Exposure method and device

Abstract

The present disclosure relates to an exposure method and device. The method includes: obtaining the Beta distribution function of each user group for each content category; obtaining the corresponding target Beta distribution function according to the content category corresponding to the content to be exposed and the group to be exposed; obtaining the estimated click-through rate of the content to be exposed according to the probability density function corresponding to the target Beta distribution function; obtaining the threshold value of the content to be exposed based on time decay according to the repeated exposure time interval corresponding to the content to be exposed; when the threshold value is greater than the estimated click-through rate, exposing the content to be exposed; when the threshold value is less than the estimated click-through rate, filtering the content to be exposed. Because it is based on feedback from factual data and the consideration of time decay, as the number of repeated exposures increases, the confidence of the exposure filtering judgment continues to increase, thereby improving the accuracy of exposure.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, and in particular to an exposure method and device. Background Art

[0002] Currently, the industry's exposure filtering solutions are usually based on rules, such as ensuring that previously exposed content does not reappear within a certain time period (a week or a month). Based on this, content is categorized, such as news content with a longer time period and other content with a shorter time period.

[0003] However, in this solution, the selection of the cycle lacks feedback from actual data, the calculation of the cycle is not accurate enough, and more suitable content may not receive appropriate re-exposure.

[0004] Therefore, how to improve the accuracy of exposure is an urgent problem to be solved. Summary of the Invention

[0005] To overcome the problems existing in the related art, the present disclosure provides an exposure method and device. The technical solution is as follows:

[0006] According to a first aspect of an embodiment of the present disclosure, there is provided an exposure method, comprising:

[0007] Obtain a Beta distribution function for each user group for each content category, where α in the Beta distribution function represents a positive sample, which is obtained based on the number of clicks on the first repeatedly exposed content and the decay value of the first repeatedly exposed content over time. β in the Beta distribution function represents a negative sample, which is obtained based on the number of unclicks on the second repeatedly exposed content and the decay value of the second repeatedly exposed content over time.

[0008] Obtain the content and groups to be exposed;

[0009] Obtaining a corresponding target Beta distribution function according to the content category corresponding to the content to be exposed and the group to be exposed;

[0010] Obtaining an estimated click-through rate of the content to be exposed according to a probability density function corresponding to the target Beta distribution function, where the estimated click-through rate is a random number generated by the probability density function corresponding to the target Beta distribution function;

[0011] Obtaining a time-attenuation threshold of the content to be exposed according to a repeated exposure time interval corresponding to the content to be exposed;

[0012] When the threshold is greater than the estimated click rate, exposing the content to be exposed;

[0013] When the threshold is less than the estimated click rate, the content to be exposed is filtered.

[0014] The present disclosure provides an exposure method, including: obtaining a Beta distribution function for each content category for each user group, where α in the Beta distribution function represents a positive sample, which is obtained based on the number of times a first repeatedly exposed content is clicked and the decay value of the first repeatedly exposed content over time, and β in the Beta distribution function represents a negative sample, which is obtained based on the number of times a second repeatedly exposed content is not clicked and the decay value of the second repeatedly exposed content over time; obtaining content to be exposed and a group to be exposed; obtaining a corresponding target Beta distribution function based on the content category and the group to be exposed corresponding to the content to be exposed; obtaining an estimated click-through rate of the content to be exposed based on a probability density function corresponding to the target Beta distribution function, where the estimated click-through rate is a random number generated by the probability density function corresponding to the target Beta distribution function; obtaining a threshold value of the content to be exposed based on time decay according to the repeated exposure time interval corresponding to the content to be exposed; exposing the content to be exposed when the threshold value is greater than the estimated click-through rate; and filtering the content to be exposed when the threshold value is less than the estimated click-through rate. Among them, based on the Beta distribution and the time decay of the sample, a probability distribution function is given to estimate the click-through rate. By comparing the estimated click-through rate with the content's time-decay threshold, it is determined whether the content should be filtered. Because it is based on feedback from factual data and takes time decay into consideration, as the number of repeated exposures increases, the confidence level of the exposure filtering judgment continues to increase, thereby improving the accuracy of exposure.

[0015] In one embodiment, obtaining α in the Beta distribution function of each user group for each content category includes:

[0016] The α is obtained according to the following formula:

[0017]

[0018] Among them, the Represents the accumulation of attenuation coefficients of positive samples that decay over time. The days represents the time between the exposure time of the content category of the current positive sample and the last exposure. The a is the first hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample, and is obtained according to the decay speed of the user's memory of the content of the positive sample in the business scenario. The b is the second hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample on the same day, and is obtained according to the degree of forgetting of the content of the positive sample on the same day by the user in the business scenario. The α0 is the third hyperparameter, which represents the default value of the positive sample when there is no data, and is obtained according to the average click-through rate of online content.

[0019] In one embodiment, obtaining β in the Beta distribution function of each user group for each content category includes:

[0020] The β is obtained according to the following formula:

[0021]

[0022] Among them, the Represents the accumulation of attenuation coefficients of negative samples that decay over time. The days′ represents the time between the exposure time of the content category of the current negative sample and the last exposure time. The a′ is the fourth hyperparameter, which represents the user's forgetting speed of the content category of the current negative sample, and is obtained according to the user's memory decay speed of the content of the negative sample in the business scenario. The b′ is the fifth hyperparameter, which represents the user's forgetting speed of the content category of the current negative sample on the same day, and is obtained according to the user's forgetting degree of the content of the negative sample on the same day in the business scenario. The β0 is the sixth hyperparameter, which represents the default value of the negative sample when there is no data, and is obtained according to the average click-through rate of online content.

