Content pushing method, related device and medium

Abstract

The invention provides a content pushing method, a related device and a medium. The method comprises the following steps: determining a target object group to which a target object belongs; a first curve of the target object group for the target category is obtained, the first curve indicates the change of the average single session object adhesion along with a first proportion, and the first proportion is the proportion of the target category in the single session; determining a first integral curve of the first curve; determining a target point on the first integral curve, and enabling an absolute value of a first difference of average curve values on two sides of the target point on the first integral curve to be maximum; and based on the target point, determining a content proportion of pushing the target category to the target object group. According to the embodiment of the invention, the content pushing accuracy after adjustment and optimization can be improved, and the content pushing adjustment and optimization overhead can be reduced. The embodiment of the invention is applied to big data, data processing and the like.

Description

Technical Field

[0001] The present disclosure relates to the field of big data, and in particular to a content push method, related devices, and media. Background Art

[0002] In the internet age, content (such as short videos) is frequently pushed to platform users. End-to-end content push optimization involves analyzing the proportion of content pushed to a specific user's device to determine if there's room for improvement. For example, if too many funny short videos are pushed to a user's device, affecting their engagement with the app, the proportion of these videos could be reduced.

[0003] Currently, end-to-end content push optimization is primarily performed through manual judgment or randomized trials. Manual judgment relies on subjective expert judgment to determine whether content push optimization should be performed for a specific target, resulting in poor accuracy. Randomized trials, which prioritize content in specific categories by gradually approaching the optimal push share for that category, require numerous trials, are costly, and fail to achieve a global optimality. Therefore, a technology that balances optimized push accuracy and optimization overhead is needed. Summary of the Invention

[0004] The present disclosure provides a content push method, related devices, and media, which can improve the accuracy of content push after optimization while reducing the content push optimization overhead.

[0005] According to one aspect of the present disclosure, a content push method is provided, comprising:

[0006] Determine the target audience group to which the target audience belongs;

[0007] Obtaining a first curve of the target object group for the target category, wherein the first curve indicates a change in average single-session object stickiness as a function of a first proportion, where the first proportion is a proportion of the number of contents of the target category in a single session to the total number of contents in the single session;

[0008] determining a first integral curve of the first curve;

[0009] determining a target point on the first integral curve so that the absolute value of a first difference between average curve values on both sides of the target point on the first integral curve is maximized;

[0010] Based on the target point, a proportion of content of the target category to be pushed to the target object group is determined.

[0011] According to one aspect of the present disclosure, there is provided a content pushing device, comprising:

[0012] A first determining unit, configured to determine a target object group to which a target object belongs;

[0013] an acquiring unit, configured to acquire a first curve of the target object group for the target category, wherein the first curve indicates a change in average single-session object stickiness as a function of a first proportion, where the first proportion is a proportion of the number of contents of the target category in a single session to the total number of contents in the single session;

[0014] a second determining unit, configured to determine a first integral curve of the first curve;

[0015] a third determining unit, configured to determine a target point on the first integral curve so that an absolute value of a first difference between average curve values on both sides of the target point on the first integral curve is maximized;

[0016] The push unit is configured to determine, based on the target point, a proportion of content of the target category to be pushed to the target object group.

[0017] Optionally, the acquiring unit is specifically configured to:

[0018] Obtaining conversations occurring in the target object group;

[0019] For each acquired session, determining the first proportion of the session and the single-session object stickiness;

[0020] determining, based on the single-session object stickiness corresponding to each of the first proportions, the average single-session object stickiness corresponding to the first proportions;

[0021] The first curve is determined based on each of the first proportions and the average single-session object stickiness corresponding to the first proportion.

[0022] Optionally, the single session object stickiness includes a single session duration and a number of contents browsed in a single session;

[0023] The acquisition unit is further specifically configured to:

[0024] determining, based on the single session duration corresponding to each of the first proportions, an average single session duration corresponding to the first proportion;

[0025] determining an average number of content viewed in a single session corresponding to the first proportion based on the number of content viewed in a single session corresponding to each of the first proportions;

[0026] The average single-session object stickiness corresponding to the first proportion is determined based on the average single-session duration and the average number of contents browsed in a single session.

[0027] Optionally, the third determining unit is specifically configured to:

[0028] Assuming that the right direction of the horizontal axis of the coordinate system where the first integral curve is located is positive, setting an objective function, wherein the objective function is equal to the first difference between the first average curve value on the right side and the second average curve value on the left side of any point on the first integral curve;

[0029] On the first integral curve, a point at which the objective function is maximized is determined as the target point.

[0030] Optionally, the first average curve value is determined by:

[0031] Obtaining the maximum first proportion of the target object group;

[0032] Determine a first integral value between the arbitrary point on the first average curve and the point corresponding to the maximum first proportion;

[0033] determining a second difference between the maximum first proportion and the abscissa of any one point;

[0034] The first average curve value is determined based on the first integrated value and the second difference.

[0035] Optionally, the second average curve value is determined by:

[0036] Obtaining the minimum first proportion of the target object group;

[0037] Determine a second integral value between the point corresponding to the minimum first proportion and the arbitrary point on the first average curve;

[0038] determining a third difference between the abscissa of the any point and the minimum first proportion;

[0039] The second average curve value is determined based on the second integrated value and the third difference.

[0040] Optionally, the target category is selected from a candidate category library;

[0041] Before obtaining the first curve of the target object group for the target category, the content pushing device further includes:

[0042] a selection unit, configured to select an unselected candidate category from the candidate category library as the target category;

[0043] After determining the proportion of content of the target category to be pushed to the target object group based on the target point, the content pushing device further includes:

[0044] a first adjusting unit, configured to make a first adjustment to the content proportions of the plurality of target categories if the sum of the content proportions of the plurality of selected target categories exceeds 100%;

[0045] a returning unit, configured to return to the step of selecting an unselected candidate category from the candidate category library as the target category if the sum of the content proportions of the selected multiple target categories does not exceed 100%;

[0046] The second adjustment unit is used to make a second adjustment to the content proportions of the multiple target categories if all the candidate categories in the candidate category library have been selected and the sum of the content proportions of the multiple selected target categories still does not exceed 100%.

[0047] Optionally, the first adjustment unit is specifically configured to:

[0048] Dividing the content proportion of each target category by the sum of the content proportions of multiple target categories to obtain a first quotient;

[0049] The first quotient is used as the first adjusted content proportion of the target category.

[0050] Optionally, the second adjustment unit is specifically configured to:

[0051] Dividing the content proportion of each target category by the sum of the content proportions of multiple target categories to obtain a second quotient;

[0052] The second quotient is used as the second adjusted content proportion of the target category.

[0053] Optionally, the selection unit is specifically configured to:

[0054] Obtaining a selection order of candidate categories in the candidate category library;

[0055] According to the candidate category selection order, an unselected candidate category is selected from the candidate category library as the target category.

[0056] Optionally, the selection unit is further configured to:

[0057] Obtaining the number of contents of each candidate category in the target content library;

[0058] Obtaining object preference for each candidate category of content in the target content library;

[0059] The candidate category selection order is determined based on the number of contents in each candidate category and the object preference of the contents in each candidate category.

[0060] Optionally, the first determining unit is specifically configured to:

[0061] Obtain platform object collection and multiple object group division parameters;

[0062] Determining an application order of target segmentation parameters based on the plurality of object group segmentation parameters;

[0063] Applying the target partitioning parameter application sequence to the platform object set to obtain a target result tree corresponding to the target partitioning parameter application sequence, wherein each leaf node of the target result tree corresponds to an object group partitioned according to the target partitioning parameter application sequence;

[0064] The divided object group to which the target object belongs is used as the target object group.

[0065] Optionally, the first determining unit is further specifically configured to:

[0066] Taking the platform object set as the object group to be divided;

[0067] Executing a first process, the first process comprising: applying unselected subject group division parameters from the plurality of subject group division parameters as subject group division parameters to be examined, respectively, to the subject groups to be divided, obtaining divided subject groups corresponding to each of the subject group division parameters to be examined, and determining, based on each of the divided subject groups, a contribution function value corresponding to each of the subject group division parameters to be examined;

[0068] Using the partitioning parameter of the to-be-investigated object group with the largest contribution function value as the target object group partitioning parameter, updating the to-be-divided object group with the partitioned object group, and returning to the first process;

[0069] The obtained multiple target object group division parameters are cascaded in the order of acquisition as the target division parameter application order.

[0070] Optionally, the first determining unit is further specifically configured to:

[0071] For each of the parameters for dividing the subject groups to be examined, determining the sub-contribution function values of the plurality of divided subject groups corresponding to the parameters for dividing the subject groups to be examined;

[0072] The contribution function value corresponding to the to-be-investigated object group division parameter is determined based on the sub-contribution function values of the plurality of divided object groups corresponding to the to-be-investigated object group division parameter.

[0073] Optionally, the first determining unit is further specifically configured to:

[0074] Taking each of the divided object groups as the target object group in turn, returning to the step of obtaining the first curve of the target object group for the target category, to obtain the absolute value of the first difference at the target point as the single-object contribution value of the divided object group;

[0075] Based on the single object contribution value, the sub-contribution function value of the divided object group is determined.

[0076] Optionally, the first determining unit is further specifically configured to:

[0077] Obtaining the number of objects in the divided object group;

[0078] The sub-contribution function value of the divided object group is determined based on the number of objects in the divided object group and the single-object contribution value.

[0079] Optionally, the determining the application order of target segmentation parameters based on the plurality of object group segmentation parameters includes: determining the application order of a plurality of target segmentation parameters based on the plurality of object group segmentation parameters;

[0080] The step of applying the target partition parameter application sequence to the platform object set to obtain a target result tree corresponding to the target partition parameter application sequence includes: applying multiple target partition parameter application sequences to the platform object set respectively to obtain the target result tree corresponding to each target partition parameter application sequence;

[0081] The step of taking the object group to which the target object belongs as the target object group includes: taking the object group to which the target object belongs in the leaf node of the target result tree corresponding to each target partition parameter application sequence as the target object group corresponding to the target partition parameter application sequence;

[0082] After determining the proportion of content of the target category to be pushed to the target object group based on the target point, the content pushing device further includes:

[0083] The fourth determining unit is configured to determine the proportion of the content of the target category to be pushed to the target object based on the proportion of the content of the target category pushed to the target object group corresponding to the determined application order of each target division parameter.

[0084] Optionally, the target category is selected from a candidate category library;

[0085] The number of target partition parameter application orders is determined by:

[0086] Obtaining the number of candidate categories in the candidate category library;

[0087] The number of application orders of the target partitioning parameters is determined to be twice the number of the candidate categories.

[0088] According to one aspect of the present disclosure, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the content push method described above when executing the computer program.

[0089] According to one aspect of the present disclosure, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the content push method described above is implemented.

[0090] According to one aspect of the present disclosure, a computer program product is provided. The computer program product includes a computer program. The computer program is read and executed by a processor of a computer device, so that the computer device executes the content pushing method described above.

[0091] The disclosed embodiment does not employ manual judgment or random experimentation as in the prior art for content push optimization. Instead, it analyzes a first curve based on the relationship between object stickiness and the ratio of the number of content items in the target category to the number of content items in a single session. A session is the period between the start and end of each session. The more content an object browses, or the longer the session, the greater the object's stickiness. If the number of content items in the target category (e.g., comedy) in a session is too small, the object may find it boring and have low stickiness. However, if there is too much content in the same category, the object may perceive it as similar, and the stickiness will also decrease. Therefore, the disclosed embodiment determines a first curve for the target category for the target object group to which the target object belongs. If the horizontal coordinate of the point with the highest stickiness on the first curve is simply used as the target category push ratio, there may be misjudgments caused by the object being offline. Because the object may leave the session midway, the statistics may show that the object browsed a lot of content or stayed for a long time at the target category push ratio, but in reality, the object was offline. To eliminate misjudgments, the disclosed embodiment integrates the first curve to obtain a first integral curve. The absolute value of the first difference between the average curve value on the right side and the average curve value on the left side of any point on the first integral curve actually reflects whether the change in object stickiness is obvious at the target category push ratio at that point. The point where the absolute value of the first difference is the largest is the point where the object stickiness reaches its peak. Using the horizontal coordinate of this point as the target category push ratio can improve object stickiness and improve the accuracy of content push. Since this solution adopts an integral method, the result obtained is accumulated over time. For the surge in object stickiness value caused by object offline, since the surge lasts for a small range on the horizontal axis, it has little effect on the integral value, thereby reducing the impact of object offline on content push optimization. The above process optimizes content push by means of curve analysis, which improves push accuracy while reducing push tuning overhead.

[0092] Other features and advantages of the present disclosure will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present disclosure. The purposes and other advantages of the present disclosure can be realized and obtained by the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] The accompanying drawings are used to provide a further understanding of the technical solution of the present disclosure and constitute a part of the specification. Together with the embodiments of the present disclosure, they are used to explain the technical solution of the present disclosure and do not constitute a limitation to the technical solution of the present disclosure.

