Media information processing method and device, storage medium and electronic equipment

Abstract

The invention discloses a media information processing method and device, a storage medium and electronic equipment. The method comprises the steps that a media information issuing request is obtained, in response to the media information issuing request, quality parameters corresponding to multiple pieces of media information to be recommended are determined, each quality parameter is determined by a target parameter model after multiple times of iteration, and the target parameter model is used for restraining the decreasing amplitude of a value parameter to meet a preset amplitude condition, in the multi-round iteration process, the target parameter model adaptively determines the number of first combined parameters of the current round of sampling and the number of second combined parameters of which the reward value of the current round meets a preset condition according to the calculation time consumption of the previous round of sampling; and determining target media information to be published at the target publishing position from the multiple pieces of media information according to the quality parameters. According to the method and the device, the technical problem that the publishing efficiency of the media information is relatively low because the quality parameters are not accurate enough is solved.

Description

Technical Field

[0001] The present application relates to the field of computers, and in particular to a method and device for processing media information, a storage medium, and an electronic device. Background Art

[0002] At present, in the media information publishing scenario, media information is generally described using quality parameters determined by multiple value parameters. For example, the click score given by the click-through rate prediction model represents the probability of the media information being clicked; the reading time score given by the reading time prediction model represents the probability of the media information being read for a long time; the timeliness score calculated by the media information mass sending time, exposure time and media information category represents the timeliness of the media information. The final ranking score of the media information is obtained by combining these different value scores through a specific fusion formula.

[0003] The combination parameters of the value formula rely on manual selection (or manually designed rules), and the algorithm has several hyperparameters and reward values, which also require manual experience to determine. Inappropriate hyperparameters and reward values ​​will cause the algorithm to become less effective as the iterations go on, which in turn will lead to inaccurate calculation of the quality parameters of media information, resulting in low efficiency in publishing media information.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present application provide a method and device for processing media information, a storage medium, and an electronic device, so as to at least solve the technical problem that the publishing efficiency of media information is low due to inaccurate quality parameters.

[0006] According to one aspect of an embodiment of the present application, a method for processing media information is provided, comprising: obtaining a media information publishing request, wherein the media information publishing request is used to request publishing of the media information at a target publishing location; in response to the media information publishing request, determining quality parameters corresponding to a plurality of media information to be recommended, wherein each of the quality parameters is determined by at least two value parameters of the corresponding media information based on a target combination parameter, wherein the target combination parameter is a combination parameter to which a target parameter model converges after multiple rounds of iterations, wherein the target parameter model is based on a training objective that a decrease in a constraint value parameter satisfies a preset amplitude condition and that the target value parameter is maximized. The model comprises at least two value parameters including the target value parameter and the constraint value parameter. The target parameter model adaptively determines the first combination parameter quantity of the current round of sampling and the second combination parameter quantity of the current round of reward value that meet the preset conditions according to the calculation time of the previous round of sampling during multiple rounds of iterations. The hyperparameters of the target parameter model include the first combination parameter quantity and the second combination parameter quantity. The reward value is adaptively determined by the change range of the target value parameter from the previous round to the current round during the multiple rounds of iterations. The target media information to be published on the target publishing position is determined from the multiple media information according to the quality parameter.

[0007] According to another aspect of an embodiment of the present application, a media information processing device is also provided, including: an acquisition module, used to obtain a media information publishing request, wherein the media information publishing request is used to request the publishing of media information at a target publishing position; a first determination module, used to determine, in response to the media information publishing request, quality parameters corresponding to a plurality of media information to be recommended, wherein each of the quality parameters is determined by at least two value parameters of the corresponding media information based on a target combination parameter, the target combination parameter is a combination parameter converged to after multiple rounds of iterations of a target parameter model, the target parameter model is based on a constraint that the decrease in the value parameter satisfies a preset amplitude condition, and the target value parameter is maximized as a training A model of a training target, wherein the at least two value parameters include the target value parameter and the constraint value parameter, and the target parameter model adaptively determines the first combination parameter quantity of the current round of sampling and the second combination parameter quantity of the current round of reward value that meet the preset conditions according to the calculation time of the previous round of sampling during multiple rounds of iterations, and the hyperparameters of the target parameter model include the first combination parameter quantity and the second combination parameter quantity, and the reward value is adaptively determined by the change range of the target value parameter from the previous round to the current round during the multiple rounds of iterations; a second determination module is used to determine the target media information to be published on the target publishing position from the multiple media information according to the quality parameter.

[0008] In an exemplary embodiment, the device is used to respond to the media information publishing request and determine the quality parameters corresponding to the multiple media information to be recommended in the following manner: in response to the media information publishing request, obtain the at least two value parameters corresponding to each of the multiple media information, wherein the multiple media information includes media information in different forms; perform the multiple rounds of iterations according to the target parameter model to determine the target combination parameter, wherein during the multiple rounds of iterations, the target parameter model is a model that takes maximizing the value parameters in all forms as the training target on the premise that the decrease amplitude of the value parameters in each single form meets the preset amplitude condition; fuse the at least two value parameters corresponding to the multiple media information according to the target combination parameter to determine a quality parameter set, wherein the quality parameter set is used to perform mixed and uniform sorting of the multiple media information in different forms to determine the target media information.

[0009] In an exemplary embodiment, the device is used to perform the multiple rounds of iterations according to the target parameter model in the following manner to determine the target combination parameters: obtaining a total consumption value parameter, a first form consumption value parameter, a second form consumption value parameter, a total duration value parameter, a first form time value parameter and a second form time value parameter corresponding to the multiple media information, wherein the target value parameter includes the total consumption value parameter and the total duration value parameter, the constraint value parameter includes the first form consumption value parameter, the second form consumption value parameter, the first form time value parameter and the second form time value parameter, the first form consumption value parameter and the second form consumption value parameter are consumption value parameters corresponding to media information of different forms, and the first form time value parameter and the second form time value parameter are time value parameters corresponding to media information of different forms; the target combination parameters are determined using the target parameter model with the first form consumption value parameter and the second form consumption value parameter not decreasing during the multiple rounds of iterations, the first form time value parameter and the second form time value parameter decreasing by no more than a preset threshold during the multiple rounds of iterations, and the total consumption value parameter and the total duration value parameter being maximized as training objectives.

[0010] In an exemplary embodiment, the device is also used to: adaptively update the reward value weight parameter corresponding to each round during multiple rounds of iterations in the following manner, wherein the reward value weight parameter is used to determine the reward value: obtain a first change amplitude of the total consumption value parameter during the iteration from the previous round to the current round and a second change amplitude of the total duration value parameter during the iteration from the previous round to the current round; normalize the first change amplitude and the second change amplitude to determine the reward value weight parameter corresponding to each round.

[0011] In an exemplary embodiment, the device is used to perform the multiple rounds of iterations according to the target parameter model to determine the target combination parameters in the following manner: obtaining an initial mean vector and an initial variance vector, wherein the initial mean vector is used to represent a set of pre-set initial combination parameters, and the initial variance vector is used to represent the difference between the maximum value and the minimum value in the set of initial combination parameters; performing multiple rounds of iterations to determine the target combination parameters in the following manner, wherein the samples sampled in each round are regarded as sample mean vectors, and the sample variance vector corresponds to the sample mean vector in each round, the sample mean vector in the first round of sampling is the initial mean vector, the combination parameters obtained in the current round of sampling are regarded as N groups of combination parameters, and the combination parameters obtained in the next round of sampling are regarded as M groups Combination parameters, N is a positive integer, M is a positive integer less than or equal to N: the N groups of combination parameters are generated according to the sample mean vector and the sample variance vector, and N reward values ​​are determined according to whether the decrease amplitude of the constraint value parameter corresponding to the N groups of combination parameters meets the preset amplitude condition; the reward values ​​in the first M descending order are determined from the N reward values; the M groups of combination parameters are determined according to the first M reward values, and the intermediate mean vector and the intermediate variance vector corresponding to the M groups of combination parameters are determined, wherein the intermediate mean vector is regarded as the sample mean vector in the next round of iteration, and the intermediate variance vector is regarded as the sample variance vector in the next round of iteration; the M groups of combination parameters that meet the preset convergence condition are determined as the target combination parameters.

[0012] In an exemplary embodiment, the apparatus is further used to: perform a perturbation operation on the intermediate variance vector to obtain a perturbation variance vector; and generate the M groups of combined parameters based on the intermediate mean vector regarded as the sample mean vector and the perturbation variance vector.

[0013] In an exemplary embodiment, the device is used to determine N reward values ​​according to whether the decrease range of the constraint value parameter corresponding to the N groups of combination parameters meets the preset range condition in the following manner: a piecewise function is set according to whether the decrease range of the constraint value parameter corresponding to the N groups of combination parameters meets the preset range condition to determine the N reward values, wherein the piecewise function is used to determine the N reward values ​​respectively in the following manner: when the first form consumption value parameter and the second form consumption value parameter corresponding to each group of combination parameters do not meet the corresponding constraint condition, and the first form time value parameter and the second form time value parameter meet the corresponding constraint condition, the reward value is set to be based on the The weighted difference between the total consumption value parameter and the total duration value parameter determined by the reward value weight parameter; when the constraint value parameters corresponding to each set of combination parameters meet the corresponding constraint conditions, the reward value is set to the weighted sum of the total consumption value parameter and the total duration value parameter determined based on the reward value weight parameter; when the first form consumption value parameter and the second form consumption value parameter corresponding to each set of combination parameters meet the corresponding constraint conditions, and / or the first form time value parameter and the second form time value parameter do not meet the corresponding constraint conditions, the reward value is set to the weighted difference between the negative value of the total consumption value parameter based on the reward value weight parameter and the total duration value parameter.

[0014] In an exemplary embodiment, the device is also used to: adaptively determine the first combination parameter number and the second combination parameter number corresponding to each round during multiple rounds of iterations in the following manner: obtain an initial combination parameter number, wherein the initial combination parameter number is a preset combination parameter number to be searched; determine the evaluation group and parameter number that meet the constraint conditions based on the total consumption value parameter, the first form consumption value parameter, the second form consumption value parameter, the first form time value parameter, and the second form time value parameter not decreasing; determine the ratio between the evaluation combination parameter number and the preset combination parameter number as the ratio between the first combination parameter number and the second combination parameter number corresponding to each round; adaptively set the first combination parameter number corresponding to the next round according to the calculation time of each round, and determine the second combination parameter number according to the ratio.