[0023] In one embodiment, obtaining a threshold value of the content to be exposed based on time decay according to the repeated exposure time interval corresponding to the content to be exposed includes:

[0024] The threshold value of the content to be exposed based on time decay is obtained according to the following formula:

[0025]

[0026] Among them, M is the threshold of the content to be exposed based on time decay, and days″ represents the length of time between the exposure time and the last exposure of the content category to be exposed. c is the sixth hyperparameter, which is obtained based on the speed of memory decay of users for content in business scenarios. d is the seventh hyperparameter, which is obtained based on the degree of forgetfulness of the content of the day by users in business scenarios.

[0027] According to a second aspect of an embodiment of the present disclosure, there is provided an exposure device, comprising:

[0028] A first acquisition module is configured to acquire a Beta distribution function for each user group for each content category, where α in the Beta distribution function represents a positive sample and is obtained based on the number of clicks on the first repeatedly exposed content and the decay value of the first repeatedly exposed content over time; and β in the Beta distribution function represents a negative sample and is obtained based on the number of unclicks on the second repeatedly exposed content and the decay value of the second repeatedly exposed content over time;

[0029] The second acquisition module is used to acquire the content to be exposed and the group to be exposed;

[0030] A third acquisition module is used to acquire a corresponding target Beta distribution function according to the content category corresponding to the content to be exposed and the group to be exposed;

[0031] A fourth acquisition module is configured to acquire an estimated click-through rate of the content to be exposed according to a probability density function corresponding to the target Beta distribution function, where the estimated click-through rate is a random number generated by the probability density function corresponding to the target Beta distribution function;

[0032] A fifth acquisition module, configured to acquire a threshold value of the content to be exposed based on time attenuation according to a repeated exposure time interval corresponding to the content to be exposed;

[0033] An exposure module, configured to expose the content to be exposed when the threshold is greater than the estimated click rate;

[0034] The filtering module is used to filter the content to be exposed when the threshold is less than the estimated click rate.

[0035] In one embodiment, the first acquisition module includes: a first acquisition submodule,

[0036] The first acquisition submodule is configured to obtain α according to the following formula:

[0037]

[0038] Among them, the Represents the accumulation of attenuation coefficients of positive samples that decay over time. The days represents the time between the exposure time of the content category of the current positive sample and the last exposure. The a is the first hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample, and is obtained according to the decay speed of the user's memory of the content of the positive sample in the business scenario. The b is the second hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample on the same day, and is obtained according to the degree of forgetting of the content of the positive sample on the same day by the user in the business scenario. The α0 is the third hyperparameter, which represents the default value of the positive sample when there is no data, and is obtained according to the average click-through rate of online content.

[0039] In one embodiment, the first acquisition module includes: a second acquisition submodule,

[0040] The second acquisition submodule is configured to obtain β according to the following formula:

[0041]

[0042] Among them, the Represents the accumulation of attenuation coefficients of negative samples that decay over time. The days′ represents the time between the exposure time of the content category of the current negative sample and the last exposure time. The a′ is the fourth hyperparameter, which represents the user's forgetting speed of the content category of the current negative sample, and is obtained according to the user's memory decay speed of the content of the negative sample in the business scenario. The b′ is the fifth hyperparameter, which represents the user's forgetting speed of the content category of the current negative sample on the same day, and is obtained according to the user's forgetting degree of the content of the negative sample on the same day in the business scenario. The β0 is the sixth hyperparameter, which represents the default value of the negative sample when there is no data, and is obtained according to the average click-through rate of online content.

[0043] In one embodiment, the fifth acquisition module includes: a third acquisition submodule,

[0044] The third acquisition submodule is configured to acquire a threshold value of the content to be exposed based on time decay according to the following formula:

[0045]

[0046] Among them, M is the threshold of the content to be exposed based on time decay, and days″ represents the length of time between the exposure time and the last exposure of the content category to be exposed. c is the sixth hyperparameter, which is obtained based on the speed of memory decay of users for content in business scenarios. d is the seventh hyperparameter, which is obtained based on the degree of forgetfulness of the content of the day by users in business scenarios.

[0047] According to a third aspect of the embodiments of the present disclosure, there is provided an exposure device, comprising:

[0048] processor;

[0049] a memory for storing processor-executable instructions;

[0050] Wherein, the processor is configured to:

[0051] Obtain a Beta distribution function for each user group for each content category, where α in the Beta distribution function represents a positive sample, which is obtained based on the number of clicks on the first repeatedly exposed content and the decay value of the first repeatedly exposed content over time. β in the Beta distribution function represents a negative sample, which is obtained based on the number of unclicks on the second repeatedly exposed content and the decay value of the second repeatedly exposed content over time.

[0052] Obtain the content and groups to be exposed;

[0053] Obtaining a corresponding target Beta distribution function according to the content category corresponding to the content to be exposed and the group to be exposed;

[0054] Obtaining an estimated click-through rate of the content to be exposed according to a probability density function corresponding to the target Beta distribution function, where the estimated click-through rate is a random number generated by the probability density function corresponding to the target Beta distribution function;

[0055] Obtaining a time-attenuation threshold of the content to be exposed according to a repeated exposure time interval corresponding to the content to be exposed;

[0056] When the threshold is greater than the estimated click rate, exposing the content to be exposed;

[0057] When the threshold is less than the estimated click rate, the content to be exposed is filtered.

[0058] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, on which computer instructions are stored, and when the instructions are executed by a processor, the steps of any one of the methods described in the first aspect are implemented.

[0059] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0061] Figure 1 is a flow chart showing an exposure method according to an exemplary embodiment.

[0062] Figure 2 is a schematic diagram of a probability density function of Beta distribution according to an exemplary embodiment.

[0063] Figure 3 It is a schematic diagram of a parameter acquisition function image according to an exemplary embodiment.

[0064] Figure 4 is a block diagram of an exposure device according to an exemplary embodiment.

[0065] Figure 5 is a block diagram of an exposure apparatus 90 according to an exemplary embodiment. DETAILED DESCRIPTION

[0066] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.

[0067] Definitions of terms in this disclosure:

[0068] Community content: Content in short. Examples include posts on Xiaohongshu, non-e-commerce videos on Douyin, and content on the Weidian Discovery page. Content can be categorized into categories, such as movies, celebrity-chasing, and food.

