Content distribution method, apparatus, server, and storage medium

Abstract

The present disclosure relates to a content distribution method, device, server and storage medium, the method comprising: obtaining a predicted click rate of a to-be-recommended account on to-be-recommended content, and calibrating the predicted click rate by using actual click rate data in a target time period to obtain a calibrated click rate; determining an account group corresponding to the to-be-recommended account; the account group is divided according to the similarity of the attribute characteristics of the to-be-recommended account; determining a distribution threshold corresponding to the account group according to the calibrated click rate, an expected click rate pre-set for the account group, and the value of the to-be-recommended content distributed to the to-be-recommended account; and determining whether to distribute the to-be-recommended content corresponding to the calibrated click rate to the to-be-recommended account according to the size relationship between the distribution threshold and the calibrated click rate. The present scheme enables dynamic adjustment of the distribution threshold, so that the actual click rate of the to-be-recommended account can always meet the expected click rate constraint, and improves the control degree of online traffic.

Description

Technical Field

[0001] This disclosure relates to the Internet field, and in particular to content distribution methods, devices, servers, and storage media. Background Technology

[0002] When recommending content such as live streams and short videos to users, a distribution control algorithm should be used to determine whether to distribute the generated recommendation results to users in order to achieve reasonable content output, bring a better user experience, meet the constraints of the overall expected click-through rate, and maximize the cumulative value.

[0003] The core of commonly used distribution control algorithms is to group users, and the distribution decision for a user is determined by the user's predicted click-through rate (CTR) and the distribution decision threshold of the user's group. When the predicted CTR of a recommended piece of content is greater than the distribution decision threshold, the recommended content is distributed to that user.

[0004] However, during their research on the algorithm, the inventors discovered that the distribution decision threshold was a pre-set hyperparameter, while online conditions were dynamically changing. This threshold could not be dynamically adjusted based on real-time online metrics. This resulted in significant fluctuations in user click-through rates when using the algorithm, making it difficult to consistently meet the overall expected click-through rate. Furthermore, it had little impact on metrics such as page views and unique visitors, leading to poor online traffic control. Summary of the Invention

[0005] This disclosure provides a content distribution method, apparatus, server, and storage medium to at least address the problems in related technologies where the click-through rate of user accounts awaiting recommendation is difficult to consistently meet the overall expected click-through rate, and where online traffic control is poor. The technical solution of this disclosure is as follows:

[0006] According to a first aspect of the present disclosure, a content distribution method is provided, comprising:

[0007] The predicted click-through rate (CTR) of the account to be recommended for the content to be recommended is obtained, and the predicted CTR is calibrated using the actual CTR data within the target time period to obtain the calibrated CTR; the account to be recommended is the account that sends the recommendation request at the current moment, and the target time period is the time period in which the time difference between the current moment and the current moment is less than a first threshold.

[0008] Determine the account group corresponding to the account to be recommended; the account group is divided according to the similarity of the attribute features of the account to be recommended;

[0009] Based on the calibrated click-through rate, the pre-set expected click-through rate for the account group, and the value of the content to be recommended distributed to the account to be recommended, the distribution threshold corresponding to the account group is determined.

[0010] Based on the relationship between the distribution threshold and the calibrated click-through rate, it is determined whether to distribute the content to be recommended corresponding to the calibrated click-through rate to the account to be recommended.

[0011] Optionally, calibrating the predicted click-through rate using actual click-through rate data within the target time period to obtain a calibrated click-through rate includes:

[0012] Determine the predicted click-through rate interval to which the predicted click-through rate belongs, and obtain the target interval;

[0013] Obtain the actual click-through rate data of the accounts to be recommended within the target interval during the target time period;

[0014] The predicted click-through rate is calibrated based on the actual click-through rate data to obtain the calibrated click-through rate of the account to be recommended for the content to be recommended.

[0015] Optionally, the actual click-through rate (CTR) data includes the actual CTR corresponding to each predicted CTR value within the target interval, and the step of calibrating the predicted CTR based on the actual CTR data includes:

[0016] Determine the linear relationship between the predicted click-through rate and the calibrated click-through rate, wherein the linear relationship includes two parameters;

[0017] The two parameters are fitted using an objective function to obtain their values; the objective function is the squared error function corresponding to the calibrated click-through rate and the actual click-through rate.

[0018] Substituting the values ​​of the two parameters into the linear relationship, we obtain the calibration relationship corresponding to the target interval;

[0019] Using the calibration formula, the calibration click-through rate of each account to be recommended for the content to be recommended within the target interval is determined.

[0020] Optionally, obtaining the predicted click-through rate of the account to be recommended for the content to be recommended includes:

[0021] The attribute features of the account to be recommended and the attribute features of the content to be recommended are input into a preset ranking model to predict the click-through rate of the account to be recommended for the content to be recommended, thus obtaining the predicted click-through rate.

[0022] Optionally, determining the distribution threshold corresponding to the account group based on the calibrated click-through rate, the pre-set expected click-through rate for the account group, and the value of distributing the content to be recommended to the account to be recommended includes:

[0023] A decision function is determined that includes the calibrated click-through rate and the distribution threshold, the value of which is determined by the relationship between the calibrated click-through rate and the distribution threshold;

[0024] Based on the calibrated click-through rate and the decision function, determine the calibrated click-through rate function for the account group;

[0025] The candidate threshold is determined based on the calibration click-through rate function and the expected click-through rate value pre-set for the accounts to be recommended in the account group;

[0026] Based on the value of the content to be recommended distributed to the account to be recommended, the distribution threshold of the account group is determined from the candidate thresholds.

