Content recommendation method, electronic device and storage medium

Abstract

The present application provides a content recommendation method, an electronic device, and a storage medium, the method comprising: the electronic device displays a first interface, the first interface includes a first control for displaying recommended content; when it is determined according to a first score that the first content meets a preset display condition, the first content is displayed in the first control within a preset time period, wherein the preset time period is a time period for displaying the first content in a preset display scheme, the first score is used to characterize the recommendation probability of the first content, the first score is generated using historical data of the user within a preset historical time length, and the historical data is used to reflect the user's historical feedback behavior on the first content displayed in the first control. The embodiment of the present application can recommend and display the first content that meets the user's needs based on the user's feedback on the first content in the historical recommendation process, thereby improving the recommendation accuracy and the user's recommendation experience.

Description

Technical Field

[0001] The present application relates to the technical field of electronic devices, and in particular to a content recommendation method, electronic device and storage medium. Background Art

[0002] With the continuous development of electronic devices, electronic devices provide users with a large amount of applications, information, services and other contents. However, when users need to use the above contents, they need to search for the required contents among the large amount of contents provided by the electronic devices. For example, since there may be a large number of applications on the desktop of the electronic device, users need to search page by page when searching, which brings inconvenience to users.

[0003] In order to facilitate users to find content, electronic devices will recommend content on the desktop. However, in the prior art, the content such as applications, information and services frequently used by users are often displayed on the desktop, and the recommendation accuracy of this recommendation method is low. Summary of the invention

[0004] In order to solve the above technical problems, the present application provides a content recommendation method, an electronic device and a storage medium. In this method, the electronic device can recommend and display the first content that meets the user's needs based on the user's feedback on the first content in the historical recommendation process, thereby improving the recommendation accuracy and the user's recommendation experience.

[0005] In a first aspect, the present application provides a content recommendation method. The content recommendation method is applied to an electronic device, and the content recommendation method includes: displaying a first interface, the first interface includes a first control, and the first control is used to display content recommended to the user. When the first score meets the preset display condition, the first content is displayed in the first control within a preset time period. The preset time period is a time period for displaying the first content in a preset display scheme. The first score is used to characterize the recommendation probability of the first content, and the first score is generated using the user's historical data within a preset historical time period, and the historical data is used to reflect the user's historical feedback behavior on the first content displayed in the first control.

[0006] In the embodiment of the present application, since the first score is generated based on historical data, the historical data can reflect the user's historical feedback behavior on the first content displayed in the first control, so the recommendation probability of the first content can be estimated based on the user's feedback on the first content in the recommended display state. In addition, since the first content is displayed when the first score meets the preset display condition, the electronic device can recommend and display the first content that meets the user's needs based on the user's feedback on the first content in the historical recommendation process, and not display the first content that does not meet the needs, that is, display suppression, thereby improving the recommendation accuracy and improving the user's recommendation experience.

[0007] Exemplarily, the first page may be a home page, a negative one screen page, a screen-off page, a lock screen page, or the like that may display recommended content.

[0008] Exemplarily, the first content may be a recommended application, a recommended card, a recommended service, recommended information, etc.

[0009] For example, the first control may be a recommended folder, such as a YOYO recommended folder. For another example, the first control may also be a display control for an application icon, a display control for a recommended card, a display control for a recommended service, a display control for recommended information, etc., without specific limitation.

[0010] Exemplarily, the historical data may be historical exposure data, and the historical data may include user feedback behavior data and exposure behavior data. The user feedback behavior data is data generated when the user performs feedback operation on the first content in the exposure state. The exposure behavior data is data generated by the exposure behavior (display behavior) of the first content. For example, the historical data may include click time, removal time, and exposure time. The click time is the time when the user clicks on the first content, the removal time is the time when the user removes the first content, and the exposure time is the time when the first content begins to be exposed.

[0011] Exemplarily, the historical data may be the historical data of the electronic device itself, to ensure that the present solution can be executed purely on the terminal side, thereby ensuring the security and privacy of the data, and ensuring that personalized exposure suppression can be performed based on the user of the electronic device, thereby improving the flexibility of exposure suppression.

[0012] In another exemplary embodiment, the historical data may be obtained by the electronic device from the cloud. In another exemplary embodiment, the first score may be a score that can reflect the display probability of the first content, which may be a Q value score or other scores.

[0013] According to the first aspect, the first score is a Q-value score. Before displaying the first content in the first control within a preset time period, the method also includes: obtaining historical data; determining, based on the historical data, a feedback behavior sequence of the user within a preset historical time length, the feedback behavior sequence including multiple feedback behaviors, and the feedback behavior is the user's operation behavior on the first content displayed in the first control; using the feedback behavior sequence to perform Q-value learning to obtain a Q-value score, the Q-value score including the Q-value of at least one display behavior of the first content; and judging whether the first content meets the preset display conditions based on the Q-value score.

[0014] In this way, the first score can be trained through Q-value learning, and the exposure probability of the first content can be accurately estimated according to discrete feedback behaviors, thereby improving the recommendation accuracy and facilitating implementation.

[0015] Exemplarily, the electronic device may divide k feedback behaviors, for example, k may be 3, and the feedback behaviors may include a click behavior, a remove behavior, and a no-operation behavior.

[0016] Exemplarily, the Q-value score may include displayed Q-values ​​and non-displayed Q-values. In one embodiment, the Q-value may be implemented as a 1*2 matrix.

[0017] In another example, the electronic device can first train to obtain a k*2 Q value matrix, where k is the number of feedback behaviors. For example, if the feedback behaviors include click behaviors, remove behaviors, and no operation behaviors, the Q value score can be implemented as a 3*2 matrix. Then, the k Q values ​​in the first column are summed to obtain the Q value of the display behavior. And, the k Q values ​​in the second column are summed to obtain the Q value of the non-display behavior.

[0018] According to the first aspect, or any implementation method of the first aspect above, the Q-value score includes the Q-values ​​of multiple display behaviors of the first content; using the feedback behavior sequence to perform Q-value learning to obtain the Q-value score, including: taking the feedback behaviors in the feedback behavior sequence as the current feedback behavior in sequence, and executing the following steps: determining the current display behavior of the first content among multiple display behaviors; based on preset reward point data, obtaining the reward point value corresponding to the current behavior combination, the current behavior combination is a combination of the current feedback behavior and the current display behavior; obtaining the Q-value corresponding to the current display behavior; updating the Q-value of the current display behavior according to the reward point value to obtain the updated Q-value of the current exposure behavior; after the Q-value update of the last feedback behavior is completed, obtaining the Q-value score.

[0019] In this way, the Q-value score can be accurately calculated by updating the Q-value score through the feedback behavior sequence, so that the calculated Q-value score can accurately estimate the display probability and non-display probability of the first content, and then exposure suppression according to this method can further improve the accuracy.

[0020] For example, when calculating the Q value, the feedback behavior can be used as the state in the Q value learning, and the display behavior can be used as the action in the Q value learning.

[0021] Exemplarily, the reward score value and the Q value before the update can be substituted into the Q value calculation function to calculate the updated Q value.

[0022] In one example, the Q value calculation function may include the following formula (1):

[0023] (1)

[0024] In another example, the Q value calculation function may include the following formula (2):

[0025] (2)

[0026] According to the first aspect, or any implementation method of the first aspect above, the Q value score includes a first Q value and a second Q value, wherein the first Q value is a displayed Q value and the second Q value is a non-displayed Q value, and judging whether the first content meets the preset display condition according to the Q value score includes: obtaining a preset judgment threshold; judging whether the first difference is less than the preset judgment threshold, wherein the first difference is the difference between the first Q value and the second Q value; judging whether the first Q value is less than the second Q value, and whether the first Q value is less than the preset threshold; when the first difference is less than the preset judgment threshold, if the first Q value is less than the second Q value, and the first Q value is less than the preset threshold, it is determined that the first content does not meet the preset display condition; when the first difference is greater than or equal to the preset judgment threshold, if the first Q value is greater than or equal to the second Q value or the first Q value is greater than or equal to the preset judgment threshold, it is determined that the first content meets the preset display condition.

[0027] In this way, it is possible to accurately determine whether the first content meets the user's recommendation requirements through the preset display conditions, thereby being able to recommend content that meets the user's recommendation requirements and suppress content that does not meet the user's requirements.

[0028] Exemplarily, the preset judgment threshold may be less than 0 to prevent excessive interception of the original exposure scheme.

[0029] According to the first aspect, or any implementation of the first aspect above, the historical data includes at least one of a first moment, a second moment and a third moment, wherein the first moment is the moment when the user performs a click operation on the first content, the second moment is the moment when the user performs a removal operation on the first content, and the third moment is the moment when the first content is displayed; according to the historical data, determining the user's feedback behavior sequence within a preset historical time period, including: dividing the preset historical time period into multiple time periods; based on the historical data, counting the first number, the second number and the third number of each time period, wherein the first number is the number of times the first content is displayed, the second number is the number of times the user clicks on the first content, and the third number is the number of times the user removes the first content; determining the feedback behavior of each time period according to the first number, the second number and the third number; based on the feedback behaviors of multiple time periods, obtaining a feedback behavior sequence, wherein the feedback behaviors of multiple time periods are sorted in chronological order in the feedback behavior sequence.

[0030] In this way, the feedback behavior of each time period can be accurately determined by dividing the time period, thereby improving the accuracy of exposure suppression.

[0031] Exemplarily, the first moment may be a click moment, the second moment may be a removal moment, and the third moment may be a display moment or an exposure moment.

[0032] Exemplarily, the first number may be the number of clicks, the second number may be the number of removals, and the third number may be the number of exposures or the number of displays.

[0033] According to the first aspect, or any implementation method of the first aspect above, the feedback behavior of each time period is determined according to the first number, the second number and the third number, including: for each time period, when the number of clicks is greater than or equal to 1, a first label is set for each time period, and the first label is used to mark the user's click behavior on the first content; when the second number is greater than or equal to 1, a second label is set for each time period, and the second label is used to mark the user's removal behavior on the first content; when the first number and the second number are equal to 0, and the third number is greater than or equal to 1, a third label is set for each time period, and the third label is used to mark the user's no operation on the first content.

[0034] In this way, the feedback behavior in each time period can be accurately determined according to the number of times, thereby improving the accuracy of subsequent Q value calculation and exposure suppression.

[0035] Exemplarily, the first tag may be a click tag, the second tag may be a remove tag, and the third tag may be an exposure tag.

[0036] Exemplarily, after obtaining multiple feedback behavior tags (collectively referred to as the first tag, the second tag, and the third tag), the feedback behavior tags can be sorted in chronological order to obtain a feedback state sequence. Then, the tags of the feedback state sequence are read, and when the first tag is read, a click behavior is added to the feedback behavior sequence; when the second tag is read, a remove behavior is added to the feedback behavior sequence; when the third tag is read, a no-operation behavior is added to the feedback behavior sequence.

[0037] According to the first aspect, or any implementation of the first aspect above, the feedback behavior includes at least one of a click behavior, a remove behavior, and no operation; and the display action includes display and non-display.

[0038] In this way, the user's feedback on the first content in the recommended state can be accurately summarized into three types: click, remove and no operation, and the display action of the first content can be accurately summarized into two types: display and non-display. On this basis, since the feedback behavior can characterize the user's demand for the first content, the Q value calculated based on the feedback behavior can be used to estimate the display probability based on the user's demand for the first content, thereby improving the recommendation accuracy.

[0039] According to the first aspect, or any implementation method of the first aspect above, the historical data includes historical data of multiple scenes; when the first score meets the preset display condition, the first content is displayed in the first control within the preset time period, including: obtaining a preset display scheme, the preset display scheme includes a display time period for each scene, and the preset time period for each scene is the display time of the first content in the first control under each scene; for each scene, according to the historical data of each scene, obtaining the first score of each scene; when the first score of each scene meets the preset display condition, in the preset display scheme, retaining the preset time period of each scene; when the first score of each scene does not meet the preset display condition, deleting the preset time period of each scene from the preset display scheme; displaying the first content in the first control during the preset time period in the adjusted preset display scheme.

[0040] In this way, after generating the preset display scheme, the electronic device can adjust the preset display scheme through the first score without changing the generation process of the exposure scheme, thereby improving the quick plug-in and pull-out characteristics of the scheme and reducing the use and development costs of the scheme.

[0041] For example, the scene may include a home scene, a company scene, and an all scene, wherein the all scene may also be referred to as a no-specific scene.

[0042] According to the first aspect, or any implementation method of the first aspect above, displaying the first content in the first control during a preset time period in an adjusted preset display scheme includes: obtaining the current scene of the electronic device; obtaining the preset time period of the current scene in the adjusted preset display scheme; and displaying the first content in the first control when the current moment is within the preset time period of the current scene.

[0043] In this way, exposure suppression can be performed in different scenarios according to the scheme after exposure suppression, so that the scheme of the embodiment of the present application can be applied to different scenarios, and different exposure suppression can be performed for different scenarios, which improves the flexibility of the scheme and further improves the user experience.

[0044] According to the first aspect, or any implementation of the first aspect above, the method further includes: if the first score does not meet a preset display condition, not displaying the first content in the first control within a preset time period.

[0045] According to the first aspect, or any implementation of the first aspect above, it is detected whether the first content has a fourth tag; if the first content does not have the fourth tag, and if the first score meets a preset display condition, the first content is displayed in the first control within a preset time period.

[0046] In this way, the technical solution of the embodiment of the present application can be executed on the first content that needs exposure suppression by setting labels in advance, and the first content that does not need exposure can be directly exposed according to the original exposure plan, thereby improving the flexibility and processing speed of the exposure plan.

[0047] According to the first aspect, or any implementation of the first aspect above, after detecting whether the first content has a fourth tag, the method further includes:

[0048] In a case where the first content has a fourth tag, the first content is displayed in the first control within a preset time period.

[0049] In a second aspect, an embodiment of the present application provides an electronic device, comprising: one or more processors, one or more memories; the one or more memories are coupled to the one or more processors, the one or more memories are used to store computer program code, the computer program code includes computer instructions, and when the one or more processors execute the computer instructions, the electronic device executes the method of the first aspect.

[0050] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, including computer instructions, which, when executed on an electronic device, causes the electronic device to execute the method of the first aspect.

[0051] In a fourth aspect, an embodiment of the present application provides a chip or a chip system, including a processing circuit and an interface circuit, the interface circuit is used to receive code instructions and transmit them to the processing circuit, and the processing circuit is used to run the code instructions to execute the method of the first aspect.

[0052] In a fifth aspect, an embodiment of the present application provides a computer program product, which includes a computer program, and when the computer program is run on a computer, it enables the computer to execute the method of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic diagram showing an application recommendation scenario provided by an embodiment of the present application is shown;

[0054] Figure 2A-2B A schematic diagram showing a set of exemplary application recommendation scenarios provided by an embodiment of the present application;

[0055] Figure 3 A schematic diagram of a main page of an electronic device is shown as an example;

[0056] Figure 4A-4B A schematic diagram showing a set of exemplary user interfaces provided in an embodiment of the present application;

[0057] Figure 5A-5CA schematic diagram showing another exemplary user interface provided in an embodiment of the present application;

[0058] Figure 6A-6E A schematic diagram showing another set of exemplary user interfaces provided in an embodiment of the present application;

[0059] Figure 7 A schematic diagram of an exemplary negative one screen page provided in an embodiment of the present application is shown;

[0060] Figure 8 A schematic diagram showing an exemplary lock screen page provided in an embodiment of the present application;

[0061] Fig. 9 A schematic diagram of the structure of an electronic device is shown;

[0062] Fig.10 is a software structure block diagram of the electronic device of the embodiment of the present application;

[0063] Fig.11 A recommended exposure flow chart in a related art is shown;

[0064] Fig.12 A logical schematic diagram showing an exemplary application exposure in the related art;

[0065] Fig.13 A logical schematic diagram showing another exemplary application exposure in a related art;

[0066] Fig.14 A schematic diagram of a process of a recommended exposure scheme provided in an embodiment of the present application is shown;

[0067] Fig.15 A logical schematic diagram of an exemplary application exposure provided by an embodiment of the present application is shown;

[0068] Fig.16 A logical schematic diagram of another exemplary application exposure provided by an embodiment of the present application is shown;

[0069] Fig.17 A schematic diagram of a data preprocessing method provided in an embodiment of the present application is shown;

[0070] Fig.18 A schematic diagram showing an exemplary historical exposure data provided by an embodiment of the present application;

[0071] Fig.19 A schematic diagram of a flow chart of an exposure model training method provided in an embodiment of the present application is shown;

[0072] Fig. 20 An exemplary Q-value learning schematic diagram provided in an embodiment of the present application is shown;

[0073] Fig.21 A schematic diagram of a process flow of an exposure scheme processing method provided in an embodiment of the present application is shown;

[0074] Fig. 22 A schematic diagram of generating an exemplary target exposure scheme provided in an embodiment of the present application is shown;

[0075] Fig.23 A schematic diagram showing an exemplary Q value prediction provided by an embodiment of the present application is shown;

[0076] Fig.24 A schematic diagram showing another application recommendation method provided in an embodiment of the present application is shown;

[0077] Fig.25 A schematic block diagram of a device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0078] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0079] The term "and / or" in this article is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone.