[0069] Community scenarios: such as Xiaohongshu and Weidian discovery page.

[0070] Exposure filtering: This means that the exposed content that has been seen by users needs to be blocked or filtered during the user's subsequent browsing process.

[0071] Scene description:

[0072] The scenario disclosed in this disclosure is mainly based on the community scenario, and determines whether the content that the user has exposed is appropriate to be exposed again.

[0073] Why limit re-exposure? Because users are prone to fatigue or boredom when they see the same content in a short period of time, and they may feel that there is too little information on the platform (the most typical example is news apps), which is not conducive to the development of the platform.

[0074] Why not permanently filter out previously exposed content? For example: (1) In many niche scenarios (such as small channels), there may be very little content that is suitable for display. In this case, secondary exposure is more appropriate than returning to zero content; or secondary exposure of high-quality content is more appropriate than initial exposure of low-quality content. (2) After a certain time interval, users are willing to see the same content again, such as when there are more new comments under the content. (3) Over time, users may forget the content that was previously exposed at a certain level.

[0075] Traditional solution:

[0076] Currently, the industry's exposure filtering solutions are typically rule-based, such as ensuring that previously exposed content does not reappear within a certain time period (a week or a month). Based on this, content is categorized, such as news content with a longer time period and other content with a shorter time period. This approach has the following problems:

[0077] 1. Weak user personalization: Different users have different sensitivities to duplicate content, and this solution does not fully consider the experience of different users.

[0078] 2. The selection of the cycle lacks feedback from actual data, the calculation of the cycle is not accurate enough, and more suitable content may not receive appropriate re-exposure.

[0079] In order to solve the above technical problems, the present disclosure proposes an exposure method.

[0080] Figure 1 FIG. 1 is a flow chart showing an exposure method according to an exemplary embodiment. Figure 1 As shown, the method includes the following steps S101-S103:

[0081] In step S101, the Beta distribution function of each user group for each content category is obtained. The α in the Beta distribution function represents a positive sample, which is obtained based on the number of clicks on the first repeatedly exposed content and the decay value of the first repeatedly exposed content over time. The β in the Beta distribution function represents a negative sample, which is obtained based on the number of unclicks on the second repeatedly exposed content and the decay value of the second repeatedly exposed content over time.

[0082] In the present disclosure, users are first clustered or classified to obtain user groups, and each user group is a collection of a type of users.

[0083] Specifically, this disclosure calculates a user's sensitivity to repeated exposure to different types of content based on real user feedback data (such as click-through rate). However, calculating this data requires a large number of repeated exposures to produce accurate and reliable results. Exposing a single user to a large amount of repeated content is clearly not suitable.

[0084] Therefore, the present disclosure first classifies / clusters users, that is, calculates feedback data of a certain group on repeated exposure to content of different types.

[0085] Although the accuracy of user personalization will be reduced to a certain extent, the number of times each user is exposed to repeated content is much less. In other words, this method strikes a balance between personalization and reducing repeated exposure.

[0086] The specific clustering / classification method can be selected according to the business scenario, for example, males aged 18 to 20 are classified as a user group.

[0087] In the present disclosure, different Beta distribution functions are maintained for different categories of content in different user groups.

[0088] The Beta distribution is a probability distribution that is often used to describe the probability distribution of a random variable between 0 and 1.

[0089] The shape of the Beta distribution is controlled by two parameters, α and β, where α represents the number of successes and β represents the number of failures.

[0090] For example, if we flip a coin 100 times and get heads 10 times and tails 90 times, we can use the Beta distribution with α = 10 and β = 90 to describe the probability distribution of heads.

[0091] Figure 2 is a schematic diagram of a probability density function of a Beta distribution according to an exemplary embodiment. Figure 2 The horizontal axis represents the value of our predicted target (the probability P of the coin landing on the top side), and the vertical axis represents the probability corresponding to this value.

[0092] Figure 2 In the middle blue area, when α = 2 and β = 5, the maximum value of P is 2 / 7. However, due to the small number of experiments, the probability of the value of P corresponding to 2 / 7 (the height of the vertical axis) is not large (the confidence level is low), and the distribution of P values is relatively scattered (too few experiments means the distribution probability of P is not concentrated).

[0093] Figure 2 In the middle yellow area, when α = 10 and β = 90, the maximum value of P is 0.1. Since the number of experiments reaches 100, the probability of a P value of 0.1 is very high (the confidence level is high), and the distribution of P values is concentrated near 0.1 (that is, the probability of P values appearing near 0.1 is high).

[0094] In this disclosure, a Beta distribution function, ranging from 0 to 1, is used to represent the estimated click-through rate (or preference value) of users for repeatedly exposed content. The Beta distribution has two parameters, α and β. α represents the number of clicks (on repeatedly exposed content), indicating the number of successful clicks (positive samples); β represents the number of unclicks (negative samples).

[0095] It is worth noting that the present disclosure also considers the time interval between repeated exposures of different contents and the last exposure, that is, α and β are both the sum of the weighted values of the samples that decay over time.

[0096] For example, let's compare two user groups. User group A had 10 impressions and 5 clicks on the second day, while user group B had 10 impressions and 5 clicks on the 20th day. Although the click-through rate is the same at 50%, we believe that user group A is more willing to be re-exposed because user group B's memory of the initial impression has faded. This means that user group A has a higher estimated click-through rate or preference score for the content they were re-exposed to.

[0097] Based on this example, the present disclosure uses a time decay function (the first time decay function and the second time decay function in the following embodiments) when calculating the Beta distribution function.