[0015] In an exemplary embodiment, the device is used to obtain the first form timeliness value parameters and the second form timeliness value parameters corresponding to multiple media information in the media information set in the following manner: obtaining a first timeliness value sub-parameter and a second timeliness value sub-parameter, wherein the first timeliness value sub-parameter and the second timeliness value sub-parameter represent exposure ratio parameters corresponding to the first media information at different release times; obtaining a third timeliness value sub-parameter and a fourth timeliness value sub-parameter, wherein the second form timeliness value parameter includes the third timeliness value sub-parameter and the fourth timeliness value sub-parameter, the third timeliness value sub-parameter and the fourth timeliness value sub-parameter represent exposure ratio parameters corresponding to the second media information at different release times, and the first media information and the second media information are media information of different forms; determining the first form timeliness value parameter according to the first timeliness value sub-parameter and the second timeliness value sub-parameter, and determining the second form timeliness value parameter according to the third timeliness value sub-parameter and the fourth timeliness value sub-parameter.

[0016] 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 method for processing media information when running.

[0017] According to another aspect of the embodiment of the present application, a computer program product or a computer program is provided, the computer program product or the computer program includes computer instructions, 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 above media information processing method.

[0018] According to another aspect of the embodiments of the present application, there is further provided an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the above-mentioned method for processing media information through the computer program.

[0019] In an embodiment of the present application, a media information publishing request is obtained, wherein the media information publishing request is used to request the publishing of media information at a target publishing position, and in response to the media information publishing request, quality parameters corresponding to the plurality of media information to be recommended are determined, wherein each quality parameter is determined by at least two value parameters of the corresponding media information based on a target combination parameter, and the target combination parameter is a combination parameter to which a target parameter model converges after multiple rounds of iterations, and the target parameter model is a model in which the decrease amplitude of the constraint value parameter satisfies a preset amplitude condition and the target value parameter is maximized as a training target, and the at least two value parameters include a target value parameter and a constraint value parameter, and the target parameter model adaptively determines the first group of samples of the current round according to the computational time of the previous round of sampling during multiple rounds of iterations. The number of combined parameters and the second number of combined parameters of the current round of reward values ​​satisfy the preset conditions, the hyperparameters of the target parameter model include the first number of combined parameters and the second number of combined parameters, the reward value is adaptively determined by the change range of the target value parameter from the previous round to the current round in the process of multiple rounds of iterations, and the method of determining the target media information to be published at the target publishing position from multiple media information according to the quality parameters, through adaptive hyperparameter selection and adaptive reward value setting, the optimization of each value parameter is more balanced and approaches the Pareto optimality, thereby realizing the technical effect of optimizing the determination method of quality parameters, improving the calculation accuracy of quality parameters, and improving the publishing efficiency of media information, thereby solving the technical problem of low publishing efficiency of media information due to inaccurate quality parameters.

[0020] In addition, the present application proposes to use some value parameters as constraint indicators, that is, instead of maximizing a single value parameter as the optimization goal, the training goal is to maximize other value parameters without decreasing the value parameter, which is more suitable for publishing scenarios involving media information in different forms. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] 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 on the present application. In the drawings:

[0022] Figure 1 is a schematic diagram of an application environment of an optional method for processing media information according to an embodiment of the present application;

[0023] Figure 2 is a flowchart of an optional method for processing media information according to an embodiment of the present application;

[0024] Figure 3 is a schematic diagram of an optional method for processing media information according to an embodiment of the present application;

[0025] Figure 4 is a schematic diagram of another optional method for processing media information according to an embodiment of the present application;

[0026] Figure 5 is a schematic diagram of another optional method for processing media information according to an embodiment of the present application;

[0027] Figure 6 is a schematic diagram of another optional method for processing media information according to an embodiment of the present application;

[0028] Figure 7 is a schematic diagram of another optional method for processing media information according to an embodiment of the present application;

[0029] Figure 8 is a schematic diagram of another optional method for processing media information according to an embodiment of the present application;

[0030] Fig. 9 is a schematic diagram of another optional method for processing media information according to an embodiment of the present application;

[0031] Fig.10 is a schematic diagram of another optional method for processing media information according to an embodiment of the present application;

[0032] Fig.11 is a schematic diagram of another optional method for processing media information according to an embodiment of the present application;

[0033] Fig.12 is a schematic structural diagram of an optional media information processing device according to an embodiment of the present application;

[0034] Fig.13 is a schematic diagram of the structure of an optional media information publishing product according to an embodiment of the present application;

[0035] Fig.14 It is a schematic diagram of the structure of an optional electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0036] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0037] 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 the data used in this way can be interchangeable where appropriate, 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 "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0038] First, some nouns or terms that appear in the description of the embodiments of the present application are subject to the following interpretation:

[0039] Session: A session is the longest list of messages that are exposed after a user enters the subscription account box once. For example, if items ranked from the first to the fifth are exposed to the user, and then the user exits the box, and the item ranked sixth is not exposed, the length of the session is five.

[0040] AB experiment: a controlled variable experiment that uses real online traffic to evaluate the quality of the model algorithm.

[0041] CTR: Click Through Rate, click-through rate.

[0042] AUC: Area under the ROC curve, an indicator used to evaluate the performance of the click-through rate prediction model. The higher the value, the better.

[0043] GAUC: An improved version of the AUC indicator, which is obtained by weighted summing the AUC of the session by the number of samples in the session.

[0044] The present application is described below in conjunction with embodiments:

[0045] According to one aspect of an embodiment of the present application, a method for processing media information is provided. Optionally, in this embodiment, the method for processing media information can be applied to: Figure 1 In the hardware environment composed of the server 101 and the terminal device 103 shown in FIG. Figure 1As shown, the server 101 is connected to the terminal 103 via a network, and can be used to provide services for the terminal device or the application installed on the terminal device. The application can be a video application, an instant messaging application, a browser application, an educational application, a game application, etc. A database 105 may be set up on the server or independently of the server to provide data storage services for the server 101, for example, a game data storage server. The above-mentioned network may include, but is not limited to, a wired network and a wireless network, wherein the wired network includes, a local area network, a metropolitan area network and a wide area network; the wireless network includes, Bluetooth, WIFI and other networks that implement wireless communication; the terminal device 103 may be a terminal configured with an application, and may include, but is not limited to, at least one of the following: a mobile phone (such as an Android phone, an iOS phone, etc.), a laptop computer, a tablet computer, a PDA, a MID (Mobile Internet Devices), a PAD, a desktop computer, a smart TV, an intelligent voice interaction device, a smart home appliance, a vehicle-mounted terminal, an aircraft, a virtual reality (VR) terminal, an augmented reality (AR) terminal, a mixed reality (MR) terminal and other computer devices; the above-mentioned server may be a single server, a server cluster consisting of multiple servers, or a cloud server.

[0046] Combination Figure 1 As shown, the above-mentioned method for processing media information can be implemented in the terminal device 103 through the following steps:

[0047] S1, obtaining a media information publishing request on the terminal device 103, wherein the media information publishing request is used to request publishing of media information at a target publishing location;

[0048] S2, in response to the media information publishing request on the terminal device 103, determine the quality parameters corresponding to the multiple media information to be recommended, wherein each quality parameter is determined by at least two value parameters of the corresponding media information based on the target combination parameter, the target combination parameter is the combination parameter converged to by the target parameter model after multiple rounds of iterations, the target parameter model is a model that takes the decrease amplitude of the constraint value parameter to meet the preset amplitude condition and the maximization of the target value parameter as the training target, the at least two value parameters include the target value parameter and the constraint value parameter, the target parameter model adaptively determines the first combination parameter number of the current round of sampling and the second combination parameter number of the current round of reward value that meets the preset condition according to the calculation time of the previous round of sampling during the multiple rounds of iterations, the hyperparameters of the target parameter model include the first combination parameter number and the second combination parameter number, and the reward value is adaptively determined by the change amplitude of the target value parameter from the previous round to the current round during the multiple rounds of iterations;

[0049] S3, determining target media information to be published at a target publishing position from multiple media information according to the quality parameter on the terminal device 103.

[0050] Optionally, in this embodiment, the above-mentioned media information processing method can also be implemented by a server, for example, Figure 1 It is implemented in the server 101 shown; or it is implemented by the terminal device and the server together.

[0051] The above is only an example and is not specifically limited in this embodiment.

[0052] Optionally, as an optional implementation, as Figure 2 As shown, the above-mentioned method for processing media information includes:

[0053] S202, obtaining a media information publishing request, wherein the media information publishing request is used to request publishing of media information at a target publishing location;

[0054] Optionally, in this embodiment, application scenarios of the above-mentioned media information processing method may include but are not limited to medical, education, gaming, social and other application scenarios.

[0055] In an exemplary embodiment, the above-mentioned media information processing method may include but is not limited to being applied to scenarios where a media information push platform pushes media information, such as a public account pushing media information, a web page pushing media information, etc.

[0056] Optionally, in this embodiment, the target publishing location may be, but is not limited to, a regional location for publishing media information online or offline, wherein the regional location may be, but is not limited to, a fixed location, such as a website homepage, an elevator entrance, etc. The regional location may also be, but is not limited to, a location flexibly preset according to preset conditions, such as displaying the regional location when the user browses the web page for a preset time.

[0057] For example, Figure 3 is a schematic diagram of an optional method for processing media information according to an embodiment of the present invention, such as Figure 3 As shown, the above-mentioned method of obtaining a media information publishing request may include but is not limited to a target publishing location 302, a group of publishing locations 304 and a terminal device display interface 306, a group of publishing locations 304 being displayed on the terminal device display interface 306, and the above-mentioned media information is published at a target publishing location 302 in the group of publishing locations 304.

[0058] Optionally, in this embodiment, the above-mentioned media information release request may include but is not limited to a media information release request generated by the system or the user according to preset conditions, for example, a media information release request generated by a user clicking a touch button of an application interface, a user refreshing a display interface, a user starting an application, etc.

