A method, apparatus, electronic device, and storage medium for pushing objects.

Abstract

This disclosure relates to an object push method, apparatus, electronic device, and storage medium. The method includes: determining estimated feedback data of a candidate object; if the estimated feedback data is less than the posterior feedback data of the sharing end corresponding to the candidate object, determining the feedback potential value of the candidate object; the feedback potential value indicates the likelihood that the candidate object will be subject to a preset feedback behavior in the future; increasing the sharing data corresponding to the candidate object according to the feedback potential value to obtain increased sharing data; and pushing the object to the account to be pushed based on the increased sharing data. Therefore, this disclosure, by adding a feedback potential value to the sharing data of candidate objects with low estimated feedback data, can increase the exposure rate of candidate objects with potential feedback, helping candidates with low exposure or in the cold start stage to quickly gain wider acceptance, improving the prediction ability of candidate objects, and increasing the efficiency of object push.

Description

Technical Field

[0001] This disclosure relates to the field of Internet technology, and in particular to an object push method, apparatus, electronic device, and storage medium. Background Technology

[0002] With the popularization of the internet, online information push systems are being used more and more widely on various internet platforms. The types of information pushed to users by online information push systems include products, videos, pictures, and even services. This model has become an indispensable paradigm for the business operations of major internet companies today.

[0003] In related technologies, information push systems employ a data-driven push model. Typically, multiple sharing endpoints provide data for the information they wish to share. The information push system then adjusts this data based on various factors, selecting suitable information to push to the user. However, current information push systems often miss out on high-potential information or information in its cold start phase. This means that information with low exposure or in a cold start phase is less competitive and less likely to be pushed to the user, potentially missing out on the high-value feedback these information could provide. Therefore, the push strategy of current information push systems needs optimization. Summary of the Invention

[0004] This disclosure provides an object push method, apparatus, electronic device, and storage medium. The technical solution of this disclosure is as follows:

[0005] According to a first aspect of the present disclosure, an object push method is provided, comprising:

[0006] Determine the estimated feedback data for candidate targets; the estimated feedback data represents the probability that the account to be pushed to will perform a preset feedback behavior on the candidate targets;

[0007] If the estimated feedback data is less than the posterior feedback data of the sharing end corresponding to the candidate object, the feedback potential value of the candidate object is determined; the posterior feedback data reflects the actual feedback of the historical sharing objects of the sharing end; the feedback potential value indicates the possibility that the candidate object will be subject to the preset feedback behavior in the future, and the feedback potential value is determined based on the historical sharing data of the sharing end.

[0008] The sharing data corresponding to the candidate object is increased based on the feedback potential value to obtain the increased sharing data; the sharing data is positively correlated with the probability of the candidate object being pushed.

[0009] Based on the enlarged sharing data, object pushes are sent to the accounts to be pushed.

[0010] In some possible embodiments, the method further includes:

[0011] Obtain the account characteristics of the account to be pushed to and the object characteristics of the candidate objects;

[0012] Operational data is estimated based on account characteristics and object characteristics to obtain predicted operational data; the predicted operational data represents the probability of the account to be pushed to click on the candidate object.

[0013] The sharing data provided by the sharing end is weighted using the estimated operation data to obtain the sharing data corresponding to the candidate object.

[0014] In some possible embodiments, the estimated feedback data for determining candidate objects includes:

[0015] Obtain the feedback data prediction model; the feedback data prediction model is updated according to the preset update frequency;

[0016] Based on the feedback data prediction model, the account characteristics of the account to be pushed and the object characteristics of the candidate object are estimated using feedback data to obtain the predicted feedback data.

[0017] In some possible embodiments, the method further includes:

[0018] Acquire the actual operation data and actual feedback data of the historical sharing objects; the actual operation data of the historical sharing objects includes the number of historical push accounts that clicked on the historical sharing objects; the actual feedback data of the historical sharing objects includes the number of historical push accounts that clicked on the historical sharing objects and performed preset feedback behaviors on the historical sharing objects.

[0019] The ratio between the actual feedback data of the historical sharing object and the actual operation data of the historical sharing object is determined as the posterior feedback data of the sharing end.

[0020] In some possible embodiments, the historical sharing data on the sharing end includes initial sharing data within a preset historical time period, feedback result data within the preset historical time period, and historical push data of candidate objects; the historical push data characterizes the interaction data between the candidate object and the historical push account during the historical push process; determining the feedback potential value of the candidate object includes:

[0021] The ratio between the feedback results data and the initial shared data is used as the first reference feedback potential value for the candidate.

[0022] Based on the historical push data of the candidate objects, the second reference feedback potential value of the candidate objects is determined by the confidence interval upper bound algorithm.

[0023] The feedback potential value of the candidate object is determined based on the first and second reference feedback potential values.

[0024] In some possible embodiments, the historical push data of the candidate object includes the historical number of times the candidate object has been displayed and the average feedback result value of the candidate object; based on the historical push data of the candidate object, a second reference feedback potential value of the candidate object is determined using a confidence interval upper bound algorithm, including:

[0025] The upper limit of the confidence interval for each candidate is determined based on the total number of push notifications, the historical display count of each candidate, and the average feedback result value of each candidate.

[0026] The upper limit of the confidence interval of a candidate is determined as the second reference feedback potential value of the candidate.

[0027] In some possible embodiments, based on the historical push data of the candidate object, a second reference feedback potential value of the candidate object is determined using a confidence interval upper bound algorithm, including:

[0028] The expected feedback result value of the candidate is determined based on the fusion characteristics of the account to be pushed to and the candidate object, and the current feature parameters of the candidate object; the fusion characteristics are determined based on the account characteristics of the account to be pushed to and the object characteristics of the candidate object; the current feature parameters are determined based on the historical push data of the candidate object.

[0029] Based on the expected feedback result value, fusion characteristics, fusion characteristic matrix, and preset exploration coefficients of the candidate objects, the upper limit of the confidence interval of the candidate objects is determined; the fusion characteristic matrix includes the historical fusion characteristics between the object characteristics and the historical object characteristics of the corresponding historical push objects of the candidate objects;

[0030] The upper limit of the confidence interval of a candidate is determined as the second reference feedback potential value of the candidate.

[0031] According to a second aspect of the present disclosure, an object pushing device is provided, comprising:

[0032] The first determination module is configured to execute the estimated feedback data for determining candidate objects; the estimated feedback data represents the probability that the account to be pushed to will perform a preset feedback behavior on the candidate object;

[0033] The second determining module is configured to determine the feedback potential value of a candidate object if the estimated feedback data is less than the posterior feedback data of the sharing end corresponding to the candidate object; the posterior feedback data reflects the actual feedback situation of the sharing objects in the past of the sharing end; the feedback potential value indicates the possibility that the candidate object will be subject to the preset feedback behavior in the future, and the feedback potential value is determined based on the historical sharing data of the sharing end.

