corresponding to the training object identifier, training ad features of the converted training ad corresponding to the training object identifier, and training tags of the converted training ad from the server providing the data service, that is, obtain training data from the server providing the data service to improve the efficiency and accuracy of obtaining training data.

[0104] S606: Input the training object information features and training ad features into the initial conversion probability prediction model for forward calculation to obtain the training conversion probability of the converted training ad.

[0105] The initial conversion probability prediction model refers to the conversion probability prediction model with initialized model parameters. These parameters can be initialized randomly, with zero initialization, or using a Gaussian distribution, or they can be initialized using pre-trained model parameters. The training conversion probability characterizes the likelihood that the training object will convert again based on a converted training ad.

[0106] Specifically, the training object information features and training ad features are input into the initial conversion probability prediction model for forward computation. This involves using the initial dimensionality reduction network within the initial conversion probability prediction model to reduce the dimensionality of both the training object information features and the training ad features, resulting in dimensionality-reduced training object information features and dimensionality-reduced training ad features. These are then input into the initial feature semantic interaction network, which concatenates the dimensionality-reduced training object information features and dimensionality-reduced ad features to obtain the concatenated features during training. Based on these concatenated features, feature semantic interaction is performed to obtain the interactive features during training. Finally, the interactive features during training are input into the initial fully connected network for fully connected operations to obtain the training conversion probability of the converted training ad.

[0107] In one embodiment, the server can also obtain a pre-trained conversion probability prediction model and use it as the pre-trained conversion probability prediction model. This pre-trained conversion probability prediction model is obtained by training a conversion probability prediction model initialized with model parameters using the unconverted training ads of the training objects and their training labels. The training labels of the unconverted training ads characterize whether the training objects perform an initial conversion on the unconverted training ads. These training labels include conversion labels and non-convert labels; the conversion label indicates that the training objects performed an initial conversion on the unconverted training ads, and the non-convert label indicates that the training objects did not perform an initial conversion on the unconverted training ads. The server takes the information features of the training objects and the training ad information features of the unconverted training ads as input and trains the model according to the training ad features and training labels. When training is complete, the pre-trained conversion probability prediction model is obtained.

[0108] S608, calculate the training loss information between the training conversion probability and the training label, and update the initial conversion probability prediction model in reverse based on the training loss information to obtain the updated conversion probability prediction model.

[0109] S610, update the conversion probability prediction model as the initial conversion probability prediction model, and return to iteratively execute the step of obtaining the training object information features corresponding to the training object identifier until the training completion condition is met, and obtain the conversion probability prediction model.

[0110] The training loss information is used to characterize the error between the training conversion probability and the training label. The training completion condition refers to the conditions under which the conversion probability prediction model is obtained, including but not limited to the model's loss information reaching a pre-set threshold, the model's parameters no longer changing, or the number of training iterations reaching the maximum number of iterations.

[0111] Specifically, the server can use a pre-set loss function to calculate the training loss information between the training conversion probability and the training label. This loss function can be a classification loss function or a regression loss function, such as the cross-entropy loss function. The server then determines whether the training completion condition is met. For example, it could determine whether a pre-set threshold has been reached based on the training loss information, whether the model parameters have changed, or whether the maximum number of training iterations has been reached. If the training completion condition is not met, the server uses the training loss information to back-update the model parameters in the initial conversion probability prediction model, obtaining an updated conversion probability prediction model. The server then uses the updated conversion probability prediction model as the initial conversion probability prediction model and iteratively executes the step of obtaining the training object information features corresponding to the training object identifier until the training completion condition is met. At this point, the initial conversion probability prediction model at the time of meeting the training completion condition is used as the final conversion probability prediction model.

[0112] In the above embodiments, by using the training object information features corresponding to the training object identifier, the training ad features of the converted training ad, and the training label of the converted training ad to the initial conversion probability prediction model, a conversion probability prediction model is obtained. This allows the conversion probability prediction model to learn information about the object's re-conversion of the converted ad through training, thereby enhancing the ability to predict the conversion probability of the converted ad and improving the accuracy of conversion probability prediction.

[0113] In one embodiment, such as Figure 6 As shown in S208, each candidate ad is filtered based on conversion probability to obtain the ads to be pushed, including:

[0114] S602, obtain the current time point and the conversion time point corresponding to the converted advertisement, and determine the decay weight corresponding to the converted advertisement based on the time period between the conversion time point and the current time point. The time period is negatively correlated with the decay weight.