[0027] Optionally, determining the candidate threshold based on the calibrated click-through rate function and the pre-set expected click-through rate value for the accounts to be recommended in the account group includes:

[0028] Set an initial value; the initial value is a value between 0 and 1.

[0029] Starting with the initial value, a binary search is used to find a candidate threshold such that the value of the average click-through rate function is greater than the expected click-through rate distribution threshold.

[0030] Optionally, determining the distribution threshold for the account group from the candidate thresholds based on the value of distributing the content to be recommended to the account to be recommended includes:

[0031] A cumulative value function is determined with the candidate threshold as a variable. The cumulative value function represents the sum of the value brought about by distributing the content to be recommended to the corresponding account in the account group.

[0032] Iterate through the candidate thresholds to determine the value of the cumulative value function corresponding to each candidate threshold;

[0033] The candidate threshold that maximizes the value of the cumulative value function is determined as the distribution threshold for the account group.

[0034] According to a second aspect of the present disclosure, a content distribution apparatus is provided, comprising:

[0035] The calibration module is configured to obtain the predicted click-through rate of the account to be recommended for the content to be recommended, and calibrate the predicted click-through rate using the actual click-through rate data within the target time period to obtain the calibrated click-through rate; the account to be recommended is the account that sent the recommendation request at the current moment, and the target time period is the time period with a time difference of less than a first threshold from the current moment;

[0036] The account group determination module is configured to determine the account group corresponding to the account to be recommended; the account group is divided according to the similarity of the attribute features of the account to be recommended.

[0037] The distribution threshold determination module is configured to determine the distribution threshold corresponding to the account group based on the calibrated click-through rate, the expected click-through rate preset for the account group, and the value of distributing the content to be recommended to the account to be recommended.

[0038] The distribution determination module is configured to determine, based on the relationship between the distribution threshold and the calibration click-through rate, whether to distribute the content to be recommended corresponding to the calibration click-through rate to the account to be recommended.

[0039] Optionally, the calibration module is specifically configured to perform:

[0040] Determine the predicted click-through rate interval to which the predicted click-through rate belongs, and obtain the target interval;

[0041] Obtain the actual click-through rate data of the accounts to be recommended within the target interval during the target time period;

[0042] The predicted click-through rate is calibrated based on the actual click-through rate data to obtain the calibrated click-through rate of the account to be recommended for the content to be recommended.

[0043] Optionally, the calibration module is also configured to perform:

[0044] Determine the linear relationship between the predicted click-through rate and the calibrated click-through rate, wherein the linear relationship includes two parameters;

[0045] The two parameters are fitted using an objective function to obtain their values; the objective function is the squared error function corresponding to the calibrated click-through rate and the actual click-through rate.

[0046] Substituting the values ​​of the two parameters into the linear relationship, we obtain the calibration relationship corresponding to the target interval;

[0047] Using the calibration formula, the calibration click-through rate of each account to be recommended for the content to be recommended within the target interval is determined.

[0048] Optionally, the calibration module is also configured to perform:

[0049] The attribute features of the account to be recommended and the attribute features of the content to be recommended are input into a preset ranking model to predict the click-through rate of the account to be recommended for the content to be recommended, thus obtaining the predicted click-through rate.

[0050] Optionally, the distribution threshold determination module is specifically configured to execute:

[0051] A decision function is determined that includes the calibrated click-through rate and the distribution threshold, the value of which is determined by the relationship between the calibrated click-through rate and the distribution threshold;

[0052] Based on the calibrated click-through rate and the decision function, determine the calibrated click-through rate function for the account group;

[0053] The candidate threshold is determined based on the calibration click-through rate function and the expected click-through rate value pre-set for the accounts to be recommended in the account group;

[0054] Based on the value of the content to be recommended distributed to the account to be recommended, the distribution threshold of the account group is determined from the candidate thresholds.

[0055] Optionally, the distribution threshold determination module is further configured to perform:

[0056] Set an initial value; the initial value is a value between 0 and 1.

[0057] Starting with the initial value, a binary search is used to find a candidate threshold such that the value of the average click-through rate function is greater than the expected click-through rate distribution threshold.

[0058] Optionally, the distribution threshold determination module is further configured to perform:

[0059] A cumulative value function is determined with the candidate threshold as a variable. The cumulative value function represents the sum of the value brought about by distributing the content to be recommended to the corresponding account in the account group.

[0060] Iterate through the candidate thresholds to determine the value of the cumulative value function corresponding to each candidate threshold;

[0061] The candidate threshold that maximizes the value of the cumulative value function is determined as the distribution threshold for the account group.

[0062] According to a third aspect of the present disclosure, a server is provided, comprising:

[0063] processor;

[0064] Memory used to store the processor's executable instructions;

[0065] The processor is configured to execute the instructions to implement the content distribution method as described in the first aspect.

[0066] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided that, when instructions in the computer-readable storage medium are executed by a processor of a server, enables the server to perform the content distribution method as described in the first aspect.

[0067] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising a computer program, characterized in that, when the computer program is executed by a processor, it implements the content distribution method described in the first aspect.

