Processing method and device of policy model and computer readable storage medium

Abstract

The application discloses a kind of processing methods of strategy model, device and computer readable storage medium, it is related to the field of financial technology and other related technical fields.Therein, the method comprises: obtaining a plurality of strategy models and the plurality of evaluation values corresponding to each strategy model, wherein each evaluation value corresponds to an evaluation index;According to evaluation value, target matrix is constructed;The target evaluation value corresponding to each strategy model is determined by target matrix, wherein the target evaluation value is used to represent the comprehensive score of each strategy model under a plurality of evaluation indexes;According to the target evaluation value corresponding to each strategy model, determine the target strategy model from the plurality of strategy models, and recommend commodity information to user by target strategy model.The application solves the technical problem of poor recommendation accuracy in the prior art when recommending commodity information to user by strategy model.

Description

Technical Field

[0001] This application relates to the fields of financial technology and other related technologies, and more specifically, to a method, apparatus and computer-readable storage medium for processing a strategy model. Background Technology

[0002] With the continuous development of asset management products in financial services, recommending wealth management products such as funds and insurance to users has become a popular practice. Current technologies typically use strategy models with recommendation functions to recommend relevant wealth management products to users. Building upon this, the key is to determine the strategy model with the best recommendation performance from among multiple existing models as the target strategy model, and then use this target strategy model in practical applications to recommend relevant wealth management products to users.

[0003] However, when determining which strategy model performs best among multiple strategy models, existing technologies typically use a single evaluation metric to assess the recommendation performance of each strategy model. Since this evaluation method measures the recommendation performance of a strategy model from a single dimension, it leads to an incomplete assessment of the strategy model's recommendation performance. Consequently, the final target strategy model may only perform well under one evaluation metric, while performing poorly under other evaluation metrics. This results in poor recommendation accuracy when using the target strategy model to recommend financial products and other product information to users. Summary of the Invention

[0004] This application provides a method, apparatus, and computer-readable storage medium for processing a strategy model, to at least solve the technical problem of poor recommendation accuracy in the prior art when recommending product information to users through a strategy model.

[0005] According to one aspect of the embodiments of this application, a method for processing strategy models is provided, comprising: obtaining multiple strategy models and multiple evaluation values ​​corresponding to each strategy model, wherein each evaluation value corresponds to an evaluation index; constructing a target matrix based on the evaluation values, wherein in the target matrix, the values ​​in each row are the evaluation values ​​of a strategy model under different evaluation indexes, and the values ​​in each column are the evaluation values ​​of multiple strategy models under the same evaluation index; determining a target evaluation value corresponding to each strategy model through the target matrix, wherein the target evaluation value is used to characterize the comprehensive score of each strategy model under multiple evaluation indexes; determining a target strategy model from the multiple strategy models based on the target evaluation value corresponding to each strategy model, and recommending product information to the user through the target strategy model.

[0006] Furthermore, the processing method for the strategy model also includes: during the training process of each strategy model, obtaining multiple first evaluation values ​​corresponding to each strategy model, wherein each first evaluation value corresponds to a first-type evaluation index, and the first-type evaluation index is the evaluation index of each strategy model during the training phase; after each strategy model is deployed online, obtaining multiple second evaluation values ​​corresponding to each strategy model, wherein each second evaluation value corresponds to a second-type evaluation index, and the second-type evaluation index is the evaluation index of each strategy model during the application phase.

[0007] Furthermore, the processing method of the strategy model also includes: randomly determining a preset number of target users from multiple users; recommending target product information to target users through each strategy model; and determining multiple second evaluation values ​​corresponding to each strategy model based on the target product information.

[0008] Furthermore, the processing method of the strategy model also includes: obtaining the user characteristics of each user among multiple users; dividing the multiple users into at least one user set according to the user characteristics, wherein the users in each user set have the same user characteristics; determining a preset number of target users from each user set; recommending target product information to the target users through each strategy model; and determining multiple second evaluation values ​​corresponding to each strategy model based on the target product information.

[0009] Furthermore, the processing method for the strategy model also includes: performing singular value decomposition on the target matrix to obtain the left singular matrix, right singular matrix, and diagonal matrix corresponding to the target matrix; and determining the target evaluation value corresponding to each strategy model based on the left singular matrix, right singular matrix, and diagonal matrix.

[0010] Furthermore, the processing method for the policy model also includes: determining the target left singular vector based on the right singular matrix and the diagonal matrix; determining the target right singular vector based on the left singular matrix and the diagonal matrix; determining the target singular value based on the diagonal matrix; and determining the target evaluation value corresponding to each policy model based on the target left singular vector, the target right singular vector, and the target singular value.

[0011] Furthermore, the processing method of the policy model also includes: obtaining multiple components in the left singular vector of the target; calculating the product of the right singular vector of the target and the singular value of the target to obtain a first value; multiplying each component in the left singular vector of the target by the first value to obtain the target evaluation value corresponding to each policy model, wherein one target evaluation value corresponds to one component.

[0012] Furthermore, the processing method for the strategy model also includes: after determining the target evaluation value corresponding to each strategy model based on the left singular matrix, the right singular matrix, and the diagonal matrix, adjusting at least one component in the target right singular vector to obtain a first right singular vector, wherein each component in the target right singular vector corresponds to an evaluation index; calculating the product of the first right singular vector and the target singular value to obtain a second value; and multiplying each component in the target left singular vector by the second value to obtain the latest evaluation value corresponding to each strategy model, wherein the latest evaluation value is the updated target evaluation value.

