Traffic distribution method and device, storage medium and electronic equipment

Abstract

The invention discloses a traffic distribution method and device, a storage medium and electronic equipment. The method comprises the steps that a target content object to be processed is acquired from a cold start content pool, and the published duration of the content object contained in the cold start content pool is smaller than a target duration threshold value; a cold start state threshold value matched with content evaluation characteristics of the target content object and accumulated exposure flow counted by the target content object in a target time period after the target content object is published are determined, and the content evaluation characteristics are used for evaluating the content propagation potential of the target content object; determining a current cold start state of the target content object according to a comparison result of the accumulated exposure flow and a cold start state threshold value; and distributing propagation flow matched with the cold start state to the target content object. The technical problem that an existing flow distribution mode is poor in accuracy is solved.

Description

Technical Field

[0001] The present application relates to the field of computers, and more particularly, to a traffic allocation method, apparatus, storage medium, and electronic device. Background Art

[0002] Nowadays, in many content recommendation platforms or content sharing communities, more and more creative accounts rely on the high-quality content they create to attract the attention of many fans. However, for some creative accounts, due to the lack of a large fan base, even if some content is published, there will be a lack of sufficient feedback data, resulting in the inability to achieve personalized recommendation for the newly published content, that is, the newly published content cannot obtain a wide enough dissemination traffic (i.e., exposure).

[0003] To overcome the above problems, the currently commonly used traffic allocation method is to provide a cold start scheme, which includes: a uniform quantity guarantee stage and a boost amplification stage. Among them, in the uniform quantity guarantee stage, the upper limit of the traffic that the creative account can obtain in this stage is restricted according to the online performance of the historical creative content of the creative account; while in the boost amplification stage, the boost category to which the creative content belongs is determined according to the real-time posterior click-through rate, and the corresponding boost traffic of the boost category is configured for it.

[0004] However, the cold start scheme for allocating dissemination traffic provided in the above related technologies only relies on the online performance of the creative content. However, due to the lack of sufficient posterior online interaction data in the initial stage, there is a certain deviation in the evaluation result of the creative content, that is, high-quality creative content cannot obtain enough dissemination traffic for distribution. On the contrary, low-quality creative content gets a large amount of exposure, resulting in a waste of traffic. In other words, the existing traffic allocation method has the problem of poor accuracy.

[0005] In view of the above problems, no effective solution has been proposed yet. Summary of the Invention

[0006] Embodiments of the present application provide a traffic allocation method, apparatus, storage medium, and electronic device to at least solve the technical problem of poor accuracy in the existing traffic allocation method.

[0007] According to one aspect of the embodiments of the present application, a traffic allocation method is provided, including: obtaining a target content object to be processed from a cold start content pool, where the published duration of the content objects included in the cold start content pool is less than a target duration threshold; determining a cold start state threshold adapted to the content evaluation characteristics of the target content object, and the cumulative exposure traffic counted within a target time period after the target content object is published, where the content evaluation characteristics are used to evaluate the content dissemination potential of the target content object itself; determining the current cold start state of the target content object according to the result of comparing the cumulative exposure traffic with the cold start state threshold; and allocating dissemination traffic adapted to the cold start state to the target content object.

[0008] According to another aspect of the embodiments of the present application, a traffic allocation device is further provided, including: an obtaining unit, configured to obtain a target content object to be processed from a cold start content pool, where the published duration of the content objects included in the cold start content pool is less than a target duration threshold; a first determination unit, configured to determine a cold start state threshold adapted to the content evaluation characteristics of the target content object, and the cumulative exposure traffic counted within a target time period after the target content object is published, where the content evaluation characteristics are used to evaluate the content dissemination potential of the target content object itself; a second determination unit, configured to determine the current cold start state of the target content object according to the result of comparing the cumulative exposure traffic with the cold start state threshold; and an allocation unit, configured to allocate dissemination traffic adapted to the cold start state to the target content object.

[0009] Optionally, in this embodiment, the above second determination unit includes: a first obtaining module, configured to obtain a cold start trial state threshold from the determined cold start state threshold, where the cold start trial state threshold is used to indicate the exposure traffic threshold for exiting the cold start trial state; a first determination module, configured to determine that the target content object is currently in the cold start trial state and retain the target content object in the cold start content pool when the cumulative exposure traffic is less than or equal to the cold start trial state threshold; and a second determination module, configured to determine that the target content object exits the cold start trial state when the cumulative exposure traffic is greater than the cold start trial state threshold.

[0010] Optionally, in this embodiment, the above-mentioned second determination unit further includes: a second acquisition module, configured to acquire a cold start acceptance state threshold from the determined cold start state thresholds, where the cold start acceptance state threshold is used to indicate the exposure traffic threshold for exiting the cold start acceptance state; a third determination module, configured to determine that the target content object is currently in the cold start acceptance state and retain the target content object in the cold start content pool when the cumulative exposure traffic is greater than the cold start trial state threshold and less than or equal to the cold start acceptance state threshold; a fourth determination module, configured to determine that the target content object currently exits the cold start acceptance state and enters the traffic competition state and delete the target content object from the cold start content pool when the cumulative exposure traffic is greater than the cold start acceptance state threshold.

[0011] Optionally, in this embodiment, the above-mentioned first determination unit includes: a fifth determination module, configured to determine the potential value of the target content object becoming a popular content object by using content evaluation features in the potential prediction model, where the potential prediction model is a decision tree model trained by using positive sample objects indicating popular content objects and negative sample objects indicating non-popular content objects; a third acquisition module, configured to acquire a trial boundary threshold configured for the cold start trial state in the cold start state; a sixth determination module, configured to determine a cold start trial state threshold adapted to the target content object by using the trial boundary threshold adjusted based on the potential value.

[0012] Optionally, in this embodiment, the above-mentioned sixth determination module is further configured to: adjust the first trial boundary threshold in the trial boundary threshold by using the potential value to obtain an updated first trial boundary threshold; compare the second trial boundary threshold and the updated first trial boundary threshold in the trial boundary threshold, where the second trial boundary threshold is less than the first trial boundary threshold; determine the second trial boundary threshold as the cold start trial state threshold when the second trial boundary threshold is greater than the updated first trial boundary threshold; determine the updated first trial boundary threshold as the cold start trial state threshold when the updated first trial boundary threshold is greater than the second trial boundary threshold.

[0013] Optionally, in this embodiment, the above-mentioned apparatus further includes: a third determination unit, configured to determine a current sample object from the acquired positive sample objects and negative sample objects; an extraction unit, configured to extract sample content evaluation features from the current sample object, where the sample content evaluation features include sample attribute features, sample quality features, and sample semantic features of the current sample object; a training unit, configured to train the initialized potential prediction model by using the sample content evaluation features until the output sample potential value reaches a first convergence condition.

[0014] Optionally, in this embodiment, the above-mentioned first determination unit also includes: a seventh determination module, which is used to determine the recommendation label category coefficient corresponding to the target content object using the content evaluation feature in the content recommendation model, wherein the content recommendation model is a decision tree model that is learned using the exposure data counted by the target content object in the cold start trial state; an eighth determination module, which is used to determine the cold start acceptance state threshold that is compatible with the target content object using the recommendation label category coefficient and the exposure factor.

[0015] Optionally, in this embodiment, the eighth determination module is used to: determine a recommended parameter based on a recommended tag category coefficient; and multiply the recommended parameter by an exposure factor to obtain a cold start acceptance state threshold.

[0016] Optionally, in this embodiment, the above-mentioned device also includes: a first acquisition unit, used to obtain sample exposure data obtained by the sample object in multiple sample collection cycles, wherein the sample exposure data includes the sample exposure flow and the sample estimated click-through rate of the sample object in each sample collection cycle; a second acquisition unit, used to obtain the reference sample exposure flow of the sample object when the sample estimated click-through rate reaches the second convergence condition based on the change of the sample estimated click-through rate in multiple sample collection cycles; a division unit, used to divide the reference sample exposure flow to obtain multiple label category coefficients matching the sample object; a first training unit, used to train the initialized content recommendation model using multiple label category coefficients and sample content evaluation features extracted from the sample object until the third convergence condition is reached.

[0017] Optionally, in this embodiment, the above-mentioned allocation unit includes: a ninth determination module, used to determine the current timestamp and the release timestamp of the target content object; a fourth acquisition module, used to obtain the account cumulative exposure of the target account that publishes the target content object; a tenth determination module, used to use the current timestamp, release timestamp, cumulative exposure traffic and account cumulative exposure to determine the traffic allocation coefficient; a multiplication module, used to multiply the traffic allocation coefficient with the recall ranking result of the target content object to obtain the propagation traffic.

[0018] According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned traffic distribution method when running.

[0019] According to another aspect of the embodiments of the present application, there is provided a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the traffic allocation method as described above.

[0020] According to another aspect of the embodiments of the present application, there is also provided an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to execute the above traffic allocation method through the computer program.