[0068] The technical solutions provided by the embodiments of this disclosure have at least the following beneficial effects:

[0069] In embodiments of this disclosure, multiple accounts that have sent recommendation requests at the current time are obtained, and multiple pieces of content to be recommended corresponding to each account are determined. The multiple pieces of content to be recommended are selected from the existing content at the current time. Based on the attribute characteristics of the accounts to be recommended, the attribute characteristics of the content to be recommended, and the actual click-through rate data of the multiple accounts to be recommended within a target time period, the click-through rate of each account to be recommended for the content to be recommended is predicted, resulting in multiple calibrated click-through rates. The target time period is a time period in which the time difference with the current time is less than a first threshold. An account group corresponding to each account to be recommended is determined. The account group is divided according to the similarity of the attribute characteristics of the accounts to be recommended. For each account group, a distribution threshold corresponding to the account group is determined based on the calibrated click-through rate, the expected click-through rate preset for the accounts to be recommended in the group, and the value of the content to be recommended distributed to the accounts to be recommended. Based on the relationship between the distribution threshold and the calibrated click-through rate, it is determined whether to distribute the content to be recommended corresponding to the calibrated click-through rate to the accounts to be recommended. In the above method, the actual click-through rate (CTR) data within the target time period is also used as a factor to determine the calibration CTR, so that the calibration CTR can adapt to real-time online metrics. The determination of the distribution threshold is also related to the calibration CTR, the expected CTR, and the value of the content to be recommended to the accounts to be recommended, so that the distribution threshold can also be dynamically adjusted, and the actual CTR of the accounts to be recommended can always meet the constraints of the overall expected CTR. Furthermore, different account groups correspond to different distribution thresholds, so that the accounts to be recommended can receive more suitable recommended content, thereby increasing the influence on page views and unique visitors, and improving the control of online traffic.

[0070] 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

[0071] 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.

[0072] Figure 1 This is a flowchart illustrating the steps of a content distribution method according to an exemplary embodiment;

[0073] Figure 2 This is a flowchart illustrating the steps of another content distribution method according to an exemplary embodiment;

[0074] Figure 3 This is a structural block diagram illustrating a distribution decision device according to an exemplary embodiment;

[0075] Figure 4 This is a block diagram illustrating a server for distributing decisions according to an exemplary embodiment. Detailed Implementation

[0076] 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.

[0077] 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 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.

[0078] Figure 1 This is a flowchart illustrating the steps of a content distribution method according to an exemplary embodiment, such as... Figure 1 As shown, the content distribution method includes the following steps.

[0079] In step S11, the predicted click-through rate of the account to be recommended for the content to be recommended is obtained, and the predicted click-through rate is calibrated using the actual click-through rate data within the target time period to obtain the calibrated click-through rate; the account to be recommended is the account that sends the recommendation request at the current moment, and the target time period is the time period with a time difference of less than a first threshold from the current moment.

[0080] When users browse content on platforms such as short videos, live streams, WeChat, Weibo, and shopping sites, they refresh the current page to get the latest content. When the platform's server detects this refresh, it treats it as a recommendation request and designates the user as a potential recommendation account. Based on the attributes of the potential recommendation account, the server uses existing algorithms to select suitable content from the current available content. One potential recommendation account typically corresponds to multiple potential recommendation items. To achieve reasonable content output, this invention will subsequently evaluate the click-through rate and calculate distribution thresholds for each of these potential recommendation items to determine the distribution decision for each item.

[0081] In subsequent steps, the prediction of an individual's click-through rate and the calculation of the distribution threshold will be based on relevant metrics from multiple other accounts that sent recommendation requests at the current moment, as well as relevant metrics from the content to be recommended for those accounts. Therefore, this step can obtain multiple accounts that sent recommendation requests at the current moment and determine multiple pieces of content to be recommended for each account.

[0082] After identifying multiple pieces of content for an account to be recommended, the predicted click-through rate for each piece of content for that account can be determined based on existing algorithms or models.

[0083] To make the predicted click-through rate more accurate, this invention also calibrates the predicted click-through rate based on the actual click-through rate data of the account to be recommended within the target time period, thus obtaining the calibrated click-through rate of the account to be recommended for the content to be recommended.

[0084] To make click-through rate (CTR) calibration more accurate, actual CTR data from the most recent time period can be used as a reference for prediction. Therefore, the target time period is a period whose time difference from the current time is less than a first threshold. This first threshold can be set to a relatively short time difference, such as 5 hours or 10 hours. Actual CTR data refers to the actual click-through rate of accounts within the target user's specified time interval for platform content within that target time period.

[0085] It's easy to understand that the first threshold can be set according to different scenarios.

[0086] In step S12, the account group corresponding to the account to be recommended is determined; the account group is divided according to the similarity of the attribute features of the account to be recommended.

[0087] This attribute feature may include: the degree of willingness of the account to be recommended to watch platform content, the age and gender of the account to be recommended, the types of content the account to be recommended is interested in, the average viewing time of the account to be recommended, etc. Based on the similarity of the above tags, all the accounts to be recommended that are currently sending recommendation requests are grouped to obtain multiple account groups.

[0088] Specifically, in order to improve the efficiency of the algorithm, all accounts that have been online for a period of time (e.g., 1 day) and are to be recommended can be grouped in advance, and the grouping results can be used directly at the current moment.

[0089] To maintain a high degree of similarity among accounts in an account group, account attribute characteristics can be updated periodically, and the account groups can be updated accordingly. For example, account groups can be updated daily.

[0090] In step S13, the distribution threshold corresponding to the account group is determined based on the calibrated click-through rate, the expected click-through rate preset for the group of accounts to be recommended, and the value of the content to be recommended distributed to the accounts to be recommended.

[0091] The distribution threshold is used to make distribution decisions, specifically to determine whether to distribute the corresponding recommended content to the account to be recommended.