[0013] According to another aspect of the embodiments of this application, a processing apparatus for strategy models is also provided, comprising: an acquisition module, configured to acquire multiple strategy models and multiple evaluation values ​​corresponding to each strategy model, wherein each evaluation value corresponds to an evaluation index; a construction module, configured to construct a target matrix based on the evaluation values, wherein in the target matrix, the values ​​in each row are the evaluation values ​​of a strategy model under different evaluation indexes, and the values ​​in each column are the evaluation values ​​of multiple strategy models under the same evaluation index; a first determination module, configured to determine the target evaluation value corresponding to each strategy model through the target matrix, wherein the target evaluation value is used to characterize the comprehensive score of each strategy model under multiple evaluation indexes; and a second determination module, configured to determine a target strategy model from the multiple strategy models based on the target evaluation value corresponding to each strategy model, and recommend product information to the user through the target strategy model.

[0014] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, which stores a computer program, wherein the computer program is configured to execute the processing method of the above-described strategy model at runtime.

[0015] According to another aspect of the embodiments of this application, an electronic device is also provided, which includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by one or more processors, the one or more processors cause the one or more processors to implement the processing method of the above-described strategy model.

[0016] The technical solution of this application adopts a method of determining the target evaluation value of each strategy model under multiple evaluation indicators, and determining the target strategy model from multiple strategy models based on the target evaluation value. First, multiple strategy models and their corresponding evaluation values ​​are obtained. Then, a target matrix is ​​constructed based on the evaluation values, and the target evaluation value for each strategy model is determined using the target matrix. Finally, the target strategy model is determined from multiple strategy models based on the target evaluation value for each strategy model, and product information is recommended to the user using the target strategy model. Each evaluation value corresponds to one evaluation indicator; in the target matrix, each row contains the evaluation value of a strategy model under different evaluation indicators, and each column contains the evaluation value of multiple strategy models under the same evaluation indicator; the target evaluation value is used to characterize the comprehensive score of each strategy model under multiple evaluation indicators.

[0017] As described above, this application constructs a target matrix and determines the target evaluation value for each strategy model using this matrix. Since the target evaluation value represents the comprehensive score of each strategy model under multiple evaluation indicators, it is actually an evaluation result obtained after comprehensively evaluating each strategy model from multiple dimensions. This avoids the problem of incomplete evaluation that exists when using a single evaluation indicator to evaluate strategy models in existing technologies. Furthermore, this application determines a target strategy model from multiple strategy models based on the target evaluation value corresponding to each strategy model. Since the target evaluation value is a comprehensive score under multiple evaluation indicators, the target strategy model is the strategy model with the best overall performance among the multiple strategy models. In other words, the target strategy model is the strategy model with the best overall recommendation effect among the multiple strategy models. Recommending product information to users using the target strategy model can improve the accuracy of product information recommendations, thereby improving the user experience.

[0018] Therefore, the technical solution of this application achieves the goal of comprehensively evaluating the strategy model based on multiple evaluation indicators, thereby improving the technical effect of improving the evaluation accuracy of the strategy model and solving the technical problem of poor recommendation accuracy in the prior art when recommending product information to users through the strategy model. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0020] Figure 1 This is a flowchart of an optional strategy model processing method according to an embodiment of this application;

[0021] Figure 2This is a schematic diagram of a processing apparatus for an optional strategy model according to an embodiment of this application;

[0022] Figure 3 This is a schematic diagram of an optional electronic device according to an embodiment of this application. Detailed Implementation

[0023] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] Furthermore, it should be noted that all information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) disclosed herein are information and data authorized by the user or fully authorized by all parties. For example, this system has interfaces with relevant users or organizations. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or organization through the interface, and obtain the relevant information only after receiving consent from the aforementioned user or organization.

[0026] Example 1

[0027] According to an embodiment of this application, a processing method embodiment for a strategy model is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0028] In addition, it should be noted that a strategy model processing system can serve as the execution entity of the strategy model processing method in the embodiments of this application.

[0029] Figure 1 This is a flowchart of an optional strategy model processing method according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0030] Step S101: Obtain multiple policy models and multiple evaluation values ​​corresponding to each policy model.

[0031] In step S101, each evaluation value corresponds to an evaluation metric. The aforementioned strategy model can be a recommendation strategy model, which has the function of recommending product information to users. The strategy model includes, but is not limited to, mathematical models and neural network models. The aforementioned evaluation metrics are indicators defined for evaluating the merits of the strategy model. These multiple evaluation metrics include at least: Accuracy, Precision, Recall, Balanced F-score (F1-score), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), Click-Through Rate (CTR), Purchase Rate (CVR), Novelty, and Diversity.

[0032] Specifically, precision, as mentioned above, is the proportion of correctly classified positive samples out of the total number of samples identified as positive by the policy model during training; recall, as mentioned above, is the proportion of correctly classified positive samples out of the total number of true positive samples during training; and accuracy, as mentioned above, represents the proportion of correctly classified samples out of the total number of samples during training. The formula for calculating accuracy is:

[0033]

[0034] Where, n in the formula correct n is the number of samples that are correctly classified. total This represents the total number of samples.