[0034] The third determination module is configured to increase the sharing data corresponding to the candidate object based on the feedback potential value, and obtain the increased sharing data; the sharing data is positively correlated with the probability of the candidate object being pushed.

[0035] The push module is configured to push objects to the target account based on the enlarged shared data.

[0036] In some possible embodiments, the apparatus further includes:

[0037] The fourth determination module is configured to perform the following actions: obtain the account characteristics of the account to be pushed to and the object characteristics of the candidate object; estimate the operation data based on the account characteristics and object characteristics to obtain the estimated operation data; the estimated operation data represents the probability that the account to be pushed to click on the candidate object; and use the estimated operation data to weight the sharing data provided by the sharing terminal to obtain the sharing data corresponding to the candidate object.

[0038] In some possible embodiments, the first determining module is further configured to execute a feedback data prediction model; the feedback data prediction model is updated according to a preset update frequency; and feedback data is estimated based on the account characteristics of the account to be pushed and the object characteristics of the candidate object to obtain the predicted feedback data.

[0039] In some possible embodiments, the apparatus further includes:

[0040] The fifth determination module is configured to retrieve the actual operation data and actual feedback data of the historical sharing object; the actual operation data of the historical sharing object includes the number of historical push accounts that clicked on the historical sharing object; the actual feedback data of the historical sharing object includes the number of historical push accounts that clicked on the historical sharing object and performed preset feedback behavior on the historical sharing object; the ratio between the actual feedback data of the historical sharing object and the actual operation data of the historical sharing object is determined as the post-feedback feedback data of the sharing end.

[0041] In some possible embodiments, the historical sharing data of the sharing terminal includes initial sharing data within a preset historical time period, feedback result data within the preset historical time period, and historical push data of the candidate object; the historical push data represents the interaction data between the candidate object and the historical push account during the historical push process;

[0042] The second determining module is further configured to perform the following: using the ratio between the feedback result data and the initial shared data as the first reference feedback potential value of the candidate object; determining the second reference feedback potential value of the candidate object based on the candidate object's historical push data using a confidence interval upper bound algorithm; and determining the feedback potential value of the candidate object based on the first and second reference feedback potential values.

[0043] In some possible embodiments, the historical push data of the candidate object includes the historical number of times the candidate object has been displayed and the average feedback result value of the candidate object;

[0044] The second determining module is also configured to determine the upper limit of the confidence interval of a candidate based on the total number of pushes, the historical number of times the candidate is displayed, and the average feedback result value of the candidate; and to determine the upper limit of the confidence interval of the candidate as the second reference feedback potential value of the candidate.

[0045] In some possible embodiments, the second determining module is further configured to perform the following steps: determining the expected feedback result value of the candidate object based on the fusion characteristics of the account to be pushed and the candidate object and the current feature parameters of the candidate object; the fusion characteristics are determined based on the account characteristics of the account to be pushed and the object characteristics of the candidate object; the current feature parameters are determined based on the historical push data of the candidate object; determining the upper limit of the confidence interval of the candidate object based on the expected feedback result value of the candidate object, the fusion characteristics, the fusion characteristic matrix and the preset exploration coefficient; the fusion characteristic matrix includes the historical fusion characteristics between the object characteristics and the historical object characteristics of the historical push objects corresponding to the candidate object; and determining the upper limit of the confidence interval of the candidate object as the second reference feedback potential value of the candidate object.

[0046] According to a third aspect of the present disclosure, an electronic device is provided, comprising:

[0047] processor;

[0048] Memory used to store processor-executable instructions;

[0049] The processor is configured to execute instructions to implement the object push method of the first aspect of the present disclosure.

[0050] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided, which, when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform the object pushing method of the first aspect of the present disclosure.

[0051] According to a fifth aspect of the present disclosure, a computer program product is provided, the computer program product including a computer program stored in a readable storage medium, wherein at least one processor of a computer device reads from the readable storage medium and executes the computer program, causing the computer device to perform the object push method of the first aspect of the present disclosure.

[0052] The technical solutions provided by the embodiments of this disclosure have at least the following beneficial effects:

[0053] By determining the estimated feedback data of candidate objects; when the estimated feedback data is less than the posterior feedback data of the sharing end corresponding to the candidate object, the feedback potential value of the candidate object is determined; the feedback potential value indicates the likelihood that the candidate object will be subject to a preset feedback behavior in the future; the sharing data corresponding to the candidate object is increased based on the feedback potential value to obtain the increased sharing data, and then the object is pushed to the accounts to be pushed based on the increased sharing data; thus, for candidate objects with low estimated feedback data, adding the feedback potential value to the candidate object's sharing data can increase the exposure rate of candidate objects with potential feedback, help candidate objects with low exposure or in the cold start stage to quickly break into the circle, improve the prediction ability of candidate objects, and improve the efficiency of object push.

[0054] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0055] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0056] Figure 1 This is a schematic diagram illustrating an application environment according to an exemplary embodiment;

[0057] Figure 2 This is a flowchart illustrating an object pushing method according to an exemplary embodiment;

[0058] Figure 3 This is a flowchart illustrating a method for obtaining shared data of candidate objects according to an exemplary embodiment;

[0059] Figure 4 This is a flowchart illustrating a method for determining estimated feedback data for candidate objects according to an exemplary embodiment;

[0060] Figure 5 This is a flowchart illustrating a method for determining the feedback potential value of a candidate object according to an exemplary embodiment;

[0061] Figure 6 This is a flowchart illustrating a method for determining a second reference feedback potential value for a candidate object according to an exemplary embodiment;

[0062] Figure 7 This is a flowchart illustrating another method for determining a second reference feedback potential value for a candidate object, according to an exemplary embodiment.

[0063] Figure 8 This is a block diagram illustrating an object pushing device according to an exemplary embodiment;

[0064] Figure 9 This is a block diagram illustrating an electronic device for object pushing according to an exemplary embodiment. Detailed Implementation

[0065] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0066] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar first 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 the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0067] It should be noted that the user information involved in this disclosure (including but not limited to user device information, user personal information, etc.) is all information authorized by the user or fully authorized by all parties.

[0068] Please see Figure 1 , Figure 1 A schematic diagram illustrating an application environment for an object push method according to an exemplary embodiment is shown below. Figure 1 As shown, it includes a client 110 and a server 120. The client 110 and the server 120 can be connected via a wired network or a wireless network, which is not limited in this disclosure.

[0069] In this embodiment, server 120 can push objects to client 110. First, server 120 determines the estimated feedback data of candidate objects, which represents the probability that the client 110's account to be pushed to will perform a preset feedback behavior on the candidate object. Then, server 120 compares the estimated feedback data with the posterior feedback data of the sharing end corresponding to the candidate object, which reflects the actual feedback of the historical sharing objects of the sharing end. If the estimated feedback data is less than the posterior feedback data, server 120 determines the feedback potential value of the candidate object. The feedback potential value indicates the possibility that the candidate object will be subject to a preset feedback behavior in the future, and the feedback potential value is determined based on the historical sharing data of the sharing end. Then, server 120 increases the sharing data according to the feedback potential value to obtain increased sharing data. Finally, server 120 pushes objects to the client 110's account to be pushed to based on the increased sharing data.