[0021] In the embodiments of the present application, a target content object to be processed is obtained from a cold start content pool, where the published duration of the content objects included in the cold start content pool is less than a target duration threshold. Then, a cold start state threshold adapted to the content evaluation feature of the target content object is determined, and the cumulative exposure traffic counted within a target time period after the target content object is published is determined, where the content evaluation feature is used to evaluate the content propagation potential of the target content object itself. Next, according to the result of comparing the cumulative exposure traffic with the cold start state threshold, the current cold start state of the target content object is determined. Furthermore, propagation traffic adapted to the cold start state is allocated to the target content object. In other words, by using the embodiments of the present application, by using the quality feature of the target content object itself to determine the cold start state threshold matching it, and then using the cold start state threshold to determine the current cold start state of the target content object, so as to allocate propagation traffic adapted to the cold start state to the target content object, the purpose of getting rid of posterior parameters in the process of traffic allocation is achieved. The problem that the existing traffic allocation method has poor accuracy is solved, and the technical effect of improving the accuracy of traffic allocation is realized. Description of the Drawings

[0022] The drawings described herein are used to provide a further understanding of the present application, and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation of the present application. In the drawings:

[0023] Figure 1 is a schematic diagram of an application environment of an optional traffic allocation method according to the embodiments of the present application;

[0024] Figure 2 is a flowchart of an optional traffic allocation method according to the embodiments of the present application;

[0025] Figure 3 is a schematic diagram of an optional traffic allocation method according to the embodiments of the present application;

[0026] Figure 4 is a flowchart of an optional traffic allocation method according to an embodiment of the present application;

[0027] Figure 5 is a flowchart of an optional traffic allocation method according to an embodiment of the present application;

[0028] Figure 6 is a flowchart of an optional traffic allocation method according to an embodiment of the present application;

[0029] Figure 7 is a flowchart of an optional traffic allocation method according to an embodiment of the present application;

[0030] Figure 8 is a schematic diagram of an optional traffic allocation method according to an embodiment of the present application;

[0031] Figure 9 is a schematic structural diagram of an optional traffic allocation device according to an embodiment of the present application;

[0032] Figure 10 is a schematic structural diagram of an optional electronic device according to an embodiment of the present application. Detailed implementation manners

[0033] In order to enable those skilled in the art to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0034] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0035] According to one aspect of the embodiments of the present application, a traffic allocation method is provided. Optionally, as an alternative implementation, the above traffic allocation can be but is not limited to being applied to, for example, Figure 1 the environment shown. As Figure 1 shown, the terminal device 102 includes a memory 104 for storing various data generated during the operation of the terminal device 102, a processor 106 for processing and computing the above various data, and a display 108 for displaying target content objects. The terminal device 102 can perform data interaction with the server 112 through the network 110. The server 112 is connected to the database 114, and the database 114 is used to store various data. The terminal device 102 can run an application program for spreading the target content object.

[0036] Furthermore, the specific application process of the above method in the Figure 1 shown environment is as follows:

[0037] Execute step S102, and the terminal device 102 sends a propagation request for requesting the propagation of the target content object to the server 112 through the network 110.

[0038] Then execute steps S104 - S110. After receiving the propagation request, the server 112 obtains the target content object to be processed from the cold start content pool, where the published duration of the content objects included in the cold start content pool is less than the target duration threshold; the server 112 obtains the target content object to be processed from the cold start content pool, where the published duration of the content objects included in the cold start content pool is less than the target duration threshold; the server 112 determines a cold start state threshold adapted to the content evaluation characteristics of the target content object, and the cumulative exposure traffic counted within the target time period after the target content object is published, where the content evaluation characteristics are used to evaluate the content propagation potential of the target content object itself; the server 112 determines the current cold start state of the target content object according to the result of comparing the cumulative exposure traffic with the cold start state threshold.

[0039] Then execute step S112, and the server 112 allocates propagation traffic adapted to the cold start state to the target content object.

[0040] Then execute step S114, and the server 112 sends a prompt message for prompting the use of the propagation traffic to propagate the target content object to the terminal device 102 through the network 110.

[0041] In an embodiment of the present application, a target content object to be processed is obtained from a cold start content pool, where the published duration of the content objects included in the cold start content pool is less than a target duration threshold. Then, a cold start state threshold adapted to the content evaluation feature of the target content object is determined, and the cumulative exposure traffic counted within a target time period after the target content object is published, where the content evaluation feature is used to evaluate the content dissemination potential of the target content object itself. Next, according to the result of comparing the cumulative exposure traffic with the cold start state threshold, the current cold start state of the target content object is determined. Furthermore, a dissemination traffic adapted to the cold start state is allocated to the target content object. In other words, by adopting the embodiment of the present application, by using the quality feature of the target content object itself to determine a cold start state threshold matching it, and then using the cold start state threshold to determine the current cold start state of the target content object, so as to allocate a dissemination traffic adapted to the cold start state to the target content object, the purpose of getting rid of posterior parameters in the process of traffic allocation is achieved. The problem that the existing traffic allocation method has poor accuracy is solved, and the technical effect of improving the accuracy of traffic allocation is realized.

[0042] Optionally, in this embodiment, the above terminal device may be a terminal device configured with a target client, and may include but are not limited to at least one of the following: mobile phone (such as Android mobile phone, iOS mobile phone, etc.), laptop computer, tablet computer, handheld computer, MID (Mobile Internet Devices), PAD, desktop computer, smart TV, etc. The target client may be a video client, an instant messaging client, a browser client, an education client, etc. The above network may include but are not limited to: wired network, wireless network, where the wired network includes: local area network, metropolitan area network and wide area network, and the wireless network includes: Bluetooth, WIFI and other networks that implement wireless communication. The above server may be a single server, or a server cluster composed of multiple servers, or a cloud server. The above is only an example, and this embodiment does not make any limitation thereto.

[0043] Optionally, as an alternative solution, as Figure 2 shown, the above traffic allocation method includes:

[0044] S202, obtain a target content object to be processed from a cold start content pool, where the published duration of the content objects included in the cold start content pool is less than a target duration threshold;

[0045] S204. Determine a cold start state threshold adapted to the content evaluation characteristics of the target content object, and the cumulative exposure traffic counted within the target time period after the target content object is published, where the content evaluation characteristics are used to evaluate the content dissemination potential of the target content object itself;

[0046] S206. Determine the current cold start state of the target content object according to the result of comparing the cumulative exposure traffic with the cold start state threshold;

[0047] S208. Allocate dissemination traffic adapted to the cold start state for the target content object.

[0048] It should be noted that the above traffic allocation method can be but is not limited to being applied to the traffic allocation scenario of new works (i.e., target content objects) published by user accounts on a network platform. Specifically, in this embodiment, the above network platform can be but is not limited to including: social media network platforms, news network platforms, e-book network platforms, song network platforms, etc., and no limitations are imposed in this regard in this embodiment. Correspondingly, the above target content objects can be but are not limited to including: short videos, articles, e-books, original songs, etc., and no limitations are imposed in this regard in this embodiment either.

[0049] Optionally, the above cold start content pool can be but is not limited to indicating a storage area in the background server corresponding to the network platform for publishing the target content object. In this area, content objects with a published duration less than the target duration threshold and awaiting traffic allocation are stored, such as newly published articles, songs, news, short videos, etc. that are awaiting traffic allocation, and no limitations are imposed in this regard in this embodiment.

[0050] Furthermore, the above content evaluation characteristics can be but are not limited to indicating the characteristics used to describe the actual content quality of the target content object. Specifically, the above content platform characteristics can be but are not limited to including: the attribute characteristics of the target content object, the quality characteristics of the target content object, the semantic characteristics of the target content object, etc.

[0051] Furthermore, the above attribute characteristics of the target content object can be but are not limited to the characteristics used to describe the actual associated information of the target content object. Specifically, the above attribute characteristics of the target content object can be but are not limited to including: the type of the target content object, the keywords corresponding to the target content object, the author information of the target content object, etc.

[0052] The quality features of the target content object may include, but are not limited to, high-order features of the target content object and quality labels of the target content object, etc., wherein the high-order features may include, but are not limited to, content quality evaluation coefficients of the target content object extracted using a network model, and the quality labels may include, but are not limited to, labels that are manually labeled according to preset rules to describe the quality level of the target content object, and different quality levels correspond to different overall effects of the content objects.

[0053] Furthermore, the semantic features of the target content object may be, but are not limited to, features used to describe the meaning attributes or characteristics of the target content object, such as the meaning, part of speech, grammatical function, emotional color, etc. of the words included in the target content object. Specifically, the semantic features of the target content object may be, but are not limited to, generated by a bidirectional pre-trained language model (Bidirectional Encoder Representations From Transformers, referred to as Bert).

[0054] Optionally, in this embodiment, the cold start state threshold may be, but is not limited to, an indicator for determining the cold start state corresponding to the target content object. Specifically, the cold start state corresponding to the target content object may include, but is not limited to, a cold start trial state and a cold start takeover state. For example, the traffic distribution diagram of the target content object in the cold start trial stage and the cold start takeover stage may be, but is not limited to, as follows: Figure 3 shown.

[0055] Furthermore, in the cold start trial phase, during the initial stage of a system or service, the system cannot accurately make personalized recommendations or predictions due to the lack of sufficient historical data or user feedback. The cold start problem is prevalent in fields such as recommendation systems, machine learning, and information retrieval. During the cold start phase, the system needs to obtain or infer the user's interests, preferences, or needs through other means in order to provide personalized recommendations or predictions. Common cold start solutions include content-based recommendations, collaborative filtering, tag or keyword analysis, user surveys, etc. By solving the cold start problem, the system can provide more accurate and personalized services and improve user experience and satisfaction. Furthermore, in this embodiment, the goal of the cold start trial phase is to accumulate original exposure and consumption data for the target content object to obtain relatively confident posterior data.

[0056] Furthermore, the cold start takeover stage refers to the stage after the cold start trial stage, which provides further exposure and promotion for the newly released content. In the cold start takeover stage, the system dynamically adjusts the exposure traffic of news or content based on the results and feedback of the cold start trial stage, combined with other factors such as content quality and user feedback. Through the strategies and mechanisms of the cold start takeover stage, the system can accelerate the growth of high-quality content into popular content and increase its exposure and influence. The goal of the cold start takeover stage is the climbing stage of traffic acquisition for the target content object. In this stage, the content recommendation model will be used to eliminate the deviation in traffic allocation to the target content object caused by factors such as user personal preferences and historical behaviors, thereby improving the fairness and accuracy of traffic allocation.