[0035] Secondly, the balanced F-score (F1-score) mentioned above can be understood as the harmonic mean of precision and recall. Precision and recall are typically contradictory concepts; a high precision often leads to a relatively low recall. To balance these two, the balanced F-score (F1-score) is defined as follows:

[0036]

[0037] Furthermore, the root mean square error (RMSE) is defined as follows:

[0038]

[0039] Among them, y i It is the true value of the i-th sample. is the predicted value of the i-th sample, and n is the number of sample points.

[0040] Finally, the click-through rate (CTR) mentioned above refers to the percentage of products that a user actually clicks on and views in the product list recommended by the strategy model; the purchase rate (CVR) mentioned above refers to the percentage of products that a user actually purchases in the product list recommended by the strategy model; novelty is used to measure the novelty of the product list recommended by the strategy model to the user, defined as the percentage of products that the user has never seen before in the recommended product list; diversity is used to measure the diversity of the product list recommended by the strategy model to the user, defined as the ratio between the number of product types recommended by the strategy model to the total number of existing product types.

[0041] It should be noted that this application obtains an evaluation value for each strategy model under each evaluation indicator, thereby obtaining multiple evaluation values. In the subsequent strategy model evaluation process, the strategy model can be comprehensively evaluated from multiple dimensions, improving the comprehensiveness of the strategy model evaluation.

[0042] Step S102: Construct the target matrix based on the evaluation values.

[0043] In step S102, in the target matrix, the value in each row is the evaluation value of a strategy model under different evaluation indicators, and the value in each column is the evaluation value of multiple strategy models under the same evaluation indicator.

[0044] Specifically, assuming there are n strategy models and multiple evaluation metrics including the aforementioned Accuracy, Precision, Recall, F1-score, Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), Click-Through Rate (CTR), Purchase Rate (CVR), Novelty, and Diversity—a total of 10 evaluation metrics—each strategy model will have 10 corresponding evaluation values ​​based on these 10 metrics. That is, one evaluation value is obtained for each metric. Based on the 10 evaluation values ​​corresponding to each strategy model, the strategy model processing system can construct the following target matrix R:

[0045]

[0046] Where n is the total number of strategy models, Model 1, Model 2 and Model 3 in the target matrix are all strategy models, and Indicator 1, Indicator 2 and Indicator 10 are the indicators among the 10 evaluation indicators mentioned above.

[0047] Step S103: Determine the target evaluation value corresponding to each strategy model through the target matrix.

[0048] In step S103, the target evaluation value is used to characterize the comprehensive score of each strategy model under multiple evaluation indicators.

[0049] Optionally, after obtaining the target matrix, the policy model processing system can perform singular value decomposition on the target matrix to obtain the left singular matrix, right singular matrix, and diagonal matrix corresponding to the target matrix, and then determine the target evaluation value corresponding to each policy model based on the left singular matrix, right singular matrix, and diagonal matrix.

[0050] Specifically, after obtaining the target matrix, the strategy model processing system can perform Singular Value Decomposition (SVD) on the target matrix to obtain the left and right singular vectors and the diagonal matrix, and then obtain the singular values ​​from the diagonal matrix. By combining the singular values ​​and the left and right singular vectors, the strategy model processing system can fuse multiple evaluation metrics to form a final comprehensive index used to evaluate the merits of a recommendation strategy. In other words, by combining the singular values ​​and the left and right singular vectors, the strategy model processing system can obtain a comprehensive score for each strategy model under multiple evaluation metrics.

[0051] Step S104: Determine the target strategy model from multiple strategy models based on the target evaluation value corresponding to each strategy model, and recommend product information to the user through the target strategy model.

[0052] In step S104, since the application scenarios of the strategy model are different, the above-mentioned product information can also be various types of product information. For example, in the financial scenario, the above-mentioned product information can be information on financial products such as funds, insurance, and stocks. In the daily consumption scenario, the above-mentioned product information can be information on consumer goods such as household goods, clothing, and food.

[0053] In addition, when determining the target strategy model from multiple strategy models based on the target evaluation value, the strategy model processing system can sort the multiple target evaluation values ​​to obtain the sorting results, and then determine the strategy model with the highest target evaluation value from multiple strategy models as the target strategy model based on the sorting results.

[0054] It is important to note that since the target evaluation value is a comprehensive score of each strategy model under multiple evaluation indicators, the target evaluation value is actually an evaluation result after evaluating the strategy model from multiple dimensions. Compared with the existing technology that only uses a single evaluation indicator to evaluate the strategy model, the target evaluation value is a more comprehensive evaluation result of the strategy model. Therefore, the target strategy model determined by the strategy model processing system based on the target evaluation value is the model with the best recommendation effect among multiple strategy models. Thus, using the target strategy model to recommend product information to users can improve the accuracy of product information recommendation, thereby improving the user experience and increasing the sales growth of related products.

