A score prediction model training method and device, electronic equipment and storage medium

Abstract

The application discloses a score prediction model training method and device, electronic equipment and a storage medium, relates to the technical field of artificial intelligence, and is used to solve the problem that the prediction result of the score prediction model trained at the present stage is not accurate enough, and comprises the following steps: obtaining sample data, wherein the sample data comprises multi-dimensional behavior characteristics of each user in a plurality of users and real user scores corresponding to the multi-dimensional behavior characteristics of each user; determining the number of times that each real user score appears in the sample data; determining target user scores corresponding to the multi-dimensional behavior characteristics of each user according to the number of times that each real user score appears in the sample data; and training an initial score prediction model according to the target user scores corresponding to the multi-dimensional behavior characteristics of each user and the multi-dimensional behavior characteristics of each user, to obtain a trained target score prediction model. The application is used for score prediction of user network perception.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a score prediction model training method and device, an electronic device, and a storage medium. BACKGROUND

[0002] Improving mobile network perception scores is a continuous concern for operators. Operators generally set the score value to 1-10, a total of 10 numbers, so that users can rate according to their own satisfaction. At present, in view of the current situation that user network perception scores are low and passive solutions to network problems do not achieve good results, operators generally use intelligent algorithms to establish a score prediction model for the network perception of all users to accurately predict user network perception scores. Then, for users with low predicted scores, operators actively take measures to ensure that the purpose of improving the perception of all users is achieved. However, the training method of the score prediction model at the present stage is not accurate enough. SUMMARY

[0003] The present application provides a score prediction model training method and device, an electronic device, and a storage medium, which can solve the problem of inaccurate prediction results of the score prediction model trained at the present stage.

[0004] To achieve the above purpose, the present application adopts the following technical solutions:

[0005] In a first aspect, the present application provides a score prediction model training method, comprising: obtaining sample data, the sample data comprising multi-dimensional behavior characteristics of each user in a plurality of users and a real user score corresponding to the multi-dimensional behavior characteristics of each user; determining the number of times each real user score appears in the sample data; determining a target user score corresponding to the multi-dimensional behavior characteristics of each user according to the number of times each real user score appears in the sample data; the target user score is used to represent the predicted value of the score of each user; and training an initial score prediction model according to the target user score corresponding to the multi-dimensional behavior characteristics of each user and the multi-dimensional behavior characteristics of each user, to obtain a trained target score prediction model.

[0006] Based on the above technical solutions, the present application corrects the initial predicted score determined by the initial score prediction model by the number of times each user score appears in the sample data, determines the final target user score, and trains the initial score prediction model accordingly to obtain the trained target score prediction model. Thus, the trained target score prediction model has higher accuracy and smaller MAE value.

[0007] In a possible implementation, the target user score corresponding to the multi-dimensional behavior feature of each user is determined according to the number of occurrences of each real user score in the sample data, specifically comprising: determining a distribution ratio of each real user score according to the number of occurrences of each real user score in the sample data; rounding the initial predicted score according to the distribution ratio of each real user score to determine the target user score corresponding to the multi-dimensional behavior feature of each user; wherein the initial predicted score is the user score determined by the initial score prediction model.

[0008] In a possible implementation, rounding the initial predicted score according to the distribution ratio of each real user score comprises: determining the distribution ratio of each two real user scores adjacent in size; and rounding the initial predicted score locally according to the distribution ratio of each two real user scores adjacent in size; wherein rounding the initial predicted score locally according to the distribution ratio of each two real user scores adjacent in size satisfies the following formula:

[0009]

[0010] wherein A represents the target user score, B1 and B2 represent each two real user scores adjacent in size, N represents the initial predicted score, and x1 and x2 represent the distribution ratio of each two real user scores adjacent in size.

[0011] In a possible implementation, after rounding the initial predicted score locally according to the distribution ratio of each two real user scores adjacent in size, the method further comprises: determining the distribution ratio of any two real user scores; and rounding the initial predicted score after local rounding globally according to the distribution ratio of any two real user scores; wherein rounding the initial predicted score after local rounding globally according to the distribution ratio of any two real user scores satisfies the following formula:

[0012]

[0013] wherein A represents the target user score, B3 and B4 represent any two real user scores, N represents the initial predicted score, and x3 and x4 represent the distribution ratio of any two real user scores.

