Content pushing method, system and device, electronic equipment, storage medium and program product

Abstract

The embodiment of the invention discloses a content pushing method, system and device, electronic equipment, a storage medium and a program product. The method comprises the following steps: in response to a content loading request, determining a target session where the content loading request is located; for a target object triggering the target session and pushed content pushed to the target object in the target session, obtaining target feedback data of the target object for the pushed content; and calling a content prediction model based on the target feedback data to obtain first candidate push content for the target object, and determining target push content according to the first candidate push content and pushing the target push content. According to the technical scheme provided by the embodiment of the invention, the adaptation degree of content pushing can be ensured.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of data processing technology, and in particular to a content push method, system, device, electronic device, storage medium and program product. Background Technology

[0002] In content delivery systems, the compatibility between users' proactive exploration behavior and the system's content supply has become a core variable affecting commercialization results. Therefore, it is crucial to push relevant content to users.

[0003] In the process of realizing this invention, the inventors discovered the following technical problems in the prior art: the adaptability of content push cannot be guaranteed, which urgently needs to be solved. Summary of the Invention

[0004] This invention provides a content push method, system, device, electronic device, storage medium, and program product, which solves the problem of not being able to guarantee the adaptability of content push.

[0005] According to one aspect of the present invention, a content push method is provided, which may include:

[0006] In response to a content loading request, determine the target session in which the content loading request is made;

[0007] For the target object that triggered the target session and the content that has been pushed to the target object in the target session, obtain the target object's target feedback data for the pushed content;

[0008] Based on the target feedback data, the content prediction model is invoked to obtain the first candidate push content for the target object, and the target push content is determined and pushed based on the first candidate push content.

[0009] According to another aspect of the present invention, a content push system is provided, which may include: a perception service, a playback service, and a model service; wherein, the perception service corresponds to a perception cache; wherein,

[0010] The awareness service is used to respond to content loading requests, determine the target session in which the content loading request is located, and retrieve the target object’s target feedback data for the pushed content from the awareness cache for the target object that triggered the target session and the pushed content that has been pushed to the target object in the target session.

[0011] The perception service can also be used to call the content prediction model based on the target feedback data, obtain the first candidate push content for the target object, and store the first candidate push content in the perception cache;

[0012] The playback service is used to retrieve the first candidate push content from the perception cache through the model service, and determine the target push content based on the first candidate push content and push it.

[0013] According to another aspect of the present invention, a content push device is provided, which may include:

[0014] The target session determination module is used to determine the target session in response to a content loading request;

[0015] The target feedback data acquisition module is used to acquire target feedback data of the target object for the target object that triggered the target session and the content that has been pushed to the target object in the target session.

[0016] The content push module is used to call the content prediction model based on the target feedback data to obtain the first candidate push content for the target object, and then determine the target push content based on the first candidate push content and push it.

[0017] According to another aspect of the present invention, an electronic device is provided, which may include:

[0018] At least one processor; and

[0019] A memory that is communicatively connected to at least one processor; wherein,

[0020] The memory stores a computer program that can be executed by at least one processor, such that when the at least one processor executes the program, it implements the content push method provided in any embodiment of the present invention.

[0021] According to another aspect of the present invention, a computer-readable storage medium is provided having computer instructions stored thereon, the computer instructions being used to cause a processor to execute and implement the content push method provided in any embodiment of the present invention.

[0022] According to another aspect of the present invention, a computer program product is provided, on which a computer program is stored, which, when executed by a processor, implements the content push method provided in any embodiment of the present invention.

[0023] The technical solution of this invention, in response to a content loading request, determines the target session in which the content loading request resides; for the target object that triggered the target session and the content already pushed to the target object in the target session, it obtains the target object's target feedback data on the pushed content; based on the target feedback data, it calls a content prediction model to obtain a first candidate push content for the target object, and then determines the target push content based on the first candidate push content and pushes it to the target object. This technical solution, through a session association mechanism, links various push contents within the same session. Based on the target object's target feedback data on the pushed content (i.e., target feedback data reflecting the target object's real-time interests), it dynamically predicts the target push content that the target object might currently be interested in using a content prediction model, thereby ensuring the adaptability between the finally determined and pushed target push content and the target object.

[0024] It should be understood that the description in this section is not intended to identify key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart of a content push method provided according to an embodiment of the present invention;

[0027] Figure 2 This is a schematic diagram of the SIC service in a content push method provided according to an embodiment of the present invention;

[0028] Figure 3 This is a flowchart of another content push method provided according to an embodiment of the present invention;

[0029] Figure 4 This is a flowchart of another content push method provided according to an embodiment of the present invention;

[0030] Figure 5 This is a signaling diagram of a content push system provided according to an embodiment of the present invention;

[0031] Figure 6a This is a schematic diagram of an optional example of a content push system provided according to an embodiment of the present invention;

[0032] Figure 6bThis is another schematic diagram of an optional example of a content push system provided according to an embodiment of the present invention;

[0033] Figure 7 This is a structural block diagram of a content push device according to an embodiment of the present invention;

[0034] Figure 8 This is a schematic diagram of the structure of an electronic device that implements the content push method of this invention. Detailed Implementation

[0035] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0036] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. The same applies to "target," "original," etc., and will not be repeated here. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0037] It should be noted that the collection, gathering, updating, analysis, processing, use, transmission, and storage of user personal information involved in the technical solution of this invention all comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken to prevent unauthorized access to user personal information data and to maintain user personal information security and network security.