[0070] The aforementioned client 110 may include, but is not limited to, smartphones, desktop computers, tablets, laptops, smart speakers, digital assistants, augmented reality (AR) / virtual reality (VR) devices, smart wearable devices, and other similar clients. It may also be software running on the aforementioned client, such as an application (App). An application can be a standalone application or a subroutine within an application. For example, an application may be a news application, a live streaming application, or a video application. Optionally, the operating system running on the client 110 may include, but is not limited to, Android, iOS, Linux, Windows, Unix, etc.

[0071] The aforementioned server 120 can be a server providing background services for the application in client 110. Specifically, the service provided by server 120 can be a topic tag recommendation service. Server 120 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing 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 (Content Delivery Network), and big data and artificial intelligence platforms. The operating system running on server 120 can include, but is not limited to, Android, iOS, Linux, Windows, Unix, etc.

[0072] It should be understood that Figure 1 The application environment shown is only an example. In actual applications, the object push method of this disclosure embodiment can be executed independently by the client or the server, or the object push method of this disclosure embodiment can be executed in cooperation between the terminal and the server. This disclosure embodiment does not limit the specific application environment.

[0073] Figure 2 This is a flowchart illustrating an object push method according to an exemplary embodiment, such as... Figure 2 As shown, the object push method can be applied to the server side or other node devices, and includes the following steps:

[0074] In step S201, the estimated feedback data of the candidate object is determined; the estimated feedback data represents the probability that the account to be pushed to will perform a preset feedback behavior on the candidate object.

[0075] In this embodiment, the server can provide an object push service. The server has a massive user base, and sharing clients with object promotion needs can request the server to promote their shared objects. Generally, the sharing client provides sharing data to the server for its shared objects, and this sharing data is positively correlated with the probability of the candidate object being pushed. In a specific application scenario, the sharing data represents the price set by the sharing client for the audience it wants to reach. For the sharing client, this price setting is crucial; the price not only determines the amount of traffic acquired but also affects the final sharing effect of the shared objects.

[0076] In real-world scenarios, after receiving a sharing request from a sharing client, the server can respond to the client user's object push request by pushing the object to that client. When pushing an object to the client, the server needs to consider multiple shared objects from different sharing clients. Based on the sharing data provided by each sharing client and other factors, it selects a suitable object from among the multiple shared objects to push to the client. This ensures that the pushed object is one that the client is genuinely interested in, and that the sharing client receives positive feedback from the push result.

[0077] In related technologies, for each of the multiple objects to be recommended, the server adjusts the sharing data provided by the sharing platform corresponding to each object based on a combination of factors. Typically, the server will evaluate the feedback that each object may receive by pushing it based on the historical feedback data of each sharing platform, and then select objects with higher feedback value for push. However, this is not friendly to objects with low exposure in historical feedback data or objects in the cold start stage. The probability of them being selected by the server to push to users is very low. If the server implements this strategy of only considering objects with high feedback in historical data for a long time, it may also miss the unexpected high results and high-value feedback that such objects with low exposure or in the cold start stage may bring.

[0078] Based on this, the object push method provided in this embodiment adjusts the sharing data provided by the sharing end. For candidate objects with low estimated feedback data, a feedback potential value is superimposed on the sharing data of the candidate objects. This can increase the possibility of objects with poor historical performance but potentially high returns being pushed to users. This is very friendly to objects with low exposure or those in the cold start stage. From the user's perspective, the objects pushed this time are different from those pushed in the past, and their novelty is improved, which may unexpectedly gain the user's interest. From the sharing end's perspective, even sharing objects with poor historical feedback data can get exposure opportunities, which can reduce sharing costs.

[0079] In some possible embodiments, in response to an object push operation by a client user, a candidate object corresponding to that client is determined; that is, the candidate object is related to the user of the client currently to be recommended. Specifically, prior to step S201 above, the object push method of this disclosure embodiment further includes: determining a set of candidate objects corresponding to the account to be recommended from an object resource pool based on the account characteristics of the account to be recommended; the set of candidate objects includes at least one candidate object; the candidate object represents an object that the account to be recommended may be interested in.

[0080] The object resource pool includes objects requested for sharing from different sharing endpoints; the objects are content that matches the actual application scenario; in some possible application scenarios, objects may include music, short videos, products, news, advertisements, etc. Account characteristics may include, but are not limited to, at least one of account name, account gender, account region, and account preferences; the above-mentioned object push operation may be based on the refresh operation of the client's recommendation page. When a user performs a refresh operation on the client's recommendation page, an object push request is sent to the server; the server responds to the object push request, identifies the account of the client user who sent the request as the account to be recommended, and obtains the account characteristics of the account to be recommended; then, based on the account characteristics, existing search algorithms, such as collaborative filtering, are used to find objects that the account to be recommended may be interested in from the object resource pool as candidate objects.

[0081] After identifying at least one candidate object that the account to be recommended might be interested in, for each candidate object, the server obtains the sharing data corresponding to that candidate object; in some possible embodiments, before step S201 above, the object push method of this disclosure embodiment may further include as follows: Figure 3 The following steps are shown:

[0082] In step S301, the account characteristics of the account to be pushed and the object characteristics of the candidate object are obtained.

[0083] Account features may include, but are not limited to, at least one of account name, account gender, account region, and account preferences. Object features match objects in the actual application scenario, and object features may include, but are not limited to, at least one of object name, object text information, object image information, and object audio information.

[0084] In step S303, operation data is estimated based on account characteristics and object characteristics to obtain estimated operation data; the estimated operation data represents the probability of the account to be pushed to click on the candidate object.

[0085] In this step, the server can use the operation data prediction model to estimate the operation data of account characteristics and object characteristics, and obtain the predicted operation data.

[0086] Specifically, the server obtains a pre-trained operation data prediction model, then inputs account features and object features into the model to estimate operation data and outputs the predicted operation data. Here, the operation data prediction model can specifically be a click-through rate (CTR) prediction model. This model can be built and trained using network structures such as Multilayer Perceptron (MLP), Squeeze-and-Excitation Networks (SENET), Factorization Machines (FM), and Field-aware Factorization Machines (FFM). Correspondingly, the predicted operation data represents the probability of the account clicking on the candidate object to be pushed to.

[0087] In step S305, the sharing data provided by the sharing end is weighted using the estimated operation data to obtain the sharing data corresponding to the candidate object.

[0088] In this step, the server uses the probability that the account to be pushed clicks on the candidate object to weight the sharing data provided by the sharing end, and obtains the weighted sharing data, which is the sharing data corresponding to the candidate object.