[0014] In a possible implementation, the multi-dimensional behavior feature of each user is determined based on the gender, age, number of uses of networks of different network modes, use time of networks of different network modes, network access rate, and downlink round-trip time (RTT) average delay of each user.

[0015] In a second aspect, the present application provides a scoring prediction model training device, comprising: the scoring prediction model training device comprises an acquisition unit and a processing unit; the acquisition unit is configured to acquire sample data, the sample data comprising multi-dimensional behavior characteristics of each user in a plurality of users and a real user score corresponding to the multi-dimensional behavior characteristics of each user; the processing unit is configured to determine a number of occurrences of each real user score in the sample data; the processing unit is further configured to determine a target user score corresponding to the multi-dimensional behavior characteristics of each user according to the number of occurrences of each real user score in the sample data; the target user score is used to represent a predicted value of the score of each user; and the processing unit is further configured to train an initial scoring prediction model according to the target user score corresponding to the multi-dimensional behavior characteristics of each user and the multi-dimensional behavior characteristics of each user, to obtain a trained target scoring prediction model.

[0016] In a possible implementation, the processing unit is further configured to determine a distribution ratio of each real user score according to the number of occurrences of each real user score in the sample data; and the processing unit is further configured to round an initial prediction score according to the distribution ratio of each real user score, to determine the target user score corresponding to the multi-dimensional behavior characteristics of each user; wherein the initial prediction score is a user score determined by the initial scoring prediction model.

[0017] In a possible implementation, the processing unit is further configured to determine a distribution ratio of each two real user scores adjacent in size; and the processing unit is further configured to locally round the initial prediction score according to the distribution ratio of each two real user scores adjacent in size; wherein the local rounding of the initial prediction score according to the distribution ratio of each two real user scores adjacent in size satisfies the following formula:

[0018]

[0019] wherein A represents the target user score, B1 and B2 represent each two real user scores adjacent in size, N represents the initial prediction score, and x1 and x2 represent the distribution ratio of each two real user scores adjacent in size.

[0020] In a possible implementation, the processing unit is further configured to determine a distribution ratio of any two real user scores; and the processing unit is further configured to globally round the initial prediction score after the local rounding according to the distribution ratio of any two real user scores; wherein the global rounding of the initial prediction score after the local rounding according to the distribution ratio of any two real user scores satisfies the following formula:

[0021]

[0022] Wherein, A represents a target user score, B3 and B4 represent any two real user scores, N represents an initial predicted score, and x3 and x4 represent distribution proportions of any two real user scores.

[0023] In a possible implementation, the multi-dimensional behavior feature of each user is determined based on the gender, age, number of times of using networks of different network modes, time length of using networks of different network modes, network access rate, and average round-trip time (RTT) of a downlink of each user.

[0024] In addition, the technical effect of the score prediction model training method of the second aspect can refer to the technical effect of the score prediction model training method of the first aspect, which will not be described here.

[0025] In a third aspect, the present application provides a computer readable storage medium storing one or more programs, the one or more programs including instructions, which when executed by an electronic device of the present application, cause the electronic device to perform the score prediction model training method as described in the first aspect and any possible implementation of the first aspect.

[0026] In a fourth aspect, the present application provides an electronic device, comprising a processor and a memory; wherein the memory is configured to store one or more programs, the one or more programs including computer execution instructions, when the electronic device is running, the processor executes the computer execution instructions stored in the memory, so that the electronic device performs the score prediction model training method as described in the first aspect and any possible implementation of the first aspect.

[0027] In a fifth aspect, the present application provides a computer program product including instructions, which when executed on a computer, cause an electronic device of the present application to perform the score prediction model training method as described in the first aspect and any possible implementation of the first aspect.

[0028] In a sixth aspect, the present application provides a chip system applied to a score prediction model training device; the chip system includes one or more interface circuits and one or more processors. The interface circuit and the processor are interconnected through a circuit; the interface circuit is configured to receive a signal from a memory of the score prediction model training device and send the signal to the processor, the signal including computer instructions stored in the memory. When the processor executes the computer instructions, the score prediction model training device performs the score prediction model training method as described in the first aspect and any possible design of the first aspect.