[0038] Before introducing the embodiments of the present invention, the application scenarios that may be involved in the embodiments of the present invention, the related content push schemes, and the reasons for the problem that the content push schemes cannot guarantee the adaptability of the content pushes will be explained by way of example, so as to better understand the content push schemes in the embodiments of the present invention.

[0039] For example, taking the push (i.e., recommendation) of advertising content to users as an example, in an advertising content push system, the compatibility between users' proactive exploration behavior and the system's content supply has become a core variable affecting commercialization results. The user's behavioral path from "active selection" to "deep browsing" is both an experience path and a value conversion chain. Since users' proactive selection is dominant, in order to better meet users' real-time interests, it is necessary to provide a more diversified content supply, including different types of advertisements (such as images, videos, and interactive advertisements) and different advertising content (such as promotional activities and new product pushes). In addition, after the advertising content is pushed to users, the depth of users' scrolling and browsing behavior will directly affect the overall advertising revenue. Therefore, big data and artificial intelligence technologies can be used to capture user interests in real time and push the most relevant (i.e., suitable) advertising content to improve user experience and advertising click-through rates and conversion rates.

[0040] Building on this, the relevant ad content push system adopts a typical "request-response" mechanism, triggering an independent request each time a user pulls down to turn a page, and then only performing calculations based on the content of the current request. That is, the system treats each page-turning request as an independent event, failing to associate multiple page-turning requests within the same session. Consequently, it cannot fully analyze user feedback behavior to different pushed content within the same session to identify real-time user interests, further resulting in a high degree of homogeneity in the pushed ad content and poor relevance to users, which urgently needs to be addressed.

[0041] Figure 1 This is a flowchart of a content push method provided in an embodiment of the present invention. This embodiment is applicable to content push scenarios, especially advertising content push scenarios. The method can be executed by the content push device provided in this embodiment of the present invention. This device can be implemented by software and / or hardware, and can be integrated into an electronic device, which can be various user terminals or servers.

[0042] See Figure 1 The method of this invention specifically includes the following steps:

[0043] S110. In response to a content loading request, determine the target session in which the content loading request is located.

[0044] In this context, a content loading request can be understood as a request triggered by the client to load content on the client side. For example, it could request to load at least one of the following: product content, novel content, or video content. The client can be understood as a program directly used and interacted with by the target object. Optionally, this program could be an application (APP), a mini-program, or a Web H5 page. All of the above are related to specific situations and are not specifically limited here. For example, a user can trigger a content loading request by performing actions such as pulling down or flipping left or right on the client side.

[0045] A target session can be understood as a session triggered by the target object through the client, specifically a session in which the target object triggers a content loading request, i.e., the session in which the content loading request resides. In this embodiment of the invention, optionally, each time the target object restarts the client, a new session is triggered. During a session, the target object may trigger one content loading request, or more likely, multiple content loading requests. Because these multiple content loading requests correspond to the same session, they are related to each other. The specific function of this relationship will be explained later.

[0046] Based on this, in response to a content loading request, the target session in which the content loading request is located is determined.

[0047] S120. For the target object that triggered the target session and the content that has been pushed to the target object in the target session, obtain the target object's target feedback data on the pushed content.

[0048] Here, "pushed content" can be understood as content that has been pushed to the target object in the target session. Specifically, the content loading request described in S110 is the most recently triggered request by the target object. For ease of distinction, the content loading request triggered by the target object in the target session before this request is referred to as the prior loading request. Therefore, the pushed content can be understood as the content determined and pushed to the target object based on the prior loading request. In this embodiment of the invention, optionally, some or all content loading requests triggered before this request can be regarded as prior loading requests. This can be set according to actual needs and is not specifically limited here.

[0049] Target feedback data can be understood as feedback data from the target audience regarding the pushed content. This feedback data can characterize whether the target audience has any interest in the pushed content. Based on this, and considering the application scenarios that may be involved in the embodiments of this invention, optionally, the feedback data can be at least one of browsing data, click data, and closing data. Specifically, browsing data can characterize whether the target audience has viewed the pushed content, thus characterizing the exposure of the pushed content. Further, browsing data can be swiping data, which can characterize whether the target audience is swiping quickly or slowly over the pushed content; browsing data can be dwell data, which can characterize whether the target audience is paying short-term or long-term attention to the pushed content; and so on. All of these can characterize the target audience's level of interest in the pushed content. Click data can characterize whether the target audience has clicked on the pushed content, i.e., whether the target audience has explored the pushed content in depth. Closing data can characterize whether the target audience has closed the pushed content, i.e., whether the target audience has no interest in the pushed content. Of course, target feedback data can also be other feedback data besides the examples above, which depends on the actual situation and is not specifically limited here.