[0115] The decay weight characterizes the degree to which the conversion probability of a converted ad needs to be attenuated. The longer the time interval, the smaller the decay weight; the shorter the time interval, the larger the decay weight. In other words, the longer the interval between two push notifications, the higher the probability of a subsequent conversion, and the smaller the decay weight. Conversely, the shorter the interval, the lower the probability of a subsequent conversion, and the larger the decay weight. A smaller decay weight indicates a smaller impact on subsequent conversions, while a larger decay weight indicates a greater impact on subsequent conversions.

[0116] Specifically, the server obtains the current time of the operating system and retrieves the conversion time corresponding to the converted ad from the database. Then, it calculates the time period between the conversion time and the current time, and determines the decay weight corresponding to the converted ad based on the pre-set correspondence between the time period and the decay weight.

[0117] S602, based on the decay weight, decays the conversion probability of the converted ad to obtain the target conversion probability of the converted ad.

[0118] S602, based on the target conversion probability of converted ads and the conversion probability of unconverted ads, filter each candidate ad to obtain the ads to be pushed.

[0119] Specifically, the server uses a decaying weight to weight the conversion probability of converted ads, obtaining the target conversion probability of the converted ads. Then, it compares the target conversion probability of the converted ads with the conversion probability of the non-converted ads, selecting the candidate ad with the highest conversion probability as the ad to be pushed. The server can sort all candidate ads from highest to lowest conversion probability, i.e., sort them according to the target conversion probability of the converted ads, improving the accuracy of the sorting of converted ads, and then select the top-ranked candidate ads as the ads to be pushed.

[0120] In one specific embodiment, the server can filter converted ads whose conversion time point is within one day prior to the current time point. Then, it can attenuate the conversion probability of converted ads whose conversion time point is within two to seven days prior to the current time point. For example, converted ads whose conversion time point is within two days prior to the current time point can receive a corresponding attenuation weight of 0.95, and converted ads whose conversion time point is within three days prior to the current time point can receive a corresponding attenuation weight of 0.92. This attenuation weight is then used to attenuate the conversion probability of converted ads, thereby further reducing the problem of excessively high conversion probabilities and improving the accuracy of conversion probability prediction.

[0121] In the above embodiment, the decay weight corresponding to the converted ad is determined by using the time period between the conversion time point corresponding to the converted ad and the current time point. Then, the conversion probability of the converted ad is decayed using the decay weight to obtain the target conversion probability of the converted ad. That is, by decaying using the decay weight, the problem of the conversion probability of the converted ad can be further reduced, and the accuracy of the conversion probability of the converted ad can be further improved. Finally, the ads to be pushed are filtered according to the target conversion probability of the converted ads, which improves the accuracy of the obtained ads to be pushed.

[0122] In one embodiment, such as Figure 7 As shown in S208, each candidate ad is filtered based on conversion probability to obtain the ads to be pushed, including:

[0123] S702, Identify relevant ads associated with converted ads from among the candidate ads, and obtain the enhanced weight of the relevant ads. The enhanced weight is determined based on the degree of relevance between the relevant ads and the converted ads.

[0124] Relevant ads refer to ads associated with converted ads. For example, an ad for "washing machine effervescent tablets" is associated with an ad for "washing machine." Based on the converted ads for "washing machine," related ads for "washing machine effervescent tablets" can be obtained. This association between converted and relevant ads is pre-mined. Enhancement weight is used to characterize the degree to which the conversion probability corresponding to a converted ad needs to be enhanced. The larger the enhancement weight, the higher the relevance between the relevant and converted ads, meaning the closer the association. The relevance between relevant and converted ads characterizes the situation where relevant and converted ads appear together in the conversion behavior information of the same object. For example, the higher the frequency of co-occurrence, the higher the relevance between the relevant and converted ads, and the larger the enhancement weight.

[0125] Specifically, the server searches for related ads associated with the converted ad among each candidate ad based on pre-saved association relationships. Then, it retrieves the relevance between the converted ad and the related ads from the database. This relevance can be directly used as the enhancement weight for the related ad, or the enhancement weight can be determined based on the mapping relationship between relevance and enhancement weight. In one embodiment, each candidate ad may include multiple converted ads, and each converted ad may also include multiple related ads. The server then determines the enhancement weight based on the enhancement weights of all related ads retrieved.

[0126] S704, based on the enhanced weight, enhances the conversion probability of related ads to obtain the related ad conversion probability.

[0127] S706: Based on the conversion probability of relevant ads and the conversion probability of irrelevant ads among the candidate ads, each candidate ad is screened to obtain the ads to be pushed.

[0128] Among them, irrelevant ads refer to ads in each candidate ad group that are not relevant ads.