[0080] The terms "first" and "second" in the description and claims of the embodiments of the present application are used to distinguish different objects rather than to describe a specific order of objects. For example, a first target object and a second target object are used to distinguish different target objects rather than to describe a specific order of target objects.

[0081] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0082] In the description of the embodiments of the present application, unless otherwise specified, the meaning of "multiple" refers to two or more than two. For example, multiple processing units refer to two or more processing units; multiple systems refer to two or more systems.

[0083] With the gradual development of the diversification and intelligence of electronic devices, electronic devices such as mobile phones provide users with a large number of applications, information, services and other content. With the increase of the above content, users need to search for the required content item by item from a large amount of content, which increases the difficulty of users. For example, when users search for applications, since there are a large number of applications on the desktop of the electronic device, users need to search page by page and folder by folder.

[0084] Therefore, in order to improve the user's convenience, the electronic device can recommend applications, information or services to the mobile phone user so that the user can quickly find the required content. Figure 1 FIG. 1 is a schematic diagram showing an application recommendation scenario provided by an embodiment of the present application. Figure 1 As shown, when user P1 wants to open the sports health application during running, the mobile phone can display the application icon A1 of the application health in the recommended folder 11 of the homepage. In this way, after opening the mobile phone, user P1 can quickly find the application icon A1 of the application health on the homepage without having to search page by page or application by application, thereby improving the user experience.

[0085] Figure 2A-2B A schematic diagram showing a set of exemplary application recommendation scenarios provided by an embodiment of the present application is shown. In some embodiments, the recommended applications exposed to the mobile phone at different times may be different. Figure 2A As shown, the recommended applications in the recommended folder 11a at 14:00 on a certain day and the recommended folder 11b at 18:00 on the same day may be different (they may be completely different or partially different, and there is no specific limitation on this). It should be noted that when the exposure scheme is not changed, the recommended applications exposed in the recommended folder 11a and the recommended folder 11b may also be the same, and there is no specific limitation on this. In other embodiments, the recommendations exposed by the mobile phone in different scenarios may be different. Figure 2B As shown, taking 14:00 on a certain day as an example, if the mobile phone is in the office scene, the mobile phone homepage displays the recommended folder 11c, through which the mobile phone can recommend to the user application icons of the first category of applications (applications related to office) that may be used in the office scene, such as office, browser, email, conference, etc. Also, if the mobile phone is in the home scene, the mobile phone homepage can display the recommended folder d, through which the mobile phone can recommend to the user application icons of the second category of applications (applications related to life and entertainment) that may be used in the home scene, such as video, chat tool A, takeaway, game, etc.

[0086] It should be noted that the above part of the embodiment of the present application is described using application recommendation as an example. It can be understood that the above recommendation method is also applicable to content recommendation scenarios such as card recommendation and information recommendation, and will not be elaborated on this.

[0087] For ease of understanding, the recommendation scenarios involved in the embodiments of the present application are described below using multiple groups of user interface schematic diagrams as an example. It is understandable that the following multiple groups of user interfaces merely illustrate a feasible user interface style for an electronic device taking a mobile phone as an example, and should not constitute a limitation on the embodiments of the present application. It should be noted that in the specific implementation of the embodiments of the present application, the electronic device can also be implemented as a computer, smart phone, tablet computer, smart watch, or other device with a content recommendation function, and the present application does not limit this.

[0088] In some embodiments of the present application, the home page of the electronic device may display recommended content. For example, Figure 3 FIG. 1 is a schematic diagram of a main page of an electronic device. Figure 3 As shown, the main page 10 may include a recommendation folder 11 , a recommendation card component 12 , a status bar 13 , a page indicator 14 , and a plurality of application icons 15 .

[0089] Specifically, the recommended folder 11 is used to place an application icon 111 of at least one recommended application. Among them, the recommended application may refer to an application recommended by the electronic device to the user, wherein each recommended application may be displayed in a preset application display area in the recommended folder. Exemplarily, the recommended folder may be called a YOYO recommended folder. In an embodiment of the present application, when a certain application needs to be exposed, the application icon of the application may be displayed in the recommended folder 11. It should be noted that the embodiment of the present application does not impose specific restrictions on the specific application type of the recommended application exposed in the recommended folder 11. Also, the embodiment of the present application does not impose specific restrictions on the number of recommended applications in the recommended folder 11. It should be noted that in an embodiment of the present application, the recommended folder 11 may also place other recommended content, such as icons of recommended services (i.e., services recommended by the electronic device to the user), etc., and no specific restrictions are imposed on this.

[0090] In the embodiment of the present application, the user can quickly open the application by clicking. Figure 4A-4B Provide explanation. Figure 4A-4B FIG. 1 is a schematic diagram showing a group of exemplary user interfaces provided by an embodiment of the present application. Figure 4A As described above, when the user clicks the application icon of the music application in the recommended folder 11 on the main page 10 of the electronic device, the electronic device can open the music application and display the following information: Figure 4BThe main page 20 of the music application shown in the figure allows the user to quickly open and use the music application. It should be noted that as the number of applications installed in the electronic device increases, through the application recommendations in the recommendation folder 11, the user can quickly open the required application without having to search page by page on the desktop, thereby improving the user experience.

[0091] In the embodiment of the present application, the user can also remove the recommended application by removing the application. Figure 5A-5C Provide explanation. Figure 5A-5C FIG. 2 shows another exemplary user interface provided by an embodiment of the present application. Figure 5A As shown, after the user performs a setting trigger operation for the reading application (i.e., an operation for triggering the relevant setting content for the selected application, such as a long press operation, etc., or can also be set to a double-click operation according to actual needs, without limitation), the electronic device can display Figure 5B The operation interface 30 shown. Figure 5B As shown, the operation interface 30 displays a recommended folder 11 and a plurality of setting options 112 for reading applications. After the user performs a removal operation (such as a click operation or other operation for removing a recommended application) for the "not interested" control, the electronic device may display Figure 5C The main interface described above. Figure 5A and Figure 5C It can be seen from the main interface that the user has deleted (or removed) the reading application from the recommended folder 11. It should be noted that the removal operation can be used to delete the reading application from the recommended folder 11, but may not be removed from the electronic device.

[0092] The recommendation card component 12 may include a recommendation card 121. The recommendation card 121 is used to display information that the user may need to view, such as Figure 3 The user's schedule card 121a shown can display the user's schedule information. For another example, the information displayed on the recommendation card 121 can also be flight information, train / bus / subway / taxi and other travel information, navigation information, express information, hotel booking information, music playback information, sports and health information, device power information, weather forecast information, etc. It should be noted that the recommendation card 121 can also display other information or services that the user needs to know, and there is no specific limitation on this. In some embodiments, the electronic device can generate one or more recommendation cards through methods such as user active addition and automatic addition when a card service is triggered. When an electronic device generates multiple recommendation cards, the multiple recommendation cards can be displayed superimposed in a card display area. Continue to see Figure 3, a recommended card is displayed on the homepage 10, and other recommended cards can be displayed by sliding. Alternatively, the homepage 10 of the electronic device can have two or more display areas. For example, two display areas can be arranged side by side on the homepage 10 of the electronic device.

[0093] Furthermore, when multiple recommendation cards are generated on the electronic device, the recommendation card component 12 may further include a card indicator 122. The card indicator 122 may be used to indicate the positional relationship between the currently displayed recommendation card and other recommendation cards.

[0094] In an embodiment of the present application, the user can quickly switch to the service page or application page corresponding to the recommended card by clicking. For example, after the user clicks on the taxi card, the electronic device can switch to display the service page of the travel application.

[0095] And, in the embodiment of the present application, the user can remove the recommended card 121 through the removal operation. Figure 6A-6E Provide explanation. Figure 6A-6E FIG. 2 shows another schematic diagram of an exemplary user interface provided by an embodiment of the present application. Fig. 6A As shown, when the user performs a setting trigger operation (i.e., an operation for triggering the setting content related to the selected card, such as a long press operation, or a double click operation according to actual needs, which is not limited) on the schedule card 121a (i.e., the currently displayed recommended card), the electronic device can display Figure 6B The operation interface 40 shown. Figure 6B As shown, the operation interface may display operation controls for removing schedule cards 121a, such as operation controls 123 for removing a "Schedule" card, and the operation interface may also display operation controls 124 for editing recommended card sets (a collection of recommended cards generated by an electronic device), such as operation controls for YOYO suggested settings, operation controls for editing default styles, operation controls for removing card sets, etc., without specific restrictions.

[0096] Continue to see Figure 6B When the user performs a preset trigger operation (such as a single click) on the operation control 123, the electronic device may display Figure 6C The main page is shown. Fig. 6A and Figure 6C The difference between the main page shown is that after the user removes the schedule card from the card set, the recommended card 121 currently displayed by the electronic device is switched from the schedule card 121a to the high-speed rail travel card 121b (which can be the next recommended card after the schedule card).

[0097] And, it should be noted that, continue to refer to Fig.6D, if the user is Figure 4B After the operation control 124a for removing the card set on the operation interface 40 shown in the figure performs a preset trigger operation (such as a single click operation), the electronic device can remove all the cards in the card set and display Fig. 6E The main page is shown. Fig. 6E The main page shown is Fig. 6A The difference of the main page shown is that no cards are displayed in the preset card display area (ie, the area for displaying recommended cards) of the main page, and the current time information can be displayed.

[0098] The status bar 13 may include one or more signal strength indicators of a mobile communication signal (also referred to as a cellular signal), a wireless fidelity (Wi-Fi) signal strength indicator, a battery status indicator, a time indicator, and the like.

[0099] The page indicator 14 can be used to indicate the positional relationship between the currently displayed page and other pages. Specifically, the page indicator 14 can be used to indicate which page the user is currently browsing is one of the multiple pages carrying multiple application icons 113. The user can swipe left and right to browse other pages.

[0100] Each application icon 15 corresponds to an application. It should be noted that the application icons 15 can be distributed on multiple pages, and the user can browse the application icons on other pages by sliding left and right. In addition, the application icon 15 can be located in the bottom fixed bar at the bottom of the interface, or it can be directly located in the application icon display area in the middle of the page, or in the application folder in the application icon display area, and its display position is not specifically limited.

[0101] It should be noted that, in addition to displaying recommended content in the above-mentioned recommended folders and preset card display areas on the main page, the electronic device can also display recommended content in locations such as the Smart Island on the main page, and there is no specific limitation on this.

[0102] In other embodiments of the present application, the electronic device may also display recommended content through the negative first screen. Figure 7 FIG. 1 shows a schematic diagram of an exemplary negative one screen page provided in an embodiment of the present application. Figure 7 As shown, the recommended content (such as YOYO recommendation) displayed in the negative one screen page 50 may include an icon 51 of a recommended service, a recommendation card 52 , and recommended news information 53 .

[0103] The recommended service is used to recommend an application or a service in an application (such as a payment service, a scan service, etc.) to the user in the form of an icon. Accordingly, the recommended service icon 51 can be an application icon or an icon of a service in an application, such as a shopping software icon, a camera icon, a scan icon, or a payment code icon, etc., without specific restrictions. It should be noted that the icons of each recommended service can be displayed in the preset service display area (i.e., the area for displaying the icons of the recommended services) of the recommended service column.

[0104] Among them, the recommendation card 52 is used to recommend content to users in the form of a card, such as a music card (displaying music playback information), a courier card (displaying courier information), a sports health card (displaying sports health information), a device power information card (displaying device power information, etc., or it can also be a recommendation of other information of the electronic device or an external device of the electronic device), a weather forecast card (displaying weather forecast information), etc. The specific form can be found in the relevant description of the recommendation card in the above part of the embodiment of the present application, and will not be repeated here.

[0105] The recommended information 53 is used to recommend news information to the user. Each information can be displayed in a preset information display area (i.e., an area for displaying information) on the negative one screen.

[0106] It should be noted that in the embodiments of the present application, the electronic device can also recommend content in forms other than application icons and cards, such as recommending quick applications, etc., and there is no specific limitation on this.

[0107] Also, it should be noted that in the embodiment of the present application, in addition to the recommended exposure on the main page and the negative one screen, the recommended content can also be displayed on the lock screen page, the screen off page, the notification display page, the control center page, etc. For example, Figure 8 FIG. 1 shows a schematic diagram of an exemplary lock screen page provided in an embodiment of the present application. Figure 8 As shown, a schedule card 61 may be displayed on the lock screen interface 61. It should be noted that the embodiment of the present application does not specifically limit the display position and display form of the recommendation card on the lock screen page.

[0108] After a preliminary introduction to the content recommendation scenario of the embodiment of the present application, the structure of the electronic device will be described next.

[0109] Fig. 9 900 is a schematic diagram showing the structure of an electronic device. It should be understood that Fig. 9The electronic device 900 shown is only one example of an electronic device, and the electronic device 900 may have more or fewer components than those shown in the figure, may combine two or more components, or may have a different component configuration. Fig. 9 The various components shown in the EMBODIMENTS 2000 may be implemented in hardware, software, or a combination of hardware and software, including one or more signal processing and / or application specific integrated circuits.

[0110] The electronic device 900 may include: a processor 910, an external memory interface 920, an internal memory 921, a Universal Serial Bus (USB) interface 930, a charging management module 940, a power management module 941, a battery 942, an antenna 1, an antenna 2, a mobile communication module 950, a wireless communication module 960, an audio module 970, a speaker 970A, a receiver 970B, a microphone 970C, an earphone interface 970D, a sensor module 980, a button 990, a motor 991, an indicator 992, a camera 993, a display screen 994, and a Subscriber Identification Module (SIM) card interface 995, etc. The sensor module 980 may include a pressure sensor, a gyroscope sensor, an air pressure sensor, a magnetic sensor, an acceleration sensor, a distance sensor, a proximity light sensor, a fingerprint sensor, a temperature sensor, a touch sensor, an ambient light sensor, a bone conduction sensor, etc.

[0111] The processor 910 may include one or more processing units, for example: the processor 910 may include an application processor (AP), a modem processor, a graphics processor (GPU), an image signal processor (ISP), a controller, a memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processor (NPU). Among them, different processing units can be independent devices or integrated in one or more processors. Among them, the controller can be the nerve center and command center of the electronic device 900. The controller can generate an operation control signal according to the instruction operation code and the timing signal to complete the control of fetching and executing instructions.

[0112] A memory may also be provided in the processor 910 for storing instructions and data. In some embodiments, the memory in the processor 910 is a high-speed cache memory. The USB interface 930 is an interface that complies with USB standard specifications, and may specifically be a Mini USB interface, a Micro USB interface, a USB Type C interface, etc. The USB interface 930 may be used to connect a charger to charge the electronic device 900, or may be used to transfer data between the electronic device 900 and a peripheral device. It may also be used to connect headphones to play audio through the headphones. The interface may also be used to connect other electronic devices, such as AR devices, etc.

[0113] The charging management module 940 is used to receive charging input from a charger. The charger may be a wireless charger or a wired charger. In some wired charging embodiments, the charging management module 940 may receive charging input from a wired charger through the USB interface 930. In some wireless charging embodiments, the charging management module 940 may receive wireless charging input through a wireless charging coil of the electronic device 900. While the charging management module 940 is charging the battery 942, it may also power the electronic device through the power management module 941.

[0114] The power management module 941 is used to connect the battery 942, the charging management module 940 and the processor 910. The power management module 941 receives input from the battery 942 and / or the charging management module 940, and supplies power to the processor 910, the internal memory 921, the external memory, the display screen 994, the camera 993, and the wireless communication module 960. The wireless communication function of the electronic device 900 can be implemented through the antenna 1, the antenna 2, the mobile communication module 950, the wireless communication module 960, the modem processor and the baseband processor.

[0115] Antenna 1 and antenna 2 are used to transmit and receive electromagnetic wave signals. Each antenna in electronic device 900 can be used to cover a single or multiple communication frequency bands. Different antennas can also be reused to improve the utilization of antennas. For example, antenna 1 can be reused as a diversity antenna for a wireless local area network. In some other embodiments, the antenna can be used in combination with a tuning switch.

[0116] The mobile communication module 950 may provide solutions for wireless communications including 2G / 3G / 4G / 5G, etc., applied to the electronic device 900. The mobile communication module 950 may include at least one filter, a switch, a power amplifier, a low noise amplifier (Low Noise Amplifier, LNA), etc.