[0094] Figure 1 is an architecture diagram of a system to which the content push method according to an embodiment of the present disclosure is applied;

[0095] Figure 2A-2C This is a schematic diagram of an interface in a scenario where an embodiment of the present disclosure is applied to an instant messaging application to view a video content push message;

[0096] Figure 3 is a flowchart of a content push method according to an embodiment of the present disclosure;

[0097] Figure 4 It is a schematic diagram of the first curve of the target category;

[0098] Figure 5 yes Figure 3 Flowchart of obtaining the first curve of the target category in step 320;

[0099] Figure 6 yes Figure 5 Schematic diagram of determining the first proportion of sessions and the stickiness of objects in a single session in step 520;

[0100] Figure 7 yes Figure 5 Flowchart of determining the average single-session object stickiness corresponding to the first proportion in step 530;

[0101] Figure 8 is a schematic diagram of a first integral curve with respect to the first curve;

[0102] Figure 9 is a schematic diagram of the average difference change curve about the first integral curve;

[0103] Figure 10 yes Figure 3 Flowchart of determining the target point on the first integral curve in step 340;

[0104] Figure 11 yes Figure 10 Flowchart of determining the first average curve value in step 1010;

[0105] Figure 12 yes Figure 3 Schematic diagram of determining a target point on the first integral curve in step 340;

[0106] Figure 13 yes Figure 10 Flowchart of determining the second average curve value in step 1010;

[0107] Figure 14 is a flowchart of a method for pushing content of multiple target categories according to an embodiment of the present disclosure;

[0108] Figure 15 yes Figure 14 Step 1410 is a flowchart for determining a target category from a candidate category library;

[0109] Figure 16 yes Figure 15 Step 1510 is a flowchart of obtaining a selection order of candidate categories in the candidate category library;

[0110] Figure 17 yes Figure 14 Step 14211 is a flowchart for performing a first adjustment on the content proportions of multiple target categories;

[0111] Figure 18 yes Figure 14 Step 14222 of the flowchart is about performing a second adjustment on the content proportions of multiple target categories;

[0112] Figure 19 yes Figure 3 Flowchart of step 310 determining the target object group to which the target object belongs;

[0113] Figure 20 yes Figure 19 Flowchart of step 1920 determining the order of application of target partitioning parameters;

[0114] Figure 21 yes Figure 20 Flowchart of determining the contribution function value corresponding to each parameter for grouping the objects to be investigated in step 2020;

[0115] Figure 22 yes Figure 21 Flowchart of determining the sub-contribution function values of the plurality of divided object groups corresponding to the division parameters of the to-be-investigated object groups in step 2110;

[0116] Figure 23 yes Figure 22 Flowchart of determining the sub-contribution function values of the divided object groups in step 2220;

[0117] Figure 24 yes Figure 20 Schematic diagram of determining the order of application of target partitioning parameters in step 2040;

[0118] Figure 25 yes Figure 19 In step 1930, a schematic diagram of a target result tree corresponding to the target partitioning parameter application order is determined;

[0119] Figure 26 is a flow chart of a content push method for applying multiple target division parameters in an order according to an embodiment of the present disclosure;

[0120] Figure 27 is a module diagram of a content push device according to an embodiment of the present disclosure;

[0121] Figure 28According to the embodiment of the present disclosure Figure 3 The terminal structure diagram of the content push method shown;

[0122] Figure 29 According to the embodiment of the present disclosure Figure 3 The server structure diagram of the content push method shown. DETAILED DESCRIPTION

[0123] In order to make the purpose, technical solutions and advantages of the present disclosure more clearly understood, the present disclosure is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present disclosure and are not intended to limit the present disclosure.

[0124] Before further explaining the embodiments of the present disclosure in detail, the nouns and terms involved in the embodiments of the present disclosure are explained. The nouns and terms involved in the embodiments of the present disclosure are subject to the following interpretations:

[0125] Push: This refers to a method whereby a system or platform proactively delivers information, content, or services to users without them explicitly requesting or searching for them. Push content is typically targeted based on user interests, preferences, past behavior, and other personal information to provide personalized content push.

[0126] Randomized controlled experiment (RCT): Also known as AB experiment or randomized trial, RCT in internet products refers to an online experimental method that uses random sampling to compare one or more product features and operational strategies to determine which strategy is more effective. The specific steps include experimental hypothesis, experimental design, experimental results analysis, and experimental decision-making.

[0127] Video Views (VVs) refer to the number of times or duration a video is viewed within a specific time period. Video views can be used to assess the popularity and dissemination of video content. High video views typically indicate audience interest and widespread dissemination. Therefore, video views are a crucial metric for video producers and content platforms, helping them understand audience feedback and guiding content production and promotion strategies. Understandably, short video businesses often aim to increase average VVs per user.

[0128] A session is the time period between when an object begins and ends browsing content. A session occurs when an object enters a short video app and exits it. A session can be used to mark the lifecycle of an object's actions.

[0129] System architecture and scenario description of the application of the embodiments of the present disclosure

[0130] Figure 1 1 is a system architecture diagram of a content push method according to an embodiment of the present disclosure, which includes: a target terminal 110 , the Internet 120 , a gateway 130 , and a push processing server 140 .

[0131] The target terminal 110 is the device used by the target to view the corresponding reminder message for the pushed content. It can be a desktop computer, laptop computer, PDA (Personal Digital Assistant), mobile phone, in-vehicle terminal, home theater terminal, dedicated terminal, and other various forms. Furthermore, it can be a single device or a collection of multiple devices. For example, multiple devices connected via a local area network and sharing a common display device can collectively constitute a terminal. The target terminal 110 can also communicate with the Internet 120 via wired or wireless means to exchange data.

[0132] Gateway 130, also known as a gateway or protocol converter, implements network interconnection at the transport layer and is a computer system or device that performs a conversion function. It acts as a translator between two systems using different communication protocols, data formats, or languages, or even completely different architectures. Gateway 130 also provides filtering and security functions. Messages sent by target terminals 110 to push processing servers 140 are sent to the corresponding push processing servers 140 via gateway 130. Messages sent by push processing servers 140 to target terminals 110 are also sent to the corresponding target terminals 110 via gateway 130.

[0133] The push processing server 140 is a computer system that provides content push services to the target terminal 110. Compared to the target terminal 110, the push processing server 140 has higher requirements in terms of stability, security, and performance. The push processing server 140 can be a single high-performance computer within a network platform, a cluster of multiple high-performance computers, a portion of a single high-performance computer (e.g., a virtual machine), or a combination of portions of multiple high-performance computers (e.g., virtual machines). The push processing server 140 can also communicate with the Internet 120 via wired or wireless means to exchange data.

[0134] The embodiments of the present disclosure can be applied in various scenarios, such as Figure 2A-2C The scenario shown is viewing a video content push message in an instant short messaging application, etc.

[0135] like Figure 2AAs shown, in the interface of the instant short message application in the target terminal 110, the instant short message application can be used to communicate with multiple contacts and can also be used to watch videos. After triggering the "Message" option in the function bar below, the short message list is displayed. The short message list contains message bars with multiple contacts. At 10:30 am, there is a content prompt ( Figure 2A (The small black dot in the upper right corner of the "Short Video" section in the notification bar reminds the recipient of new content push notifications. The recipient can view the pushed content by triggering the "Short Video" option in the function bar below. The recipient can also view pushed content by triggering the "Short Video Selection" push bar.

[0136] like Figure 2B As shown, at 12:00 am, the subject sees the "small black dot" in the interface of the instant short message application and chooses to trigger the "short video" in the function bar below.

[0137] like Figure 2C As shown, after triggering the "Short Video" option, the instant messaging application opens the content push interface and plays the pushed content, which is a video V. The content push interface also displays the number of likes (176,000), favorites (2,392), and comments (2,965) the video has received before 12:00 am. The subject can also like, favorite, and comment on the video in the interface. The video content pushed to the subject is determined by the push processing server 140.

[0138] Current end-to-end content push optimization is primarily performed through manual judgment or randomized trials. Manual judgment relies on subjective expert judgment to determine whether content push optimization should be performed for a specific object, resulting in poor content push accuracy. Randomized trials, by gradually approximating the optimal push share for a specific category by increasing or decreasing the ranking of content within that category, require numerous trials, incur high overhead, and fail to achieve a global optimum. Therefore, a technique is needed that balances post-optimization push accuracy and optimization overhead. The disclosed embodiment of the present invention employs neither manual judgment nor randomized trials for content push optimization, as in related techniques. Instead, it analyzes a first curve representing the relationship between object stickiness and the ratio of the number of content within the target category to the total number of content within a single session. This disclosed embodiment utilizes an integral approach, yielding cumulative results over time. A surge in object stickiness caused by an object going offline has a minimal impact on the integral value, as this surge persists over a relatively short period on the horizontal axis. This minimizes the impact of the object going offline on content push optimization. This process optimizes content push through curve analysis, improving push accuracy while reducing push optimization overhead.

[0139] General description of the embodiments of the present disclosure

[0140] According to an embodiment of the present disclosure, a content pushing method is provided.

[0141] The content push method determines the target content (e.g., funny short videos) of a target category and pushes a notification message about the target content to the target terminal 110. The target object is the person who wants to view the target content. The target content can be a variety of types, such as images, videos, articles, and applications.

[0142] In the internet age, content (such as short videos) is frequently pushed to platform users. End-to-end content push optimization involves analyzing the proportion of content pushed to a specific user's device to determine if there's room for improvement. For example, if too many funny short videos are pushed to a user's device, affecting their engagement with the app, the proportion of these videos could be reduced.

[0143] Currently, end-to-end content push optimization is primarily performed through manual judgment or randomized trials. Manual judgment relies on subjective expert judgment to determine whether content push optimization is appropriate for a specific target, resulting in low accuracy. Manual judgment can determine whether push traffic is significantly concentrated in a single category. For example, if the games category accounts for 80% of push traffic, this is clearly unhealthy and can be manually adjusted. However, when the problem is less obvious, such as when all categories are evenly split across all push traffic, manual judgment cannot provide specific recommendations on whether the current situation is healthy, which direction should be adjusted, or what proportion of traffic each category should receive. Randomized trials, which gradually approach the optimal push share for a specific category by increasing or decreasing the ranking of content within that category, require numerous trials, are costly, and cannot achieve a global optimality. Furthermore, randomized trials can only verify whether the current experimental strategy is superior to the control group; they do not reveal whether there is room for further improvement, making them unable to provide systematic and stable guidance for the business.

[0144] In the disclosed embodiment, content push optimization is performed neither by manual judgment nor by random experimentation, as in related art. Instead, a first curve is analyzed, comparing the object's stickiness to the ratio of the number of target category content items to the number of content items in a single session. The more content an object browses in each session, or the longer the session, the greater the object's stickiness. If the number of target category content items (e.g., funny) in a session is too small, the object may find it uninteresting and have low stickiness. However, if there is too much content in the same category, the object may also perceive it as similar content, and the stickiness will also decrease. Therefore, the disclosed embodiment determines a first curve for the target category for the target object group to which the target object belongs. If the horizontal coordinate of the point with the highest stickiness on the first curve is simply determined as the target category push ratio, there may be misjudgments caused by the object being offline. Because the object may leave midway, the statistics may show that the object browsed a lot of content or stayed for a long time at the target category push ratio, but in fact, it was caused by the object being offline. To eliminate misjudgments, the disclosed embodiment integrates the first curve to obtain a first integral curve. The absolute value of the first difference between the average curve value to the right and the average curve value to the left at any point on the first integral curve actually reflects whether the change in object stickiness is significant at the target category push ratio at that point. The point with the largest absolute value of this first difference is the point where object stickiness reaches its peak. Using the horizontal coordinate of this point as the target category push ratio can improve object stickiness and enhance content push accuracy.

[0145] The content push method of the embodiment of the present disclosure is executed on the server 140. After the target content to be pushed is determined, a reminder message corresponding to the target content is pushed to the target terminal 110 via a network (eg, the Internet 120).

[0146] like Figure 3 As shown, according to one embodiment of the present disclosure, the content push method includes:

[0147] Step 310: Determine the target object group to which the target object belongs;

[0148] Step 320: Obtain a first curve of the target object group for the target category, wherein the first curve indicates a change in average single-session object stickiness as a function of a first proportion, where the first proportion is the proportion of the number of contents of the target category in a single session to the total number of contents in the single session;

[0149] Step 330: determine a first integral curve of the first curve;

[0150] Step 340: determine a target point on the first integral curve so that the absolute value of a first difference between average curve values on both sides of the target point on the first integral curve is maximized;

[0151] Step 350: Based on the target point, determine the proportion of content of the target category to be pushed to the target object group.

[0152] The advantage of the embodiment of step 310-step 350 is that it does not use manual judgment as in the related art, nor does it use random experiments to optimize content push. It uses the method of analyzing the first curve between the object stickiness and the ratio of the number of content in the target category to the number of content in a single session. Since this solution uses the integral method, the result obtained is accumulated over time. For the surge in the object stickiness value caused by the object being offline, since the surge lasts for a small range on the horizontal axis, it has little effect on the integral value, thereby reducing the impact of the object being offline on the content push optimization. The above process optimizes content push by means of curve analysis, which improves the push accuracy while reducing the push optimization overhead.

[0153] The above steps 310 to 350 are described in detail below.

[0154] Detailed description of step 310

[0155] In step 310, the target object group refers to the group to which the target object belongs. The target object group can be a group consisting of all objects on the application platform. The target object group can also be a group consisting of objects with similar characteristics to the target object.

[0156] Target audience characteristics refer to the characteristics of the target audience. Target audiences with common target audience characteristics are likely to share certain common viewing patterns. Therefore, using target audience characteristics for push processing predictions can help improve push processing effectiveness. Target audience characteristics include education level, years of work experience, job characteristics, hobbies, and historical viewing patterns.

[0157] Regarding the educational level feature, subjects with the same educational level tend to be more likely to choose the same type of content related to their educational level. For example, a subject with a college education level may tend to watch videos introducing graduate schools.

[0158] Regarding the work experience feature, subjects with a certain number of years of work experience are more likely to choose content related to that year of work experience. For example, a subject with 10 years of work experience may prefer to watch videos about career advancement.

[0159] In terms of job characteristics, subjects with the same job are more likely to choose the same type of content that is frequently required for their job. For example, subjects who work as lawyers tend to watch videos about legal education.

[0160] For hobby features, subjects with the same hobby may tend to watch the same content. In one example, a subject with a hobby of swimming likes to watch swimming instruction videos.

[0161] For historical viewing content features, an object may like to watch content with the same historical viewing content features. In one example, an object that has watched a 60-second swimming instruction video may also like to watch a swimming venue introduction video.

[0162] In one embodiment, the target object characteristics of the target object may be obtained by obtaining target object characteristics selected in advance by the target object, or by obtaining target object characteristics statistically obtained from the target object's viewing content log.

[0163] Please note that when obtaining the target subject's target characteristics, you must obtain the target subject's consent in advance. Furthermore, the collection, use, and processing of these characteristics will comply with relevant laws, regulations, and standards. When obtaining the target subject's consent, you can obtain their individual permission or consent through a pop-up window or by redirecting them to a confirmation page.

