A first picture recommendation method and device and a storage medium

Abstract

The application provides a first picture recommendation method, comprising: obtaining at least two candidate pictures of a use user; determining scores of the at least two candidate pictures based on a picture evaluation model of the use user; selecting one of the at least two candidate pictures as a first picture to be displayed based on the scores; adjusting the picture evaluation model of the use user according to feedback of a browsing user on the displayed candidate picture of the use user, so that the scores change with the feedback of the browsing user on the candidate picture of the use user until the change of the scores of the candidate picture is stable within a first threshold; taking the picture evaluation model in which the change of the scores of the candidate picture is stable within the first threshold as a first picture recommendation model of the use user; and determining a first picture of the use user based on the first picture recommendation model. The application solves the technical problems of large ineffective exposure and low matching efficiency of a social platform, improves the probability that a first picture of a user is loved, and reduces exposure waste.

Description

Technical Field

[0001] This application relates to the field of image recommendation, and more particularly to a method, apparatus and storage medium for recommending a first image. Background Technology

[0002] In stranger-based social networking scenarios, the appropriateness of the homepage image selection is one of the key factors influencing whether users can be successfully matched. However, most users are unsure which images are more likely to be popular with similar users when choosing their homepage image. Exposing a less attractive homepage image fails to increase the user's chances of social interaction, resulting in a large amount of ineffective exposure on the social platform and making it difficult to improve the chances of being popular. Summary of the Invention

[0003] In view of this, this application provides a method, apparatus and storage medium for recommending homepage images, so as to increase the probability of a user's homepage image being liked and reduce exposure waste.

[0004] The first aspect of this invention provides a method for recommending header images, comprising:

[0005] Obtain at least two alternative images from the user;

[0006] Determine the scores of the at least two candidate images;

[0007] Based on the score, one of the at least two candidate images is selected as the first image to be displayed;

[0008] The image evaluation model for the user is adjusted based on the feedback from browsing users to the displayed alternative images for the user, so that the score changes with the feedback from browsing users to the alternative images for the user, until the change in the score of the alternative images stabilizes within a first threshold.

[0009] The image evaluation model whose score changes are stable within a first threshold is used as the first image recommendation model for the user.

[0010] The user's homepage image is determined based on the homepage recommendation model.

[0011] In some embodiments, the image evaluation model is controlled by at least a first parameter and a second parameter. If the browsing user provides positive feedback on the candidate images displayed by the user, the first parameter is adjusted; if the browsing user does not provide positive feedback on the candidate images displayed by the user, the second parameter is adjusted.

[0012] In some embodiments, the selection of one of the at least two candidate images as the primary image is based on the score:

[0013] Calculate the display probability of the score, and select the candidate image with the highest display probability at the current moment as the display image;

[0014] If, based on the user's feedback at the current moment, the displayed image's score becomes lower than the user's other candidate images, then the displayed image with a score lower than the user's other candidate images has a higher probability of being displayed than the user's other candidate images.

[0015] In some embodiments, the image evaluation model is to calculate the scores of the at least two candidate images using the probability density function of the beta distribution;

[0016] The step of adjusting the user's image evaluation model based on the browsing user's feedback on the displayed alternative images, so that the score changes with the browsing user's feedback on the user's alternative images, includes:

[0017] If the browsing user provides positive feedback on the candidate images selected by the user, then the value of alpha in the probability density function of the beta distribution is increased;

[0018] If the browsing user does not provide positive feedback on the candidate images selected by the user, then the value of beta in the probability density function of the beta distribution is increased.

[0019] In some embodiments, if the first image is displayed more than a second threshold number of times, and the score of the candidate images does not stabilize within a first threshold, then the image evaluation model updated after displaying the second threshold number of times is used as the first image recommendation model for the user.

[0020] In some embodiments, determining the user's homepage image based on the homepage image recommendation model includes:

[0021] Based on the aforementioned first image recommendation model, the order of the candidate images selected by the user is adjusted, and the user selects any one of the sorted candidate images as the first image.

