Resource Recommendation Method, Apparatus, Electronic Device, and Storage Medium

Abstract

The present disclosure relates to a resource recommendation method, apparatus, electronic device, and storage medium. The method includes: determining an initial evaluation result of multiple behavioral objectives of candidate resources; obtaining a prior distribution of historical evaluation results of each behavioral objective, and determining a prior distribution parameter corresponding to the initial evaluation result of each behavioral objective according to the prior distribution; generating a resource evaluation result of the candidate resources according to the prior distribution parameters of the multiple behavioral objectives of the candidate resources; and determining a recommended resource from the candidate resources according to the resource evaluation result, where the recommended resource is used for resource recommendation. According to the solution of the present disclosure, the resource evaluation result obtained based on the prior distribution can better fit the positive feedback degree of the user account to the candidate resources, which helps to improve the accuracy of resource recommendation.

Description

Technical Field

[0001] The present disclosure relates to the field of Internet technologies, and in particular, to a resource recommendation method, apparatus, electronic device, computer-readable storage medium, and computer program product. Background Art

[0002] The optimization goal of a resource recommendation system can be to improve positive feedback behavior goals. Among them, positive feedback behavior goals include usage duration goals (such as play duration, play progress), and interaction behavior goals (such as like, follow). Correspondingly, negative feedback behavior goals include usage duration goals (such as short play, exit play), and interaction behavior goals (such as not interested, complaint reporting). In the sorting stage of the resource recommendation system for multiple candidate resources, the sorting model outputs the predicted values of each behavior goal for each candidate resource. Then, the multiple predicted values of each candidate resource are fused to obtain a comprehensive score. Finally, the multiple candidate resources are sorted according to the comprehensive score to determine the candidate resources that can be recommended.

[0003] In the related art, since the data distributions corresponding to the predicted values of each behavior goal are inconsistent and the absolute values of the predicted values vary greatly, in the fusion sorting stage, it is necessary to normalize each predicted value from the same data dimension. For example, a linear function can be used to map the predicted value to a new value, and multiple candidate resources are sorted based on the new value. However, the normalization process in the related art ignores the difference degree between each predicted value, which easily leads to inaccurate resource recommendation. Summary of the Invention

[0004] The present disclosure provides a resource recommendation method, apparatus, electronic device, computer-readable storage medium, and computer program product to at least solve the problem that the normalization method in the related art ignores the difference degree between each predicted value, which easily leads to inaccurate resource recommendation. The technical solution of the present disclosure is as follows:

[0005] According to a first aspect of an embodiment of the present disclosure, a resource recommendation method is provided, including:

[0006] Determine the initial evaluation results of multiple behavior goals of candidate resources;

[0007] Obtain the prior distribution of the historical evaluation results of each behavior goal, and determine the prior distribution parameters corresponding to the initial evaluation results of each behavior goal according to the prior distribution;

[0008] Generate a resource evaluation result of the candidate resource according to the prior distribution parameters of the multiple behavior goals of the candidate resource;

[0009] Determine the recommended resources from the candidate resources according to the resource evaluation results, and the recommended resources are used for resource recommendation.

[0010] In one embodiment, the prior distribution adopts a beta distribution, and the beta distribution of each behavior target is generated according to the historical evaluation results of each behavior target, and the historical evaluation results are generated according to the historical behavior data of the user account for multiple recommended resources.

[0011] In one embodiment, the generation method of the prior distribution of each behavior target includes:

[0012] Obtain multiple historical evaluation results of each behavior target, and generate an expectation and a variance according to the multiple historical evaluation results of each behavior target;

[0013] Generate the alpha coefficient and beta coefficient of the beta distribution according to the expectation and the variance;

[0014] Generate the cumulative distribution of the beta distribution of each behavior target according to the alpha coefficient and the beta coefficient, and use it as the prior distribution of each behavior target.

[0015] In one embodiment, the obtaining of multiple historical evaluation results of each behavior target includes:

[0016] Obtain multiple historical evaluation results of each behavior target within a preset time period before the current moment.

[0017] In one embodiment, the prior distribution of each behavior target is generated after determining the candidate resources;

[0018] Alternatively, the prior distribution of each behavior target is periodically generated before determining the candidate resources.

[0019] In one embodiment, the generating of the resource evaluation result of the candidate resource according to the prior distribution parameters of the multiple behavior targets of the candidate resource includes:

[0020] Obtain the adjustment parameter corresponding to each behavior target;

[0021] Generate the resource evaluation result of the candidate resource according to the adjustment parameter corresponding to each behavior target and the prior distribution parameter.

[0022] In one embodiment, the generating of the resource evaluation result of the candidate resource according to the adjustment parameter corresponding to each behavior target and the prior distribution parameter includes:

[0023] Obtain the resource evaluation result of the candidate resource in any of the following ways:

[0024] Obtain the weighted sum of the adjustment parameters of the multiple behavioral objectives and the prior distribution parameters, and use the weighted sum as the resource evaluation result;

[0025] Obtain the power of the adjustment parameter of the prior distribution parameter of the behavioral objective, and generate the resource evaluation result according to the power of the adjustment parameter of the prior distribution parameter.

[0026] According to the second aspect of the embodiments of the present disclosure, there is provided a resource recommendation device, including:

[0027] An initial result determination module, configured to execute to determine the initial evaluation results of multiple behavioral objectives of a candidate resource;

[0028] A prior parameter determination module, configured to execute to obtain the prior distribution of the historical evaluation results of each behavioral objective, and determine the prior distribution parameters corresponding to the initial evaluation results of each behavioral objective according to the prior distribution;

[0029] A resource result generation module, configured to execute to generate the resource evaluation result of the candidate resource according to the prior distribution parameters of the multiple behavioral objectives of the candidate resource;

[0030] A recommendation module, configured to execute to determine the recommended resources from the candidate resources according to the resource evaluation result, and the recommended resources are used for resource recommendation.