[0098] In one embodiment, α in the Beta distribution function is obtained according to the following formula:

[0099]

[0100] in, It represents the accumulation of decay coefficients of positive samples decaying over time, which is also the first time decay function. Days represents the time between the exposure time of the content category of the current positive sample and the last exposure (for example, the duration can be: days, hours, minutes or seconds). A is the first hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample, and is obtained according to the memory decay speed of the user's positive sample content in the business scenario. B is the second hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample on the same day, and is obtained according to the degree of forgetting of the user's positive sample content on the same day in the business scenario. α0 is the third hyperparameter, which represents the default value of the positive sample when there is no data, and is obtained according to the average click-through rate of online content.

[0101] Because the first-time attenuation function takes the inverse, positive samples with a longer time since the last exposure have a smaller effect, while those with a closer time have a larger effect. This means that samples with longer repeated exposures have a smaller effect and need to be attenuated. The speed of attenuation is controlled by hyperparameters. In the formula, a and b are greater than 0 to make the formula more suitable for the business.

[0102] Assume there are three positive samples, which are click contents after repeated exposure for 10 days, 3 days, and 5 days respectively. Then α = initial value + b / e^(a*10) + b / e^(a*3) + b / e^(a*5).

[0103] For example, the parameter a is used to control the speed of time decay. The larger the parameter, the faster the decay. To put it simply, you can draw a graph (assuming b = 1). Figure 3 The vertical axis represents the attenuation coefficient when a=0.5, 1, or 2, and the horizontal axis represents the number of days.

[0104] Observing the graph, we find that when a = 0.5, the number approaches 0 after about 10 days, indicating that it is forgotten. This is relatively reasonable, so we choose a = 0.5. In contrast, when a = 2, the number approaches 2 after the third day, and when a = 1, the number approaches 0 after the sixth day, indicating that it is forgotten too quickly.

[0105] The value of b is usually set to 1 by default, and its main meaning is the degree of attenuation on that day (i.e., when the denominator = 1).

[0106] The original meaning of α is the number of times, but the present disclosure takes the attenuation into consideration, that is, a single behavior is discounted. The data is expressed as the above formula, that is, a single behavior after time decay may only be counted as 0.5 behaviors after discount.

[0107] In one embodiment, β in the Beta distribution function is obtained according to the following formula:

[0108]

[0109] in, Represents the accumulation of decay coefficients of negative samples decaying over time, which is also the second time decay function. Days′ represents the time between the exposure time of the current negative sample content category and the last exposure (for example, the duration can be: days, hours, minutes or seconds). a′ is the fourth hyperparameter, which represents the user's forgetting speed of the current negative sample content category, and is obtained according to the user's memory decay speed of the negative sample content in the business scenario. b′ is the fifth hyperparameter, which represents the user's forgetting speed of the current negative sample content category on the day, and is obtained according to the degree of forgetting of the user's negative sample content on the day in the business scenario. β0 is the sixth hyperparameter, which represents the default value of the negative sample when there is no data, and is obtained according to the average click-through rate of online content.

[0110] Because the second time decay function takes the inverse, positive samples with a longer time since the last exposure have a smaller effect, while those with a closer time have a larger effect. This means that samples with longer repeated exposures have a smaller effect and need to be attenuated. The speed of attenuation is controlled by hyperparameters. a′ and b′ in the formula are greater than 0 to make the formula more suitable for the business.

[0111] In the above formula, the parameters of the Beta distribution are updated in real time based on user feedback data. The initial value is a hyperparameter, that is, a default value when the user does not have any repeated exposure data.

[0112] α0 can be combined with online data. For example, if the click-through rate of most online ads is about 0.2, and if more than 5 impressions are considered relatively reliable, then the initial value of α is α0 = 1, and the initial value of β is β0 = 4. That is, click-through rate = 0.2 = 1 / (1+4).

[0113] In step S102, the content to be exposed and the group to be exposed are obtained;

[0114] In step S103, a corresponding target Beta distribution function is obtained according to the content category corresponding to the content to be exposed and the group to be exposed;

[0115] In step S104, an estimated click-through rate of the content to be exposed is obtained according to the probability density function corresponding to the target Beta distribution function. The estimated click-through rate is a random number generated by the probability density function corresponding to the target Beta distribution function.

[0116] by Figure 2 For example, Figure 2 There are various values in the beta distribution and their corresponding probabilities, so we can program the output of a random number P.

[0117] When α=10 and β=90, the distribution of generated random numbers will be concentrated around 0.1.

[0118] When α=2 and β=5, the generated random numbers will be scattered and not concentrated enough.

[0119] In step S105, a threshold value of the content to be exposed based on time decay is obtained according to the repeated exposure time interval corresponding to the content to be exposed;

[0120] In one embodiment, the threshold value of the content to be exposed based on time decay is obtained according to the following formula:

[0121]

[0122] Where M is the time-attenuation threshold for the content to be exposed, "days" represents the time between the exposure time and the last exposure time of the content category to be exposed. c is the sixth hyperparameter, which is obtained based on the speed of memory decay of users for content in business scenarios. d is the seventh hyperparameter, which is obtained based on the degree of forgetfulness of the content of the day in the business scenario.

[0123] This formula means that the longer the exposure interval, the higher the probability of repeated exposure, that is, the larger the M value c and d are greater than 0 and are business-defined hyperparameters.

[0124] In step S106, when the threshold of the content to be exposed based on time decay is greater than the estimated click rate, the content to be exposed is exposed;

[0125] In step S107 , when the threshold of the content to be exposed based on time decay is less than the estimated click rate, the content to be exposed is filtered.

[0126] To determine whether a piece of content warrants repeated exposure, first determine the content's category and the target user group. Find the corresponding Beta distribution function. Based on its probability density, generate a random number as the estimated click-through rate (P). Then, based on the interval between repeated exposures, determine a threshold M for the content to be exposed based on time decay.

[0127] M is between 0 and 1 and can be directly compared with P. Specifically, when M>P, it is considered that the content can be repeatedly exposed, otherwise it is considered that the content needs to be exposed and filtered.