[0092] To obtain a more suitable distribution threshold, we can take into account the overall situation of groups with high similarity and determine a distribution threshold for each account group.

[0093] To ensure that the distribution threshold meets the actual online conditions, that the actual click-through rate always meets the overall expected click-through rate, and that the value brought by the content to be recommended to the recommended account is maximized, the distribution threshold can be determined based on the calibrated click-through rate of each account to be recommended in the account group, the pre-set expected click-through rate for the accounts to be recommended in the group, and the value brought by the content to be recommended to the recommended account.

[0094] The value brought by distributing content to the recommended account mainly refers to the cost savings and efficiency improvement, e-commerce value, user development value, and value-added service value that the platform gains after the content is distributed to the recommended account.

[0095] In step S14, based on the relationship between the distribution threshold and the calibration click-through rate, it is determined whether to distribute the content to be recommended corresponding to the calibration click-through rate to the account to be recommended.

[0096] Specifically, it can be set so that if the calibrated click-through rate is greater than or equal to the distribution threshold, the content to be recommended corresponding to the calibrated click-through rate is distributed to the corresponding account to be recommended; if the calibrated click-through rate is less than the distribution threshold, the content to be recommended corresponding to the calibrated click-through rate is not distributed to the corresponding account to be recommended.

[0097] The calibration click-through rate (CTR) is determined based on the predicted CTR and the actual CTR data of multiple accounts to be recommended within the recent time period, so the accuracy of the calibration CTR is high. The distribution threshold is determined based on the calibration CTR of each account to be recommended in the account group, the pre-set expected CTR for the group of accounts to be recommended, and the value of the content to be recommended distributed to the accounts to be recommended. Therefore, the distribution threshold can meet the actual online situation, ensuring that the actual CTR always meets the constraint of the overall expected CTR and that the content distributed to the accounts to be recommended brings greater value.

[0098] In summary, in the embodiments of this disclosure, the predicted click-through rate (CTR) of the account to be recommended for the content to be recommended is obtained, and the predicted CTR is calibrated using the actual CTR data within the target time period to obtain the calibrated CTR; the account to be recommended is the account that sends the recommendation request at the current moment, and the target time period is the time period with a time difference from the current moment that is less than a first threshold; the account group corresponding to the account to be recommended is determined; the account group is divided according to the similarity of the attribute features of the account to be recommended; the distribution threshold corresponding to the account group is determined based on the calibrated CTR, the expected CTR preset for the account group, and the value of the content to be recommended distributed to the account to be recommended; and the relationship between the distribution threshold and the calibrated CTR determines whether to distribute the content to be recommended corresponding to the calibrated CTR to the account to be recommended. In the above method, the actual click-through rate (CTR) data within the target time period is also used as a factor to determine the calibration CTR, so that the calibration CTR can adapt to real-time online metrics. The determination of the distribution threshold is also related to the calibration CTR, the expected CTR, and the value of the content to be recommended to the accounts to be recommended, so that the distribution threshold can also be dynamically adjusted, and the actual CTR of the accounts to be recommended can always meet the constraints of the overall expected CTR. Furthermore, different account groups correspond to different distribution thresholds, so that the accounts to be recommended can receive more suitable recommended content, thereby increasing the influence on page views and unique visitors, and improving the control of online traffic.

[0099] Figure 2 This is a flowchart illustrating the steps of another content distribution method according to an exemplary embodiment, such as... Figure 2 As shown, the content distribution method includes the following steps.

[0100] In step S21, the attribute features of the account to be recommended and the attribute features of the content to be recommended are input into a preset ranking model to predict the click-through rate of the account to be recommended for the content to be recommended, thus obtaining the predicted click-through rate.

[0101] The ranking model is a pre-trained model that uses historical data to predict click-through rates (CTR). The attribute features of the account to be recommended (u) and the attribute features of the content to be recommended (i) are input into the ranking model to predict the CTR of account u for content i, thus obtaining the predicted CTR (pctr). ui .

[0102] This operation is performed on each account to be recommended, ultimately yielding multiple predicted click-through rates (pctr). ui .

[0103] While ranking models can guarantee the order of recommendations based on predicted click-through rates (CTR), they can deviate from the actual CTR. In distribution decisions, real-time online metrics must be met, and relying solely on predicted CTR can introduce significant errors. Therefore, the predicted CTR can be calibrated to approximate the actual CTR.

[0104] In step S22, the predicted click-through rate interval to which the predicted click-through rate belongs is determined, thus obtaining the target interval.

[0105] Specifically, steps S22-S24 are steps for calibrating the predicted click-through rate.

[0106] To improve the accuracy of the calibration, the predicted click-through rate (CTR) for each account to be recommended obtained in the previous steps can be divided into multiple predicted CTR intervals according to the range of values, and the predicted CTR within each predicted CTR interval can be calibrated.

[0107] For example, if the predicted click-through rate (CTR) ranges from (0,1), the predicted CTR can be divided into M intervals. Depending on the actual account to be recommended, the spacing between each predicted CTR interval can be the same or different.

[0108] It's understandable that the rules for dividing the predicted click-through rate (CTR) intervals can be adjusted based on real-time online metrics. To improve algorithm efficiency, the CTR interval division rules can be predetermined and used directly here. Furthermore, the CTR interval division rules can be updated periodically, for example, every 10 hours, to better adapt to real-time online conditions.

[0109] Specifically, obtain the latest predicted click-through rate (CTR) interval division rules, then determine the predicted CTR interval to which the predicted CTR belongs, and use the predicted CTR to which it belongs as the target interval.