[0089] For example, in a bidding push scenario, the server identifies three candidate accounts to be recommended: candidate A, candidate B, and candidate C. Candidate A is shared by sharing platform A, candidate B by sharing platform B, and candidate C by sharing platform C. Sharing platform A provides an initial bid (price A) of 100 for candidate A, sharing platform B provides an initial bid (price B) of 94 for candidate B, and sharing platform C provides an initial bid (price C) of 96 for candidate C. The server uses click-through rate (CTR) prediction... The model estimates the probability (ctrA) of the account to be recommended clicking on candidate A as 0.92, the probability (ctrB) of the account to be recommended clicking on candidate B as 0.98, and the probability (ctrC) of the account to be recommended clicking on candidate C as 0.9. Then, the server uses the click-through rate of each candidate to weight the initial bids, i.e., price*ctr, to obtain the weighted initial bids priceA' of candidate A as 92, priceB' of candidate B as 92.12, and priceC' of candidate C as 86.4.

[0090] In the above embodiments, the server estimates the probability of the recommended account clicking on the candidate object based on account characteristics and object characteristics; using the estimated click-through rate of the candidate object, the sharing data of the candidate object provided by the sharing end is initially adjusted. Therefore, even if the sharing data of the candidate object provided by the sharing end is low, as long as the estimated click-through rate of the candidate object is high, the corresponding sharing data of the candidate object can be increased, which helps to increase the probability of the candidate object being pushed, shorten the cold start time of the candidate object, and increase the exposure speed.

[0091] In this embodiment, the server can estimate the probability that the account to be pushed to will perform a preset feedback behavior on each candidate object, thus obtaining the estimated feedback data for each candidate object. The preset feedback behavior can match the actions that an object can perform in a real-world application scenario. For example, in some application scenarios, the preset feedback behavior may include purchasing or downloading. In actual business scenarios, feedback data can also be referred to as conversion rate, which is a metric for measuring the sharing effectiveness of the sharing end.

[0092] In some possible embodiments, step S201 above, determining the estimated feedback data of the candidate object, may include, for example: Figure 4 The following steps are shown:

[0093] In step S401, the feedback data prediction model is obtained; the feedback data prediction model is updated according to a preset update frequency.

[0094] In actual business scenarios, feedback data prediction models can also be called conversion rate prediction models, which are used to predict conversion rates; optionally, conversion rate prediction models can also be built based on the above-mentioned MLP, SENET, FM, and FFM.

[0095] Here, the feedback data prediction model can be trained online in real time. Each time a push is made, the current feedback data prediction model can be obtained first, updated, and then the updated feedback data prediction model can be used for the current prediction.

[0096] Specifically, during each push, the server can determine the estimated loss between the historical estimated feedback data of the historical sharing object based on the current feedback data prediction model and the actual feedback data of the historical sharing object obtained from the online log data. Then, the parameters of the current feedback data prediction model are updated using this estimated loss to obtain the updated feedback data prediction model.

[0097] In detail, firstly, the server can obtain the current feedback data prediction model and the historical predicted feedback data of previously shared objects. The historical predicted feedback data is obtained by using the current feedback data prediction model to predict the conversion rate of previously shared objects. Here, a historically shared object is an object that the sharing client previously requested from the server and that the server actually recommended to the user. These objects include music, short videos, products, news, advertisements, etc. The historical predicted feedback data refers to the conversion rate of a previously shared object, predicted by the server using a conversion rate prediction model before recommending it. For example, before recommending a product advertisement, the server uses a conversion rate prediction model to predict the probability that a user will click on the advertisement and purchase the product; similarly, before pushing a live stream, the server uses a conversion rate prediction model to predict the probability that a user will click on the live stream and place an order there.

[0098] Then, the server can obtain online log data, which includes actual feedback data from historical sharing objects. Based on the actual feedback data and historical estimated feedback data, the server can update the current feedback data prediction model, resulting in an updated feedback data prediction model.

[0099] In step S403, the feedback data is estimated based on the account characteristics of the account to be pushed and the object characteristics of the candidate object according to the feedback data prediction model to obtain the predicted feedback data.

[0100] In this step, the server can use the updated feedback data prediction model to estimate the account characteristics of the account to be pushed and the object characteristics of the candidate object to obtain the predicted feedback data.

[0101] In this step, during the current push process, the server inputs the account characteristics of the account to be pushed and the object characteristics of the candidate object into the updated feedback data prediction model, performs feedback data estimation on the candidate object, and obtains the predicted feedback data, that is, the probability that the account to be pushed will perform the preset feedback behavior on the candidate object.

[0102] After the server uses the updated feedback data prediction model to estimate the feedback data of the candidate objects, it uses the updated feedback data prediction model as the current feedback data prediction model. In the next push, the server can obtain the actual feedback data of the candidate objects estimated in this push from the online log data, and then compare the actual feedback data of the candidate objects with the predicted feedback data output by the model to determine the prediction loss, and update the model again.

[0103] Based on the example above, for candidate A, candidate B, and candidate C, the server uses the updated conversion rate prediction model to predict the probability pre_cvrA of the account to be pushed to candidate A, the probability pre_cvrB of the account to be pushed to candidate B, and the probability pre_cvrC of the account to be pushed to candidate C.

[0104] In the above embodiments, the server updates the current feedback data prediction model each time a push is made, using the historical predicted feedback data of the historical sharing objects and the actual feedback data of the historical sharing objects obtained from the online log data. In this way, by making full use of the timeliness of the data, continuously training the model online, and updating and optimizing the model parameters in a timely manner, the accuracy of the model prediction can be improved.

[0105] In step S203, if the estimated feedback data is less than the posterior feedback data of the sharing end corresponding to the candidate object, the feedback potential value of the candidate object is determined; the posterior feedback data reflects the actual feedback situation of the historical sharing objects of the sharing end; the feedback potential value indicates the possibility that the candidate object will be subject to the preset feedback behavior in the future, and the feedback potential value is determined based on the historical sharing data of the sharing end.

[0106] In this embodiment, in addition to determining the estimated feedback data for each candidate object, the server also determines the post-feedback feedback data for each sharing end corresponding to each candidate object. Here, the post-feedback feedback data reflects the actual feedback of the sharing end's historical sharing objects. Historical sharing objects are those objects that the sharing end has previously shared, which were successfully pushed to at least one account by the server, and at least one account has performed an operation (e.g., click) and / or a preset feedback behavior (e.g., purchase, download) on that historical sharing object. The post-feedback feedback data can be calculated using a specific calculation method; optionally, it can be calculated using the following embodiments.