[0128] It's worth noting that because the Beta distribution reflects the increasing confidence in the estimated CTR as the sample size increases, directly comparing the number of clicks / impressions to the threshold will be highly uncertain. For example, if there is one impression and zero clicks, the estimated CTR is 0, which is always lower than the threshold, so no more impressions are considered. This is obviously unreasonable.

[0129] Using the Beta distribution to generate random numbers can effectively solve this problem. Furthermore, as the confidence level increases, the fluctuation of the random numbers gradually decreases, and the comparison result with the threshold becomes more reliable.

[0130] The above embodiments use community scenarios as an example and assume that users are passively exposed to content, such as the Xiaohongshu homepage, and do not consider scenarios such as active searches. Only repeated exposure is considered, without specifically distinguishing between secondary and tertiary exposures. The classification of content and users is simplified and not explained in detail. The data on users' repeated exposure to content only considers click-through rates, with higher click-through rates indicating greater user liking of the content. Users generally dislike repeated exposure in the short term. However, users' memory of previously exposed content increases over time, and the more they forget, the higher the probability of repeated exposure. The numbers are for illustration only, and specific values can be determined based on the characteristics of different services.

[0131] The present disclosure provides an exposure method, including: obtaining a Beta distribution function for each content category for each user group, where α in the Beta distribution function represents a positive sample, which is obtained based on the number of times a first repeatedly exposed content is clicked and the decay value of the first repeatedly exposed content over time, and β in the Beta distribution function represents a negative sample, which is obtained based on the number of times a second repeatedly exposed content is not clicked and the decay value of the second repeatedly exposed content over time; obtaining content to be exposed and a group to be exposed; obtaining a corresponding target Beta distribution function based on the content category and the group to be exposed corresponding to the content to be exposed; obtaining an estimated click-through rate of the content to be exposed based on a probability density function corresponding to the target Beta distribution function, where the estimated click-through rate is a random number generated by the probability density function corresponding to the target Beta distribution function; obtaining a threshold value of the content to be exposed based on time decay according to the repeated exposure time interval corresponding to the content to be exposed; exposing the content to be exposed when the threshold value is greater than the estimated click-through rate; and filtering the content to be exposed when the threshold value is less than the estimated click-through rate. Among them, based on the Beta distribution and the time decay of the sample, a probability distribution function is given to estimate the click-through rate. By comparing the estimated click-through rate with the content's time-decay threshold, it is determined whether the content should be filtered. Because it is based on feedback from factual data and takes time decay into consideration, as the number of repeated exposures increases, the confidence level of the exposure filtering judgment continues to increase, thereby improving the accuracy of exposure.

[0132] The present disclosure allows for personalized determination of exposure filtering, specifically including: 1. User personalization: different types of users have different exposure filtering periods. 2. Content personalization: different content has different exposure filtering periods.

[0133] Based on feedback from factual data and consideration of time decay, the confidence level of exposure filtering judgments increases with the number of repeated exposures.

[0134] Clustering / classifying similar users and reducing the number of times users are repeatedly exposed to content when the confidence level of exposure filtering calculation is low improves the user experience during the initial cold start.

[0135] In addition, similar users are clustered / classified. When the confidence of exposure filtering calculation is low, the number of times users are repeatedly exposed to content is reduced, which improves the user experience during the initial cold start.

[0136] Specifically, when determining whether repeatedly exposed content requires filtering, a function is used to simulate the time-dependent decay of memory, incorporating the Beta probability distribution into the calculation. This function is continuously updated based on real-time user data, continuously increasing confidence in exposure filtering decisions.

[0137] When there is little feedback data in the early stages, users are clustered / classified to balance the contradiction between the accuracy of user personalized exposure filtering and the reduction in user experience caused by repeated exposure.

[0138] The following are embodiments of the apparatus disclosed herein, which can be used to execute embodiments of the method disclosed herein.

[0139] Figure 4 FIG. 1 is a block diagram of an exposure device according to an exemplary embodiment. Figure 4 As shown, the exposure device includes:

[0140] A first acquisition module 11 is configured to obtain a Beta distribution function for each content category for each user group, where α in the Beta distribution function represents a positive sample, which is obtained based on the number of clicks on the first repeatedly exposed content and the decay value of the first repeatedly exposed content over time; and β in the Beta distribution function represents a negative sample, which is obtained based on the number of unclicks on the second repeatedly exposed content and the decay value of the second repeatedly exposed content over time.

[0141] The second acquisition module 12 is used to acquire the content to be exposed and the group to be exposed;

[0142] A third acquisition module 13 is configured to acquire a corresponding target Beta distribution function according to the content category corresponding to the content to be exposed and the group to be exposed;

[0143] A fourth acquisition module 14 is configured to acquire an estimated click-through rate of the content to be exposed based on a probability density function corresponding to the target Beta distribution function, where the estimated click-through rate is a random number generated by the probability density function corresponding to the target Beta distribution function;

[0144] A fifth acquisition module 15 is configured to acquire a time-attenuation-based threshold of the content to be exposed according to a repeated exposure time interval corresponding to the content to be exposed;

[0145] An exposure module 16, configured to expose the content to be exposed when the threshold is greater than the estimated click rate;

[0146] The filtering module 17 is configured to filter the content to be exposed when the threshold is less than the estimated click rate.

[0147] In one embodiment, the first acquisition module includes: a first acquisition submodule,

[0148] The first acquisition submodule is configured to obtain α according to the following formula:

[0149]

[0150] Among them, the Represents the accumulation of attenuation coefficients of positive samples that decay over time. The days represents the time between the exposure time of the content category of the current positive sample and the last exposure. The a is the first hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample, and is obtained according to the decay speed of the user's memory of the content of the positive sample in the business scenario. The b is the second hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample on the same day, and is obtained according to the degree of forgetting of the content of the positive sample on the same day by the user in the business scenario. The α0 is the third hyperparameter, which represents the default value of the positive sample when there is no data, and is obtained according to the average click-through rate of online content.