[0031] In one embodiment, the prior distribution adopts a beta distribution, and the beta distribution of each behavioral objective is generated according to the historical evaluation results of each behavioral objective, and the historical evaluation results are generated according to the historical behavior data of the user account for multiple recommended resources.

[0032] In one embodiment, the device further includes:

[0033] A historical result acquisition module, configured to execute to obtain multiple historical evaluation results of each behavioral objective;

[0034] An expectation and variance generation module, configured to execute to generate an expectation and a variance according to the multiple historical evaluation results of each behavioral objective;

[0035] A coefficient generation module, configured to execute to generate the alpha coefficient and the beta coefficient of the beta distribution according to the expectation and the variance;

[0036] A beta distribution generation module, configured to generate a cumulative distribution of a beta distribution for each of the behavior targets according to the alpha coefficient and the beta coefficient, as a prior distribution for each of the behavior targets.

[0037] In one embodiment, the historical result acquisition module is configured to acquire a plurality of historical evaluation results of each of the behavior targets within a preset time period before the current moment.

[0038] In one embodiment, the prior distribution of each of the behavior targets is generated after determining the candidate resources;

[0039] Alternatively, the prior distribution of each of the behavior targets is periodically generated before determining the candidate resources.

[0040] In one embodiment, the resource result generation module is configured to perform operations including:

[0041] An adjustment parameter acquisition unit, configured to acquire an adjustment parameter corresponding to each of the behavior targets;

[0042] A resource result generation unit, configured to generate a resource evaluation result of the candidate resources according to the adjustment parameter corresponding to each of the behavior targets and the prior distribution parameter.

[0043] In one embodiment, the resource result generation unit is configured to perform:

[0044] Obtain the resource evaluation result of the candidate resources by any one of the following methods:

[0045] Obtain the weighted sum of the adjustment parameters and the prior distribution parameters of the plurality of behavior targets, and use the weighted sum as the resource evaluation result;

[0046] Obtain the power of the adjustment parameter of the prior distribution parameter of the behavior target, and generate the resource evaluation result according to the power of the adjustment parameter of the prior distribution parameter.

[0047] According to a third aspect of the embodiments of the present disclosure, there is provided an electronic device, including:

[0048] A processor;

[0049] A memory for storing executable instructions of the processor;

[0050] Wherein, the processor is configured to execute the instructions to implement the resource recommendation method according to any one of the embodiments of the first aspect above.

[0051] According to a fourth aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided. When instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the resource recommendation method according to any one of the embodiments in the first aspect as described above.

[0052] According to a fifth aspect of the embodiments of the present disclosure, a computer program product is provided. The computer program product includes instructions that, when executed by a processor of an electronic device, enable the electronic device to execute the resource recommendation method according to any one of the embodiments in the first aspect as described above.

[0053] The technical solutions provided by the embodiments of the present disclosure at least bring the following beneficial effects:

[0054] After determining the initial evaluation results of multiple behavioral objectives of candidate resources, obtain the prior distribution of the historical evaluation results of each behavioral objective, and determine the prior distribution parameters corresponding to the initial evaluation results of each behavioral objective according to the prior distribution. By mapping the initial evaluation results to prior distribution parameters, not only can the initial evaluation results with different dimensions be mapped to a unified data dimension, but also the discrimination information reflecting the differences between the initial evaluation results of different behavioral objectives can be well reflected. Then, according to the prior distribution parameters of multiple behavioral objectives of the candidate resources, generate the resource evaluation results of the candidate resources, and determine the recommended resources from the candidate resources according to the resource evaluation results, so that the resource evaluation results obtained based on the prior distribution can better fit the positive feedback degree of the user account to the candidate resources, which helps to improve the accuracy of resource recommendation and the overall index effect of the resource recommendation system.

[0055] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and do not limit the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure and do not constitute an improper limitation of the present disclosure.

[0057] Figure 1 is an application environment diagram of a resource recommendation method shown according to an exemplary embodiment.

[0058] Figure 2 is a flowchart of a resource recommendation method shown according to an exemplary embodiment.

[0059] Figure 3 is a flowchart of generating a beta distribution shown according to an exemplary embodiment.

[0060] Figure 4It is a curve graph of a beta distribution shown according to an exemplary embodiment.

[0061] Figure 5 It is a schematic diagram showing the difference reflected by the beta distribution according to an exemplary embodiment.

[0062] Figure 6 It is a flowchart of a resource recommendation method shown according to an exemplary embodiment.

[0063] Figure 7 It is a block diagram of a resource recommendation device shown according to an exemplary embodiment.

[0064] Figure 8 It is a block diagram of an electronic device shown according to an exemplary embodiment. Detailed implementation manners

[0065] In order to enable those of ordinary skill in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0066] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the present disclosure. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0067] It should also be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in the present disclosure are all information and data authorized by the user or fully authorized by all parties.