[0164] The advantage of the above embodiment is that by determining the target object group to which the target object belongs, the audience group can be located more accurately, avoiding pushing to irrelevant audiences, thereby pushing content more targetedly and improving push efficiency and accuracy.

[0165] Detailed description of step 320

[0166] In step 320, a first curve of the target object group for the target category is obtained, wherein the first curve indicates a change in average single-session object stickiness with a first proportion, where the first proportion is the proportion of the number of contents of the target category in a single session to the number of contents in a single session.

[0167] Target categories refer to the pre-defined types of content that are pushed to a specific target audience. For short video push scenarios, target categories include film and television variety shows, games, anime, comedy, news, knowledge, food, lifestyle, sports, etc.

[0168] The first curve refers to a curve indicating the change in the average single-session object stickiness of the target object group as a function of the first proportion. The first curve may be a curve connected based on the values corresponding to each divided proportion interval. The proportion interval is the interval of the first proportion pre-divided by the first curve. The values corresponding to adjacent proportion intervals are connected as a line segment, so the first curve may be in the form of a broken line composed of line segments. The average single-session object stickiness refers to the numerical value of the object stickiness of the target object group in a single session of a single target category. Object stickiness refers to the viewing time of an object in a session or the number of contents viewed in a session. The higher the average single-session object stickiness of a target category, the longer the target object group browses the content of the target category or the more contents viewed in a session.

[0169] The first proportion refers to the ratio of the number of target category contents in a single session to the total number of contents in a single session. In other words, the first proportion is the ratio of the number of target category contents (such as funny categories) viewed by the subject in a session to the total number of contents viewed by the subject in the session. Figure 4 As shown, if the target category is the funny category, the average single-session object stickiness of the target object group for the funny category is used as the vertical coordinate, and the first proportion of the target object group for the funny category is used as the horizontal coordinate to construct the first curve of the target object group for the funny category. Wherein, the first proportion is divided based on the interval of 10%, and the divided proportion intervals may include [0, 10%], (10%, 20%], (20%, 30%], (30%, 40%], (40%, 50%], (50%, 60%], (60%, 70%], (70%, 80%], (80%, 90%], (90%, 100%]. Therefore, the connection of the average single-session object stickiness corresponding to each divided proportion interval constitutes the first curve.

[0170] It should be noted that for Figure 4 The first curve shown, the curve function corresponding to this curve can be used to define the category preference function f ik (t). f ik (t) represents the average single-session object stickiness of group k when the distribution share of target category i is t. Therefore, the goal of the first curve is to find the first share t corresponding to the maximum average single-session object stickiness. The process of determining the first share corresponding to the maximum average single-session object stickiness of the first curve is shown in the following formula 1:

[0171] t best =argmaxf ik (t) (Formula 1)

[0172] In formula 1, t bestIndicates the first proportion corresponding to the maximum average single-session object stickiness, f ik (t) represents the category preference function corresponding to the first curve, t represents the first proportion, argmaxf ik (t) represents the solution function f ik (t) The course of the value of the first ratio at the maximum.

[0173] The related art tends to believe that the average single-session object stickiness in the interval of (10%, 20%) is the highest, so the corresponding first proportion is the optimal proportion. However, in reality, there is a misjudgment caused by the object being offline. For example, if the object leaves midway, the final statistics show that the object stickiness at the target category push ratio suddenly reaches a very high level. This is caused by the discreteness of the object and is not the optimal proportion. Therefore, this application will explain in detail in subsequent embodiments how to avoid such problems, find the optimal proportion of the target category, improve object stickiness, and thus improve push accuracy. No specific explanation will be given here.

[0174] Understandably, the target audience's willingness to spend may follow a trend of initially increasing and then decreasing. In other words, if the average per-session engagement of the target audience is too high, it can lead to fatigue or boredom, while if the average per-session engagement is too low, the categories of interest to the target audience will be underexposed, leaving their needs unsatisfied. Therefore, it's important to find the optimal target category share to increase engagement and, consequently, improve push accuracy.

[0175] In one embodiment, Figure 5 As shown, step 320 includes:

[0176] Step 510: Obtain the conversations occurring in the target object group;

[0177] Step 520: For each acquired session, determine a first session percentage and a single session object stickiness;

[0178] Step 530: Determine an average single-session object stickiness corresponding to the first proportion based on the single-session object stickiness corresponding to each first proportion;

[0179] Step 540: Determine a first curve based on each first proportion and the average single-session object stickiness corresponding to the first proportion.

[0180] In step 510, the conversation occurring in the target object group refers to the conversation occurring in any object in the target object group. Figure 6As shown, the target group includes subject A. Subject A's session S1 represents the time period from when subject A starts browsing content to when he stops. Subject A's session S1 runs from 4:20:10 PM to 4:30:10 PM. In session S1, subject A browsed five videos: comedy video V1, comedy video V2, music video V3, food video V4, and game video V5.

[0181] In step 520, the object stickiness of a single session refers to the object stickiness based on a single session. The object stickiness of a single session includes the duration of a single session and the number of contents viewed in a single session. The duration of a single session refers to the total duration of a push content viewed by a single session object. The number of contents viewed in a single session refers to the total number of push contents viewed by a single session object. In this case, the first proportion of the session refers to the proportion of the number of contents of the target category in a single session occurring in the target object group to the number of contents in the single session. Figure 6 As shown, the conversation S1 that occurred in the target object group includes four target categories: comedy, music, food, and games. For the first proportion of each target category in this conversation: comedy accounts for 20%, music accounts for 10%, food accounts for 10%, and games accounts for 10%. Therefore, it can be determined that the single conversation duration included in the single conversation object stickiness is 10 minutes (that is, from 4:20:10 pm to 4:30:10 pm), and the number of content browsed in a single conversation is 5.

[0182] In step 530 , the average single-session object stickiness refers to a value representing the single-session object stickiness of all the sessions occurring in the target object group.

[0183] In one embodiment, if the single-session object stickiness only includes the single-session duration, then based on the single-session object stickiness corresponding to each first proportion, determining the average single-session object stickiness corresponding to the first proportion includes: determining the average single-session duration corresponding to the first proportion based on the single-session duration corresponding to each first proportion; and determining the average single-session object stickiness corresponding to the first proportion based on the average single-session duration. The average single-session object stickiness in this case can be the average single-session duration itself, or it can be the result of multiplying the average single-session duration by a preset duration weight or duration adjustment coefficient. For example, a single session includes four target categories: comedy, music, food, and games. The single-session duration corresponding to the first proportion of each target category in the session is: comedy is 5 minutes, music is 3 minutes, food is 5 minutes, and games is 3 minutes. In this case, the average single-session duration corresponding to the first proportion is (5+3+5+3) / 4 = 4 minutes. Therefore, the average single-session object stickiness can be calculated as the average single-session duration of 4 minutes. If the duration adjustment factor is 2, the average single-session object stickiness can be 2×4=8 minutes; if the duration weight is 0.8, the average single-session object stickiness can be 0.8×4=3.2 minutes.

[0184] In another embodiment, if the single-session object stickiness only includes the number of contents viewed in a single session, then the average single-session object stickiness corresponding to the first percentage is determined based on the single-session object stickiness corresponding to each first percentage. This includes: determining the average number of contents viewed in a single session corresponding to the first percentage based on the number of contents viewed in a single session corresponding to each first percentage; and determining the average single-session object stickiness corresponding to the first percentage based on the average number of contents viewed in a single session. In this case, the average single-session object stickiness can be the average number of contents viewed in a single session itself, or it can be the result of multiplying the average number of contents viewed in a single session by a preset duration weight or duration adjustment coefficient. For example, a single session includes four target categories: comedy, music, food, and games. The number of contents viewed in a single session corresponding to the first percentage for each target category in the session includes: six for comedy, four for music, four for food, and two for games. In this case, the average number of content viewed per session corresponding to the first percentage is (6 + 4 + 4 + 2) / 4 = 4. Therefore, the average object stickiness per session can be 4. If the duration adjustment factor is 3, the average object stickiness per session can be 3 × 4 = 12; if the duration weight is 0.5, the average object stickiness per session can be 0.5 × 4 = 2.

[0185] In another embodiment, Figure 7As shown, step 530 includes:

[0186] Step 710: Determine an average single session duration corresponding to each first proportion based on the single session duration corresponding to each first proportion.

[0187] Step 720: Determine the average number of content views in a single session corresponding to each first proportion based on the number of content views in a single session corresponding to each first proportion.

[0188] Step 730: Determine the average single-session object stickiness corresponding to the first proportion based on the average single-session duration and the average number of contents browsed in a single session.

[0189] In step 710, the average single session duration is the average of the single session durations corresponding to the various first proportions of a session. For example, a session includes four target categories: comedy, music, food, and games. The single session durations corresponding to the first proportions of each target category in this session are: comedy is 6 minutes, music is 2 minutes, food is 4 minutes, and games is 8 minutes. In this case, the average single session duration corresponding to the first proportions is (6 + 2 + 4 + 8) / 4 = 5 minutes.

[0190] In step 720, the average number of content views per session refers to the average number of content views per session corresponding to the various first percentages for a session. For example, a session includes four target categories: comedy, music, food, and games. The number of content views per session corresponding to the first percentage for each target category in that session is: comedy (4 minutes), music (8), food (1), and games (7). In this case, the average number of content views per session corresponding to the first percentage is (4 + 8 + 1 + 7) / 4 = 5.

[0191] In step 730, based on the average single session duration and the average number of contents viewed in a single session, the average single session object stickiness corresponding to the first proportion is determined. This can be done by calculating the average or weighted average of the average single session duration and the average number of contents viewed in a single session.

[0192] When calculating the average single-session object stickiness for each percentage, the average of the average single-session duration and the average number of content viewed per session is used. For example, if the average single-session duration is 10 minutes and the average number of content viewed per session is 6, the average single-session object stickiness is (10 + 6) / 2 = 8. The advantage of using the average to calculate the average single-session object stickiness is that it equally reflects the impact of the average single-session duration and the average number of content viewed per session on determining the average single-session object stickiness.

[0193] When calculating the average per-session object stickiness for each percentage using the weighted average of average session duration and average number of content viewed per session, for example, if the weights for average session duration and average number of content viewed per session are set to 0.6 and 0.4, respectively, and the average session duration is 10 minutes and the average number of content viewed per session is 6, then the average per-session object stickiness is 10 × 0.6 + 6 × 0.4 = 8.4. The advantage of using the weighted average to calculate average per-session object stickiness is that different weights can be set for average session duration and average number of content viewed per session, increasing flexibility in determining average per-session object stickiness.

[0194] In step 540, based on each first proportion and the average single-session object stickiness corresponding to the first proportion, corresponding curve points are determined in the coordinate system where the first curve is located, and the curve points of adjacent proportion intervals are sequentially connected to obtain the first curve.

[0195] The advantage of the above embodiment is that by acquiring conversations occurring within the target group and performing a detailed analysis of each conversation's first percentage and single-session object stickiness, a more comprehensive understanding of the target group's behavior and interactions is achieved. Based on this data, a first curve is constructed, which more accurately captures the target group's behavioral characteristics and trends, thereby improving the accuracy and effectiveness of content push.

[0196] Detailed description of step 330

[0197] In step 330, the first integral curve of the first curve is determined. If the horizontal coordinate of the point with the highest stickiness on the first curve is simply determined as the target category push ratio, there may be misjudgment caused by the object being offline. Since the object may leave halfway, it is statistically shown that the object browsed a lot of content or stayed for a long time at the target category push ratio, but in fact it was caused by the object being offline. In order to eliminate this effect, after obtaining the first curve, the embodiment of the present disclosure will integrate the first curve to determine the first integral curve of the first curve. The first integral curve is used to represent a cumulative preference function.

[0198] The integral of a function represents the cumulative effect of the function over a certain interval and can be understood as the area under the function curve. For example, if there is a function f(x), then its integral can be expressed as ∫f(x)dx, where ∫ is the integral symbol, f(x) is the function to be integrated, and dx represents the integration of x. Therefore, combined with Formula 1, the formula for determining the first integral curve of the first curve in the embodiment of the present disclosure is shown in Formula 2 below:

[0199]

[0200] In Formula 2, F ik (t0) represents the function corresponding to the first integral curve, F ik (t0) represents the cumulative result of the area under the function curve of the first curve before the first proportion t0, that is, the cumulative object stickiness of the average single-session object stickiness of group k when the distribution proportion of target category i is t. For example, Figure 8 Shows the Figure 4 The first integral curve after the integration of the first curve is shown. Due to the existence of outliers, the highest value of the category preference function of the first curve is 20%, but according to the cumulative preference function F ik (t0), we can see that the highest value is around 60%, that is Figure 8 The position of the asterisk.

[0201] Figure 9 The average difference change curve of the first integral curve is shown. The function value corresponding to each proportion in the average difference change curve represents the slope of the function corresponding to the first integral curve, that is, the slope of the function represents the rate of change or growth rate of the function curve at a certain point. For example, for the function f(x), the slope at a certain point x=a can be expressed as f′(a), that is, the derivative value of the function f(x) at the point x=a. If the slope of the function curve at a certain point is positive, it means that the function is increasing at this point; if the slope is negative, it means that the function is decreasing at this point; if the slope is zero, it means that the function is stable at this point. Since the slope of the curve (that is, the first-order derivative) should also be steepest in the area around the optimal distribution ratio, at this time, Figure 9 The highest value of the slope is also around 60%, that is Figure 9 The position of the asterisk.

[0202] The disclosed embodiment does not directly seek the optimal solution for the function of the first curve that is sensitive to outliers. Instead, it relies on a higher-level cumulative preference function, namely, a function of the first integral curve obtained by integrating the first curve. Because this solution uses an integral approach, the result is accumulated over time. The surge in object stickiness caused by an object going offline has a small impact on the integral value because the surge lasts for a short period of time on the horizontal axis, thereby reducing the impact of the object going offline on content push optimization.

[0203] Detailed description of step 340

[0204] In step 340, a target point on the first integral curve is determined so that the absolute value of the first difference between the average curve values on both sides of the target point on the first integral curve is maximized. The first integral curve has been described in detail in the above embodiment and will not be repeated here.