[0055] Based on steps S101 to S104, the technical solution of this application adopts a method of determining the target evaluation value of each strategy model under multiple evaluation indicators, and determining the target strategy model from multiple strategy models based on the target evaluation value. First, multiple strategy models and multiple evaluation values ​​corresponding to each strategy model are obtained. Then, a target matrix is ​​constructed based on the evaluation values, and the target evaluation value corresponding to each strategy model is determined through the target matrix. Finally, the target strategy model is determined from multiple strategy models based on the target evaluation value corresponding to each strategy model, and product information is recommended to the user through the target strategy model. Each evaluation value corresponds to one evaluation indicator; in the target matrix, the values ​​in each row are the evaluation values ​​of a strategy model under different evaluation indicators, and the values ​​in each column are the evaluation values ​​of multiple strategy models under the same evaluation indicator; the target evaluation value is used to characterize the comprehensive score of each strategy model under multiple evaluation indicators.

[0056] As described above, this application constructs a target matrix and determines the target evaluation value for each strategy model using this matrix. Since the target evaluation value represents the comprehensive score of each strategy model under multiple evaluation indicators, it is actually an evaluation result obtained after comprehensively evaluating each strategy model from multiple dimensions. This avoids the problem of incomplete evaluation that exists when using a single evaluation indicator to evaluate strategy models in existing technologies. Furthermore, this application determines a target strategy model from multiple strategy models based on the target evaluation value corresponding to each strategy model. Since the target evaluation value is a comprehensive score under multiple evaluation indicators, the target strategy model is the strategy model with the best overall performance among the multiple strategy models. In other words, the target strategy model is the strategy model with the best overall recommendation effect among the multiple strategy models. Recommending product information to users using the target strategy model can improve the accuracy of product information recommendations, thereby improving the user experience.

[0057] Therefore, the technical solution of this application achieves the goal of comprehensively evaluating the strategy model based on multiple evaluation indicators, thereby improving the technical effect of improving the evaluation accuracy of the strategy model and solving the technical problem of poor recommendation accuracy in the prior art when recommending product information to users through the strategy model.

[0058] In one optional embodiment, the strategy model processing system acquires multiple first evaluation values ​​for each strategy model during the training process. Each first evaluation value corresponds to a first-type evaluation metric, which is the evaluation metric for each strategy model during the training phase. Simultaneously, after each strategy model is deployed online, the system acquires multiple second evaluation values ​​for each strategy model. Each second evaluation value corresponds to a second-type evaluation metric, which is the evaluation metric for each strategy model during the application phase.

[0059] Specifically, the aforementioned first-category evaluation metrics include at least the following: Accuracy, Precision, Recall, Balanced F-score (F1-score), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). For these six evaluation metrics, the relevant data for calculating the evaluation values ​​is generated during the training phase of the policy model. The policy model processing system can calculate the first evaluation value based on the calculation formula corresponding to each first-category evaluation metric and the relevant data.

[0060] In addition, the second category of evaluation indicators mentioned above includes at least the click-through rate (CTR), purchase rate (CVR), novelty, and diversity.

[0061] In an optional embodiment, for the second type of evaluation index mentioned above, the strategy model processing system can sample the recommendation results of the strategy model using random sampling or stratified sampling, and calculate the second evaluation value based on the sampling results.

[0062] Specifically, when sampling the recommendation results of the strategy model using random sampling, the strategy model processing system first randomly determines a preset number of target users from multiple users, then recommends target product information to the target users through each strategy model, and finally determines multiple second evaluation values ​​corresponding to each strategy model based on the target product information.

[0063] For example, for each strategy model, 20% of users can be randomly selected from a pool of existing fixed users as target users. Then, the strategy model recommends target product information to these target users. In other words, each strategy model recommends a product list to the target users. Based on the product list and the calculation method for each of the second-type evaluation metrics mentioned above, the strategy model can calculate a second evaluation value corresponding to each target user. For instance, if strategy model A recommends a product list to 10 target users, then under the second-type evaluation metric of novelty, strategy model A can obtain 10 novelty evaluation values, meaning each novelty evaluation value corresponds to one target user. Finally, the strategy model processing system calculates the average of the 10 novelty evaluation values ​​and uses this average as the second evaluation value of strategy model A under the second-type evaluation metric of novelty.

[0064] Furthermore, when sampling the recommendation results of the strategy model using stratified sampling, the strategy model processing system first obtains the user characteristics of each user among multiple users. Then, based on the user characteristics, the multiple users are divided into at least one user set, and a preset number of target users are determined from each user set. Finally, the strategy model processing system recommends target product information to the target users through each strategy model and determines multiple second evaluation values ​​corresponding to each strategy model based on the target product information. Users in each user set have the same user characteristics.

[0065] Because different user groups may have varying degrees of acceptance of the same strategy model, the strategy model processing system pre-acquires the user characteristics of each user among multiple users to eliminate these biases. Then, based on these characteristics, the system divides the multiple users into at least one user set—essentially, selecting users based on their characteristics. For example, user characteristics could include age, gender, total assets, and cash flow. The system can group users with similar characteristics into the same user set. After obtaining at least one user set, the system can determine a predetermined number of target users from each set. For instance, it might randomly select 20% of the users in each set as target users. After obtaining the target users, the system recommends target product information to them using the strategy model. The process of determining multiple second evaluation values ​​corresponding to each strategy model based on the target product information is the same as the process described above for calculating the second evaluation values ​​based on random sampling results, and will not be elaborated further here.