[0151] In one embodiment, the first acquisition module includes: a second acquisition submodule,

[0152] The second acquisition submodule is configured to obtain β according to the following formula:

[0153]

[0154] Among them, the Represents the accumulation of attenuation coefficients of negative samples that decay over time. The days′ represents the time between the exposure time of the content category of the current negative sample and the last exposure time. The a′ is the fourth hyperparameter, which represents the user's forgetting speed of the content category of the current negative sample, and is obtained according to the user's memory decay speed of the content of the negative sample in the business scenario. The b′ is the fifth hyperparameter, which represents the user's forgetting speed of the content category of the current negative sample on the same day, and is obtained according to the user's forgetting degree of the content of the negative sample on the same day in the business scenario. The β0 is the sixth hyperparameter, which represents the default value of the negative sample when there is no data, and is obtained according to the average click-through rate of online content.

[0155] In one embodiment, the fifth acquisition module includes: a third acquisition submodule,

[0156] The third acquisition submodule is configured to acquire a threshold value of the content to be exposed based on time decay according to the following formula:

[0157]

[0158] Among them, M is the threshold of the content to be exposed based on time decay, and days″ represents the length of time between the exposure time and the last exposure of the content category to be exposed. c is the sixth hyperparameter, which is obtained based on the speed of memory decay of users for content in business scenarios. d is the seventh hyperparameter, which is obtained based on the degree of forgetfulness of the content of the day by users in business scenarios.

[0159] According to a third aspect of the embodiments of the present disclosure, there is provided an exposure device, comprising:

[0160] processor;

[0161] a memory for storing processor-executable instructions;

[0162] The processor is configured as follows:

[0163] Obtain a Beta distribution function for each user group for each content category, where α in the Beta distribution function represents a positive sample, which is obtained based on the number of clicks on the first repeatedly exposed content and the decay value of the first repeatedly exposed content over time. β in the Beta distribution function represents a negative sample, which is obtained based on the number of unclicks on the second repeatedly exposed content and the decay value of the second repeatedly exposed content over time.

[0164] Obtain the content and groups to be exposed;

[0165] Obtaining a corresponding target Beta distribution function according to the content category corresponding to the content to be exposed and the group to be exposed;

[0166] Obtaining an estimated click-through rate of the content to be exposed according to a probability density function corresponding to the target Beta distribution function, where the estimated click-through rate is a random number generated by the probability density function corresponding to the target Beta distribution function;

[0167] Obtaining a time-attenuation threshold of the content to be exposed according to a repeated exposure time interval corresponding to the content to be exposed;

[0168] When the threshold is greater than the estimated click rate, exposing the content to be exposed;

[0169] When the threshold is less than the estimated click rate, the content to be exposed is filtered.

[0170] The above processor may also be configured to:

[0171] In one embodiment, obtaining α in the Beta distribution function of each user group for each content category includes:

[0172] The α is obtained according to the following formula:

[0173]

[0174] Among them, the Represents the accumulation of attenuation coefficients of positive samples that decay over time. The days represents the time between the exposure time of the content category of the current positive sample and the last exposure. The a is the first hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample, and is obtained according to the decay speed of the user's memory of the content of the positive sample in the business scenario. The b is the second hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample on the same day, and is obtained according to the degree of forgetting of the content of the positive sample on the same day by the user in the business scenario. The α0 is the third hyperparameter, which represents the default value of the positive sample when there is no data, and is obtained according to the average click-through rate of online content.

[0175] In one embodiment, obtaining β in the Beta distribution function of each user group for each content category includes:

[0176] The β is obtained according to the following formula:

[0177]

[0178] Among them, the Represents the accumulation of attenuation coefficients of negative samples that decay over time. The days′ represents the time between the exposure time of the content category of the current negative sample and the last exposure time. The a′ is the fourth hyperparameter, which represents the user's forgetting speed of the content category of the current negative sample, and is obtained according to the user's memory decay speed of the content of the negative sample in the business scenario. The b′ is the fifth hyperparameter, which represents the user's forgetting speed of the content category of the current negative sample on the same day, and is obtained according to the user's forgetting degree of the content of the negative sample on the same day in the business scenario. The β0 is the sixth hyperparameter, which represents the default value of the negative sample when there is no data, and is obtained according to the average click-through rate of online content.

[0179] In one embodiment, obtaining a threshold value of the content to be exposed based on time decay according to the repeated exposure time interval corresponding to the content to be exposed includes:

[0180] The threshold value of the content to be exposed based on time decay is obtained according to the following formula:

[0181]

[0182] Among them, M is the threshold of the content to be exposed based on time decay, and days″ represents the length of time between the exposure time and the last exposure of the content category to be exposed. c is the sixth hyperparameter, which is obtained based on the speed of memory decay of users for content in business scenarios. d is the seventh hyperparameter, which is obtained based on the degree of forgetfulness of the content of the day by users in business scenarios.

[0183] Regarding the apparatus in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0184] Figure 5 FIG2 is a block diagram illustrating an apparatus 90 for exposure according to an exemplary embodiment. For example, apparatus 90 may be provided as a server. Apparatus 90 includes a processing component 902, which further includes one or more processors, and memory resources represented by memory 903 for storing instructions, such as applications, executable by processing component 902. The applications stored in memory 903 may include one or more modules, each corresponding to a set of instructions. Furthermore, processing component 902 is configured to execute the instructions to perform the above-described method.

[0185] The device 90 may further include a power supply component 906 configured to perform power management of the device 90, a wired or wireless network interface 905 configured to connect the device 90 to a network, and an input / output (I / O) interface 908. The device 90 may operate based on an operating system stored in the memory 903, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, or the like.