[0022] Alternatively, the candidate image with the highest score based on the aforementioned first image recommendation model can be used as the first image.

[0023] In some embodiments, after a third time interval, the first image recommendation model is updated. The update method is as follows: the first image recommendation model of the user is adjusted based on the feedback of the browsing user to the candidate images displayed to the user, so that the score changes with the feedback of the browsing user to the candidate images of the user, until the score of the candidate images stabilizes within a third threshold. The first image recommendation model in which the score of the candidate images stabilizes within the third threshold is used as the updated first image recommendation model of the user.

[0024] In some embodiments, the initial image evaluation model is determined as follows:

[0025] An initial image evaluation model is established based on the popularity of images from active users on the platform, and the scores of the at least two candidate images are determined using the initial image evaluation model.

[0026] A second aspect of the present invention provides a first image recommendation device, comprising:

[0027] The image acquisition unit is configured to acquire at least two candidate images that are distinguishable to the user.

[0028] A static unit, wherein the static assignment unit is configured to determine the scores of the at least two candidate images based on a user-defined image evaluation model;

[0029] The display unit is configured to select one of the at least two candidate images as the first image to be displayed based on the score; and after the dynamic unit obtains a stable first image recommendation model, it determines the first image of the user based on the first image recommendation model.

[0030] A dynamic unit is configured to adjust the user's image evaluation model based on the browsing user's feedback on the displayed candidate images, such that the score changes with the browsing user's feedback on the candidate images until the score of the candidate images stabilizes within a first threshold; and the image evaluation model whose score of the candidate images stabilizes within the first threshold is used as the user's first image recommendation model.

[0031] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the steps of the first image recommendation method described above. Attached Figure Description

[0032] Figure 1 This is a flowchart illustrating a first-image recommendation method as an exemplary embodiment of the present invention.

[0033] Figure 2This is a comparison diagram illustrating the benefits and effects of a first image recommendation method as an exemplary embodiment of the present invention.

[0034] Figure 3 This is a schematic block diagram of a recommended device in some exemplary embodiments of the present invention. Detailed Implementation

[0035] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0036] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0037] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0038] Stranger social networking scenarios typically involve matching users on both sides. On one hand, as browsing users, the sheer volume of users on the social platform often leads to quick judgments based solely on the first image displayed, making its selection crucial. On the other hand, as users, the platform allows them to upload multiple images and choose one as their first image. However, most users are unsure which images will be more appealing to similar browsing users when selecting their first image. Consequently, the images they prioritize are not necessarily those most likely to be liked by their intended target audience, thus reducing the success rate of matching.

[0039] Traditional methods for determining the primary image default to using the user-defined primary image as the main image for display. This means that even if the platform determines a high degree of matching between the two parties based on other information modeling, the matching may still fail because the user's primary image selection is less attractive to the browsing user.

[0040] With the increasing application of data science-based recommendation and ranking technologies in social scenarios, techniques such as random exploration have been introduced into the recommendation process for the first image. For example, random exploration is performed in the early stages of the display, and the better image is selected for display after the random exploration. However, the above methods have two main problems:

[0041] First, the initial exploration process was highly random. The purely random approach did not utilize prior information about the images, and there was a significant risk that poorly exposed images could cause considerable harm to users.

[0042] Secondly, the use of fixed images for display in the later stages did not take into account the randomness of the early exploration process and the time fluidity of platform users. More specifically, it did not take into account the potential differences in the platform's audience between the early and later stages.

[0043] In view of this, the present invention provides a method for recommending the first image, which combines a prior image evaluation model to determine the best image during static startup, dynamically adjusts the image evaluation model until it meets expectations, and uses the adjusted image evaluation model as the model algorithm for recommending the first image, thereby reducing unnecessary exposure waste and increasing the chances of the user being popular.

[0044] Figure 1 This is a flowchart of a header image recommendation method in some exemplary embodiments.

[0045] Please refer to Figure 1 One method for recommending the main image includes:

[0046] S101 retrieves at least two alternative images from the user.