[0107] In some possible embodiments, determining the post-feedback data of the sharing end may include: obtaining the actual operation data and actual feedback data of the historical sharing object; wherein, the actual operation data of the historical sharing object includes the number of historical push accounts that clicked on the historical sharing object; the actual feedback data of the historical sharing object includes the number of historical push accounts that clicked on the historical sharing object and performed a preset feedback behavior on the historical sharing object; both the actual operation data and the actual feedback data of the historical sharing object can be obtained from online log data; then, the ratio between the actual feedback data of the historical sharing object and the actual operation data of the historical sharing object is determined as the post-feedback data of the sharing end.

[0108] Specifically, as mentioned above, feedback data represents conversion rate in actual business scenarios; correspondingly, the actual feedback data of the aforementioned historical sharing objects includes the number of historical push accounts that have had preset conversion behaviors towards the historical sharing objects; the actual operation data of the historical sharing objects includes the number of historical push accounts that have clicked on the historical sharing objects; it should be noted here that the premise for an account to implement preset conversion behaviors is that it has already clicked on the historical sharing objects. Therefore, a historical push account that has had preset conversion behaviors towards a historical sharing object must have clicked on the historical sharing object.

[0109] Based on this, the server can directly calculate the ratio between the number of historical push accounts that have had preset conversion behavior towards the historical shared object and the number of historical push accounts that have clicked on the historical shared object for each sharing terminal, and thus obtain the retrospective conversion rate of each sharing terminal.

[0110] Taking the sharing client A in the example above as an example, when the server determines the posterior feedback data, i.e. the posterior conversion rate, of sharing client A, it obtains the actual feedback data and actual operation data of the historical sharing object D that sharing client A has shared in the past. Assuming that the historical sharing object D was successfully pushed to 1000 accounts by the server, among these 1000 accounts, the number of accounts that have clicked on the historical sharing object D is 800, and the number of accounts that have purchased the historical sharing object D is 400. Of course, these 400 accounts are the accounts in the aforementioned 800 accounts; thus, the server can calculate that the posterior conversion rate of sharing client A is 400 / 800, which is 0.5.

[0111] In the above embodiments, the server calculates the posterior feedback data of each sharing end based on the actual operation data and actual feedback data of the historical sharing objects of each sharing end. Thus, the posterior feedback data can reflect the actual feedback of the historical sharing objects of the sharing end, which is beneficial for subsequent comprehensive consideration in combination with the estimated feedback data, so as to further optimize and adjust the sharing data.

[0112] In related technologies, after determining the estimated feedback data of the candidate and the posterior feedback data of the corresponding sharing end, if the estimated feedback data is greater than or equal to the posterior feedback data, it indicates that the candidate has a high expectation of receiving positive feedback from the account to be pushed to. In this case, the server increases the sharing data to increase the probability that the candidate will be pushed to the account to be pushed to. Conversely, if the estimated feedback data is less than the posterior feedback data, it indicates that the candidate has a low expectation of receiving feedback from the account to be pushed to. In this case, the server decreases the sharing data to decrease the probability that the candidate will be pushed to the account to be pushed to.

[0113] In actual business scenarios, when implementing the above embodiments, the server improves the ranking of candidate objects with high estimated conversion rates among multiple candidate objects, while suppressing other candidate objects with low estimated conversion rates. This can improve the post-conversion rate of the sharing end. However, after a period of full-scale operation, a problem of difficulty in scaling up will occur. That is, when comparing the post-conversion rate and the estimated conversion rate, the server usually tends to select candidate objects with higher estimated conversion rates than post-conversion rates. This will cause the post-conversion rate to continue to increase, leading to a gradual increase in the requirement for the estimated conversion rate. Consequently, fewer and fewer candidate objects meet the requirements, and scaling up the candidate objects becomes increasingly difficult.

[0114] Based on this, in this embodiment of the disclosure, if the estimated feedback data is less than the posterior feedback data, the server determines the feedback potential value of the candidate object; here, the feedback potential value indicates the likelihood that the candidate object will be subject to a preset feedback behavior in the future. Specifically, the feedback potential value is determined based on the historical sharing data of the sharing end; the calculation method of the feedback potential value will be described in the specific embodiments below, and will not be elaborated here.

[0115] This disclosure primarily considers the potential of a candidate to receive feedback when the estimated feedback data is less than the posterior feedback data, aiming to improve the candidate's ranking. For cases where the estimated feedback data is greater than or equal to the posterior feedback data, the server can directly set the feedback potential value of a candidate to a fixed value, such as 1. That is, for candidates with high expected feedback, there is no need to further consider their ability to receive feedback. Thus, the server further adjusts the shared data for each candidate using their feedback potential value, resulting in adjusted shared data.

[0116] In some possible embodiments, in addition to determining the feedback potential value of a candidate object by directly judging the magnitude of the predicted feedback data and the posterior feedback data, the server can also set specific conditions. The server can perform simple calculations on the predicted feedback data and the posterior feedback data, and under certain specific conditions, adopt the corresponding method for determining the feedback potential value to obtain the feedback potential value of different candidate objects.

[0117] In a specific example, this particular condition could be, when the following conditions are met... In this case, the feedback potential value of the candidate object is determined according to the specific implementation method given below; where pre_cvr is the estimated feedback data of the candidate object; post_cvr is the post-feedback feedback data of the sharing end corresponding to the candidate object; α and β are hyperparameters that can be customized to meet the business needs of different application scenarios.

[0118] The following describes the situation where the predicted feedback data is less than the posterior feedback data, or, if the following conditions are met... In this case, a specific implementation method for determining the feedback potential value of a candidate object.

[0119] In some possible embodiments, the historical sharing data on the sharing end includes initial sharing data within a preset historical time period, feedback result data within the preset historical time period, and historical push data of candidate objects. The preset historical time period can be any historical time span before the current moment. The preset historical time period can be selected based on the actual data volume; if the daily data volume is high, the preset historical time period can be set shorter. For reference, the preset historical time period can be the past 1 day, 3 days, 5 days, several weeks, or several months. Initial sharing data can be the initial data invested when sharing historical sharing objects, which can be understood as cost. Correspondingly, feedback result data can be the benefits brought by the conversion behavior of historical sharing objects.

[0120] Therefore, the feedback potential value for determining candidate objects mentioned above may include, for example: Figure 5 The following steps are shown:

[0121] In step S501, the ratio between the feedback result data and the initial shared data is used as the first reference feedback potential value of the candidate object.

[0122] In this step, when calculating the feedback potential value of a candidate object, the server first considers the overall sharing benefits of the sharing end corresponding to the candidate object from the perspective of the sharing end. In a specific application scenario, the return on investment (ROI) can be used to characterize the first reference feedback potential value of the candidate object.

[0123] In real-world push notification scenarios, Return on Investment (ROI) refers to the ratio of input to output, serving as a crucial reference for measuring performance. Therefore, the server uses ROI calculation methods to determine the ratio between the revenue generated by the conversion behavior of historically shared objects within a preset historical time period and the initial data input (cost) for sharing those historically shared objects. This ratio is then used as the first reference potential value for candidate objects.