[0029] In this application, the name of the above-mentioned score prediction model training apparatus does not constitute a limitation on the device or functional unit itself, and in actual implementation, these devices or functional units can appear in other names. As long as the function of each device or functional unit is similar to that of the present application, it belongs to the scope of the claims of the present application and its equivalent technologies. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 A structural schematic diagram of a score prediction model training apparatus provided by an embodiment of the present application;

[0031] Figure 2 A flowchart of a score prediction model training method provided by an embodiment of the present application;

[0032] Figure 3 A flowchart of another score prediction model training method provided by an embodiment of the present application;

[0033] Figure 4 A flowchart of another score prediction model training method provided by an embodiment of the present application;

[0034] Figure 5 A structural schematic diagram of a score prediction model training apparatus provided by an embodiment of the present application;

[0035] Figure 6 A structural schematic diagram of another score prediction model training apparatus provided by an embodiment of the present application. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0037] In this document, the character " / " generally represents that the associated objects before and after are in an "or" relationship. For example, A / B can be understood as A or B.

[0038] The terms "first" and "second" in the specification and claims of the present application are used to distinguish different objects, not to describe the specific order of the objects. For example, the first edge service node and the second edge service node are used to distinguish different edge service nodes, not to describe the characteristic order of the edge service nodes.

[0039] Moreover, the terms "comprising" and "having" and any variations thereof in the present description are intended to cover a non-exclusive inclusion. For example, a process, method, system, product, or apparatus that comprises a list of steps or elements is not necessarily limited to those listed steps or elements but can include other not-listed steps or elements, or can include other steps or elements that are inherent to such process, method, product, or apparatus.

[0040] In addition, in the embodiments of the present application, the words "exemplary" and "for example" are used on the basis and in the sense of assisting in general understanding of the application. Any embodiment or design scheme described as "exemplary" or "for example" in the present application should not be interpreted as being more preferred or having more advantages than other embodiments or design schemes. In fact, the use of the words "exemplary" or "for example" is intended to present the concept in a specific manner.

[0041] At present, improving the mobile network perception score is a continuous concern of operators. Operators generally set the score value to 1-10, so that users can score according to their own satisfaction. If the user scores 10 points, it means that the user is very satisfied; if the user scores 1 point, it means that the user is very dissatisfied.

[0042] In view of the current situation that the user network perception score is low and passive solution to network problems does not achieve good results, operators generally use intelligent algorithms to establish a scoring prediction model for the network perception of all users in the network to accurately predict the user network perception score. Then, for users with low predicted scores, operators actively take measures to ensure key points to improve the network perception of all users.

[0043] In order to measure the accuracy of the network perception score prediction model, the Mean Absolute Error (MAE) value is generally used for evaluation. The MAE value describes the average of the absolute error. For the scoring prediction model involved in the present application, the smaller the MAE value, the closer the predicted score result is to the actual score result of the user, and the better the prediction effect.

[0044] At present, when establishing a network perception score prediction model, two methods are usually adopted: classification method and regression method. When the classification method is used to predict the score of user network perception, the target value is 10 score values from 1 to 10. The user features are taken as the model input, and the classification model is trained to predict that the user score belongs to a certain category among the 10 numbers. However, the classification method also has some disadvantages: according to the analysis of the actual data, it is found that the user scores are extremely unevenly distributed, and the number of scores of 10 and 1 is the largest, and the number of intermediate scores such as 3 and 4 is smaller, so misclassification is likely to occur during model training. If the classification method is used to predict the score of user network perception, the classification error will greatly affect the evaluation index MAE value and the prediction accuracy.

[0045] When the regression method is used to predict the score of user network perception, the target value is still 10 different scores. Then the user features are taken as the input to train the regression model, and the user score value can be predicted. However, the regression method also has some disadvantages: the regression method usually takes the evaluation index, i.e. the MAE value index, as the objective function of the model, and the value of the objective function tends to be smaller during the training process of the model; based on this principle, the model trained has a prediction result that is mostly not an integer, but a decimal such as 3.2 and 5.8, which does not occur in actual user scoring and deviates from the actual situation, and the prediction accuracy is poor.

[0046] Exemplarily, to solve the above problems, the following two schemes are provided at present:

[0047] Scheme one: a user score prediction model is constructed by using deep learning technology; the purchase history and score record of a user for a tourism resource are obtained, and the user score prediction model is trained using the tourism resource purchased by the user; the predicted score is obtained by inputting the tourism resource not purchased by the user into the user score prediction model, and the first c tourism resources with the highest predicted score and not purchased by the user are recommended to the user. This method does not consider the possible defects of the classification and regression methods in the user score prediction model, and is not optimized.