[0059] S204, in response to the media information publishing request, determining the quality parameters respectively corresponding to the plurality of media information to be recommended, wherein each quality parameter is determined by at least two value parameters of the corresponding media information based on a target combination parameter, the target combination parameter is a combination parameter to which the target parameter model converges after multiple rounds of iterations, the target parameter model is a model that takes the decrease amplitude of the constraint value parameter to meet the preset amplitude condition and the maximization of the target value parameter as the training target, the at least two value parameters include the target value parameter and the constraint value parameter, the target parameter model adaptively determines the first combination parameter number of the current round of sampling and the second combination parameter number of the current round of reward value that meets the preset condition according to the calculation time of the previous round of sampling during the multiple rounds of iterations, the hyperparameters of the target parameter model include the first combination parameter number and the second combination parameter number, and the reward value is adaptively determined by the change amplitude of the target value parameter from the previous round to the current round during the multiple rounds of iterations;

[0060] Optionally, in this embodiment, the above-mentioned response to the media information publishing request can be understood as receiving the media information publishing request or obtaining the task that requires publishing media information, etc., that is, the above-mentioned media information publishing request can come from the user's active triggering through interactive operations, or it can come from the media information publishing task pre-configured by the media information publishing platform.

[0061] In an exemplary embodiment, Figure 4 is a schematic diagram of another optional method for processing media information according to an embodiment of the present invention. Figure 4 As shown, the above-mentioned method for processing media information can be applied on a live broadcast platform, wherein the live broadcast platform can include multiple types such as a video live broadcast room 402, a text live broadcast room 404, an audio live broadcast room 406, and a graphic live broadcast room. One type of live broadcast room corresponds to one media information, and by calculating the quality parameters corresponding to multiple live broadcast rooms, several live broadcast rooms with the highest ranking are selected and displayed on the homepage of the live broadcast platform, so as to push live broadcast rooms with higher value to users.

[0062] In another exemplary embodiment, Figure 5 is a schematic diagram of another optional method for processing media information according to an embodiment of the present invention. Figure 5 As shown, the above-mentioned method for processing media information can be applied in the official account mini-program, wherein the resources allowed to be pushed by the official account may include multiple media information such as media information 502, media information 504 and media information 506, and different media information may be composed of the same or different forms of content. By calculating the quality parameters corresponding to the multiple media information, several media information with the highest ranking are selected and displayed on the homepage of the official account, so as to push higher-value media information to users.

[0063] In another exemplary embodiment, Figure 6 is a schematic diagram of another optional method for processing media information according to an embodiment of the present invention, such as Figure 6 As shown, the above-mentioned method for processing media information can be applied in the official account mini-program, wherein the resources allowed to be pushed by the official account may include multiple media information such as media information 602, media information 604 and media information 606, and different media information are respectively composed of content in different forms. By calculating the quality parameters corresponding to the multiple media information, several media information with the highest ranking are selected and displayed on the homepage of the official account, so as to push higher-value media information to users.

[0064] Optionally, in this embodiment, the above-mentioned multiple media information to be recommended may include but is not limited to being composed of media information of multiple forms, each media information may include content of different forms, or may be media information of different forms, and the above-mentioned quality parameters are obtained by fusing at least two value parameters of the corresponding media information based on the target combination parameters.

[0065] Among them, the above-mentioned value parameters may include but are not limited to the click score given by the click rate prediction model, which represents the probability of the media information being clicked; the time score given by the reading time prediction model, which represents the probability of the media information being read for a long time; the timeliness score calculated by the mass sending time, exposure time and message category of the media information, which represents the timeliness of the media information.

[0066] The above target combination parameters can be obtained by combining them through a specific fusion formula, and the weight corresponding to each value parameter in the fusion formula is the above target combination parameter.

[0067] Exemplarily, the above-mentioned quality parameters can be understood as the final ranking scores of media information. These different value scores are combined through a specific fusion formula to obtain the above-mentioned quality parameters, and then multiple quality parameters are sorted to obtain the final ranking score, and the media information whose ranking scores meet specific conditions is determined as the target media information, wherein the specific conditions can be understood as the first N media information in the ranking being the target media information, the first media information in the ranking being the target media information, etc.

[0068] It should be noted that the above target combination parameters are combination parameters determined by the target parameter model through multiple rounds of iterations through the cross entropy algorithm, wherein the cross entropy algorithm includes the following process:

[0069] S1, initialize the mean μ vector and variance σ vector of the Gaussian distribution. For example, μ is set to a set of combination parameters that have been verified to have good performance, and σ is set to the difference between the maximum and minimum values ​​of the pre-set combination parameters;

[0070] S2, randomly generate N groups of combination parameters according to Gaussian distribution, obtain N formulas, calculate their reward values ​​(Reward, for example, set as α*consumption index + β*duration index + γ*timeliness index), and select the top N groups of parameters with the best Reward;

[0071] S3: Count the mean and variance of the top N groups of parameters, make a small perturbation to the variance (to prevent falling into the local optimum too early) to obtain a new Gaussian distribution, and continue sampling according to the new Gaussian distribution to obtain new parameters to be evaluated;

[0072] S4, after several iterations of S2 and S3, will eventually converge to a set of better combination parameters.

[0073] In an exemplary embodiment, Figure 7 is a schematic diagram of another optional method for processing media information according to an embodiment of the present invention, such as Figure 7 As shown, the following steps are included:

[0074] S1, starting from the normal distribution N(μ,σ2);

[0075] S2, use this distribution to generate N vectors;

[0076] S3, evaluates each vector and selects the best one, indicated by shading;

[0077] S4, calculate the mean and standard deviation of the optimal vector;

[0078] S5, adding a noise term to the standard deviation to avoid premature convergence to the local optimum;

[0079] S6, this mean and standard deviation define the normal distribution for the next iteration.

[0080] Optionally, in this embodiment, the above-mentioned target parameter model is a model that constrains the decrease range of the value parameter to meet the preset range condition, and the target value parameter maximization is used as the training goal. It can be understood as maximizing other value parameters under the premise that some value parameters do not decrease, which can be achieved by subdividing the evaluation indicators and constructing a reward formula with additional constraints.

[0081] It should be noted that the above preset amplitude conditions can be set manually as needed, for example:

[0082] For each target value parameter or constraint value parameter, the following indicators can be used to evaluate it: 1) Total click consumption indicator: all-format content GAUC (denoted as M1); 2) Single-format content consumption indicator: single-format content GAUC, such as subscription text and picture GAUC and video account short video GAUC (denoted as M2 and M3); 3) Total duration indicator: the average reading time (or playback time) of the top 4 items in the session (denoted as M4); 4) Single-format content timeliness indicator: the exposure share of subscription text and picture content published within 3 hours / 24 hours, and the exposure share of video account short video content published within 2 hours / 24 hours (denoted as M5, M6, M7, M8).

[0083] These indicators under the baseline formula are represented by the subscript _base, and the goal of value formula combination optimization is:

[0084]

[0085]

[0086] That is, in each iteration, M1 and M4 are maximized, M2 and M3 do not decrease, and M5, M6, M7, and M8 decrease by no more than 0.5%.

[0087] Optionally, in this embodiment, the target value parameter may be understood as a value parameter that needs to be maximized as a target, and the constraint value parameter may be understood as a value parameter that is not allowed to decrease or the decrease range cannot exceed a threshold.

[0088] Optionally, in this embodiment, the target parameter model adaptively determines the number of first combination parameters for sampling in the current round and the number of second combination parameters for reward values ​​in the current round that meet preset conditions according to the calculation time of the previous round of sampling during multiple rounds of iterations. This can be understood as the target parameter model adaptively adjusting the number of combination parameters for random search in each round during multiple rounds of iterations.

[0089] In the cross entropy algorithm, an inappropriate hyperparameter will cause the new Gaussian distribution mean to fall into a poor search space, causing the search results to deviate from the global optimum. Therefore, before the target parameter model starts calculating the quality parameters or before each round of executing the cross entropy algorithm, a small-scale random search algorithm can be used in advance to evaluate the difficulty of searching for a better combination of parameters under hard constraints, including the following steps:

[0090] S1, sets the search scale of random search, that is, the number of combination parameters to be searched N1;

[0091] S2, according to the hard constraints that the timeliness index (M5, M6, M7, M8), single-mode GAUC (M2, M3) and total GAUC (M1) do not decrease, evaluate the number of combined parameters N2 that meet the condition;

[0092] S3, then in the cross entropy algorithm, the ratio between the number of sampling parameters in each round and the number of parameters constituting the new Gaussian distribution (called the effective search ratio) should satisfy: Top N / N≤N1 / N2, where (Top N corresponds to the number of the first combination of parameters mentioned above, and N corresponds to the number of the second combination of parameters mentioned above);

[0093] The search time of the cross entropy algorithm is proportional to the number of parameters sampled in each round of iteration. Therefore, based on the evaluation of computational time and search efficiency, for example, N=500 is set and Top N is set to Top N=N*(N1 / N2)=10. In this way, it can be ensured that the Gaussian distribution mean obtained by the statistics of the Top N sampling advantages also has good performance. In short, for a new fusion formula or new constraints, analyzing the effective search ratio of the cross entropy algorithm in advance, combined with the tolerance for the algorithm time consumption, can assist in the selection of hyperparameters.

[0094] It should be noted that the above reward value is adaptively determined by the change range of the target value parameter from the previous round to the current round during multiple iterations. It can be understood that the calculation formula for the reward value that maximizes the target value parameter is R=α*M1+β*M4. α and β are dynamically calculated during the algorithm iteration process, so that the algorithm can dynamically adjust the optimization direction according to the optimization difficulty of each value parameter as the target, avoiding falling into the search space dominated by a certain indicator.

[0095] Among them, the improvement of M1 index before and after the algorithm is δ(M1 new -M1 old ), the improvement of M4 index before and after the algorithm is δ(M4 new -M4 old ), then the weighted weight of the reward value is obtained by the inverse of the normalized improvement. The adaptive reward value is defined as: R = α*M1+β*M4, where:

[0096]

[0097]

[0098] Therefore, value targets that are easier to improve have lower reward weights, making the optimization of each value target more balanced.

[0099] S206: Determine target media information to be published at a target publishing position from the plurality of media information according to the quality parameter.

[0100] Optionally, in this embodiment, the target media information may be understood as media information determined based on the quality parameter ranking, and may be the media information with the highest quality parameter ranking, or may be a plurality of media information with the highest quality parameter ranking.