[0050] S130. Based on the target feedback data, call the content prediction model to obtain the first candidate push content for the target object, and determine the target push content based on the first candidate push content and push it.

[0051] Here, the content directly obtained in response to a content loading request is referred to as the target loaded content. This target loaded content can be understood as the content originally requested by the target loading request (or the target object). Correspondingly, the target pushed content can be understood as the content pushed (i.e., returned) to the target object along with the target loaded content, intended for the target object to receive. This content can be advertising content, promotional content, event content, or benefit content, depending on the specific circumstances, and is not specifically limited here. For example, assuming the target loaded content is a brand of mobile phone, the target pushed content could be another mobile phone that the brand owner wants the target object to purchase; in this case, the target pushed content is equivalent to advertising content.

[0052] As explained above, there is a relationship between pre-load requests and content loading requests. This relationship is mainly reflected in the fact that both are requests triggered by the target object within a single session of the target session. Typically, a session lasts for a short time, meaning the time interval between their triggering times is short. This implies that when responding to a content loading request, the target object's feedback data regarding the already pushed content corresponding to the pre-load request can be considered real-time feedback, reflecting the target object's real-time interests. Therefore, the target pushed content corresponding to the content loading request can be determined based on this target feedback data, integrating all data within the target session (such as requests, pushed content, and feedback data) to improve the compatibility between the target pushed content and the target object.

[0053] Specifically, a content prediction model is invoked based on the target feedback data to obtain the first candidate push content for the target audience. This content prediction model can be understood as a model with content prediction capabilities, specifically a model used to predict content that the target audience might be interested in. In this embodiment, optionally, the model can be a deep learning model or an algorithm model such as linear regression or logistic regression, which can be set according to actual needs and is not specifically limited here. The first candidate push content can be understood as content that can be considered as target push content based on the output of the content prediction model.

[0054] Furthermore, the target push content is determined based on the first candidate push content. For example, the first candidate push content can be directly used as the target push content, or it can be determined by combining the first candidate push content with other content, etc., without specific limitations. The target push content is then pushed to the client operated by the target object, hoping that the target object will browse and gain a deeper understanding of the target push content through the client. Of course, at the same time, the target loading content can also be pushed to the target object.

[0055] In this embodiment of the invention, by introducing a model-aware mechanism, the target feedback data is analyzed in depth, thereby gaining a deeper understanding of the evolution of the target object's interests and adjusting the first candidate push content accordingly, thus ensuring the continued relevance and attractiveness of the target push content determined based on the first candidate push content.

[0056] Based on this, for example, a Field Perception Center (SIC) service can be pre-built. Then, the SIC service can be used to execute S110 to determine the target session, execute S220 to obtain target feedback data, and execute S230 to obtain the first candidate push content by calling the content prediction model. Based on this, the SIC service can be used to determine the target push content based on the first candidate push content and push it. Alternatively, other services or systems can be used to determine the target push content and push it, etc. This can be set according to actual needs and is not specifically limited here.

[0057] Based on this, see Figure 2 The SIC service constructed above may have at least one of the following capabilities:

[0058] 1. It can handle content loading requests from different sources (such as different product business lines of an APP), thereby enabling various content loading requests to be processed effectively, providing a solid data foundation for subsequent calculations and analysis, that is, it has strong multi-source heterogeneous data processing capabilities.

[0059] 2. It can provide low cost, high availability and powerful computing resources, support data calculation between multiple content loading requests within the same session (such as target feedback data integration, which is equivalent to preprocessing), so that it can efficiently (i.e. real-time) calculate large amounts of data and quickly respond to user requests, that is, it has strong computing power.

[0060] 3. By using high-performance caching services (such as Redis cluster), other services can write data to the Redis cluster through the write interface and read data from the Redis cluster through the read interface, thereby providing efficient data writing and querying capabilities and ensuring the real-time availability and consistency of data.

[0061] This example demonstrates how building a SIC service can collect, calculate, and integrate all target feedback data from a target object within the same session in real time. This significantly enhances the relevance and coherence of the data within the session, enabling the system to accurately capture and respond to the target object's immediate interests in real time, thereby providing the target object with more personalized and diverse targeted content.

[0062] The technical solution of this invention, in response to a content loading request, determines the target session in which the content loading request resides; for the target object that triggered the target session and the content already pushed to the target object in the target session, it obtains the target object's target feedback data on the pushed content; based on the target feedback data, it calls a content prediction model to obtain a first candidate push content for the target object, and then determines the target push content based on the first candidate push content and pushes it to the target object. This technical solution, through a session association mechanism, links various push contents within the same session. Based on the target object's target feedback data on the pushed content (i.e., target feedback data reflecting the target object's real-time interests), and using a content prediction model, it dynamically predicts the target push content that the target object might currently be interested in, thereby ensuring the fit between the finally determined and pushed target push content and the target object.