[0117] The wireless communication module 960 can provide wireless communication solutions for application in the electronic device 900, including wireless local area networks (WLAN) (such as wireless fidelity (Wi-Fi) networks), Bluetooth (BT), global navigation satellite system (GNSS), frequency modulation (FM), near field communication technology (NFC), infrared technology (IR), etc.

[0118] In some embodiments, antenna 1 of electronic device 900 is coupled to mobile communication module 950, and antenna 2 is coupled to wireless communication module 960, so that electronic device 900 can communicate with the network and other devices through wireless communication technology.

[0119] The electronic device 900 implements the display function through a GPU, a display screen 994, and an application processor. The GPU is a microprocessor for image processing, which connects the display screen 994 and the application processor. The GPU is used to perform mathematical and geometric calculations for graphics rendering. The processor 910 may include one or more GPUs, which execute program instructions to generate or change display information.

[0120] The display screen 994 is used to display images, videos, etc. The display screen 994 includes a display panel. The display panel can be a liquid crystal display (LCD), an organic light-emitting diode (OLED), etc. In some embodiments, the electronic device 900 may include 1 or M display screens 994, where M is a positive integer greater than 1. In the embodiment of the present application, the electronic device 900 can display the display capability provided by the display screen 994. Figure 1 , Figure 2A-2B , Figure 3 , Figure 4A-4B , Figure 5A-5C , Figure 6A-6E , Figure 7 , Figure 8 The user interface shown includes recommended folders, application icons of recommended applications, recommended cards, recommended information and other image resources displayed in the interface.

[0121] The electronic device 900 can realize the shooting function through an ISP, a camera 993, a video codec, a GPU, a display screen 994, and an application processor.

[0122] The camera 993 is used to capture still images or videos. The object generates an optical image through the lens and projects it onto the photosensitive element. The photosensitive element can be a charge coupled device (CCD) or a complementary metal oxide semiconductor (CMOS) phototransistor. The photosensitive element converts the optical signal into an electrical signal, and then transmits the electrical signal to the ISP to be converted into a digital image signal. The ISP outputs the digital image signal to the DSP for processing. The DSP converts the digital image signal into an image signal in a standard RGB, YUV or other format. In some embodiments, the electronic device 900 may include 1 or N cameras 993, where N is a positive integer greater than 1.

[0123] The external memory interface 920 can be used to connect an external memory card, such as a Micro SD card, to expand the storage capacity of the electronic device 900. The external memory card communicates with the processor 910 through the external memory interface 920 to implement a data storage function, such as storing music, video and other files in the external memory card.

[0124] The internal memory 921 can be used to store computer executable program codes, which include instructions. The processor 910 executes various functional applications and data processing of the electronic device 900 by running the instructions stored in the internal memory 921. The internal memory 921 may include a program storage area and a data storage area. Among them, the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc. The data storage area may store data created during the use of the electronic device 900 (such as audio data, a phone book, etc.), etc. In addition, the internal memory 921 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, a universal flash storage (Universal Flash Storage, UFS), etc.

[0125] The electronic device 900 can implement audio functions such as music playing and recording through the audio module 970, the speaker 970A, the receiver 970B, the microphone 970C, the headphone jack 970D, and the application processor.

[0126] The audio module 970 is used to convert digital audio information into analog audio signal output, and is also used to convert analog audio input into digital audio signal. The audio module 970 can also be used to encode and decode audio signals. In some embodiments, the audio module 970 can be arranged in the processor 910, or some functional modules of the audio module 970 can be arranged in the processor 910.

[0127] A touch sensor is also called a "touch panel". The touch sensor can be arranged on the display screen 994, and the touch sensor and the display screen 994 form a touch screen, also called a "touch screen". The touch sensor is used to detect touch operations acting on or near it. The touch sensor can pass the detected touch operation to the application processor to determine the type of touch event. Visual output related to the touch operation can be provided through the display screen 994. In other embodiments, the touch sensor can also be arranged on the surface of the electronic device 900, which is different from the position of the display screen 994. In an embodiment of the present application, the electronic device 900 can detect user operations such as single-clicking, long pressing, and sliding on the display screen 994 through the touch detection capability provided by the touch sensor, thereby controlling the opening or removal of recommended applications, recommended cards, and recommended information, and realizing jump switching between different user interfaces.

[0128] The key 990 includes a power key, a volume key, etc. The key 990 may be a mechanical key or a touch key. The electronic device 900 may receive key input and generate key signal input related to user settings and function control of the electronic device 900.

[0129] Motor 991 can generate vibration prompts. Motor 991 can be used for incoming call vibration prompts, and can also be used for touch vibration feedback. For example, touch operations acting on different applications (such as taking pictures, audio playback, etc.) can correspond to different vibration feedback effects. Indicator 992 can be an indicator light, which can be used to indicate charging status, power changes, messages, missed calls, notifications, etc.

[0130] In passing Fig. 9 After introducing the hardware structure of electronic equipment, we will continue to combine Figure 8 The software structure of the electronic device is described.

[0131] The software system of the electronic device 900 may adopt a layered architecture, an event-driven architecture, a micro-core architecture, a micro-service architecture, or a cloud architecture. The embodiment of the present application takes the Android system of the layered architecture as an example to exemplify the software structure of the electronic device 900.

[0132] Fig.10 It is a software structure block diagram of the electronic device 900 according to an embodiment of the present application.

[0133] The layered architecture of the electronic device 900 divides the software into several layers, each layer has a clear role and division of labor. The layers communicate with each other through software interfaces. In some embodiments, the Android system is divided into four layers, from top to bottom, namely, the application layer, the framework layer, the hardware abstraction layer (HAL), and the kernel layer.

[0134] The application layer can include a series of application packages.

[0135] like Fig.10 As shown, the application package may include applications such as calls, music, and videos.

[0136] The application framework layer provides an application programming interface (API) and a programming framework for the applications in the application layer. The application framework layer includes some predefined functions.

[0137] like Fig.10 As shown, the application framework layer may include a window manager, a content provider, a view system, a phone manager, a resource manager, a notification manager, and the like.

[0138] The hardware abstraction layer is an interface layer located between the application framework layer and the kernel layer, and is used to provide a virtual hardware platform for the operating system. The hardware abstraction layer may include a camera HAL, an audio HAL, a WiFi HAL, etc. In an embodiment of the present application, the audio HAL may include a perception module, an exposure scheme generation module, an exposure scheme update module, and a service presentation module. It should be noted that the above modules can also be implemented by other software structure layers, such as by one or more applications in the software program layer, etc., and there is no specific limitation on this.

[0139] The perception module, which runs permanently or in a low-power form, has the ability to perceive external facts or environments. The perception module can detect relevant events and obtain the status of relevant events from the application program layer, application framework layer, system layer or kernel layer through the application programming interface (API), such as detecting Bluetooth connection, network connection, monitoring user text messages, custom timers, etc. In an embodiment of the present application, the perception module can monitor whether an exposure event of content recommendation (such as card recommendation, application recommendation, information recommendation, etc.) occurs, and can notify the exposure plan generation module of the monitored exposure event.

[0140] An exposure scheme generation module is used to generate an original exposure scheme. In the embodiment of the present application, taking a single application as an example, the original exposure scheme may include the exposure time of the application (e.g., 8:00-12:00). Alternatively, the original exposure scheme may include the exposure time of the application in different scenarios.

[0141] The exposure plan update module can obtain the user's operation behavior (click, remove or no operation) on the recommended content in the historical event segment, and then determine the Q value score used to characterize the user's click probability on the recommended content based on the operation behavior, and determine whether the recommended content should be exposed based on the Q value score to update the original exposure plan. Alternatively, it can determine whether the recommended content should be exposed within a certain original exposure time period based on the Q value score. Alternatively, it can also determine whether the recommended content should be exposed in different scenarios based on the Q value score.

[0142] The business presentation module (such as YOYO suggestions) is used to obtain the updated exposure plan, and update and expose the recommendation cards displayed on the negative one screen, home page and other desktops, the application icons of recommended applications in the recommendation folder, recommended information, etc. based on the updated exposure plan.

[0143] The kernel layer is the layer between hardware and software. The kernel layer contains at least display driver, camera driver, audio driver, and sensor driver.

[0144] Understandably, Fig.10 The layers in the software structure shown and the components included in each layer do not constitute a specific limitation on the electronic device 900. In other embodiments of the present application, the electronic device 900 may include more or fewer layers than shown, and each layer may include more or fewer components, which is not limited in the present application.

[0145] After the hardware structure and software structure of the electronic device 900 are preliminarily introduced, the technical terms involved in the electronic device are explained next.

[0146] (1) Exposure and non-exposure. In the embodiments of the present application, exposure may refer to displaying the target recommended content (such as the application icon of the recommended application, the recommended card, the recommended information, the recommended service, etc.) in the target display area (i.e., the location for displaying the recommended content, such as the preset application display location in the recommended folder, the preset card display area, the resource display location on the negative one screen, or the display location of the recommended service). Similarly, non-exposure refers to not displaying the target recommended content in the target display area.

[0147] (2) Click, remove, and no operation (or no response). In the embodiments of the present application, the user's click behavior on recommended content (recommended applications, recommended cards, recommended information, etc.) can be called a click (a click can be an operation used to open the recommended content), the user's removal behavior on recommended content can be called removal, and the user's failure to perform an operation on the recommended content can be called no operation (or other behaviors of the user on the recommended content other than click operations or removal operations can also be called no operation, such as invalid click operations, etc.).

[0148] (3) Q-value learning, i.e., a reinforcement learning method based on value function, can learn the optimal action selection strategy by estimating the value (Q-value) of each state-action pair. In the embodiment of the present application, the user's triggering behavior for recommended content can be summarized into three operation actions: click, remove, and no operation, and the above three operation actions are determined as three states in Q-value learning. In addition, whether the application is displayed can be summarized into two display actions: exposure and non-exposure, and the above two display actions are determined as two actions in Q-value learning.

[0149] (4) Grid search, a parameter tuning method, can help find the best model parameters. In grid search, we first specify the candidate value range of the parameter, then enumerate all possible parameter combinations and calculate the performance index of each model. Finally, the parameter combination with the best performance index is selected as the final model parameter.

[0150] (5) Scenario, that is, the usage scenario of the electronic device.

[0151] In some embodiments, the scene of the electronic device may be related to the location of the electronic device. For example, the scene may include one or more scenes of special geographical locations such as home scenes, company scenes, subway scenes, etc. Exemplarily, the electronic device can obtain the location information of its current location, such as the latitude and longitude information collected by the positioning module, or the base station identifier carried in the base station signal sent by the base station after the electronic device enters the coverage range of the base station, such as the cellular cell identity (cellID). If the location information matches (for example, is consistent with) the location of the special geographical scene, it can be determined that the electronic device is in a special geographical scene. For example, if the cellID of the current location of the electronic device is consistent with the cellID of the pre-stored company, it can be determined that the electronic device is in the company scene.

[0152] In other embodiments, the scene of the electronic device may also be related to the user's usage status, usage environment or usage needs, such as meeting scenes, game scenes, audio and video scenes, focus scenes, etc., which can be set by the user on the electronic device.

[0153] It should be noted that during the use of the electronic device, it can also be set to other usage scenarios according to the user's usage scenarios and usage requirements, and there is no specific limitation on this.

[0154] After introducing the above content, the following will take application recommendation as an example to continue to explain content recommendation. It should be noted that, for ease of understanding, the following part of the embodiment of the present application takes application recommendation as an example to explain the content recommendation scheme, but it should be understood that the scheme provided in the embodiment of the present application is also applicable to content recommendations such as recommendation cards, recommendation services, and recommendation information.

[0155] In a related technique, Fig.11 FIG. 2 shows a recommended exposure flow chart in a related art. Fig.11 As shown, after the electronic device generates the original exposure plan, it can expose the application according to the original exposure plan. The original exposure plan is used to record the recommended applications of the electronic device (i.e., the applications that need to be recommended for exposure), as well as the application exposure status or application exposure plan of the recommended applications. For example, Fig.12 FIG. 2 shows a logic diagram of an exemplary application exposure in a related art. Fig.12 As shown, if the exposure time periods in the original exposure plan of application A are 8:00-14:00 and 16:00-20:00, then application A is exposed during 8:00-14:00 and 16:00-20:00 (i.e., the application icon of application A is displayed in the recommended folder), and is not exposed during other time periods (i.e., the application icon of application A is not displayed in the recommended folder). In another exemplary embodiment, Fig.13 FIG. 2 shows another exemplary logic diagram of application exposure in a related art. Fig.13 As shown, if the original exposure scheme of application A includes: the exposure time periods of application A in scene 1 are 8:00-14:00 and 16:00-20:00, and the exposure time periods of application A in scene 2 are 10:00-12:00 and 18:00-19:00. If the electronic device is in scene 1 before 9:00 and after 17:00, then before 9:00 and after 17:00, the electronic device can be exposed according to the exposure time period of application A in scene 1, such as 8:00-9:00 exposure and 17:00-20:00 exposure. And, if the electronic device is in scene 2 during the time period of 9:00-17:00, then during this time period, the electronic device can be exposed according to the exposure time period of application A in scene 2, such as 10:00-12:00 exposure.

[0156] Regarding the specific form of the original exposure scheme, in one example, the original exposure scheme may include exposure time periods for multiple applications. In another embodiment, the original exposure scheme may include application identifiers of exposure applications, etc. In yet another embodiment, the original exposure scheme may include exposure applications for multiple time periods. In yet another embodiment, the original exposure scheme may include exposure schemes for different scenarios. For example, exposure time periods for multiple applications in each scenario, etc. It should be noted that the original exposure scheme may include different contents according to actual conditions and specific scenarios, and no specific limitation is imposed on this.

[0157] In some embodiments, the electronic device may determine the original exposure plan according to the leisure time period and leisure intention by predicting the user's leisure time period (i.e., the time period during which applications can be recommended to the user, such as 20:00-22:00 every day) and the probability of leisure intention.

[0158] Among them, the time when the electronic device opens applications (such as applications in a preset application set, which are applications that can be recommended to users) within a historical time period (such as 90 days before the current date) can be counted, and based on the counted time, the probability of opening applications in multiple time periods within a day (for example, a day can be divided into 48 time periods) is estimated through a probability prediction model, and based on the probability of opening applications in each time period, the idle time period within a day is determined (for example, one or more time periods with the highest probability or a relatively high probability, or one or more time periods with a probability higher than a threshold value are determined as idle time periods). It should be noted that the idle time period can also be pre-set uniformly, or set by the user on the electronic device, and there is no specific restriction on this.

[0159] When applications can be divided into multiple categories, the probability of opening applications in each category during the idle time period can be determined. Applications in one or more categories with the highest probability are selected for exposure recommendation. It should be noted that the probability of users opening applications can also be determined in other ways, and then recommendations are made, such as determining the opening probability of each application, etc., and there is no specific limitation on this.

[0160] In other embodiments, the electronic device may also determine the original recommendation scheme according to the frequency of use of each application by the user. For example, the application with high frequency of use may be determined as the recommended application. For another example, the application with high frequency of use in each time period may be determined as the recommended application, and then the recommended application may be exposed in each time period.

[0161] In some other embodiments, the electronic device may also determine the original recommendation scheme according to the user's usage habits. For example, if the user is accustomed to using food delivery software during the noon time period, the electronic device may determine the food delivery software as a recommended application during the noon time period, and determine the noon time period as the exposure period for the food delivery application.

[0162] In some further embodiments, the electronic device may also determine the original recommendation scheme according to the scenario the user is in. For example, if the user is in a company scenario, office applications, conference applications, etc. may be determined as recommended applications.

[0163] It should be noted that in the embodiments of the present application, the original recommendation plan can also be determined based on actual recommendation needs and specific recommendation requirements, according to the usage habits and usage of the electronic device, etc., and there is no specific limitation on this.

[0164] However, the prediction accuracy of the original recommendation scheme in the above-mentioned related schemes is often low. The inventors have found through research that the above-mentioned prediction process is often based on the user's application usage history in the past period of time, and does not consider the user's feedback behavior on the exposed application (or recommended application) after the application is exposed. This makes it impossible for the original recommendation scheme of the related technology to timely understand the user's attitude towards the recommendation, such as whether the recommended application is accurate, whether it meets the user's needs, and whether the user is interested in the recommended application.

[0165] For example, the recommended applications determined by the above-mentioned schemes may be exposed for a long time but not clicked by users. This may cause inaccurate recommendations and users to be uninterested (unsensitive) to the recommendations, which means that the recommended applications cannot meet user needs. This not only wastes exposure costs, but also causes other applications that users may be interested in to be unable to be exposed due to occupying exposure positions, further reducing user experience.