[0205] The target point is the point on the first integral curve where the absolute value of the first difference between the average curve values on both sides is the largest. The horizontal coordinate of the target point on the first integral curve is the optimal first proportion. Figure 8 As shown, the highest value is around 60%, so the point in the 60% interval is the target point.

[0206] The average curve value is the ratio of the accumulated result of the curve in a cumulative interval to the length of the curve mapped on the horizontal coordinate of the cumulative interval. The average curve value on both sides of the target point is the average curve value calculated by dividing the target point's horizontal coordinate by the interval on both sides of the curve.

[0207] The first difference refers to the difference between the average curve values on either side of any point on the first integral curve. Since the difference between the average curve values on either side of any point on the first integral curve may be positive, negative, or zero, in order to be able to compare any multiple points on the first integral curve, the disclosed embodiment takes the absolute value of the difference in the obtained average curve values. In this case, the point with the largest absolute value of the first difference between the average curve values on either side can be used as the target point. The absolute value of the first difference between the average curve values on the right and the average curve values on the left of any point on the first integral curve actually reflects whether the change in object stickiness is significant at the content proportion of the target category at that point. The point with the largest absolute value of the first difference is the point where object stickiness reaches its peak. Using the horizontal coordinate of this point as the content proportion of the target category can improve object stickiness and improve the accuracy of content push. In other words, the absolute value of the slope at the target point is also the largest. For example, the first integral curve includes four points: point A, point B, point C, and point D. The first difference of the average curves on both sides of point A is 3, the first difference of the average curves on both sides of point B is -2, the first difference of the average curves on both sides of point C is 3, and the first difference of the average curves on both sides of point D is -5. At this time, |-5|>|-3|=|3|>|-2|, that is, the absolute value of the first difference of the average curve values on both sides of point D is the largest, and point D is the target point on the first integral curve.

[0208] In one embodiment, Figure 10 As shown, step 340 includes:

[0209] Step 1010: Assuming that the right direction of the horizontal axis of the coordinate system where the first integral curve is located is positive, set an objective function, where the objective function is equal to the first difference between the first average curve value on the right side and the second average curve value on the left side of any point on the first integral curve;

[0210] Step 1020: On the first integral curve, determine the point that maximizes the objective function as the target point.

[0211] In step 1010, if Figure 8As shown, the right side of the horizontal axis of the coordinate system where the first integral curve is located is positive, and the horizontal axis of the first integral curve is the first proportion. The objective function is used to represent the first difference between the first average curve value to the right and the second average curve value to the left of any point on the first integral curve.

[0212] The first average curve value refers to the average cumulative result of the interval to the right of any point on the first integral curve. Figure 11 As shown, the first average curve value is determined by:

[0213] Step 1110: Obtain the maximum first proportion of the target object group;

[0214] Step 1120: Determine a first integral value between any point on the first integral curve and the point corresponding to the maximum first proportion;

[0215] Step 1130: Determine a second difference between the maximum first proportion and the horizontal coordinate of any point;

[0216] Step 1140: Determine a first average curve value based on the first integral value and the second difference.

[0217] In step 1110, the maximum first proportion refers to the maximum value of the target object group on the horizontal axis of the coordinate system where the first integral curve is located. In practice, the maximum first proportion is infinitely close to 1 (or 100%), but does not reach 1 (or 100%). Figure 12 As shown, the maximum first proportion may be 98%, 99%, etc., which is not specifically limited here.

[0218] In step 1120, the process of calculating the first integral value is shown in the following formula 3:

[0219]

[0220] In formula 3, L1 represents the first integral value, t represents any ratio between the first ratio and the maximum first ratio at any point on the first integral curve, and t i Represents the first proportion of any point on the first integral curve, t max Indicates the largest first proportion, F ik (t) represents the function of the first integral curve, that is, the cumulative object stickiness of the average single-session object stickiness of group k when the distribution proportion of target category i is t.

[0221] In step 1130, the second difference refers to the difference between the maximum first proportion and the horizontal coordinate of any point. Figure 12 As shown, for example, the maximum first proportion t max is 98%, the horizontal coordinate t of any point iis 56%, then the second difference is 98%-56%=42% (or 0.42).

[0222] In step 1140, the first average curve value refers to the ratio of the first integral value to the second difference. Therefore, the process of calculating the first average curve value is shown in the following formula 4:

[0223]

[0224] In formula 4, D1 represents the first average curve value, t i Represents the first proportion of any point on the first integral curve, t max Indicates the largest first proportion, t max -t i Indicates the second worst, Indicates that the distribution ratio of calculation group k in target category i is t i is a function of the first integral value on the right side.

[0225] The advantage of the above steps 1110 to 1140 is that the result accumulated over time by the target object group to the right of any point on the first average curve can be determined through the first integral value between any point on the first integral curve and the maximum first proportion. As for the surge in object stickiness value caused by the object being offline, since the surge lasts for a small range on the horizontal axis, it has little effect on the integral value, thereby reducing the impact of the object being offline on the content push optimization. By using the ratio of the first integral value to the second difference, the first average curve value to the right of any point on the first integral curve can be determined, thereby determining the target point based on the difference between the average curve values on the left and right sides of any point. Therefore, the embodiment of the present disclosure optimizes content push by means of curve analysis, which can effectively reduce push tuning overhead while improving push accuracy.

[0226] The second average curve value refers to the average cumulative result of the interval to the left of any point on the first integral curve. Figure 13 As shown, the second average curve value is determined by:

[0227] Step 1310: Obtain the minimum first proportion of the target object group;

[0228] Step 1320: Determine a second integral value between the point corresponding to the minimum first proportion and any point on the first average curve;

[0229] Step 1330: Determine the third difference between the horizontal coordinate of any point and the minimum first proportion;

[0230] Step 1340: Determine a second average curve value based on the second integral value and the third difference.

[0231] In step 1310, the minimum first proportion refers to the minimum value of the target object group on the horizontal axis of the coordinate system where the first integral curve is located. In practice, the minimum first proportion is infinitely close to 0 (or 0%), but does not reach 0 (or 0%). Figure 12 As shown, the minimum first proportion may be 1%, 5%, etc., which is not specifically limited here.

[0232] In step 1320, the process of calculating the second integral value is shown in the following formula 5:

[0233]

[0234] In Formula 5, L2 represents the second integral value, t represents any ratio between the minimum first ratio and the first ratio at any point on the first integral curve, and t i Represents the first proportion of any point on the first integral curve, t min Indicates the minimum first proportion, F ik (t) represents the function of the first integral curve, that is, the cumulative object stickiness of the average single-session object stickiness of group k when the distribution proportion of target category i is t.

[0235] In step 1330, the third difference refers to the difference between the horizontal coordinate of any point and the minimum first ratio. Figure 12 As shown, for example, the minimum first proportion t max is 2%, the horizontal coordinate t of any point i is 56%, then the third difference is 56%-2%=54% (or 0.54).

[0236] In step 1340, the second average curve value refers to the ratio of the second integral value to the third difference. Therefore, the process of calculating the second average curve value is shown in the following formula 6:

[0237]

[0238] In Formula 6, D2 represents the second average curve value, t i Represents the first proportion of any point on the first integral curve, t min Indicates the minimum first proportion, t i -t min Indicates the third difference, Indicates that the distribution ratio of calculation group k in target category i is t i is a function of the second integral value on the left.

[0239] The advantage of the above steps 1310 to 1340 is that the result accumulated over time by the target object group to the left of any point on the first integral curve can be determined by the second integral value between any point on the first integral curve and the minimum first proportion of the target object group. As for the surge in object stickiness value caused by the object being offline, since the surge lasts for a small range on the horizontal axis, it has little effect on the integral value, thereby reducing the impact of the object being offline on the content push optimization. By using the ratio of the second integral value to the third difference, the second average curve value to the left of any point on the first integral curve can be determined, thereby determining the target point based on the difference between the average curve values on the left and right sides of any point. Therefore, the embodiment of the present disclosure optimizes content push by means of curve analysis, which can effectively reduce push tuning overhead while improving push accuracy.

[0240] The objective function is equal to the first difference between the first average curve value on the right side and the second average curve value on the left side of any point on the first integral curve. Therefore, the process of calculating the objective function is shown in the following formula 7:

[0241]

[0242] In formula 7, D(t i ) represents the objective function, D1 represents the first average curve value, and D2 represents the second average curve value.

[0243] In step 1020, on the first integral curve, the point that maximizes the absolute value of the objective function is determined as the target point. Figure 12 As shown, if the first point of the funny category with a proportion of 60% is the point where the absolute value of the objective function is the largest, then based on the target point, it is determined that the proportion of funny category content pushed to the target object group is 60%.

[0244] The advantage of the above steps 1010 to 1020 is that by setting the objective function and determining the point on the first integral curve that maximizes the objective function, the target point that maximizes the objective function can be accurately found. Therefore, the embodiment of the present disclosure adopts an integral method to obtain a result accumulated over time. For the surge in the object stickiness value caused by the object being offline, since the surge lasts for a small range on the horizontal axis, it has little effect on the integral value, thereby reducing the impact of the object being offline on the content push optimization. The above process optimizes content push by means of curve analysis, which improves push accuracy while reducing push tuning overhead.

[0245] Detailed description of step 350

[0246] In step 350, based on the target point, the percentage of content of the target category to be pushed to the target audience group is determined. The process for determining the target point has been described in detail in the above embodiment and will not be repeated here. A first integral curve corresponds to a target category to be pushed to the target audience group. Based on the first percentage of the target point in the first curve, the percentage of content of the target category to be pushed to the target audience group can be determined.

[0247] For a single target category, a separate analysis is performed. After determining the optimal target point for the target object group under that target category, the first proportion of the target point corresponding to each target category can be used to determine the proportion of content in the target category to be pushed to the target object group. For example, the target categories include comedy, music, news, food, and games. After determining the target points based on curve analysis according to the disclosed embodiment, it is determined that the optimal proportion of content in the comedy category to be pushed to the target object group is 40%, the optimal proportion of content in the music category to be pushed to the target object group is 15%, the optimal proportion of content in the news category to be pushed to the target object group is 30%, the optimal proportion of content in the food category to be pushed to the target object group is 2%, and the optimal proportion of content in the game category to be pushed to the target object group is 13%. At this time, the proportion of content in the target category is pushed to the target object group based on the determined content proportion. In other words, the objects in the target object group do not like to watch content in the food category very much, but prefer to watch content in the comedy category. Therefore, the disclosed embodiment can effectively reduce the push tuning overhead while improving push accuracy.

[0248] Detailed description of content push based on multi-category combination in the embodiment of the present disclosure

[0249] In one embodiment, multiple target categories can be pushed to a target group, and each target category is selected from a candidate category library. The candidate category library is a library structure that stores multiple candidate categories. The candidate categories stored in the candidate category library include film and television variety shows, games, anime, comedy, news, knowledge, gourmet food, lifestyle, and sports.

[0250] In one embodiment, Figure 14 As shown, before step 320, the content push method of the embodiment of the present disclosure further includes:

[0251] Step 1410: Select an unselected candidate category from the candidate category library as the target category.

[0252] For content push scenarios involving multiple categories, an unselected candidate category is selected from the candidate category library as the target category. After determining the content percentage of this target category, the next unselected candidate category is selected from the candidate category library as the target category. This allows the optimal target point corresponding to the target group across multiple target categories to be determined. Based on the first percentage of the target points corresponding to the multiple target categories, the percentage of content within the target category within the multi-category combination to be pushed to the target group is determined.

[0253] In one embodiment, the embodiment of the present disclosure may randomly select an unselected candidate category from the candidate category library as the target category. That is, the embodiment of the present disclosure may randomly determine the order of selecting categories.

[0254] In another embodiment, Figure 15 As shown, step 1410 includes:

[0255] Step 1510: Obtain the selection order of candidate categories in the candidate category library;

[0256] Step 1520: According to the candidate category selection order, select an unselected candidate category from the candidate category library as the target category.

[0257] In step 1510, the candidate category selection order refers to the pre-set order in which candidate categories are selected from the candidate category library. Based on this order, it can be determined that some candidate categories whose combined content percentage exceeds 100% will be excluded from consideration. Furthermore, by progressively selecting candidate categories according to this order, we ensure that the required candidate categories are fully evaluated, avoiding the omission or neglect of certain categories.

[0258] In one embodiment, Figure 16 As shown, step 1510 includes:

[0259] Step 1610: Obtain the number of content in each candidate category in the target content library;

[0260] Step 1620: Obtain the object preference of each candidate category of content in the target content library;

[0261] Step 1630: Determine the candidate category selection order based on the number of contents in each candidate category and the object preference of the contents in each candidate category.

[0262] In step 1610, the target content library refers to a library structure that stores multiple contents of multiple candidate categories. The more content a candidate category has, the greater the proportion of the candidate category in the entire target content library, and the higher the selection order of the candidate category.

[0263] In step 1620, the object preference of the content refers to the degree of preference of the target object group for the content of the corresponding candidate category. The object preference can reflect the degree of positive feedback of the object to the content of the category, such as likes, comments, fans, forwarding, etc. Since each content's positive feedback such as likes, comments, fans, and forwarding can correspond to a score, the higher the score, the higher the degree of positive feedback, such as the higher the number of likes, the more comments, etc. Therefore, by adding these scores together, a score for characterizing the object preference can be obtained. In order to simplify the calculation, this score can be mapped to any value in [0-10]. The higher the object preference of the content, the more the target object group likes the content of the candidate category, and the selection order of the candidate category can be higher to better meet the needs of the target object group.

[0264] In step 1630, based on the number of contents in each candidate category and the object preference of the contents in each candidate category, the candidate category selection order is determined, including: determining a first score for each candidate category based on the number of contents in each candidate category; determining a second score for each candidate category based on the object preference of the contents in each candidate category; determining a total score for each candidate category based on the first score and the second score of each candidate category; and determining a candidate category selection order based on the total score of each candidate category.