[0057] It should be noted that in this embodiment, the cumulative exposure traffic of the target content object counted in the target time period after being published and the above can be obtained from the remote dictionary service storage system (Remote Dictionary Server, Redis for short) in the backend server corresponding to the network platform for publishing the target content object, but is not limited to. Among them, Redis is an open source memory data structure storage system that supports multiple data structures, such as strings, hash tables, lists, sets and ordered sets. Redis has the characteristics of fast read and write speed and low latency, and is suitable for high-concurrency data access scenarios. It also provides rich functions, such as transaction support, publish and subscribe, persistence and cluster mode. Redis is widely used in cache, message queue, real-time statistics and distributed locks. The above cumulative exposure traffic can be, but is not limited to, used to indicate the traffic counted in the target time period after the target content object is published. Such as the number of visits, click-through rate, etc. counted in the target time period after the target content object is published.

[0058] Optionally, in this embodiment, the above-mentioned allocation of propagation traffic that is compatible with the cold start state to the target content object may include, but is not limited to: utilizing the current timestamp and the release timestamp of the target content object and the cumulative account exposure and cumulative exposure traffic of the target account of the target content object and the recall sorting result of the target content object to determine the propagation traffic that is compatible with the cold start state.

[0059] As an optional implementation, assuming that the above traffic distribution method is applied to a scenario where traffic is distributed to a new article, using Figure 4 The following steps are shown to illustrate the above method by example:

[0060] Step S402 is executed to obtain a new article to be non-allocated traffic from the cold start content pool (ie, the target content is an object).

[0061] Then, step S404 is executed to obtain the cold start trial state threshold, the cold start succession state threshold and the cumulative exposure flow counted within the target time period after the new article is published that are adapted to the content evaluation characteristics of the new article.

[0062] Then execute step S406 to determine whether the cumulative exposure flow is greater than the cold start trial state threshold. When the cumulative exposure flow is less than the cold start trial state threshold, execute step S408-1 to determine that the new article is currently in a cold start trial state, and retain the new article in the cold start content pool. Then execute step S404 again. When the cumulative exposure flow is greater than the cold start trial state threshold and less than or equal to the cold start takeover state threshold, execute step S408-2 to determine that the new article is currently in a cold start takeover state, and retain the new article in the cold start content pool. Then execute step S410, and when the new article is currently in a cold start takeover state, continue to determine whether the cumulative exposure flow of the new article is greater than the cold start takeover state threshold.

[0063] When the cumulative exposure flow is greater than the cold start acceptance state threshold, step S412-1 is executed to delete the new article from the cold start content pool. When the cumulative exposure flow is less than or equal to the cold start acceptance state threshold, step S412-2 is executed to keep the new article in the cold start content pool, allocate the new article with the propagation flow that matches the cold start state, and execute step S408-2 again.

[0064] In an embodiment of the present application, a target content object to be processed is obtained from a cold start content pool, wherein the published duration of the content objects contained in the cold start content pool is less than the target duration threshold. Then, a cold start state threshold adapted to the content evaluation feature of the target content object and the cumulative exposure flow counted in the target time period after the target content object is published are determined, wherein the content evaluation feature is used to evaluate the content dissemination potential of the target content object itself. Then, according to the result of comparing the cumulative exposure flow with the cold start state threshold, the current cold start state of the target content object is determined. Then, the target content object is allocated a propagation flow adapted to the cold start state. In other words, by using the embodiment of the present application, the target content object is determined by using its own quality characteristics to determine the cold start state threshold matched therewith, and then the cold start state threshold is used to determine the current cold start state of the target content object, thereby allocating a propagation flow adapted to the cold start state to the target content object, achieving the purpose of getting rid of the a posteriori parameters in the process of flow distribution. The problem of poor accuracy in the existing flow distribution method is solved, and the technical effect of improving the accuracy of flow distribution is achieved.

[0065] Optionally, as an alternative solution, determining the current cold start state of the target content object according to the result of comparing the cumulative exposure traffic with the cold start state threshold includes:

[0066] Obtain the cold start trial state threshold from the determined cold start state thresholds, where the cold start trial state threshold is used to indicate the exposure traffic threshold for exiting the cold start trial state;

[0067] When the cumulative exposure traffic is less than or equal to the cold start trial state threshold, determine that the target content object is currently in the cold start trial state and retain the target content object in the cold start content pool;

[0068] When the cumulative exposure traffic is greater than the cold start trial state threshold, determine that the target content object exits the cold start trial state.

[0069] It should be noted that in this embodiment, during the cold start trial stage, at the initial stage of a system or service, due to the lack of sufficient historical data or user feedback, the system cannot accurately perform personalized recommendation or prediction. The cold start problem is common in fields such as recommendation systems, machine learning, and information retrieval. During the cold start stage, the system needs to use other means to obtain or infer the interests, preferences, or needs of users in order to provide personalized recommendations or predictions. Common cold start solutions include content-based recommendations, collaborative filtering, tag or keyword analysis, user surveys, etc. By solving the cold start problem, the system can provide more accurate and personalized services, improving the user experience and satisfaction. Further, in this embodiment, the goal of the cold start trial stage is to accumulate original exposure and consumption data for the target content object to obtain relatively reliable posterior data.

[0070] Further, in this embodiment, the following steps may be used, but are not limited to, to determine the cold start trial state threshold: Use the content evaluation features in the potential prediction model to determine the potential value of the target content object becoming a popular content object, where the potential prediction model is a decision tree model trained using positive sample objects indicating popular content objects and negative sample objects indicating non-popular content objects. Then, obtain the trial boundary threshold configured for the cold start trial state in the cold start state. Furthermore, use the trial boundary threshold adjusted based on the potential value to determine the cold start trial state threshold adapted to the target content object.

[0071] In an embodiment of the present application, a cold start tentative state threshold is obtained from the determined cold start state threshold, wherein the cold start tentative state threshold is used to indicate an exposure flow threshold for exiting the cold start tentative state. When the cumulative exposure flow is less than or equal to the cold start tentative state threshold, it is determined that the target content object is currently in a cold start tentative state, and the target content object is retained in the cold start content pool. When the cumulative exposure flow is greater than the cold start tentative state threshold, it is determined that the target content object exits the cold start tentative state. In other words, using an embodiment of the present application, by comparing the cumulative exposure flow of the target content object with the cold start tentative state threshold, it is quickly determined whether the current target content object is in a cold start tentative state. Thereby achieving the technical effect of improving the efficiency of determining the cold start state of the target content object.

[0072] Optionally, as an optional solution, according to the result of comparing the cumulative exposure flow with the cold start state threshold, determining the current cold start state of the target content object further includes:

[0073] Obtaining a cold start acceptance state threshold from the determined cold start state threshold, wherein the cold start acceptance state threshold is used to indicate an exposure flow threshold for exiting the cold start acceptance state;

[0074] When the cumulative exposure flow is greater than the cold start trial state threshold and less than or equal to the cold start acceptance state threshold, determining that the target content object is currently in the cold start acceptance state, and retaining the target content object in the cold start content pool;

[0075] When the accumulated exposure traffic is greater than the cold start acceptance state threshold, it is determined that the target content object currently exits the cold start acceptance state and enters the traffic competition state, and the target content object is deleted from the cold start content pool.

[0076] It should be noted that, in this embodiment, the cold start takeover stage refers to the stage after the cold start trial stage, which provides further exposure and promotion for the newly released content. In the cold start takeover stage, the system dynamically adjusts the exposure traffic of news or content based on the results and feedback of the cold start trial stage, combined with other factors such as content quality, user feedback, etc. Through the strategies and mechanisms of the cold start takeover stage, the system can accelerate the growth of high-quality content into popular content and increase its exposure and influence. The goal of the cold start takeover stage is the climbing stage of traffic acquisition for the target content object. In this stage, the content recommendation model will be used to eliminate the deviation in traffic allocation to the target content object caused by factors such as user personal preferences and historical behaviors, thereby improving the fairness and accuracy of traffic distribution.

[0077] Optionally, in this embodiment, the following steps may be used, but are not limited to, to determine the cold start acceptance state threshold: Use content evaluation features in the content recommendation model to determine the recommended label category coefficient corresponding to the target content object, where the content recommendation model is a decision tree model learned using the exposure data statistically obtained in the cold start trial state of the target content object. Then, use the recommended label category coefficient and the exposure volume factor to determine the cold start acceptance state threshold adapted to the target content object.

[0078] Further, the above traffic competition state may be, but is not limited to, a state where the target content object has passed the cold start state, and after obtaining traffic that matches its own content, it completely attracts users' attention and exposure by content.

[0079] In the embodiment of the present application, obtain the cold start acceptance state threshold from the determined cold start state thresholds, where the cold start acceptance state threshold is used to indicate the exposure traffic threshold for exiting the cold start acceptance state. Then, when the cumulative exposure traffic is greater than the cold start trial state threshold and less than or equal to the cold start acceptance state threshold, determine that the target content object is currently in the cold start acceptance state, and retain the target content object in the cold start content pool. Next, when the cumulative exposure traffic is greater than the cold start acceptance state threshold, determine that the target content object currently exits the cold start acceptance state and enters the traffic competition state, and delete the target content object from the cold start content pool. In other words, in the embodiment of the present application, when the target content object has entered the cold start trial state, by comparing the cumulative exposure traffic of the target content object with the cold start trial state threshold, it is quickly determined whether the current target content object is in the cold start acceptance state. Thus, the technical effect of improving the efficiency of determining the cold start state of the target content object is achieved.