[0101] Through this embodiment, a media information publishing request is obtained, wherein the media information publishing request is used to request the publishing of media information at a target publishing position, and in response to the media information publishing request, quality parameters corresponding to a plurality of media information to be recommended are determined, wherein each quality parameter is determined by at least two value parameters of the corresponding media information based on a target combination parameter, and the target combination parameter is a combination parameter to which a target parameter model converges after multiple rounds of iterations, and the target parameter model is a model that uses the decrease amplitude of the constraint value parameter to satisfy a preset amplitude condition and the maximization of the target value parameter as a training target, and the at least two value parameters include a target value parameter and a constraint value parameter, and the target parameter model adaptively determines the first combination of the current round of sampling according to the calculation time of the previous round of sampling during multiple rounds of iterations. The number of parameters and the second number of combined parameters of the current round of reward values ​​satisfy the preset conditions, the hyperparameters of the target parameter model include the first number of combined parameters and the second number of combined parameters, the reward value is adaptively determined by the change range of the target value parameter from the previous round to the current round during multiple rounds of iterations, the method of determining the target media information to be published at the target publishing position from multiple media information according to the quality parameters, through adaptive hyperparameter selection and adaptive reward value setting, the optimization of each value parameter is more balanced and approaches the Pareto optimality, thereby realizing the technical effect of optimizing the determination method of quality parameters, improving the calculation accuracy of quality parameters, and improving the publishing efficiency of media information, thereby solving the technical problem of low publishing efficiency of media information due to inaccurate quality parameters.

[0102] As an optional scheme, in response to a media information publishing request, quality parameters corresponding to multiple media information to be recommended are determined, including: in response to the media information publishing request, at least two value parameters corresponding to each of the multiple media information are obtained, wherein the multiple media information include media information in different forms; multiple rounds of iterations are performed according to a target parameter model to determine a target combination parameter, wherein during the multiple rounds of iterations, the target parameter model is a model that takes maximizing the value parameters in all forms as a training target on the premise that the decrease in the value parameters in each single form meets a preset amplitude condition; at least two value parameters corresponding to the multiple media information are merged according to the target combination parameter to determine a quality parameter set, wherein the quality parameter set is used to perform mixed and uniform sorting of multiple media information in different forms to determine the target media information.

[0103] Optionally, in this embodiment, the above-mentioned media information in different forms can be understood as one piece of media information including content in different forms or media information in different single forms.

[0104] It should be noted that the above-mentioned value parameters under the single form can be understood as the GAUC of single-form content, for example, the GAUC for subscription pictures and texts and the GAUC for short videos on video accounts.

[0105] As an optional solution, multiple rounds of iterations are performed based on the target parameter model to determine the target combination parameters, including:

[0106] Obtaining a total consumption value parameter, a first form consumption value parameter, a second form consumption value parameter, a total duration value parameter, a first form time value parameter, and a second form time value parameter corresponding to a plurality of media information, wherein the target value parameter includes the total consumption value parameter and the total duration value parameter, the constraint value parameter includes the first form consumption value parameter, the second form consumption value parameter, the first form time value parameter, and the second form time value parameter, the first form consumption value parameter and the second form consumption value parameter are consumption value parameters corresponding to media information of different forms, and the first form time value parameter and the second form time value parameter are time value parameters corresponding to media information of different forms;

[0107] The target parameter model is used to determine the target combination parameters, with the first-form consumption value parameters and the second-form consumption value parameters not decreasing during multiple rounds of iterations, the first-form time value parameters and the second-form time value parameters not decreasing by more than the preset threshold during multiple rounds of iterations, and the total consumption value parameters and the total time value parameters being maximized as the training objectives.

[0108] Optionally, in this embodiment, for each optimization target or constraint target, the following indicators can be used to evaluate it: 1) Total click consumption indicator: all-form content GAUC (denoted as M1, corresponding to the aforementioned total consumption value parameter); 2) Single-form content consumption indicator: single-form content GAUC, such as subscription text and picture GAUC and video number short video GAUC (denoted as M2 and M3, corresponding to the aforementioned first-form consumption value parameter and second-form consumption value parameter); 3) Total duration indicator: the average reading time (or playback time) of the top 4 items in the session (denoted as M4, corresponding to the aforementioned total duration value parameter); 4) Single-form content timeliness indicator: the exposure share of subscription text and picture content published within 3 hours / 24 hours, and the exposure share of video number short video content published within 2 hours / 24 hours (denoted as M5, M6, M7, M8, where M5 and M6 correspond to the aforementioned first-form timeliness value parameters, and M7 and M8 correspond to the aforementioned second-form timeliness value parameters).

[0109] As an optional solution, the above method also includes: adaptively updating the reward value weight parameters corresponding to each round during multiple rounds of iterations in the following manner, wherein the reward value weight parameters are used to determine the reward value: obtaining a first change range of the total consumption value parameter during the iteration from the previous round to the current round and a second change range of the total duration value parameter during the iteration from the previous round to the current round; normalizing the first change range and the second change range to determine the reward value weight parameters corresponding to each round.

[0110] In an exemplary embodiment, the calculation formula of the reward value under the above maximization objective is R=α*M1+β*M4. The reward value weight parameters α and β are dynamically calculated during the algorithm iteration process, so that the algorithm can dynamically adjust the optimization direction according to the optimization difficulty of each value objective, avoiding falling into the search space dominated by a certain indicator.

[0111] Among them, the first change range of M1 index before and after the algorithm is δ(M1 new -M1 old ), the second change range of the M4 index before and after the algorithm is δ(M4 new -M4 old ), then the reward value weight parameter is obtained by the inverse of the normalized improvement. The adaptive reward value is defined as: R = α*M1+β*M4, where:

[0112]

[0113]

[0114] Therefore, value targets that are easier to improve have lower reward weights, making the optimization of each value target more balanced.

[0115] As an optional solution, multiple rounds of iterations are performed according to the target parameter model to determine the target combination parameters, including: obtaining an initial mean vector and an initial variance vector, wherein the initial mean vector is used to represent a set of pre-set initial combination parameters, and the initial variance vector is used to represent the difference between the maximum value and the minimum value in a set of initial combination parameters; multiple rounds of iterations are performed in the following manner to determine the target combination parameters, wherein the samples sampled in each round are regarded as sample mean vectors, the sample variance vectors corresponding to the sample mean vectors in each round, the sample mean vectors in the first round of sampling are the initial mean vectors, the combination parameters obtained in the current round of sampling are regarded as N groups of combination parameters, and the combination parameters obtained in the next round of sampling are regarded as M groups Combination parameters, N is a positive integer, M is a positive integer less than or equal to N: N groups of combination parameters are generated according to the sample mean vector and the sample variance vector, and N reward values ​​are determined according to whether the decrease amplitude of the constraint value parameters corresponding to the N groups of combination parameters meets the preset amplitude condition; the reward values ​​in the first M descending order are determined from the N reward values; M groups of combination parameters are determined according to the first M reward values, and the intermediate mean vector and the intermediate variance vector corresponding to the M groups of combination parameters are determined, wherein the intermediate mean vector is regarded as the sample mean vector in the next round of iteration, and the intermediate variance vector is regarded as the sample variance vector in the next round of iteration; the M groups of combination parameters that meet the preset convergence conditions are determined as the target combination parameters.

[0116] In an exemplary embodiment, Figure 8 is a schematic diagram of another optional method for processing media information according to an embodiment of the present invention, such as Figure 8 As shown, the above-mentioned multiple rounds of iterations are performed according to the target parameter model to determine the target combination parameters, which may include but is not limited to the following steps:

[0117] S802, setting the mean μ vector (corresponding to the aforementioned initial mean vector) and the variance σ vector (corresponding to the aforementioned sample variance vector) of the initial Gaussian distribution, such as μ is set to a set of combined parameters of a value formula that has been verified to perform well, and σ is set to the difference between the maximum and minimum values ​​of the pre-set combined parameters;

[0118] S804, randomly generate N groups of combination parameters according to Gaussian distribution, calculate the reward value corresponding to each group of combination parameters, and select the top N (corresponding to the aforementioned M) groups of parameters with the best reward;

[0119] As an optional solution, the method further includes: performing a perturbation operation on the intermediate variance vector to obtain a perturbation variance vector; and generating M groups of combination parameters according to the intermediate mean vector regarded as the sample mean vector and the perturbation variance vector.

[0120] S806: Count the mean and variance of the top N groups of parameters, make a small perturbation to the variance (to prevent falling into the local optimum too early) to obtain a new Gaussian distribution, and continue sampling according to the new Gaussian distribution to obtain new parameters to be evaluated;

[0121] S808, repeating S804 and S806, during which the reward value weight is dynamically updated;

[0122] S810: Converge to a set of better parameters and end the search.

[0123] As an optional solution, according to whether the decline of the constraint value parameters corresponding to the N groups of combination parameters meets the preset amplitude condition, N reward values ​​are determined, including: according to whether the decline of the constraint value parameters corresponding to the N groups of combination parameters meets the preset amplitude condition, a piecewise function is set to determine the N reward values, wherein the piecewise function is used to determine the N reward values ​​respectively in the following manner: when the first form consumption value parameter and the second form consumption value parameter corresponding to each group of combination parameters do not meet the corresponding constraint condition, and the first form time value parameter and the second form time value parameter meet the corresponding constraint condition, the reward value is set to be based on the reward value weight The reward value is set to the weighted sum of the total consumption value parameter and the total duration value parameter determined by the weight parameter; when the constraint value parameters corresponding to each set of combination parameters meet the corresponding constraint conditions, the reward value is set to the weighted sum of the total consumption value parameter and the total duration value parameter determined by the reward value weight parameter; when the first form consumption value parameter and the second form consumption value parameter corresponding to each set of combination parameters meet the corresponding constraint conditions, and / or the first form time value parameter and the second form time value parameter do not meet the corresponding constraint conditions, the reward value is set to the weighted difference between the negative value of the total consumption value parameter based on the reward value weight parameter and the total duration value parameter.

[0124] Optionally, in this embodiment, in order to realize the search of the optimization target with constraints, the reward value Reward of the cross entropy algorithm is set to a piecewise function:

[0125]

[0126] When the value formula satisfies all constraints, Reward is equal to the optimization target; when the single-form content GAUC is not satisfied, the reward brought by the total duration indicator M4 is set to a negative value, driving the search algorithm to improve the total GAUC target; in other cases, both click and duration rewards are set to negative values, driving the search algorithm to leave the area near the secondary advantage.

[0127] As an optional solution, the above method also includes: adaptively determining the number of first combination parameters and the number of second combination parameters corresponding to each round during multiple rounds of iterations in the following manner: obtaining an initial number of combination parameters, wherein the initial number of combination parameters is a pre-set number of combination parameters to be searched; determining the number of evaluation groups and parameters that meet the constraints according to the total consumption value parameter, the first form consumption value parameter, the second form consumption value parameter, the first form time value parameter, and the second form time value parameter not decreasing as constraints; determining the ratio between the number of evaluation combination parameters and the preset number of combination parameters as the ratio between the number of first combination parameters and the number of second combination parameters corresponding to each round; adaptively setting the number of first combination parameters corresponding to the next round according to the calculation time of each round, and determining the number of second combination parameters according to the ratio.