[0124] In step S503, based on the historical push data of the candidate object, the second reference feedback potential value of the candidate object is determined using the confidence interval upper bound algorithm.

[0125] Among them, historical push data represents the interaction data between the candidate and the historical push account during the historical push process.

[0126] In this step, the server also considers the potential feedback capability of the candidate object from the perspective of the candidate object; thus, based on the candidate object's historical push data, the server uses the confidence interval upper bound algorithm to determine the second reference feedback potential value of the candidate object; the specific calculation method will be described in detail below, and will not be elaborated here.

[0127] It should be noted that other bandit algorithms, similar to the confidence interval upper bound algorithm, can also be used in other embodiments; the bandit algorithm aims to balance the exploration and exploitation problems. Exploitation utilizes the user's historical behavior to discover the user's interests and uses the currently optimal solution, i.e., selecting the object with the highest predicted feedback value by the model for recommendation. Obviously, if recommendations are only made based on the user's historical behavior, similar products may be continuously pushed to the user, ignoring other user interests. Therefore, exploration explores the user's potential interests. Thus, the bandit algorithm can minimize the loss caused by exploration, thereby converging to an approximate globally optimal strategy.

[0128] This disclosure utilizes a typical bandit algorithm, namely the confidence interval upper bound algorithm, to improve the exploratory nature of the algorithm when pushing based on shared data. Other embodiments may also employ algorithms such as Thompson sampling (TS) or Neural-linear algorithms.

[0129] In step S505, the feedback potential value of the candidate object is determined based on the first reference feedback potential value and the second reference feedback potential value.

[0130] In this step, the server can directly sum the first and second reference feedback potential values, or perform a weighted summation, and use the summation result as the feedback potential value of the candidate object.

[0131] In the above embodiments, the server considers the feedback potential value of the candidate object from the perspective of both the candidate object and the sharing end corresponding to the candidate object. Based on the historical sharing benefit ratio of the sharing end and the historical push effect of the candidate object, the server comprehensively considers the feedback potential value of the candidate object, that is, the possibility of obtaining high-value feedback. This value is then used to further adjust the sharing data in subsequent steps to obtain the final sharing data of the candidate object. In this way, compared with the traditional push method of directly selecting candidate objects based on the sharing data, i.e., the initial bid, this disclosure superimposes a bandit strategy on the traditional method and uses the feedback potential value to adjust the sharing data. This can enhance the server's ability to explore objects with potential high-value feedback during the push process. For candidate objects whose estimated feedback data is less than the posterior feedback data, the server can increase the probability of them being "revived" and improve their ranking in the push sorting. This can not only improve the diversity of object push but also bring unexpected high results and high-value feedback to the sharing end.

[0132] The following describes two specific embodiments of step S503 above.

[0133] First, in some possible embodiments, the historical push data of the candidate object may include the historical number of times the candidate object was displayed and the average feedback result value of the candidate object;

[0134] Accordingly, step S503 above, which determines the second reference feedback potential value of the candidate object based on the candidate object's historical push data and using a confidence interval upper bound algorithm, may include, for example: Figure 6 The following steps are shown:

[0135] In step S601, the upper limit of the confidence interval of the candidate object is determined based on the total number of pushes, the historical display number of the candidate object, and the average feedback result value of the candidate object.

[0136] In this step, the server uses the most basic calculation method of the Upper Confidence Bound (UCB) algorithm. The core idea is to select the candidate with the highest upper confidence interval among all candidates as the target object. Specifically, the server can calculate the upper confidence interval of the candidate object according to the following formula (1):

[0137]

[0138] Where, x j This represents the upper limit of the confidence interval for the candidate object; The average feedback result value of candidate j is represented by n; n represents the total number of push notifications; n j This indicates the number of times a candidate has been displayed in history.

[0139] In step S603, the upper limit of the confidence interval of the candidate object is determined as the second reference feedback potential value of the candidate object.

[0140] In this step, the server calculates the upper limit of the confidence interval of the candidate object based on the total number of pushes, the historical display number of the candidate object and the average feedback result value of the candidate object using the above formula (1), and then directly determines the upper limit of the confidence interval of the candidate object as the second reference feedback potential value of the candidate object.

[0141] Secondly, in some other possible embodiments, step S503 above, which determines the second reference feedback potential value of the candidate object based on the candidate object's historical push data and using a confidence interval upper bound algorithm, may also include, for example... Figure 7 The following steps are shown:

[0142] In step S701, the expected feedback result value of the candidate object is determined based on the fusion characteristics of the account to be pushed and the candidate object and the current characteristic parameters of the candidate object.

[0143] Among them, the fusion features are determined based on the account features of the account to be pushed to and the object features of the candidate objects; the current feature parameters are determined based on the historical push data of the candidate objects.

[0144] The UCB algorithm is a context-free bandit algorithm, while the LinUCB algorithm introduces feature information into UCB. In this embodiment, the server determines the upper limit of the confidence interval of candidate objects based on the LinUCB algorithm. Specifically, the server first determines the fusion feature representing the combination of the account to be pushed and the candidate object based on the account features of the account to be pushed and the object features of the candidate object. At each selection, based on the fusion feature between the account to be pushed and each candidate object, the expected feedback result value of each candidate object is estimated, i.e., the expected benefit. This benefit can be represented by the probability of an action such as clicking, the probability of implementing a preset conversion action such as purchasing or downloading, or other forms.

[0145] In the LinUCB algorithm, there is a linear relationship between the expected return of the candidate object and the fusion feature. The current feature parameter is the linear parameter between the expected return obtained online and the fusion feature. The current feature parameter will be updated in the next calculation based on the expected return calculated each time. In the next calculation, the expected return is compared with the actual online feedback data to complete the update of the current feature parameter.

[0146] In step S703, the upper limit of the confidence interval of the candidate object is determined based on the expected feedback result value, fusion feature, fusion feature matrix and preset exploration coefficient of the candidate object.

[0147] The fusion feature matrix includes historical fusion features between the object features and the historical object features of the historical push objects corresponding to the candidate objects.

[0148] In this step, the server determines the upper limit of the confidence interval of the candidate object based on the expected feedback result value, fusion feature, fusion feature matrix and preset exploration coefficient of the candidate object using the following formula (2):

[0149]

[0150] Where, p t,a This represents the upper limit of the confidence interval for candidate object a; The expected feedback result value of candidate object a is represented by the superscript T, which indicates the transpose of the matrix; the superscript -1 indicates the inverse of the matrix; A a The fusion feature matrix of the candidate objects; x a This represents the fusion characteristics of the candidate and the account to be recommended; α represents the preset exploration coefficient, which determines the degree of algorithm exploration. The smaller α is, the more likely it is to select candidate with higher expected returns at present, and the larger α is, the more likely it is to select candidate with poor historical returns.

[0151] In step S705, the upper limit of the confidence interval of the candidate object is determined as the second reference feedback potential value of the candidate object.