[0068] The resource recommendation method provided by the present disclosure can be applied to, for example Figure 1In the application environment shown. Among them, the terminal 110 interacts with the server 120 through the network. An application program is installed in the terminal 110, and the application program can be an application program such as a video type, an instant messaging type, an e-commerce type, etc. The terminal 110 can display resources to the user through these application programs. A resource recommendation system is configured in the server 120. In response to a resource recommendation request of a user account, the server 120 recalls a plurality of candidate resources from the resource library through the resource recommendation system. Coarse ranking and / or fine ranking are performed on these candidate resources to obtain initial evaluation results of multiple behavioral objectives of each candidate resource. The server 120 obtains the prior distribution of the historical evaluation results of each behavioral objective, and determines the prior distribution parameters corresponding to the initial evaluation results of each behavioral objective according to the prior distribution. The prior distribution parameters of multiple behavioral objectives of each candidate resource are fused to generate a resource evaluation result of the candidate resource. Multiple candidate resources are ranked according to the resource evaluation result, a resource recommended to the user account is determined from the ranked candidate resources, and the recommended resource is sent to the terminal 110 where the user account is located.

[0069] Among them, the terminal 110 can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, portable wearable devices. The server 120 can be implemented by an independent server or a server cluster composed of multiple servers.

[0070] Figure 2 It is a flowchart of a resource recommendation method shown according to an exemplary embodiment. As Figure 2 shown, the resource recommendation method is used in a server and includes the following steps.

[0071] In step S210, initial evaluation results of multiple behavioral objectives of candidate resources are determined.

[0072] Among them, the resources can be data-transmissible contents such as videos, commodities, news, articles, music, etc. The candidate resources can be resources to be coarsely ranked that are screened from all objects or a specified object set through a recall policy and enter the coarse ranking stage; they can also be resources to be finely ranked that enter the fine ranking stage after being screened by the recall policy and the coarse ranking stage. Among them, the recall policy can adopt a multi-way recall policy. In this case, operations such as interest tag matching, interest entity matching, collaborative filtering, and geographical location matching can be performed according to the account-related information of the user account to screen a batch of resources from all resources or a specified resource set. Among them, the account-related information of the user account is not limited to attribute information corresponding to the user account, historical behavior data, etc.

[0073] Coarse ranking and fine ranking are relative concepts. Coarse ranking can refer to the process of quickly ranking based on relatively few user account features, resource features, etc. Fine ranking can refer to the process of precisely ranking based on more user account features, resource features, etc. Since the accuracy of the fine ranking result is high, but the detection process of the fine ranking model is usually more complex and requires excessive resources, in some embodiments, a batch of resources obtained by recall can be first coarsely ranked to determine a certain number (e.g., 200) of resources; then a certain number of resources can be finely ranked, and resources are recommended to the user account based on the ranking result output in the fine ranking stage.

[0074] Multiple behavior targets can include positive feedback type behavior targets and negative feedback type behavior targets. Among them, positive feedback type behavior targets can include but are not limited to: positive feedback duration type targets (such as the play duration, play progress, effective play, long play, complete play of a resource); positive feedback interaction type behavior targets (such as like, follow, forward, share, comment, favorite, download, enter the personal homepage). Negative feedback type behavior targets can include but are not limited to: negative feedback duration type targets (such as short play, exit play); negative feedback interaction type behavior targets (such as choose not to be interested, complain and report).

[0075] The initial evaluation result is used to quantify the prediction effect of the behavior target and can be represented by probability.

[0076] In one embodiment, the server receives a resource recommendation request sent by a user account. The resource recommendation request can be triggered in any case such as when starting the application, when the user clicks on a specified control, when opening a specified page, etc. In response to the resource recommendation request, the server filters a batch of resources to be coarsely ranked through a recall policy as candidate resources. The server can input each candidate resource into the coarse ranking model to obtain the initial evaluation result corresponding to each behavior target for each candidate resource.

[0077] In another embodiment, the server can also determine multiple resources from the multiple resources to be coarsely ranked obtained by recall based on the coarse ranking model as candidate resources. The server can input each candidate resource into the fine ranking model to obtain the initial evaluation result corresponding to each behavior target for each candidate resource.

[0078] In another embodiment, the number of candidate resources can also be one. In this case, the server can cancel the fusion ranking process and directly send the candidate resource to the user account.

[0079] In step S220, obtain the prior distribution of the historical evaluation result of each behavior target, and determine the prior distribution parameter corresponding to the initial evaluation result of each behavior target according to the prior distribution.

[0080] Among them, the historical evaluation result can be the evaluation result output by the rough ranking model or the fine ranking model; it can also be the evaluation result generated according to the feedback data of the user account on the recommended resources (for example, effective play duration, like information, etc.). The historical evaluation result can be characterized by probability.

[0081] The prior distribution is used to represent the probability distribution of the historical evaluation result. For example, it can be a beta distribution, a gamma distribution, etc. The prior distribution parameter is used to represent the probability that the initial evaluation result may occur in the prior distribution.

[0082] Specifically, after the server obtains the multiple initial evaluation results of the multiple behavioral objectives of each candidate resource, it maps the initial evaluation result of each behavioral objective to the prior distribution corresponding to the behavioral objective to obtain the corresponding prior distribution parameter, and uses this prior distribution parameter as the final evaluation result of the behavioral objective.

[0083] In step S230, according to the prior distribution parameters of the multiple behavioral objectives of the candidate resource, a resource evaluation result of the candidate resource is generated.

[0084] Among them, the resource evaluation result is used to represent the degree of positive feedback of the user account on the candidate resource. The higher the value of the resource evaluation result of the candidate resource, the more likely it is that the user account will generate positive feedback behaviors on the candidate resource.

[0085] Specifically, the server performs a fusion process on the multiple prior distribution parameters of the multiple behavioral objectives of each candidate resource through a pre-deployed fusion strategy to obtain the resource evaluation result of each candidate resource. For example, the average value of the multiple prior distribution parameters can be obtained and used as the resource evaluation result.