[0166] Based on this, the embodiment of the present application provides a content recommendation scheme, where the electronic device can adjust the exposure scheme according to the user's feedback (or feedback behavior) on the exposed recommended applications, and timely intercept the exposure of recommended applications that do not need to be exposed (applications that have been continuously exposed for multiple times but have not been clicked). In some embodiments, Fig.14 The following is a schematic diagram showing the process flow of the recommended exposure scheme provided in the embodiment of the present application. Fig.14 and Fig.11It can be seen that in the technical solution of the embodiment of the present application, after obtaining the original exposure plan, the electronic device does not directly perform exposure according to the original recommended plan, but suppresses the original exposure plan through the user's feedback data on the recommended application (that is, data that can reflect / record the user's relevant feedback behavior on the exposed application) (recommended application does not perform exposure, or part or all of the exposure period of the recommended application does not perform exposure, etc., which can suppress the exposure process of the recommended application), and updates to obtain a new exposure plan. Then, the electronic device can perform exposure according to the new exposure plan.

[0167] For example, Fig.15 The following is a logical diagram of an exemplary application exposure provided by an embodiment of the present application. Fig.15 and Fig.13 It can be seen that the exposure time periods of the original exposure scheme of application A can be 8:00-14:00 and 16:00-20:00, and if exposure suppression is performed on application A, application A may not be exposed during the above exposure time periods. It should be noted that in the scheme provided in the embodiment of the present application, exposure may not be performed during some of the above exposure time periods, and there is no specific limitation on this.

[0168] As another example, Fig.16 A logical schematic diagram of another exemplary application exposure provided by an embodiment of the present application is shown. Fig.16 and Fig.12 The difference is that if application A is suppressed from exposure in scene 2, application A is suppressed from exposure in the updated exposure scheme of scene 2. If the electronic device is in scene 1 before 9:00 and after 17:00, the electronic device can be exposed according to the exposure period of application A in scene 1 before 9:00 and after 17:00 (i.e., exposed according to the original exposure scheme), such as 8:00-9:00 and 17:00-20:00. Also, the electronic device is in scene 2 during the time period of 9:00-17:00, and since application A is suppressed from exposure in scene 2, application A is not exposed during the time period of 9:00-17:00.

[0169] Through the embodiments of the present application, the electronic device can further improve the accuracy of content recommendation based on the user's feedback on the recommended application, further improve the user's experience of content recommendation, and increase the user's click rate on the recommended content.

[0170] Optionally, if the electronic device can perform multiple content recommendation services (such as performing content recommendation at different locations or different recommendation functions, or when multiple recommendation algorithms are used to perform the same recommendation, different recommendation algorithms are determined as different content recommendation services), the recommendation algorithms used by different content recommendation services can be the same or different. Each content recommendation service can generate an original recommendation scheme based on its own recommendation algorithm, and then after each content recommendation service, the content recommendation method of the embodiment of the present application can be used to perform separate exposure suppression.

[0171] After a preliminary understanding of the technical solution of the embodiment of the present application through the above examples, the content recommendation method of the embodiment of the present application is described in conjunction with the flow chart. It should be noted that, for the sake of convenience, the following part of the embodiment of the present application takes application recommendation as an example to illustrate the technical solution of the embodiment of the present application, but it should be understood that the content recommendation method provided by the embodiment of the present application can also be applied to content recommendations such as recommendation cards, recommendation information, and recommendation services, which will not be repeated here.

[0172] The content recommendation process in the embodiment of the present application can be divided into three parts. Specifically, the first part is the data preprocessing part, the second part is the exposure model training (learning) part, and the third part is the exposure scheme processing part. The contents of each part will be explained one by one below.

[0173] First, for the data preprocessing part, it is used to sort out the user's feedback behavior on the exposed applications to be processed from the historical exposure data, and organize it into a feedback behavior list in a preset format, so as to facilitate the subsequent part to train the exposure model based on the feedback behavior list in the preset format.

[0174] Specifically, Fig.17 FIG. 1 is a flow chart of a data preprocessing method provided in an embodiment of the present application. Fig.17 As shown, the data preprocessing part may include steps S101 to S105, and then Figure 17 to Figure 18 The data preprocessing part is explained.

[0175] S101: A data preprocessing module of an electronic device obtains historical exposure data of an application to be processed.

[0176] In an embodiment of the present application, the application to be processed may be an application with exposure authority (i.e., it can be recommended / exposed in the recommendation folder). In some embodiments, the applications with exposure authority can be respectively used as applications to be processed. For example, if the original exposure scheme can select one or more applications from K applications (applications with exposure authority) for recommendation, then the K applications with exposure authority can be respectively used as applications to be processed, where K is a positive integer greater than 2. For example, the above K applications can constitute a set of applications to be recommended (application pool of applications to be recommended), and the applications in the set of applications to be recommended can be respectively used as applications to be processed. Exemplarily, if the recommendation service (or recommendation algorithm, recommendation strategy) for recommendation using leisure time can select applications from application A-application E for application recommendation, then application A-application E can be respectively used as applications to be processed. For example, the electronic device can obtain the historical exposure data of the last 90 days recommended by YOYO. Optionally, the electronic device can use the application with the original exposure scheme generated as the application to be processed. For example, if application A generates the original exposure scheme on the current date, application A can be used as the application to be processed.

[0177] And, in an embodiment of the present application, historical exposure data (also referred to as historical data) may include data generated during the exposure process of the application to be processed within the historical time period. That is to say, if the application to be processed is exposed multiple times within the historical time period, the historical exposure data may include data generated by the application to be processed at each exposure. Among them, if the exposure of the application can be divided into multiple exposure cycles (for example, one day is used as an exposure cycle), the historical time period may include a preset duration (such as a preset number of historical days) before the current exposure cycle (or current moment). For example, the historical time period may be 90 days, such as 90 days before the current date. It should be noted that the historical time period can also be set to other time lengths according to actual conditions and specific needs, such as 180 days, 30 days, etc., and there is no specific limitation on this.

[0178] In some embodiments, the historical exposure data may include relevant data that can reflect the user's feedback status or feedback situation on the application to be processed in the exposure state.

[0179] In one embodiment, the historical exposure data may include user feedback behavior data. The user feedback behavior data is data generated when a user performs a feedback operation on an application to be processed in an exposed state. In other words, the user feedback behavior data is used to characterize the user's feedback behavior on an application to be processed in an exposed state.

[0180] Exemplarily, since the user's feedback behavior on the application to be processed in the exposure state may include operations such as clicking (such as a click operation for viewing specific recommended content), removing (a removal operation for removing recommended content), etc., accordingly, the user feedback behavior data may include a click moment (i.e., the moment when the electronic device detects that the user performs a click operation on the application to be processed in the recommended file, which may be referred to as the first moment) and / or a removal moment (i.e., the moment when the electronic device detects that the user removes the application icon of the application to be processed from the recommended folder, which may be referred to as the second moment). Specifically, the format of the click moment and the removal moment may include a date and a specific time, wherein the specific time may be accurate to minutes, or may be accurate to hours or seconds, etc. according to actual needs, and no specific restrictions are made to this. For example, if the user clicks the application icon of the application to be processed at 18:45 on January 1, 2024, the click moment recorded by the electronic device may include 20240101 (date information, the date format is only for example, and it may also be other formats) and 18:45 (specific time). Among them, the date and the specific time may be one piece of information or multiple pieces of information, and no restrictions are made to this. The specific format of the removal time is similar to that of the click time. Please refer to the above part of the embodiment of the present application for the specific description of the click time, which will not be repeated here.

[0181] Optionally, in order to make personalized application recommendations based on usage scenarios, the user feedback behavior data may also include click scene information and / or removal scene information. Among them, the click scene information may be the usage scenario of the electronic device when the click behavior occurs, and the removal scene information may be the usage scenario of the electronic device when the removal operation occurs. For example, if the electronic device is in the home scene when the user performs the click operation, the user feedback behavior data may also include "home" (i.e., scene information). Specifically, the electronic device may obtain and record the scene information of the electronic device when the user performs the click operation or the removal operation. Alternatively, the electronic device may also use other methods to record the scene, such as obtaining and recording the current scene of the electronic device when the scene changes, or obtaining and recording the current scene of the electronic device at intervals of a preset duration, and then determining the click scene information, removal scene information, etc. based on the current scene obtained. In addition, it should be noted that the above-mentioned user feedback behavior data may also be implemented as other data according to other user feedback operations in the actual exposure scene and specific exposure requirements, and there is no specific limitation on this.

[0182] In another embodiment, the historical exposure data may further include exposure behavior data, wherein the exposure behavior data is data generated by the exposure behavior of the application to be processed, or the exposure behavior data is used to characterize the exposure behavior of the application to be processed.

[0183] Exemplarily, the exposure behavior data may include the exposure time of the application to be processed (i.e., the time when the application to be processed starts to be exposed, which may be referred to as the second time. For example, if the application to be processed is exposed in the time period of 8:00-10:00 on January 2, 2024, the exposure time is 8:00 on January 2, 2024). It should be noted that the specific format of the exposure time may be similar to the click time, which may refer to the description of the click time in the above part of the embodiment of the present application, and will not be repeated here.

[0184] Optionally, in order to be able to make personalized application recommendations based on usage scenarios, the exposure behavior data may also include exposure scenario information, that is, the usage scenario of the electronic device when the application to be processed is exposed, wherein the exposure scenario information is similar to the click scenario information, and reference may be made to the above description of the click scenario information in the above embodiment of the present application, which will not be described in detail. Also, it should be noted that the above exposure behavior data may also be implemented as other data based on the actual exposure scenario and specific exposure requirements, such as the time to end the exposure, etc., and there is no specific limitation on this.

[0185] In a specific example, Fig.18 FIG. 2 is a schematic diagram showing an exemplary historical exposure data provided by an embodiment of the present application. Fig.18 As shown in the figure, if the application to be processed is exposed three times in a certain day, the first exposure starts at time t1 (i.e., start time t1) and ends at time t2 (i.e., end time t2); the second exposure starts at time t3 (i.e., start time t3) and ends at time t4 (i.e., end time t4); the third exposure starts at time t5 (i.e., start time t5) and ends at time t6 (i.e., end time t6). Then the historical exposure data can record three exposure times, i.e., start time t1, start time t3, and start time t5.

[0186] Also, during the first exposure of the application to be processed, the user performs a click operation at time Ta, and the historical exposure data can record the click time Ta. Also, during the second exposure of the application to be processed, the user performs a removal operation at time Tb, and the historical exposure data can record the removal time Tb.

[0187] Also, it should be noted that historical exposure data may also include other data based on actual exposure scenarios and specific exposure needs, such as information on other dimensions besides the time dimension, such as the user's location at the time of exposure, the scene of the electronic device, etc., without any specific restrictions on this.

[0188] S102: A data preprocessing module of the electronic device screens the historical exposure data to obtain feedback behavior data.

[0189] In the embodiment of the present application, the feedback behavior data may be data that can characterize the feedback state or feedback behavior of the user to the application to be processed in the exposure state. Exemplarily, the feedback behavior data may include user feedback behavior data. In another exemplary embodiment, the feedback behavior data may also include exposure behavior data. Among them, the user feedback behavior data and the exposure behavior data can refer to the relevant description of the above part of the embodiment of the present application, which will not be repeated here.

[0190] In one embodiment, Fig.17 As shown, the feedback behavior data may include exposure time, click time, and removal time. Optionally, the feedback behavior data may also include exposure scene information, click scene information, removal scene information, etc. It should be noted that, depending on the specific scenario and actual needs, the feedback behavior data may also include other information that can be used in model training or content recommendation, which will not be described in detail. For example, whether the date of exposure behavior, click behavior, and removal behavior is a holiday or a working day.

[0191] In S102, since the information recorded in the historical exposure data is relatively complex, such as including information such as judgment conditions, S102 can accurately obtain the required feedback behavior data from the historical exposure data to facilitate subsequent processing. Optionally, in actual scenarios, the electronic device can also directly obtain the required information without screening, and there is no limitation on this.

[0192] S103: The data preprocessing module of the electronic device removes abnormal data from the screened feedback behavior data.

[0193] In some embodiments, data that meets the abnormal data determination condition may be processed, wherein the abnormal data determination condition may be set according to actual needs and experience, and no specific limitation is imposed on this.

[0194] In one embodiment, for exposure moments, if the time interval between multiple exposure moments is less than or equal to a preset time interval (for example, 5 minutes, which can be set to other values ​​based on actual experience or specific needs, and there is no specific limitation on this), then the multiple exposure moments can be merged and only the first exposure moment (for example, the exposure moment with the smallest time value among the multiple exposure moments) is retained.

[0195] For example, if the exposure moments in the feedback behavior data include 11:00, 11:01, and 11:03 on the same day, 11:00 can be retained, and 11:01 and 11:03 can be removed from the feedback behavior data. For example, if the time interval between two consecutive exposure moments does not exceed 5 minutes, they can be merged, and only the first exposure moment is retained.

[0196] Optionally, the click moment and / or the removal moment may be merged in the same manner. The specific implementation method thereof may refer to the relevant description of the exposure moment, which will not be described in detail.

[0197] It should be noted that when the electronic device collects feedback behavior data abnormally, the data collection is inaccurate, or the electronic device has abnormal continuous exposure, this processing method can avoid the abnormal situation where the user only sees one exposure but collects multiple exposure data, improves the accuracy of the collected data, and further improves the accuracy of content recommendation. Similarly, when similar merging processing is used for click time, removal time, etc., it can avoid data anomalies caused by abnormal data collection, inaccurate data collection, users mistakenly clicking continuously, or electronic devices detecting abnormal user click behavior, improve the accuracy of collected data, and further improve the accuracy of content recommendation.

[0198] Furthermore, it should be noted that in S103, other abnormal data may also be processed, such as eliminated or corrected, and for example, data with wrong data format or unreasonable data may be processed, without any specific limitation.

[0199] S104, the data preprocessing module of the electronic device divides the historical time period into a plurality of preset time slots, determines the feedback behavior label in each preset time slot according to the feedback behavior data, and generates a feedback status list based on the feedback behavior label.

[0200] For the preset time period, the electronic device can divide the 24 hours of each day into multiple preset time periods (slots), and the length of each preset time period (slot_size) is the preset length. For example, if each day is divided into 48 preset time periods, and the preset length of each preset time period is 0.5 hours, the historical time period can be divided into 90*48 time periods. For example, continue to refer to Fig.18 , 0:00-0:30 of each day is the first preset time period of the day (i.e., time period 1), 0:30-1:00 is the second preset time period of the day (i.e., time period 2), and so on. Optionally, in order to accurately suppress the exposure of the original exposure plan, the division method of the preset time period can be the same as the division method of the preset time period in the original exposure plan. For example, if the original exposure plan is determined based on the leisure time period and the leisure exposure intention, the daily time period in the historical time cycle is divided into 48 time periods when the leisure time period is predicted, then S104 can also divide the daily time period into 48 time periods. It should be understood that the above description is based on the example of 48 preset time periods and the duration of each preset time period is 0.5 hours; the present application does not limit the number of preset time periods and the duration of the preset time periods.

[0201] For each feedback behavior label within a preset time period, it is used to describe the user's feedback behavior on the application to be processed in the exposure state within the preset time period. Exemplarily, the feedback behavior label may include behavior labels of different feedback behaviors. For example, if the feedback behavior includes exposure behavior, click behavior and removal behavior, the feedback behavior label may include an exposure label (exposure_label), a click label (click_label) and a removal label (remove_label). Among them, the exposure label is used to mark that the application to be processed has been exposed within the preset time period, or the exposure label can also be used to mark that the application to be processed has been exposed within the preset time period and the user has no feedback operation. The click label is used to mark that the user has clicked on the application icon of the application to be processed in the recommended folder within the preset time period. The removal label is used to mark that the user has removed the application icon of the application to be processed from the recommended folder within the preset time period. Optionally, if the feedback behavior includes exposure behavior and click behavior, the feedback behavior label may include an exposure label and a click label. Optionally, if the feedback behavior includes a click behavior and a removal behavior, the feedback behavior label may include a click label and a removal label.

[0202] Furthermore, the method for determining the feedback behavior label in each preset time period may be determined based on the statistical number of each feedback behavior in the preset time period. Specifically, the step of the electronic device determining the feedback behavior label may include step A1 and step A2. Step A1, based on the feedback behavior data, counting the number of exposures, clicks, and removals in each preset time period. For example, continue to refer to Fig.18 , the number of exposures in time period 1 (0:00-0:30) is 1, the number of clicks is 0, and the number of removals is 0; the number of exposures in time period 2 (0:30-1:00) is 0, the number of clicks is 1, and the number of removals is 0; ...; time period 12 (5:30-6:00). Step A2, the number of exposures, clicks, and removals in each preset time period are used to determine the feedback behavior label for each time period. Exemplarily, the specific definition rules of the feedback behavior label can be as follows:

[0203] Define rule 1: if the number of clicks in a preset time period is greater than 0, the click label of the preset time period is defined as 1, i.e., click_label = 1. Similarly, if the number of clicks in the preset time period is 0, the click label of the preset time period is defined as 0, i.e., click_label = 0. Optionally, the specific value of the click label can also be set to other values ​​according to actual conditions and specific needs, such as the number of clicks, etc., and there is no specific restriction on this.