[0063] Based on this, an optional technical solution is to obtain target feedback data of the target object regarding the pushed content, including:

[0064] Obtain the session identifier assigned to the target session and determine the request identifier associated with the session identifier;

[0065] For the pre-load request corresponding to the pushed content, determine the target identifier corresponding to the pre-load request from all request identifiers, and obtain the feedback data corresponding to the target identifier as the target feedback data for the pushed content.

[0066] In this approach, a session identifier is pre-assigned to the target session. This allows each loading request triggered by the target object within the target session (such as the content loading request and prior loading requests described above) to be associated with this session identifier, thus effectively associating multiple loading requests within the target session. Optionally, in this technical solution, the session identifier can be represented by a page number (page_id), and the request identifier can be represented by a page number (page). Of course, these can be set according to actual needs; this is merely an example and not a specific limitation.

[0067] Based on this, when it is necessary to obtain target feedback data, the session identifier of the corresponding target session can be obtained first, then the request identifier associated with the session identifier can be obtained, and then for the pre-load request corresponding to the pushed content (that is, the pushed content is obtained by responding to the pre-load request), the target identifier used to identify the pre-load request can be determined from each request identifier, and then the feedback data corresponding to the target identifier in the various stored feedback data can be used as the target feedback data of the target object for the pushed content.

[0068] The above technical solution associates sessions and loading requests based on identifiers, thereby enabling the rapid and accurate acquisition of target feedback data.

[0069] Figure 3 This is a flowchart of another content push method provided by an embodiment of the present invention. This embodiment is based on and optimized from the above-described technical solutions. Optionally, in this embodiment, the above-described content push method further includes: acquiring multiple preset candidate push contents, and determining a second candidate push content from the multiple preset candidate push contents according to a content loading request; determining target push content based on the first candidate push content and pushing it, including: determining the target push content based on the first candidate push content and the second candidate push content and pushing it. The explanations of terms that are the same as or corresponding to those in the above embodiments are not repeated here.

[0070] See Figure 3 The method in this embodiment may specifically include the following steps:

[0071] S210. In response to a content loading request, determine the target session in which the content loading request is located.

[0072] S220. For the target object that triggered the target session and the content that has been pushed to the target object in the target session, obtain the target object's target feedback data on the pushed content.

[0073] S230. Based on the target feedback data, call the content prediction model to obtain the first candidate push content for the target object.

[0074] S240. Obtain multiple preset candidate push content, and determine the second candidate push content from the multiple preset candidate push content according to the content loading request.

[0075] The preset candidate push content can be understood as pre-set content that can be pushed to objects. These objects include at least the target object, and may also include other objects besides the target object. This can be set according to actual needs and is not specifically limited here. In this embodiment of the invention, optionally, the multiple preset candidate push contents can be obtained from a preset push candidate pool.

[0076] As explained above, a content loading request can represent the content originally requested by the target object, meaning it can also represent the target object's real-time interests to a certain extent. Therefore, a second candidate push content (i.e., the preset candidate push content that the target object may be interested in) can be determined from multiple preset candidate push content based on the content loading request. For example, the second candidate push content can be determined based on the triggering time and / or triggering method of the content loading request. In particular, the target loading content requested by the content loading request can be determined first, and then the second candidate push content can be determined based on the target loading content. Compared with the content loading request, the target loading content can more intuitively reflect the target object's real-time interests, thereby improving the adaptability between the second candidate push content and the target object, and further improving the adaptability between the target push content determined in this way and the target object.

[0077] S250. Based on the first candidate push content and the second candidate push content, determine the target push content and push it.

[0078] Specifically, the target push content is determined by combining the first candidate push content with the second candidate push content. The application of the second candidate push content further ensures the appropriateness of the content push.

[0079] The technical solution of this invention determines a second candidate push content from multiple preset candidate push content through a content loading request. Then, the first and second candidate push content are combined to determine the target push content. This achieves two goals: first, by associating the target feedback data and content loading request within the target session according to the session association mechanism, data continuity and integrity are ensured; second, this effectively uses the first candidate push content as part of the push candidate pool to determine the target push content, thus enabling dynamic expansion of the push candidate pool. These two points work together to further improve the adaptability between the target push content and the target object.

[0080] Figure 4This is a flowchart of another content push method provided by an embodiment of the present invention. This embodiment is based on the above-mentioned technical solutions and optimized. In this embodiment, optionally, the content prediction model is trained on the basis of the first prediction model. The content prediction model is pre-trained through the following steps: for the sample object that triggers the sample session and the sample push content pushed to the sample object in the sample session, the first feedback data of the sample object to the sample push content is obtained, so as to obtain a first sequence based on multiple first feedback data obtained sequentially in the sample session; a first conditional probability is determined according to the first prediction model and the first sequence, and a loss is calculated according to the first conditional probability to obtain a first loss; the parameters in the first prediction model are adjusted according to the first loss to obtain the content prediction model. The explanations of the same or corresponding terms as those in the above embodiments are not repeated here.