[0128] For example, Fig. 9 is a schematic diagram of another optional method for processing media information according to an embodiment of the present invention, such as Fig. 9 As shown, before the above step S802, the following steps are also included but not limited to:

[0129] S902, setting the search scale of the random search, that is, the number of combination parameters to be searched N1;

[0130] S904, according to the hard constraint conditions that the timeliness index (M5, M6, M7, M8), the single-mode GAUC (M2, M3) and the total GAUC (M1) do not decrease, evaluate the number of combination parameters N2 that meet the condition, that is, the effective search ratio is Top N / N=N1 / N2 (corresponding to the ratio between the aforementioned evaluation combination parameter number and the preset combination parameter number is determined as the ratio between the first combination parameter number and the second combination parameter number corresponding to each round);

[0131] S906, determine the N value in the cross entropy algorithm based on the computing power and time consumption evaluation, and then set the Top N value = N*(N1 / N2).

[0132] As an optional solution, obtaining the first form time validity value parameter and the second form time validity value parameter corresponding to the plurality of media information in the media information set includes:

[0133] Obtaining a first timeliness value sub-parameter and a second timeliness value sub-parameter, wherein the first timeliness value sub-parameter and the second timeliness value sub-parameter represent exposure ratio parameters corresponding to the first media information at different release times;

[0134] Obtaining a third timeliness value sub-parameter and a fourth timeliness value sub-parameter, wherein the second form timeliness value parameter includes the third timeliness value sub-parameter and the fourth timeliness value sub-parameter, the third timeliness value sub-parameter and the fourth timeliness value sub-parameter represent exposure ratio parameters corresponding to the second media information at different release times, and the first media information and the second media information are media information of different forms;

[0135] The first form time-limit value parameter is determined according to the first time-limit value sub-parameter and the second time-limit value sub-parameter, and the second form time-limit value parameter is determined according to the third time-limit value sub-parameter and the fourth time-limit value sub-parameter.

[0136] The present application is further explained below with reference to specific examples:

[0137] This application proposes an improved multi-objective cross-entropy algorithm (Improved Multi-Objective Cross-Entropy Method) for media information recommendation, which is used to combine multiple value scores (such as click scores, duration scores, timeliness scores, etc.) related to the media information to be recommended to obtain a final ranking score.

[0138] This application proposes the following improvements based on the cross entropy algorithm:

[0139] 1) Combinatorial optimization with soft-constraint values ​​(soft-constraint): The algorithm is allowed to regard some value targets as constraint indicators, that is, instead of maximizing a single value as the optimization goal, other value targets are maximized on the premise that the value does not decrease.

[0140] 2) Adaptive hyper-parameter settings: In the cross entropy algorithm, "how many new evaluation combinations to sample" and "how many optimal combinations to select" are hyper-parameters that need to be carefully adjusted. This application automatically determines these two hyper-parameters through pre-processing random search, so that the reward can increase monotonically with iterations.

[0141] 3) Adaptive reward: In the cross entropy algorithm, the design of the reward formula for evaluating the quality of the combination parameters relies on manual experience. This application dynamically updates the calculation composition of the reward value in each round of iteration by calculating the optimization difficulty of each value target, so that the optimization of each value target is more balanced and close to the Pareto optimality.

[0142] In an exemplary embodiment, taking the subscription account information flow scenario as an example, a message is described by multiple value scores. For example, the click score (pctr) given by the click-through rate prediction model represents the probability of being clicked; the duration score (preadtime) given by the reading time prediction model represents the probability of being read for a long time; the timeliness score calculated by the message group sending time, exposure time and message category represents timeliness (ptimebonus). The final ranking score of the message is obtained by combining these different value scores through a specific fusion formula (called a value formula). For example, score = w1*pctr+w2*preadtime+w3*ptimebonus, where w1, w2, and w3 are combination parameters.

[0143] For the combination parameters of the value formula, the specific value needs to be determined through the search algorithm. The random search algorithm is the most intuitive and easy-to-use algorithm. For a given search space, the algorithm traverses each combination parameter, evaluates its various indicators, and finally selects the optimal point through manually set rules.

[0144] Fig.10 is a schematic diagram of another optional method for processing media information according to an embodiment of the present invention, such as Fig.10 As shown: In the two-dimensional coordinate system, each point represents the GAUC index and average duration index of the corresponding sorting result under a combination of parameters. Fig.10 The midpoint 1002 represents the structure sorted only by pctr (i.e. w1=1, w2=w3=0). The remaining points are the results of the combination value formula using pctr and preadtime. The random search algorithm relies on manual selection (or manually designed rules) to select point 1004 in the upper right corner of the coordinate system and identify it as the optimal combination parameter.

[0145] Among them, the cross entropy algorithm is a combinatorial optimization algorithm. Compared with the random search algorithm that exhaustively evaluates the effects of all combination parameters, the cross entropy algorithm starts from the initial parameters and iteratively converges to a better combination parameter, which can avoid exploring in the combination parameter space with poor effects and reduce more invalid calculations.

[0146] However, the algorithm has some shortcomings. First, the algorithm has several hyperparameters that require manual experience to determine. Inappropriate hyperparameters may cause the algorithm to degrade in effectiveness as the iterations progress. Second, the design of the reward function also relies on manual experience. Inappropriate rewards will cause the algorithm to converge to a non-global optimal point.

[0147] For the above two shortcomings, the improved algorithms proposed in this application are as follows:

[0148] 1) Adaptive hyperparameter selection. This application automatically determines these two hyperparameters through preprocessing random search;

[0149] 2) Adaptive reward value: This application calculates the optimization difficulty of each value target and dynamically updates the calculation method of the reward value in each round of iteration, so that the optimization of each value target is more balanced and close to the Pareto optimality.

[0150] At the same time, this application proposes to adjust some value targets to constraint indicators, that is, instead of maximizing a single value as the optimization goal, other value targets are maximized on the premise that the value does not decrease. This makes the cross entropy algorithm more suitable for subscription account information flow sorting scenarios.

[0151] Application scenarios of the algorithm: recommendation and sorting technology of pushed message cards (including those subscribed by users themselves or actively recommended by the platform) in the subscription account message box. It is mainly used in the unified sorting scenario of all content messages, that is, it is necessary to sort different forms of content (such as public account articles, public account pictures, video account videos, etc.) in a mixed and unified manner.

[0152] In this scenario, there are business requirements for the consumption levels of different content forms, that is, the consumption of a certain form of content cannot be increased at the expense of the consumption level of another form of content. At the same time, the ranking of messages should take into account multiple consumption indicators and platform ecological indicators such as click-through rate, user stay time, and message timeliness.

[0153] Under this complex goal, it is necessary to introduce better combinatorial optimization algorithms to balance multiple constraints and multiple optimization tasks.

[0154] (1) Cross entropy algorithm process:

[0155] S1, initialize the mean μ vector and variance σ vector of the Gaussian distribution, such as μ is set to a set of combined parameters of a value formula that has been verified to perform well, and σ is set to the difference between the maximum and minimum values ​​of the pre-set combined parameters;

[0156] S2, randomly generate N groups of combination parameters according to Gaussian distribution, obtain N formulas, calculate their reward values ​​(Reward, for example, set as α*consumption index + β*duration index + γ*timeliness index), and select the top N groups of parameters with the best Reward;

[0157] S3: Count the mean and variance of the top N groups of parameters, make a small perturbation to the variance (to prevent falling into the local optimum too early) to obtain a new Gaussian distribution, and continue sampling according to the new Gaussian distribution to obtain new parameters to be evaluated;

[0158] S4, after several iterations of S2 and S3, will eventually converge to a set of better parameters.

[0159] (2) Combinatorial optimization with soft constraint value:

[0160] In the iteration of subscription account scenarios, it is not required to maximize multiple indicators such as duration, timeliness, and consumption of various forms of content at the same time, but to transform them into a more practical optimization goal: maximize other key indicators while maintaining the premise that some indicators do not decline. This is achieved by subdividing evaluation indicators and building a constrained reward formula.

[0161] For each optimization goal or constraint goal, the following indicators can be used to evaluate it: 1) Total click consumption indicator: all-format content GAUC (denoted as M1); 2) Single-format content consumption indicator: single-format content GAUC, such as subscription text and picture GAUC and video account short video GAUC (denoted as M2 and M3); 3) Total duration indicator: the average reading time (or playback time) of the top 4 items in the session (denoted as M4); 4) Single-format content timeliness indicator: the exposure share of subscription text and picture content published within 3 hours / 24 hours, and the exposure share of video account short video content published within 2 hours / 24 hours (denoted as M5, M6, M7, M8).

[0162] These indicators under the baseline formula are represented by the subscript _base, and the goal of value formula combination optimization is:

[0163]

[0164]

[0165] That is, the goal is to maximize total clicks and total consumption, α and β are the weight values ​​of the combination of the two goals, and the consumption and timeliness exposure indicators of single-form content are used as constraint targets. The former requires that it should not decrease, and the latter requires that its maximum decrease threshold does not exceed 0.5%.

[0166] In order to achieve the search of constrained optimization objectives, the reward value Reward of the cross entropy algorithm is set as a piecewise function:

[0167]

[0168] When the value formula satisfies all constraints, Reward is equal to the optimization target; when the single-form content GAUC is not satisfied, the reward brought by the total duration indicator M4 is set to a negative value, driving the search algorithm to improve the total GAUC target; in other cases, both click and duration rewards are set to negative values, driving the search algorithm to leave the area near the secondary advantage.

[0169] (3) Adaptive hyperparameter selection:

[0170] In the cross entropy algorithm, an inappropriate hyperparameter will cause the new Gaussian distribution mean to fall into a poor search space, causing the search results to deviate from the global optimum.