[0080] Optionally, as an alternative solution, determining the cold start state threshold adapted to the content evaluation features of the target content object includes:

[0081] Use content evaluation features in the potential prediction model to determine the potential value of the target content object becoming a popular content object, where the potential prediction model is a decision tree model trained using positive sample objects indicating popular content objects and negative sample objects indicating non-popular content objects;

[0082] Obtain the trial boundary threshold configured for the cold start trial state in the cold start state;

[0083] Use the trial boundary threshold adjusted based on the potential value to determine the cold start trial state threshold adapted to the target content object.

[0084] It should be noted that in this embodiment, the above potential prediction model can be trained by, but not limited to, the following steps: using the sample content evaluation features corresponding to the positive sample object and the negative sample object respectively to train the initialized potential prediction model until the output sample potential value reaches the first convergence condition.

[0085] Further, in this embodiment, the potential value of the above target content object becoming a popular content object can be used to indicate, but not limited to, the probability that the above target content object becomes a popular content object, and the value range of this probability is [0, 1].

[0086] Optionally, in this embodiment, the above trial boundary threshold can include, but not limited to, the predefined upper limit threshold and lower limit threshold of traffic exposure obtained through a large number of offline experimental analyses. Further, determining the cold start trial state threshold adapted to the target content object by using the trial boundary threshold adjusted based on the potential value can include, but not limited to: adjusting the first trial boundary threshold (i.e., the upper limit threshold of traffic exposure) in the trial boundary threshold by using the potential value to obtain the updated first trial boundary threshold. Then, comparing the second trial boundary threshold (i.e., the lower limit threshold of traffic exposure) in the trial boundary threshold with the updated first trial boundary threshold, where the second trial boundary threshold is less than the first trial boundary threshold. Then, in the case where the second trial boundary threshold is greater than the updated first trial boundary threshold, the second trial boundary threshold is determined as the cold start trial state threshold. In the case where the updated first trial boundary threshold is greater than the second trial boundary threshold, the updated first trial boundary threshold is determined as the cold start trial state threshold.

[0087] Optionally, as an alternative implementation manner, the following steps shown in Figure 5 are used to give an example and explanation of the above traffic allocation method:

[0088] Execute steps S502 - S506, input the content evaluation features into the potential prediction model to obtain the potential value of the target content object becoming a popular content object. Obtain the upper limit threshold and lower limit threshold of traffic exposure configured for the cold start trial state in the cold start state. Then, adjust the upper limit threshold of traffic exposure by using the potential value to obtain the updated first trial boundary threshold.

[0089] Then, step S508 is executed to determine whether the updated first trial boundary threshold is greater than the traffic exposure lower limit threshold. If the updated first trial boundary threshold is greater than the traffic exposure lower limit threshold, step S510-1 is executed to determine the updated first trial boundary threshold as the cold start trial state threshold. If the updated first trial boundary threshold is less than the traffic exposure lower limit threshold, step S510-2 is executed to determine the traffic exposure lower limit threshold as the cold start trial state threshold.

[0090] In this embodiment, the potential value of the target content object to become a popular content object is used to constrain the upper threshold of traffic exposure. The higher the potential value of the target content object to become a popular content object, the more traffic it is likely to obtain. When the potential value of the target content object to become a popular content object is low, that is, when it is impossible to become a popular content object, at least the minimum traffic is selected as a guarantee. This avoids the problem of a continuously low cold start threshold.

[0091] It should be noted that, in this embodiment, after being generated, the above cold start acceptance state threshold will be stored in the Redis in the backend server corresponding to the network platform for publishing the target content object. During the application process, the above cold start acceptance state threshold can be directly extracted from Redis.

[0092] In an embodiment of the present application, the potential value of the target content object to become a popular content object is determined by using the content evaluation feature in the potential prediction model, wherein the potential prediction model is a decision tree model trained by using positive sample objects for indicating popular content objects and negative sample objects for indicating non-popular content objects. Then, the trial boundary threshold configured for the cold start trial state in the cold start state is obtained. Next, the cold start trial state threshold adapted to the target content object is determined by using the trial boundary threshold adjusted based on the potential value. In other words, in an embodiment of the present application, the cold start state threshold matching the target content object is determined by using the quality characteristics of the target content object itself, and then the cold start state threshold is used to determine the current cold start state of the target content object, thereby allocating the propagation flow adapted to the cold start state to the target content object, thereby achieving the purpose of getting rid of the posterior parameter in the process of flow distribution. The problem of poor accuracy in the existing flow distribution method is solved, and the technical effect of improving the accuracy of flow distribution is achieved.

[0093] Optionally, as an optional solution, using the trial boundary threshold adjusted based on the potential value to determine the cold start trial state threshold adapted to the target content object includes:

[0094] Using the potential value to adjust a first tentative boundary threshold in the tentative boundary thresholds to obtain an updated first tentative boundary threshold;

[0095] Compare the second trial boundary threshold and the updated first trial boundary threshold in the trial boundary threshold, where the second trial boundary threshold is less than the first trial boundary threshold;

[0096] In the case where the second trial boundary threshold is greater than the updated first trial boundary threshold, determine the second trial boundary threshold as the cold start trial state threshold;

[0097] In the case where the updated first trial boundary threshold is greater than the second trial boundary threshold, determine the updated first trial boundary threshold as the cold start trial state threshold.

[0098] It should be noted that in this embodiment, the above first trial boundary threshold can be but is not limited to the pre-defined upper limit threshold of traffic exposure, and the above second trial boundary threshold can be but is not limited to indicating the pre-defined lower limit threshold of traffic exposure.

[0099] Optionally, in this embodiment, the above adjustment of the first trial boundary threshold in the trial boundary threshold by using the potential value to obtain the updated first trial boundary threshold can be but is not limited to including: determining the calculation result obtained by calculating the potential value and the first trial boundary threshold as the updated first trial boundary threshold. For example, perform a multiplication operation on the potential value and the first trial boundary threshold to obtain the product between the potential value and the first trial boundary threshold. Then, use the product between the potential value and the first trial boundary threshold to obtain the updated first trial boundary threshold. That is, the first trial boundary threshold = ceil(T upper × s), where T upper is the first trial boundary threshold, s is the potential value, and ceil is the rounding-down calculation.

[0100] Furthermore, for example, the above comparison of the second trial boundary threshold and the updated first trial boundary threshold in the trial boundary threshold to obtain the cold start trial state threshold can be but is not limited to including: t = max(T lower , ceil(T upper × s)), where the above t is the cold start trial state threshold, and the above T lower is the second trial boundary threshold.

[0101] In an embodiment of the present application, the first trial boundary threshold in the trial boundary threshold is adjusted using the potential value to obtain an updated first trial boundary threshold. Then, the second trial boundary threshold in the trial boundary threshold is compared with the updated first trial boundary threshold, where the second trial boundary threshold is less than the first trial boundary threshold. In the case where the second trial boundary threshold is greater than the updated first trial boundary threshold, the second trial boundary threshold is determined as the cold start trial state threshold. In the case where the updated first trial boundary threshold is greater than the second trial boundary threshold, the updated first trial boundary threshold is determined as the cold start trial state threshold. By determining the cold start state threshold matching the target content object using the quality characteristics of the target content object itself, and then using the cold start state threshold to determine the current cold start state of the target content object, so as to allocate the propagation traffic adapted to the cold start state for the target content object, the purpose of getting rid of the posterior parameters in the process of traffic allocation is achieved. The problem that the existing traffic allocation method has poor accuracy is solved, and the technical effect of improving the accuracy of traffic allocation is realized.

[0102] Optionally, as an alternative solution, before obtaining the target content object to be processed from the cold start content pool, it further includes:

[0103] Determine the current sample object from the obtained positive sample objects and negative sample objects;

[0104] Extract sample content evaluation features from the current sample object, where the sample content evaluation features include sample attribute features, sample quality features, and sample semantic features of the current sample object;

[0105] Train the initialized potential prediction model using the sample content evaluation features until the output sample potential value reaches the first convergence condition.

[0106] It should be noted that in this embodiment, the above positive sample object may but is not limited to a sample content object whose corresponding page view (PV) reaches a predetermined PV threshold and whose corresponding click through rate (CTR) reaches a predetermined CTR threshold. The above negative sample object may but is not limited to a sample content object whose corresponding page view (PV) does not reach the predetermined PV threshold and / or whose corresponding click through rate (CTR) does not reach the predetermined CTR threshold. Among them, the above predetermined PV threshold and the above CTR threshold are empirical parameters obtained through a large number of offline experimental analyses. The predetermined PV thresholds corresponding to different types of sample content objects are different, and the predetermined CTR thresholds corresponding to different types of sample content objects are also different.

[0107] For example, as shown in Table 1, a label value of 1 is used to indicate that the sample content object is a positive sample object, and a label value of 0 is used to indicate that the sample content object is a negative sample object. Among them, the conditions for negative sample objects of the graphic and text type are that PV < 100 or (PV > 1000 and CTR < 0.03), and the conditions for negative sample objects of the video type are that PV < 200 or ((PV > 2000 and (CTR < 0.03 or VTR < 0.1)); the conditions for positive sample objects of the graphic and text type are that PV > 1000 and CTR > 0.1, and the conditions for negative sample objects of the video type are that PV > 2000 and (CTR > 0.1 or VTR > 0.3).

[0108] Table 1

[0109]

[0110] Furthermore, the extraction of sample content evaluation features from the above-mentioned current sample objects can be, but is not limited to, features used to describe the actual content quality corresponding to the current sample objects. Specifically, the above-mentioned attribute features can be, but are not limited to, features used to describe the actual associated information of the current sample objects, such as the type of the current sample object, the corresponding keywords, and the author information, etc.