[0152] In this step, the server calculates the upper limit of the confidence interval of the candidate object based on the expected feedback result value, fusion feature, fusion feature matrix and preset exploration coefficient of the candidate object using the above formula (2), and then directly determines the upper limit of the confidence interval of the candidate object as the second reference feedback potential value of the candidate object.

[0153] In the two embodiments described above, different upper bound algorithms for confidence intervals are used to determine the second reference feedback potential value of a candidate object, thereby determining the candidate object's feedback potential value. This allows for adjustments to the shared data of the candidate object based on its feedback potential value when the estimated feedback data is low, assigning a higher exploration score to the candidate object's shared data. This gives previously potentially rejected candidates a chance to be exposed, effectively mitigating the cold start problem and the long-tail distribution problem. Furthermore, the upper bound algorithms for confidence intervals are completely independent and highly robust. If an online incident occurs and the "exploration" capability needs to be removed, the calculation of the second reference feedback potential value can be directly omitted, and the system can still operate correctly and stably.

[0154] In step S205, the sharing data corresponding to the candidate object is increased based on the feedback potential value to obtain the increased sharing data; the sharing data is positively correlated with the probability of the candidate object being pushed.

[0155] In this embodiment of the disclosure, for candidate objects whose estimated feedback data is less than their posterior feedback data, the server adjusts the sharing data according to the feedback potential value after determining its feedback potential value. Specifically, the sharing data corresponding to the candidate object is increased according to the feedback potential value to obtain the increased sharing data. Thus, candidate objects with higher increased sharing data have a higher probability of being pushed.

[0156] In some possible embodiments, the process of increasing the shared data corresponding to the candidate object based on the feedback potential value may include the following steps: the server inputs the feedback potential value of the candidate object into an existing fusion function such as a linear function or a sigmoid function to calculate the corresponding function value, and then sums the function value with the shared data to obtain the increased shared data.

[0157] In step S207, based on the enlarged sharing data, object push is performed to the account to be pushed.

[0158] In this embodiment of the disclosure, for each candidate among multiple candidate objects of the account to be recommended, the shared data of each candidate object is determined, wherein the shared data of some candidate objects is increased; thus, when the server pushes objects to the account to be pushed, it can sort the multiple candidate objects based on the shared data of each candidate object, take the top N candidate objects as the target objects, and then push the target objects to the account to be pushed; wherein N is a natural number greater than or equal to 1.

[0159] In summary, the object push method provided in this disclosure determines the estimated feedback data of a candidate object; when the estimated feedback data is less than the posterior feedback data of the sharing end corresponding to the candidate object, it determines the feedback potential value of the candidate object; the feedback potential value indicates the likelihood that the candidate object will be subject to a preset feedback behavior in the future; the sharing data corresponding to the candidate object is increased based on the feedback potential value to obtain increased sharing data, and then the object is pushed to the account to be pushed based on the increased sharing data. Thus, for candidate objects with low estimated feedback data, adding a feedback potential value to the sharing data can increase the exposure rate of candidate objects with potentially high feedback probability, giving wings to the traditional object push algorithm based on sharing data, helping candidate objects with low exposure or in the cold start stage to quickly gain wider acceptance, improving the prediction ability of candidate objects, and increasing the efficiency of object push.

[0160] Figure 8 This is a block diagram illustrating an object pushing device according to an exemplary embodiment. (Refer to...) Figure 8 The device includes a first determining module 801, a second determining module 802, a third determining module 803, and a pushing module 804.

[0161] The first determining module 801 is configured to execute the estimated feedback data for determining candidate objects; the estimated feedback data represents the probability that the account to be pushed to will perform a preset feedback behavior on the candidate object;

[0162] The second determining module 802 is configured to determine the feedback potential value of a candidate object if the estimated feedback data is less than the posterior feedback data of the sharing end corresponding to the candidate object; the posterior feedback data reflects the actual feedback situation of the sharing objects in the past of the sharing end; the feedback potential value indicates the possibility that the candidate object will be subject to the preset feedback behavior in the future, and the feedback potential value is determined based on the historical sharing data of the sharing end.

[0163] The third determining module 803 is configured to perform an adjustment process on the sharing data corresponding to the candidate object based on the feedback potential value, so as to obtain the adjusted sharing data; the sharing data is positively correlated with the probability of the candidate object being pushed.

[0164] Push module 804 is configured to perform object push to the target account based on the enlarged sharing data.

[0165] In some possible embodiments, the apparatus further includes:

[0166] The fourth determination module is configured to perform the following actions: obtain the account characteristics of the account to be pushed to and the object characteristics of the candidate object; estimate the operation data based on the account characteristics and object characteristics to obtain the estimated operation data; the estimated operation data represents the probability that the account to be pushed to click on the candidate object; and use the estimated operation data to weight the sharing data provided by the sharing terminal to obtain the sharing data corresponding to the candidate object.

[0167] In some possible embodiments, the first determining module 801 is further configured to execute the acquisition of feedback data prediction model; the feedback data prediction model is updated according to a preset update frequency; and feedback data is estimated based on the account characteristics of the account to be pushed and the object characteristics of the candidate object to obtain the predicted feedback data.

[0168] In some possible embodiments, the apparatus further includes:

[0169] The fifth determination module is configured to retrieve the actual operation data and actual feedback data of the historical sharing object; the actual operation data of the historical sharing object includes the number of historical push accounts that clicked on the historical sharing object; the actual feedback data of the historical sharing object includes the number of historical push accounts that clicked on the historical sharing object and performed preset feedback behavior on the historical sharing object; the ratio between the actual feedback data of the historical sharing object and the actual operation data of the historical sharing object is determined as the post-feedback feedback data of the sharing end.

[0170] In some possible embodiments, the historical sharing data of the sharing terminal includes initial sharing data within a preset historical time period, feedback result data within the preset historical time period, and historical push data of the candidate object; the historical push data represents the interaction data between the candidate object and the historical push account during the historical push process;

[0171] The second determining module 802 is further configured to perform the following: using the ratio between the feedback result data and the initial shared data as the first reference feedback potential value of the candidate object; using the confidence interval upper bound algorithm to determine the second reference feedback potential value of the candidate object based on the historical push data of the candidate object; and determining the feedback potential value of the candidate object based on the first reference feedback potential value and the second reference feedback potential value.

[0172] In some possible embodiments, the historical push data of the candidate object includes the historical number of times the candidate object has been displayed and the average feedback result value of the candidate object;

[0173] The second determining module 802 is further configured to determine the upper limit of the confidence interval of a candidate object based on the total number of pushes, the historical display count of the candidate object, and the average feedback result value of the candidate object; and to determine the upper limit of the confidence interval of the candidate object as the second reference feedback potential value of the candidate object.