[0047] When acquiring potential images from users, guide them to upload multiple distinctive images as options to expand the types and range of choices. For example, after a user registers and enters the application, interactive prompts can guide them to upload photos of lifestyle, sports, and other types, including those based on their registered interests. As users' historical behavior accumulates, they can be further guided to add new types of images based on their past behavior.

[0048] Stranger social networking typically relies on comprehensive data matching and recommendation to guide users to upload a variety of distinctive images. The types of these images are related to their registration information or historical behavior to some extent, ensuring that there are enough high-quality images as alternatives during the initial image recommendation process, thus increasing the chances of being liked by browsing users who are matched with the data after exposure.

[0049] S102 determines the scores of the at least two candidate images.

[0050] An image rating model built using images from active users on the platform scores the popularity of at least two candidate images submitted by the user. This model employs regression modeling. The input to the regression model is the images uploaded by active users; the output is the popularity score for each of these images. Any regression model can be used to determine the image rating model; this embodiment does not limit this approach. Furthermore, the regression model is merely a modeling method for determining the image rating model, and the image rating model itself is not limited to regression models.

[0051] As an exception, for users who have registered for a period of time and have only uploaded a single image, the popularity is determined by the user's average like rate over the most recent period, such as, but not limited to, the most recent 30 days.

[0052] Understandably, to improve the effectiveness of subsequent matching, it is possible to further limit the scoring to the popularity of the user's images by active users related to the user's data.

[0053] In some embodiments, the image evaluation model is controlled by at least a first parameter and a second parameter. These parameters change dynamically based on user feedback after the system goes live.

[0054] After determining the initial image evaluation model based on the above method, the initial image evaluation model is used to predictively score the candidate images selected by the user.

[0055] This concludes the static cold start score assignment process for the first image recommendation method. The next steps involve dynamic exploration after deployment.

[0056] S103 selects one of the at least two candidate images as the first image to be displayed based on the score.

[0057] It is understandable that there is randomness in the process of selecting one of the candidate images as the main image. Considering the characteristics of the application scenario of this method, the main image is quickly viewed by a large number of users. As the main image is frequently clicked on or swiped by users, the image evaluation model is rapidly iterated.

[0058] The process of selecting one of the at least two candidate images as the first image based on the score also includes further statistical distribution calculation of the score, using the calculated value as the basis for the first image display. Symmetrical distribution models such as the normal distribution can be used as the above statistical distribution calculation method, so that the selection of the first image has a certain degree of randomness around the score: candidate images with higher scores have a higher probability of being the first image, candidate images with medium scores have a medium probability of being the first image, candidate images with lower scores have a lower probability of being the first image, and the candidate image with the highest score is not guaranteed to be the first image.

[0059] By adding a method that uses statistical distribution calculations based on scores, rather than absolutely relying on score-based ranking for the first image display, the flexibility and accuracy of the first image recommendation are improved. This avoids situations where other potentially more popular images are not shown as the first image due to the first image not changing, thus missing out on better first image selection opportunities.

[0060] S104 Adjusts the user's image evaluation model based on the browsing user's feedback on the displayed candidate images, so that the score changes with the browsing user's feedback on the user's candidate images until the score of the candidate images stabilizes within a first threshold. The image evaluation model whose score of the candidate images stabilizes within the first threshold is used as the user's first image recommendation model.

[0061] In an embodiment of the image evaluation model controlled by at least the first parameter and the second parameter involved in S102, if the browsing user provides positive feedback on the candidate image displayed by the user, the first parameter is adjusted; if the browsing user does not provide positive feedback on the candidate image displayed by the user, the second parameter is adjusted.

[0062] Of course, to improve the effectiveness of subsequent matching, the aforementioned browsing users can be further limited to active users related to matching using user data.