[0265] The first score of each candidate category is determined based on the number of contents in each candidate category, and can be obtained by searching a first comparison table. Table 1 shows an example of the first comparison table.

[0266] Number of contents First Score Within 10 20 Greater than or equal to 10 and less than 100 40 Greater than or equal to 100 and less than 500 60 Greater than or equal to 500 and less than 1000 80 Greater than or equal to 1000 100

[0267] Table 1

[0268] Based on the above example, the number of content in the candidate category is 200. By looking up Table 1, the corresponding first score is 60; the number of content in the candidate category is 550. By looking up Table 1, the corresponding first score is 80.

[0269] The second score for each candidate category is determined based on the object preference of the content of each candidate category, which can be found by searching the second comparison table. The object preference of the content can be any value between [0-10]. Table 2 shows an example of the second comparison table.

[0270] Object preference of content Second score Less than 2 10 Greater than or equal to 2 and less than 4 40 Greater than or equal to 4 and less than 6 60 Greater than or equal to 6 and less than 8 80 Greater than or equal to 8 100

[0271] Table 2

[0272] Based on the above example, the object preference of the content of the candidate category is 5. Looking up Table 2, the corresponding second score is 60; the object preference of the content of the candidate category is 7. Looking up Table 2, the corresponding second score is 80.

[0273] Based on the first score and the second score of each candidate category, the total score of each candidate category is determined, which can be obtained by averaging or weighted averaging.

[0274] In one embodiment, the total score is determined by taking the average of the first and second scores. For example, if the first score, determined based on the number of candidate category content, is 60, and the second score, determined based on the object preference of the candidate category content, is 80, then the total score is (60 + 80) / 2 = 70. The advantage of using the average to calculate the total score is that the number of candidate category content and the object preference of the content have the same impact on the calculation of the total score, improving the fairness of determining the candidate category selection order.

[0275] In another embodiment, the total score is calculated using a weighted average of the first score and the second score. In this embodiment, it is first necessary to set weights for the first score corresponding to the number of contents in the candidate category and the second score corresponding to the object preference of the contents in the candidate category. For example, the weight of the first score is 0.6, the weight of the second score is 0.4, the first score is 60, and the second score is 80, then the total score is 60*0.6+80*0.4=68. The advantage of using the weighted average method to calculate the total score is that different weights can be flexibly set for the first score corresponding to the number of contents in the candidate category and the second score corresponding to the object preference of the contents in the candidate category according to the needs of the actual application, thereby improving the flexibility of determining the selection order of the candidate categories.

[0276] Based on the total score of each candidate category, they are ranked from highest to lowest to determine the candidate category selection order. The higher the total score, the higher the candidate category is ranked in the candidate category selection order. For example, the candidate categories stored in the candidate category library include film and television variety shows, games, anime and manga, comedy, and news. If the total score is determined by taking the average of the first and second scores, by calculating the total score of each candidate category, the total score for the film and television variety show category can be determined to be 15, the total score for the games category to be 40, the total score for the anime and manga category to be 25, the total score for the comedy category to be 80, and the total score for the news category to be 50. Therefore, since 80 > 50 > 40 > 25 > 15, the candidate category selection order is: comedy > news > games > anime and manga > film and television variety shows.

[0277] It should be noted that for multiple candidate categories with the same total score, the order can be randomly determined and is not specifically limited here.

[0278] The advantage of the above steps 1610 to 1630 is that the total score is determined based on the number of contents in each candidate category and the object preference of the contents in each candidate category. The scores corresponding to the number of contents in each candidate category and the object preference of the contents in each candidate category can be adjusted according to the needs of actual application, thereby adjusting the candidate category selection order and improving the flexibility of determining the candidate category selection order.

[0279] In step 1520 , after the candidate category selection order is determined, an unselected candidate category is selected from the candidate category library in sequence according to the candidate category selection order as the target category.

[0280] The advantage of the above steps 1510 to 1520 is that by determining the order of candidate category selection and selecting an unselected candidate category from the candidate category library as the target category, candidate categories with higher importance can be analyzed first, ensuring that the required candidate categories are fully evaluated to avoid missing or ignoring certain categories.

[0281] In one embodiment, Figure 14 As shown, after step 350, the content push method of the embodiment of the present disclosure further includes:

[0282] Step 1420: Adjust the content proportions of the multiple target categories according to the sum of the content proportions of the multiple selected target categories.

[0283] It is understandable that the sum of the final optimal content proportions of multiple target categories pushed to the target object group is 100% (or 1).

[0284] In one embodiment, Figure 14 As shown, step 1420 includes:

[0285] Step 14211: If the sum of the content proportions of the selected multiple target categories exceeds 100%, perform a first adjustment on the content proportions of the multiple target categories.

[0286] The disclosed embodiment selects an unselected candidate category from the candidate category library as the target category, that is, the order of selecting categories can be randomly determined. If the total resource amount is 100%, after the optimal content ratio of a target category is selected, the total resource amount is subtracted from the optimal content ratio, and then the optimal content ratio is selected for the next selected target category until the total resource amount is exhausted. In this way, if the sum of the content ratios of multiple selected target categories exceeds 100%, a first adjustment is made to the content ratios of multiple target categories that have exceeded 100%. In this way, the sum of the content ratios of the multiple target categories after adjustment is 100%.

[0287] In one embodiment, Figure 17 As shown, the first adjustment is made to the content proportion of multiple target categories, including:

[0288] Step 1710: Divide the content proportion of each target category by the sum of the content proportions of multiple target categories to obtain a first quotient;

[0289] Step 1720: Use the first quotient as the first adjusted content proportion of the target category.

[0290] In step 1710, the sum of the content proportions is the sum of the content proportions of the selected target categories that exceed 100%. The first quotient is the ratio of the content proportion of each target category to the sum of the content proportions of the multiple target categories. The first quotient is equivalent to the result of normalizing the content proportion of each target category based on the sum of the content proportions of the multiple target categories.

[0291] In step 1720, after the first adjustment, the content proportions of the selected target categories, the sum of the content proportions of multiple target categories is 100%. For example, the categories stored in the candidate category library include film and television variety category, game category, two-dimensional category, funny category, and news category. Select one candidate category that has not been selected from the selection category library as the target category in turn. In this way, the optimal content proportion of the film and television variety category is determined to be 35%, the optimal content proportion of the game category is 25%, the optimal content proportion of the two-dimensional category is 10%, and the optimal content proportion of the funny category is 40%. Since the sum of the content proportions of the film and television variety category, the game category, the two-dimensional category, and the funny category is 35% + 25% + 10% + 40% = 110%, it has exceeded 100%. In this way, the content proportions of film and television variety shows, game categories, two-dimensional categories, and comedy categories are adjusted according to the sum of the current content proportions. That is, the optimal content proportion of film and television variety shows is adjusted to 35% / 110%≈31.8%, the optimal content proportion of game categories is 25% / 110%≈22.7%, the optimal content proportion of two-dimensional categories is 10% / 110%≈9.1%, and the optimal content proportion of comedy categories is 40% / 110%≈36.4%.

[0292] The advantage of the above steps 1710 to 1720 is that the constraint that the total category proportion is 100% can be fully considered, and the first adjustment of the content proportion of multiple target categories can be made. In this way, the impact of the content proportion exceeding 100% on the selected target category with a higher content proportion can be avoided, while improving the push efficiency and effectively reducing the push tuning overhead.

[0293] In another embodiment, step 1420 includes: if the sum of the content proportions of the multiple target categories that have been selected exceeds 100%, and all the candidate categories in the candidate category library have been selected, a third adjustment is made to the content proportions of the multiple target categories. In this way, all categories in the candidate category library can be fully considered, and the push accuracy can be effectively improved. For example, the categories stored in the candidate category library include film and television variety categories, game categories, two-dimensional categories, funny categories, and news categories. Select one candidate category that has not been selected from the selection category library as the target category in turn, so that the optimal content proportion of the film and television variety category is determined to be 35%, the optimal content proportion of the game category is 25%, the optimal content proportion of the two-dimensional category is 10%, the optimal content proportion of the funny category is 40%, the optimal content proportion of the two-dimensional category is 10%, and the optimal content proportion of the news category is 30%. Since the sum of the content proportions of film and television variety shows, game categories, two-dimensional categories, and funny categories is 35% + 25% + 10% + 40% + 30% = 140%, it has exceeded 100%. After all the candidate categories in the candidate category library have been selected, the content proportion of each category is adjusted according to the sum of the content proportions of all categories, that is, the optimal content proportion of the film and television variety show category is adjusted to 35% / 140% = 25%, the optimal content proportion of the game category is 25% / 140% ≈ 17.9%, the optimal content proportion of the two-dimensional category is 10% / 140% ≈ 7.1%, the optimal content proportion of the funny category is 40% / 140% ≈ 28.6%, and the optimal content proportion of the news category is 30% / 140% ≈ 21.4%. Therefore, the embodiment of the present disclosure will adjust the content proportions of all target categories, so as to fully consider the impact of the content proportion of each category on the content proportion of the target category that is finally pushed to the target object group, which can effectively improve the push accuracy.

[0294] In another embodiment, Figure 14 As shown, step 1420 includes:

[0295] Step 14221: If the sum of the content proportions of the selected multiple target categories does not exceed 100%, return to the step of selecting an unselected candidate category from the candidate category library as the target category;

[0296] Step 14222: If all candidate categories in the candidate category library have been selected, and the sum of the content proportions of the selected multiple target categories still does not exceed 100%, a second adjustment is made to the content proportions of the multiple target categories.

[0297] In step 14221, if the total percentage of content in the selected target categories does not exceed 100%, the process returns to the step of selecting an unselected candidate category from the candidate category library as the target category, and reselects the next unselected candidate category as the target category. After determining the percentage of content in the next target category, step 14221 is executed again.

[0298] In step 14222, if all candidate categories in the candidate category library have been selected, and the sum of the content proportions of the selected multiple target categories still does not exceed 100%, a second adjustment is made to the content proportions of the multiple target categories. Figure 18 As shown, a second adjustment is made to the content proportions of multiple target categories, including:

[0299] Step 1810: Divide the content proportion of each target category by the sum of the content proportions of multiple target categories to obtain a second quotient;

[0300] Step 1820: Use the second quotient as the second adjusted content proportion of the target category.

[0301] In step 1810, the sum of the content percentages is the sum of the content percentages of all selected target categories. The second quotient is the ratio of the content percentage of each target category to the sum of the content percentages of all target categories. This second quotient is equivalent to normalizing the content percentage of each target category based on the sum of the content percentages of all target categories.

[0302] In step 1820, after the second adjustment, the content proportions of the selected target categories, the sum of the content proportions of multiple target categories is 100%. For example, the categories stored in the candidate category library include film and television variety category, game category, two-dimensional category, funny category, and news category. Select one candidate category that has not been selected from the selection category library as the target category in turn. In this way, the optimal content proportion of the film and television variety category is determined to be 15%, the optimal content proportion of the game category is 10%, the optimal content proportion of the two-dimensional category is 5%, the optimal content proportion of the funny category is 20%, and the optimal content proportion of the news category is 2%. Since all the candidate categories in the candidate category library have been selected, and the sum of the content proportions of the film and television variety category, game category, two-dimensional category, funny category, and news category is 15% + 10% + 5% + 20% + 2% = 52%, it does not reach more than 100%. In this way, according to the sum of the current proportions of all content, the content proportions of film and television variety shows, games, two-dimensional, comedy, and news categories are adjusted for the second time. That is, the optimal content proportion of film and television variety shows is adjusted to 15% / 52%≈28.8%, the optimal content proportion of games is 10% / 52%≈19.2%, the optimal content proportion of two-dimensional categories is 5% / 52%≈9.6%, the optimal content proportion of comedy categories is 20% / 52%≈38.5%, and the optimal content proportion of news categories is 2% / 52%≈3.9%.

[0303] The advantage of the above steps 1810 to 1820 is that each candidate category in the candidate category library can be fully considered, and a second adjustment can be made to the content proportion of multiple target categories. In this way, omission or neglect of certain categories can be avoided, while improving the push accuracy and effectively reducing the push tuning overhead.

[0304] The advantage of the above steps 14221 and 14222 is that they fully consider the situations where the sum of the content proportions of the selected multiple target categories does not exceed 100% and exceeds 100%, and optimize content push through curve analysis, thereby improving push accuracy and reducing push tuning overhead.

[0305] Detailed description of content push compatible with heterogeneous objects in the disclosed embodiment

[0306] Heterogeneity refers to the differences or diversity among different subjects in certain specific aspects or characteristics. Since subjects within the same target group share similar characteristics, such as education, years of experience, job, hobbies, and historical viewing history, determining the target group to which a target subject belongs can effectively mitigate the impact of heterogeneity on content push.

[0307] In one embodiment, Figure 19 As shown, step 310 includes:

[0308] Step 1910: Obtain a platform object set and multiple object group division parameters;

[0309] Step 1920: Determine the application order of the target segmentation parameters based on the multiple object group segmentation parameters;

[0310] Step 1930: Apply the target partitioning parameter application sequence to the platform object set to obtain a target result tree corresponding to the target partitioning parameter application sequence, where each leaf node of the target result tree corresponds to an object group partitioned according to the target partitioning parameter application sequence.

[0311] Step 1940: The divided object group to which the target object belongs is used as the target object group.

[0312] In step 1910, the platform object set refers to the set of all objects included in the push platform. For example, the push platform may be an instant messaging application. The object grouping parameters refer to the parameters used to classify different objects in the push platform. Based on the object grouping parameters, multiple objects in the push platform can be divided into different object groups. The multiple object grouping parameters can be constructed based on education level characteristics, years of work experience characteristics, job characteristics, hobby characteristics, historical viewing content characteristics, and so on.

[0313] In step 1920, different object groups are generated due to different object group segmentation parameter orders. Therefore, a target segmentation parameter application order can be determined based on multiple object group segmentation parameters. The target segmentation parameter application order refers to the reference order of the object group segmentation parameters selected when segmenting the object groups.