[0174] In some possible embodiments, the second determining module 802 is further configured to perform the following steps: determining the expected feedback result value of the candidate object based on the fusion characteristics of the account to be pushed and the candidate object and the current feature parameters of the candidate object; the fusion characteristics are determined based on the account characteristics of the account to be pushed and the object characteristics of the candidate object; the current feature parameters are determined based on the historical push data of the candidate object; determining the upper limit of the confidence interval of the candidate object based on the expected feedback result value of the candidate object, the fusion characteristics, the fusion characteristic matrix and the preset exploration coefficient; the fusion characteristic matrix includes the historical fusion characteristics between the object characteristics and the historical object characteristics of the historical push objects corresponding to the candidate object; and determining the upper limit of the confidence interval of the candidate object as the second reference feedback potential value of the candidate object.

[0175] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0176] Figure 9 This is a block diagram illustrating an electronic device for object pushing according to an exemplary embodiment.

[0177] This electronic device can be a server or a terminal device, and its internal structure diagram can be as follows: Figure 9As shown, this electronic device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When executed by the processor, the computer program implements an object pushing method.

[0178] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the electronic device to which the present disclosure is applied. A specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0179] In an exemplary embodiment, an electronic device is also provided, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the object push method as described in the embodiments of this disclosure.

[0180] In an exemplary embodiment, a computer-readable storage medium is also provided, which, when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform the object pushing method of the present disclosure embodiments.

[0181] In an exemplary embodiment, a computer program product is also provided, comprising a computer program stored in a readable storage medium, wherein at least one processor of a computer device reads from the readable storage medium and executes the computer program, causing the computer device to perform the object push method of the present disclosure embodiments.

[0182] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This 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 of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0183] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0184] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. An object push method, characterized in that, include: Determine the estimated feedback data for candidate objects; The estimated feedback data represents the probability that the account to be pushed to will perform a preset feedback behavior on the candidate object; If the estimated feedback data is less than the posterior feedback data of the sharing end corresponding to the candidate object, the initial sharing data of the sharing end within a preset historical time period, the feedback result data within the preset historical time period, and the historical push data of the candidate object are obtained; the historical push data represents the interaction data between the candidate object and the historical push account during the historical push process; the posterior feedback data reflects the actual feedback of the historical sharing object of the sharing end. The ratio between the feedback result data and the initial shared data is used as the first reference feedback potential value of the candidate object. Based on the historical push data of the candidate objects, the second reference feedback potential value of the candidate objects is determined using the confidence interval upper bound algorithm; The first reference feedback potential value and the second reference feedback potential value are summed, and the summation result is used as the feedback potential value of the candidate object; The feedback potential value indicates the likelihood that the candidate object will be subject to the preset feedback behavior in the future; Based on the feedback potential value, the shared data corresponding to the candidate object is increased to obtain the increased shared data. The shared data is positively correlated with the probability that the candidate object is pushed to the system. Based on the increased sharing data, object pushes are performed to the accounts to be pushed.

2. The object push method according to claim 1, characterized in that, The method further includes: Obtain the account characteristics of the account to be pushed to and the object characteristics of the candidate object; Based on the account characteristics and the object characteristics, operation data is estimated to obtain predicted operation data; the predicted operation data represents the probability that the account to be pushed to clicks the candidate object. The sharing data provided by the sharing terminal is weighted using the estimated operation data to obtain the sharing data corresponding to the candidate object.

3. The object push method according to claim 1, characterized in that, The predicted feedback data for determining candidate objects includes: Obtain a feedback data prediction model; the feedback data prediction model is updated according to a preset update frequency; The predicted feedback data is obtained by estimating the account characteristics of the account to be pushed and the object characteristics of the candidate object based on the feedback data prediction model.

4. The object push method according to claim 1, characterized in that, The method further includes: Obtain the actual operation data and actual feedback data of the historical sharing object; the actual operation data of the historical sharing object includes the number of historical push accounts that clicked on the historical sharing object; the actual feedback data of the historical sharing object includes the number of historical push accounts that clicked on the historical sharing object and performed the preset feedback behavior on the historical sharing object; The ratio between the actual feedback data of the historical sharing object and the actual operation data of the historical sharing object is determined as the posterior feedback data of the sharing end.

5. The object push method according to claim 1, characterized in that, The historical push data of the candidate object includes the historical number of times the candidate object was displayed and the average feedback result value of the candidate object. The step of determining the second reference feedback potential value of the candidate object based on the candidate object's historical push data and using a confidence interval upper bound algorithm includes: The upper limit of the confidence interval of the candidate object is determined based on the total number of push notifications, the historical display count of the candidate object, and the average feedback result value of the candidate object; The upper limit of the confidence interval of the candidate object is determined as the second reference feedback potential value of the candidate object.

6. The object push method according to claim 1, characterized in that, The step of determining the second reference feedback potential value of the candidate object based on the candidate object's historical push data and using a confidence interval upper bound algorithm includes: The expected feedback result value of the candidate object is determined based on the fusion characteristics of the account to be pushed to and the candidate object and the current feature parameters of the candidate object; the fusion characteristics are determined based on the account characteristics of the account to be pushed to and the object characteristics of the candidate object; the current feature parameters are determined based on the historical push data of the candidate object. Based on the expected feedback result value of the candidate object, the fusion feature, the fusion feature matrix, and the preset exploration coefficient, the upper limit of the confidence interval of the candidate object is determined; the fusion feature matrix includes the historical fusion features between the object features and the historical object features of the historical push objects corresponding to the candidate object; The upper limit of the confidence interval of the candidate object is determined as the second reference feedback potential value of the candidate object.

7. An object pushing device, characterized in that, include: The first determination module is configured to execute the estimated feedback data for determining candidate objects; The estimated feedback data represents the probability that the account to be pushed to will perform a preset feedback behavior on the candidate object; The second determining module is configured to, if the estimated feedback data is less than the posterior feedback data of the sharing end corresponding to the candidate object, acquire the initial sharing data of the sharing end within a preset historical time period, the feedback result data within the preset historical time period, and the historical push data of the candidate object; the historical push data represents the interaction data between the candidate object and the historical push account during the historical push process; the posterior feedback data reflects the actual feedback situation of the historical sharing object of the sharing end; the ratio between the feedback result data and the initial sharing data is used as the first reference feedback potential value of the candidate object; and based on the historical push data of the candidate object, a second reference feedback potential value of the candidate object is determined using a confidence interval upper bound algorithm. The first reference feedback potential value and the second reference feedback potential value are summed, and the summation result is used as the feedback potential value of the candidate object; The feedback potential value indicates the likelihood that the candidate object will be subject to the preset feedback behavior in the future; The third determining module is configured to perform an enlargement process on the shared data corresponding to the candidate object based on the feedback potential value, so as to obtain the enlarged shared data. The shared data is positively correlated with the probability that the candidate object is pushed to the system. The push module is configured to push the shared data based on the candidate object to the account to be pushed.

8. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the object pushing method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is able to perform the object pushing method as described in any one of claims 1-6.

Images