[0204] Define rule 2: if the number of removals within a preset time period is greater than 0, then the removal label of the preset time period is defined as 1, that is, remove_label=1. Similarly, if the number of removals within the preset time period is 0, then the removal label of the preset time period is defined as 0, that is, remove_label=0. Optionally, the specific value of the removal label can also be set to other values ​​according to actual conditions and specific needs, such as the number of removals, etc., and there is no specific restriction on this. Optionally, in order to facilitate the determination of the feedback behavior of the preset time period in subsequent steps, if the number of clicks and the number of removals in a preset time period are both greater than 0, the removal label can be defined as 1 and the click label can be defined as 0.

[0205] Definition rule 3: If the number of removals and clicks within a preset time period are both 0, and the number of exposures is greater than 1, then the exposure label within the preset time period is defined as 1, that is, exposure_label=1. Similarly, if the number of removals within the preset time period is greater than 0 or the number of clicks within the preset time period is greater than 0, then the exposure label within the preset time period is defined as 0, that is, exposure_label=0. And, if the number of removals, clicks, and exposures within the preset time period are all 0, then the exposure label within the preset time period is defined as 0. Through this definition rule 3, if the exposure label is defined as 1, it means that the user has no feedback operation on the pending application in the exposed state during the preset time period. Optionally, the electronic device may also define the exposure label as 1 when the number of exposures is greater than 0; and define the exposure label as 0 when the number of exposures is 0. At this time, the exposure label indicates whether the pending application has been exposed.

[0206] In one example, referring to 18, the number of exposures in period 1 (0:00-0:30) is 1, the number of clicks is 0, and the number of removals is 0, then the exposure_label of period 1 is 1, click_label=0, and remove_label=0. Also, the number of exposures in period 2 (0:30-1:00) is 0, the number of clicks is 1, and the number of removals is 0, then the exposure_label of period 2 is 0, click_label=1, and remove_label=0. Similarly, the feedback behavior labels of other time periods can be determined, which will not be described in detail.

[0207] The feedback status list may include feedback behavior labels for multiple preset time periods within the historical time period. Exemplarily, for ease of processing, the feedback behavior labels for multiple preset time periods in the feedback status list may be arranged in chronological order. Optionally, for ease of processing, the feedback status list is used to record feedback behavior labels for valid time periods, wherein the value of one or more of the exposure label, click label, and removal label for the valid time period is 1, that is, exposure_label, click_label, and remove_label are not all 0. For example, continue to refer to Fig.17 , the feedback status list may include the feedback behavior label of time period 1, the feedback behavior label of time period 2, ..., the feedback behavior label of time period n. Wherein, time period 1 to time period n may all be the above-mentioned valid time periods, and n is an integer greater than or equal to 1. It should be noted that the feedback status list may also be set to other forms according to actual conditions and specific values, such as retaining the feedback behavior labels of all preset time periods, and there is no specific limitation on this.

[0208] Optionally, if the electronic device includes multiple scenes, the number of feedback behaviors in multiple preset time periods in each scene can be determined based on the historical exposure data of each scene, and the feedback behavior labels in multiple preset time periods in the scene can be determined based on the number of feedback behaviors in multiple preset time periods. And, the feedback status lists of multiple scenes are spliced ​​into one feedback status list, for example, the list can be named exposureClickLocData.

[0209] Exemplarily, the feedback status list exposureClickLocData may be as shown in Table 1 below.

[0210] Table 1 Feedback status list (exposureClickLocData)

[0211]

[0212] As shown in Table 1 above, the feedback status list exposureClickLocData can be arranged according to the scene type (such as the Loc field in the above table), and the data for the same scene can be sorted in chronological order (date, slot fields in the above table) from small to large. It should be noted that the order of sorting by scene in Table 1 is only for illustrative purposes, and the specific arrangement can also be set according to actual needs, without limitation. And, for the sake of ease of expression, each row of data can be referred to as feedback behavior label data for a time period.

[0213] Also, it should be noted that, in an embodiment of the present application, the all scene may include all scenes and no scene, and the feedback behavior label of the all scene may include the feedback behavior labels of all preset time periods, for example, it may include the feedback behavior labels of all sub-scenes (home scene, company scene, etc.).

[0214] Optionally, the feedback status list may also include information related to subsequent model training and exposure suppression, such as information indicating whether it is a weekday, or information indicating the day of the week, without specific limitation.

[0215] Optionally, it should also be noted that in the embodiment of the present application, the feedback behavior labels of multiple time periods or multiple moments may be determined without time division, and there is no specific limitation on this.

[0216] S105 , the data preprocessing module of the electronic device generates a feedback behavior list (feedbackDict) based on the feedback state list (exposureClickLocData).

[0217] The feedback behavior list is used to record the feedback behaviors in each preset time period. Fig.17 , the feedback behavior list may include feedback behavior in time period 1, feedback behavior in time period 2, ..., feedback behavior in time period n. Exemplarily, feedback behavior may include one or more of click, remove, and no operation. Among them, no operation may mean that the application to be processed has been exposed, but the user has not performed a triggering operation on it. It should be noted that in the embodiments of the present application, it can also be set to other feedback data according to actual conditions and specific needs, such as no operation and click, etc., and there is no specific limitation on this.

[0218] In one embodiment, the feedback behavior of each time period can be determined based on the feedback behavior label of the time period. Exemplarily, the electronic device can first define an empty feedback behavior list (or feedback dictionary), that is, define feedbackDict={}. Then, for the feedback behavior label data of each scene, determine the feedback sequence of the scene. Then write the feedback sequence of the scene into the feedback behavior list. Among them, the feedback sequence of each scene can include feedback behaviors of multiple time periods under the scene.

[0219] Exemplarily, the feedback behavior list may be as shown in Table 2 below.

[0220] Table 2 Feedback behavior list (feedbackDict)

[0221]

[0222] As described in Table 2 above, the feedback behavior list may include feedback sequences of multiple scenarios. For example, the all scenario corresponds to the feedback sequence feedbackList, the home scenario may correspond to the feedback sequence feedbackList2, and the company scenario corresponds to the feedback sequence feedbackList3.

[0223] Exemplarily, the method of generating the feedback behavior list may include: for each scenario, executing the following steps D1 to D3.

[0224] Step D1, for a single scenario, such as the all scenario, define an empty feedback sequence, that is, define feedbackList=[].

[0225] Step D2, for each time period under the scenario in the feedback status list, take it as the target time period in turn, and execute the following steps D21 to D23.

[0226] Step D21: For the target time period, if the tag value of its removal tag is 1, then add the identification data of the removal behavior in the feedback sequence, such as adding "remove" at the end of the feedback sequence. Then the next time period of the target time period can be used as the new target time period to determine the feedback behavior of the next time period.

[0227] Step D22: For the target time period, if the tag value of the click tag is 1, the identification data of the click behavior is added to the feedback sequence, for example, "click" can be added at the end of the feedback sequence. Then the next time period of the target time period can be used as the new target time period to determine the feedback behavior of the next time period.

[0228] Step D23, if the tag value of the exposure tag is 1, then add identification data of no operation behavior to the feedback sequence, such as adding "nothing" to the end of the feedback sequence. For example, "click" can be added to the end of the feedback sequence. Then the next time period of the target time period can be used as the new target time period to determine the feedback behavior of the next time period. Optionally, if the tag values ​​of the click tag and the removal tag are both 1, select the tag data of the behavior with higher priority and add it to the feedback sequence.

[0229] Step D3, after determining the feedback actions for all time periods of a certain scene, write the obtained feedback sequence into the feedback behavior list. For example, after obtaining the feedback sequence feedbackList for all scenes, it can be written into the feedback behavior list feedbackDict. For example, feedbackDict['all']=feedbackList.

[0230] In one example, continuing to take the feedback status list (exposureClickLocData) shown in Table 1 as an example, after determining the feedback sequence feedbackList2 corresponding to the company scene, the feedback sequence feedbackList of the all scene can be determined. Specifically, for the all scene, first define the initial feedback sequence feedbackList=[].

[0231] Then, for the preset time period 1 of 20240101 in the all scenario, the label value of its exposure label (exposure_label) is 1, and "nothing" can be added to the feedback sequence feedbackList to obtain the feedback sequence feedbackList=[nothing].

[0232] Then, for the preset time period 2 of 20240101 in the all scenario, the label value of the removal label (remove_label) is 1, and "remove" can be added to the feedback sequence feedbackList obtained in the previous step to obtain the feedback sequence feedbackList=[nothing,remove].

[0233] Then, for the preset time period 1 of 20240102 in the all scenario, the label value of the click label (click_label) is 1, so "click" can be added to the feedback sequence feedbackList obtained in the previous step, and so on, and the feedback actions of other time periods are added to the feedback list until the feedback sequence feedbackList=[nothing,remove,click,…] is obtained.

[0234] Then, write the feedbackList of all scenes into feedbackDict. For example, if the feedback behavior list feedbackDict before the feedbackList of all scenes is written is {"home":[nothing,remove,…], "company":[click,…]}, where the feedback sequence feedbackList1 in the home scene is [nothing,remove,…], and the feedback sequence feedbackList2 in the company scene is [click,…]. Then the feedback behavior list feedbackDict after the feedbackList of all scenes is written is {"home":[nothing,remove,…], "company":[click,…], "all":[nothing,remove,click,…]}. The data length of each feedback sequence may refer to the number of identification data in the feedback sequence. For example, the feedback sequence feedbackList1 in the home scene contains two identification data "nothing" and "remove", and the data length of the feedback sequence feedbackList1 in the home scene is 2. Similarly, the data length of the feedback sequence feedbackList2 in the company scenario is 1, and the data length of the feedback sequence feedbackList in the all scenario is 3.

[0235] Correspondingly, the result example of S105 output data is as follows: feedbackDict={"home":[nothing,remove,…],"company":[click,…],"all":[nothing,remove,click,…]}. It should be noted that the scenes in the example and the identification data of each feedback behavior in the feedback behavior list are only used as examples and are not limited to them in the actual process. If the electronic device does not include a specific scene, the feedback behavior list feedbackDict may only include the feedback sequence of all scenes. Optionally, the output result of S105 can also be implemented in other forms, such as directly outputting the feedback sequence of each scene.

[0236] Optionally, in order to improve the feasibility of subsequent model training, such as the credibility of the Q value obtained by training, if the data length in the feedback sequence of a certain scenario is short, the feedback sequence can be cleared. Specifically, S105 may include step B1 and step B2. Step B1, obtain the data length of the feedback sequence of a certain scenario, wherein the data length may be the number of feedback behaviors in the feedback sequence. Step B2, when the data length is less than the preset data length threshold, the data of the feedback sequence of the scenario may be cleared. Among them, the preset data length threshold may be 10, and its value may be determined according to the actual scenario, specific training accuracy, data volume, etc., and its value is not adjusted. For example, if the historical time period is 180 days, the preset data length threshold may be selected as a value greater than 10.

[0237] It should be noted that in S105, the electronic device may determine the feedback behavior directly based on the exposure time, click time, and removal time without using the feedback behavior label, or determine the feedback behavior for each preset time period based on the number of exposures, clicks, and removals, without specific restrictions.

[0238] Also, it should be noted that the scene of the electronic device can also be a meeting scene, a game scene, etc., and the processing method is similar to the above-mentioned home scene and company scene, which will not be repeated here.

[0239] Furthermore, if the scenario of the electronic device is not considered, a feedback sequence (similar to feedbackList of all scenarios) can be obtained as a feedback behavior list according to the historical exposure data of the electronic device.

[0240] Furthermore, if the electronic device includes multiple content recommendation services, the historical exposure data of each content recommendation service may be processed using the method of steps S101 to S105 to obtain a feedback behavior list of the content recommendation service.

[0241] In some embodiments, if a day is divided into multiple exposure periods, the feedback behavior list for each exposure period can be determined based on the exposure data of each exposure period in the historical exposure data. For example, if 8:00-10:00 of a day is determined as an exposure period, the feedback behavior list for 8:00-10:00 can be determined through the above S101 to S105 based on the historical exposure data of 8:00-10:00 every day in the previous 90 days, so as to determine the Q value corresponding to 8:00-10:00 based on the feedback behavior list for 8:00-10:00.

[0242] In other embodiments, if the Q values ​​of weekdays and weekends are calculated separately, a feedback behavior list of weekdays can be determined based on the historical exposure data of weekdays in the previous 90 days, so that the Q values ​​corresponding to weekdays can be calculated according to the feedback behavior list of weekdays in the subsequent process. Similarly, a feedback behavior list of weekends can be determined according to the historical exposure data of weekends, so that the Q values ​​corresponding to weekends can be calculated according to the feedback behavior list of weekends in the subsequent process.

[0243] It should be noted that in the embodiment of the present application, historical exposure data of at least two categories can be determined in the historical exposure data, and then the feedback behavior list of each category can be determined based on the historical exposure data of each category, so as to calculate the Q value corresponding to each category in the subsequent process. The method of determining the feedback behavior list of each category is similar to the feedback behavior list of different scenarios, which can be referred to the above description and will not be repeated here.

[0244] After introducing the data preprocessing part through the above content, we will continue to explain the exposure model training part.

[0245] In an embodiment of the present application, the exposure model can be used to predict whether the application to be processed needs to be exposed. The electronic device can use the Q learning (Q value learning) reinforcement learning model to determine whether the application needs to be exposed. Optionally, in an embodiment of the present application, other models, such as deep neural models or other reinforcement learning models, can also be used as exposure models, without specific restrictions. Next, the training process of the exposure model is explained using the Q learning reinforcement learning model as an example.

[0246] Since the Q learning reinforcement learning model needs to involve state (state, )、action ) and Q value calculation function, before introducing the Q learning reinforcement learning model, we first )、action ), and the Q value calculation function are explained in turn.

[0247] State: In the embodiment of the present application, the feedback behavior can be determined as a state. For example, in the embodiment of the present application, three states can be included: click, remove, and nothing.

[0248] Action: In the embodiment of the present application, two actions may be included, namely ACTIONS=[“Yes”, “NO”]. “Yes” means exposure, and “No” means no exposure.

[0249] The Q value calculation function is used to calculate the Q value score of the current exposure cycle according to the feedback action, exposure action, and reward score value in the feedback behavior list.

[0250] In some embodiments, the Q value calculation function may be as shown in the following formula (1).

[0251] (1)

[0252] in, is a hyperparameter, which can be expressed as the ratio of behavioral benefits. Can be a value less than 1 and greater than 0. For example, The value of can be 0.2. Specifically, The value of can be set according to actual experience or recommended business requirements, or can be determined by parameter tuning methods such as grid search, and there is no restriction on its setting method.

[0253] is the proportion of future income, Can be a value less than 1 and greater than 0. For example, The value of can be 0.9. It should be noted that The value of can be set to other values ​​according to specific circumstances and actual needs. It can be set based on experience or by parameter tuning, and there is no specific limitation on this.

[0254] as well as, is a reward score value, which may include the user's score on whether the application is exposed (action) when performing a feedback action (state). In an embodiment of the present application, the electronic device may pre-set reward score data (hereinafter referred to as preset reward score data), and the preset reward score data may be represented as a k*p matrix, where each element in the matrix is ​​a reward score value, where k is the number of feedback actions, and p is the number of application exposure actions (such as whether the application is exposed). Exemplarily, the preset reward score data may be represented as a 3*2 matrix, where rows represent feedback actions (states) and columns represent application exposure actions (actions).

[0255] For example, the preset reward score data R={"Yes":{"click":40,"remove":-20,"nothing":-6},"No":{"click":-50,"remove":25,"nothing":0}}. That is to say, the score of the behavior combination of application exposure (Yes)-click (click) (the state-action pair of "click"-"Yes", i.e., the situation where the recommended application that the user will click is exposed) is 40 points, and the score of the behavior combination of application non-exposure (No)-click (click) (the state-action pair of "click"-"No", i.e., the situation where the recommended application that the user will click is not exposed) is -50 points; the score of the behavior combination of application exposure (Yes)-remove (remove) (the state-action pair of "remove"-"Yes", i.e., the situation where the recommended application that the user will remove is exposed) is -20 points, and the score of application non-exposure (No)-click (click) (the state-action pair of "click"-"No", i.e., the situation where the recommended application that the user will remove is exposed) is -50 points. The score for the behavior combination of no exposure (No) - removal (remove) (the state-action pair of "remove" - ​​"No", i.e., the recommended apps that users will remove are not exposed) is 25 points; the score for the behavior combination of app exposure (Yes) - no operation (nothing) (the state-action pair of "nothing" - "Yes", i.e., the recommended apps that users will not operate are exposed) is -6 points; the score for the behavior combination of app non-exposure (No) - no operation (nothing) (the state-action pair of "nothing" - "No", i.e., the recommended apps that users will not operate) is 0 points.