[0314] In one embodiment, based on multiple object group segmentation parameters, a target segmentation parameter application order can be randomly determined based on experience. For example, the target segmentation parameter application order can be education level characteristics - years of work characteristics - work characteristics - hobby characteristics - historical viewing content characteristics, etc., and this is not specifically limited here.

[0315] In another embodiment, Figure 20 As shown, step 1920 includes:

[0316] Step 2010: Use the platform object set as the object group to be divided;

[0317] Step 220: Execute a first process, which includes: applying unselected object group segmentation parameters from the plurality of object group segmentation parameters as the to-be-examined object group segmentation parameters to the to-be-examined object groups, obtaining segmented object groups corresponding to each to-be-examined object group segmentation parameter, and determining contribution function values corresponding to each to-be-examined object group segmentation parameter based on each segmented object group;

[0318] Step 2030: Use the partitioning parameter of the to-be-investigated object group with the largest contribution function value as the partitioning parameter of the target object group, update the to-be-divided object group with the partitioned object group, and return to the first process;

[0319] Step 2040: Concatenate the obtained multiple target object group division parameters in the order of acquisition as the target division parameter application order.

[0320] In step 2010, when determining the application order of the target partitioning parameters, the platform object set is first used as the group of objects to be partitioned. The group of objects to be partitioned is used to represent a group variable in the first process of the subsequent loop.

[0321] In step 220, the first process refers to the process of determining the target object group division parameters. The object group division parameters to be examined are used to represent a parameter variable in the first process of the loop. The object group division parameters that have not been selected from the multiple object group division parameters are respectively used as the object group division parameters to be examined. The object groups to be divided are divided into groups based on the object group division parameters to be examined each time, and the divided object groups corresponding to the object group division parameters to be examined are obtained. For example, the multiple object group division parameters include parameters constructed based on education level characteristics and working years characteristics. If the parameters corresponding to the education level characteristics (such as undergraduate degree or below and undergraduate degree or above) are selected from the multiple object group division parameters, the parameters are applied to the object groups to be divided, and the divided object groups obtained include group A and group B, group A represents the group composed of objects with undergraduate degree or below, and group B represents the group composed of objects with undergraduate degree or above. If a parameter corresponding to a working experience characteristic (such as less than three years of work, more than three years of work but less than ten years of work, and more than ten years of work) is selected from multiple object group division parameters, and the parameter is applied to the object group to be divided, the object groups after division include group C, group D, and group E. Group C represents a group composed of objects who have worked for less than three years, group D represents a group composed of objects who have worked for more than three years but less than ten years, and group E represents a group composed of objects who have worked for more than ten years.

[0322] The contribution function value is the push loss value corresponding to the parameters for the group of subjects to be examined. The contribution function value is a numerical value. The larger the contribution function value, the greater the impact of the parameters for the group of subjects to be examined on content push, and thus the more priority should be given.

[0323] In one embodiment, Figure 21 As shown, in step 2020, based on each divided object group, the contribution function value corresponding to each division parameter of the to-be-investigated object group is determined, including:

[0324] Step 2110: for each parameter for dividing the subject group to be examined, determine the sub-contribution function values of the multiple divided subject groups corresponding to the parameter for dividing the subject group to be examined;

[0325] Step 2120: Determine the contribution function value corresponding to the partition parameter of the object group to be examined based on the sub-contribution function values of the multiple divided object groups corresponding to the partition parameter of the object group to be examined.

[0326] In step 2110 , the sub-contribution function value refers to the total loss value when the partitioning parameter of the object group to be investigated is partitioned on the corresponding multiple partitioned object groups.

[0327] In one embodiment, Figure 22 As shown, in step 2110, determining the sub-contribution function values of the plurality of divided object groups corresponding to the division parameters of the to-be-investigated object groups includes:

[0328] Step 2210: Use each divided object group as the target object group in turn, and return to the step of obtaining the first curve of the target object group for the target category to obtain the absolute value of the first difference at the target point as the single object contribution value of the divided object group;

[0329] Step 2220: Determine the sub-contribution function values of the divided object groups based on the single-object contribution values.

[0330] In step 2210, each of the divided subject groups is used as the target subject group in turn, and steps 310 to 340 are performed. In this case, the single-subject contribution value is the absolute value of the first difference between the average curve values on both sides of the target point on the first integral curve when the absolute value is the largest. The calculation process of the single-subject contribution value is shown in Formula 7 above.

[0331] In step 2220 , after determining the single-object contribution value of each divided object group corresponding to each to-be-investigated object group division parameter, the sub-contribution function value of the divided object group is determined according to the multiple single-object contribution values.

[0332] In one embodiment, Figure 23 As shown, step 2220 includes:

[0333] Step 2310: Obtain the number of objects in the divided object group;

[0334] Step 2320: Determine the sub-contribution function value of the divided object group based on the number of objects in the divided object group and the single-object contribution value.

[0335] In step 2310 , the number of objects in the divided object group refers to the number of objects included in the divided object group.

[0336] In step 2320, the sub-contribution function value of the divided object group is determined based on the number of objects in the divided object group and the single-object contribution value. For example, if the object group to be examined is divided into multiple divided object groups after the object group is divided by the object group division parameter, the sub-contribution function value of each divided object group is calculated as shown in the following formula 8:

[0337] H(S la , t i )=N la *D 2 (t i ) (Formula 8)

[0338] In formula 8, H(S la , t i ) represents the sub-contribution function value of the lath object group divided by the object group division parameter to be investigated. For example, if the object group division parameter to be investigated corresponds to two object groups after division, the subscripts are l1 and l2 respectively, and l represents the leaf node number in the target result tree corresponding to the application order of the target division parameter; S la represents the sample set of the object group after the lath label division, N la Denotes the number of objects in the object group after the lath label division; D 2 (t i ) represents the square value of the single-object contribution value based on the divided object group; H(S la , t i ) represents the sub-contribution function value of the object group after division; t i Indicates the first proportion corresponding to the absolute value of the first difference at the target point corresponding to the single object contribution value.

[0339] In step 2120, the contribution function value corresponding to the partition parameter of the subject group to be examined is determined based on the sub-contribution function values of the multiple partitioned subject groups corresponding to the partition parameter of the subject group to be examined. This includes: determining a candidate contribution function value corresponding to the partition parameter of the subject group to be examined in a single target category based on the sub-contribution function values of the multiple partitioned subject groups corresponding to the partition parameter of the subject group to be examined; and determining a contribution function value corresponding to the partition parameter of the subject group to be examined based on the candidate contribution function values corresponding to the partition parameter of the subject group to be examined in multiple target categories. Therefore, if the partition parameter of the subject group to be examined is divided into two partitioned subject groups after the partition parameter of the subject group to be examined is divided, the process of calculating the contribution function value corresponding to the partition parameter of the subject group to be examined is shown in the following formula 9:

[0340]

[0341] In Formula 9, H l It represents the contribution function value corresponding to the parameter of the group division of the object to be investigated, H(S l1 , t i ) represents the sub-contribution function value of the object group after the l1th label division, k′ represents the number of categories of the target category of the object group after the l1th label division, S l1 represents the sample set of the object group after the l1th label division, H(S l2 , t i ) represents the sub-contribution function value of the l2th numbered object group after the object group division parameter is divided, k″ represents the number of categories of the target category of the l2th numbered object group after the division, S l2 Represents the sample set of the object group after division with the l2th label.

[0342] In step 2030, the parameter for the group of objects to be examined with the largest contribution function value is used as the target group of objects to be examined parameter, and the group of objects to be divided is updated with the divided group of objects, and the process returns to the first step. At this point, the unselected group of objects to be examined parameters are used as the target group of objects to be examined parameters and applied to the updated group of objects to be divided. In this way, multiple target group of objects to be examined parameters are obtained, and the number of target group of objects to be examined parameters is the same as the number of pre-set group of objects to be divided parameters. Therefore, the process of calculating the target group of objects to be divided parameters is shown in Formula 10 below:

[0343]

[0344] In formula 10, HD represents the contribution function value of the parameter for dividing the group of objects to be investigated with the largest contribution function value. It represents the process of calculating the maximum value of the contribution function value of the group division parameters of multiple objects to be examined.

[0345] In step 2040, the multiple target object grouping parameters obtained are concatenated in the order in which they were obtained, which serves as the target grouping parameter application order. The target grouping parameter application order refers to the order in which the target object grouping parameters are used to determine the leaf nodes at each level of the target result tree. Because target object grouping parameters can be classified into various types, such as education level, hobbies, and occupations, the target result tree obtained will differ depending on which parameter is used for segmentation. Therefore, the target grouping parameter application order is the basis for determining the fixed form of the target result tree.

[0346] like Figure 24 As shown, multiple object group division parameters include parameters constructed based on work characteristics (such as those who have worked for less than ten years and those who have worked for more than ten years), hobby characteristics (such as those who like music and those who don't like music), geographical region characteristics (such as the south and the north), education level characteristics (college and above and other), and historical viewing content characteristics (such as the average vv per person in the past seven days is greater than a preset vv number and less than a preset vv number). The platform object set is used as the object group to be divided, and the object group division parameters that have not been selected from the multiple object group division parameters are respectively used as the object group division parameters to be examined. They are applied to the object groups to be divided, and the object groups corresponding to each object group division parameter to be examined are obtained. Based on each object group, the contribution function value corresponding to each object group division parameter to be examined is determined. In the first process, the contribution function value of the object group divided corresponding to the work characteristic is 10, the contribution function value of the object group divided corresponding to the hobby characteristic is 15, the contribution function value of the object group divided corresponding to the geographical region characteristic is 8, the contribution function value of the object group divided corresponding to the education level characteristic is 12, and the contribution function value of the object group divided corresponding to the historical viewing content characteristic is 6. By comparing the values of multiple contribution functions, it is determined that the contribution function value of the object group segmentation parameter corresponding to the hobby feature is the largest. The hobby feature is then used as the first target object group segmentation parameter. The segmented object group is updated with the object group to be segmented, and the process returns to the first step. Because each first step only uses the unselected object group segmentation parameters from the multiple object group segmentation parameters as the object group segmentation parameters to be examined, multiple target object group segmentation parameters can be determined. Therefore, after multiple first steps, the target object group segmentation parameters are sequentially determined in the following order: hobby features, education level features, work features, geographic area features, and historical viewing content features. At this point, the target segmentation parameters are applied in the following order: "hobby features - education level features - work features - geographic area features - historical viewing content features."

[0347] The advantage of steps 2010-2040 is that the target object grouping parameters are determined based on the contribution function values corresponding to the various target object grouping parameters to be considered. This allows the target grouping parameters to be applied in a cascaded order based on the obtained multiple target object grouping parameters, in the order in which they were obtained. Therefore, the disclosed embodiment can consider the degree to which different target object grouping parameters affect grouping, thereby determining a more accurate target grouping parameter application order and improving push accuracy.

[0348] In step 1930, the target partitioning parameter application order is applied to the platform object set to obtain a target result tree corresponding to the target partitioning parameter application order, where each leaf node of the target result tree corresponds to an object group divided according to the target partitioning parameter application order.

[0349] like Figure 25 As shown, if the target partitioning parameter application order is determined to be "educational level characteristics-work characteristics-geographical area characteristics", the target partitioning parameter application order can be applied to the platform object set. At this time, the root node of the first layer of the target result tree is the platform object set, and the leaf nodes of the second layer of the target result tree are the two object groups divided based on the target object group partitioning parameters corresponding to the education level characteristics, namely object group G11 and object group G12. The leaf nodes of the third layer of the target result tree are the target object group partitioning parameters corresponding to the work characteristics, and the object groups after each object group of the second layer is divided are object group G21 and object group G22 corresponding to object group G11, and object group G23 and object group G24 corresponding to object group G12. The leaf nodes of the fourth layer of the target result tree are the target object group division parameters corresponding to the geographical area characteristics. The object groups after each object group of the third layer is divided are object group G31 and object group G32 corresponding to object group G21, object group G33 and object group G34 corresponding to object group G22, object group G35 and object group G36 corresponding to object group G23, and object group G37 and object group G38 corresponding to object group G24.

[0350] In step 1940, the group of objects to which the target object belongs is defined as the target object group. After determining the target object tree, the target object's features are matched against the target object's features according to the target partitioning parameter application order corresponding to the target result tree to determine the group of objects that successfully matches the target object's features. Thus, the group of objects to which the target object belongs is defined as the target object group.

[0351] The advantage of the above steps 1910-1940 is that the present application divides the platform object set by traversing the target object group partitioning parameters in the target partitioning parameter application order to iteratively construct a complete target result tree. At this time, the tree structure of the target result tree is a partitioning structure of an object group, and each layer of leaf nodes corresponds to a partitioning of a target object group partitioning parameter, and each leaf node corresponds to an object group. Therefore, by determining the target object group to which the target object belongs based on the target result tree, the impact of the heterogeneity of different objects on the content push of the target object can be effectively alleviated. At the same time, since the embodiment of the present disclosure adopts an integral method, the result obtained is accumulated over time. For the object stickiness value caused by the object being offline, the surge is small, because the surge lasts for a small range on the horizontal axis, and has little effect on the integral value, thereby reducing the impact of the object being offline on the content push optimization. The above process optimizes content push by means of curve analysis, which improves push accuracy while reducing push tuning overhead.

[0352] Detailed description of mitigating randomness of content push in the disclosed embodiment

[0353] In practice, the results of a single target result tree are closely related to the order in which categories are extracted and the optimal solutions for each category are calculated. However, the order in which target partitioning parameters are applied is random, which results in instability in target result trees constructed multiple times. Therefore, the disclosed embodiment alleviates this randomness by constructing multiple target result trees. Based on these multiple target result trees, the proportion of content in the target category ultimately pushed to the target object is comprehensively considered, effectively improving push accuracy.

[0354] In one embodiment, Figure 26 As shown, step 1920 includes:

[0355] Step 19201: Based on multiple object group segmentation parameters, determine the application order of multiple target segmentation parameters.