[0171] Therefore, before executing the cross entropy algorithm, a small-scale random search algorithm can be used to evaluate the difficulty of searching for a better combination of parameters under hard constraints, including the following steps:

[0172] S1, sets the search scale of random search, that is, the number of combination parameters to be searched N1;

[0173] S2, according to the hard constraints that the timeliness index (M5, M6, M7, M8), single-mode GAUC (M2, M3) and total GAUC (M1) do not decrease, evaluate the number of combined parameters N2 that meet the condition;

[0174] S3, then in the cross entropy algorithm, the ratio between the number of sampled parameters in each round and the number of parameters constituting the new Gaussian distribution (called the effective search ratio) should satisfy: Top N / N≤N1 / N2;

[0175] The search time of the cross entropy algorithm is proportional to the number of parameters sampled in each round of iteration. Therefore, based on the evaluation of computational time and search efficiency, for example, N=500 is set and Top N is set to Top N=N*(N1 / N2)=10. In this way, it can be ensured that the Gaussian distribution mean obtained by the statistics of the Top N sampling advantages also has good performance. In short, for a new fusion formula or new constraints, analyzing the effective search ratio of the cross entropy algorithm in advance, combined with the tolerance for the algorithm time consumption, can assist in the selection of hyperparameters.

[0176] (4) Adaptive reward value:

[0177] The calculation formula of the reward value under the maximization goal is R = α*M1+β*M4. α and β are dynamically calculated during the algorithm iteration process, so that the algorithm can dynamically adjust the optimization direction according to the optimization difficulty of each value target, avoiding falling into the search space dominated by a certain indicator.

[0178] Fig.11 is a schematic diagram of another optional method for processing media information according to an embodiment of the present invention, such as Fig.11 As shown, it includes a reward value contour map under the condition of fixed α=β=1 in a two-dimensional search space. In the upper left area of ​​the space, since the duration indicator increases faster as w2 increases, the reward value also changes at a faster rate. Therefore, the cross entropy algorithm tends to select a combination parameter with large w2 and small w1. However, this may cause the importance of the click GAUC indicator to be ignored.

[0179] Therefore, the improvement of M1 index before and after the algorithm is δ(M1 new -M1 old ), the improvement of M4 index before and after the algorithm is δ(M4 new -M4 old ), then the weighted weight of the reward value is obtained by the inverse of the normalized improvement. The adaptive reward value is defined as: R = α*M1+β*M4, where:

[0180]

[0181]

[0182] Therefore, value targets that are easier to improve have lower reward weights, making the optimization of each value target more balanced.

[0183] In summary, the above improved cross entropy algorithm process includes but is not limited to:

[0184] S1, random search preprocessing;

[0185] S1-1, set the search scale of random search, that is, the number of combination parameters to be searched N1;

[0186] S1-2, according to the hard constraints that the timeliness index (M5, M6, M7, M8), the single-mode GAUC (M2, M3) and the total GAUC (M1) do not decrease, evaluate the number of combined parameters N2 that meet the conditions; the effective search ratio Top N / N = N1 / N2;

[0187] S1-3, determine the N value in the cross entropy algorithm based on the computing power and time consumption evaluation, and then the Top N value = N*(N1 / N2);

[0188] S2, set the mean μ vector and variance σ vector of the initial Gaussian distribution, such as μ is set to a set of combination parameters of a value formula that has been verified to have good performance, and σ is set to the difference between the maximum and minimum values ​​of the pre-set combination parameters;

[0189] S3, randomly generates N groups of combined parameters according to Gaussian distribution, calculates their reward values, and selects the top N groups of parameters with the best reward;

[0190] S4: Count the mean and variance of the top N groups of parameters, make a small perturbation to the variance (to prevent falling into the local optimum too early) to obtain a new Gaussian distribution, and continue sampling according to the new Gaussian distribution to obtain new parameters to be evaluated;

[0191] S5, repeating S3 and S4, dynamically updating the reward value weight during the process;

[0192] S6, converge to a set of better parameters and end the search.

[0193] Through this embodiment, the existing cross-entropy algorithm is difficult to find a better parameter combination through iteration when applied to the subscription account information flow scenario, while the improved cross-entropy algorithm adopted in this embodiment has achieved better performance. Specifically, compared with the random search algorithm, with fewer searches and calculation time, it can search for a better combination of parameters for sub-indicators, and automatically meet the constraints without the need for complex manual debugging.

[0194] In an exemplary embodiment, the above-mentioned media information processing method is applied to the subscription account message scenario. When the text and short video messages are uniformly sorted, compared with the baseline search algorithm, the average time per person is increased by +0.4%, and the average number of clicks per person is increased by +0.1%.

[0195] It is understandable that in the specific implementation of this application, related data such as user information is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0196] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0197] According to another aspect of the embodiments of the present application, a media information processing device for implementing the above-mentioned media information processing method is also provided. Fig.12 As shown, the device comprises:

[0198] The acquisition module 1202 is used to acquire a media information publishing request, wherein the media information publishing request is used to request to publish the media information at a target publishing location;

[0199] The first determination module 1204 is used to determine the quality parameters corresponding to the multiple media information to be recommended in response to the media information release request, wherein each quality parameter is determined by at least two value parameters of the corresponding media information based on the target combination parameter, the target combination parameter is the combination parameter converged to by the target parameter model after multiple rounds of iterations, the target parameter model is a model that takes the decrease amplitude of the constraint value parameter to meet the preset amplitude condition and the maximization of the target value parameter as the training target, the at least two value parameters include the target value parameter and the constraint value parameter, the target parameter model adaptively determines the number of first combination parameters of the current round of sampling and the number of second combination parameters of the current round of reward values ​​that meet the preset conditions according to the calculation time of the previous round of sampling during the multiple rounds of iterations, the hyperparameters of the target parameter model include the number of first combination parameters and the number of second combination parameters, and the reward value is adaptively determined by the change amplitude of the target value parameter from the previous round to the current round during the multiple rounds of iterations;

[0200] The second determining module 1206 is used to determine the target media information to be published at the target publishing position from the multiple media information according to the quality parameter.

[0201] As an optional solution, the device is used to respond to a media information publishing request and determine the quality parameters corresponding to multiple media information to be recommended in the following manner: in response to the media information publishing request, obtain at least two value parameters corresponding to each of the multiple media information, wherein the multiple media information includes media information in different forms; perform multiple rounds of iterations according to a target parameter model to determine a target combination parameter, wherein during the multiple rounds of iterations, the target parameter model is a model that takes maximizing the value parameters under all forms as a training target on the premise that the decrease in the value parameters under each single form meets a preset amplitude condition; fuse at least two value parameters corresponding to the multiple media information according to the target combination parameter to determine a quality parameter set, wherein the quality parameter set is used to perform mixed and uniform sorting of multiple media information in different forms to determine the target media information.

[0202] As an optional solution, the device is used to perform multiple rounds of iterations according to the target parameter model in the following manner to determine the target combination parameters: obtain the total consumption value parameter, the first form consumption value parameter, the second form consumption value parameter, the total duration value parameter, the first form time value parameter and the second form time value parameter corresponding to multiple media information, wherein the target value parameter includes the total consumption value parameter and the total duration value parameter, the constraint value parameter includes the first form consumption value parameter, the second form consumption value parameter, the first form time value parameter and the second form time value parameter, the first form consumption value parameter and the second form consumption value parameter are consumption value parameters corresponding to media information of different forms, and the first form time value parameter and the second form time value parameter are time value parameters corresponding to media information of different forms; the first form consumption value parameter and the second form consumption value parameter do not decrease during multiple rounds of iterations, the first form time value parameter and the second form time value parameter do not decrease during multiple rounds of iterations, and the total consumption value parameter and the total duration value parameter are maximized as training objectives, and the target combination parameters are determined using the target parameter model.

[0203] As an optional solution, the device is also used to: adaptively update the reward value weight parameters corresponding to each round during multiple rounds of iterations in the following manner, wherein the reward value weight parameters are used to determine the reward value: obtain a first change range of the total consumption value parameter during the iteration from the previous round to the current round and a second change range of the total duration value parameter during the iteration from the previous round to the current round; normalize the first change range and the second change range to determine the reward value weight parameters corresponding to each round.

[0204] As an optional solution, the device is used to perform multiple rounds of iterations according to the target parameter model in the following manner to determine the target combination parameters: obtain an initial mean vector and an initial variance vector, wherein the initial mean vector is used to represent a set of pre-set initial combination parameters, and the initial variance vector is used to represent the difference between the maximum value and the minimum value in a set of initial combination parameters; perform multiple rounds of iterations in the following manner to determine the target combination parameters, wherein the samples sampled in each round are regarded as sample mean vectors, the sample variance vectors corresponding to the sample mean vectors in each round, the sample mean vectors in the first round of sampling are the initial mean vectors, the combination parameters obtained in the current round of sampling are regarded as N groups of combination parameters, and the combination parameters obtained in the next round of sampling are regarded as is M groups of combination parameters, N is a positive integer, and M is a positive integer less than or equal to N: N groups of combination parameters are generated according to the sample mean vector and the sample variance vector, and N reward values ​​are determined according to whether the decrease amplitude of the constraint value parameters corresponding to the N groups of combination parameters meets the preset amplitude condition; the reward values ​​in the first M descending order are determined from the N reward values; M groups of combination parameters are determined according to the first M reward values, and the intermediate mean vector and the intermediate variance vector corresponding to the M groups of combination parameters are determined, wherein the intermediate mean vector is regarded as the sample mean vector in the next round of iteration, and the intermediate variance vector is regarded as the sample variance vector in the next round of iteration; the M groups of combination parameters that meet the preset convergence condition are determined as the target combination parameters.

[0205] As an optional solution, the device is also used to: perform a perturbation operation on the intermediate variance vector to obtain a perturbation variance vector; and generate M groups of combination parameters based on the intermediate mean vector regarded as a sample mean vector and the perturbation variance vector.

[0206] As an optional solution, the device is used to determine N reward values ​​according to whether the decrease range of the constraint value parameters corresponding to the N groups of combination parameters meets the preset range condition in the following manner: a piecewise function is set according to whether the decrease range of the constraint value parameters corresponding to the N groups of combination parameters meets the preset range condition to determine the N reward values, wherein the piecewise function is used to determine the N reward values ​​respectively in the following manner: when the first form consumption value parameter and the second form consumption value parameter corresponding to each group of combination parameters do not meet the corresponding constraint condition, and the first form time value parameter and the second form time value parameter meet the corresponding constraint condition, the reward value is set to be based on The reward value weight parameter determines the weighted difference between the total consumption value parameter and the total duration value parameter; when the constraint value parameters corresponding to each set of combination parameters meet the corresponding constraint conditions, the reward value is set to the weighted sum of the total consumption value parameter and the total duration value parameter determined based on the reward value weight parameter; when the first form consumption value parameter and the second form consumption value parameter corresponding to each set of combination parameters meet the corresponding constraint conditions, and / or the first form time value parameter and the second form time value parameter do not meet the corresponding constraint conditions, the reward value is set to the weighted difference between the negative value of the total consumption value parameter based on the reward value weight parameter and the total duration value parameter.