[0063] Taking an image rating model based on the beta distribution as an exemplary embodiment, the beta distribution is a type of probability distribution, typically used to describe values ​​between 0 and 1, such as probability, ratio, or score. The parameters of the beta distribution can be estimated using methods such as maximum likelihood estimation or Bayesian inference. In practical applications, the beta distribution is often used to model the prior distribution of binomial distribution parameters, or as the posterior distribution of the probability of success in a Bernoulli trial. When the prior distribution is a beta distribution, its posterior distribution has the same form. The aforementioned properties of the beta distribution are used to predict and score candidate images.

[0064] Specifically, the probability density function of the beta distribution can be used to calculate the score of the candidate images.

[0065] The probability density function of the beta distribution is:

[0066]

[0067] Where α and β are distribution parameters that control the shape of the probability density function, x is a random variable, and B(α,β) is the beta function.

[0068] If the browsing user provides positive feedback on the candidate images selected by the user, the value of α in the probability density function of the beta distribution is increased; if the browsing user does not provide positive feedback on the candidate images selected by the user, the value of β in the probability density function of the beta distribution is increased.

[0069] As exposure increases rapidly, user feedback on the user's candidate images iterates quickly, leading to a gradual stabilization of the like rate and a stable beta distribution of the like rate for each candidate image. In other words, this can be understood as the image evaluation model for user-selected images, which is related to the user's data, stabilizing within the current time period. Using an image evaluation model where the score changes of candidate images remain stable within a first threshold within the current time period, we can serve as the user's first image recommendation model, thereby reducing unnecessary exposure waste and increasing the user's chances of being popular.

[0070] Because there is a high degree of randomness among active users, both as users and as browsing users, the use of user-uploaded images as candidate images is highly subjective. Even if data matching is introduced during the selection process of browsing users, there is still a possibility that browsing users may not like all the candidate images selected by the user.

[0071] In this scenario, if the first image is displayed more than the second threshold number of times, and the scores of the candidate images do not stabilize within the first threshold, then the image evaluation model updated after displaying the second threshold number of times is used as the first image recommendation model for the user. This approach, considering computational efficiency, ensures that even if the image evaluation model does not meet expectations, it can still optimize the recommendation effect of the first image to a certain extent without introducing excessive resource consumption.

[0072] Since the first image recommendation model is determined based on feedback from users browsing within a specific time period, user preferences may change over time. By updating the first image recommendation model at regular intervals, it can adapt to these changing preferences, better reflect current user preferences, and iterate effectively to avoid wasted exposure due to outdated feedback.

[0073] The method for updating the first image recommendation model is similar to the method for determining the first image recommendation model. Specifically, the first image recommendation model for the user is adjusted based on the browsing user's feedback on the candidate images displayed to the user, so that the score changes with the browsing user's feedback on the candidate images to the user, until the score change of the candidate images stabilizes within a third threshold. The first image recommendation model where the score change of the candidate images stabilizes within the third threshold is used as the updated first image recommendation model for the user.

[0074] It is worth noting that if a new image uploaded by the user is used as a candidate image, or if one or more of the candidate images are updated, the above steps need to be repeated for the candidate images to determine a new primary image recommendation model.

[0075] S105 determines the user's homepage image based on the homepage recommendation model.

[0076] Similar to S103, which selects one of the at least two candidate images as the first image to be displayed based on the score, S105, in the process of determining the first image of the user based on the first image recommendation model, also includes calculating the statistical distribution of the dynamically adjusted score to obtain the display probability.

[0077] In some embodiments, the display probability of the score is calculated, and the candidate image with the highest display probability at the current moment is selected as the display image. If, based on the user's feedback on the display image at the current moment, the score of the display image becomes lower than that of the user's other candidate images, then the display image with a lower score than the user's other candidate images has a higher probability of having a lower display probability than the user's other candidate images.

[0078] When determining the main image, you can rely entirely on the main image recommendation model, or you can provide users with some freedom of choice based on the main image recommendation model.

[0079] For example, the candidate image with the highest score at the current moment can be selected as the display image. Alternatively, if the display image's score decreases below the user's other candidate images due to feedback from the browsing user at the current moment, then the candidate image with the highest score among the current display images will be selected as the display image for the next moment.