[0086] In step S240, the recommended resources are determined from the candidate resources according to the resource evaluation result.

[0087] Specifically, the server sorts the candidate resources in descending or ascending order according to the resource evaluation result, and screens out the preset number of resources with the highest resource evaluation results from the sorted candidate resources. If the candidate resource is the resource to be roughly ranked output in the recall stage, the server can input the preset number of resources into the fine ranking model and repeat the process of steps S210 to S240 to obtain the resources that can be recommended. If the candidate resource is the resource to be finely ranked output in the rough ranking stage, the server can use the preset number of resources as the resources that can be recommended. The server sends the resources that can be recommended to the user account so that the client where the user account is located can display the resources that can be recommended.

[0088] In the above resource recommendation method, after determining the initial evaluation results of multiple behavioral objectives of candidate resources, the prior distribution of the historical evaluation results of each behavioral objective is obtained, and the prior distribution parameters corresponding to the initial evaluation results of each behavioral objective are determined according to the prior distribution. By mapping the initial evaluation results to prior distribution parameters, not only can the initial evaluation results with different dimensions be mapped to a unified data dimension, but also the discrimination information reflecting the differences between the initial evaluation results of different behavioral objectives can be well represented. Then, according to the prior distribution parameters of multiple behavioral objectives of candidate resources, a resource evaluation result of the candidate resources is generated, and the recommended resources are determined from the candidate resources according to the resource evaluation result, so that the resource evaluation result obtained based on the prior distribution can better fit the positive feedback degree of the user account to the candidate resources, which helps to improve the accuracy of resource recommendation and the overall index effect of the resource recommendation system.

[0089] In an exemplary embodiment, the prior distribution adopts a beta distribution. The beta distribution of each behavioral objective is generated according to the historical evaluation results of each behavioral objective, and the historical evaluation results are generated according to the historical behavior data of the user account for multiple recommended resources.

[0090] Among them, the beta distribution is a family of continuous probability distributions defined on the interval [0, 1], and it has two positive coefficients, which are respectively called the alpha (α) coefficient and the beta (β) coefficient.

[0091] The historical behavior data is used to reflect the behaviors of the user account for the recommended resources. For example, the playing duration, whether to like, whether to play, etc.

[0092] Specifically, whenever the server sends multiple recommended resources to the client, the server obtains the historical behavior data of the user account for each recommended resource reported by each client, and summarizes the historical behavior data of the multiple recommended resources to generate the historical evaluation results of each behavioral objective. For example, the recommended resource A is pushed to 10 user accounts. Among them, 5 user accounts have a like behavior for the recommended resource A, then the historical evaluation result corresponding to the behavioral objective "like rate" can be 0.5. The server obtains multiple historical evaluation results corresponding to each behavioral objective, and constructs the beta distribution of each behavioral objective according to the multiple historical evaluation results.

[0093] In this embodiment, by constructing a beta distribution according to the historical behavior data of the user account for the recommended resources, the beta distribution can reflect the more real behavior tendency of the user account, which helps to improve the prediction accuracy of the resource evaluation results of the resources.

[0094] In an exemplary embodiment, as Figure 3 shown, the generation method of the prior distribution of each behavioral objective includes:

[0095] In step S310, multiple historical evaluation results of each behavioral objective are obtained, and the expectation and variance are generated based on the multiple historical evaluation results of each behavioral objective.

[0096] In step S320, according to the expectation and variance, the alpha coefficient and beta coefficient of the beta distribution are generated.

[0097] In step S330, according to the alpha coefficient and beta coefficient, the cumulative distribution of the beta distribution of each behavioral objective is generated as the prior distribution of each behavioral objective.

[0098] Specifically, for each behavioral objective, after obtaining multiple historical evaluation results corresponding to each behavioral objective, the mean of the multiple historical evaluation results can be generated, and this mean is used as the expectation of the beta distribution. Further, the variance of the multiple historical evaluation results is generated based on the mean of the multiple historical evaluation results, and this variance is used as the variance of the beta distribution. Furthermore, the alpha coefficient and beta coefficient are calculated according to the expectation formula and variance formula of the beta distribution. The cumulative distribution function of the beta distribution is constructed based on the alpha times and beta coefficient, and the cumulative distribution function is used as the prior distribution of each behavioral objective.

[0099] In an example, the multiple historical evaluation results corresponding to a certain behavioral objective are (X 1 , X 2 ,..., X n ), where n is a positive integer greater than 1. Then it can be obtained that:

[0100] The expectation of the beta distribution is:

[0101] The variance of the beta distribution is:

[0102]

[0103] The alpha coefficient α and beta coefficient β are calculated according to the above expectation formula and variance formula.

[0104] Then, the probability density function f(x; α, β) of the beta distribution is generated based on the alpha coefficient α and beta coefficient β:

[0105]

[0106]

[0107] Finally, the cumulative distribution function F(x; α, β) of the beta distribution is generated based on the probability density function:

[0108]

[0109]

[0110] After obtaining the cumulative distribution function of the beta distribution for each behavioral objective, the initial evaluation result of each behavioral objective can be used as the independent variable x of the cumulative distribution function F(x; α, β), and then the cumulative distribution function value can be calculated. This cumulative distribution function value is used as the prior distribution parameter for each behavioral objective.