[0129] Specifically, the server uses enhanced weights to weighted calculate the conversion probability of relevant ads, obtaining the conversion probability of relevant ads. Then, it compares the conversion probability of relevant ads among all candidate ads with the conversion probability of irrelevant ads among all candidate ads, selecting the candidate ad with the highest conversion probability as the ad to be pushed. The server can sort all candidate ads from highest to lowest conversion probability, i.e., sort by the conversion probability of relevant ads, improving the accuracy of the relevant ad sorting, and then select the top-ranked candidate ads as the ads to be pushed. In one embodiment, the relevant ad can also be a converted ad among the candidate ads. In this case, the server uses both decay and enhancement weights to weighted calculate the conversion probability of the relevant ad, obtaining the final conversion probability of the relevant ad.

[0130] In the above embodiments, by obtaining the enhanced weights of related ads associated with converted ads, and then enhancing the conversion probability of related ads, the conversion probability of related ads is obtained. That is, by enhancing the conversion probability of related ads through enhanced weights, the probability of pushing related ads associated with converted ads is increased. Finally, ads to be pushed are selected according to the conversion probability of related ads. This allows for the pushing of related ads associated with converted ads, thereby increasing the likelihood of the target audience converting from the pushed ads and improving the effectiveness of ad pushing.

[0131] In one embodiment, prior to S702, i.e. before determining the relevant ads associated with the converted ads from among the candidate ads and obtaining the enhanced weights of the relevant ads, the following steps are included:

[0132] Obtain the first and second converted ads for each target object identifier within the target time period; count the number of times the first and second converted ads co-occur, which represents the number of times the first and second converted ads correspond to the same target object identifier within the target time period; count the total number of objects for each target object identifier and calculate the ratio of the number of co-occurrences to the total number of objects to obtain the correlation between the first and second converted ads; when the correlation exceeds a preset correlation threshold, establish the association between the first and second converted ads.

[0133] The target object identifier can be any one of the various object identifiers. These object identifiers can be those with corresponding converted ads or those without. The target time period represents the maximum time interval between the conversion time of the first converted ad and the conversion time of the second converted ad, and can be pre-set. For example, the target time period can be set to within one week, then the maximum time interval between the conversion time of the first and second converted ads is one week. That is, the conversion time of the first and second converted ads are both within the target time period. The first and second converted ads are any two converted ads whose conversion time points are within the target time period. The number of times they co-occur is used to represent the number of times the first and second converted ads co-correspond to the same object identifier within the target time period. For example, if the first and second converted ads appear in the conversion behavior information list of the same object within one week, then the number of times they co-correspond to the same object is counted to obtain the number of times they co-occur. The more times ads appear together, the more likely an audience is to convert from one ad within the same timeframe as another ad. This indicates a high correlation between the two ads; if an audience converts from one ad, they are more likely to convert from the other ad. A preset relevance threshold refers to a pre-set threshold for the degree of relevance. This threshold can be set based on human experience or statistically derived from historical data.

[0134] Specifically, the server retrieves all object identifiers from the database, then obtains the identifiers of the first and second converted ads for the target object identifiers within the target time period. Based on these identifiers, it searches the historical conversion behavior information of all object identifiers for the frequency of co-occurrence of the first and second converted ads. The server then calculates the ratio of this co-occurrence frequency to the total number of objects to determine the relevance between the first and second converted ads. At this point, the server determines whether the relevance exceeds a preset threshold. If the relevance exceeds the threshold, it indicates that the first and second converted ads are related, and the server establishes an association between them; for example, it can save the association between the first and second converted ads. If the relevance exceeds the threshold, it indicates that the first and second converted ads are not related, and no action is taken.

[0135] In the above embodiments, by counting the number of times the first converted ad and the second converted ad co-occur, and then using the number of co-occurrences and the total number of objects to calculate the correlation between the first converted ad and the second converted ad, the accuracy of the obtained correlation is improved. Then, based on the correlation between the first converted ad and the second converted ad, it is determined whether the first converted ad and the second converted ad are related, that is, the accuracy of the obtained association is improved.

[0136] In one embodiment, S204, namely obtaining the advertising features of each candidate advertisement corresponding to the identifier of the object to be pushed, includes the following steps:

[0137] Get the converted ads corresponding to the target object identifier, and get each related ad of the converted ad; use each related ad as a candidate ad, and get the ad features of each candidate ad.