[0081] See Figure 4 The method in this embodiment may specifically include the following steps:

[0082] S310. For the sample object that triggers the sample session and the sample push content pushed to the sample object in the sample session, obtain the first feedback data of the sample object to the sample push content, so as to obtain a first sequence based on multiple first feedback data obtained sequentially in the sample session.

[0083] Here, a sample session can be understood as a session used during model training, a sample object can be understood as the object that triggers the session, and sample push content can be understood as the content pushed to the object in the session.

[0084] Specifically, each time a sample object triggers a loading request (referred to as a sample loading request for clarity), the sample push content obtained in response to that loading request is pushed to the sample object, thereby obtaining the sample object's first feedback data regarding that pushed content. In other words, each time a sample loading request is detected, the corresponding first feedback data can be obtained. Based on these sequentially obtained first feedback data, a first sequence can be derived. It can be understood that in this first sequence, the first feedback data obtained earlier is listed first, and the first feedback data obtained later is listed last.

[0085] S320. Determine the first conditional probability based on the first prediction model and the first sequence, and calculate the loss based on the first conditional probability to obtain the first loss.

[0086] As explained above, during the model inference phase, the content prediction model can predict future feedback data that has not yet been generated based on the generated target feedback data, and then perform content prediction based on the future feedback data. Therefore, during the model training phase, conditional probability can be used to calculate the loss.

[0087] Specifically, the first prediction model can be understood as a model that currently requires training. This model can be a completely untrained model or a model obtained after pre-training, which can be set according to actual needs, and no specific limitation is made here. Obtain the first prediction model.

[0088] A first conditional probability is determined based on a first prediction model and a first sequence. The number of such first conditional probabilities can be one or more, which is related to the process of determining the first conditional probability and is not specifically limited here.

[0089] For example, assuming the first sequence includes n first feedback data, the first n-1 first feedback data can be input into the first prediction model to obtain the first conditional probability of the first prediction model predicting the nth first feedback data based on the first n-1 first feedback data. At this time, there is only one first conditional probability.

[0090] For another example, each first feedback data point from the second to the last position in the first sequence is referred to as a subsequent feedback data point. For each subsequent feedback data point, the first feedback data point preceding it in the first sequence (referred to as the prior feedback data for clarity) is input into the first prediction model to obtain the first conditional probability that the first prediction model predicts the true value of the subsequent feedback data point based on the prior feedback data point. This first conditional probability can be understood as the probability that the sample object will give the subsequent feedback data point after giving the prior feedback data point. There are n-1 first conditional probabilities in this case. Based on this, the prior feedback data point can be directly input into the first prediction model, or the quantized semantics can be obtained first from the prior feedback data point, and then this quantized semantics can be used as the prior feedback data point and input into the first prediction model; etc. This can be set according to actual needs and is not specifically limited here.

[0091] Furthermore, the loss is calculated based on the first conditional probability to obtain the first loss. For example, the first conditional probability can be used directly as the first loss, or the sum of multiple first conditional probabilities or the sum of logarithmic operations can be used as the first loss. This can be set according to actual needs and is not specifically limited here.

[0092] S330. Adjust the parameters in the first prediction model according to the first loss to obtain the content prediction model.

[0093] S340. In response to a content loading request, determine the target session in which the content loading request is located.

[0094] S350. For the target object that triggered the target session and the content that has been pushed to the target object in the target session, obtain the target object's target feedback data on the pushed content.

[0095] S360. Based on the target feedback data, call the content prediction model to obtain the first candidate push content for the target object, and determine the target push content based on the first candidate push content and push it.

[0096] The technical solution of this invention determines a first conditional probability by using a first prediction model and a first sequence given by a sample object within a sample session. Then, a first loss is calculated based on the first conditional probability. Subsequently, a first prediction model is trained based on the first loss to obtain a content prediction model. The implementation of calculating the first loss based on the first conditional probability ensures that the content prediction model trained thereby can accurately predict future feedback data based on the target feedback data, and then accurately predict content based on the future feedback data.

[0097] Based on this, an optional technical solution is to train the first prediction model on the basis of the second prediction model, which is pre-trained through the following steps:

[0098] Obtain the second sequence, wherein multiple second feedback data in the second sequence are data fed back by the sample object before the sample session is triggered;

[0099] The second conditional probability is determined based on the second prediction model and the second sequence, and the loss is calculated based on the second conditional probability to obtain the second loss;

[0100] The parameters in the second prediction model are adjusted based on the second loss to obtain the first prediction model.

[0101] In this case, considering that there is no target feedback data when the target object first triggers the content loading request to start the target session, in order for the content prediction model to still make accurate content predictions in this situation, the first prediction model can be trained based on the second feedback data (or historical feedback data), so that the content prediction model trained on the basis of the first prediction model can make predictions based on the second feedback data.

[0102] Specifically, for ease of explanation, the concept of historical loading requests is given here. Historical loading requests can be understood as loading requests triggered by the sample object before the sample session is started. There can be multiple loading requests, and these multiple loading requests can be associated with or not associated with the same session. This can be selected according to actual needs, and no specific limitation is made here.