[0207] As an optional solution, the device is also used to: adaptively determine the number of first combination parameters and the number of second combination parameters corresponding to each round during multiple rounds of iterations in the following manner: obtain an initial number of combination parameters, wherein the initial number of combination parameters is a pre-set number of combination parameters to be searched; determine the evaluation group and the number of parameters that meet the constraints according to the total consumption value parameter, the first form consumption value parameter, the second form consumption value parameter, the first form time value parameter, and the second form time value parameter not decreasing as constraints; determine the ratio between the number of evaluation combination parameters and the preset number of combination parameters as the ratio between the number of first combination parameters and the number of second combination parameters corresponding to each round; adaptively set the number of first combination parameters corresponding to the next round according to the calculation time of each round, and determine the number of second combination parameters according to the ratio.

[0208] As an optional scheme, the device is used to obtain the first form timeliness value parameters and the second form timeliness value parameters corresponding to multiple media information in a media information set in the following manner: obtain the first timeliness value sub-parameter and the second timeliness value sub-parameter, wherein the first timeliness value sub-parameter and the second timeliness value sub-parameter represent the exposure ratio parameters corresponding to the first media information at different release times; obtain the third timeliness value sub-parameter and the fourth timeliness value sub-parameter, wherein the second form timeliness value parameter includes the third timeliness value sub-parameter and the fourth timeliness value sub-parameter, the third timeliness value sub-parameter and the fourth timeliness value sub-parameter represent the exposure ratio parameters corresponding to the second media information at different release times, and the first media information and the second media information are media information of different forms; determine the first form timeliness value parameter according to the first timeliness value sub-parameter and the second timeliness value sub-parameter, and determine the second form timeliness value parameter according to the third timeliness value sub-parameter and the fourth timeliness value sub-parameter.

[0209] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0210] According to one aspect of the present application, a computer program product is provided, the computer program product comprising a computer program / instruction, the computer program / instruction comprising a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through the communication part 1309, and / or installed from a removable medium 1311. When the computer program is executed by the central processor 1301, various functions provided in the embodiments of the present application are executed.

[0211] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0212] Fig.13 The structure block diagram of a computer system for implementing an electronic device according to an embodiment of the present application is schematically shown.

[0213] It should be noted that Fig.13 The computer system 1300 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0214] like Fig.13As shown, the computer system 1300 includes a central processing unit 1301 (CPU), which can perform various appropriate actions and processes according to the program stored in the read-only memory 1302 (ROM) or the program loaded from the storage part 1308 to the random access memory 1303 (RAM). Various programs and data required for system operation are also stored in the random access memory 1303. The central processing unit 1301, the read-only memory 1302 and the random access memory 1303 are connected to each other through a bus 1304. The input / output interface 1305 (Input / Output interface, i.e., I / O interface) is also connected to the bus 1304.

[0215] The following components are connected to the input / output interface 1305: an input section 1306 including a keyboard, a mouse, etc.; an output section 1307 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1308 including a hard disk, etc.; and a communication section 1309 including a network interface card such as a LAN card, a modem, etc. The communication section 1309 performs communication processing via a network such as the Internet. A drive 1310 is also connected to the input / output interface 1305 as needed. A removable medium 1311, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1310 as needed so that a computer program read therefrom is installed into the storage section 1308 as needed.

[0216] In particular, according to an embodiment of the present application, the process described in each method flow chart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer readable medium, and the computer program contains a program code for executing the method shown in the flow chart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 1309, and / or installed from the removable medium 1311. When the computer program is executed by the central processor 1301, various functions defined in the system of the present application are executed.

[0217] According to another aspect of the embodiment of the present application, an electronic device for implementing the above-mentioned media information processing method is also provided. The electronic device may be Figure 1 The terminal device or server shown in the figure. This embodiment is described by taking the electronic device as a terminal device as an example. Fig.14As shown, the electronic device includes a memory 1402 and a processor 1404. The memory 1402 stores a computer program, and the processor 1404 is configured to execute the steps in any of the above method embodiments through the computer program.

[0218] Optionally, in this embodiment, the electronic device may be located in at least one network device among a plurality of network devices of a computer network.

[0219] Optionally, in this embodiment, the processor may be configured to perform the following steps through a computer program:

[0220] S1, obtaining a media information publishing request, wherein the media information publishing request is used to request publishing of media information at a target publishing location;

[0221] S2, in response to a media information publishing request, determining quality parameters respectively corresponding to a plurality of media information to be recommended, wherein each quality parameter is determined by at least two value parameters of the corresponding media information based on a target combination parameter, the target combination parameter is a combination parameter to which a target parameter model converges after multiple rounds of iterations, the target parameter model is a model that takes the decrease amplitude of a constraint value parameter to meet a preset amplitude condition and the maximization of a target value parameter as a training target, the at least two value parameters include a target value parameter and a constraint value parameter, the target parameter model adaptively determines the first combination parameter quantity of a current round of sampling and the second combination parameter quantity of a current round of reward value that meets a preset condition according to the calculation time of a previous round of sampling during multiple rounds of iterations, the hyperparameters of the target parameter model include the first combination parameter quantity and the second combination parameter quantity, and the reward value is adaptively determined by the change amplitude of the target value parameter from the previous round to the current round during multiple rounds of iterations;

[0222] S3, determining target media information to be published at a target publishing position from multiple media information according to the quality parameter.

[0223] Alternatively, a person skilled in the art may understand that: Fig.14 The structure shown is for illustration only, and the electronic device may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (MID), a PAD, and other terminal devices. Fig.14 The electronic device and the electronic equipment described above are not limited in structure. Fig.14 More or fewer components (such as network interfaces, etc.) as shown in, or with Fig.14 Different configurations are shown.

[0224] Among them, the memory 1402 can be used to store software programs and modules, such as the program instructions / modules corresponding to the media information processing method and device in the embodiments of the present application. The processor 1404 executes various functional applications and data processing by running the software programs and modules stored in the memory 1402, that is, realizing the above-mentioned media information processing method. The memory 1402 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 memory, or other non-volatile solid-state memory. In some instances, the memory 1402 may further include a memory remotely located relative to the processor 1404, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. Among them, the memory 1402 may be specifically, but not limited to, used to store information such as the above-mentioned value parameters. As an example, such as Fig.14 As shown, the memory 1402 may include but is not limited to the acquisition module 1202, the first determination module 1204 and the second determination module 1206 in the media information processing device. In addition, it may also include but is not limited to other module units in the media information processing device, which will not be repeated in this example.

[0225] Optionally, the transmission device 1406 is used to receive or send data via a network. Specific examples of the above-mentioned network may include a wired network and a wireless network. In one example, the transmission device 1406 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices and routers via a network cable so as to communicate with the Internet or a local area network. In one example, the transmission device 1406 is a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0226] In addition, the electronic device further includes: a display 1408 for displaying the media information; and a connection bus 1410 for connecting various module components in the electronic device.

[0227] In other embodiments, the terminal device or server may be a node in a distributed system, wherein 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. The nodes may form a peer-to-peer network, and any form of computing device, such as a server, terminal or other electronic device, may become a node in the blockchain system by joining the peer-to-peer network.

[0228] According to one aspect of the present application, a computer-readable storage medium is provided, and 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 media information processing method provided in various optional implementations of the above-mentioned media information publishing aspects.

[0229] Optionally, in this embodiment, the computer-readable storage medium may be configured to store a computer program for performing the following steps:

[0230] S1, obtaining a media information publishing request, wherein the media information publishing request is used to request publishing of media information at a target publishing location;

[0231] S2, in response to a media information publishing request, determining quality parameters respectively corresponding to a plurality of media information to be recommended, wherein each quality parameter is determined by at least two value parameters of the corresponding media information based on a target combination parameter, the target combination parameter is a combination parameter to which a target parameter model converges after multiple rounds of iterations, the target parameter model is a model that takes the decrease amplitude of a constraint value parameter to meet a preset amplitude condition and the maximization of a target value parameter as a training target, the at least two value parameters include a target value parameter and a constraint value parameter, the target parameter model adaptively determines the first combination parameter quantity of a current round of sampling and the second combination parameter quantity of a current round of reward value that meets a preset condition according to the calculation time of a previous round of sampling during multiple rounds of iterations, the hyperparameters of the target parameter model include the first combination parameter quantity and the second combination parameter quantity, and the reward value is adaptively determined by the change amplitude of the target value parameter from the previous round to the current round during multiple rounds of iterations;

[0232] S3, determining target media information to be published at a target publishing position from multiple media information according to the quality parameter.

[0233] Optionally, in this embodiment, a person of ordinary skill in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing hardware related to the terminal device through a program, and the program may be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0234] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0235] If the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application.

[0236] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0237] In the several embodiments provided in the present application, it should be understood that the disclosed client can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. 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 mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0238] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0239] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0240] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for processing media information, characterized in that: include: Obtaining a media information publishing request, wherein the media information publishing request is used to request publishing of media information at a target publishing location; In response to the media information publishing request, determine the quality parameters respectively corresponding to the plurality of media information to be recommended, wherein each of the quality parameters is determined by at least two value parameters of the corresponding media information based on a target combination parameter, the target combination parameter is a combination parameter converged to by a target parameter model after multiple rounds of iterations, the target parameter model is a model that takes the decrease amplitude of the constraint value parameter to meet a preset amplitude condition and the maximization of the target value parameter as a training target, the at least two value parameters include the target value parameter and the constraint value parameter, the target parameter model adaptively determines the first combination parameter quantity of the current round of sampling and the second combination parameter quantity of the current round of reward value that meets the preset condition according to the calculation time consumption of the previous round of sampling during the multiple rounds of iterations, the hyperparameters of the target parameter model include the first combination parameter quantity and the second combination parameter quantity, and the reward value is adaptively determined by the change amplitude of the target value parameter from the previous round to the current round during the multiple rounds of iterations; The target media information to be published on the target publishing location is determined from the plurality of media information according to the quality parameter.

