Service recommendation method, device and electronic equipment
By obtaining and updating the paid feature data set of users in multiple preset periods, and combining the paid feature data of other users, the problem of inaccurate payment level prediction caused by relying on long-term historical features in the prior art is solved, and more accurate Service recommendations improve user experience.
Patent Information
- Application Number
- CN202011368461.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-11-27
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2040-11-27
AI Technical Summary
The existing service recommendation method relies on long-term historical payment characteristics in the payment level prediction, resulting in large fluctuations in the recent payment characteristics of users but small fluctuations in the overall feature, resulting in the long-term unchanged results of the payment level prediction. The recommended service information does not match the current situation of the user, and the user experience Difference.
By obtaining the paid feature data set of users in multiple preset periods, updating the weight according to the interval duration, inputting the classification model to obtain the paid level, and recommending service information based on the paid level, combining the paid feature data of other users to construct the data sequence and Weight adjustment to ensure that the prediction results are in line with the current situation.
Improve the accuracy of paid level prediction, make the recommendation service information more in line with the current user situation and improves the user experience.
Smart Images

Figure CN112446763B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of computer technology, and in particular to a service recommendation method, device and electronic device. Background Art
[0002] With the development of the Internet, the number of services that can be recommended to users has gradually increased. In order to meet user needs, in some fields, such as the gaming field, it is usually necessary to formulate corresponding services based on the user's payment habits to meet user needs. Therefore, it is necessary to predict the user's payment level and recommend corresponding activities or virtual items to users based on different payment habits to improve user experience.
[0003] The existing service recommendation method extracts the user's payment features in the current game and then inserts them into the prediction model to determine the user's payment level based on various payment features, and then recommends corresponding service information based on the user's payment level. In order to improve the accuracy of service recommendations, long-term historical payment features are usually required. However, when using long-term payment features to predict payment levels, due to the large amount of data, even if the user's recent payment features fluctuate greatly, the overall payment features fluctuate less, making the final payment level prediction result may remain unchanged for a long time, which in turn causes the service information recommended to the user to be inconsistent with the user's current actual situation, resulting in a poor user experience. Summary of the invention
[0004] The purpose of this application is to solve at least one of the technical problems existing in the prior art, and to provide a service recommendation method, device and electronic device to improve the accuracy of service recommendation and enhance user experience.
[0005] The present application provides a service recommendation method, including:
[0006] Respond to service recommendation instructions;
[0007] According to the service recommendation instruction, obtaining a first payment feature data set of the target user in multiple preset time periods, wherein the first payment feature data set includes multiple first payment feature data;
[0008] According to the interval between each of the preset time periods and the current moment, the initial weight of each of the first payment feature data sets is updated to obtain the preset weight of each of the first payment feature data sets, wherein the interval is negatively correlated with the preset weight;
[0009] Inputting each of the first payment feature data sets and each of the preset weights into a classification model, and outputting the payment level of the target user based on the classification model;
[0010] According to the payment level, corresponding recommendation information is sent to the target user.
[0011] Furthermore, the step of obtaining a first payment feature data set of a target user in a plurality of preset time periods according to the service recommendation instruction includes:
[0012] According to the service recommendation instruction, the first payment feature data set of the target user in the multiple preset time periods is obtained from each target application, wherein each target application is a target application logged in by the target user through the same user information.
[0013] Furthermore, before the interval between each of the preset time periods and the current time, the method further includes:
[0014] Extracting a second payment feature data set of each non-target user in the preset time period, wherein the second payment feature data set includes a plurality of second payment feature data;
[0015] A data sequence is formed according to the order of the first payment feature data set in each of the second payment feature data sets in the corresponding preset time period, so as to obtain the initial weight of the first payment feature data set in the corresponding preset time period according to the data sequence.
[0016] Furthermore, in the embodiment of the present application, it also includes:
[0017] The payment level of each of the historical users is updated according to the data sequence.
[0018] Furthermore, in the embodiment of the present application, it also includes:
[0019] The associated data of the target user is stored in a training set, wherein the associated data includes each of the first payment feature data sets, a preset weight of each of the first payment feature data sets, and a payment level of the target user.
[0020] Furthermore, storing the associated data of the target user in the training set includes:
[0021] According to the small class samples in the training set, the corresponding first payment feature data is extracted from the first payment feature data set to perform data amplification on the small class samples.
[0022] Furthermore, the outputting the payment level of the target user based on the classification model includes:
[0023] According to the preset weight of each of the first payment feature data sets, each of the first payment feature data sets is weighted by the classification model, so as to obtain the payment level of the target user according to the weighted result.
[0024] Furthermore, the multiple preset time periods constitute a continuous time series.