[0080] In this embodiment, by introducing modeling based on historical pictures during static cold start in the display of the user's premier picture, combined with the method of dynamic exploration, on the basis of fully considering the confidence of dynamic exploration, the current optimal alternative picture of the user is intelligently used as the premier picture for exposure display, improving the overall like rate of the user, increasing the probability of forming likes or matches between the matching parties, and reducing exposure waste.

[0081] The following is a simple example of another exemplary dynamic exploration process.

[0082] Use two pictures, picture IMG1 and picture IMG2 uploaded by the user, as alternative pictures for the premier picture. Among them, the popularity level output by the initial picture evaluation model for picture IMG1 is 0.5, and the popularity level output by the initial picture evaluation model for picture IMG2 is 0.4. Then the initialization parameters of the beta distribution corresponding to picture IMG1 are: α = 5, β = 5, denoted as Beta1(5,5); the initialization parameters of the beta distribution corresponding to picture IMG1 are: α = 4, β = 6, denoted as Beta2(4,6).

[0083] At time T, sample scores from Beta1(5,5) and Beta2(4,6) respectively to obtain scores p1 and p2. If p1 < p2, display picture IMG2 of the user, and vice versa. [[ID=ill]]

[0084] At time T, if the selected picture for display is picture IMG2 and the browsing user clicks "like" after exposure, update Beta2(4,6) to Beta2(5,6), otherwise update it to Beta2(4,7).

[0085] At time T, if the selected picture for display is picture IMG1 and the browsing user clicks "like" after exposure, update Beta2(5,5) to Beta2(6,5), otherwise update it to Beta2(5,6).

[0086] After calculating time T, calculate the updated scores p1 and p2. If p1 < p2, display picture IMG2 of the user, and vice versa.

[0087] Repeat the above steps until the popularity level of the alternative pictures of the user tends to be stable or the total number of display times reaches the preset value of 100 times.

[0088] Of course, the above simple example of the exemplary dynamic exploration process can also be applied to any number of alternative picture sets with more than two alternative pictures.

[0089] Figure 2This is a comparative diagram illustrating the benefits and effects of a first-image recommendation method according to an exemplary embodiment of the present invention. This exemplary embodiment randomly selects several groups of users online and applies the aforementioned first-image recommendation method and a traditional random exploration method, comparing the difference in the overall like rate of users as time and the number of times the image is displayed increases. The random exploration process can be configured with thresholds of 50, 100, 200, 400, etc., depending on the degree of exploration. These thresholds represent the number of attempts to randomly select images in the initial stage of exposure. Once the preset threshold is reached, the optimal image obtained from the current exploration is selected for subsequent exposure and display.

[0090] This exemplary embodiment involves two main processes to quickly explore and obtain the most popular images for users: a static cold start and a dynamic exploration. The static cold start is used to obtain the prior distribution of the popularity of each image, while the dynamic exploration is used to continuously update the posterior distribution using observed data.

[0091] Please continue to refer to Figure 2 The horizontal axis represents the number of exposures, and the vertical axis represents the average like rate. Figure 2 The like rate of the first image recommendation method in this embodiment and the traditional method using random exploration were compared. It can be observed that as the number of exposures increases, the traditional method, if it can conduct more exploration attempts in the early stage, has a higher final cumulative exposure like rate, but it is still lower than the first image recommendation method in this embodiment, and the random exploration process in the early stage causes greater loss to users.

[0092] Figure 3 This is a schematic block diagram of a first-figure recommendation device in some exemplary embodiments.

[0093] Please refer to Figure 3 A first image recommendation device includes:

[0094] Image acquisition unit 301, the image acquisition unit is configured to acquire at least two candidate images of the user.

[0095] Static unit 302, the static assignment unit is configured to determine the scores of the at least two candidate images.

[0096] Display unit 303 is configured to select one of the at least two candidate images as the first image to be displayed based on the score; and after the dynamic unit obtains a stable first image recommendation model, it determines the first image of the user based on the first image recommendation model.