[0111] The quality features of the above-mentioned current sample objects can be, but are not limited to, including the high-order features of the current sample objects and the high-quality labels of the current sample objects, etc. Among them, the above-mentioned high-order features can be, but are not limited to, used to indicate the content quality evaluation coefficient of the current sample object extracted by using a network model, and the above-mentioned high-quality labels can be, but are not limited to, labels used to indicate the high-quality level of the current sample object marked manually according to preset rules. The overall effects of content objects corresponding to different high-quality levels are also different.

[0112] The semantic features of the above-mentioned current sample objects can be, but are not limited to, features used to describe the meaning attributes or characteristics of the current sample objects, such as features in aspects such as the word meaning, part of speech, grammatical function, and emotional color of the words included in the current sample object. Specifically, the semantic features of the above-mentioned current sample objects can be, but are not limited to, generated by a bidirectional pre-trained language model (Bidirectional Encoder Representations From Transformers, abbreviated as Bert).

[0113] Optionally, in this embodiment, the initialized potential prediction model may be, but is not limited to, a model based on the gradient decision tree algorithm (EXtreme Gradient Boosting, abbreviated as XGBoost). Specifically, XGBoost is an efficient and flexible machine learning model that performs well on large-scale datasets and has achieved excellent results in many machine learning competitions. XGBoost combines gradient boosting trees and regularization techniques, and can effectively handle complex non-linear relationships and a large number of features. It also has good performance in feature selection, feature importance evaluation, missing value processing, etc.

[0114] Optionally, as an alternative implementation, it may be, but is not limited to, based on the following steps as Figure 6 shown to illustrate the above traffic allocation method by way of example:

[0115] Execute steps S602 - S608 to label positive and negative samples for the sample content object based on a predetermined access volume threshold and a predetermined click-through rate threshold, so as to obtain positive sample objects and negative sample objects. Determine the current sample object from the positive sample objects and negative sample objects. Extract sample content evaluation features from the current sample object. Input the content evaluation features into the potential prediction model to obtain the sample potential value of the current sample object becoming a popular content object.

[0116] Then execute step S610 to determine whether the sample potential value is consistent with the sample label corresponding to the current sample object. In the case where the predicted potential value is inconsistent with the sample label corresponding to the current sample object, execute step S604 again, that is, obtain the next sample object as the current sample object. In the case where the sample potential value is consistent with the sample label corresponding to the current sample object, execute step S612 to determine that the potential prediction model has reached the first sub-convergence condition, and accumulate and increment by 1 the number of times the predicted potential value has reached the first sub-convergence condition.

[0117] Then execute step S614 to determine whether the number of times the predicted potential value has reached the first sub-convergence condition is greater than the first convergence threshold, where the first convergence threshold is a positive integer greater than 2. In the case where it is determined that the number of times the predicted potential value has reached the first sub-convergence condition is less than or equal to the first convergence threshold, execute step S604 again, that is, obtain the next sample object as the current sample object. In the case where it is determined that the number of times the predicted potential value has reached the first sub-convergence condition is greater than the first convergence threshold, execute step S616 to determine that the sample potential value has reached the first convergence condition, that is, the potential prediction model training is completed.

[0118] In an embodiment of the present application, a current sample object is determined from the acquired positive sample objects and negative sample objects. Then, a sample content evaluation feature is extracted from the current sample object, wherein the sample content evaluation feature includes a sample attribute feature, a sample quality feature, and a sample semantic feature of the current sample object. Next, the initialized potential prediction model is trained using the sample content evaluation feature until the output sample potential value reaches the first convergence condition. In other words, using an embodiment of the present application, the potential prediction model is trained using the actual content evaluation features of the sample object itself, thereby achieving the technical effect of improving the robustness of the potential prediction model, thereby further improving the accuracy of traffic distribution.

[0119] Optionally, as an optional solution, determining a cold start state threshold that matches the content evaluation feature of the target content object includes:

[0120] In the content recommendation model, the content evaluation feature is used to determine the recommended label category coefficient corresponding to the target content object, wherein the content recommendation model is a decision tree model that is learned using the exposure data counted for the target content object in a cold start trial state;

[0121] The recommended tag category coefficient and exposure factor are used to determine the cold start acceptance state threshold that is suitable for the target content object.

[0122] Optionally, in this embodiment, the content recommendation model may also use XGBoost, but is not limited to it. Further, the content evaluation feature is used in the content recommendation model to determine the recommended label category coefficient corresponding to the target content object, but is not limited to, inputting the content evaluation feature corresponding to the target content object into the content recommendation model, and determining the output result of the content recommendation model as the recommended label category coefficient corresponding to the target content object.

[0123] It should be noted that the above-mentioned recommendation label category coefficient can be, but is not limited to, used to indicate a numerical value, and different numerical values ​​correspond to failed recommendation levels. In other words, there is a mapping relationship between different data and different recommendation levels. The above-mentioned exposure factor can be, but is not limited to, used to indicate the exposure ratio obtained through a large amount of experimental data.

[0124] Furthermore, the above-mentioned use of the recommendation tag category coefficient and the exposure factor to determine the cold start acceptance state threshold adapted to the target content object may include, but is not limited to: determining a recommendation parameter based on the recommendation tag category coefficient. Multiplying the recommendation parameter with the exposure factor to obtain the cold start acceptance state threshold.

[0125] In an embodiment of the present application, the content evaluation features are used in the content recommendation model to determine the recommended label category coefficient corresponding to the target content object, wherein the content recommendation model is a decision tree model that uses the exposure data counted by the target content object in the cold start trial state for learning. Then, the recommended label category coefficient and the exposure factor are used to determine the cold start acceptance state threshold that is compatible with the target content object. The cold start acceptance state threshold is then used to determine the current cold start state of the target content object so as to allocate the target content object with the propagation traffic that is compatible with the cold start state, thereby achieving the purpose of getting rid of the a posteriori parameters in the process of traffic distribution. The problem of poor accuracy of the existing traffic distribution method is solved, and the technical effect of improving the accuracy of traffic distribution is achieved.

[0126] Optionally, as an optional solution, using the recommendation tag category coefficient and the exposure factor to determine the cold start acceptance state threshold adapted to the target content object includes:

[0127] Determine the recommendation parameters based on the recommendation tag category coefficient;

[0128] The recommended parameter is multiplied by the exposure factor to obtain the cold start acceptance state threshold.

[0129] For example, the recommendation parameter determined based on the recommendation tag category coefficient may include but is not limited to: recommendation parameter = 2 l+1 , where l is the above-mentioned recommended label category coefficient. Further, the recommended parameter is multiplied by the exposure factor to obtain the cold start acceptance state threshold, which may include but is not limited to: q = T × 2 l+1 , where q is the cold start acceptance state threshold.

[0130] In the embodiment of the present application, the recommendation parameter is determined based on the recommendation tag category coefficient. Then, the recommendation parameter is multiplied by the exposure factor to obtain the cold start acceptance state threshold. The cold start acceptance state threshold is then used to determine the current cold start state of the target content object, thereby allocating the target content object with a propagation flow that is compatible with the cold start state, thereby achieving the purpose of getting rid of the a posteriori parameters in the process of flow distribution. The problem of poor accuracy in the existing flow distribution method is solved, and the technical effect of improving the accuracy of flow distribution is achieved.

[0131] Optionally, as an optional solution, before obtaining the target content object to be processed from the cold start content pool, the method further includes:

[0132] Obtaining sample exposure data obtained by the sample object in multiple sample collection cycles, wherein the sample exposure data includes the sample exposure flow and the sample estimated click rate of the sample object in each sample collection cycle;

[0133] Based on the changes in the predicted click-through rate of samples in multiple sample collection cycles, obtain the reference sample exposure traffic of the sample object when the predicted click-through rate of the sample reaches the second convergence condition;

[0134] Divide the reference sample exposure traffic to obtain multiple label category coefficients matching the sample object;

[0135] Use multiple label category coefficients and the sample content evaluation features extracted from the sample object to train the initialized content recommendation model until the third convergence condition is reached.

[0136] Optionally, in this embodiment, the above second convergence condition may but is not limited to indicating that the change trend corresponding to the predicted click-through rate of the sample in the above multiple sample collection cycles reaches a stable trend. Specifically, it may but is not limited to performing linear regression processing on the predicted click-through rate of the sample in multiple sample collection cycles to obtain the slope corresponding to the linear regression processing. Furthermore, when the slope corresponding to the linear regression processing is less than or equal to the target slope value, it is determined that the predicted click-through rate of the sample reaches the second convergence condition.

[0137] Furthermore, the above use of multiple label category coefficients and the sample content evaluation features extracted from the sample object to train the initialized content recommendation model until the third convergence condition is reached may but is not limited to including: inputting the sample content evaluation features extracted from the sample object into the content recommendation model to obtain the predicted label category coefficients corresponding to the sample object. When the predicted label category coefficients are consistent with the label category coefficients corresponding to the sample object, and the number of times the predicted label category coefficients are consistent with the label category coefficients corresponding to the sample object reaches the third convergence threshold, it is determined that the content recommendation model has reached the third convergence condition, where the third convergence threshold is a positive integer greater than 2.