[0048] Scheme two: sample songs are obtained, the sample songs contain pre-labeled song labels; a user score matrix is obtained, the user score matrix contains at least one score of a sample song by a user, and the score is calculated according to the operation behavior of the user on the sample song; a song classifier is generated according to the user score matrix and the song labels of the sample songs; and the song labels are assigned to each song in the song library by using the song classifier. However, the influence of the unbalanced user scores on the accuracy of song classification is not considered.

[0049] The above scheme cannot solve the problem of low user score prediction accuracy when the user network perception score is predicted by regression or classification alone. In summary, the training method of the current score prediction model and the prediction result of the trained score prediction model are not accurate enough.

[0050] To solve the problem of the current score prediction model training method and the prediction result of the trained score prediction model not being accurate enough, the present application provides a score prediction model training method, which takes user behavior features as input data to predict the user score based on the classic machine learning regression model Lightgbm. The optimal rounding and reclassification are performed on the predicted decimal result to form a user score prediction model with higher accuracy and smaller MAE value.

[0051] Exemplarily, as shown in Figure 1 The structure of the score prediction model training device provided by the present application is shown in the figure. The score prediction model training device 10 specifically includes a sample data module 11, a user score processing module 12, and a model training module 13.

[0052] The sample data module 11 is used to obtain sample data. The sample data includes multi-dimensional behavior features of each user in multiple users and real user scores corresponding to the multi-dimensional behavior features of each user. The sample data module 11 is also used to determine the number of occurrences of each real user score in the sample data.

[0053] The user score processing module 12 is used to determine the target user score corresponding to the multi-dimensional behavior features of each user according to the number of occurrences of each real user score in the sample data.

[0054] The model training module 13 is used to construct an initial score prediction model and train the initial score prediction model according to the target user score corresponding to the multi-dimensional behavior features of each user and the multi-dimensional behavior features of each user, to obtain a trained target score prediction model.

[0055] It should be noted that in the score prediction model training method provided by the present application, the execution subject is a score prediction model training device. The score prediction model training device can be an electronic device (such as a computer terminal, a server), a processor in an electronic device, a control module for score prediction model training in an electronic device, or a client for score prediction model training in an electronic device.

[0056] The flow of the score prediction model training method provided by the present embodiment is described below.

[0057] Exemplarily, as shown in Figure 2As shown, the score prediction model training method provided in the present application specifically includes the following S201-S204:

[0058] S201, the score prediction model training device acquires sample data.

[0059] The sample data includes the multi-dimensional behavior characteristics of each user in the plurality of users and the real user score corresponding to the multi-dimensional behavior characteristics of each user.

[0060] Optionally, the multi-dimensional behavior characteristics of each user are determined according to the gender, age, usage frequency of different network modes, usage time length of different network modes, network connection rate and average delay of downlink round-trip time (RTT).

[0061] Exemplarily, the score prediction model training device can acquire the sample data through an operator data bureau.

[0062] Optionally, the real user score is set to 10 numbers from 1 to 10, which is obtained by the user according to the satisfaction degree. The user score of 10 points indicates that the user is very satisfied, and the user score of 1 point indicates that the user is very dissatisfied.

[0063] It can be understood that the real user score is the real score value of each user based on the multi-dimensional behavior characteristics of the user to the network provided by the operator.

[0064] S202, the score prediction model training device determines the number of times each real user score appears in the sample data.

[0065] Optionally, the score prediction model training device, after acquiring the sample data, according to the setting condition of the real user score, counts the number of times each score appears in the sample data. Exemplarily, combined with the example in S201, the score prediction model training device counts the number of times from 1 to 10 in the sample data.

[0066] S203, the score prediction model training device determines the target user score corresponding to the multi-dimensional behavior characteristics of each user according to the number of times each real user score appears in the sample data.

[0067] Optionally, the score prediction model training device determines the distribution proportion of each real user score according to the number of times each real user score appears in the sample data.

[0068] Further, the score prediction model training apparatus rounds the initial prediction score according to a distribution proportion of each real user score, to determine a target user score corresponding to the multi-dimensional behavior feature of each user. It can be understood that the target user score is a prediction value for embodying the score of each user in this embodiment.