[0111] For example, assume that the behavioral objectives include the effective play rate and the short play rate. According to the above formula, the beta distribution of the effective play rate is calculated to be B(2, 2), and the beta distribution of the short play rate is B(2, 5). Figure 4 In (1), the probability density function f(x; 2, 2) of the effective play rate and the probability density function f(x; 2, 5) of the short play rate are exemplarily shown, as Figure 4 shown in (1), the area between the probability density function curve and the horizontal axis is equal to 1, which reflects the probability magnitude falling in different intervals. Figure 4 In (2), the cumulative distribution functions of the effective play rate and the short play rate are exemplarily shown. Figure 4 In (2), the cumulative distribution function F(x; 2, 2) of the effective play rate and the cumulative distribution function F(x; 2, 5) of the short play rate are exemplarily shown. As Figure 4 shown in (2), the cumulative distribution function is a monotonically increasing function, the function value range is 0 - 1, and the cumulative distribution function curve has the characteristic that the function value differentiation between the head and the tail is relatively large.

[0112] Figure 5 An exemplary schematic diagram of obtaining the prior distribution parameter according to the cumulative distribution function is shown. As Figure 5 shown, the cumulative distribution function can better reflect the differentiation degree of the differences between different evaluation results. The following takes a specific example to illustrate. Assume that the initial evaluation result of the effective play rate in the rough sorting stage is 0.5, and the initial evaluation result in the fine sorting stage is 0.51. The initial evaluation result of the short play rate in the rough sorting stage is 0.27, and the initial evaluation result in the fine sorting stage is 0.28. The difference magnitude between the effective play rate and the short play rate in both stages is 0.01. However, since the initial evaluation result value of the short play rate is relatively small, the differentiation degree of the difference magnitude of the short play rate is more significant. The fusion sorting method in the related art cannot accurately identify such a differentiation degree. In this embodiment, by mapping the initial evaluation result to the cumulative distribution function value of the beta distribution, as Figure 5 shown, the slope of the cumulative distribution function F(x; 2, 5) of the short play rate at the independent variable 0.27 is 1, which is greater than the slope of the cumulative distribution function F(x; 2, 2) of the effective play rate at the independent variable 0.5, indicating that the differentiation degree of the short play rate at 0.27 is greater than that of the effective play rate at 0.5.

[0113] In this embodiment, by using the beta distribution to map the initial evaluation results to the corresponding beta distribution parameters, not only can the initial evaluation results with different dimensions be mapped to a unified data dimension, but also the discrimination information of the differences between the initial evaluation results of different behavioral goals can be well reflected.

[0114] In an exemplary embodiment, another generation method of the beta distribution is described. In this embodiment, the alpha coefficient and the beta coefficient can be obtained based on the first moment and the second moment. In one example, multiple historical evaluation results corresponding to the behavioral goal are obtained as (X 1 , X 2 ,..., X n ), where n is a positive integer greater than 1. Then it can be obtained that:

[0115] The expectation of the beta distribution is:

[0116] where the first moment k = 1; the second moment k = 2. After calculating the alpha coefficient and the beta coefficient according to the above formula, the cumulative distribution function of the beta distribution can be constructed with reference to the above embodiment, which will not be elaborated here.

[0117] In this embodiment, by providing multiple generation methods of the beta distribution, the construction of the beta distribution of the behavioral goal is made more flexible, enriching the functions of the resource recommendation system.

[0118] In an exemplary embodiment, obtaining multiple historical evaluation results of each behavioral goal includes: obtaining multiple historical evaluation results of each behavioral goal within a preset time period before the current moment. Among them, the preset time period before the current moment can be one hour, one day, one month, etc. before the current moment, and the preset time period for obtaining historical evaluation results can be configured according to actual needs. The server generates the prior distribution of each behavioral goal based on the historical evaluation results within this preset time period. By adopting this method, the amount of data input can be reduced, thereby reducing the computing pressure on the server; in addition, the real-time nature and accuracy of the data can be ensured, improving the correlation degree between the prior distribution and the user account, and thus contributing to improving the accuracy of resource recommendation.

[0119] In an exemplary embodiment, the prior distribution of each behavioral goal is generated in real time after determining the candidate resources. That is, whenever the server determines the initial evaluation results of the candidate resources in response to a resource recommendation request, it then obtains multiple historical evaluation results corresponding to each behavioral goal, and generates the prior distribution of each behavioral goal based on the multiple historical evaluation results. The prior distribution constructed in this way can most reflect the current interests of the user account.

[0120] Alternatively, the prior distribution of each behavioral objective is periodically generated before determining the candidate resources. The period can be set according to actual requirements. For example, every hour, every day, every month, etc. That is, whenever the server determines that the current time meets the generation time of the prior distribution, it obtains multiple historical evaluation results corresponding to each behavioral objective, and generates the prior distribution of each behavioral objective based on the multiple historical evaluation results. Adopting this method can reduce the real-time data processing pressure on the server. When the server makes resource recommendations, it only needs to call the already generated prior distribution, which can also improve the resource recommendation efficiency.

[0121] In an exemplary embodiment, in step S230, according to the prior distribution parameters of multiple behavioral objectives of the candidate resources, a resource evaluation result of the candidate resources is generated, including: obtaining a regulation parameter corresponding to each behavioral objective; generating a resource evaluation result of the candidate resources according to the regulation parameter and the prior distribution parameter corresponding to each behavioral objective.

[0122] Among them, the regulation parameter can be used to reflect the importance of the behavioral objective. The more important the behavioral objective, the higher the regulation parameter can be configured. The regulation parameter can be a pre-configured constant; it can also be updated online in real time or offline regularly according to the current resource recommendation requirements. For example, the regulation parameter corresponding to each behavioral objective is predicted by a deep learning model based on historical recommendation records. The deep learning model can be any model capable of predicting the regulation parameter, such as a linear model, a neural network model, a support vector machine, a logistic regression model, etc.