[0356] In step 19001, after determining multiple object group segmentation parameters, the multiple object group segmentation parameters can be randomly arranged to obtain multiple target segmentation parameter application orders. For example, if the multiple object group segmentation parameters include parameters constructed based on work characteristics, hobby characteristics, and geographic area characteristics, the multiple target segmentation parameter application orders in this case include: work characteristics-hobby characteristics-geographic area characteristics, work characteristics-geographic area characteristics-hobby characteristics, hobby characteristics-geographic area characteristics-work characteristics, hobby characteristics-work characteristics-geographic area characteristics, geographic area characteristics-work characteristics-hobby characteristics, and geographic area characteristics-hobby characteristics-work characteristics. In this way, six target segmentation parameter application orders can be obtained.

[0357] In another embodiment, the target category is selected from a candidate category library, and the number of target segmentation parameter application orders is determined by:

[0358] Get the number of candidate categories in the candidate category library;

[0359] The number of application orders of the target partitioning parameters is determined to be twice the number of candidate categories.

[0360] It is understandable that the number of target partitioning parameter application orders determined at this time is twice the number of candidate categories in the candidate category library. For example, the candidate category library contains five candidate categories, namely, funny category, music category, news category, food category, and game category. In this way, the number of target partitioning parameter application orders is 2×5=10. This ensures that each candidate category has the opportunity to appear at the front of the random order. At the same time, the average results obtained by modeling the target result tree corresponding to each target partitioning parameter application order will gradually converge and eventually stabilize. Moreover, the embodiment of the present disclosure determines the number of target partitioning parameter application orders to be twice the number of candidate categories, which is also a result proven by a large number of experiments and can effectively improve the push accuracy.

[0361] Based on this, Figure 26 As shown, step 1930 includes:

[0362] Step 19301: Apply multiple target partitioning parameter application sequences to the platform object set respectively to obtain the target result tree corresponding to each target partitioning parameter application sequence.

[0363] In step 19301, the process of applying the various target partitioning parameter application sequences to the platform object set to obtain the target result tree corresponding to each target partitioning parameter application sequence is similar to the above step 1930, and the target result tree constructed can refer to the above Figure 25 , I will not go into details here.

[0364] Based on this, Figure 26 As shown, step 1940 includes:

[0365] Step 19401: The object group to which the target object belongs in the leaf node of the target result tree corresponding to each target partitioning parameter application sequence is used as the target object group corresponding to the target partitioning parameter application sequence.

[0366] In step 19401, since multiple target partitioning parameter application sequences and the target result trees corresponding to each target partitioning parameter application sequence have been determined, the target object group to which the target object belongs in the leaf node of the target result tree corresponding to each target partitioning parameter application sequence can be used as the target object group corresponding to that target partitioning parameter application sequence. In this way, multiple target object groups to which the target object belongs can be determined.

[0367] In this way, if Figure 26 As shown, after step 350, the content push method of the embodiment of the present disclosure further includes:

[0368] Step 2610: Determine the proportion of content of the target category to be pushed to the target object based on the proportion of content of the target category to be pushed to the target object group corresponding to the determined application order of each target division parameter.

[0369] At this time, the proportion of content of the target category pushed to the target object can be the average or weighted average of the proportion of content of the target category pushed to the target object group corresponding to each target division parameter application order. For example, when the proportion of content of the target category pushed to the target object is the average of the proportion of content of the target category pushed to the target object group corresponding to each target division parameter application order, if there are two target division parameter application orders, the proportion of content of the funny category pushed to the target object group corresponding to the first target division parameter application order is 40%, and the proportion of content of the funny category pushed to the target object group corresponding to the second target division parameter application order is 50%. In this way, the proportion of content of the funny category pushed to the target object is (40%+50%) / 2=45%. Therefore, the advantage of using the average method to determine the proportion of content of the target category pushed to the target object is that the influence of each target division parameter application order on the proportion of content of the target category pushed to the target object is the same, thereby improving the fairness of determining the proportion of content of the target category pushed to the target object.

[0370] In another embodiment, when calculating the percentage of target category content pushed to a target object, and taking the weighted average of the percentage of target category content pushed to the target object group corresponding to each target partitioning parameter application order, if there are two target partitioning parameter application orders, a corresponding weight is set for the target object group corresponding to each target partitioning parameter application order. For example, the weight of the target object group corresponding to the first target partitioning parameter application order is 0.7, and the percentage of funny category content pushed to the target object group corresponding to the first target partitioning parameter application order is 40%. The weight of the target object group corresponding to the second target partitioning parameter application order is 0.3, and the percentage of funny category content pushed to the target object group corresponding to the second target partitioning parameter application order is 50%. In this way, the percentage of funny category content pushed to the target object is 0.7 × 40% + 0.3 × 50% = 43%. Therefore, the advantage of using a weighted average to determine the percentage of target category content pushed to the target object is that different weights can be flexibly set for the target object groups corresponding to each target partitioning parameter application order according to actual application needs, thereby increasing the flexibility of determining the percentage of target category content pushed to the target object.

[0371] It should be noted that, in actual application, the content push method provided by the embodiment of the present disclosure was tested on the short video data of XX / XX / XXXX, and the results of the optimal proportion of each target category were obtained. It can be found that the current distribution proportion of the sports category is 10%, but the optimal proportion should be 25%. At this time, the distribution of the sports category should be increased for 85% of the objects in the market. The current distribution proportion of the funny category is 30%, but the optimal proportion should be 15%. At this time, the distribution of the funny category should be reduced for 85% of the users in the market. Based on this, the results of the optimal proportion of each target category were further verified by experiments. The experimental results show that: raising the weight of the sports category (i.e., increasing the distribution proportion of the content) by 1.5 times can increase the average vv of the market by 1.18%, the average time per person by 0.38%, and the average vv of new users by 5.36%, and the average time per person by 5.12%. Demoting the funny category by 0.5 times (i.e. reducing the distribution ratio of the content) can increase the average VV of the market by 0.70%, the average duration of use by 0.34%, and the average VV of new users by 4.76%, the average duration of use by 4.34%, and the login duration by 3.20%.

[0372] Therefore, the disclosed embodiments utilize an integral approach, yielding results accumulated over time. A surge in object stickiness caused by an object going offline has a minimal impact on the integral value due to the short duration of the surge on the horizontal axis, thus minimizing the impact of the object going offline on content push optimization. The above process optimizes content push through curve analysis, improving push accuracy while reducing push tuning overhead.

[0373] Description of the apparatus and device of the present disclosure

[0374] It is to be understood that, although the steps in the above-mentioned flowcharts are shown in sequence according to the arrow representations, these steps are not necessarily performed in sequence according to the order represented by the arrows. Unless otherwise specified in the present embodiment, there is no strict order restriction on the execution of these steps, and these steps can be performed in other orders. Moreover, at least a portion of the steps in the above-mentioned flowcharts may include multiple steps or multiple stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with other steps or at least a portion of the steps or stages in other steps.

[0375] It should be noted that in each specific embodiment of the present application, when it comes to the need to perform relevant processing based on data related to the target object, such as the target object group or the stickiness of the session object, the permission or consent of the target content will be obtained first, and the collection, use and processing of such data will comply with relevant laws, regulations and standards. In addition, when the embodiment of the present application needs to obtain relevant information about the target object group, it will obtain the separate permission or consent of the target object group through a pop-up window or jump to a confirmation page. After clearly obtaining the separate permission or consent of the target object group, the necessary relevant data for the normal operation of the embodiment of the present application will be obtained.

[0376] Figure 27 This is a schematic diagram of the structure of a content push device 2700 provided in an embodiment of the present disclosure. The content push device 2700 includes:

[0377] A first determining unit 2710 is configured to determine a target object group to which a target object belongs;

[0378] An acquisition unit 2720 is configured to acquire a first curve of the target object group for the target category, wherein the first curve indicates a change in average single-session object stickiness as a function of a first proportion, where the first proportion is a ratio of the number of contents of the target category in a single session to the total number of contents in the single session;

[0379] A second determining unit 2730 is configured to determine a first integral curve of the first curve;

[0380] The third determining unit 2740 is configured to determine a target point on the first integral curve so that the absolute value of a first difference between average curve values on both sides of the target point on the first integral curve is maximized;

[0381] The push unit 2750 is configured to determine a proportion of content of a target category to be pushed to a target object group based on the target point.

[0382] Optionally, the acquiring unit 2720 is specifically configured to:

[0383] Get the conversations that occurred in the target group;

[0384] For each acquired session, determine the first percentage of the session and the object stickiness of a single session;

[0385] determining an average single-session object stickiness corresponding to the first proportion based on the single-session object stickiness corresponding to each first proportion;

[0386] A first curve is determined based on each first proportion and the average single-session object stickiness corresponding to the first proportion.

[0387] Optionally, the single-session object stickiness includes the single-session duration and the number of contents viewed in a single session;

[0388] The acquiring unit 2720 is further specifically configured to:

[0389] Determine an average single session duration corresponding to the first proportion based on the single session duration corresponding to each first proportion;

[0390] Determine, based on the number of content views in a single session corresponding to each first proportion, an average number of content views in a single session corresponding to the first proportion;

[0391] Based on the average single session duration and the average number of contents viewed in a single session, the average single session object stickiness corresponding to the first proportion is determined.

[0392] Optionally, the third determining unit 2740 is specifically configured to:

[0393] Assuming that the right direction of the horizontal axis of the coordinate system where the first integral curve is located is positive, set the objective function, which is equal to the first difference between the first average curve value on the right side and the second average curve value on the left side of any point on the first integral curve;

[0394] On the first integral curve, a point that maximizes the objective function is determined as the target point.

[0395] Optionally, the first average curve value is determined by:

[0396] Get the largest first percentage of the target audience group;

[0397] Determine a first integral value between any point on the first average curve and the point corresponding to the maximum first proportion;

[0398] determining a second difference between the maximum first proportion and the abscissa of any point;

[0399] Based on the first integrated value and the second difference, a first average curve value is determined.

[0400] Optionally, the second average curve value is determined by:

[0401] Get the minimum first percentage of the target group;

[0402] Determine a second integral value between a point corresponding to the minimum first proportion and any point on the first average curve;

[0403] Determine the third difference between the abscissa of any point and the minimum first proportion;

[0404] Based on the second integrated value and the third difference, a second average curve value is determined.

[0405] Optionally, the target category is selected from a candidate category library;

[0406] Before obtaining the first curve of the target object group for the target category, the content push device further includes:

[0407] A selection unit (not shown), configured to select an unselected candidate category from the candidate category library as a target category;

[0408] After determining the proportion of content of the target category to be pushed to the target object group based on the target point, the content pushing device further includes:

[0409] a first adjusting unit (not shown), configured to make a first adjustment to the content proportions of the selected multiple target categories if the sum of the content proportions of the selected multiple target categories exceeds 100%;

[0410] a returning unit (not shown), configured to return to the step of selecting an unselected candidate category from the candidate category library as the target category if the sum of the content proportions of the selected multiple target categories does not exceed 100%;

[0411] The second adjustment unit (not shown) is used to make a second adjustment to the content proportions of the multiple target categories if all candidate categories in the candidate category library have been selected and the sum of the content proportions of the selected multiple target categories still does not exceed 100%.

[0412] Optionally, the first adjustment unit (not shown) is specifically configured to:

[0413] Divide the content share of each target category by the sum of the content shares of multiple target categories to obtain the first quotient;

[0414] Use the first business as the first adjusted content proportion of the target category.

[0415] Optionally, the second adjustment unit (not shown) is specifically configured to:

[0416] Divide the content share of each target category by the sum of the content shares of multiple target categories to obtain the second quotient;

[0417] Use the second business as the second adjusted content proportion of the target category.

[0418] Optionally, the selection unit (not shown) is specifically configured to:

[0419] Obtain the selection order of candidate categories in the candidate category library;

[0420] According to the candidate category selection order, select an unselected candidate category from the candidate category library as the target category.

[0421] Optionally, the selection unit (not shown) is further configured to:

[0422] Obtain the number of content in each candidate category in the target content library;

[0423] Obtaining the object preference of each candidate category of content in the target content library;

[0424] The candidate category selection order is determined based on the number of contents in each candidate category and the object preference of the contents in each candidate category.

[0425] Optionally, the first determining unit 2710 is specifically configured to:

[0426] Obtain platform object collection and multiple object group division parameters;

[0427] Determining the order of applying the target segmentation parameters based on multiple object group segmentation parameters;

[0428] Apply the target partitioning parameter application sequence to the platform object set to obtain a target result tree corresponding to the target partitioning parameter application sequence, where each leaf node of the target result tree corresponds to an object group partitioned according to the target partitioning parameter application sequence;

[0429] The divided object group to which the target object belongs is used as the target object group.

[0430] Optionally, the first determining unit 2710 is further specifically configured to:

[0431] Taking the platform object set as the object group to be divided;

[0432] Executing a first process, the first process comprising: applying unselected subject group segmentation parameters from the plurality of subject group segmentation parameters as subject group segmentation parameters to be examined, respectively, to the subject groups to be divided, obtaining segmented subject groups corresponding to the respective subject group segmentation parameters to be examined, and determining contribution function values corresponding to the respective subject group segmentation parameters to be examined based on the respective segmented subject groups;

[0433] The partitioning parameter of the to-be-investigated object group with the largest contribution function value is used as the partitioning parameter of the target object group, and the to-be-divided object group is updated with the partitioned object group, and the process returns to the first step;

[0434] The obtained multiple target object group division parameters are cascaded in the order of acquisition as the target division parameter application order.

[0435] Optionally, the first determining unit 2710 is further specifically configured to:

[0436] For each parameter for dividing the subject group to be examined, determining the sub-contribution function values of the plurality of divided subject groups corresponding to the parameter for dividing the subject group to be examined;

[0437] The contribution function value corresponding to the partition parameter of the to-be-investigated object group is determined based on the sub-contribution function values of the plurality of divided object groups corresponding to the partition parameter of the to-be-investigated object group.

[0438] Optionally, the first determining unit 2710 is further specifically configured to:

[0439] Each divided object group is used as the target object group in turn, and the step of obtaining the first curve of the target object group for the target category is returned to obtain the absolute value of the first difference at the target point as the single object contribution value of the divided object group;

[0440] Based on the single object contribution value, the sub-contribution function value of the divided object group is determined.