[0110] In step S23, the actual click-through rate data of the accounts to be recommended within the target interval during the target time period is obtained.

[0111] Obtain the actual click-through rate (CTR) data of the accounts to be recommended within the target time period. This target time period can be the most recent time period for which actual CTR data is available.

[0112] Within the target time period, if the total number of clicks corresponding to the predicted click-through rate value of the recommended account in the target interval s is click... t The total number of recommended content items distributed corresponding to the t-th predicted click-through rate value is shown. t Then, the actual click-through rate (CTR) data for the recommended account corresponding to the t-th predicted CTR value is: the ratio of total clicks to total distributions. t / show t .

[0113] In step S24, the predicted click-through rate is calibrated based on the actual click-through rate data to obtain the calibrated click-through rate of the account to be recommended for the content to be recommended.

[0114] The predicted click-through rate is calibrated based on actual click-through rate data, making the predicted click-through rate more accurate and more in line with real-time online metrics.

[0115] In steps S22-S24, the predicted click-through rate (CTR) is divided into multiple intervals according to its value range. The predicted CTR within each interval is then calibrated, ensuring that the actual CTR data used in the calibration process better matches the characteristics of the account to be recommended, thus further improving the accuracy of the calibrated CTR. Furthermore, the two parameters are fitted using data from the target time period. The fitted calibrated CTR approximately matches the CTR of the most recent time period, meaning that the calibrated CTR is obtained by referencing the CTR of the most recent time period. This makes the calibrated CTR more consistent with real-time online metrics and more accurate.

[0116] Optionally, step S24 includes the following steps S241-S244:

[0117] S241. Determine the linear relationship between the predicted click-through rate and the calibrated click-through rate, wherein the linear relationship includes two parameters.

[0118] Here, it is assumed that the predicted click-through rate and the calibrated click-through rate within each interval are linearly related, that is, for the target interval s, there is a parameter a. s and b s This ensures that the calibrated click-through rate (cctr) and the predicted click-through rate (pctr) satisfy equation (1):

[0119] cctr=a s*pctr+b s (1)

[0120] Where pctr is the predicted click-through rate of the accounts to be recommended within the target interval s, cctr is the calibrated click-through rate of the accounts to be recommended within the target interval s, and a s and b s These are two fitting parameters.

[0121] S242. Fit the two parameters using the objective function to obtain the values ​​of the two parameters; the objective function is: the squared error function corresponding to the calibrated click-through rate and the actual click-through rate.

[0122] a s and b s A target function can be used for fitting. Specifically, the target function is a function composed of the sum of squares of the differences between the calibration click-through rate and the actual click-through rate, as shown in equation (2):

[0123]

[0124] Where E is the name of the squared error function, n s It is the number of values ​​in the pctr interval within the target interval s, U S It is the set of value intervals of the target interval s, cctr t The calibrated click-through rate corresponding to the t-th pctr value within the target interval s. This represents the actual click-through rate corresponding to the t-th pctr value within the target interval s.

[0125] When the above squared error function reaches its minimum value, the corresponding value of a is... s and b s The values ​​of the two parameters are obtained as a and b.

[0126] S243. Substitute the values ​​of the two parameters into the linear relationship to obtain the calibration relationship corresponding to the target interval.

[0127] Substituting the values ​​of a and b obtained from step S252 into the linear relationship, we obtain the calibration relationship (3) corresponding to the target interval:

[0128] cctr=a*pctr+b (3)

[0129] S244. Using the calibration formula, determine the calibration click-through rate of each account to be recommended for the content to be recommended within the target interval.

[0130] Specifically, by substituting the predicted click-through rate (pctr) value into the calibration formula, the calibrated click-through rate (cctr) corresponding to that pctr can be obtained.

[0131] In steps S241-S244, a s and b s By fitting data from the most recent time period, the resulting CTR approximately matches the click-through rate (CTR) of that period. This means the CTR is derived with reference to the CTR of the most recent time period, making it more consistent with real-time online metrics and more accurate. Furthermore, the predicted CTR is divided into multiple intervals according to its value range, and the CTR within each interval is calibrated. This ensures that the actual CTR data referenced during the calibration process better matches the characteristics of the recommended account corresponding to the predicted CTR, further improving the accuracy of the calibrated CTR.

[0132] In step S25, the account group corresponding to each of the accounts to be recommended is determined; the account group is divided according to the similarity of the attribute features of the accounts to be recommended.

[0133] In the embodiments of this disclosure, step S25 can refer to step S12, and will not be repeated here.

[0134] In step S26, a decision function is determined that includes the calibrated click-through rate and the distribution threshold, the value of which is determined by the relationship between the calibrated click-through rate and the distribution threshold.

[0135] The decision function, with the calibration click-through rate (CTR) as the independent variable, determines whether to distribute the content to the recommended account based on the relationship between the CTR and the distribution threshold. This function is piecewise: it takes the value 1 when the CTR is greater than or equal to the distribution threshold, and 0 when the CTR is less than the distribution threshold.

[0136] The decision function can be expressed as shown in equation (4):

[0137]

[0138] Where π is the name of the decision function, d = 1 indicates that the live stream will be distributed to the account to be recommended, u represents the account to be recommended, p represents pctr, and α is the distribution threshold. A value of 1 on the right side of the equation indicates that the event d = 1 occurs under the current conditions of u and p, while a value of 0 on the right side indicates that the event d = 1 does not occur under the current conditions of u and p.

[0139] In step S27, the calibration click-through rate function for the account group is determined based on the calibration click-through rate and the decision function.