[0069] It can be understood that the initial prediction score is a user score determined by an initial score prediction model. The initial score prediction model is a model constructed based on a classical machine learning regression model Lightgbm, and the user score predicted by the initial score prediction model is the initial prediction score. The initial prediction score may have decimal places based on the characteristics of the algorithm itself. However, in the actual user scoring process, the score value will not have decimal places, so the score prediction model training apparatus will round the initial prediction score.

[0070] In a possible implementation, the score prediction model training apparatus rounds the initial prediction score locally according to a distribution proportion of each two real user scores adjacent in size. For details of the process in which the score prediction model training apparatus rounds the initial prediction score locally, refer to S301-S302 below, which will not be described here.

[0071] In another possible implementation, the score prediction model training apparatus rounds the initial prediction score locally according to a distribution proportion of each two real user scores adjacent in size, and then rounds the initial prediction score after local rounding globally according to a distribution proportion of any two real user scores. For details of the process in which the score prediction model training apparatus rounds the initial prediction score locally and globally, refer to S401-S402 below, which will not be described here.

[0072] It can be understood that the rounding of the initial prediction score by the score prediction model training apparatus makes the prediction score more consistent with the actual situation, thereby improving the accuracy of score prediction.

[0073] S204, the score prediction model training apparatus trains the initial score prediction model according to the target user score corresponding to the multi-dimensional behavior feature of each user and the multi-dimensional behavior feature of each user, to obtain a trained target score prediction model.

[0074] Optionally, after the score prediction model training apparatus obtains the target user score corresponding to the multi-dimensional behavior feature of each user, the score prediction model training apparatus takes the target user score as a prediction result of the initial score prediction model to calculate an MAE value index and determine whether the MAE value meets a preset requirement. For example, the preset requirement that the MAE value meets can be that the MAE value is less than or equal to a preset value, or a function related to the MAE value reaches convergence, which is not limited in the embodiments of the present application.

[0075] Understandably, if the MAE value of the initial rating prediction model meets the preset requirements after the rating prediction model training device uses the target user rating as the prediction result of the initial rating prediction model, then the rating prediction model training device determines that the initial rating prediction model training is complete and determines the initial rating prediction model at this time as the trained target rating prediction model.

[0076] Similarly, if the MAE (Maximum Averaging Effect) of the initial rating prediction model fails to meet the preset requirements after the rating prediction model training device uses the target user rating as the prediction result, iterative training of the initial rating prediction model will continue based on the sample data. Furthermore, in each iteration, the rating prediction model training device will round the prediction result of the initial rating prediction model according to the frequency of each real user rating in the sample data to determine the target user rating corresponding to each user's multi-dimensional behavioral characteristics. After this, the rating prediction model training device will again use the target user rating determined in this iteration as the prediction result of the initial rating prediction model, until the MAE value of the initial rating prediction model meets the preset requirements.

[0077] Based on the above technical solution, this application embodiment corrects the initial predicted rating determined by the initial rating prediction model by the frequency of each user rating appearing in the sample data, thereby determining the final target user rating. The initial rating prediction model is then trained based on this result to obtain a trained target rating prediction model. Consequently, the trained target rating prediction model has higher accuracy and a lower MAE value.

[0078] For example, combined Figure 2 ,like Figure 3 As shown, in the rating prediction model training method provided in this application, the rating prediction model training device determines the target user rating corresponding to the multi-dimensional behavioral characteristics of each user based on the number of times each real user rating appears in the sample data, specifically including the following S301-S302:

[0079] S301, The rating prediction model training device determines the distribution ratio of every two real user ratings that are adjacent in size.

[0080] It is understandable that the rating prediction model training device determines the distribution ratio of every two adjacent real user ratings in order to round down the decimal places in the initial predicted ratings. If the two real user ratings determined by the rating prediction model training device are not adjacent in size, the number of decimal places in the initial predicted rating will also change, which does not conform to the purpose of rounding down the decimal places in the initial predicted rating in this embodiment.

[0081] It should be noted that the method for determining the distribution ratio of a certain real user score is described above in S202, which will not be repeated here.

[0082] S302, the score prediction model training apparatus locally rounds the initial predicted score according to the distribution ratio of each two real user scores adjacent in size.