[0138] In this embodiment, the server can obtain all relevant ads for converted ads of the target object identifier, and then calculate the conversion probability of each relevant ad as a candidate ad to obtain the conversion probability of each relevant ad. Then, the server can filter the relevant ads according to their conversion probabilities to obtain the ads to be pushed. That is, the server obtains the ad features of each relevant ad, which may include both converted and unconverted ads. The server then uses the target object information features and the unconverted ad information features of the unconverted ads to calculate the conversion probability of the unconverted ads, and simultaneously uses the target object information features and the converted ad information features of the converted ads to calculate the conversion probability of the converted ads. Finally, the server compares the conversion probabilities of all relevant ads and selects the relevant ad with the highest conversion probability to push to the terminal corresponding to the target object identifier. The server can sort all the conversion probabilities of all relevant ads and then select the top-ranked ads.

[0139] In the above embodiments, by obtaining each related advertisement corresponding to the converted advertisement of the target object identifier, and then using each related advertisement as a candidate advertisement, the conversion probability of each related advertisement can be calculated. Then, the advertisement to be pushed is obtained from each related advertisement and pushed to the terminal corresponding to the target object identifier, so that the pushed advertisement is an advertisement associated with the converted advertisement, thereby further improving the accuracy of advertisement push.

[0140] In a specific embodiment, such as Figure 8 As shown, an advertising push method is provided, executed by a computer device, which can be a server or a terminal. Preferably, the computer device can be a server, and specifically performs the following steps:

[0141] S802, obtain the object information features corresponding to the object identifier to be pushed, and obtain the advertising information of each unconverted advertisement in each candidate advertisement corresponding to the object identifier to be pushed. Extract features from the advertising information of each unconverted advertisement to obtain the unconverted advertisement information features of each unconverted advertisement.

[0142] S804: Obtain at least two converted ads and their corresponding conversion time points from each candidate ad for the target object identifier. Filter the at least two converted ads according to their conversion time points and a preset conversion filtering time period to obtain each target converted ad. Obtain the ad information and conversion behavior information for each target converted ad.

[0143] S806, extract features from the advertising information of each target converted ad to obtain the converted ad information features of each target converted ad, and extract features from the conversion behavior information of each target converted ad to obtain the converted features of each target converted ad.

[0144] S808, input the object information features, the unconverted ad information features of each unconverted ad, the converted ad information features of each target converted ad, and the converted features of each target converted ad into the conversion probability prediction model to obtain the output conversion probability of each unconverted ad and the conversion probability of each target converted ad.

[0145] S810: Obtain the current time point and the conversion time point corresponding to each target converted ad, and determine the decay weight corresponding to each target converted ad based on the time period between the conversion time point and the current time point. Based on the decay weight, decay the conversion probability of each target converted ad to obtain the target conversion probability of each target converted ad.

[0146] S812, determine each relevant ad associated with each target converted ad from each unconverted ad and each target converted ad, obtain the enhancement weight of the relevant ads, enhance the conversion probability of each relevant ad based on the enhancement weight, and obtain the relevant ad conversion probability of each relevant ad.

[0147] S814: Based on the conversion probability of each relevant advertisement and the conversion probability of each irrelevant advertisement, filter each unconverted advertisement and each target converted advertisement to obtain the advertisement to be pushed, and push the advertisement to be pushed to the terminal corresponding to the target identifier.

[0148] In the above embodiments, by filtering converted ads whose conversion time points fall within a preset conversion filtering time period, duplicate pushes of converted ads within a short period can be avoided. Then, the conversion probability of target converted ads is calculated using converted features, reducing the problem of overestimating conversion results when converting ads are used for conversion prediction, thus improving the accuracy of the obtained conversion probability. Furthermore, by attenuating the conversion probability of each target converted ad with a decay weight, the target conversion probability of each target converted ad is obtained, further reducing the problem of overestimating conversion results when converting ads are used for conversion prediction, and further improving the accuracy of the obtained conversion probability. Finally, by enhancing the conversion probability of related ads with enhancement weights, the conversion probability of each related ad is enhanced, thus increasing the likelihood of related ad pushes. Finally, ads are filtered according to conversion probability to obtain the ads to be pushed, and then pushed, improving the accuracy of ad pushes.