[0256] It should be noted that the above The scoring is an example. In the actual process, it can be set as an experience value or set by grid search. There is no restriction on the setting method and specific setting score. ) is higher, the more the state-action pair meets the user's needs, and the lower the reward score is, the less the state-action meets the user's needs. Therefore, since users expect to expose the clicked recommended applications, the application exposure (Yes)-click (click) behavior combination can be configured with a score greater than 0 (such as 40 points as mentioned above), and the application non-exposure (No)-click (click) behavior combination can be configured with a score less than 0 (such as -50 points as mentioned above); and, since users expect not to expose the removed recommended applications, the application non-exposure (No)-remove (remove) behavior combination can be configured with a score greater than 0 (such as 25 points as mentioned above), and the application exposure (Yes)-remove (remove) behavior combination can be configured with a score less than 0 (such as -20 points as mentioned above). Optionally, since the situation of continuous exposure without clicking does not meet the user's needs, a score less than 0 can be configured for the situation of single exposure without clicking, that is, a score less than 0 (such as the above-mentioned -6 points) is configured for the behavior combination of application exposure (Yes)-no operation (nothing), but the score is greater than other negative scores, such as greater than the score of the behavior combination of application non-exposure (No)-click (click), and greater than the score less than 0 (such as the above-mentioned -20 points) for the configuration of application exposure (Yes)-remove (remove). In other embodiments, the Q value calculation function can be shown as the following formula (2).

[0257] (2)

[0258] The meaning of each parameter in Formula 2 can be found in the relevant description of Formula (1), which will not be repeated here.

[0259] After introducing the above related content, we will continue to explain the training method of the exposure model.

[0260] Fig.19 FIG. 1 is a flow chart of an exposure model training method provided in an embodiment of the present application. Fig.19 As shown, the exposure model training method may include the following steps S201 to S212.

[0261] S201, initialization parameters of the model training module.

[0262] In some embodiments, the model training module may initialize the Q value in S201.

[0263] In one embodiment, the Q value may be implemented as a k*p matrix. For example, the Q value may be implemented as a 3*2 matrix.

[0264] At this time, Q={"Yes":{"click":s11,"remove":s21,"nothing":s31},"No":{"click":s12,"remove":s22, "nothing":s32}}.

[0265] Exemplarily, the Q value can be expressed as shown in Table 3 below.

[0266] Table 3 Q value matrix

[0267]

[0268] Among them, s11 is the Q value of the behavior combination of application exposure (Yes)-click (click), s12 is the Q value of the behavior combination of application non-exposure (No)-click (click); s21 is the Q value of the behavior combination of application exposure (Yes)-remove (remove); s22 is the Q value of the behavior combination of application non-exposure (No)-remove (remove); s31 is the Q value of the behavior combination of application exposure (Yes)-no operation (nothing); s32 is the Q value of the behavior combination of application non-exposure (No)-no operation (nothing).

[0269] During the initialization of the Q value, the model training module can set each Q value in the Q value matrix to 0, that is, Q={"Yes":{"click":0,"remove":0,"nothing":0},"No":{"click":0,"remove":0, "nothing":0}}. It should be noted that the initial Q value can also be determined during training optimization, or can be set to other values ​​based on experience and business needs. For example, s11=1 is set (i.e., Q[Yes][click]=1), and other Q values ​​are set to 0. The embodiment of the present application does not impose specific restrictions on the setting of the initial Q value.

[0270] In another embodiment, the Q value can be implemented as a 1*p matrix. For example, a 1*2 matrix, Q={"Yes":v0;"No":v1}. Among them, the initial Q value can be set to 0, that is, Q={"Yes":0;"No":0}. Or v0=1 and v1 are set to 0, that is, Q={"Yes":1;"No":0}. Its setting method is similar to the Q value setting method in the previous embodiment, and the relevant description of the previous embodiment can be referred to, which will not be repeated here.

[0271] In some embodiments, the model training module can also set learning parameters and preset reward points data in S210. The learning parameters can be the ratio of the above behavior benefits. , the proportion of future income The learning parameters and the preset reward score data R can be found in the relevant descriptions of the above section and will not be repeated here.

[0272] In some embodiments, the model training module can also perform other parameters required for Q value learning in S201, such as random probability threshold , which represents the probability of randomly selecting whether to expose during training. For example, it can be less than 1 and greater than 0. For example, the random probability threshold It can be set to 0.95. It should be noted that it can also be set to other values ​​according to actual experience and training requirements, and there is no specific limitation on this.

[0273] In some embodiments, the electronic device may further define a Q value calculation function in S210. For example, the electronic device may define a Q value calculation function by the following example: def calculate_Q( , , next_max):

[0274] = + return newQ

[0275] Through this example, the electronic device can define the name, parameters and specific calculation function of the Q-value calculation function through def, and define the return value newQ of the Q-value calculation function through return.

[0276] And, optionally, the model training model can also define the Q value calculation function as the above formula (2), and its definition method is similar to that of this example, which will not be repeated here.

[0277] In some embodiments, the model training module may also determine the exposure status (“Yes” and “No”) of the application to be processed as an action in Q-value learning.

[0278] S202, the model training module obtains a feedback behavior list (feedbackDict) of the application to be processed.

[0279] Among them, the content of the feedback behavior list can refer to the above description of the embodiment of the present application, and will not be repeated here.

[0280] S203, the model training module obtains the feedback sequence of each scene based on the feedback behavior sequence. For example, Fig. 20 FIG. 2 shows an exemplary Q value learning schematic diagram provided in an embodiment of the present application. Fig. 20As shown, the model training module can obtain the feedback sequence feedbackList2001 of all scenarios, that is, feedbackList=[nothing, remove, click, ...]. Alternatively, if there is only one scenario, or the scenario is not considered, it can be processed directly according to the feedback sequence.

[0281] S204, the model training module obtains the current feedback behavior based on the feedback sequence. In some embodiments, for each scene, the feedback behaviors in the feedback sequence of the scene can be used as the current feedback behavior in sequence. Then, based on each current feedback behavior, the Q value is updated in one round using the Q value calculation function, wherein the following steps S204 to S212 can be performed in each round of update. That is, if the feedback sequence of the current scene includes L feedback actions, the Q value can be updated in L rounds, where L is a positive integer greater than or equal to 2.

[0282] Continuing with the previous example, see Fig. 20 , you can use nothing in the feedback sequence feedbackList2001 of the all scene as the current feedback behavior. After the first round of Q value update, remove is used as the current feedback behavior, and then the second round of Q value update is performed until the last round of update is completed using the last feedback behavior in the feedback sequence feedbackList2001.

[0283] S205: The model training module determines the current exposure behavior (action) of the application to be processed.

[0284] In some embodiments, S205 may include: when the following random selection condition 1 or random selection condition 2 is met, the model training module randomly selects the current exposure behavior (action) as exposure or non-exposure.

[0285] Random selection condition 1: The model training module generates a random number rand, and the random number rand is greater than the random probability threshold Among them, the value range of the random number rand is [0,1], that is, the generated random number is greater than or equal to 0 and less than or equal to 1. By randomly selecting condition 1, through the random number and random probability threshold, the current exposure behavior can be randomly selected with a certain probability, thereby preventing Q value learning from falling into the local optimum, improving the accuracy of Q value learning, and further improving the accuracy of content recommendation.

[0286] Random selection condition 2: The model training module obtains the Q value score Q[Yes][state] of the current feedback behavior-exposure behavior combination, and obtains the Q value score Q[No][state] of the current feedback behavior-non-exposure behavior combination, Q[Yes][state]=Q[No][state]. For example, continue to refer to Fig. 20 , the current feedback behavior is nothing, then read Q[Yes][nothing] and Q[No][nothing] from the Q value data 2002. If the two are equal, then randomly select one between exposure and non-exposure as the current exposure behavior, for example, randomly select non-exposure as the current exposure behavior.

[0287] If neither of the above random selection conditions 1 and 2 is satisfied, when Q[Yes][nothing] is greater than Q[No][nothing], the model training module can determine that the current exposure behavior is exposure. Also, when Q[Yes][nothing] is less than Q[No][nothing], the model training module can determine that the current exposure behavior is non-exposure.

[0288] It should be noted that in the embodiments of the present application, according to the actual scenario and specific needs, the model training module can also select the current exposure behavior through other rules or methods, and there is no specific limitation on this.

[0289] Optionally, when the Q value is a 1*2 matrix, the model training module can determine the current exposure behavior based on the size of Q[Yes] and Q[No]. For example, when the random number rand is greater than the random probability threshold , or when Q[Yes] is equal to Q[No], randomly select one of exposure and non-exposure as the current exposure behavior. Otherwise, when Q[Yes] is greater than Q[No], the current exposure behavior is determined to be exposure; and when Q[Yes] is less than Q[No], the current exposure behavior is determined to be non-exposure.

[0290] S206, the model training module reads the Q value score of the behavior combination of the current feedback behavior and the current exposure behavior from the Q value data 2002 before the update as the Q value score oldQ to be updated. The Q value data before the update may be the Q value data updated in the previous round. In the first round of update, the model training module may use the initialized Q value as the Q value data 2002 before the update.

[0291] For example, see Fig. 20, Q value data 2002 is the initialized Q value (i.e., a 3*2 matrix, each Q value in the matrix is ​​0). Since the current feedback behavior in this round of update is nothing and the current exposure behavior is No, the model training module can use the Q value score Q[No][nothing] of the no operation-no exposure behavior combination as the Q value score oldQ to be updated.

[0292] Optionally, if the Q value data is a 1*2 matrix, the Q value of the current exposure behavior may be read. For example, if the current exposure behavior is no exposure, Q[No] may be determined as the Q value score oldQ to be updated.

[0293] S207, the model training module reads the reward score value raward of the behavior combination of the current feedback behavior and the current exposure behavior. Fig. 20 The model training module can read the reward score value reward of the behavior combination of the current feedback behavior and the current exposure behavior from the preset reward score data 2003, that is, R[No][nothing]=0.

[0294] S208, the model training module reads the next feedback action of the current feedback action in the feedback sequence, and determines the future maximum benefit corresponding to the next feedback action. Fig. 20 , the next feedback action next_state after nothing in the feedback sequence feedbackList2001 is remove, and the model training module can determine the next feedback action of the current feedback action as remove. Optionally, in the case where the current feedback action is the last feedback action of the feedback sequence, the model training module can determine the current feedback action as the next feedback action.

[0295] And, the model training module can obtain the Q-value score Q[No][next_state] of the next feedback behavior-not exposing the behavior combination and the Q-value score Q[Yes][next_state] of the next feedback behavior-exposure of the behavior combination. Then, the larger value of the two Q-value scores is determined as the future maximum benefit next_max. That is, next_max=max(Q[Yes][next_state], Q[No][next_state]). Optionally, in the case where the current feedback behavior is the last feedback action in the feedback sequence, the value of the future maximum benefit next_max can be 0.

[0296] For example, see Fig. 20, the model training module can determine the larger value of Q[Yes][remove] and Q[No][remove], that is, max(Q[Yes][remove],Q[No][remove]) and determine the larger value as next_max.

[0297] It should be noted that if the Q-value calculation function does not include the maximum future benefit, the model training module may not execute step S208.

[0298] S209, the model training module generates an updated Q-value score based on the reward score value raward, the future maximum benefit next_max, and the Q-value score to be updated oldQ, and updates the Q-value data using the updated Q-value score.

[0299] For example, see Fig. 20 , the model training module can substitute the reward score value raward, the future maximum benefit next_max, and the Q value score to be updated oldQ into the Q value calculation function 2005 (such as the above formula 1) to obtain the updated Q value score newQ. Then, the model training module can replace the Q value score to be updated oldQ in the Q value data 2002 with the updated Q value score newQ to obtain the Q value data after this round of update. For example, continue to refer to Fig. 20 , after calculating the updated Q-value score newQ, Q[No][nothing] in the Q-value data can be updated.

[0300] It should be noted that when the Q-value calculation function does not include the future maximum benefit next_max, the reward score value raward and the Q-value score to be updated oldQ can be substituted into the Q-value calculation function (such as the above formula 2) to obtain the updated Q-value score newQ.

[0301] S210, the model training module determines whether the current feedback behavior is the last feedback behavior in the feedback sequence. If the determination result is no, step S211 is executed. If the determination result is yes, step S212 is executed.

[0302] S211: When the current feedback behavior is not the last feedback behavior in the feedback sequence, the model training module determines the next feedback behavior of the current feedback behavior as a new current feedback behavior, and returns to step S204.

[0303] S212, when the current feedback behavior is the last feedback behavior in the feedback sequence, the model training module obtains the Q value of the application to be processed in the scenario based on the updated Q value data.

[0304] In some embodiments, the Q value of the to-be-processed application in the scene may include: an exposed Q value score and a non-exposed Q value score, wherein a higher Q value score indicates a higher expectation.

[0305] For example, the model training module can obtain the Q value score Q[Yes] of exposure based on the sum of Q[Yes][click], Q[Yes][remove], and Q[Yes][nothing] in the Q value data obtained after the loop ends, and obtain the Q value score Q[No] of non-exposure based on the sum of Q[No][click], Q[No][remove], and Q[No][nothing] in the Q value data obtained after the loop ends.

[0306] For example, if the Q value data obtained in the loop is Q = {'Yes': {'click': 73, 'remove': 0, 'nothing': 9}, 'No': {'click': -5, 'remove': 0, 'nothing': 28}}. Then the Q value of the application to be processed in this scene can be expressed as Q = {'Yes': 82, 'No': 23}. In another exemplary embodiment, the model training module can also set different weights for Q[Yes][click], Q[Yes][remove] and Q[Yes][nothing], and then perform weighted summation on Q[Yes][click], Q[Yes][remove] and Q[Yes][nothing], and determine the weighted summation result as the Q value score of the exposure. Similarly, the model training module can also perform weighted summation on Q[No][click], Q[No][remove], and Q[No][nothing], and determine the weighted summation result as the Q-value score of non-exposure.

[0307] In other embodiments, if the Q value data is a 1*2 matrix, the model training module can determine the Q value data obtained after the loop ends as the Q value to be processed and applied in the scene, that is, the two values ​​of the Q value data can be: the exposed Q value score Q[Yes] and the non-exposed Q value score Q[No].

[0308] It should be noted that the model training module can determine the Q value of other scenarios through the above steps S201 to S211, and its specific implementation method is similar to the above content and will not be repeated here.

[0309] In some embodiments, the model training module can generate a Q-value learning result QValueTable based on the Q-values ​​of multiple scenes. Exemplarily, the Q-value learning result QValueTable={"home": {"Yes":v11, "No":v12}, "company": {"Yes":v21, "No":v22}, "all": {"Yes":v31, "No":v32}}. Among them, v11 is the Q-value score of the module to be processed exposed in the home scene, v12 is the Q-value score of the module to be processed not exposed in the home scene, v21 is the Q-value score of the module to be processed exposed in the company scene, v22 is the Q-value score of the module to be processed not exposed in the company scene, v31 is the Q-value score of the module to be processed exposed in the all scene, and v32 is the Q-value score of the module to be processed not exposed in the all scene.

[0310] It should be noted that in the embodiments of the present application, other scenarios can also be set according to actual business conditions and specific business needs, and there is no specific limitation on this.

[0311] After introducing the model training part, we will continue to explain the exposure scheme processing part.

[0312] Fig.21 FIG. 1 is a flow chart of an exposure scheme processing method provided in an embodiment of the present application. Fig.21 As shown, the exposure scheme processing method may include the following steps S301 to S306.

[0313] S301, the exposure scheme processing module obtains the original exposure scheme of the application to be processed.

[0314] In some embodiments, the exposure plan generation module can generate an original exposure plan for the application to be processed. The original exposure plan can be an exposure plan for a preset exposure date (such as the current date or tomorrow, etc., the date for which the exposure plan needs to be estimated) or a preset time period of the preset exposure date. The specific content of the original exposure plan can refer to the relevant description of the above part of the embodiment of the present application, and no specific limitation is made to this.

[0315] Exemplarily, the original exposure scheme in the embodiment of the present application may include exposure time periods for multiple scenes. For example, Fig. 22 FIG. 2 shows a schematic diagram of generating an exemplary target exposure scheme provided in an embodiment of the present application. Fig. 22As shown, the original exposure scheme 2201 may include: original exposure time periods for home scenes, such as 7:00-8:00 and 12:00-13:00; original exposure time periods for company scenes, such as 9:00-9:30 and 10:00-11:00; original exposure time periods for all scenes, such as 7:00-8:00, 9:00-9:30, 10:00-11:00 and 12:00-13:00.