[0083] In a possible implementation, the score prediction model training apparatus locally rounds the initial predicted score according to the distribution ratio of each two real user scores adjacent in size, which satisfies the following formula:

[0084]

[0085] Wherein, A represents the target user score, B1 and B2 represent each two real user scores adjacent in size, N represents the initial predicted score, and x1 and x2 represent the distribution ratio of each two real user scores adjacent in size.

[0086] The above describes the specific process of the score prediction model training apparatus determining the target user score corresponding to the multi-dimensional behavior feature of each user according to the number of occurrences of each real user score in the sample data in the score prediction model training method provided by the present application.

[0087] For example, in combination with Figure 3 As shown in Figure 4 The score prediction model training method provided by the present application, the score prediction model training apparatus locally rounds the initial predicted score according to the distribution ratio of each two real user scores adjacent in size, and then globally rounds the initial predicted score after local rounding according to the distribution ratio of any two real user scores, which specifically includes the following S401-S402:

[0088] S401, the score prediction model training apparatus determines the distribution ratio of any two real user scores.

[0089] It can be understood that the score prediction model training apparatus determines the distribution ratio of any two real user scores in order to globally round the initial predicted score and achieve the purpose of re-splitting the initial predicted score.

[0090] It should be noted that the method for determining the distribution ratio of a certain real user score is described above in S202, which will not be repeated here.

[0091] S402, the score prediction model training apparatus globally rounds the initial predicted score after local rounding according to the distribution ratio of any two real user scores.

[0092] In a possible implementation, the score prediction model training apparatus performs global rounding on the initial predicted score after the local rounding according to the distribution proportion of any two real user scores, and the global rounding satisfies the following formula:

[0093]

[0094] wherein A represents the target user score, B3 and B4 represent the any two real user scores, N represents the initial predicted score, and x3 and x4 represent the distribution proportion of the any two real user scores.

[0095] It should be understood that, based on the above S301-S302 and S401-S402, the score prediction model training apparatus determines the target user score corresponding to the multi-dimensional behavior feature of each user, which is equivalent to determining the clustering center, and the clustering center can reflect the actual user score situation analyzed according to the sample data, and the initial score prediction model is classified again according to the proportion, so as to reduce the MAE index of the prediction result of the score prediction model and improve the accuracy of the score prediction.

[0096] The embodiments of the present application can divide the function modules or function units of the score prediction model training apparatus according to the above method examples. For example, each function module or function unit can be divided according to each function, or two or more functions can be integrated into one processing module. The above integrated module can be realized in the form of hardware or software function module or function unit. The division of modules or units in the embodiments of the present application is illustrative, and is only a logical function division. In actual implementation, another division mode can be used.

[0097] Exemplarily, as shown in Figure 5 , a possible structure schematic diagram of a score prediction model training apparatus related to the embodiments of the present application is shown. The score prediction model training apparatus 500 includes an acquisition unit 501 and a processing unit 502.

[0098] The acquisition unit 501 is configured to acquire sample data, and the sample data includes the multi-dimensional behavior feature of each user in a plurality of users and the real user score corresponding to the multi-dimensional behavior feature of each user.

[0099] The processing unit 502 is configured to determine the number of times of occurrence of each real user score in the sample data.

[0100] The processing unit 502 is further configured to determine the target user score corresponding to the multi-dimensional behavior feature of each user according to the number of times of occurrence of each real user score in the sample data, and the target user score is used to represent the predicted value of the score of each user.

[0101] The processing unit 502 is further configured to train the initial score prediction model according to the target user score corresponding to each user's multi-dimensional behavior feature and the initial score prediction model according to each user's multi-dimensional behavior feature, to obtain a trained target score prediction model.

[0102] Optionally, the processing unit 502 is further configured to determine a distribution ratio of each real user score according to the number of times each real user score appears in the sample data.

[0103] Optionally, the processing unit 502 is further configured to round the initial prediction score according to the distribution ratio of each real user score to determine the target user score corresponding to each user's multi-dimensional behavior feature; wherein the initial prediction score is the user score determined by the initial score prediction model.

[0104] Optionally, the processing unit 502 is further configured to determine the distribution ratio of each two real user scores adjacent in size.