[0103] A second sequence can be pre-constructed, so that each time a historical loading request is detected, the historical push content corresponding to the historical loading request can be pushed to the sample object, thereby obtaining the sample object's second feedback data for the historical push content, and then adding the second feedback data to the second sequence, so that it includes multiple second feedback data in sequential order.

[0104] In this way, the second conditional probability can be determined based on the second prediction model and the second sequence. Then, the loss can be calculated based on the second conditional probability to obtain the second loss. Subsequently, the parameters in the second prediction model can be adjusted based on the second loss to obtain the first prediction model. This implementation process is similar to S320 and S330, and will not be described in detail here.

[0105] The above technical solution trains the first prediction model by using the second sequence fed back by the sample object before triggering the sample session, thereby enabling it to capture the long-term dependencies of historical feedback data. As a result, the content prediction model trained on it can still accurately predict content even when the target object first triggers a content loading request to start the target session (i.e., there is no target feedback data).

[0106] Therefore, the above text describes the model training process under different application scenarios. However, it should be noted that during the model inference process, there is no need to distinguish between the first visit and subsequent visits, because the only difference between the two is whether the target feedback data exists. Therefore, if the target feedback data can be obtained, it will be used as the input information of the content prediction model; otherwise, it will not be used as the input information.

[0107] Based on this, in order to better understand the training process of each model in the embodiments of the present invention, a complete example is given here in conjunction with the page turning request in the example above.

[0108] For example, as explained above, the content prediction model is trained based on a first prediction model, which in turn is trained based on a second prediction model. This second prediction model can be a large language model based on a decoder-only architecture. Therefore, the content prediction model trained on conditional probability can be considered a type of sequential autoregressive generative model. The model training process will be described in detail below.

[0109] When the target audience visits for the first time (i.e., on the homepage), the content prediction model primarily relies on historical feedback data (i.e., second feedback data) for content prediction. Therefore, the first step involves pre-training the second prediction model using an autoregressive mechanism based on the second sequence. This enables the trained first prediction model to capture long-term dependencies in historical feedback data, as detailed below:

[0110] a) Model input: Sample objects in historical time periods within The quantified semantic ID of each historical feedback data point (such as clicks, favorites, add-to-cart, and purchases) is denoted as... ;

[0111] b) Second Loss: Autoregressive pre-training is performed by maximizing the sequence generation probability (i.e., the second conditional probability). The second loss function is: ,in This indicates that the representation is based on the first t-1 quantized semantics. Generate the t-th quantization semantic The second conditional probability.

[0112] At this point, the first prediction model has been trained.

[0113] Furthermore, for subsequent page-turning requests, the content prediction model uses the collected target feedback data within the current real-time page to model the target object's interest divergence process and perform content prediction. Therefore, the second step involves fine-tuning the first prediction model within the page based on the first sequence, especially the first and second sequences, so that it can capture the target object's real-time behavior patterns within the page, thus obtaining the content prediction model. Specifically,

[0114] a) Model input: Sample objects The quantized semantic ID of each historical feedback data is counted as: The sequence of page exposures of the sample object within the sample session. The sequence of page clicks by the sample object within the sample session is denoted as... , here and This is the first sequence; the data clicked by the user on the page after pagination is recorded as... ,Right now ;

[0115] b) First loss: Supervised fine-tuning is performed by maximizing the sequence generation probability (i.e., the first conditional probability), and the second loss function is... .

[0116] At this point, the content prediction model has been trained.

[0117] In summary, the above example combines historical feedback data and real-time feedback data (i.e., the first feedback data, which in this example is obtained through...). and The model is trained by representing the target object, and the resulting content prediction model can guarantee the accuracy of content prediction in both homepage access and page-turning access scenarios.

[0118] Figure 5 This is a structural block diagram of a content push system provided in an embodiment of the present invention. This embodiment is applicable to content push scenarios, especially advertising content push scenarios.

[0119] See Figure 5The content push system of this invention includes: a perception service 410, a playback service 430, and a model service 420; wherein, the perception service 410 corresponds to a perception cache 440; wherein,

[0120] The perception service 410 can be used to respond to a content loading request, determine the target session where the content loading request is located, and obtain the target object’s target feedback data for the pushed content from the perception cache 440 for the target object that triggered the target session and the pushed content that has been pushed to the target object in the target session.

[0121] The perception service 410 is also used to call the content prediction model based on the target feedback data, obtain the first candidate push content for the target object, and store the first candidate push content in the perception cache 440;

[0122] The playback service 430 can be used to obtain the first candidate push content from the perception cache 440 through the model service 420, and determine the target push content based on the first candidate push content and push it.

[0123] In this embodiment of the invention, optionally, the perception service 410 (i.e., the SIC service exemplified above), playback service 430, and model service 420 can be deployed on different electronic devices, or at least some services can be deployed on the same electronic device. Alternatively, the perception cache 440 can be implemented using a Redis cluster or a distributed memory grid, etc. All of the above can be configured according to actual needs, and no specific limitations are made here.