[0066] In one optional embodiment, the policy model processing system determines the target evaluation value corresponding to each policy model through the target matrix. Specifically, the policy model processing system first performs singular value decomposition on the target matrix to obtain the left singular matrix, right singular matrix, and diagonal matrix corresponding to the target matrix, and then determines the target evaluation value corresponding to each policy model based on the left singular matrix, right singular matrix, and diagonal matrix.

[0067] Optionally, when determining the target evaluation value corresponding to each policy model based on the left singular matrix, right singular matrix, and diagonal matrix, the policy model processing system first determines the target left singular vector based on the right singular matrix and diagonal matrix, then determines the target right singular vector based on the left singular matrix and diagonal matrix, and finally determines the target singular value based on the diagonal matrix. Then, the policy model processing system determines the target evaluation value corresponding to each policy model based on the target left singular vector, target right singular vector, and target singular value.

[0068] Specifically, taking the target matrix R mentioned above as an example, the policy model processing system first performs singular value decomposition on the target matrix. The decomposition process is shown in the following formula:

[0069] R=U∑V T

[0070] Where U is called the n*10 left singular matrix, V is called the 10*10 right singular matrix, and ∑ is a diagonal matrix composed of singular values ​​arranged in descending order. The singular values ​​are based on matrix R. T The result of taking the square root of the eigenvalues ​​of R, with the superscript T indicating the transpose of the target matrix R. Each column of the left singular matrix U is called a left singular vector, and each column of the right singular matrix V is called a right singular vector.

[0071] After obtaining the left singular matrix U, the right singular matrix V, and the diagonal matrix ∑, the policy model processing system can determine the first singular value σ1 in the diagonal matrix ∑ as the target singular value. It then determines the target left singular vector u1 based on the right singular matrix V and the diagonal matrix ∑, and the target right singular vector v1 based on the left singular matrix U and the diagonal matrix ∑. It's important to note that multiple left singular vectors can be calculated using the right singular matrix V and the diagonal matrix ∑. Typically, the policy model processing system selects the first left singular vector as the target left singular vector u1. Similarly, multiple right singular vectors can be calculated using the left singular matrix U and the diagonal matrix ∑. Typically, the policy model processing system selects the first right singular vector as the target right singular vector v1.

[0072] Furthermore, according to matrix computation theory, the strategy model processing system can obtain the recovery matrix E by calculating the product of σ1u1v1, with an error of ||R-σ1u1v1||. F =σ2.

[0073] In one optional embodiment, after obtaining the target right singular vector, the target left singular vector, and the target singular value, the policy model processing system further obtains multiple components in the target left singular vector and calculates the product of the target right singular vector and the target singular value to obtain a first value. Finally, the policy model processing system multiplies each component in the target left singular vector with the first value to obtain the target evaluation value corresponding to each policy model, wherein one target evaluation value corresponds to one component.

[0074] Optionally, still taking the target matrix R as an example, after obtaining the target right singular vector v1, the target left singular vector u1, and the target singular value σ1, the policy model processing system calculates the product of σ1v1 as a weighted weight vector, and obtains multiple components in the target left singular vector u1.

[0075] For policy model 1 corresponding to the target matrix R, the policy model processing system determines that the target evaluation value corresponding to policy model 1 is the product of the first component of the target left singular vector u1 and σ1v1, i.e., u 11 ·σ1v1;

[0076] For policy model 2 corresponding to the target matrix R, the policy model processing system determines that the target evaluation value corresponding to policy model 2 is the product of the second component of the target left singular vector u1 and σ1v1, i.e., u 12 ·σ1v1;

[0077] Similarly, for the policy model n corresponding to the target matrix R, the policy model processing system determines that the target evaluation value corresponding to the policy model n is the product of the nth component of the target left singular vector u1 and σ1v1, i.e., u 1n ·σ1v1.

[0078] Through the above process, the strategy model processing system can obtain the target evaluation value corresponding to each strategy model. Since the target evaluation value is the comprehensive score of each strategy model under multiple evaluation indicators, the strategy model processing system actually integrates the original multiple evaluation indicators into a comprehensive indicator and evaluates each strategy model based on this comprehensive indicator. By adopting the strategy model evaluation method of this application, each strategy model can be evaluated more comprehensively, thereby determining the strategy model with the best recommendation effect from multiple strategy models.

[0079] In one optional embodiment, after determining the target evaluation value corresponding to each policy model based on the left singular matrix, the right singular matrix, and the diagonal matrix, the policy model processing system can adjust at least one component in the target right singular vector to obtain a first right singular vector, then calculate the product of the first right singular vector and the target singular value to obtain a second value. Finally, the policy model processing system multiplies each component in the target left singular vector with the second value to obtain the latest evaluation value corresponding to each policy model, wherein each component in the target right singular vector corresponds to an evaluation index, and the latest evaluation value is the updated target evaluation value.