[0316] For example, the original exposure scheme can be expressed as {'home':[[7:00,8:00],[12:00,13:00]],'company':[[9:00,9:30],[10:00,11:00]],'all':[[7:00,8:00],[9:00,9:30],[10:00,11:00],[12:00,13:00]]}.

[0317] It should be noted that the original exposure scheme can also be implemented in other forms according to actual scenarios and needs. Please refer to the relevant description of the above part of the embodiment of the present application, and will not be repeated here.

[0318] S302, the exposure scheme processing module obtains the Q value learning result of the application to be processed. Exemplarily, the Q value learning result may include the Q value of different scenarios, wherein the Q of each scenario may include Q[Yes] and Q[No], Q[Yes] represents the intention probability of application exposure in the scenario, and Q[No] represents the intention probability of application non-exposure in the scenario. For example, continue to refer to Fig. 22 , the Q-value learning result 2202 can be expressed as {"home":{"Yes":-4.96,"No":-1.0},"company":{"Yes":v21,"No":v22},"all": {"Yes":v31, "No":v32}}.

[0319] In some embodiments, the exposure scheme generation module may generate a Q-value learning result and send it to the exposure scheme processing module. The Q-value learning result may be generated based on historical exposure data of a historical period before a preset exposure date. For example, Fig.23 FIG. 4 is a schematic diagram showing an exemplary Q value prediction provided by an embodiment of the present application. Fig.23As shown, when the preset exposure date is the 91st day D91, the electronic device can use the historical exposure data of the 90 days before the preset exposure date, that is, the historical exposure data of the 1st day D1-the 90th day D90, to calculate the Q-value learning result Qa. Similarly, when the preset exposure date is the 92nd day D92, the electronic device can use the historical exposure data of the 90 days before the preset exposure date, that is, the historical exposure data of the 2nd day D2-the 91st day D91, to calculate the Q-value learning result Qb. Similarly, when the preset exposure date is the 93rd day D93, the electronic device can use the historical exposure data of the 90 days before the preset exposure date, that is, the historical exposure data of the 3rd day D3-the 92nd day D92, to calculate the Q-value learning result Qc. Among them, the specific content of the Q-value learning result can refer to the relevant description of the above part of the embodiment of the present application, which will not be repeated here.

[0320] It should be noted that Fig.23 An implementation method of updating the Q-value learning results daily is shown. In actual practice, the model training module can also use other time frequencies to update the Q-value learning results, such as updating every 7 days or every half a day, etc., without specific limitations.

[0321] Also, it should be noted that the above embodiment shows an implementation method in which the model training module uses the historical exposure data of the previous 90 days to update the Q-value learning result of the current date. In the actual process, the model training module can also use the historical exposure data of the previous 90 days to calculate tomorrow's Q-value learning result according to specific requirements and actual conditions. For example, the historical exposure data from the 1st to the 90th day is used to update the Q-value learning result of the 92nd day, etc. There is no specific limitation on this.

[0322] S303: The exposure scheme processing module obtains a preset judgment threshold.

[0323] The preset judgment threshold thres may be determined by a grid search. Alternatively, the preset judgment threshold may be set according to actual experience or needs, and no specific limitation is made thereto. Exemplarily, the preset judgment threshold may be less than 0. By setting the preset judgment threshold, excessive interception of the original exposure scheme may be prevented, further improving the user experience of the application recommendation function.

[0324] In some embodiments, each scene may correspond to a preset judgment threshold, and the preset judgment thresholds of different scenes may be the same or different, and there is no limitation on this. For example, the preset judgment threshold may be expressed as thres={'home':-1, 'company':-2, 'all':-1}. In other words, continue to refer to Fig. 22, the preset judgment threshold th1 = -1 in the home scenario, the preset judgment threshold th2 = -2 in the company scenario, and the preset judgment threshold th3 = -1 in the "all" scenario.

[0325] In some other embodiments, without considering the scenario, the threshold generation module may generate a preset judgment threshold.

[0326] S304. The exposure scheme processing module determines whether the exposure suppression condition is satisfied according to the preset judgment threshold and the Q-value learning result. If the judgment result is yes, step S305 is executed; if the judgment result is no, step S306 is executed.

[0327] In some embodiments, the exposure suppression condition may refer to the conditions that need to be satisfied when the application to be processed needs to be suppressed from being exposed. Exemplarily, the exposure suppression condition can be expressed as: Q[Yes] - Q[No] < thres | (Q[Yes] < Q[No] & Q[Yes] < thres). Among them, when at least one of the following exposure suppression sub-conditions 1 and exposure suppression sub-conditions 2 is satisfied, the application to be processed satisfies the exposure suppression condition.

[0328] Exposure suppression sub-condition 1: Q[Yes] - Q[No] < thres.

[0329] Exposure suppression sub-condition 2: Q[Yes] < Q[No] and Q[Yes] < thres

[0330] Exemplarily, for each scenario, the exposure scheme processing module may obtain the preset exposure threshold of the scenario and the Q-value of the scenario, and then determine whether the exposure suppression condition is satisfied. For example, continuing to refer to Fig. 22 , for the home scenario, Q[Yes] = -4.96, Q[No] = -1.0 in its Q-value, and the preset judgment threshold th1 = -1. It satisfies exposure suppression sub-condition 1 and exposure suppression sub-condition 2. At this time, it can be determined that the application to be processed satisfies the exposure suppression condition in the home scenario.

[0331] It should be noted that according to actual requirements and specific application scenarios, the exposure suppression condition in the embodiments of the present application can also be set in other forms, such as Q[Yes] being less than Q[No], or Q[Yes] being less than Q[No] and Q[Yes] < thres, etc., and no specific limitation is made thereto.

[0332] S305. When the application to be recommended satisfies the exposure suppression condition, the exposure scheme processing module deletes the original exposure time period in the original exposure scheme.

[0333] In some embodiments, the exposure scheme processing module may delete the original exposure time period of the preset exposure date to achieve exposure suppression of the original exposure scheme. Fig.23 The exposure scheme processing module can use the Q value learning result Qa (from the first day D1 to the 90th day D90) to suppress the exposure of the original exposure scheme on the 91st day D91. And the electronic device can use the Q value learning result Qb (from the second day D2 to the 91st day D91) to suppress the exposure of the original exposure scheme on the 92nd day D92.

[0334] In some embodiments, for each scene, the exposure scheme processing module may delete the original exposure time period of the scene if it is determined that the application to be processed meets the exposure suppression condition in the scene. Fig. 22 , if the application to be processed meets the exposure suppression condition in the home scene, the original exposure time periods [7:00-8:00] and [12:00-13:00] of the home scene are deleted. In other words, if the electronic device is in the home scene at [7:00-8:00] and [12:00-13:00] on the current date, the application to be processed will not be exposed.

[0335] S306: When the application to be recommended does not meet the exposure suppression condition, the exposure scheme processing module performs exposure according to the original exposure scheme.

[0336] For example, see Fig. 22 In the target exposure scheme 2203, if the application to be processed does not meet the exposure suppression condition in the company scene and the general scene, the original exposure time period of the company scene and the general scene is retained.

[0337] It should be noted that, in the embodiment of the present application, steps S304 to S306 may be performed for each scene to respectively determine whether the electronic device needs to be exposed according to the original exposure scheme in each scene.

[0338] In some embodiments, before S304, the exposure scheme processing module may also determine whether the current scene needs to be subjected to exposure suppression. If it is determined that the current scene needs to be subjected to exposure suppression, S304 is continued to be executed. If it is determined that the current scene does not need to be subjected to exposure suppression, the exposure is directly performed according to the original exposure scheme. Alternatively, when the scene does not need to be considered, it may be determined whether the application to be processed needs to be subjected to exposure suppression.

[0339] In one embodiment, the electronic device can set an exposure tag for a preset scene, and the exposure tag is used to mark that the preset scene needs to be exposed according to the original exposure scheme. Accordingly, when the exposure scheme processing module detects the exposure tag, it can be determined that the current scene does not need to be exposed.

[0340] In one example, the electronic device can set an exposure label for the target application, such as adding exposure labels to applications that do not require exposure suppression, such as applications with high user click probability, high usage frequency, and commonly used applications, or adding exposure labels to some scenes of the application. Alternatively, the electronic device can add exposure labels to appropriate applications based on actual experience and specific needs, without any restrictions.

[0341] In another example, after step B2, that is, after determining that the number of feedback behaviors in the feedback sequence is insufficient during data preprocessing, the electronic device sets an exposure label for a preset scene corresponding to the feedback sequence after clearing the feedback behaviors in the feedback sequence.

[0342] In another example, the electronic device can determine the confidence of the Q value. When the confidence is less than a preset confidence threshold, an exposure label can be added for the scene in which the application to be processed corresponding to the Q value is located (when Q value learning is performed according to the scene) or the application to be processed corresponding to the Q value (when Q value learning is performed with the application to be processed as the granularity).

[0343] It should be noted that when historical exposure data is divided into multiple categories (such as weekends, weekdays, and multiple exposure periods every day) for exposure suppression respectively, exposure labels can be set for the multiple categories.

[0344] In other embodiments, the electronic device may also set Q[Yes] and Q[No] of applications that do not require exposure, or scenes where parts of applications do not require exposure, to 0, or other scores that can make them not exposed, such as Q[Yes] being greater than Q[No], without specific limitation.

[0345] In the embodiment of the present application, applications that do not require exposure suppression or scenarios where applications do not require exposure suppression may also be marked in other ways, and there is no specific limitation on this.

[0346] The embodiment of the present application can use historical exposure data to judge the user's recommendation needs for the exposed applications according to the user's feedback on the exposed applications, so that unnecessary applications can be exposed and intercepted in time, reducing disturbance to the user, improving the user's recommendation experience, and increasing the user's exposure click rate.

[0347] Furthermore, it should be noted that, by configuring the reward point data, the embodiment of the present application can timely intercept the exposure of applications that are not needed by users, such as those that have been continuously exposed for multiple times but have not been clicked, thereby avoiding disturbing the users and improving the users' recommendation experience and exposure click-through rate.

[0348] And, compared with the use of manual rules to configure exposure suppression strategies such as continuous exposure for a preset number of times without clicking, continuous exposure for a preset number of days without clicking, etc., manual rules often require the use of manual experience, which is inefficient, inaccurate, and inflexible. The embodiment of the present application uses historical exposure data for Q-value learning, and uses the Q-value learning results to intercept exposure. Since the Q-value learning results can estimate the exposure probability of the application to be processed based on the user's feedback on the application to be processed, and since the user's feedback on the application to be processed can reflect the user's exposure needs for the application to be processed, the exposure suppression based on the Q-value learning results can determine whether the user has an exposure need for the application to be processed, and timely and accurately intercept unnecessary exposure periods, thereby improving exposure accuracy. And since the Q-value learning results can be updated regularly every day, the flexibility of exposure interception is improved.

[0349] In addition, the embodiment of the present application is equivalent to adding a judgment gate after the native exposure plan is generated and before exposure, without changing the generation process of the original exposure plan, and the embodiment of the present application only needs to configure a small number of parameters to be used, and can realize quick plug-in and unplug operations, which is easy to implement.

[0350] In some embodiments, the application recommendation method of the embodiment of the present application may further include the step of determining preset reward score data and preset judgment threshold. Specifically, the threshold acquisition module of the electronic device may determine the reward score and the preset judgment threshold through grid search. Specifically, the application recommendation method may further include the following step E1.

[0351] Step E1, the threshold acquisition module determines multiple candidate reward point data, and determines multiple candidate judgment thresholds. Among them, the specific form of each candidate reward point data is similar to the preset reward point data in the above-mentioned part of the embodiment of the present application, and you can refer to the relevant description thereof, which will not be repeated here. And, the candidate judgment threshold is similar to the specific form of the preset judgment threshold in the above-mentioned part of the embodiment of the present application, and you can refer to the relevant description of the above-mentioned embodiment, which will not be repeated here.

[0352] Step E2, the threshold acquisition module determines multiple parameter combinations based on multiple candidate reward score data and multiple candidate judgment thresholds, wherein each parameter combination includes a candidate reward score and a candidate judgment threshold.

[0353] For example, if expressed in a table form, the parameter combination can be expressed as shown in Table 4 below.

[0354] Table 4

[0355]

[0356] As shown in Table 4 above, if there are p candidate reward score data and q candidate judgment thresholds q, p*q parameter combinations can be generated.

[0357] Step E3, the threshold acquisition module uses a grid search method to determine the parameter combination corresponding to the optimal click parameter, determines the candidate reward score in the parameter combination as the preset reward score data, and determines the candidate judgment threshold in the parameter combination as the preset judgment threshold.

[0358] In step E3, the click parameter may refer to parameters such as click-through rate (CTR), user click-through rate (UCTR), number of click users, total recommendation revenue, recommendation return rate (ROI), etc. that can reflect the user's feedback attitude towards exposure. Among them, the click rate may refer to the ratio of the number of clicks to the number of exposures. The user click rate may refer to the ratio of the number of click users to the number of exposed users, and the recommendation return rate may be the ratio of click revenue to exposure cost. It should be noted that in the embodiment of the present application, the click parameter may also be other parameters that the application recommendation service pays attention to, such as the user's feedback score for the application recommendation, etc., and there is no specific limitation on this.

[0359] And, in step E3, in some embodiments, a cross-search method can be used to determine the preset bonus point data and the preset judgment threshold. Exemplarily, the better preset bonus point data can be determined based on the click parameters, and then the better candidate judgment threshold can be determined in the parameter combination corresponding to the better candidate judgment threshold, and then the preset bonus point data can be tuned based on the better candidate judgment threshold, ..., until the parameter combination corresponding to the optimal click parameter is determined. In other embodiments, the preset bonus point data can be determined based on the click parameters, and then the preset judgment threshold can be determined based on the click parameters. In some other embodiments, the click parameters corresponding to each parameter combination can be calculated, and then the parameter combination corresponding to the optimal click parameters can be selected.

[0360] In one embodiment, in step E3, for each parameter combination, the threshold acquisition module may execute the following steps E31 to E33.

[0361] Step E31, the threshold acquisition module acquires a verification data set. The verification data set includes historical exposure data of a historical verification duration. The historical verification duration is longer than the historical time period. For example, if the historical verification duration is a first preset historical number of days (such as 180 days, etc.), and the historical time period is a second preset historical number of days (such as 90 days, etc.), the first preset historical number of days is longer than or equal to the second preset historical number of days.

[0362] The verification data set may include historical exposure data of a single user or historical exposure data of multiple users, without any specific limitation.

[0363] In one example, the threshold acquisition module can use a pure end-side optimization method to perform a grid search on the preset judgment threshold and preset reward points data. In this way, the verification data set can be the historical exposure data of the electronic device itself to perform a grid search on the preset judgment threshold and preset reward points data. Through this embodiment, since the historical feedback data does not need to be uploaded to the cloud at all, the confidentiality and security of user data can be further improved, and the application recommendation method provided in the embodiment of the present application can also be promoted and applied to more businesses that do not meet the cloud conditions. In addition, it gets rid of the limitations of business dependence, data loss during cloud collection, end-cloud inconsistency, and the preset judgment threshold and preset reward points data obtained are the most suitable for users. In addition, the grid search has a timeliness guarantee, does not rely on cloud-optimized parameters, and there is no result deviation caused by parameter failure.

[0364] In another example, the threshold acquisition module can use a cloud-side + terminal-side combination to perform a grid search, that is, the threshold acquisition module can use the historical exposure data on the cloud to perform a grid search of preset judgment thresholds and preset reward points data based on the historical exposure data of the electronic device itself. Exemplarily, the cloud can use the historical exposure data on the cloud to calculate the reference judgment threshold and reference reward points data through a grid search. The cloud sends the reference judgment threshold and reference reward points data to the threshold acquisition module of the electronic device. The threshold acquisition module uses the received parameters as seeds and uses its own historical behavior data to spread around to find the parameter combination corresponding to the optimal click parameters.

[0365] Optionally, in order to improve the accuracy of the preset judgment threshold and preset reward point data, the cloud can select historical exposure data that matches the electronic device. For example, the cloud can select historical exposure data of scenes similar to the current scene. It should be noted that the cloud can also select historical exposure data of similar users to the user of the electronic device (for example, users with similar usage habits, the same user portrait, etc.).

[0366] Compared with the pure end-side method, this example can further improve the speed of obtaining the preset judgment threshold and preset reward point data, and ensure the privacy of the end-side data.

[0367] In another embodiment, when the electronic device's own historical exposure data is insufficient, for example, the amount of historical exposure data is less than 90 days, or the feedback behaviors generated by the historical exposure data are too few (for example, the data links are less than 10), the historical exposure data in the cloud can be obtained, or the preset reward point data and preset judgment threshold pre-generated in the cloud can be directly obtained, thereby achieving a cold start of the electronic device.