[0149] In a specific embodiment, such as Figure 9The diagram illustrates an ad push architecture. Specifically, when a converted user requests an ad push, step 1.a filters out converted ads with the same ad identifier and conversion behavior from all recalled candidate ads within one day. Then, ad push prediction is performed via the main ad push path. This main path, LTR (Large Path for Consistent Recall Towards the Final Goal), performs a coarse ranking of the filtered candidate ads, resulting in a coarsely ranked list of candidate ads. This list is then finely ranked, using a conversion probability prediction model to calculate the conversion probability of each candidate ad. Specifically, the conversion rate prediction model in step 1.c uses the ad features of the coarsely ranked candidate ads and the user's object information features for calculation. These ad features include the conversion characteristics of converted ads, which is a long-term conversion sequence feature. This long-term conversion sequence feature includes the user identifier, ad identifiers converted by the user within the past 180 days, ad conversion behavior, and the time of conversion. Then, step 1.b calculates the decay weight of converted ads within 2 to 7 days and decays the conversion probability of the calculated converted ads, reducing their competitiveness in the ranking process. Simultaneously, the relevance of related ads associated with converted ads is calculated through the associated push branch, and step 2.c determines the enhancement weight of related ads based on the relevance, then enhances the conversion probability of related ads to improve their competitiveness in the ranking process. Before executing the associated push branch, historical conversion information of the target audience is obtained in advance, including ad identifiers and ad conversion behavior information. Then, step 2.a uses the historical conversion information to mine related ads of converted ads, i.e., establishing the association relationship between ads. Finally, the server obtains the final conversion probability of each candidate ad, ranks the candidate ads using the final conversion probability, and then selects an appropriate number of ads to push to the target audience, thereby improving the specificity of ad push. The ads can be performance ads, which are a pay-per-performance online advertising model whose core objective is to achieve the advertiser's desired conversion behavior, such as clicks, downloads, registrations, activations, and lead generation. In oCPA (Optimized Cost per Action) performance advertising, advertisers bid for the desired conversion behavior, and the advertising platform uses its recommendation capabilities to expose the advertiser's ads to the right audience on the right media traffic, thus achieving the goal of ad exposure and performance within the cost contract stipulated by the advertiser's bid.

[0150] In one embodiment, this ad push method is applied to an instant messaging application. Specifically: when a user uses the instant messaging application, the application's server receives an ad push request sent by the user and then obtains user information features based on the user identifier in the request. Simultaneously, the server retrieves the ad features of each candidate ad corresponding to the user from the database. The user information features and ad features are input into a pre-deployed conversion probability prediction model to obtain the conversion probability of each candidate ad. This model can use the decay weight of converted ads or the enhancement weight of related ads to weight the conversion probability. Finally, the server selects the ad with the highest conversion probability and pushes it to the user's terminal. When the instant messaging application on the user's terminal receives the pushed ad, it displays the ad. Figure 10 The image shown is a schematic diagram of a page displaying an advertisement in an instant messaging application. This page displays the content of the pushed application advertisement, including the application advertisement text, the application advertisement image, and the application download link "Check it out." When a user clicks the application download link, they can download the corresponding application.

[0151] In one embodiment, the ad push method is applied to a social media platform. Specifically, when a user uses the social media platform, they can send an ad push request to the platform's server. The social media platform's server responds to the ad push request, searches its database for the user's user information features and the video ad features of each candidate video ad corresponding to the user. These video ad features include the conversion features of converted video ads. Then, a conversion probability prediction model is used to obtain the conversion probability of each candidate video ad. This can be achieved by using the decay weight of converted video ads to weight the conversion probability, or by using the enhancement weight of related ads to weight the conversion probability, or by filtering the user's converted video ads within one day before calculating the conversion probability. Finally, the server sorts the conversion probabilities of each candidate video ad and can select the candidate video ad with the highest conversion probability to push to the user's terminal. When the user's terminal receives the pushed video ad, the video ad is played on the social media platform. Figure 11 The image shown is a screenshot of a video advertisement displayed on a social media platform. The video advertisement is 1 minute and 30 seconds long, and the screenshot is from the 2-second mark. This screenshot displays the content of the product advertisement, including the advertisement text, the video screenshot, and a link to the advertised product page. When a user clicks the link, they are redirected to the corresponding product page.

[0152] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0153] Based on the same inventive concept, this application also provides an advertising push device for implementing the advertising push method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more advertising push device embodiments provided below can be found in the limitations of the advertising push method described above, and will not be repeated here.

[0154] In one embodiment, such as Figure 12 As shown, an advertising push device 1200 is provided, including: an object feature acquisition module 1202, an advertising feature acquisition module 1204, a conversion calculation module 1206, and a push module 1208, wherein:

[0155] The object feature acquisition module 1202 is used to acquire the object information features corresponding to the identifier of the object to be pushed.

[0156] The advertising feature acquisition module 1204 is used to acquire the advertising features of each candidate advertisement corresponding to the identifier of the object to be pushed. Each candidate advertisement includes unconverted advertisements and converted advertisements. The advertising features of converted advertisements include converted advertisement information features and converted features. The converted features are obtained by feature extraction of the conversion behavior information of converted advertisements. The advertising features of unconverted advertisements include unconverted advertisement information features.