[0097] Dynamic unit 304 is configured to adjust the user's image evaluation model based on the browsing user's feedback on the displayed candidate images, such that the score changes with the browsing user's feedback on the candidate images until the score of the candidate images stabilizes within a first threshold; and the image evaluation model whose score of the candidate images stabilizes within the first threshold is used as the user's first image recommendation model.

[0098] When acquiring images from a user's candidate images, the image acquisition unit 301 guides the user to upload multiple distinctive images as candidate images to expand the types and range of candidate images. For example, after a user registers and enters the application, the unit interactively guides the user to upload photos of various categories, such as lifestyle photos, sports photos, and photos of other types selected by the user during registration. As the user's historical behavior increases, the unit can also guide the user to supplement with new types of images based on their historical behavior characteristics.

[0099] Stranger social networking typically relies on comprehensive data matching and recommendation to guide users to upload a variety of distinctive images. The types of these images are related to their registration information or historical behavior to some extent, ensuring that there are enough high-quality images as alternatives during the initial image recommendation process, thus increasing the chances of being liked by browsing users who are matched with the data after exposure.

[0100] Static unit 302 uses an image evaluation model built based on images from active users on the platform to score the popularity of at least two candidate images from the user. The image evaluation model built based on active users on the platform employs regression modeling. The input to this regression model is: images uploaded by active users; the output of this regression model is: the popularity score of each of the aforementioned images. Any regression model can be used as the modeling model to determine the image evaluation model; this embodiment does not limit this, and the regression model is only used as a modeling method to determine the image evaluation model, while the image evaluation model itself is not limited to regression models.

[0101] As an exception, for users who have registered for a period of time and have only uploaded a single image, the popularity is determined by the user's average like rate over the most recent period, such as, but not limited to, the most recent 30 days.

[0102] Understandably, to improve the effectiveness of subsequent matching, it is possible to further limit the scoring to the popularity of the user's images by active users related to the user's data.

[0103] In some embodiments, the image evaluation model is controlled by at least a first parameter and a second parameter. These parameters change dynamically based on user feedback after the system goes live.

[0104] After determining the initial image evaluation model based on the above method, the initial image evaluation model is used to predictively score the candidate images selected by the user.

[0105] The dynamic unit 304 takes, for example, the image evaluation model controlled by at least the first parameter and the second parameter, which corresponds to the static unit: if the browsing user provides positive feedback on the candidate image displayed by the user, the first parameter is adjusted; if the browsing user does not provide positive feedback on the candidate image displayed by the user, the second parameter is adjusted.

[0106] Of course, to improve the effectiveness of subsequent matching, the aforementioned browsing users can be further limited to active users related to matching using user data.

[0107] Taking an image rating model based on the beta distribution as an exemplary embodiment, the beta distribution is a type of probability distribution, typically used to describe values ​​between 0 and 1, such as probability, ratio, or score. The parameters of the beta distribution can be estimated using methods such as maximum likelihood estimation or Bayesian inference. In practical applications, the beta distribution is often used to model the prior distribution of binomial distribution parameters, or as the posterior distribution of the probability of success in a Bernoulli trial. When the prior distribution is a beta distribution, its posterior distribution has the same form. The aforementioned properties of the beta distribution are used to predict and score candidate images.

[0108] Specifically, the probability density function of the beta distribution can be used to calculate the score of the candidate images.

[0109] The probability density function of the beta distribution is:

[0110]

[0111] Where α and β are distribution parameters that control the shape of the probability density function, x is a random variable, and B(α,β) is the beta function.

[0112] If the browsing user provides positive feedback on the candidate images selected by the user, the value of α in the probability density function of the beta distribution is increased; if the browsing user does not provide positive feedback on the candidate images selected by the user, the value of β in the probability density function of the beta distribution is increased.