[0123] Specifically, after the server obtains the prior distribution parameters corresponding to each behavioral objective, it obtains the regulation parameter corresponding to each behavioral objective. The resource evaluation result of the candidate resources can be generated by any one of the following methods:

[0124] (1) Obtain the weighted sum of the regulation parameters and the prior distribution parameters of multiple behavioral objectives, and use the weighted sum as the resource evaluation result. Assuming that the behavioral objectives include effective play rate, like rate,..., short play rate, then the resource evaluation result of each candidate resource is:

[0125] G(x) = G((F(effective play rate), F(like rate),..., F(short play rate))

[0126] = w 1 ×F(effective play rate) + w 2 ×F(like rate) +... + w n ×F(short play rate)

[0127] Among them, G(x) represents the resource evaluation result; F(effective play rate), F(like rate),..., F(short play rate) represent the prior distribution parameters; w n represents the regulation parameter.

[0128] (2) Obtain the power of the adjustment parameter of the prior distribution parameter of the behavior target, and generate a resource evaluation result according to the power of the adjustment parameter of the prior distribution parameter. Assume that the behavior targets include effective play rate, like rate,..., short play rate, then the resource evaluation result of each candidate resource can be:

[0129] G(x) = G((F(effective play rate), F(like rate),..., F(short play rate))

[0130] = F(effective play rate)^w 1 ×F(like rate)^w 2 ×...×F(short play rate)^w n

[0131] Alternatively, the resource evaluation result of each candidate resource can be:

[0132] G(x) = G((F(effective play rate), F(like rate),..., F(short play rate))

[0133] = F(effective play rate)^w 1 +F(like rate)^w 2 +...+F(short play rate)^w n

[0134] where G(x) represents the resource evaluation result; F(effective play rate), F(like rate),..., F(short play rate) represent the prior distribution parameters; w n represents the adjustment parameter.

[0135] In this embodiment, by deploying multiple selectable evaluation result fusion methods, the flexibility of resource recommendation can be improved.

[0136] Figure 6 is a flowchart of a resource recommendation method shown according to an exemplary embodiment. Taking the resource as a video as an example, as Figure 6 shown, the resource recommendation method is used in a server and includes the following steps.

[0137] In step S602, in response to a video recommendation request of a user account, recall multiple videos to be roughly ranked through a recall policy.

[0138] In step S604, input each video to be roughly ranked into a rough ranking model to obtain the initial rough ranking evaluation results of multiple behavior targets of each video to be roughly ranked.

[0139] In step S606, obtain multiple historical evaluation results of each behavior target within a preset time period before the current moment. Among them, the obtaining method of the multiple historical evaluation results can refer to the above embodiment and will not be specifically elaborated here.

[0140] In step S608, according to multiple historical evaluation results of each behavior target, a beta distribution of each behavior target is generated. Among them, the specific generation method of the beta distribution can refer to the above embodiments and will not be elaborated here.

[0141] In step S610, the initial rough ranking evaluation result of each behavior target is used as an independent variable and substituted into the cumulative distribution function of each behavior target to obtain the rough ranking cumulative distribution function value (i.e., the prior distribution parameter).

[0142] In step S612, according to the multiple rough ranking cumulative distribution function values corresponding to each video to be roughly ranked, and the rough ranking adjustment parameters corresponding to each rough ranking cumulative distribution function value, a rough ranking video evaluation result corresponding to each video to be roughly ranked is generated.

[0143] In step S614, the multiple videos to be roughly ranked are sorted in descending order according to the rough ranking video evaluation results, and a certain number of videos with the highest rankings are selected from the sorted multiple videos to be roughly ranked as the videos to be finely ranked.

[0144] In step S616, each video to be finely ranked is input into the fine ranking model to obtain the initial fine ranking evaluation results of multiple behavior targets of each video to be finely ranked.

[0145] In step S618, the initial fine ranking evaluation result of each behavior target is used as an independent variable and substituted into the cumulative distribution function of each behavior target to obtain the fine ranking cumulative distribution function value.

[0146] In step S620, according to the multiple fine ranking cumulative distribution function values corresponding to each video to be finely ranked, and the fine ranking adjustment parameters corresponding to each fine ranking cumulative distribution function value, a fine ranking video evaluation result corresponding to each video to be finely ranked is generated.

[0147] In step S622, the multiple videos to be finely ranked are sorted in descending order according to the fine ranking video evaluation results, and a certain number of videos with the highest rankings are selected from the sorted multiple videos to be finely ranked as the recommendable videos, and the recommendable videos are sent to the user account.

[0148] In step S624, the behavior data of the user account for the recommendable videos is updated to the online recommendation record, and the beta distribution of each behavior target is dynamically updated according to the online recommendation record.

[0149] It should be understood that although the steps in the above flowcharts are shown sequentially according to the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless explicitly stated in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, at least some of the steps in the above flowcharts may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least some of the steps or stages in other steps or other steps.

[0150] It can be understood that the same / similar parts among the various embodiments of the above methods in this specification can be referred to each other. Each embodiment focuses on the differences from other embodiments. For the relevant parts, refer to the descriptions of other method embodiments.

[0151] Figure 7 It is a block diagram of a resource recommendation device 700 shown according to an exemplary embodiment. Refer to Figure 7 , the device includes an initial result determination module 702, a prior parameter determination module 704, a resource result generation module 706, and a recommendation module 708.