[0186] A non-transitory computer-readable storage medium, when the instructions in the storage medium are executed by the processor of the device 90, enables the device 90 to perform the above-mentioned exposure method, the method comprising:

[0187] Obtain a Beta distribution function for each user group for each content category, where α in the Beta distribution function represents a positive sample, which is obtained based on the number of clicks on the first repeatedly exposed content and the decay value of the first repeatedly exposed content over time. β in the Beta distribution function represents a negative sample, which is obtained based on the number of unclicks on the second repeatedly exposed content and the decay value of the second repeatedly exposed content over time.

[0188] Obtain the content and groups to be exposed;

[0189] Obtaining a corresponding target Beta distribution function according to the content category corresponding to the content to be exposed and the group to be exposed;

[0190] Obtaining an estimated click-through rate of the content to be exposed according to a probability density function corresponding to the target Beta distribution function, where the estimated click-through rate is a random number generated by the probability density function corresponding to the target Beta distribution function;

[0191] Obtaining a time-attenuation threshold of the content to be exposed according to a repeated exposure time interval corresponding to the content to be exposed;

[0192] When the threshold is greater than the estimated click rate, exposing the content to be exposed;

[0193] When the threshold is less than the estimated click rate, the content to be exposed is filtered.

[0194] In one embodiment, obtaining α in the Beta distribution function of each user group for each content category includes:

[0195] The α is obtained according to the following formula:

[0196]

[0197] Among them, the Represents the accumulation of attenuation coefficients of positive samples that decay over time. The days represents the time between the exposure time of the content category of the current positive sample and the last exposure. The a is the first hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample, and is obtained according to the decay speed of the user's memory of the content of the positive sample in the business scenario. The b is the second hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample on the same day, and is obtained according to the degree of forgetting of the content of the positive sample on the same day by the user in the business scenario. The α0 is the third hyperparameter, which represents the default value of the positive sample when there is no data, and is obtained according to the average click-through rate of online content.

[0198] In one embodiment, obtaining β in the Beta distribution function of each user group for each content category includes:

[0199] The β is obtained according to the following formula:

[0200]

[0201] Among them, the Represents the accumulation of attenuation coefficients of negative samples that decay over time. The days′ represents the time between the exposure time of the content category of the current negative sample and the last exposure time. The a′ is the fourth hyperparameter, which represents the user's forgetting speed of the content category of the current negative sample, and is obtained according to the user's memory decay speed of the content of the negative sample in the business scenario. The b′ is the fifth hyperparameter, which represents the user's forgetting speed of the content category of the current negative sample on the same day, and is obtained according to the user's forgetting degree of the content of the negative sample on the same day in the business scenario. The β0 is the sixth hyperparameter, which represents the default value of the negative sample when there is no data, and is obtained according to the average click-through rate of online content.

[0202] In one embodiment, obtaining a threshold value of the content to be exposed based on time decay according to the repeated exposure time interval corresponding to the content to be exposed includes:

[0203] The threshold value of the content to be exposed based on time decay is obtained according to the following formula:

[0204]

[0205] Among them, M is the threshold of the content to be exposed based on time decay, and days″ represents the length of time between the exposure time and the last exposure of the content category to be exposed. c is the sixth hyperparameter, which is obtained based on the speed of memory decay of users for content in business scenarios. d is the seventh hyperparameter, which is obtained based on the degree of forgetfulness of the content of the day by users in business scenarios.

[0206] Other embodiments of the present disclosure will readily occur to those skilled in the art after considering the specification and practicing the disclosure herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the following claims.

[0207] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. An exposure method, characterized in that: include: Obtain a Beta distribution function for each user group for each content category, where α in the Beta distribution function represents a positive sample, which is obtained based on the number of clicks on the first repeatedly exposed content and the decay value of the first repeatedly exposed content over time. β in the Beta distribution function represents a negative sample, which is obtained based on the number of unclicks on the second repeatedly exposed content and the decay value of the second repeatedly exposed content over time. Obtain the content and groups to be exposed; Obtaining a corresponding target Beta distribution function according to the content category corresponding to the content to be exposed and the group to be exposed; Obtaining an estimated click-through rate of the content to be exposed according to a probability density function corresponding to the target Beta distribution function, where the estimated click-through rate is a random number generated by the probability density function corresponding to the target Beta distribution function; Obtaining a time-attenuation threshold of the content to be exposed according to a repeated exposure time interval corresponding to the content to be exposed; When the threshold is greater than the estimated click rate, exposing the content to be exposed; When the threshold is less than the estimated click rate, filtering the content to be exposed; The acquiring, according to the repeated exposure time interval corresponding to the content to be exposed, a threshold value based on time attenuation of the content to be exposed includes: The threshold value of the content to be exposed based on time decay is obtained according to the following formula: Among them, M is the threshold of the content to be exposed based on time decay, and days″ represents the length of time between the exposure time and the last exposure of the content category to be exposed. c is the sixth hyperparameter, which is obtained based on the speed of memory decay of users for content in business scenarios. d is the seventh hyperparameter, which is obtained based on the degree of forgetfulness of the content of the day by users in business scenarios.

2. The method according to claim 1, characterized in that Obtain the α in the Beta distribution function for each user group for each content category, including: The α is obtained according to the following formula: Among them, the Represents the accumulation of attenuation coefficients of positive samples that decay over time. The days represents the time between the exposure time of the content category of the current positive sample and the last exposure. The a is the first hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample, and is obtained according to the decay speed of the user's memory of the content of the positive sample in the business scenario. The b is the second hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample on the same day, and is obtained according to the degree of forgetting of the content of the positive sample on the same day by the user in the business scenario. The α0 is the third hyperparameter, which represents the default value of the positive sample when there is no data, and is obtained according to the average click-through rate of online content.