[0138] As an alternative implementation, the following steps are used to illustrate the above traffic allocation method by way of example as shown in Figure 7 :

[0139] Execute steps S702 - S708 to obtain the sample exposure traffic {y1, y2,..., y k} and the predicted click-through rate of the sample {x1, x2,..., x k} obtained by the sample object in k sample collection cycles. Perform linear regression processing on the predicted click-through rate of the sample {x1, x2,..., x k} to obtain the slope a. Determine the sample exposure traffic {y1, y2,..., y k} corresponding to when the slope a reaches the target slope value as the reference sample exposure traffic. Divide the reference sample exposure traffic according to quantiles to obtain multiple label category coefficients matching the sample object;

[0140] Next, step S710 is executed to determine the current sample object from the sample objects.

[0141] Then, steps S712 - S714 are executed. The sample content evaluation features extracted from the current sample object are input into the content recommendation model to obtain the predicted label category coefficients corresponding to the current sample object. It is determined whether the label category coefficients corresponding to the current sample object are consistent with the predicted label category coefficients corresponding to the current sample object.

[0142] In the case where the label category coefficients corresponding to the current sample object are not consistent with the predicted label category coefficients corresponding to the current sample object, step S710 is executed again, that is, the next current sample object is obtained. In the case where the label category coefficients corresponding to the current sample object are consistent with the predicted label category coefficients corresponding to the current sample object, step S716 is executed to determine that the content recommendation model has reached the third sub - convergence condition, and the number of times the predicted label category coefficients are consistent with the label category coefficients corresponding to the sample object is incremented by 1. Next, step S718 is executed to determine whether the number of times the predicted label category coefficients are consistent with the label category coefficients corresponding to the sample object is greater than the third convergence threshold.

[0143] In the case where the number of times the predicted label category coefficients are consistent with the label category coefficients corresponding to the sample object is not greater than the third convergence threshold, step S710 is executed again, that is, the next current sample object is obtained. In the case where the number of times the predicted label category coefficients are consistent with the label category coefficients corresponding to the sample object is greater than the third convergence threshold, step S720 is executed to determine that the content recommendation model has reached the third convergence condition.

[0144] In an embodiment of the present application, sample exposure data obtained by a sample object within multiple sample collection cycles is acquired, where the sample exposure data includes the sample exposure traffic and the sample estimated click-through rate of the sample object in each sample collection cycle. Then, based on the change in the sample estimated click-through rate in multiple sample collection cycles, the reference sample exposure traffic of the sample object when the sample estimated click-through rate reaches the second convergence condition is obtained. Next, the reference sample exposure traffic is divided to obtain multiple label category coefficients matching the sample object. Furthermore, the initialized content recommendation model is trained using the multiple label category coefficients and the sample content evaluation features extracted from the sample object until the third convergence condition is reached. In other words, in an embodiment of the present application, the multiple label category coefficients matching the sample object are obtained by using the sample exposure traffic and the sample estimated click-through rate of the sample object in each sample collection cycle. Thus, the initialized content recommendation model is trained using the multiple label category coefficients and the sample content evaluation features extracted from the sample object to obtain a content recommendation model with higher accuracy. Furthermore, the technical effects of improving the accuracy of the content recommendation model and the accuracy of traffic allocation are achieved.

[0145] Optionally, as an alternative solution, allocating propagation traffic adapted to the cold start state to the target content object includes:

[0146] Determine the current timestamp and the publication timestamp of the target content object;

[0147] Obtain the cumulative exposure volume of the target account that publishes the target content object;

[0148] Use the current timestamp, the publication timestamp, the cumulative exposure traffic, and the cumulative exposure of the account to determine the traffic allocation coefficient;

[0149] Multiply the traffic allocation coefficient by the recall sorting result of the target content object to obtain the propagation traffic.

[0150] It should be noted that the recall sorting result of the above target content object can be generated but is not limited to the following steps:

[0151] S1. Obtain multiple recalls after the target content object is published. Specifically, multiple recalls are mainly divided into three categories: content-based, collaborative filtering, and Embedding vector recall. Content-based recall includes recalls of popularity, attributes, and new product strategies; collaborative filtering includes User Based and Item Based; Embedding vector recall includes Word2vec and Bert. The recall link processes a large amount of data and the complexity cannot be too high. The design of multiple recalls can facilitate the addition of new strategies or algorithms. Establishing perfect metrics early and tracking the effects of each recall route helps to select the superior and eliminate the inferior. The effect of recall is not that the more complex the recall algorithm is, the better. Different business characteristics may be suitable for different recalls. As more and more recall algorithms emerge, new recalls need to be different from and complementary to existing recalls to have value. The recall link also undertakes the mission of business and platform construction, such as cold start of users and items, business traffic support, etc. The quality of the recall link directly determines the upper limit of the subsequent links.

[0152] S2. Coarsely rank the target content object. Specifically, simple fusion strategies are often used for coarse ranking. There are many combinations of strategies and the test cycle is long. The application of the two-tower model not only solves the efficiency problem of multiple recall combinations but also avoids the performance problems of fine ranking. The two-tower model has good scalability and is convenient for freely adding custom networks. The User and Item towers are decoupled, and the dot product calculation requires less computing power. In order to ensure the real-time nature of the coarse ranking inference data, the generation of User vectors and the dot product calculation are both real-time.

[0153] S3. Fine rank the target content object to obtain the above recall ranking result (i.e., the recommendation scoring result for the target content object). Specifically, fine ranking is directly responsible for accuracy and relatively easy to obtain direct benefits. The investment in fine ranking is relatively large, and the benefits obtained from it are also relatively large. Among them, fine ranking has gone through three stages: logistic regression and FM of linear models, XGBoost of tree models, and DeepFM of neural network models.

[0154] Furthermore, for example, the above method of determining the traffic distribution coefficient using the current timestamp, publication timestamp, cumulative exposure traffic, and account cumulative exposure may include but is not limited to: The traffic distribution coefficient is inversely related to the time interval since the publication of the target content object, that is, the larger the time interval, the smaller the traffic distribution coefficient, and the less the corresponding allocated propagation traffic. In addition, the traffic distribution coefficient is positively related to the historical cumulative exposure data of the target content object, that is, the more the historical cumulative exposure, the larger the traffic distribution coefficient, and the more the corresponding allocated propagation traffic.

[0155] For example, the above process can refer to the following formula:

[0156] b = (βe-0.15×(24-hour(now()-pt)) +(1-β)e -0.05×real_expose ) 1+cp_pub_num (1)

[0157] Among them, the above-mentioned b is the traffic distribution coefficient, the above-mentioned pt is the post timestamp of the target content object (i.e., the above-mentioned release timestamp), the above-mentioned now() is the current timestamp, the above-mentioned real_expose is the cumulative exposure volume of the content of the target content object (i.e., the above-mentioned cumulative exposure traffic), and the above-mentioned cp_pub_num is the cumulative number of published contents of the publishing account corresponding to the target content object (i.e., the target account) (i.e., the above-mentioned account cumulative exposure volume).

[0158] Furthermore, multiply the traffic distribution coefficient by the recall sorting result of the target content object to obtain the propagation traffic. For example, the calculation formula can be as follows:

[0159] Prob_boost=α×b(2)

[0160] Among them, the above-mentioned Prob_boost is the propagation traffic, the above-mentioned α is the recall sorting result of the target content object, and the above-mentioned b is the traffic distribution coefficient.

[0161] It should be noted that in this embodiment, it is also necessary to perform a diversity scattering process on the content objects in the cold start content pool to avoid the cold start of content objects of the same theme and the same creator being concentrated too quickly. Among them, the above-mentioned diversity scattering process can be but is not limited to using the Shuffle method, or any other diversity scattering method, and this is not limited in this embodiment. In addition, during the cold start process, it is also necessary to adopt a traffic control strategy, that is, to avoid the content objects published by user accounts with low activity and the content objects published by first-time viewing user accounts. In other words, the content objects published by user accounts with low activity and first-time viewing user accounts are not processed by the cold start scheme, so as to avoid the negative effects on sensitive user accounts during the cold start trial stage.

[0162] In the embodiment of the present application, the current timestamp and the post timestamp of the target content object are determined. Then, the account cumulative exposure volume of the target account that publishes the target content object is obtained. Furthermore, the traffic distribution coefficient is determined by using the current timestamp, the post timestamp, the cumulative exposure traffic, and the account cumulative exposure. Then, multiply the traffic distribution coefficient by the recall sorting result of the target content object to obtain the propagation traffic. In other words, in the embodiment of the present application, by introducing the traffic distribution coefficient, the traffic of the target content object is adaptively adjusted, thereby further improving the accuracy of traffic distribution.

[0163] Optionally, as an alternative implementation, such as Figure 8The architecture shown in the figure is used to explain the above-mentioned traffic distribution method as a whole by the following steps:

[0164] The target content object to be processed is obtained from the cold start content pool 802 , wherein the published duration of the content object included in the cold start content pool is less than the target duration threshold.

[0165] Next, in the feature extraction module 804-1 and the labeling module 804-2 of the data preprocessing module 804, the historical consumption log and the refined ranking model log of the target content object provided by the online service of the remote dictionary service storage system 812 are used to obtain the content evaluation features and the recommended label category coefficients of the target content object.

[0166] Furthermore, in the cold start trial module 806, the potential prediction model is used to determine the potential value of the target content object to become a hot content object using the content evaluation features, thereby obtaining the trial boundary threshold configured for the cold start trial state in the cold start state. The cold start trial state threshold adapted to the target content object is determined using the trial boundary threshold adjusted based on the potential value. In the cold start acceptance module 808, the cold start acceptance state threshold adapted to the target content object is determined using the recommended tag category coefficient and the exposure factor.