[0152] The initial result determination module 702 is configured to execute to determine an initial evaluation result of multiple behavioral objectives of candidate resources; the prior parameter determination module 704 is configured to execute to obtain a prior distribution of the historical evaluation results of each behavioral objective, and determine prior distribution parameters corresponding to the initial evaluation results of each behavioral objective according to the prior distribution; the resource result generation module 706 is configured to execute to generate a resource evaluation result of candidate resources according to the prior distribution parameters of multiple behavioral objectives of candidate resources; the recommendation module 708 is configured to execute to determine recommended resources from candidate resources according to the resource evaluation results for resource recommendation.

[0153] In an exemplary embodiment, the prior distribution adopts a beta distribution. The beta distribution of each behavioral objective is generated according to the historical evaluation results of each behavioral objective, and the historical evaluation results are generated according to the historical behavioral data of the user account for multiple recommended resources.

[0154] In an exemplary embodiment, the apparatus 700 further includes: a historical result acquisition module configured to acquire multiple historical evaluation results of each behavior target; an expectation and variance generation module configured to generate an expectation and a variance according to the multiple historical evaluation results of each behavior target; a coefficient generation module configured to generate an alpha coefficient and a beta coefficient of a beta distribution according to the expectation and the variance; and a beta distribution generation module configured to generate a cumulative distribution of the beta distribution of each behavior target according to the alpha coefficient and the beta coefficient as a prior distribution of each behavior target.

[0155] In an exemplary embodiment, the historical result acquisition module is configured to acquire multiple historical evaluation results of each behavior target within a preset time period before the current moment.

[0156] In an exemplary embodiment, the prior distribution of each behavior target is generated after determining candidate resources; or, the prior distribution of each behavior target is periodically generated before determining candidate resources.

[0157] In an exemplary embodiment, the resource result generation module 706 is configured to perform operations including: a regulation parameter acquisition unit configured to acquire a regulation parameter corresponding to each behavior target; and a resource result generation unit configured to generate a resource evaluation result of candidate resources according to the regulation parameter corresponding to each behavior target and the prior distribution parameter.

[0158] In an exemplary embodiment, the resource result generation unit is configured to obtain the resource evaluation result of candidate resources in any one of the following manners: obtain a weighted sum of the regulation parameters and the prior distribution parameters of multiple behavior targets and use the weighted sum as the resource evaluation result; obtain a power of the regulation parameter of the prior distribution parameter of the behavior target and generate a resource evaluation result according to the power of the regulation parameter of the prior distribution parameter.

[0159] Regarding the apparatus in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated herein.

[0160] Figure 8 is a block diagram of an electronic device S00 for recommending resources to a terminal according to an exemplary embodiment. For example, the electronic device S00 may be a server. Refer to Figure 8, the electronic device S00 includes a processing component S20, which further includes one or more processors, and memory resources represented by a memory S22 for storing instructions executable by the processing component S20, such as application programs. The application programs stored in the memory S22 may include one or more modules each corresponding to a set of instructions. In addition, the processing component S20 is configured to execute instructions to perform the above method.

[0161] The electronic device S00 may further include: a power supply component S24 configured to perform power management of the electronic device S00, a wired or wireless network interface S26 configured to connect the electronic device S00 to a network, and an input / output (I / O) interface S28. The electronic device S00 may operate based on an operating system stored in the memory S22, such as Windows Server, Mac OSX, Unix, Linux, FreeBSD or the like.

[0162] In an exemplary embodiment, there is also provided a computer-readable storage medium including instructions, such as the memory S22 including instructions, and the above instructions can be executed by the processor of the electronic device S00 to complete the above method. The storage medium may be a computer-readable storage medium. For example, the computer-readable storage medium may be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0163] In an exemplary embodiment, there is also provided a computer program product, which includes instructions that can be executed by the processor of the electronic device S00 to complete the above method.

[0164] It should be noted that the above-mentioned device, electronic device, computer-readable storage medium, computer program product, etc. may also include other embodiments according to the description of the method embodiments. The specific implementation manners may refer to the description of the relevant method embodiments and will not be elaborated here one by one.

[0165] Those skilled in the art will readily conceive of other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common general knowledge or conventional technical means in the technical field not disclosed by the present disclosure. The specification and embodiments are only to be considered as exemplary, and the true scope and spirit of the present disclosure are pointed out by the claims.

[0166] It should be understood that the present disclosure is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present disclosure is only limited by the appended claims.

Claims

1. A resource recommendation method, characterized in that, it includes: Determine the initial evaluation results of multiple behavioral objectives of the candidate resources, where the behavioral objectives include positive feedback duration type objectives, positive feedback interaction type behavioral objectives, negative feedback duration type objectives, and negative feedback interaction type behavioral objectives; Obtain multiple historical evaluation results of each of the behavioral objectives, and generate an expectation and a variance based on the multiple historical evaluation results of each of the behavioral objectives; Generate the alpha coefficient and beta coefficient of the beta distribution according to the expectation and the variance; Generate the probability density function of the beta distribution of each of the behavioral objectives according to the alpha coefficient and the beta coefficient, generate the cumulative distribution function of the beta distribution of each of the behavioral objectives according to the probability density function, and use the cumulative distribution function as the prior distribution of each of the behavioral objectives, where the cumulative distribution function is used to reflect the discrimination degree of the differences between different evaluation results; Map the initial evaluation result of each of the behavioral objectives as the independent variable of the cumulative distribution function to the prior distribution corresponding to each of the behavioral objectives, calculate the cumulative distribution function value, and use the cumulative distribution function value as the prior distribution parameter corresponding to the initial evaluation result of each of the behavioral objectives; Generate the resource evaluation result of the candidate resources according to the prior distribution parameters of the multiple behavioral objectives of the candidate resources; Determine the recommended resources from the candidate resources according to the resource evaluation result, and the recommended resources are used for resource recommendation.