[0113] As exposure increases rapidly, user feedback on the user's candidate images iterates quickly, leading to a gradual stabilization of the like rate and a stable beta distribution of the like rate for each candidate image. In other words, this can be understood as the image evaluation model for user-selected images, which is related to the user's data, stabilizing within the current time period. Using an image evaluation model where the score changes of candidate images remain stable within a first threshold within the current time period, we can serve as the user's first image recommendation model, thereby reducing unnecessary exposure waste and increasing the user's chances of being popular.

[0114] Because there is a high degree of randomness among active users, both as users and as browsing users, the use of user-uploaded images as candidate images is highly subjective. Even if data matching is introduced during the selection process of browsing users, there is still a possibility that browsing users may not like all the candidate images selected by the user.

[0115] In this scenario, if the first image is displayed more than the second threshold number of times, and the scores of the candidate images do not stabilize within the first threshold, then the image evaluation model updated after displaying the second threshold number of times is used as the first image recommendation model for the user. This approach, considering computational efficiency, ensures that even if the image evaluation model does not meet expectations, it can still optimize the recommendation effect of the first image to a certain extent without introducing excessive resource consumption.

[0116] Since the first image recommendation model is determined based on feedback from users browsing within a specific time period, user preferences may change over time. By updating the first image recommendation model at regular intervals, it can adapt to these changing preferences, better reflect current user preferences, and iterate effectively to avoid wasted exposure due to outdated feedback.

[0117] The method for updating the first image recommendation model is similar to the method for determining the first image recommendation model. Specifically, the first image recommendation model for the user is adjusted based on the browsing user's feedback on the candidate images displayed to the user, so that the score changes with the browsing user's feedback on the candidate images to the user, until the score change of the candidate images stabilizes within a third threshold. The first image recommendation model where the score change of the candidate images stabilizes within the third threshold is used as the updated first image recommendation model for the user.

[0118] When determining the first image, the display unit 303 may further perform statistical distribution calculations on the scores, using the calculated values ​​as the basis for displaying the first image. Symmetrical distribution models such as the normal distribution can be used as the statistical distribution calculation method, making the selection of the first image somewhat random around the score: candidate images with higher scores have a higher probability of being the first image, candidate images with medium scores have a medium probability of being the first image, candidate images with lower scores have a lower probability of being the first image, and the candidate image with the highest score is not guaranteed to be the first image.

[0119] By adding a method that uses statistical distribution calculations based on scores, rather than absolutely relying on score-based ranking for the first image display, the flexibility and accuracy of the first image recommendation are improved. This avoids situations where other potentially more popular images are not shown as the first image due to the first image not changing, thus missing out on better first image selection opportunities.

[0120] In some embodiments, the display probability of the score is calculated, and the candidate image with the highest display probability at the current moment is selected as the display image. If, based on the user's feedback on the display image at the current moment, the score of the display image becomes lower than that of the user's other candidate images, then the display image with a lower score than the user's other candidate images has a higher probability of having a lower display probability than the user's other candidate images.

[0121] When determining the main image, you can rely entirely on the main image recommendation model, or you can provide users with some freedom of choice based on the main image recommendation model.

[0122] For example, the candidate image with the highest score at the current moment can be selected as the display image. Alternatively, if the display image's score decreases below the user's other candidate images due to feedback from the browsing user at the current moment, then the candidate image with the highest score among the current display images will be selected as the display image for the next moment.

[0123] It is understood that the apparatus of this embodiment is used to perform, for example... Figure 1 The other implementation principles and technical effects of the aforementioned first image recommendation method are similar, and will not be elaborated here.

[0124] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the head-figure recommendation methods provided in this application.

[0125] Specifically, computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media, and memory devices, such as semiconductor memory devices (e.g., EPROM, EEPROM, and flash memory devices), magnetic disks (e.g., internal hard disks or removable disks), magneto-optical disks, and CD-ROM and DVD-ROM disks.

[0126] Specifically, the electronic device can be a mobile phone, tablet, computer, etc.