[0167] Next, the cold start state determination module 810 uses the cumulative exposure flow, cold start trial state threshold, and cold start takeover state threshold in the real-time consumption log of the target content object obtained from the online service of the remote dictionary service storage system 812 to determine the current cold start state of the target content object. Specifically, when the cumulative exposure flow is less than or equal to the cold start trial state threshold, it is determined that the target content object is currently in the cold start trial state, and the target content object is retained in the cold start content pool; when the cumulative exposure flow is greater than the cold start trial state threshold, it is determined that the target content object exits the cold start trial state. When the cumulative exposure flow is greater than the cold start trial state threshold and less than or equal to the cold start takeover state threshold, it is determined that the target content object is currently in the cold start takeover state, and the target content object is retained in the cold start content pool; when the cumulative exposure flow is greater than the cold start takeover state threshold, it is determined that the target content object currently exits the cold start takeover state and enters the flow competition state, and the target content object is deleted from the cold start content pool.

[0168] Further, allocate propagation traffic adapted to the cold start state for the target content object. Specifically, determine the current timestamp and the release timestamp of the target content object. Then, obtain the cumulative exposure volume of the target account that released the target content object. Next, use the current timestamp, the release timestamp, the cumulative exposure traffic, and the cumulative exposure of the account to determine the traffic allocation coefficient. Furthermore, multiply the traffic allocation coefficient by the recall sorting result of the target content object to obtain the propagation traffic.

[0169] In this embodiment, by using the quality characteristics of the target content object itself to determine the cold start state threshold matching it, and then using the cold start state threshold to determine the current cold start state of the target content object, thereby allocating propagation traffic adapted to the cold start state for the target content object, the purpose of getting rid of posterior parameters in the process of traffic allocation is achieved. It solves the problem that the existing traffic allocation method has poor accuracy and realizes the technical effect of improving the accuracy of traffic allocation.

[0170] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that this application is not limited by the described action sequence, because according to this application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0171] According to another aspect of the embodiments of the present application, there is also provided a traffic allocation device for implementing the above traffic allocation method. As Figure 9 shown, the device includes:

[0172] An acquisition unit 902, configured to acquire a target content object to be processed from the cold start content pool, where the published duration of the content objects included in the cold start content pool is less than the target duration threshold;

[0173] A first determination unit 904, configured to determine a cold start state threshold adapted to the content evaluation characteristics of the target content object, and the cumulative exposure traffic counted within a target time period after the target content object is released, where the content evaluation characteristics are used to evaluate the content propagation potential of the target content object itself;

[0174] A second determination unit 906, configured to determine the current cold start state of the target content object according to the result of comparing the cumulative exposure traffic with the cold start state threshold;

[0175] An allocation unit 908, configured to allocate propagation traffic adapted to the cold start state for the target content object.

[0176] Optionally, in this embodiment, the second determination unit includes: a first acquisition module, used to obtain a cold start trial state threshold from the determined cold start state threshold, wherein the cold start trial state threshold is used to indicate an exposure flow threshold for exiting the cold start trial state; a first determination module, used to determine that the target content object is currently in a cold start trial state and retain the target content object in the cold start content pool when the cumulative exposure flow is less than or equal to the cold start trial state threshold; and a second determination module, used to determine that the target content object exits the cold start trial state when the cumulative exposure flow is greater than the cold start trial state threshold.

[0177] Optionally, in this embodiment, the second determination unit further includes: a second acquisition module, used to acquire a cold start takeover state threshold from the determined cold start state threshold, wherein the cold start takeover state threshold is used to indicate an exposure traffic threshold for exiting the cold start takeover state; a third determination module, used to determine that the target content object is currently in a cold start takeover state when the cumulative exposure traffic is greater than the cold start trial state threshold and less than or equal to the cold start takeover state threshold, and retain the target content object in the cold start content pool; a fourth determination module, used to determine that the target content object currently exits the cold start takeover state and enters a traffic competition state when the cumulative exposure traffic is greater than the cold start takeover state threshold, and delete the target content object from the cold start content pool.

[0178] Optionally, in this embodiment, the above-mentioned first determination unit includes: a fifth determination module, which is used to determine the potential value of the target content object to become a popular content object using content evaluation features in a potential prediction model, wherein the potential prediction model is a decision tree model trained using positive sample objects for indicating popular content objects and negative sample objects for indicating non-popular content objects; a third acquisition module, which is used to obtain a trial boundary threshold configured for a cold start trial state in a cold start state; and a sixth determination module, which is used to determine a cold start trial state threshold that is adapted to the target content object using the trial boundary threshold adjusted based on the potential value.

[0179] Optionally, in this embodiment, the above-mentioned sixth determination module is also used to: use the potential value to adjust the first tentative boundary threshold in the tentative boundary threshold to obtain an updated first tentative boundary threshold; compare the second tentative boundary threshold in the tentative boundary threshold with the updated first tentative boundary threshold, wherein the second tentative boundary threshold is less than the first tentative boundary threshold; when the second tentative boundary threshold is greater than the updated first tentative boundary threshold, determine the second tentative boundary threshold as the cold start tentative state threshold; when the updated first tentative boundary threshold is greater than the second tentative boundary threshold, determine the updated first tentative boundary threshold as the cold start tentative state threshold.

[0180] Optionally, in this embodiment, the above-mentioned device also includes: a third determination unit, used to determine the current sample object from the acquired positive sample objects and negative sample objects; an extraction unit, used to extract sample content evaluation features from the current sample object, wherein the sample content evaluation features include sample attribute features, sample quality features, and sample semantic features of the current sample object; a training unit, used to train the initialized potential prediction model using the sample content evaluation features until the output sample potential value reaches the first convergence condition.

[0181] Optionally, in this embodiment, the above-mentioned first determination unit also includes: a seventh determination module, which is used to determine the recommendation label category coefficient corresponding to the target content object using the content evaluation feature in the content recommendation model, wherein the content recommendation model is a decision tree model that is learned using the exposure data counted by the target content object in the cold start trial state; an eighth determination module, which is used to determine the cold start acceptance state threshold that is compatible with the target content object using the recommendation label category coefficient and the exposure factor.

[0182] Optionally, in this embodiment, the eighth determination module is used to: determine a recommended parameter based on a recommended tag category coefficient; and multiply the recommended parameter by an exposure factor to obtain a cold start acceptance state threshold.

[0183] Optionally, in this embodiment, the above-mentioned device also includes: a first acquisition unit, used to obtain sample exposure data obtained by the sample object in multiple sample collection cycles, wherein the sample exposure data includes the sample exposure flow and the sample estimated click-through rate of the sample object in each sample collection cycle; a second acquisition unit, used to obtain the reference sample exposure flow of the sample object when the sample estimated click-through rate reaches the second convergence condition based on the change of the sample estimated click-through rate in multiple sample collection cycles; a division unit, used to divide the reference sample exposure flow to obtain multiple label category coefficients matching the sample object; a first training unit, used to train the initialized content recommendation model using multiple label category coefficients and sample content evaluation features extracted from the sample object until the third convergence condition is reached.

[0184] Optionally, in this embodiment, the above-mentioned allocation unit includes: a ninth determination module, used to determine the current timestamp and the release timestamp of the target content object; a fourth acquisition module, used to obtain the account cumulative exposure of the target account that publishes the target content object; a tenth determination module, used to use the current timestamp, release timestamp, cumulative exposure traffic and account cumulative exposure to determine the traffic allocation coefficient; a multiplication module, used to multiply the traffic allocation coefficient with the recall ranking result of the target content object to obtain the propagation traffic.

[0185] For specific embodiments, reference may be made to the examples shown in the above traffic allocation method, and details thereof will not be elaborated herein.

[0186] According to another aspect of the embodiments of the present application, there is also provided an electronic device for implementing the above traffic allocation method. The electronic device may be Figure 1 the terminal device or server shown. In this embodiment, the electronic device is taken as an example of the terminal device for illustration. As Figure 10 shown, the electronic device includes a memory 1002 and a processor 1004. A computer program is stored in the memory 1002, and the processor 1004 is configured to execute the steps in any of the above method embodiments through the computer program.

[0187] Optionally, in this embodiment, the above electronic device may be at least one of multiple network devices in a computer network.

[0188] Optionally, in this embodiment, the above processor may be configured to execute the following steps through a computer program:

[0189] S1. Obtain a target content object to be processed from a cold start content pool, where the published duration of the content objects included in the cold start content pool is less than a target duration threshold;

[0190] S2. Determine a cold start state threshold adapted to the content evaluation characteristics of the target content object, and the cumulative exposure traffic counted within a target time period after the target content object is published, where the content evaluation characteristics are used to evaluate the content dissemination potential of the target content object itself;

[0191] S3. Determine the current cold start state of the target content object according to the result of comparing the cumulative exposure traffic with the cold start state threshold;

[0192] S4. Allocate a dissemination traffic adapted to the cold start state for the target content object.

[0193] Optionally, those of ordinary skill in the art can understand that Figure 10 the structure shown is only schematic, and the electronic device may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a handheld computer, and a mobile Internet device (MID), a PAD, and other terminal devices. Figure 10 It does not limit the structure of the above electronic device. For example, the electronic device may further include more or fewer components (such as a network interface, etc.) than those shown Figure 10 , or have a different configuration from that shown Figure 10 .

[0194] Among them, the memory 1002 can be used to store software programs and modules, such as the program instructions / modules corresponding to the traffic allocation method and device in the embodiments of the present application. The processor 1004 executes various functional applications and data processing by running the software programs and modules stored in the memory 1002, that is, implements the above-mentioned traffic allocation method. The memory 1002 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 1002 may further include a memory remotely disposed relative to the processor 1004, and these remote memories can be connected to the terminal through a network. Examples of the above network include but are not limited to the Internet, enterprise intranet, local area network, mobile communication network, and combinations thereof. As an example, as Figure 10 shown, the above memory 1002 may but is not limited to include the acquisition unit 902, the first determination unit 904, the second determination unit 906, and the allocation unit 908 in the above traffic allocation device. In addition, it may also include but is not limited to other module units in the above traffic allocation device, which will not be elaborated in this example.