[0368] Step E32, for each parameter combination, if the verification data set includes r days of historical exposure data, for i days therein, the threshold acquisition module can use the historical exposure data from the i-th day to the i+89th day (a total of 90 days), use the candidate reward score data of the parameter combination as the preset reward score, and use the candidate judgment threshold of the parameter combination as the preset judgment threshold to calculate the Q value, and then use the calculated Q value to suppress the exposure plan of the i+90th day, and obtain the target exposure plan of the i+90th day, and obtain s target exposure plans in total. Wherein, i is any positive integer, i+89 is less than or equal to r, and s is an integer greater than or equal to 1.

[0369] Step E33, the threshold acquisition module calculates the click parameter corresponding to the parameter combination based on the click data of the s target exposure schemes. For example, taking the number of clicks as the click rate, if the application to be processed is exposed 80 times in total and the user clicks 9 times within s days corresponding to the s target exposure schemes, the click rate is 9 / 80. It should be noted that the calculation method of other click parameters is similar to that of the embodiment of the present application, which will not be described in detail.

[0370] Exemplarily, if the validation data set includes 180 days of historical exposure data, the Q value can be calculated using the historical exposure data from the 1st day to the 90th day, and the calculation result is used to suppress the original exposure scheme on the 91st day to obtain the number of clicks and exposures on the 91st day after the exposure suppression. Also, the Q value is calculated using the historical exposure data from the 2nd day to the 91st day, and the calculation result is used to suppress the original exposure scheme on the 92nd day to obtain the number of clicks and exposures on the 92nd day after the exposure suppression, ..., the Q value is calculated using the historical exposure data from the 90th day to the 179th day, and the calculation result is used to suppress the original exposure scheme on the 180th day to obtain the number of clicks and exposures on the 180th day after the exposure suppression. The sum of the number of clicks and the sum of the number of exposures from the 91st day to the 180th day is calculated, and the ratio of the sum of the number of clicks to the sum of the number of exposures is calculated to obtain the click-through rate.

[0371] Optionally, the preset reward point data and the preset judgment threshold may be tuned based on the daily added feedback data, so as to ensure the accuracy of the preset reward point data and the preset judgment threshold.

[0372] In some embodiments, the embodiments of the present application also provide another application recommendation method. The difference from the above embodiments is that the historical exposure data of the application to be processed can be divided into multiple categories. For each category, Q value calculation, exposure suppression, etc. can be performed according to the method provided in the embodiments of the present application. The specific implementation method is similar to multiple scenes. Please refer to the relevant description of multiple scenes in the above section, which will not be repeated here.

[0373] In one embodiment, if a day is divided into multiple exposure cycles, the data preprocessing module can generate a feedback list of multiple exposure cycles based on the historical exposure data of each exposure cycle in the historical exposure data. Then, the model training module generates the Q values ​​of the multiple exposure cycles based on the feedback list of the multiple exposure cycles. Then, the exposure scheme processing module can determine whether the exposure cycle needs exposure suppression based on the Q value of each exposure cycle and the preset recommended threshold. If the exposure threshold is required, the original exposure time period of the exposure cycle is deleted from the original exposure scheme. For example, if a day can be divided into a time period of every two time periods, taking the [16:00-18:00] period as an example, if it is determined that the period needs exposure suppression, if the original exposure scheme includes the [16:00-16:30] time period, the time period can be deleted from the original exposure scheme.

[0374] In other embodiments, if a week is divided into working days and weekends, the data preprocessing module can generate a feedback list for working days based on the historical exposure data of working days, and generate a feedback list for weekends based on the historical exposure data of weekends. Then, the model training module generates a Q value for working days based on the feedback list for working days; and, based on the feedback list for weekends, generates a Q value for weekends. Then, the exposure scheme processing module can determine whether exposure suppression is required for working days based on the Q value of working days and the preset recommended threshold. If an exposure threshold is required, the original time period of working days is deleted from the original exposure scheme. Similarly, the exposure scheme processing module can also determine whether exposure suppression is required for weekends based on the Q value of weekends and the preset recommended threshold. If an exposure threshold is required, the original exposure time period of weekends is deleted from the original exposure scheme. For example, if the original exposure scheme is the original exposure scheme for Monday, the original exposure time period in the original exposure scheme can be deleted. Alternatively, weekends and working days can be further divided into different scenarios, and it is determined whether the application to be processed needs exposure suppression in different scenarios of weekends and different scenarios of working days.

[0375] In some embodiments, the present application embodiment also provides another application recommendation method. The difference from the above embodiment is that: Fig.24 FIG. 2 is a schematic diagram showing another application recommendation method provided by an embodiment of the present application. Fig.24 As shown, the electronic device includes multiple application recommendation services, such as application recommendation service 1 to application recommendation service T, where T is any integer greater than or equal to 2. Each application recommendation service can generate a historical exposure plan, such as an original exposure plan T, according to a recommendation algorithm, wherein the recommendation algorithms used by different application recommendation services can be the same or different.

[0376] And, if Fig.24As shown, each application recommendation service can perform Q value learning based on its own historical exposure data to obtain its own Q value learning result, and then perform a separate exposure suppression process on its own original exposure scheme based on its own Q value learning result to obtain target exposure data. It should be noted that each application recommendation service can have its own application pool, and then select applications from its own application pool as applications to be processed for exposure suppression.

[0377] Furthermore, it should be noted that the preset judgment thresholds and preset reward points data for different application recommendation services may be different or the same, and there is no limitation on this.

[0378] In some embodiments, the embodiments of the present application also provide another application recommendation method. The difference from the above embodiments is that the electronic device can set a switch control for exposure suppression in the setting interface. After the user performs an on operation on the switch control, the electronic device can execute the application recommendation method of the embodiment of the present application to suppress exposure, that is, the electronic device can obtain historical exposure data, and calculate the Q value based on the historical exposure data, and perform exposure suppression on the application to be processed according to the Q value calculation result. And, after the user performs a off operation on the switch control, the electronic device may not suppress exposure, that is, the electronic device can expose according to the original exposure scheme.

[0379] It is understandable that, in order to realize the above functions, the electronic device includes hardware and / or software modules corresponding to the execution of each function. In combination with the algorithm steps of each example described in the embodiments disclosed herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application in combination with the embodiments, but such implementation should not be considered to be beyond the scope of the present application.

[0380] In one example, Fig.25 A schematic block diagram of a device 2500 according to an embodiment of the present application is shown. The device 2500 may include: a processor 2501 and a transceiver / transceiver pin 2502 , and optionally, a memory 2503 .

[0381] The various components of the device 2500 are coupled together via a bus 2504, wherein the bus 2504 includes a power bus, a control bus, and a status signal bus in addition to a data bus. However, for the sake of clarity, various buses are referred to as bus 2504 in the figure.

[0382] Optionally, the memory 2503 may be used for the instructions in the aforementioned method embodiment. The processor 2501 may be used to execute the instructions in the memory 2503, and control the receiving pin to receive a signal, and control the sending pin to send a signal.

[0383] The device 2500 may be the electronic device or a chip of the electronic device in the above method embodiment.

[0384] Among them, all relevant contents of each step involved in the above method embodiment can be referred to the functional description of the corresponding functional module, and will not be repeated here.

[0385] The steps performed by the electronic device in the content recommendation method provided in the above embodiment of the present application may also be performed by a chip system included in the electronic device, wherein the chip system may include a processor and a Bluetooth chip. The chip system may be coupled to a memory so that when the chip system is running, the computer program stored in the memory is called to implement the steps performed by the above terminal 100. The processor in the chip system may be an application processor or a processor other than an application processor.

[0386] It should be noted that, as used in the specification and the appended claims of the present application, the singular expressions "a", "a", "said", "above", "the" and "this" are intended to also include plural expressions, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in the present application refers to and includes any or all possible combinations of one or more listed items. As used in the above embodiments, the term "when..." may be interpreted to mean "if..." or "after..." or "in response to determining..." or "in response to detecting...", depending on the context. Similarly, the phrase "when determining..." or "if (stated condition or event) is detected" may be interpreted to mean "if determining..." or "in response to determining..." or "when (stated condition or event) is detected" or "in response to detecting (stated condition or event)", depending on the context.

[0387] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A content recommendation method, characterized in that: The method is applied to an electronic device, and the method comprises: Displaying a first interface, wherein the first interface includes a first control, and the first control is used to display content recommended to the user; Generate a preset display scheme, the preset display scheme including an exposure period of the first content, the preset display scheme being used to record the first content recommended for exposure by the electronic device and an exposure plan for the first content; Based on the first score, updating the preset display scheme to obtain an updated exposure scheme; Among them, updating the preset display scheme includes: Determining whether the first score meets a preset display condition; If the first score satisfies a preset display condition, displaying the first content in the first control within the exposure period, where the exposure period is the display time of the first content in the first control; If the first score does not satisfy a preset display condition, deleting the exposure period in the preset display scheme, so as not to display the first content in the first control within the exposure period; The exposure period is a time period in a preset display scheme for displaying the first content. The first score is used to characterize the recommendation probability of the first content, the first score is generated by using historical data of the user in a historical time period, the historical data includes data used to reflect the user's historical feedback behavior on the first content displayed in the first control, and the first score includes a Q value of exposure behavior and a Q value of non-exposure behavior; And, the method further comprises: Filtering the historical data to obtain an exposure moment of the first content in a historical time period, a click moment of the first content displayed in the first control by a user in the historical time period, and a removal moment of the first content displayed in the first control by a user in the historical time period; Determine a feedback behavior sequence in a historical time period based on the exposure moment, the click moment, and the removal moment, wherein the feedback behavior sequence includes feedback behaviors in multiple time periods within the historical time period, and the feedback behaviors in the multiple time periods are sorted in chronological order in the feedback behavior sequence, wherein the feedback behavior is a click behavior, a removal behavior, or a no-operation behavior; The feedback behaviors in the feedback behavior sequence are sequentially used as current feedback behaviors, and the following steps are performed: Determining a current display behavior of the first content, where the current display behavior includes exposure or non-exposure; Based on the current behavior combination consisting of the current feedback behavior and the current display behavior, determine the reward score value of the current behavior combination in the preset reward score data, wherein the preset reward score data includes the reward score values ​​of the behavior combinations consisting of different feedback behaviors and different display behaviors, and the high or low reward score value is related to the user's demand for the behavior combination; The Q value of the current display behavior is updated according to the reward score value to obtain the updated Q value of the current display behavior, and the first score is obtained after the Q value update of the last feedback behavior is completed. The first control is a display control for a recommended folder, a recommended card, or a display control for a recommended service, and the first content is a recommended application, a recommended card, or a recommended service.

2. The method according to claim 1, characterized in that The first score includes a first Q value and a second Q value, wherein the first Q value is the Q value of the exposure behavior, and the second Q value is the Q value of the non-exposure behavior, The determining whether the first score satisfies a preset display condition includes: Obtaining a preset judgment threshold; Determine whether a first difference is less than the preset determination threshold, wherein the first difference is a difference between the first Q value and the second Q value; Determine whether the first Q value is less than the second Q value, and whether the first Q value is less than a preset threshold; In the case where the first difference is less than the preset judgment threshold, if the first Q value is less than the second Q value, and the first Q value is less than the preset threshold, it is determined that the first content does not meet the preset display condition; In the case where the first difference is greater than or equal to the preset judgment threshold, if the first Q value is greater than or equal to the second Q value or the first Q value is greater than or equal to the preset judgment threshold, it is determined that the first content meets the preset display condition.

3. The method according to claim 1, characterized in that The determining of the feedback behavior sequence in the historical time period based on the exposure time, the click time, and the removal time includes: dividing the historical time period into a plurality of time periods; Based on the historical data, count the first number, the second number, and the third number in each time period, wherein the first number is the number of times the first content is displayed, the second number is the number of times the user clicks on the first content, and the third number is the number of times the user removes the first content; Determining a feedback behavior in each time period according to the first number, the second number, and the third number; Based on the feedback behaviors in the multiple time periods, the feedback behavior sequence is acquired.

4. The method according to claim 3, characterized in that The step of determining the feedback behavior in each time period according to the first number of times, the second number of times, and the third number of times includes: For each time period, when the number of clicks is greater than or equal to 1, a first label is set for each time period, where the first label is used to mark the user's click behavior on the first content; When the second number is greater than or equal to 1, a second label is set for each time period, where the second label is used to mark the user's removal behavior of the first content; When the first number and the second number are equal to 0, and the third number is greater than or equal to 1, a third label is set for each time period, and the third label is used to mark that the user has no operation on the first content.

5. The method according to claim 1, characterized in that The historical data includes historical data of multiple scenarios; When the first score satisfies a preset display condition, displaying the first content in the first control within the exposure period includes: Acquire the preset display scheme, where the preset display scheme includes a display time period for each scene, and the exposure time period for each scene is the display time of the first content in the first control in each scene; For each scenario, obtaining a first score for each scenario according to historical data of the scenario; In the case where the first score of each scene satisfies the preset display condition, retaining the exposure period of each scene in the preset display scheme; If the first score of each scene does not satisfy the preset display condition, deleting the exposure period of each scene from the preset display scheme; During the adjusted exposure period in the preset display scheme, the first content is displayed in the first control.

6. The method according to claim 5, characterized in that The step of displaying the first content in the first control during the exposure period in the adjusted preset display scheme includes: Acquire the current scene of the electronic device; In the adjusted preset display scheme, obtaining an exposure period of the current scene; When the current moment is within the exposure period of the current scene, the first content is displayed in the first control.

7. The method according to claim 1, characterized in that detecting whether the first content has a fourth tag; When the first content does not have the fourth tag and the first score satisfies a preset display condition, the first content is displayed in the first widget within the exposure period.

8. The method according to claim 7, characterized in that After detecting whether the first content has a fourth tag, the method further includes: In a case where the first content has a fourth tag, the first content is displayed in the first widget within the exposure period.

9. An electronic device, characterized in that: include: one or more processors; Memory; and one or more computer programs, wherein the one or more computer programs are stored on the memory and when the computer programs are executed by the one or more processors, cause the processors to perform the following steps: Displaying a first interface, wherein the first interface includes a first control, and the first control is used to display content recommended to the user; Generate a preset display scheme, the preset display scheme including an exposure period of the first content, the preset display scheme being used to record the first content recommended for exposure by the electronic device and an exposure plan for the first content; Based on the first score, updating the preset display scheme to obtain an updated exposure scheme; Among them, updating the preset display scheme includes: Determining whether the first score meets a preset display condition; If the first score satisfies a preset display condition, displaying the first content in the first control within the exposure period, where the exposure period is the display time of the first content in the first control; If the first score does not satisfy a preset display condition, deleting the exposure period in the preset display scheme, so as not to display the first content in the first control within the exposure period; The exposure period is a time period in a preset display scheme for displaying the first content. The first score is used to characterize the recommendation probability of the first content, the first score is generated by using historical data of the user in a historical time period, the historical data is used to reflect the user's historical feedback behavior on the first content displayed in the first control, and the first score includes a Q value of exposure behavior and a Q value of non-exposure behavior; And, also includes: Filtering the historical data to obtain an exposure moment of the first content in a historical time period, a click moment of the first content displayed in the first control by a user in the historical time period, and a removal moment of the first content displayed in the first control by a user in the historical time period; Determine a feedback behavior sequence in a historical time period based on the exposure moment, the click moment, and the removal moment, wherein the feedback behavior sequence includes feedback behaviors in multiple time periods within the historical time period, and the feedback behaviors in the multiple time periods are sorted in chronological order in the feedback behavior sequence, wherein the feedback behavior is a click behavior, a removal behavior, or a no-operation behavior; The feedback behaviors in the feedback behavior sequence are sequentially used as current feedback behaviors, and the following steps are performed: Determining a current display behavior of the first content, where the current display behavior includes exposure or non-exposure; Based on the current behavior combination consisting of the current feedback behavior and the current display behavior, determine the reward score value of the current behavior combination in the preset reward score data, wherein the preset reward score data includes the reward score values ​​of the behavior combinations consisting of different feedback behaviors and different display behaviors, and the high or low reward score value is related to the user's demand for the behavior combination; The Q value of the current display behavior is updated according to the reward score value to obtain the updated Q value of the current display behavior, and the first score is obtained after the Q value update of the last feedback behavior is completed. The first control is a display control for a recommended folder, a recommended card, or a display control for a recommended service, and the first content is a recommended application, a recommended card, or a recommended service.

10. A computer-readable storage medium, characterized in that: The invention comprises a computer program, which, when executed on an electronic device, enables the electronic device to execute the content recommendation method according to any one of claims 1 to 8.

11. A chip, characterized in that: It comprises a processing circuit and an interface circuit, wherein the interface circuit is used to receive code instructions and transmit them to the processing circuit, and the processing circuit is used to run the code instructions to execute the content recommendation method according to any one of claims 1-8.

Images