2. The method according to claim 1, characterized in that, The prior distribution adopts the beta distribution, and the beta distribution of each of the behavioral objectives is generated according to the historical evaluation results of each of the behavioral objectives, and the historical evaluation results are generated according to the historical behavior data of the user account for multiple recommended resources.

3. The method according to claim 1, characterized in that, The obtaining of multiple historical evaluation results of each of the behavioral objectives includes: Obtain multiple historical evaluation results of each of the behavioral objectives within a preset time period before the current moment.

4. The method according to claim 1, characterized in that, The prior distribution of each of the behavioral objectives is generated after determining the candidate resources; Or, the prior distribution of each of the behavioral objectives is generated periodically before determining the candidate resources.

5. The method according to claim 1, characterized in that, The generating of the resource evaluation result of the candidate resources according to the prior distribution parameters of the multiple behavioral objectives of the candidate resources includes: Obtain the adjustment parameter corresponding to each of the behavioral objectives; Generate the resource evaluation result of the candidate resources according to the adjustment parameter corresponding to each of the behavioral objectives and the prior distribution parameter.

6. The method according to claim 5, characterized in that, The generating of the resource evaluation result of the candidate resources according to the adjustment parameter corresponding to each of the behavioral objectives and the prior distribution parameter includes: Obtain the resource evaluation result of the candidate resources by any one of the following methods: Obtain the weighted sum of the adjustment parameters of the multiple behavioral targets and the prior distribution parameters, and use the weighted sum as the resource evaluation result; Obtain the power of the adjustment parameter of the prior distribution parameter of the behavioral target, and generate the resource evaluation result according to the power of the adjustment parameter of the prior distribution parameter.

7. A resource recommendation device, Characterized in that, Comprising: An initial result determination module, configured to execute and determine the initial evaluation results of multiple behavioral targets of candidate resources, where the behavioral targets include positive feedback duration type targets, positive feedback interaction type behavioral targets, negative feedback duration type targets, and negative feedback interaction type behavioral targets; A historical result acquisition module, configured to execute and obtain multiple historical evaluation results of each of the behavioral targets; An expectation and variance generation module, configured to execute and generate an expectation and a variance according to the multiple historical evaluation results of each of the behavioral targets; A coefficient generation module, configured to execute and generate an alpha coefficient and a beta coefficient of a beta distribution according to the expectation and the variance; A beta distribution generation module, configured to execute and generate a probability density function of the beta distribution of each of the behavioral targets according to the alpha coefficient and the beta coefficient, generate a cumulative distribution function of the beta distribution of each of the behavioral targets according to the probability density function, and use the cumulative distribution function as the prior distribution of each of the behavioral targets, where the cumulative distribution function is used to reflect the discrimination degree of the differences between different evaluation results; A prior parameter determination module, configured to execute and obtain the prior distribution of the historical evaluation results of each of the behavioral targets, map the initial evaluation result of each of the behavioral targets to the prior distribution corresponding to each of the behavioral targets as the independent variable of the cumulative distribution function, calculate the cumulative distribution function value, and use the cumulative distribution function value as the prior distribution parameter corresponding to the initial evaluation result of each of the behavioral targets; A resource result generation module, configured to execute and generate a resource evaluation result of the candidate resource according to the prior distribution parameters of the multiple behavioral targets of the candidate resource; A recommendation module, configured to execute and determine the recommended resources from the candidate resources according to the resource evaluation result, where the recommended resources are used for resource recommendation.

8. The device according to claim 7, Characterized in that, The prior distribution adopts a beta distribution, and the beta distribution of each of the behavioral targets is generated according to the historical evaluation results of each of the behavioral targets, and the historical evaluation results are generated according to the historical behavior data of the user account for multiple recommended resources.

9. The device according to claim 7, Characterized in that, The historical result acquisition module is configured to execute and obtain multiple historical evaluation results of each of the behavioral targets within a preset time period before the current moment.

10. The device according to claim 7, Characterized in that, The prior distribution of each of the behavioral targets is generated after the candidate resources are determined; Alternatively, the prior distribution of each of the behavioral targets is periodically generated before the candidate resources are determined.

11. The apparatus according to claim 7, wherein, the resource result generation module is configured to perform, including: a regulation parameter acquisition unit configured to acquire regulation parameters corresponding to each of the behavior objectives; a resource result generation unit configured to generate a resource evaluation result of the candidate resource according to the regulation parameters corresponding to each of the behavior objectives and the prior distribution parameters.

12. The apparatus according to claim 11, wherein, the resource result generation unit is configured to perform: obtaining the resource evaluation result of the candidate resource in any one of the following manners: obtaining a weighted sum of the regulation parameters and the prior distribution parameters of the plurality of behavior objectives, and using the weighted sum as the resource evaluation result; obtaining a regulation parameter power of the prior distribution parameter of the behavior objective, and generating the resource evaluation result according to the regulation parameter power of the prior distribution parameter.

13. An electronic device, wherein, it includes: a processor; a memory for storing executable instructions of the processor; wherein, the processor is configured to execute the instructions to implement the resource recommendation method according to any one of claims 1 to 6.

14. A computer-readable storage medium, wherein, when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the resource recommendation method according to any one of claims 1 to 6.

15. A computer program product including instructions, wherein, when the instructions are executed by a processor of an electronic device, the electronic device is enabled to execute the resource recommendation method according to any one of claims 1 to 6.

Images