[0140] The calibration click-through rate (CTR) is the ratio of predicted total clicks to predicted total distributions. Predicted total clicks and predicted total distributions are determined by the CTR and distribution threshold for each account to be recommended. Therefore, the CTR function for an account group can be expressed as shown in Equation 5:

[0141]

[0142] Among them, cctr ui This represents the calibration click-through rate of the account u to be recommended for the content i to be recommended. π ui This indicates whether to distribute the content to be recommended i to the account to be recommended u. If it is distributed, π is 1; if it is not distributed, π is 0. The decision function in Formula 3 determines whether to distribute the content to be recommended i to the account to be recommended u.

[0143] f(x) represents the ratio of the predicted total clicks of all accounts to be recommended in the same account group to the predicted total distribution of all accounts to be recommended. In other words, f(x) is the calibration click-through rate of the account group.

[0144] In step S28, a candidate threshold is determined based on the calibration click-through rate function and the expected click-through rate value preset for the accounts to be recommended in the account group.

[0145] The candidate threshold should satisfy the condition that the calibrated click-through rate is greater than or equal to the expected click-through rate value, that is, the candidate threshold should make equation (6) true.

[0146]

[0147] Where c is the expected click-through rate.

[0148] In steps S26-S28, candidate thresholds are determined based on the calibration click-through rate function and the pre-set expected click-through rate values ​​for the accounts to be recommended in the account group. Then, a distribution threshold for the account group is determined from the candidate thresholds based on the value of the content to be recommended distributed to the accounts. Here, the candidate thresholds are thresholds that satisfy the expected click-through rate values, while the distribution threshold is the threshold selected from the candidate thresholds that brings greater value. This ensures that the content to be recommended distributed to the accounts, determined according to the distribution threshold, both satisfies the expected click-through rate and brings maximum value, thus improving the effectiveness of content distribution.

[0149] Optionally, step S28 includes the following steps S281-S282:

[0150] S281. Set the starting value of the distribution threshold;

[0151] S282. Starting with the initial value, a binary search is used to find a candidate threshold such that the value of the average click-through rate function is greater than the distribution threshold of the expected click-through rate.

[0152] In steps S281-S282, a binary search method can be used to search for candidate thresholds in equation (6).

[0153] The binary search approach works as follows: First, take the midpoint of the distribution threshold range (0,1) as the starting value. Here, we set the starting value to 0.5. Substituting 0.5 into formula 3, we can obtain the values ​​of each cctr in the account group. ui The corresponding π ui .

[0154] Then cctr ui π ui Substituting into equation (6), we can obtain the value on the left side of equation (6). If the value on the left side satisfies equation (6), then this initial value is used as a candidate threshold; if the initial value does not satisfy equation (6), since the result on the left side can be considered positively correlated with α, then it means that the value in the left interval will not satisfy equation (6), and we can continue searching in the right interval. Take the middle element 0.75 of the right interval (0.5,1) to determine the cctr corresponding to 0.75. ui π ui Can equation (6) be made true? If equation (6) can be made true, then 0.75 is a candidate threshold. If equation (6) cannot be made true, then the previous steps are repeated. The search ends when the interval range of a certain step is 0. After the search ends, multiple candidate thresholds can be obtained.

[0155] Using a binary search method to search for candidate thresholds results in fewer search attempts, higher efficiency, and the ability to quickly obtain accurate candidate thresholds.

[0156] In step S29, the distribution threshold for the account group is determined from the candidate thresholds based on the value of the content to be recommended distributed to the account to be recommended.

[0157] After determining multiple candidate thresholds, each candidate threshold is substituted into the decision function to determine whether to distribute the content to be recommended i to the account to be recommended u. Then, the value brought by distributing all the content to be recommended in the account group to the account to be recommended is calculated, thereby determining the cumulative value of the entire account group.

[0158] In this way, the cumulative value corresponding to each candidate threshold is obtained, and the candidate threshold with the largest cumulative value is selected as the distribution threshold for the account group.

[0159] Optionally, step S29 includes the following steps S291-S293:

[0160] S291. Determine a cumulative value function with the candidate threshold as a variable, wherein the cumulative value function represents the sum of the value brought about by distributing the content to be recommended to the corresponding account in the account group.

[0161] First, determine the value function v. ui v ui This represents the value derived from distributing the content to be recommended, i, to the account to be recommended, u. Then, the cumulative value function corresponding to the account group is determined, which can be expressed as equation (7):

[0162]

[0163] S292. Traverse the candidate thresholds to determine the value of the cumulative value function corresponding to the candidate thresholds;

[0164] S293. The candidate threshold that maximizes the value of the cumulative value function is determined as the distribution threshold for the account group.

[0165] In steps S292-S293, the candidate thresholds are traversed to find the candidate threshold that maximizes the cumulative value function as the distribution threshold. Specifically, the distribution threshold can be selected from the candidate thresholds according to the following equation (8):

[0166]

[0167] in, This indicates the distribution threshold.

[0168] In steps S291-S293, the candidate threshold with the largest cumulative value is selected as the distribution threshold for the account group, so that the distribution threshold can bring a larger cumulative value.

[0169] In step S30, based on the relationship between the distribution threshold and the calibration click-through rate, it is determined whether to distribute the content to be recommended corresponding to the calibration click-through rate to the account to be recommended.