[0105] Optionally, the processing unit 502 is further configured to locally round the initial prediction score according to the distribution ratio of each two real user scores adjacent in size; wherein the local rounding of the initial prediction score according to the distribution ratio of each two real user scores adjacent in size satisfies the following formula:

[0106]

[0107] Wherein, A represents the target user score, B1 and B2 represent each two real user scores adjacent in size, N represents the initial prediction score, and x1 and x2 represent the distribution ratio of each two real user scores adjacent in size.

[0108] Optionally, the processing unit 502 is further configured to determine the distribution ratio of any two real user scores.

[0109] Optionally, the processing unit 502 is further configured to globally round the initial prediction score after local rounding according to the distribution ratio of any two real user scores; wherein the global rounding of the initial prediction score after local rounding according to the distribution ratio of any two real user scores satisfies the following formula:

[0110]

[0111] Wherein, A represents the target user score, B3 and B4 represent any two real user scores, N represents the initial prediction score, and x3 and x4 represent the distribution ratio of any two real user scores.

[0112] Optionally, the score prediction model training apparatus 500 can further include a storage unitFigure 5 The storage unit stores a program or an instruction, which, when executed by the processing unit 501, causes the score prediction model training apparatus to perform the score prediction model training method described in the above method embodiments.

[0113] In addition, Figure 5 The technical effects of the score prediction model training apparatus can refer to those of the score prediction model training method described in the above embodiments, which will not be repeated here.

[0114] Exemplarily, Figure 6 Another possible structure of the score prediction model training apparatus involved in the above embodiments is shown in FIG. 6. As shown in FIG. 6, Figure 6 The score prediction model training apparatus 600 includes a processor 602.

[0115] The processor 602 is configured to control and manage the actions of the score prediction model training apparatus, for example, to perform the steps performed by the obtaining unit 501 and the processing unit 502, and / or to perform other processes of the technical solutions described herein.

[0116] The processor 602 can be various exemplary logical blocks, modules and circuits described in combination with the content of the present application. The processor can be a central processing unit, a general purpose processor, a digital signal processor, an application specific integrated circuit, a field programmable gate array or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It can implement or execute various exemplary logical blocks, modules and circuits described in combination with the content of the present application. The processor can also be a combination of computing functions, such as one or more microprocessor combinations, DSP and microprocessor combinations, etc.

[0117] Optionally, the score prediction model training apparatus 600 can further include a communication interface 603, a memory 601 and a bus 604. The communication interface 603 is configured to support the communication between the score prediction model training apparatus 600 and other network entities. The memory 601 is configured to store the program code and data of the score prediction model training apparatus.

[0118] The memory 601 can be a memory in the score prediction model training apparatus, which can include a volatile memory such as a random access memory; the memory can also include a non-volatile memory such as a read-only memory, a flash memory, a hard disk or a solid state disk; the memory can also include a combination of the above kinds of memories.

[0119] The bus 604 can be an Extended Industry Standard Architecture (EISA) bus or the like. The bus 604 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 Only one thick line is used to represent the bus in the middle, but it does not mean that there is only one bus or only one type of bus.

[0120] Through the description of the above embodiments, those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above division of functional modules is taken as an example, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device and module described above can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0121] The embodiment of the present application provides a computer program product containing instructions, when the computer program product runs on the electronic device of the present application, so that the computer executes the score prediction model training method described in the method embodiment.

[0122] The embodiment of the present application also provides a computer readable storage medium, the computer readable storage medium stores instructions, when the computer executes the instructions, the electronic device of the present application executes each step executed by the score prediction model training device in the method flow shown in the method embodiment.

[0123] The computer readable storage medium, for example, can be, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium include: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), registers, a hard disk, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. A tangible, non-transitory computer-readable storage medium can be coupled to a processor such that the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium can be integral to the processor. The processor and the storage medium can reside in an application-specific integrated circuit (ASIC). In an embodiment of the application, the computer readable storage medium can be any tangible medium that can retain, or store, program code in the form of instructions or data, which can be executed by a processor or computing device.

[0124] The above description is merely illustrative of the application, and not restrictive. The scope of the application should be determined by the appended claims, along with their full scope of equivalents.