[0124] Based on this, in order to better understand the content push system described in the embodiments of the present invention, the following is combined with Figure 6a and Figure 6b The example provided illustrates how advertising content is pushed based on real-time user feedback.

[0125] For example, when the ad entry detects a request triggered by a user through the client (such as a homepage request or a page-turning request), the ad entry can forward the request to the SIC service and the playback service in parallel.

[0126] The SIC service, upon receiving a request, utilizes a session association mechanism to retrieve the corresponding target feedback data from the SIC cache. This data is then sent to the feature service via a routing service. The feature service extracts features from the received target feedback data to obtain quantified semantics. This quantified semantics is then sent to the content prediction model via the routing service. The content prediction model predicts and returns the first candidate push content based on the received quantified semantics. This first candidate push content represents the user's real-time feedback. The SIC service can then store the received first candidate push content in the SIC cache. In this way, the playback service can access the model service in real-time to retrieve the model calculation results (i.e., the first candidate push content) from the SIC cache and perform a coarse ranking of the first candidate push content.

[0127] At the same time, the playback service can recall second candidate push content from the push candidate pool based on the received request, and perform coarse ranking of each second candidate push content.

[0128] Furthermore, the playback service can perform fine ranking based on the results of the two coarse rankings to obtain target push content that is suitable for the user (i.e., the advertising content that the user is expected to receive), and push the advertising content to the client.

[0129] In the example above, the SIC service obtains target feedback data from the SIC cache and uses a model-aware mechanism to deeply understand the evolution of user interests, obtaining first candidate push content that reflects the user's real-time interests and storing it in the SIC cache. Then, the playback service can obtain the first candidate push content from the SIC cache through the model service and combine it with the second candidate push content recalled from the push candidate pool to determine the advertising content that is more suitable for the user. This reduces the homogenization of advertisements and maximizes the cost per thousand impressions (CPM) of advertisements, thereby optimizing the advertising performance.

[0130] Figure 7 This is a structural block diagram of a content push device provided in an embodiment of the present invention. This device is used to execute the content push method provided in any of the above embodiments. This device and the content push methods of the above embodiments belong to the same inventive concept. Details not described in detail in the embodiments of the content push device can be found in the embodiments of the above content push methods. See also... Figure 7 Specifically, the device may include: a target session determination module 510, a target feedback data acquisition module 520, and a content push module 530.

[0131] The target session determination module 510 can be used to determine the target session in which the content loading request is located in response to the content loading request.

[0132] The target feedback data acquisition module 520 can be used to acquire target feedback data of the target object for the target object that triggered the target session and the content that has been pushed to the target object in the target session.

[0133] The content push module 530 can be used to call the content prediction model based on the target feedback data to obtain the first candidate push content for the target object, and then determine the target push content and push it based on the first candidate push content.

[0134] Optionally, the above-mentioned content push device may further include:

[0135] The second candidate push content determination module can be used to obtain multiple preset candidate push contents and determine the second candidate push content from the multiple preset candidate push contents according to the content loading request.

[0136] Content push module 530 may include:

[0137] The content push unit can be used to determine the target push content and push it based on the first candidate push content and the second candidate push content.

[0138] Based on this, the optional second candidate push content determination module may include:

[0139] The second candidate push content determination unit can be used to determine the target content requested by the content loading request, and determine the second candidate push content from multiple preset candidate push contents based on the target content.

[0140] Optionally, the content prediction model is trained based on the first prediction model. The content prediction model is pre-trained through the following modules:

[0141] The first sequence acquisition module can be used to obtain the first feedback data of the sample object to the sample push content for the sample object that triggers the sample session and the sample push content pushed to the sample object in the sample session, so as to obtain the first sequence based on multiple first feedback data obtained sequentially in the sample session.

[0142] The first loss acquisition module is used to determine the first conditional probability based on the first prediction model and the first sequence, and to calculate the loss based on the first conditional probability to obtain the first loss;

[0143] The content prediction model acquisition module can be used to adjust the parameters in the first prediction model based on the first loss to obtain the content prediction model.

[0144] Based on this, an optional first loss deriving module may include:

[0145] The first conditional probability determination unit can be used to input the prior feedback data that precedes the subsequent feedback data in the first sequence into the first prediction model for each subsequent feedback data in the first sequence, so as to obtain the first conditional probability of the first prediction model predicting the subsequent feedback data based on the prior feedback data.

[0146] Alternatively, the first prediction model is trained based on the second prediction model, which is pre-trained using the following modules:

[0147] The second sequence acquisition module can be used to acquire a second sequence, wherein multiple second feedback data in the second sequence are data fed back by the sample object before the sample session is triggered;

[0148] The second loss acquisition module is used to determine the second conditional probability based on the second prediction model and the second sequence, and to calculate the loss based on the second conditional probability to obtain the second loss.

[0149] The pre-trained model module can be used to adjust the parameters in the second prediction model based on the second loss to obtain the first prediction model.