[0080] Optionally, in practical applications, if business personnel deem one or more of the multiple metrics particularly important, they can flexibly adjust the weights of those metrics when obtaining the target evaluation value. For example, regarding the multiple target evaluation values ​​u obtained above... 11 ·σ1v1,u 12 ·σ1v1,...,u 1n • σ1v1: If business personnel need to increase the weight of one or more evaluation indicators to emphasize their importance, they can adjust the component corresponding to that indicator in the target right singular vector v1 through the strategy model processing system. Since there is a one-to-one correspondence between the components in the target right singular vector v1 and the evaluation indicators, the value of each component in the target right singular vector v1 is positively correlated with the weight of the corresponding evaluation indicator. It's important to note that after adjusting the components in the target right singular vector v1, the strategy model processing system divides the adjusted component value by the square root of the sum of all components in v1, thus normalizing the target right singular vector v1. The normalized target right singular vector v1 can then be considered the first right singular vector. The strategy model processing system will calculate the product of the first right singular vector and the target singular value to obtain the second value, and then multiply each component in the target left singular vector by the second value to obtain the latest evaluation value for each strategy model. This latest evaluation value is the updated target evaluation value.

[0081] Therefore, the technical solution of this application can flexibly adjust the weight of each evaluation indicator among multiple evaluation indicators, thereby achieving the goal of flexibly evaluating the strategy model through multiple evaluation indicators in different application scenarios. This enables the selection of the target strategy model with the best recommendation effect from multiple strategy models, which can not only improve the user experience, but also increase the sales volume of related products.

[0082] Example 2

[0083] According to embodiments of this application, a processing apparatus for a strategy model is also provided, wherein, Figure 2 This is a schematic diagram of a processing apparatus for an optional strategy model according to an embodiment of this application, such as... Figure 2 As shown, the device includes: an acquisition module 201, used to acquire multiple strategy models and multiple evaluation values ​​corresponding to each strategy model, wherein each evaluation value corresponds to an evaluation index; a construction module 202, used to construct a target matrix based on the evaluation values, wherein in the target matrix, the values ​​in each row are the evaluation values ​​of a strategy model under different evaluation indexes, and the values ​​in each column are the evaluation values ​​of multiple strategy models under the same evaluation index; a first determination module 203, used to determine the target evaluation value corresponding to each strategy model through the target matrix, wherein the target evaluation value is used to characterize the comprehensive score of each strategy model under multiple evaluation indexes; and a second determination module 204, used to determine a target strategy model from multiple strategy models based on the target evaluation value corresponding to each strategy model, and recommend product information to the user through the target strategy model.

[0084] It should be noted that the above-mentioned acquisition module 201, construction module 202, first determination module 203, and second determination module 204 correspond to steps S101 to S104 in the above embodiment 1, respectively. The four modules and the corresponding steps implement the same examples and application scenarios, but are not limited to the content disclosed in the above embodiment 1.

[0085] Optionally, the acquisition module further includes a first acquisition unit and a second acquisition unit. The first acquisition unit is used to acquire multiple first evaluation values ​​corresponding to each policy model during the training process of each policy model, wherein each first evaluation value corresponds to a first-type evaluation metric, and the first-type evaluation metric is the evaluation metric for each policy model during the training phase. The second acquisition unit is used to acquire multiple second evaluation values ​​corresponding to each policy model after each policy model is deployed online, wherein each second evaluation value corresponds to a second-type evaluation metric, and the second-type evaluation metric is the evaluation metric for each policy model during the application phase.

[0086] Optionally, the second acquisition unit further includes: a first determining subunit, a first recommending subunit, and a second determining subunit. The first determining subunit is used to randomly determine a preset number of target users from multiple users; the first recommending subunit is used to recommend target product information to the target users through each strategy model; and the second determining subunit is used to determine multiple second evaluation values ​​corresponding to each strategy model based on the target product information.

[0087] Optionally, the second acquisition unit further includes: a first acquisition subunit, a segmentation unit, a third determination subunit, a second recommendation subunit, and a fourth determination subunit. The first acquisition subunit is used to acquire user characteristics of each user among multiple users; the segmentation unit is used to segment the multiple users into at least one user set based on the user characteristics, wherein users in each user set have the same user characteristics; the third determination subunit is used to determine a preset number of target users from each user set; the second recommendation subunit is used to recommend target product information to the target users through each strategy model; and the fourth determination subunit is used to determine multiple second evaluation values ​​corresponding to each strategy model based on the target product information.

[0088] Optionally, the first determining module further includes a decomposition unit and a first determining unit. The decomposition unit is used to perform singular value decomposition on the target matrix to obtain the left singular matrix, right singular matrix, and diagonal matrix corresponding to the target matrix; the first determining unit is used to determine the target evaluation value corresponding to each policy model based on the left singular matrix, right singular matrix, and diagonal matrix.

[0089] Optionally, the first determining unit further includes: a fifth determining subunit, a sixth determining subunit, a seventh determining subunit, and an eighth determining subunit. The fifth determining subunit is used to determine the target left singular vector based on the right singular matrix and the diagonal matrix; the sixth determining subunit is used to determine the target right singular vector based on the left singular matrix and the diagonal matrix; the seventh determining subunit is used to determine the target singular value based on the diagonal matrix; and the eighth determining subunit is used to determine the target evaluation value corresponding to each policy model based on the target left singular vector, the target right singular vector, and the target singular value.

[0090] Optionally, the eighth determining subunit further includes: a first acquisition submodule, a first calculation submodule, and a second calculation submodule. The first acquisition submodule is used to acquire multiple components in the target left singular vector; the first calculation submodule is used to calculate the product of the target right singular vector and the target singular value to obtain a first value; the second calculation submodule is used to multiply each component in the target left singular vector by the first value to obtain the target evaluation value corresponding to each policy model, wherein one target evaluation value corresponds to one component.