[0441] Optionally, the first determining unit 2710 is further specifically configured to:

[0442] Get the number of objects in the divided object group;

[0443] The sub-contribution function value of the divided object group is determined based on the number of objects in the divided object group and the single-object contribution value.

[0444] Optionally, determining the application order of target partitioning parameters based on the plurality of object group partitioning parameters includes: determining the application order of a plurality of target partitioning parameters based on the plurality of object group partitioning parameters;

[0445] Applying the target partition parameter application sequence to the platform object set to obtain a target result tree corresponding to the target partition parameter application sequence, including: applying multiple target partition parameter application sequences to the platform object set respectively to obtain a target result tree corresponding to each target partition parameter application sequence;

[0446] Taking the divided object group to which the target object belongs as the target object group includes: taking the object group to which the target object belongs in the leaf node of the target result tree corresponding to each target partition parameter application sequence as the target object group corresponding to the target partition parameter application sequence;

[0447] After determining the proportion of content of the target category to be pushed to the target object group based on the target point, the content pushing device further includes:

[0448] The fourth determining unit (not shown) is configured to determine the proportion of the target category content to be pushed to the target object based on the proportion of the target category content pushed to the target object group corresponding to the determined application order of each target division parameter.

[0449] Optionally, the target category is selected from a candidate category library;

[0450] The number of application orders of the target partitioning parameters is determined as follows:

[0451] Get the number of candidate categories in the candidate category library;

[0452] The number of application orders of the target partitioning parameters is determined to be twice the number of candidate categories.

[0453] Reference Figure 28 , Figure 28 This is a block diagram of the structure of the target terminal 110 for implementing the content push method according to an embodiment of the present disclosure. The terminal includes: a radio frequency (RF) circuit 2810, a memory 2815, an input unit 2830, a display unit 2840, a sensor 2850, an audio circuit 2860, a wireless fidelity (WiFi) module 2870, a processor 2880, and a power supply 2890. Those skilled in the art will understand that Figure 28 The structure of the target terminal 110 shown does not constitute a limitation on a mobile phone or a computer, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0454] The RF circuit 2810 may be used for receiving and sending signals during information transmission or calls. In particular, after receiving downlink information from the base station, it is sent to the processor 2880 for processing. In addition, the designed uplink data is sent to the base station.

[0455] The memory 2815 may be used to store software programs and modules. The processor 2880 executes the software programs and modules stored in the memory 2815 to execute various functional applications and data processing of the content terminal.

[0456] The input unit 2830 may be configured to receive input digital or character information and generate key signal input related to the settings and function control of the content terminal. Specifically, the input unit 2830 may include a touch panel 2831 and other input devices 2832.

[0457] The display unit 2840 may be configured to display input information or provided information and various menus of the content terminal. The display unit 2840 may include a display panel 2841.

[0458] The audio circuit 2860 , the speaker 2861 , and the microphone 2862 may provide an audio interface.

[0459] In this embodiment, the processor 2880 included in the target terminal 110 can execute the content pushing method of the previous embodiment.

[0460] The target terminal 110 of the embodiment of the present disclosure includes but is not limited to mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle terminals, aircraft, etc. The embodiment of the present invention can be applied to various scenarios, including but not limited to content recommendation, data screening, etc.

[0461] Figure 29 This is a structural block diagram of a portion of the push processing server 140 for implementing the content push method of an embodiment of the present disclosure. The push processing server 140 may have relatively large differences due to different configurations or performance, and may include one or more central processing units (CPUs) 2922 (for example, one or more processors) and memories 2932, and one or more storage media 2930 (for example, one or more mass storage devices) for storing application programs 2942 or data 2944. Among them, the memories 2932 and the storage media 2930 can be temporary storage or permanent storage. The program stored in the storage medium 2930 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Furthermore, the central processing unit 2922 can be configured to communicate with the storage medium 2930 to execute a series of instruction operations in the storage medium 2930 on the server.

[0462] The push processing server 140 may also include one or more power supplies 2926, one or more wired or wireless network interfaces 2950, one or more input and output interfaces 2958, and / or one or more operating systems 2941, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0463] The central processor 2922 in the push processing server 140 may be configured to execute the content push method of the embodiment of the present disclosure.

[0464] The embodiments of the present disclosure further provide a computer-readable storage medium, which is used to store program codes, and the program codes are used to execute the content push methods of the aforementioned embodiments.

[0465] The present disclosure also provides a computer program product, which includes a computer program. A processor of a computer device reads and executes the computer program, so that the computer device implements the above-mentioned content push method.

[0466] The terms "first," "second," "third," "fourth," and the like (if any) in the specification of the present disclosure and the accompanying drawings are used to distinguish between similar contents and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present disclosure described herein, for example, can be implemented in orders other than those illustrated or described herein. In addition, the terms "comprises" and "comprising," and any variations thereof, are intended to cover non-exclusive inclusions, e.g., a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0467] It should be understood that in the present disclosure, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated content, indicating that there can be three relationships. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and following associated content is in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0468] It should be understood that in the description of the embodiments of the present disclosure, the meaning of multiple (or multiple items) is more than two, greater than, less than, exceed, etc. are understood to exclude the number itself, and above, below, within, etc. are understood to include the number itself.

[0469] In the several embodiments provided in the present disclosure, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0470] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0471] In addition, in the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program that has a predetermined function and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories) or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0472] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc. Various media that can store program codes.

[0473] It should also be understood that the various implementations provided in the embodiments of the present disclosure can be arbitrarily combined to achieve different technical effects.

[0474] The above is a specific description of the implementation methods of the present disclosure, but the present disclosure is not limited to the above implementation methods. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present disclosure. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present disclosure.

Claims

1. A content push method, characterized in that: include: Determine the target audience group to which the target audience belongs; Obtaining a first curve of the target object group for the target category, wherein the first curve indicates a change in average single-session object stickiness as a function of a first proportion, where the first proportion is a proportion of the number of contents of the target category in a single session to the total number of contents in the single session; determining a first integral curve of the first curve; determining a target point on the first integral curve so that the absolute value of a first difference between average curve values on both sides of the target point on the first integral curve is maximized; Based on the target point, a proportion of content of the target category to be pushed to the target object group is determined.

2. The content push method according to claim 1, characterized in that: The obtaining of a first curve of the target object group with respect to the target category includes: Obtaining conversations occurring in the target object group; For each acquired session, determining the first proportion of the session and the single-session object stickiness; determining, based on the single-session object stickiness corresponding to each of the first proportions, the average single-session object stickiness corresponding to the first proportions; The first curve is determined based on each of the first proportions and the average single-session object stickiness corresponding to the first proportion.

3. The content push method according to claim 2, characterized in that: The single session object stickiness includes the single session duration and the number of contents browsed in a single session; The determining, based on the single-session object stickiness corresponding to each of the first proportions, the average single-session object stickiness corresponding to the first proportions includes: determining, based on the single session duration corresponding to each of the first proportions, an average single session duration corresponding to the first proportion; determining an average number of content viewed in a single session corresponding to the first proportion based on the number of content viewed in a single session corresponding to each of the first proportions; The average single-session object stickiness corresponding to the first proportion is determined based on the average single-session duration and the average number of contents browsed in a single session.

4. The content push method according to claim 1, wherein: The determining of the target point on the first integral curve includes: Assuming that the right direction of the horizontal axis of the coordinate system where the first integral curve is located is positive, setting an objective function, wherein the objective function is equal to the first difference between the first average curve value on the right side and the second average curve value on the left side of any point on the first integral curve; On the first integral curve, a point at which the objective function is maximized is determined as the target point.

5. The content push method according to claim 4, characterized in that: The first average curve value is determined by: Obtaining the maximum first proportion of the target object group; Determine a first integral value between the arbitrary point on the first integral curve and the point corresponding to the maximum first proportion; determining a second difference between the maximum first proportion and the abscissa of any one point; The first average curve value is determined based on the first integrated value and the second difference.

6. The content push method according to claim 4, characterized in that: The second average curve value is determined by: Obtaining the minimum first proportion of the target object group; Determine a second integral value between the point corresponding to the minimum first proportion and the arbitrary point on the first integral curve; determining a third difference between the abscissa of the any point and the minimum first proportion; The second average curve value is determined based on the second integrated value and the third difference.

7. The content push method according to claim 1, characterized in that: The target category is selected from a candidate category library; Before obtaining the first curve of the target object group for the target category, the content push method further includes: selecting an unselected candidate category from the candidate category library as the target category; After determining the proportion of content of the target category to be pushed to the target object group based on the target point, the content pushing method further includes: If the sum of the content proportions of the selected multiple target categories exceeds 100%, performing a first adjustment on the content proportions of the multiple target categories; If the sum of the content proportions of the selected multiple target categories does not exceed 100%, returning to the step of selecting an unselected candidate category from the candidate category library as the target category; If all the candidate categories in the candidate category library have been selected, and the sum of the content proportions of the selected multiple target categories still does not exceed 100%, a second adjustment is made to the content proportions of the multiple target categories.

8. The content push method according to claim 7, characterized in that: The first adjustment of the content proportions of the plurality of target categories includes: Dividing the content proportion of each target category by the sum of the content proportions of multiple target categories to obtain a first quotient; The first quotient is used as the first adjusted content proportion of the target category.

9. The content push method according to claim 7, characterized in that: The second adjustment of the content proportions of the plurality of target categories includes: Dividing the content proportion of each target category by the sum of the content proportions of multiple target categories to obtain a second quotient; The second quotient is used as the second adjusted content proportion of the target category.

10. The content push method according to claim 1, characterized in that: Determining the target object group to which the target object belongs includes: Obtain platform object collection and multiple object group division parameters; Determining an application order of target segmentation parameters based on the plurality of object group segmentation parameters; Applying the target partitioning parameter application sequence to the platform object set to obtain a target result tree corresponding to the target partitioning parameter application sequence, wherein each leaf node of the target result tree corresponds to an object group partitioned according to the target partitioning parameter application sequence; The divided object group to which the target object belongs is used as the target object group.

11. The content push method according to claim 10, characterized in that: The determining the application order of the target segmentation parameters based on the plurality of object group segmentation parameters includes: Taking the platform object set as the object group to be divided; Executing a first process, the first process comprising: applying unselected subject group division parameters from the plurality of subject group division parameters as subject group division parameters to be examined, respectively, to the subject groups to be divided, obtaining divided subject groups corresponding to each of the subject group division parameters to be examined, and determining, based on each of the divided subject groups, a contribution function value corresponding to each of the subject group division parameters to be examined; Using the partitioning parameter of the to-be-investigated object group with the largest contribution function value as the target object group partitioning parameter, updating the to-be-divided object group with the partitioned object group, and returning to the first process; The obtained multiple target object group division parameters are cascaded in the order of acquisition as the target division parameter application order.

12. The content push method according to claim 11, characterized in that: The determining, based on each of the divided object groups, a contribution function value corresponding to each of the division parameters of the to-be-investigated object groups, includes: For each of the parameters for dividing the subject groups to be examined, determining the sub-contribution function values of the plurality of divided subject groups corresponding to the parameters for dividing the subject groups to be examined; The contribution function value corresponding to the to-be-investigated object group division parameter is determined based on the sub-contribution function values of the plurality of divided object groups corresponding to the to-be-investigated object group division parameter.

13. The content push method according to claim 12, characterized in that: The determining of the sub-contribution function values of the plurality of divided object groups corresponding to the division parameters of the to-be-investigated object groups includes: Taking each of the divided object groups as the target object group in turn, returning to the step of obtaining the first curve of the target object group for the target category, to obtain the absolute value of the first difference at the target point as the single-object contribution value of the divided object group; Based on the single object contribution value, the sub-contribution function value of the divided object group is determined.

14. The content push method according to claim 13, characterized in that: The determining, based on the single object contribution value, the sub-contribution function value of the divided object group includes: Obtaining the number of objects in the divided object group; The sub-contribution function value of the divided object group is determined based on the number of objects in the divided object group and the single-object contribution value.

15. The content push method according to claim 10, characterized in that: The determining the application order of target segmentation parameters based on the plurality of object group segmentation parameters includes: determining the application order of the plurality of target segmentation parameters based on the plurality of object group segmentation parameters; The step of applying the target partition parameter application sequence to the platform object set to obtain a target result tree corresponding to the target partition parameter application sequence includes: applying multiple target partition parameter application sequences to the platform object set respectively to obtain the target result tree corresponding to each target partition parameter application sequence; The step of taking the object group to which the target object belongs as the target object group includes: taking the object group to which the target object belongs in the leaf node of the target result tree corresponding to each target partition parameter application sequence as the target object group corresponding to the target partition parameter application sequence; After determining the proportion of content of the target category to be pushed to the target object group based on the target point, the content pushing method further includes: Based on the determined proportion of the content of the target category pushed to the target object group corresponding to each application order of the target division parameters, the proportion of the content of the target category pushed to the target object is determined.

16. The content push method according to claim 15, characterized in that: The target category is selected from a candidate category library; The number of target partition parameter application orders is determined by: Obtaining the number of candidate categories in the candidate category library; The number of application orders of the target partitioning parameters is determined to be twice the number of the candidate categories.

17. A content push device, characterized in that: include: A first determining unit, configured to determine a target object group to which a target object belongs; an acquiring unit, configured to acquire a first curve of the target object group for the target category, wherein the first curve indicates a change in average single-session object stickiness as a function of a first proportion, where the first proportion is a proportion of the number of contents of the target category in a single session to the total number of contents in the single session; a second determining unit, configured to determine a first integral curve of the first curve; a third determining unit, configured to determine a target point on the first integral curve so that an absolute value of a first difference between average curve values on both sides of the target point on the first integral curve is maximized; The push unit is configured to determine, based on the target point, a proportion of content of the target category to be pushed to the target object group.

18. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the content pushing method according to any one of claims 1 to 16 is implemented.

19. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the content pushing method according to any one of claims 1 to 16 is implemented.

20. A computer program product, comprising a computer program, wherein the computer program is read and executed by a processor of a computer device, so that the computer device executes the content pushing method according to any one of claims 1 to 16.

Images