[0170] In the embodiments of this disclosure, once the distribution threshold corresponding to each account group is determined, the distribution decision for the content to be recommended can be determined. Specifically, when the calibrated click-through rate of the account to be recommended is greater than or equal to the distribution threshold of its account group, the content to be recommended is distributed to the account to be recommended; otherwise, the content to be recommended is not distributed to the account to be recommended.

[0171] In conclusion, Figure 2 The content distribution method in the middle, in addition to having Figure 1 In addition to the beneficial effects of the content distribution method in [the context], the following beneficial effects also apply:

[0172] The predicted click-through rate (CTR) is divided into multiple intervals according to its value range, and the predicted CTR within each interval is calibrated. This makes the actual CTR data referenced in the calibration process more consistent with the characteristics of the recommended account corresponding to the predicted CTR, thereby further improving the accuracy of the calibrated CTR.

[0173] Furthermore, the two parameters are fitted with data from the most recent time period. The fitted calibrated click-through rate approximately matches the click-through rate from the most recent time period. This means that the calibrated click-through rate is obtained with reference to the click-through rate from the most recent time period, making the calibrated click-through rate more consistent with real-time online metrics and more accurate.

[0174] Furthermore, the binary search method is used to search for candidate thresholds, which requires fewer searches and is more efficient. It can quickly obtain accurate candidate thresholds and select the candidate threshold with the highest cumulative value as the distribution threshold for the account group, so that the distribution threshold can bring greater cumulative value.

[0175] Figure 3 This is a structural block diagram of a distribution decision device according to an exemplary embodiment.

[0176] like Figure 3 As shown, the distribution decision device 40 includes:

[0177] The content determination module 41 is configured to retrieve multiple accounts that have sent recommendation requests at the current time, and determine multiple pieces of content to be recommended for each account; the multiple pieces of content to be recommended are selected from the existing content at the current time.

[0178] Click-through rate prediction module 42 is configured to perform a prediction of the click-through rate of each account to be recommended for the content to be recommended based on the attribute characteristics of the account to be recommended, the attribute characteristics of the content to be recommended, and the actual click-through rate data of the multiple accounts to be recommended within a target time period, thereby obtaining multiple calibrated click-through rates; the target time period is a time period in which the time difference with the current time is less than a first threshold.

[0179] The account group determination module 43 is configured to determine the account group corresponding to each of the accounts to be recommended; the account groups are divided according to the similarity of the attribute features of the accounts to be recommended.

[0180] The threshold determination module 44 is configured to determine the distribution threshold corresponding to each account group based on the calibrated click-through rate, the expected click-through rate preset for the accounts to be recommended in the group, and the value of the content to be recommended distributed to the accounts to be recommended.

[0181] The distribution determination module 45 is configured to determine, based on the relationship between the distribution threshold and the calibration click-through rate, whether to distribute the content to be recommended corresponding to the calibration click-through rate to the account to be recommended.

[0182] 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.

[0183] Figure 4 This is a block diagram illustrating a server for distributing decisions according to an exemplary embodiment, the internal structure of which can be shown as follows: Figure 4 As shown, the server includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The server's memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The server's network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a content distribution method.

[0184] Those skilled in the art will understand that Figure 4 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 server to which the present disclosure is applied. A specific server may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0185] In an exemplary embodiment, a server 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 content distribution method as described in the embodiments of this disclosure.

[0186] In an exemplary embodiment, a computer-readable storage medium is also provided, which, when executed by a server's processor, enables the server to perform the content distribution method of this disclosure. The computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc.

[0187] In an exemplary embodiment, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform the content distribution method of the present disclosure embodiments.

[0188] 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.

[0189] 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.

[0190] 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. A content distribution method, characterized in that, include: The predicted click-through rate (CTR) of the account to be recommended for the content to be recommended is obtained, and the predicted CTR is calibrated using the actual CTR data within the target time period to obtain the calibrated CTR; the account to be recommended is the account that sends the recommendation request at the current moment, and the target time period is the time period in which the time difference between the current moment and the current moment is less than a first threshold. Determine the account group corresponding to the account to be recommended; The account groups are divided based on the similarity of the attribute features of the accounts to be recommended; Based on the calibrated click-through rate, the pre-set expected click-through rate for the account group, and the value of the content to be recommended distributed to the account to be recommended, the distribution threshold corresponding to the account group is determined. Based on the relationship between the distribution threshold and the calibration click-through rate, determine whether to distribute the content to be recommended corresponding to the calibration click-through rate to the account to be recommended; The step of determining the distribution threshold corresponding to the account group based on the calibrated click-through rate, the pre-set expected click-through rate for the account group, and the value of distributing the content to be recommended to the account to be recommended includes: A decision function is determined that includes the calibrated click-through rate and the distribution threshold, the value of which is determined by the relationship between the calibrated click-through rate and the distribution threshold; Based on the calibrated click-through rate and the decision function, determine the calibrated click-through rate function for the account group; Candidate thresholds are determined based on the calibration click-through rate function and the pre-set expected click-through rate value for the account group; Based on the value of the content to be recommended distributed to the account to be recommended, the distribution threshold of the account group is determined from the candidate thresholds.

2. The method according to claim 1, characterized in that, The step of calibrating the predicted click-through rate using actual click-through rate data within the target time period to obtain the calibrated click-through rate includes: Determine the predicted click-through rate interval to which the predicted click-through rate belongs, and obtain the target interval; Obtain the actual click-through rate data of the accounts to be recommended within the target interval during the target time period; The predicted click-through rate is calibrated based on the actual click-through rate data to obtain the calibrated click-through rate of the account to be recommended for the content to be recommended.