[0157] The conversion calculation module 1206 is used to calculate the conversion probability of an advertisement based on the characteristics of the object information and the characteristics of the advertisement, and to obtain the conversion probability of each candidate advertisement. The conversion probability is used to characterize the likelihood that the object identifier to be pushed will convert the candidate advertisement.

[0158] The push module 1208 is used to filter each candidate advertisement based on the conversion probability, obtain the advertisement to be pushed, and push the advertisement to be pushed to the terminal corresponding to the target identifier.

[0159] In one embodiment, the advertising feature acquisition module 1204 is further configured to acquire advertising information of unconverted ads in each candidate ad, extract features from the advertising information of unconverted ads to obtain unconverted ad information features of unconverted ads; acquire advertising information of converted ads and conversion behavior information of converted ads in each candidate ad; extract features from the advertising information of converted ads to obtain converted ad information features of converted ads, and extract features from the conversion behavior information of converted ads to obtain converted features of converted ads.

[0160] In one embodiment, converted ads include at least two;

[0161] The ad feature acquisition module 1204 is also used to acquire at least two converted ads corresponding to the target object identifier and the conversion time points corresponding to the at least two converted ads respectively; to filter the at least two converted ads according to the conversion time points and the preset conversion filtering time period to obtain the target converted ads; and to acquire the ad information of the target converted ads and the conversion behavior information of the target converted ads.

[0162] In one embodiment, the conversion calculation module 1206 is further configured to calculate the conversion probability of an ad based on object information features and unconverted ad information features to obtain the conversion probability of an unconverted ad; and to calculate the conversion probability of an ad based on object information features, converted ad information features, and converted features to obtain the conversion probability of a converted ad.

[0163] In one embodiment, the conversion calculation module 1206 is further configured to reduce the dimensionality of the object information features and the advertising features respectively to obtain the target object features and the target advertising features; concatenate the target object features and the target advertising features to obtain the concatenated features; perform feature semantic interaction based on the concatenated features to obtain the interaction features; and perform a full connection operation based on the interaction features to obtain the conversion probability of each candidate advertisement.

[0164] In one embodiment, the conversion calculation module 1206 is further configured to input object information features and advertising features into the conversion probability prediction model; reduce the dimensionality of the object information features and advertising features respectively through the dimensionality reduction network in the conversion probability prediction model to obtain target object features and target advertising features; concatenate the target object features and target advertising features through the feature semantic interaction network in the conversion probability prediction model to obtain concatenated features, and perform feature semantic interaction based on the concatenated features to obtain interactive features; and perform fully connected operations on the interactive features through the fully connected network in the conversion probability prediction model to obtain the conversion probability of each candidate advertisement.

[0165] In one embodiment, the advertising push device 1200 further includes:

[0166] The model training module is used to obtain the training object information features corresponding to the training object identifier; obtain the training ad features and training tags of the converted training ads corresponding to the training object identifier. The training tags are determined based on the conversion result information of the converted training ads, which is obtained after pushing the converted training ads to the terminals corresponding to the training object identifier. The training ad features include the converted training ad information features and the converted training features. The converted training features are obtained by feature extraction of the conversion behavior information of the converted training ads; the training object information features and the training ad features are input into the initial conversion probability prediction model for forward calculation to obtain the training conversion probability of the converted training ads; the training loss information between the training conversion probability and the training tags is calculated, and the initial conversion probability prediction model is updated in reverse based on the training loss information to obtain the updated conversion probability prediction model; the updated conversion probability prediction model is used as the initial conversion probability prediction model, and the step of obtaining the training object information features corresponding to the training object identifier is returned and iteratively executed until the training completion condition is met to obtain the conversion probability prediction model.

[0167] In one embodiment, the push module 1208 is further configured to obtain the current time point and the conversion time point corresponding to the converted advertisement, and determine the decay weight corresponding to the converted advertisement based on the time period between the conversion time point and the current time point, wherein the time period is negatively correlated with the decay weight; decay the conversion probability of the converted advertisement based on the decay weight to obtain the target conversion probability of the converted advertisement; and filter each candidate advertisement based on the target conversion probability of the converted advertisement and the conversion probability of the unconverted advertisement to obtain the advertisement to be pushed.

[0168] In one embodiment, the push module 1208 is further configured to determine the relevant ads associated with the converted ads from each candidate ad and obtain the enhancement weight of the relevant ads, the enhancement weight being determined based on the degree of relevance between the relevant ads and the converted ads; enhance the conversion probability of the relevant ads based on the enhancement weight to obtain the relevant ad conversion probability of the relevant ads; and filter each candidate ad based on the relevant ad conversion probability of the relevant ads and the conversion probability of the irrelevant ads in each candidate ad to obtain the ads to be pushed.