3. The method according to claim 1, characterized in that Obtain the β in the Beta distribution function for each user group for each content category, including: The β is obtained according to the following formula: Among them, the The cumulative decay coefficients of negative samples decay over time are represented by days′, which represents the time between the exposure time and the last exposure time of the content category of the current negative sample. a′ is the fourth hyperparameter, which represents the user's forgetting speed of the content category of the current negative sample, and is obtained based on the user's memory decay speed of the negative sample content in the business scenario. b′ is the fifth hyperparameter, which represents the user's forgetting speed of the content category of the current negative sample on the day corresponding to the current negative sample, and is obtained based on the user's forgetting degree of the negative sample content on the day in the business scenario. The β0 is the sixth hyperparameter, which represents the default value of negative samples when there is no data and is obtained based on the average click-through rate of online content.

4. An exposure device, characterized in that include: A first acquisition module is configured to acquire a Beta distribution function for each user group for each content category, where α in the Beta distribution function represents a positive sample and is obtained based on the number of clicks on the first repeatedly exposed content and the decay value of the first repeatedly exposed content over time; and β in the Beta distribution function represents a negative sample and is obtained based on the number of unclicks on the second repeatedly exposed content and the decay value of the second repeatedly exposed content over time; The second acquisition module is used to acquire the content to be exposed and the group to be exposed; A third acquisition module is used to acquire a corresponding target Beta distribution function according to the content category corresponding to the content to be exposed and the group to be exposed; A fourth acquisition module is configured to acquire an estimated click-through rate of the content to be exposed according to a probability density function corresponding to the target Beta distribution function, where the estimated click-through rate is a random number generated by the probability density function corresponding to the target Beta distribution function; A fifth acquisition module, configured to acquire a threshold value of the content to be exposed based on time attenuation according to a repeated exposure time interval corresponding to the content to be exposed; An exposure module, configured to expose the content to be exposed when the threshold is greater than the estimated click rate; A filtering module, configured to filter the content to be exposed when the threshold is less than the estimated click rate; The acquiring, according to the repeated exposure time interval corresponding to the content to be exposed, a threshold value based on time attenuation of the content to be exposed includes: The threshold value of the content to be exposed based on time decay is obtained according to the following formula: Among them, M is the threshold of the content to be exposed based on time decay, and days″ represents the length of time between the exposure time and the last exposure of the content category to be exposed. c is the sixth hyperparameter, which is obtained based on the speed of memory decay of users for content in business scenarios. d is the seventh hyperparameter, which is obtained based on the degree of forgetfulness of the content of the day by users in business scenarios.

5. The device according to claim 4, characterized in that The first acquisition module includes: a first acquisition submodule, which is configured to acquire α according to the following formula: Among them, the Represents the accumulation of attenuation coefficients of positive samples that decay over time. The days represents the time between the exposure time of the content category of the current positive sample and the last exposure. The a is the first hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample, and is obtained according to the decay speed of the user's memory of the content of the positive sample in the business scenario. The b is the second hyperparameter, which represents the user's forgetting speed of the content category of the current positive sample on the same day, and is obtained according to the degree of forgetting of the content of the positive sample on the same day by the user in the business scenario. The α0 is the third hyperparameter, which represents the default value of the positive sample when there is no data, and is obtained according to the average click-through rate of online content.

6. The device according to claim 4, characterized in that The first acquisition module includes: a second acquisition submodule, The second acquisition submodule is used to obtain the β according to the following formula: Among them, the Represents the accumulation of attenuation coefficients of negative samples that decay over time. The days′ represents the time between the exposure time of the content category of the current negative sample and the last exposure time. The a′ is the fourth hyperparameter, which represents the user's forgetting speed of the content category of the current negative sample, and is obtained according to the user's memory decay speed of the content of the negative sample in the business scenario. The b′ is the fifth hyperparameter, which represents the user's forgetting speed of the content category of the current negative sample on the same day, and is obtained according to the user's forgetting degree of the content of the negative sample on the same day in the business scenario. The β0 is the sixth hyperparameter, which represents the default value of the negative sample when there is no data, and is obtained according to the average click-through rate of online content.

7. The device according to claim 4, characterized in that The fifth acquisition module includes: a third acquisition submodule, The third acquisition submodule is configured to acquire a threshold value of the content to be exposed based on time decay according to the following formula: Among them, M is the threshold of the content to be exposed based on time decay, and days″ represents the length of time between the exposure time and the last exposure of the content category to be exposed. c is the sixth hyperparameter, which is obtained based on the speed of memory decay of users for content in business scenarios. d is the seventh hyperparameter, which is obtained based on the degree of forgetfulness of the content of the day by users in business scenarios.

8. An exposure device, characterized in that: include: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to: Obtain a Beta distribution function for each user group for each content category, where α in the Beta distribution function represents a positive sample, which is obtained based on the number of clicks on the first repeatedly exposed content and the decay value of the first repeatedly exposed content over time. β in the Beta distribution function represents a negative sample, which is obtained based on the number of unclicks on the second repeatedly exposed content and the decay value of the second repeatedly exposed content over time. Obtain the content and groups to be exposed; Obtaining a corresponding target Beta distribution function according to the content category corresponding to the content to be exposed and the group to be exposed; Obtaining an estimated click-through rate of the content to be exposed according to a probability density function corresponding to the target Beta distribution function, where the estimated click-through rate is a random number generated by the probability density function corresponding to the target Beta distribution function; Obtaining a time-attenuation threshold of the content to be exposed according to a repeated exposure time interval corresponding to the content to be exposed; When the threshold is greater than the estimated click rate, exposing the content to be exposed; When the threshold is less than the estimated click rate, filtering the content to be exposed; The acquiring, according to the repeated exposure time interval corresponding to the content to be exposed, a threshold value based on time attenuation of the content to be exposed includes: The threshold value of the content to be exposed based on time decay is obtained according to the following formula: Among them, M is the threshold of the content to be exposed based on time decay, and days″ represents the length of time between the exposure time and the last exposure of the content category to be exposed. c is the sixth hyperparameter, which is obtained based on the speed of memory decay of users for content in business scenarios. d is the seventh hyperparameter, which is obtained based on the degree of forgetfulness of the content of the day by users in business scenarios.

9. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instruction is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

Images