3. The method according to claim 2, characterized in that, The actual click-through rate (CTR) data includes the actual CTR corresponding to each predicted CTR value within the target interval. The step of calibrating the predicted CTR based on the actual CTR data includes: Determine the linear relationship between the predicted click-through rate and the calibrated click-through rate, wherein the linear relationship includes two parameters; The two parameters are fitted using an objective function to obtain their values; the objective function is the squared error function corresponding to the calibrated click-through rate and the actual click-through rate. Substituting the values ​​of the two parameters into the linear relationship, we obtain the calibration relationship corresponding to the target interval; Using the calibration formula, the calibration click-through rate of each account to be recommended for the content to be recommended within the target interval is determined.

4. The method according to claim 1, characterized in that, The process of obtaining the predicted click-through rate of the account to be recommended for the content to be recommended includes: The attribute features of the account to be recommended and the attribute features of the content to be recommended are input into a preset ranking model to predict the click-through rate of the account to be recommended for the content to be recommended, thus obtaining the predicted click-through rate.

5. The method according to claim 1, characterized in that, The step of determining the candidate threshold based on the calibration click-through rate function and the pre-set expected click-through rate value for the account group includes: Set an initial value; the initial value is a value between 0 and 1. Starting with the initial value, a binary search is used to find a candidate threshold such that the value of the average click-through rate function is greater than the expected click-through rate distribution threshold.

6. The method according to claim 1, characterized in that, The step of determining the distribution threshold for the account group from the candidate thresholds based on the value of the content to be recommended to the account to be recommended includes: A cumulative value function is determined with the candidate threshold as a variable. The cumulative value function represents the sum of the value brought about by distributing the content to be recommended to the corresponding account in the account group. Iterate through the candidate thresholds to determine the value of the cumulative value function corresponding to each candidate threshold; The candidate threshold that maximizes the value of the cumulative value function is determined as the distribution threshold for the account group.

7. A content distribution device, characterized in that, include: The calibration module is configured to obtain the predicted click-through rate of the account to be recommended for the content to be recommended, and calibrate the predicted click-through rate using the actual click-through rate data within the target time period to obtain the calibrated click-through rate; the account to be recommended is the account that sent the recommendation request at the current moment, and the target time period is the time period with a time difference of less than a first threshold from the current moment; The account group determination module is configured to determine the account group corresponding to the account to be recommended; The account groups are divided based on the similarity of the attribute features of the accounts to be recommended; The distribution threshold determination module is configured to determine the distribution threshold corresponding to the account group based on the calibrated click-through rate, the expected click-through rate preset for the account group, and the value of distributing the content to be recommended to the account to be recommended. The distribution determination module is configured to determine, based on the relationship between the distribution threshold and the calibration click-through rate, whether to distribute the content to be recommended corresponding to the calibration click-through rate to the account to be recommended; The distribution threshold determination module is specifically configured to execute: A decision function is determined that includes the calibrated click-through rate and the distribution threshold, the value of which is determined by the relationship between the calibrated click-through rate and the distribution threshold; Based on the calibrated click-through rate and the decision function, determine the calibrated click-through rate function for the account group; Candidate thresholds are determined based on the calibration click-through rate function and the pre-set expected click-through rate value for the account group; Based on the value of the content to be recommended distributed to the account to be recommended, the distribution threshold of the account group is determined from the candidate thresholds.

8. The apparatus according to claim 7, characterized in that, The calibration module is specifically configured to perform: Determine the predicted click-through rate interval to which the predicted click-through rate belongs, and obtain the target interval; Obtain the actual click-through rate data of the accounts to be recommended within the target interval during the target time period; The predicted click-through rate is calibrated based on the actual click-through rate data to obtain the calibrated click-through rate of the account to be recommended for the content to be recommended.

9. The apparatus according to claim 8, characterized in that, The calibration module is also configured to perform: Determine the linear relationship between the predicted click-through rate and the calibrated click-through rate, wherein the linear relationship includes two parameters; The two parameters are fitted using an objective function to obtain their values; the objective function is the squared error function corresponding to the calibrated click-through rate and the actual click-through rate. Substituting the values ​​of the two parameters into the linear relationship, we obtain the calibration relationship corresponding to the target interval; Using the calibration formula, the calibration click-through rate of each account to be recommended for the content to be recommended within the target interval is determined.

10. The apparatus according to claim 7, characterized in that, The calibration module is also configured to perform: The attribute features of the account to be recommended and the attribute features of the content to be recommended are input into a preset ranking model to predict the click-through rate of the account to be recommended for the content to be recommended, thus obtaining the predicted click-through rate.

11. The apparatus according to claim 7, characterized in that, The distribution threshold determination module is also configured to perform: Set an initial value; the initial value is a value between 0 and 1. Starting with the initial value, a binary search is used to find a candidate threshold such that the value of the average click-through rate function is greater than the expected click-through rate distribution threshold.

12. The apparatus according to claim 7, characterized in that, The distribution threshold determination module is also configured to perform: A cumulative value function is determined with the candidate threshold as a variable. The cumulative value function represents the sum of the value brought about by distributing the content to be recommended to the corresponding account in the account group. Iterate through the candidate thresholds to determine the value of the cumulative value function corresponding to each candidate threshold; The candidate threshold that maximizes the value of the cumulative value function is determined as the distribution threshold for the account group.

13. A server, 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 content distribution method as described in any one of claims 1 to 6.

14. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the server, the server is able to perform the content distribution method as described in any one of claims 1 to 6.

15. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the content distribution method according to any one of claims 1 to 6.

Images