Claims

1. A method for training a rating prediction model, characterized in that, The method comprises: obtaining sample data, wherein the sample data comprises multi-dimensional behavior characteristics of each user in a plurality of users and real user scores corresponding to the multi-dimensional behavior characteristics of the each user; the multi-dimensional behavior characteristics are determined based on gender, age, usage times of networks of different network modes, usage times of networks of different network modes, network access rates and average downlink round-trip delay of the each user; determining a number of occurrences of each of the real user scores in the sample data; determining a distribution proportion of each of the real user scores according to the number of occurrences of the each of the real user scores in the sample data; rounding an initial prediction score according to the distribution proportion of each of the real user scores to determine a target user score corresponding to the multi-dimensional behavior characteristics of the each user; the target user score is used to represent a prediction value of a score of the each user; the initial prediction score is a user score determined by an initial score prediction model; training the initial score prediction model according to the target user score corresponding to the multi-dimensional behavior characteristics of the each user and the multi-dimensional behavior characteristics of the each user to obtain a trained target score prediction model; wherein rounding the initial prediction score according to the distribution proportion of each of the real user scores comprises: determining a distribution proportion of each of two real user scores adjacent in size; performing local rounding on the initial prediction score according to the distribution proportion of each of the two real user scores adjacent in size; wherein the local rounding on the initial prediction score according to the distribution proportion of each of the two real user scores adjacent in size satisfies the following formula: wherein A represents the target user score, B1 and B2 represent each of the two real user scores adjacent in size, N represents the initial prediction score, and x1 and x2 represent the distribution proportion of each of the two real user scores adjacent in size; determining a distribution proportion of any two real user scores; performing global rounding on the initial prediction score after the local rounding according to the distribution proportion of the any two real user scores; wherein the global rounding on the initial prediction score after the local rounding according to the distribution proportion of the any two real user scores satisfies the following formula: wherein A represents the target user score, B3 and B4 represent the any two real user scores, N represents the initial prediction score, and x3 and x4 represent the distribution proportion of the any two real user scores. 2.A scoring prediction model training apparatus, characterized by, The score prediction model training device comprises an acquisition unit and a processing unit. The acquisition unit is configured to acquire sample data, wherein the sample data comprises multi-dimensional behavior characteristics of each user in a plurality of users and real user scores corresponding to the multi-dimensional behavior characteristics of the each user; the multi-dimensional behavior characteristics are determined based on gender, age, usage times of networks of different network modes, usage times of networks of different network modes, network access rates and average downlink round-trip delay of the each user; The processing unit is configured to determine a number of times each real user score appears in the sample data; The processing unit is further configured to determine a distribution proportion of each real user score according to the number of times each real user score appears in the sample data; The processing unit is further configured to round an initial predicted score according to the distribution proportion of each real user score to determine a target user score corresponding to the multi-dimensional behavior feature of each user, wherein the target user score is used to represent a predicted value of the score of each user, and the initial predicted score is a user score determined by an initial score prediction model; The processing unit is further configured to train an initial score prediction model according to the target user score corresponding to the multi-dimensional behavior feature of each user and the multi-dimensional behavior feature of each user to obtain a trained target score prediction model; The rounding of the initial predicted score according to the distribution proportion of each real user score includes: determining a distribution proportion of each two real user scores adjacent in size; performing local rounding of the initial predicted score according to the distribution proportion of each two real user scores adjacent in size, wherein the local rounding of the initial predicted score according to the distribution proportion of each two real user scores adjacent in size satisfies the following formula: wherein A represents the target user score, B1 and B2 represent each two real user scores adjacent in size, N represents the initial predicted score, and x1 and x2 represent the distribution proportion of each two real user scores adjacent in size; The processing unit is further configured to determine a distribution proportion of any two real user scores; The processing unit is further configured to perform global rounding of the initial predicted score after the local rounding according to the distribution proportion of any two real user scores, wherein the global rounding of the initial predicted score after the local rounding according to the distribution proportion of any two real user scores satisfies the following formula: wherein A represents the target user score, B3 and B4 represent any two real user scores, N represents the initial predicted score, and x3 and x4 represent the distribution proportion of any two real user scores.

3. An electronic device, comprising: The electronic device comprises a processor and a memory, wherein the memory is configured to store computer execution instructions, and when the electronic device is running, the processor executes the computer execution instructions stored in the memory to enable the electronic device to perform the score prediction model training method according to claim 1. The computer readable storage medium comprises instructions, and when the instructions in the computer readable storage medium are executed by an electronic device, the electronic device is enabled to perform the score prediction model training method according to claim 1.

4. A computer-readable storage medium, characterized in that, ​

Images