[0169] In one embodiment, the advertising push device 1200 further includes:

[0170] The association establishment module is used to obtain the first and second converted ads of the target object identifier in the target time period among each object identifier; count the number of times the first and second converted ads co-occur, which is used to characterize the number of times the first and second converted ads correspond to the same object identifier in the target time period; count the total number of objects for each object identifier and calculate the ratio of the number of co-occurrences to the total number of objects to obtain the correlation degree between the first and second converted ads; when the correlation degree exceeds the preset correlation threshold, the association relationship between the first and second converted ads is established.

[0171] In one embodiment, the advertising feature acquisition module 1204 is further configured to acquire the converted advertisement corresponding to the identifier of the object to be pushed, and acquire each related advertisement of the converted advertisement; take each related advertisement as a candidate advertisement, and acquire the advertising features of each candidate advertisement.

[0172] Each module in the aforementioned advertising push device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0173] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 13 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores object information features, advertising features of various candidate advertisements, and model training data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements an advertising push method.

[0174] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 14As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements an advertising push method. The display unit of the computer device is used to form a visually visible image. It can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0175] Those skilled in the art will understand that Figure 13 or Figure 14 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0176] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0177] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0178] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0179] It should be noted that the object information (including but not limited to object device information, object personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the object or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0180] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this 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, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0181] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0182] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An advertising push method, characterized in that, The method includes: Obtain the object information features corresponding to the identifier of the object to be pushed; Obtain the advertising features of each candidate advertisement corresponding to the identifier of the object to be pushed. Each candidate advertisement includes unconverted advertisements and converted advertisements. The advertising features of the converted advertisements include converted advertisement information features and converted features. The converted features are obtained by feature extraction of the conversion behavior information of the converted advertisements. The advertising features of the unconverted advertisements include unconverted advertisement information features. Based on the object information features and the advertising features, the conversion probability of the advertisement is calculated to obtain the conversion probability of each candidate advertisement. The conversion probability is used to characterize the likelihood that the object identifier to be pushed will convert the candidate advertisement. Based on the conversion probability, each candidate advertisement is filtered to obtain the advertisement to be pushed, and the advertisement to be pushed is pushed to the terminal corresponding to the target identifier.

2. The method according to claim 1, characterized in that, The step of obtaining the advertising features of each candidate ad corresponding to the identifier of the object to be pushed includes: Obtain the advertising information of the unconverted ads in each candidate ad, extract features from the advertising information of the unconverted ads, and obtain the unconverted ad information features of the unconverted ads; Obtain the advertising information of the converted ads and the conversion behavior information of the converted ads in each of the candidate ads; Feature extraction is performed on the advertising information of the converted advertisement to obtain the converted advertisement information features, and feature extraction is performed on the conversion behavior information of the converted advertisement to obtain the converted features of the converted advertisement.

3. The method according to claim 2, characterized in that, The converted ads include at least two; Obtaining the advertising information of converted ads and the conversion behavior information of the converted ads from each candidate ad group, including: Obtain at least two converted ads corresponding to the identifier of the target object and the conversion time points corresponding to the at least two converted ads respectively; The at least two converted ads are filtered according to the conversion time point and the preset conversion filtering time period to obtain the target converted ads; Obtain the advertising information of the target converted advertisement and the conversion behavior information of the target converted advertisement.

4. The method according to claim 2, characterized in that, The calculation of the conversion probability of each candidate ad based on the object information features and the ad features includes: The conversion probability of the ad is calculated based on the object information features and the unconverted ad information features to obtain the conversion probability of the unconverted ad. Based on the object information features, the converted advertisement information features, and the converted features, the conversion probability of the advertisement is calculated to obtain the conversion probability of the converted advertisement.

5. The method according to claim 1, characterized in that, The calculation of the conversion probability of each candidate ad based on the object information features and the ad features includes: The object information features and the advertisement features are dimensionality reduced respectively to obtain the target object features and the target advertisement features; The target object features and the target advertisement features are concatenated to obtain concatenated features, and feature semantic interaction is performed based on the concatenated features to obtain interactive features; Based on the interaction features, a fully connected operation is performed to obtain the conversion probability of each candidate advertisement.

6. The method according to claim 1, characterized in that, The calculation of the conversion probability of each candidate ad based on the object information features and the ad features includes: The object information features and the advertising features are input into the conversion probability prediction model; The object information features and the advertising features are reduced in dimensionality using the dimensionality reduction network in the conversion probability prediction model to obtain the target object features and the target advertising features. The feature semantic interaction network in the conversion probability prediction model concatenates the target object features and the target advertisement features to obtain concatenated features, and performs feature semantic interaction based on the concatenated features to obtain interactive features. The conversion probability of each candidate ad is obtained by performing a fully connected operation on the interaction features using the fully connected network in the conversion probability prediction model.