[0091] Optionally, the strategy model processing device further includes: an adjustment module, a first calculation module, and a second calculation module. The adjustment module is used to adjust at least one component of the target right singular vector to obtain a first right singular vector, wherein each component of the target right singular vector corresponds to an evaluation index; the first calculation module is used to calculate the product of the first right singular vector and the target singular value to obtain a second value; the second calculation module is used to multiply each component of the target left singular vector by the second value to obtain the latest evaluation value corresponding to each strategy model, wherein the latest evaluation value is the updated target evaluation value.

[0092] Example 3

[0093] According to an embodiment of this application, a computer-readable storage medium is also provided, in which a computer program is stored, wherein the computer program is configured to execute the strategy model processing method in Embodiment 1 above at runtime.

[0094] Example 4

[0095] According to an embodiment of this application, an embodiment of an electronic device is also provided, wherein, Figure 3 This is a schematic diagram of an optional electronic device according to an embodiment of this application, such as... Figure 3 As shown, the electronic device includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps:

[0096] Obtain multiple strategy models and their corresponding evaluation values, where each evaluation value corresponds to an evaluation metric. Construct a target matrix based on the evaluation values, where each row of the target matrix contains the evaluation values ​​of a strategy model under different evaluation metrics, and each column contains the evaluation values ​​of multiple strategy models under the same evaluation metric. Determine the target evaluation value for each strategy model using the target matrix, where the target evaluation value represents the comprehensive score of each strategy model under multiple evaluation metrics. Based on the target evaluation value corresponding to each strategy model, determine the target strategy model from the multiple strategy models, and recommend product information to the user using the target strategy model.

[0097] Optionally, the processor may further implement the following steps when executing the program: during the training process of each policy model, obtain multiple first evaluation values ​​corresponding to each policy model, wherein each first evaluation value corresponds to a first-type evaluation metric, and the first-type evaluation metric is the evaluation metric of each policy model during the training phase; after each policy model is deployed online, obtain multiple second evaluation values ​​corresponding to each policy model, wherein each second evaluation value corresponds to a second-type evaluation metric, and the second-type evaluation metric is the evaluation metric of each policy model during the application phase.

[0098] Optionally, the processor may also perform the following steps when executing the program: randomly select a preset number of target users from multiple users; recommend target product information to the target users through each strategy model; and determine multiple second evaluation values ​​corresponding to each strategy model based on the target product information.

[0099] Optionally, the processor may also perform the following steps when executing the program: obtaining user characteristics for each of the multiple users; dividing the multiple users into at least one user set based on the user characteristics, wherein users in each user set have the same user characteristics; determining a preset number of target users from each user set; recommending target product information to the target users through each strategy model; and determining multiple second evaluation values ​​corresponding to each strategy model based on the target product information.

[0100] Optionally, the processor may also perform the following steps when executing the program: perform singular value decomposition on the target matrix to obtain the left singular matrix, right singular matrix and diagonal matrix corresponding to the target matrix; determine the target evaluation value corresponding to each policy model based on the left singular matrix, right singular matrix and diagonal matrix.

[0101] Optionally, the processor may also perform the following steps when executing the program: determining the target left singular vector based on the right singular matrix and the diagonal matrix; determining the target right singular vector based on the left singular matrix and the diagonal matrix; determining the target singular value based on the diagonal matrix; and determining the target evaluation value corresponding to each policy model based on the target left singular vector, the target right singular vector, and the target singular value.

[0102] Optionally, the processor may also perform the following steps when executing the program: obtain multiple components in the left singular vector of the target; calculate the product of the right singular vector of the target and the singular value of the target to obtain a first value; multiply each component in the left singular vector of the target by the first value to obtain the target evaluation value corresponding to each policy model, wherein one target evaluation value corresponds to one component.

[0103] Optionally, the processor, when executing the program, also implements the following steps: after determining the target evaluation value corresponding to each policy model based on the left singular matrix, the right singular matrix, and the diagonal matrix, at least one component in the target right singular vector is adjusted to obtain a first right singular vector, wherein each component in the target right singular vector corresponds to an evaluation index; the product of the first right singular vector and the target singular value is calculated to obtain a second value; each component in the target left singular vector is multiplied by the second value to obtain the latest evaluation value corresponding to each policy model, wherein the latest evaluation value is the updated target evaluation value.

[0104] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0105] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0106] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between units or modules may be electrical or other forms.