[0195] Optionally, the above transmission device 1006 is used to receive or send data via a network. Specific examples of the above network may include a wired network and a wireless network. In one instance, the transmission device 1006 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices and routers through a network cable, thereby enabling communication with the Internet or a local area network. In one instance, the transmission device 1006 is a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0196] In addition, the above electronic device further includes: a connection bus 1008, which is used to connect each module component in the above electronic device.

[0197] In other embodiments, the above terminal device or server may be a node in a distributed system. Among them, the distributed system may be a blockchain system, and the blockchain system may be a distributed system formed by connecting the multiple nodes through network communication. Among them, the nodes can form a point-to-point network, and any form of computing device, such as a server, a terminal, and other electronic devices, can become a node in the blockchain system by joining the point-to-point network.

[0198] According to one aspect of the present application, a computer program product is provided. The computer program product includes a computer program / instructions, and the computer program / instructions contain program code for executing the above method. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part, and / or installed from a removable medium. When the computer program is executed by a central processing unit, various functions provided by the embodiments of the present application are executed.

[0199] According to one aspect of the present application, a computer-readable storage medium is provided. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above method.

[0200] Optionally, in this embodiment, the above computer-readable storage medium may be configured to store a computer program for executing the following steps:

[0201] S1. Obtain a target content object to be processed from a cold start content pool, where the published duration of the content objects included in the cold start content pool is less than a target duration threshold;

[0202] S2. Determine a cold start state threshold adapted to the content evaluation characteristics of the target content object, and the cumulative exposure traffic counted within a target time period after the target content object is published, where the content evaluation characteristics are used to evaluate the content dissemination potential of the target content object itself;

[0203] S3. Determine the current cold start state of the target content object according to the result of comparing the cumulative exposure traffic with the cold start state threshold;

[0204] S4. Allocate dissemination traffic adapted to the cold start state to the target content object.

[0205] Optionally, in the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program with a predetermined function, which works together with other related parts to achieve a predetermined goal, and can be fully or partially implemented by using software, hardware (such as a processing circuit or a memory), or a combination thereof. Similarly, one 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 a part of an overall module or unit including the functions of the module or unit.

[0206] Optionally, in this embodiment, those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the relevant hardware of the terminal device through a program, and this program can be stored in a computer-readable storage medium. The storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, an optical disk, etc.

[0207] If the integrated unit in the above embodiment is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in the above computer-readable storage medium. Based on such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing one or more computer devices (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0208] In the above embodiments of this application, the descriptions of the various embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0209] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. Among them, the device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, 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 displayed or discussed mutual coupling or direct coupling or communication connection can be through some interfaces, and the indirect coupling or communication connection of units or modules can be in an electrical or other form.

[0210] The unit described as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it can be located in one place, or it can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0211] In addition, the functional units in the various embodiments of this application can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0212] The above are only the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A flow allocation method, characterized in that, include: Acquire a target content object to be processed from a cold start content pool, wherein the published duration of the content object contained in the cold start content pool is less than a target duration threshold; Determine a cold start state threshold adapted to the content evaluation feature of the target content object, and the cumulative exposure traffic counted for the target content object within a target time period after being released, wherein the content evaluation feature is used to evaluate the content dissemination potential of the target content object itself; Determining the current cold start state of the target content object according to a result of comparing the cumulative exposure flow with the cold start state threshold; A propagation flow rate adapted to the cold start state is allocated to the target content object.

2. The method according to claim 1, wherein The step of determining the current cold start state of the target content object according to a result of comparing the cumulative exposure flow with the cold start state threshold comprises: Acquire a cold start trial state threshold from the determined cold start state threshold, wherein the cold start trial state threshold is used to indicate an exposure flow threshold for exiting the cold start trial state; When the accumulated exposure flow is less than or equal to the cold start trial state threshold, determining that the target content object is currently in the cold start trial state, and retaining the target content object in the cold start content pool; When the cumulative exposure flow is greater than the cold start trial state threshold, it is determined that the target content object exits the cold start trial state.

3. The method according to claim 2, wherein The step of determining the current cold start state of the target content object according to a result of comparing the cumulative exposure flow with the cold start state threshold further includes: Acquire a cold start takeover state threshold from the determined cold start state threshold, wherein the cold start takeover state threshold is used to indicate an exposure flow threshold for exiting the cold start takeover state; When the accumulated exposure flow is greater than the cold start trial state threshold and less than or equal to the cold start acceptance state threshold, determining that the target content object is currently in the cold start acceptance state, and retaining the target content object in the cold start content pool; When the accumulated exposure traffic is greater than the cold start acceptance state threshold, it is determined that the target content object currently exits the cold start acceptance state and enters a traffic competition state, and the target content object is deleted from the cold start content pool.

4. The method according to claim 2, wherein The step of determining a cold start state threshold value that matches the content evaluation feature of the target content object includes: Determining the potential value of the target content object to become a popular content object by using the content evaluation feature in a potential prediction model, wherein the potential prediction model is a decision tree model trained by using positive sample objects indicating popular content objects and negative sample objects indicating non-popular content objects; Obtaining a probe boundary threshold configured for a cold start probe state in the cold start state; The cold start trial state threshold adapted to the target content object is determined by using the trial boundary threshold adjusted based on the potential value.

5. The method according to claim 4, wherein The step of determining a cold start trial state threshold adapted to the target content object by using the trial boundary threshold adjusted based on the potential value includes: Using the potential value to adjust a first tentative boundary threshold among the tentative boundary thresholds to obtain an updated first tentative boundary threshold; Comparing a second tentative boundary threshold among the tentative boundary thresholds with the updated first tentative boundary threshold, wherein the second tentative boundary threshold is less than the first tentative boundary threshold; In a case where the second probing boundary threshold is greater than the updated first probing boundary threshold, determining the second probing boundary threshold as the cold start probing state threshold; In a case where the updated first probing boundary threshold is greater than the second probing boundary threshold, the updated first probing boundary threshold is determined as the cold start probing state threshold.

6. The method according to claim 4, wherein Before acquiring the target content object to be processed from the cold start content pool, the method further includes: Determining a current sample object from the acquired positive sample objects and the acquired negative sample objects; Extracting sample content evaluation features from the current sample object, wherein the sample content evaluation features include sample attribute features, sample quality features, and sample semantic features of the current sample object; The initialized potential prediction model is trained using the sample content evaluation features until the output sample potential value reaches a first convergence condition.

7. The method according to claim 3, wherein The step of determining a cold start state threshold value that matches the content evaluation feature of the target content object includes: Determining a recommendation label category coefficient corresponding to the target content object by using the content evaluation feature in a content recommendation model, wherein the content recommendation model is a decision tree model learned by using exposure data counted for the target content object in the cold start trial state; The cold start acceptance state threshold value adapted to the target content object is determined by using the recommendation tag category coefficient and the exposure factor.

8. The method according to claim 7, characterized in that, The step of using the recommendation tag category coefficient and the exposure factor to determine the cold start acceptance state threshold value adapted to the target content object includes: Determining a recommendation parameter based on the recommendation tag category coefficient; The recommended parameter is multiplied by the exposure factor to obtain the cold start acceptance state threshold.

9. The method according to claim 7, wherein Before acquiring the target content object to be processed from the cold start content pool, the method further includes: Acquire sample exposure data obtained by the sample object in multiple sample collection cycles, wherein the sample exposure data includes the sample exposure flow and the sample estimated click rate of the sample object in each of the sample collection cycles; Based on the change of the estimated click-through rate of the sample in the multiple sample collection cycles, obtaining the reference sample exposure flow of the sample object when the estimated click-through rate of the sample reaches a second convergence condition; Dividing the reference sample exposure flow to obtain a plurality of label category coefficients matching the sample object; Training the initialized content recommendation model using the multiple tag category coefficients and the sample content evaluation features extracted from the sample objects until the third convergence condition is reached.

10. The method according to any one of claims 1 to 9, characterized in that, Assigning the propagation traffic adapted to the cold start state to the target content object includes: Determining the current timestamp and the publication timestamp of the target content object; Obtaining the cumulative exposure volume of the target account that published the target content object; Determining a traffic allocation coefficient using the current timestamp, the publication timestamp, the cumulative exposure traffic, and the cumulative exposure of the account; Multiplying the traffic allocation coefficient by the recall ranking result of the target content object to obtain the propagation traffic.

11. A flow distribution device, characterized in that, Including: An obtaining unit, configured to obtain a target content object to be processed from a cold start content pool, where the published duration of the content objects included in the cold start content pool is less than a target duration threshold; A first determination unit, configured to determine a cold start state threshold adapted to the content evaluation features of the target content object, and the cumulative exposure traffic counted within a target time period after the target content object is published, where the content evaluation features are used to evaluate the content propagation potential of the target content object itself; A second determination unit, configured to determine the current cold start state of the target content object according to the result of comparing the cumulative exposure traffic with the cold start state threshold; An allocation unit, configured to assign propagation traffic adapted to the cold start state to the target content object.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, where the program, when run by a processor, executes the method described in any one of claims 1 to 10.

13. A computer program product, comprising a computer program / instructions, characterized in that, The computer program / instructions, when executed by a processor, implement the steps of the method described in any one of claims 1 to 10.

14. An electronic device, comprising a memory and a processor, characterized in that, A computer program is stored in the memory, and the processor is configured to execute the method described in any one of claims 1 to 10 through the computer program.

Images