2. The method according to claim 1, characterized in that The step of determining, in response to the media information publishing request, quality parameters respectively corresponding to the plurality of media information to be recommended includes: In response to the media information publishing request, obtaining the at least two value parameters corresponding to each of the plurality of media information, wherein the plurality of media information includes media information in different forms; Performing the multiple rounds of iterations according to the target parameter model to determine the target combination parameters, wherein during the multiple rounds of iterations, the target parameter model is a model that takes maximizing the value parameters under all forms as the training target, on the premise that the decrease amplitude of the value parameters under each single form meets the preset amplitude condition; The at least two value parameters corresponding to the multiple media information are merged according to the target combination parameter to determine a quality parameter set, wherein the quality parameter set is used to perform mixed and uniform sorting on the multiple media information in different forms to determine the target media information.

3. The method according to claim 2, characterized in that The performing the multiple rounds of iterations according to the target parameter model to determine the target combination parameters includes: Obtaining a total consumption value parameter, a first form consumption value parameter, a second form consumption value parameter, a total duration value parameter, a first form time-limited value parameter, and a second form time-limited value parameter corresponding to the plurality of media information, wherein the target value parameter includes the total consumption value parameter and the total duration value parameter, the constraint value parameter includes the first form consumption value parameter, the second form consumption value parameter, the first form time-limited value parameter, and the second form time-limited value parameter, the first form consumption value parameter and the second form consumption value parameter are consumption value parameters corresponding to media information of different forms, and the first form time-limited value parameter and the second form time-limited value parameter are time-limited value parameters corresponding to media information of different forms; The target parameter model is used to determine the target combination parameters, taking the first form consumption value parameter and the second form consumption value parameter not decreasing during the multiple rounds of iterations, the first form time value parameter and the second form time value parameter decreasing by no more than a preset threshold during the multiple rounds of iterations, and maximizing the total consumption value parameter and the total duration value parameter as training objectives.

4. The method according to claim 3, characterized in that The method further comprises: In the process of performing multiple rounds of iterations, the reward value weight parameter corresponding to each round is adaptively updated in the following manner, wherein the reward value weight parameter is used to determine the reward value: Obtaining a first change range of the total consumption value parameter during an iteration from a previous round to a current round and a second change range of the total duration value parameter during an iteration from a previous round to a current round; The first variation range and the second variation range are normalized to determine the reward value weight parameter corresponding to each round.

5. The method according to claim 4, characterized in that The performing the multiple rounds of iterations according to the target parameter model to determine the target combination parameters includes: Obtaining an initial mean vector and an initial variance vector, wherein the initial mean vector is used to represent a set of preset initial combination parameters, and the initial variance vector is used to represent the difference between the maximum value and the minimum value in the set of initial combination parameters; The target combination parameters are determined by performing multiple rounds of iterations in the following manner, wherein the samples sampled in each round are regarded as sample mean vectors, the sample variance vector corresponds to the sample mean vector in each round, the sample mean vector in the first round of sampling is the initial mean vector, the combination parameters obtained in the current round of sampling are regarded as N groups of combination parameters, and the combination parameters obtained in the next round of sampling are regarded as M groups of combination parameters, N is a positive integer, and M is a positive integer less than or equal to N: Generate the N groups of combined parameters according to the sample mean vector and the sample variance vector, and determine N reward values ​​according to whether the decrease amplitude of the constraint value parameters corresponding to the N groups of combined parameters meets the preset amplitude condition; Determine the top M reward values ​​in descending order from the N reward values; Determine the M groups of combination parameters according to the first M reward values, and determine the intermediate mean vector and the intermediate variance vector corresponding to the M groups of combination parameters, wherein the intermediate mean vector is regarded as the sample mean vector in the next round of iteration, and the intermediate variance vector is regarded as the sample variance vector in the next round of iteration; The M groups of combined parameters that meet a preset convergence condition are determined as the target combined parameters.

6. The method according to claim 5, characterized in that The method further comprises: Performing a perturbation operation on the intermediate variance vector to obtain a perturbation variance vector; The M groups of combined parameters are generated according to the intermediate mean vector regarded as the sample mean vector and the perturbation variance vector.

7. The method according to claim 5, characterized in that The determining of N reward values ​​according to whether the decrease range of the constraint value parameters corresponding to the N groups of combined parameters meets the preset range condition includes: A piecewise function is set according to whether the decrease range of the constraint value parameter corresponding to the N groups of combined parameters meets the preset range condition to determine the N reward values, wherein the piecewise function is used to determine the N reward values ​​respectively in the following manner: When the first form consumption value parameter and the second form consumption value parameter corresponding to each group of combination parameters do not satisfy the corresponding constraint conditions, and the first form time value parameter and the second form time value parameter satisfy the corresponding constraint conditions, the reward value is set to the weighted difference between the total consumption value parameter and the total time value parameter determined based on the reward value weight parameter; When the constraint value parameters corresponding to each group of combination parameters satisfy the corresponding constraint conditions, the reward value is set to the weighted sum of the total consumption value parameter and the total duration value parameter determined based on the reward value weight parameter; When the first form consumption value parameter and the second form consumption value parameter corresponding to each set of combination parameters satisfy the corresponding constraints, and / or the first form time value parameter and the second form time value parameter do not satisfy the corresponding constraints, the reward value is set to the weighted difference between the negative value of the total consumption value parameter based on the reward value weight parameter and the total duration value parameter.

8. The method according to claim 3, characterized in that The method further comprises: In the process of performing multiple rounds of iterations, the number of the first combination parameters and the number of the second combination parameters corresponding to each round are adaptively determined in the following manner: Acquire an initial number of combination parameters, wherein the initial number of combination parameters is a preset number of combination parameters to be searched; According to the total consumption value parameter, the first form consumption value parameter, the second form consumption value parameter, the first form time value parameter and the second form time value parameter not decreasing as the constraint conditions, determining the evaluation groups and the number of parameters that meet the constraint conditions; Determine the ratio between the number of evaluation combination parameters and the number of preset combination parameters as the ratio between the number of the first combination parameters and the number of the second combination parameters corresponding to each round; The number of the first combination parameters corresponding to the next round is adaptively set according to the time consumption of each round of calculation, and the number of the second combination parameters is determined according to the ratio.

9. The method according to claim 3, characterized in that: The obtaining of the first form time validity value parameter and the second form time validity value parameter corresponding to the plurality of media information in the media information set includes: Obtaining a first timeliness value sub-parameter and a second timeliness value sub-parameter, wherein the first timeliness value sub-parameter and the second timeliness value sub-parameter represent exposure ratio parameters corresponding to the first media information at different release times; Obtaining a third time-effectiveness value sub-parameter and a fourth time-effectiveness value sub-parameter, wherein the second form time-effectiveness value parameter includes the third time-effectiveness value sub-parameter and the fourth time-effectiveness value sub-parameter, the third time-effectiveness value sub-parameter and the fourth time-effectiveness value sub-parameter represent exposure ratio parameters corresponding to the second media information at different release durations, and the first media information and the second media information are media information of different forms; The first form time-limit value parameter is determined according to the first time-limit value sub-parameter and the second time-limit value sub-parameter, and the second form time-limit value parameter is determined according to the third time-limit value sub-parameter and the fourth time-limit value sub-parameter.

10. A media information processing device, characterized in that: include: An acquisition module, used for acquiring a media information publishing request, wherein the media information publishing request is used for requesting to publish the media information at a target publishing location; A first determination module is used to determine the quality parameters respectively corresponding to the plurality of media information to be recommended in response to the media information publishing request, wherein each of the quality parameters is determined by at least two value parameters of the corresponding media information based on a target combination parameter, the target combination parameter is a combination parameter converged to by a target parameter model after multiple rounds of iterations, the target parameter model is a model that takes the decrease amplitude of the constraint value parameter to meet a preset amplitude condition and the maximization of the target value parameter as a training target, the at least two value parameters include the target value parameter and the constraint value parameter, the target parameter model adaptively determines the first combination parameter quantity of the current round of sampling and the second combination parameter quantity of the current round of reward value that meets the preset condition according to the calculation time consumption of the previous round of sampling during the multiple rounds of iterations, the hyperparameters of the target parameter model include the first combination parameter quantity and the second combination parameter quantity, and the reward value is adaptively determined by the change amplitude of the target value parameter from the previous round to the current round during the multiple rounds of iterations; The second determining module is used to determine the target media information to be published at the target publishing position from the multiple media information according to the quality parameter.

11. The device according to claim 10, characterized in that The device is used to respond to the media information publishing request in the following manner to determine the quality parameters respectively corresponding to the plurality of media information to be recommended: In response to the media information publishing request, obtaining the at least two value parameters corresponding to each of the plurality of media information, wherein the plurality of media information includes media information in different forms; Performing the multiple rounds of iterations according to the target parameter model to determine the target combination parameters, wherein during the multiple rounds of iterations, the target parameter model is a model that takes maximizing the value parameters under all forms as the training target, on the premise that the decrease amplitude of the value parameters under each single form meets the preset amplitude condition; The at least two value parameters corresponding to the multiple media information are merged according to the target combination parameter to determine a quality parameter set, wherein the quality parameter set is used to perform mixed and uniform sorting on the multiple media information in different forms to determine the target media information.

12. The device according to claim 11, characterized in that The device is used to perform the multiple rounds of iterations according to the target parameter model to determine the target combination parameter in the following manner: Obtaining a total consumption value parameter, a first form consumption value parameter, a second form consumption value parameter, a total duration value parameter, a first form time-limited value parameter, and a second form time-limited value parameter corresponding to the plurality of media information, wherein the target value parameter includes the total consumption value parameter and the total duration value parameter, the constraint value parameter includes the first form consumption value parameter, the second form consumption value parameter, the first form time-limited value parameter, and the second form time-limited value parameter, the first form consumption value parameter and the second form consumption value parameter are consumption value parameters corresponding to media information of different forms, and the first form time-limited value parameter and the second form time-limited value parameter are time-limited value parameters corresponding to media information of different forms; The target parameter model is used to determine the target combination parameters, taking the first form consumption value parameter and the second form consumption value parameter not decreasing during the multiple rounds of iterations, the first form time value parameter and the second form time value parameter decreasing by no more than a preset threshold during the multiple rounds of iterations, and maximizing the total consumption value parameter and the total duration value parameter as training objectives.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program can be executed by a terminal device or a computer to execute the method described in any one of claims 1 to 9.

14. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method described in any one of claims 1 to 9 are implemented.

15. 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 according to any one of claims 1 to 9 through the computer program.

Images