[0107] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0108] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0109] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, 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. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0110] The above are merely preferred embodiments of this application. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for processing a strategy model, characterized in that, The method comprises the following steps: obtaining a plurality of strategy models and a plurality of evaluation values corresponding to each strategy model, wherein each evaluation value corresponds to an evaluation index, and the plurality of evaluation indexes include a first type of evaluation index of the strategy model in a training stage and a second type of evaluation index of the strategy model in an application stage, the first type of evaluation index at least includes accuracy, precision, recall, balanced F-score, root mean square error and mean absolute percentage error, and the second type of evaluation index at least includes click rate, purchase rate, novelty and diversity; constructing a target matrix according to the evaluation values, wherein in the target matrix, the values in each row are the evaluation values of a strategy model under different evaluation indexes respectively, and the values in each column are the evaluation values of the plurality of strategy models under the same evaluation index; determining a target evaluation value corresponding to each strategy model through the target matrix, comprising: singular value decomposition of the target matrix to obtain a left singular matrix, a right singular matrix and a diagonal matrix corresponding to the target matrix; determining the target evaluation value corresponding to each strategy model according to the left singular matrix, the right singular matrix and the diagonal matrix, wherein the target evaluation value is used to represent the comprehensive score of each strategy model under a plurality of evaluation indexes; determining a target strategy model from the plurality of strategy models according to the target evaluation value corresponding to each strategy model, and recommending commodity information to a user through the target strategy model.

2. The method of claim 1, wherein, obtaining a plurality of strategy models and a plurality of evaluation values corresponding to each strategy model comprises: obtaining a plurality of first evaluation values corresponding to each strategy model during the training process of each strategy model, wherein each first evaluation value corresponds to a first type of evaluation index, and the first type of evaluation index is an evaluation index of each strategy model in a training stage; obtaining a plurality of second evaluation values corresponding to each strategy model after the online deployment of each strategy model, wherein each second evaluation value corresponds to a second type of evaluation index, and the second type of evaluation index is an evaluation index of each strategy model in an application stage.

3. The method of claim 2, wherein, obtaining a plurality of second evaluation values corresponding to each strategy model after the online deployment of each strategy model comprises: randomly determining a preset number of target users from a plurality of users; recommending target commodity information to the target users through each strategy model; determining a plurality of second evaluation values corresponding to each strategy model according to the target commodity information.

4. The method of claim 2, wherein, obtaining a plurality of second evaluation values corresponding to each strategy model after the online deployment of each strategy model comprises: obtaining user features of each user in a plurality of users; dividing the plurality of users into at least one user set according to the user features, wherein the users in each user set have the same user features; determining a preset number of target users from each user set; recommending target commodity information to the target users through each strategy model; determining a plurality of second evaluation values corresponding to each strategy model according to the target commodity information.

5. The method of claim 1, wherein, The method further comprises: determining a target left singular vector according to the right singular matrix and the diagonal matrix; determining a target right singular vector according to the left singular matrix and the diagonal matrix; determining a target singular value according to the diagonal matrix; determining the target evaluation value corresponding to each strategy model according to the target left singular vector, the target right singular vector and the target singular value.

6. The method of claim 5, wherein, The method further comprises: obtaining a plurality of components in the target left singular vector; calculating a product of the target right singular vector and the target singular value to obtain a first numerical value; multiplying each component in the target left singular vector with the first numerical value respectively to obtain the target evaluation value corresponding to each strategy model, wherein one target evaluation value corresponds to one component.

7. The method of claim 6, wherein, After determining the target evaluation value corresponding to each strategy model according to the left singular matrix, the right singular matrix and the diagonal matrix, the method further comprises: adjusting at least one component in the target right singular vector to obtain a first right singular vector, wherein each component in the target right singular vector corresponds to one evaluation index; calculating a product of the first right singular vector and the target singular value to obtain a second numerical value; multiplying each component in the target left singular vector with the second numerical value respectively to obtain a latest evaluation value corresponding to each strategy model, wherein the latest evaluation value is an updated target evaluation value.

8. A processing device of a policy model, characterized by, The method further comprises: an obtaining module, configured to obtain a plurality of strategy models and a plurality of evaluation values corresponding to each strategy model, wherein each evaluation value corresponds to one evaluation index, and the plurality of evaluation indexes include a first type of evaluation index when the strategy model is in a training stage and a second type of evaluation index when the strategy model is in an application stage, the first type of evaluation index at least includes accuracy, precision, recall, balanced F-score, root mean square error and mean absolute percentage error, and the second type of evaluation index at least includes click rate, purchase rate, novelty and diversity; a constructing module, configured to construct a target matrix according to the evaluation values, wherein in the target matrix, the values in each row are evaluation values of one strategy model under different evaluation indexes, and the values in each column are evaluation values of the plurality of strategy models under the same evaluation index; a first determining module, configured to determine a target evaluation value corresponding to each strategy model through the target matrix, wherein the target evaluation value is used to represent a comprehensive score of the each strategy model under a plurality of evaluation indexes. The first determining module comprises: a decomposition unit, configured to perform singular value decomposition on the target matrix to obtain a left singular matrix, a right singular matrix and a diagonal matrix corresponding to the target matrix; and a first determining unit, configured to determine the target evaluation value corresponding to each strategy model according to the left singular matrix, the right singular matrix and the diagonal matrix. The second determining module is configured to determine a target strategy model from the plurality of strategy models according to the target evaluation value corresponding to each strategy model, and recommend commodity information to a user through the target strategy model.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, wherein the computer program is configured to execute the processing method of the strategy model in any one of claims 1-7 when running.

10. An electronic device, comprising: The computer readable storage medium stores a computer program, wherein the computer program is configured to execute the processing method of the strategy model in any one of claims 1-7 when running. The computer readable storage medium stores a computer program, wherein the computer program is configured to execute the processing method of the strategy model in any one of claims 1-7 when running.

Images