[0150] Optionally, the target feedback data acquisition module 520 may include:

[0151] The request identifier determination unit is used to obtain the session identifier assigned to the target session and determine the request identifier associated with the session identifier;

[0152] The target feedback data acquisition unit can be used to determine the target identifier corresponding to the pre-load request from all request identifiers for the pre-load request corresponding to the pushed content, and obtain the feedback data corresponding to the target identifier as the target feedback data of the target object for the pushed content.

[0153] The content push device provided in this embodiment of the invention can ensure the adaptability of the content push through the cooperation of various modules.

[0154] The content push device provided in the embodiments of the present invention can execute the content push method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0155] It is worth noting that in the embodiments of the above-mentioned content push device, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the scope of protection of the present invention.

[0156] Figure 8A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. This electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0157] like Figure 8 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0158] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0159] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as content push methods.

[0160] In some embodiments, the content push method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the content push method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the content push method by any other suitable means (e.g., by means of firmware).

[0161] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-chips or system-on-a-chips (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0162] Computer programs used to implement the methods of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs can be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0163] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0164] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0165] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0166] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0167] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication unit 19, or installed from storage unit 18, or installed from ROM 12. When the computer program is executed by processor 11, it performs the functions defined in the methods of the embodiments of the present invention.

[0168] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0169] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A content push method, characterized in that, include: In response to a content loading request, determine the target session in which the content loading request is located; For the target object that triggered the target session and the pushed content that has been pushed to the target object in the target session, obtain the target object's target feedback data for the pushed content; Based on the target feedback data, a content prediction model is invoked to obtain the first candidate push content for the target object, so as to determine the target push content and push it according to the first candidate push content.

2. The method according to claim 1, characterized in that, Also includes: Obtain multiple preset candidate push content, and determine a second candidate push content from the multiple preset candidate push content according to the content loading request; The step of determining the target push content based on the first candidate push content and pushing it includes: Based on the first candidate push content and the second candidate push content, the target push content is determined and pushed.

3. The method according to claim 2, characterized in that, The step of determining the second candidate push content from a plurality of preset candidate push content according to the content loading request includes: The target content to be loaded in the content loading request is determined, and a second candidate push content is determined from a plurality of preset candidate push content based on the target content.

4. The method according to claim 1, characterized in that, The content prediction model is trained based on the first prediction model, and the content prediction model is pre-trained through the following steps: For the sample object that triggers the sample session and the sample push content pushed to the sample object in the sample session, obtain the first feedback data of the sample object to the sample push content, so as to obtain a first sequence based on multiple first feedback data obtained sequentially in the sample session; A first conditional probability is determined based on the first prediction model and the first sequence, and a first loss is obtained by calculating the loss based on the first conditional probability. The parameters in the first prediction model are adjusted based on the first loss to obtain the content prediction model.

5. The method according to claim 4, characterized in that, Determining the first conditional probability based on the first prediction model and the first sequence includes: For each subsequent feedback data in the first sequence, the preceding feedback data in the first sequence is input into the first prediction model to obtain a first conditional probability that the first prediction model predicts the subsequent feedback data based on the preceding feedback data.

6. The method according to claim 4, characterized in that, The first prediction model is trained based on the second prediction model, which is pre-trained through the following steps: Obtain a second sequence, wherein multiple second feedback data in the second sequence are data fed back by the sample object before the sample session is triggered; The second conditional probability is determined based on the second prediction model and the second sequence, and the loss is calculated based on the second conditional probability to obtain the second loss; The parameters in the second prediction model are adjusted based on the second loss to obtain the first prediction model.

7. The method according to claim 1, characterized in that, The step of obtaining the target object's target feedback data for the pushed content includes: Obtain the session identifier assigned to the target session, and determine the request identifier associated with the session identifier; For the prior loading request corresponding to the pushed content, a target identifier corresponding to the prior loading request is determined from all the request identifiers, and the feedback data corresponding to the target identifier is obtained as the target feedback data of the target object for the pushed content.

8. A content push system, characterized in that, include: The system includes a perception service, a playback service, and a model service; wherein the perception service corresponds to a perception cache. The perception service is used to respond to a content loading request, determine the target session in which the content loading request is located, and obtain target feedback data of the target object for the pushed content from the perception cache for the target object that triggered the target session and the pushed content that has been pushed to the target object in the target session. The perception service is also used to call the content prediction model based on the target feedback data to obtain the first candidate push content for the target object, and store the first candidate push content in the perception cache; The playback service is used to obtain the first candidate push content from the perception cache through the model service, and determine the target push content based on the first candidate push content and push it.

9. A content push device, characterized in that, include: The target session determination module is used to determine the target session in which the content loading request is located in response to the content loading request; The target feedback data acquisition module is used to acquire target feedback data of the target object to the pushed content, for the target object that triggered the target session and the pushed content that has been pushed to the target object in the target session. The content push module is used to call the content prediction model based on the target feedback data to obtain the first candidate push content for the target object, so as to determine the target push content and push it according to the first candidate push content.

10. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor to cause the at least one processor to perform the content push method as described in any one of claims 1-7.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute and implement the content push method as described in any one of claims 1-7.

12. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the content push method as described in any one of claims 1-7.

Images