[0127] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for recommending a first image, characterized in that, include: Obtain at least two alternative images from the user; Determine the scores of the at least two candidate images; Based on the score, one of the at least two candidate images is selected as the first image to be displayed; The step of selecting one of the at least two candidate images as the first image to be displayed based on the score includes: calculating the display probability of the score, selecting the candidate image with the highest display probability at the current moment as the display image; if the display image's score is lower than the user's other candidate images due to the user's feedback on the display image at the current moment, then the display image with a lower score than the user's other candidate images has a higher probability of having a lower display probability than the user's other candidate images. The image evaluation model for the user is adjusted based on the feedback from browsing users to the displayed candidate images, such that the score changes with the feedback from browsing users to the candidate images, until the score of the candidate images stabilizes within a first threshold. This adjustment includes: if the browsing user provides positive feedback to the candidate images, increasing the alpha value in the probability density function of the beta distribution; if the browsing user does not provide positive feedback to the candidate images, increasing the beta value in the probability density function of the beta distribution. The image evaluation model whose score changes are stable within a first threshold is used as the first image recommendation model for the user; the image evaluation model uses the probability density function of beta distribution to calculate the scores of the at least two candidate images; The user's homepage image is determined based on the homepage recommendation model.

2. The first image recommendation method according to claim 1, characterized in that, If the first image is displayed more than the second threshold number of times, and the score of the candidate images does not stabilize within the first threshold, then the image evaluation model updated after the second threshold number of times is displayed is used as the first image recommendation model for the user.

3. The first image recommendation method according to claim 1, characterized in that, The process of determining the user's homepage image based on the homepage recommendation model includes: Based on the aforementioned first image recommendation model, the order of the candidate images selected by the user is adjusted, and the user selects any one of the sorted candidate images as the first image. Alternatively, the candidate image with the highest score based on the aforementioned first image recommendation model can be used as the first image.

4. The first image recommendation method according to claim 1, characterized in that, After a third time interval, the first image recommendation model is updated. The update method is as follows: the first image recommendation model of the user is adjusted based on the feedback of the browsing user to the candidate images displayed to the user, so that the score changes with the feedback of the browsing user to the candidate images of the user, until the score of the candidate images stabilizes within a third threshold. The first image recommendation model with the score of the candidate images stabilizing within the third threshold is used as the updated first image recommendation model of the user.

5. The first image recommendation method according to claim 1, characterized in that, An initial image evaluation model is established based on the popularity of images from active users on the platform, and the scores of the at least two candidate images are determined using the initial image evaluation model.

6. A first image recommendation device, characterized in that, include: The image acquisition unit is configured to acquire at least two candidate images from the user. A static unit, configured to determine the scores of the at least two candidate images; A display unit is configured to select one of the at least two candidate images as the first image to be displayed based on the score; The step of selecting one of the at least two candidate images as the first image to be displayed based on the score includes: calculating the display probability of the score, selecting the candidate image with the highest display probability at the current moment as the display image; if the display image's score is lower than the user's other candidate images due to the user's feedback at the current moment, then the display image with a lower score than the user's other candidate images has a higher probability of having a lower display probability than the user's other candidate images; and after obtaining a stable first image recommendation model in the dynamic unit, determining the user's first image based on the first image recommendation model; A dynamic unit is configured to adjust the user's image evaluation model based on the browsing user's feedback on the displayed candidate images, such that the score changes with the browsing user's feedback on the candidate images until the score of the candidate images stabilizes within a first threshold. The adjustment of the user's image evaluation model based on the browsing user's feedback on the displayed candidate images, such that the score changes with the browsing user's feedback on the candidate images, includes: if the browsing user provides positive feedback on the candidate images, increasing the alpha value in the probability density function of the beta distribution; if the browsing user does not provide positive feedback on the candidate images, increasing the beta value in the probability density function of the beta distribution; and using the image evaluation model where the score of the candidate images stabilizes within the first threshold as the user's first image recommendation model; the image evaluation model uses the probability density function of the beta distribution to calculate the scores of the at least two candidate images.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that... When the program is executed by the processor, it implements the steps of the first image recommendation method according to any one of claims 1-5.

Images