7. The method according to claim 6, characterized in that, The training of the conversion probability prediction model includes the following steps: Obtain the training object information features corresponding to the training object identifier; The training ad features and training tags of the converted training ads corresponding to the training object identifier are obtained. The training tags are determined based on the conversion result information corresponding to the converted training ads. The conversion result information is obtained after pushing the converted training ads to the terminal corresponding to the training object identifier. The training ad features include converted training ad information features and converted training features. The converted training features are obtained by feature extraction of the conversion behavior information of the converted training ads. The training object information features and the training advertisement features are input into the initial conversion probability prediction model for forward calculation to obtain the training conversion probability of the converted training advertisement. Calculate the training loss information between the training conversion probability and the training label, and update the initial conversion probability prediction model in reverse based on the training loss information to obtain the updated conversion probability prediction model; The updated conversion probability prediction model is used as the initial conversion probability prediction model, and the step of obtaining the training object information features corresponding to the training object identifier is iteratively executed until the training completion condition is met, thus obtaining the conversion probability prediction model.

8. The method according to claim 1, characterized in that, The step of filtering the candidate ads based on the conversion probability to obtain the ads to be pushed includes: Obtain the current time point and the conversion time point corresponding to the converted advertisement, and determine the decay weight corresponding to the converted advertisement based on the time period between the conversion time point and the current time point, wherein the time period is negatively correlated with the decay weight; The conversion probability of the converted ad is attenuated based on the attenuation weight to obtain the target conversion probability of the converted ad. Based on the target conversion probability of the converted ads and the conversion probability of the unconverted ads, the candidate ads are filtered to obtain the ads to be pushed.

9. The method according to claim 1, characterized in that, The step of filtering the candidate ads based on the conversion probability to obtain the ads to be pushed includes: From the candidate ads, identify the related ads associated with the converted ads and obtain the enhancement weight of the related ads, which is determined based on the degree of relevance between the related ads and the converted ads; The conversion probability of the relevant advertisement is enhanced based on the enhanced weight, thereby obtaining the relevant advertisement conversion probability; Based on the conversion probability of the relevant ads and the conversion probability of the irrelevant ads among the candidate ads, the candidate ads are filtered to obtain the ads to be pushed.

10. The method according to claim 1, characterized in that, Before determining the relevant ads associated with the converted ads from the candidate ads and obtaining the enhanced weights of the relevant ads, the method further includes: Obtain the first converted ad and the second converted ad of the target object identifier in the target time period from each of the object identifiers; The number of times the first converted ad and the second converted ad co-occurs is counted, and the number of times they co-occur is used to characterize the number of times the first converted ad and the second converted ad co-correspond to the same object identifier in the target time period; The total number of objects identified by each object is counted, and the ratio of the number of times they co-occur to the total number of objects is calculated to obtain the correlation between the first converted advertisement and the second converted advertisement. When the relevance exceeds a preset relevance threshold, an association is established between the first converted advertisement and the second converted advertisement.

11. The method according to claim 1, characterized in that, The step of obtaining the advertising features of each candidate advertisement corresponding to the identifier of the object to be pushed includes: Obtain the converted advertisement corresponding to the identifier of the object to be pushed, and obtain each related advertisement of the converted advertisement; Each relevant advertisement is used as a candidate advertisement, and the advertising features of each candidate advertisement are obtained.

12. An advertising push device, characterized in that, The device includes: The object feature acquisition module is used to acquire the object information features corresponding to the identifier of the object to be pushed; The advertising feature acquisition module is used to acquire the advertising features of each candidate advertisement corresponding to the identifier of the object to be pushed. The candidate advertisements include unconverted advertisements and converted advertisements. The advertising features of the converted advertisements include converted advertisement information features and converted features. The converted features are obtained by feature extraction of the conversion behavior information of the converted advertisements. The advertising features of the unconverted advertisements include unconverted advertisement information features. The conversion calculation module is used to calculate the conversion probability of an advertisement based on the object information features and the advertisement features, and to obtain the conversion probability of each candidate advertisement. The conversion probability is used to characterize the likelihood that the object identifier to be pushed will convert the candidate advertisement. The push module is used to filter each candidate advertisement based on the conversion probability to obtain the advertisement to be pushed, and push the advertisement to be pushed to the terminal corresponding to the target identifier.

13. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 11.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 11.

15. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 11.

Images