[0025] Furthermore, in an embodiment of the present application, a service recommendation device is also provided, including:
[0026] An instruction response module, used to respond to a service recommendation instruction;
[0027] A data acquisition module, configured to acquire a first payment feature data set of a target user in a plurality of preset time periods according to the service recommendation instruction, wherein the first payment feature data set includes a plurality of first payment feature data;
[0028] A weight updating module, configured to update the initial weight of each of the first payment feature data sets according to the interval between each of the preset time periods and the current moment, to obtain the preset weight of each of the first payment feature data sets, wherein the interval is negatively correlated with the preset weight;
[0029] a level classification module, configured to input each of the first payment feature data sets and each of the preset weights into a classification model, and output a payment level of the target user based on the classification model;
[0030] The service recommendation module is used to send corresponding recommendation information to the target user according to the payment level.
[0031] Furthermore, the data acquisition module is specifically used for:
[0032] According to the service recommendation instruction, the first payment feature data set of the target user in the multiple preset time periods is obtained from each target application, wherein each target application is a target application logged in by the target user through the same user information.
[0033] Furthermore, an embodiment of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the service recommendation method as described in the above embodiment is implemented.
[0034] Furthermore, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the service recommendation method as described in the above embodiment.
[0035] Compared with the prior art, the above embodiment obtains the payment feature data set of the user in multiple preset time periods, so that there is no need to use long-term payment features to predict the payment level, and then obtains the preset weight of the payment feature data set according to the interval length, so that the obtained payment level can fully reflect the payment characteristics of each time period, and at the same time, the service information recommended based on the payment level can better fit the user's current actual situation, thereby improving the user experience.
[0036] The above embodiment obtains a payment feature data set from multiple target applications logged in by the user terminal through the same user information, so that payment prediction does not need to be performed after obtaining payment feature data for a long period of time. At the same time, the payment feature data obtained can more accurately reflect the user's payment habits, thereby improving the accuracy of payment level prediction, and further recommending more accurate service information to users.
[0037] The above-mentioned embodiment obtains the payment feature data set of other users in a preset time period to form a data sequence with the payment feature data set of the target user, thereby obtaining the initial weight of the payment feature data set of the target user, so that the initial weight of the payment feature data set of the target user can be comprehensively set by considering the levels of payment features of other users, so that the predicted payment level of the target user can be adapted to the current payment situation, thereby improving the accuracy of the prediction and recommending more accurate service information to the user.
[0038] The above embodiment updates the payment level of historical users by forming a data sequence based on the payment feature data set of other users in a preset time period and the payment feature data set of the target user, so as to update the users included in each payment level, improve the accuracy of the subsequent user payment level prediction, and further improve the accuracy of service information recommendation.
[0039] The above embodiment ensures real-time updating and training of the training set by storing the payment feature data set of the target user, the preset weight of the data set and the payment level of the user in the training set, thereby improving the accuracy of service recommendation.
[0040] The above embodiment can increase the training set data accordingly by classifying the small category samples in the training set and extracting the corresponding payment feature data from the payment feature data set to perform data amplification on the small category samples, thereby ensuring the updating training effect of the training set and improving the accuracy of service recommendations.
[0041] The above embodiment improves the accuracy of the payment level prediction result obtained after the payment feature data set is input into the classification model by weighted calculation.
[0042] The above embodiment adopts a continuous time series as the preset time period, so that the obtained payment level can better reflect the user's continuous payment situation, improve the accuracy of payment level prediction, and thus improve the accuracy of service recommendation. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] The present application is further described below in conjunction with the accompanying drawings and embodiments;
[0044] Figure 1 An application environment diagram of a service recommendation method in an embodiment;
[0045] Figure 2 A flowchart of a service recommendation method in one embodiment;
[0046] Figure 3 It is a structural block diagram of a service recommendation device in one embodiment;
[0047] Figure 4 It is a structural block diagram of a service recommendation device in another embodiment;
[0048] Figure 5 FIG. 4 is a structural block diagram of a computer device in one embodiment. DETAILED DESCRIPTION
[0049] This section will describe in detail the specific embodiments of the present application. The preferred embodiments of the present application are shown in the accompanying drawings. The purpose of the accompanying drawings is to supplement the description of the text part of the specification with graphics, so that people can intuitively and vividly understand each technical feature and the overall technical solution of the present application, but it cannot be understood as a limitation on the scope of protection of the present application.
[0050] The existing service recommendation method extracts the user's payment features in the current game and then inserts them into the prediction model to determine the user's payment level based on various payment features, and then recommends corresponding service information based on the user's payment level. In order to improve the accuracy of service recommendations, long-term historical payment features are usually required. However, when using long-term payment features to predict payment levels, due to the large amount of data, even if the user's recent payment features fluctuate greatly, the overall payment features fluctuate less, making the final payment level prediction result may remain unchanged for a long time, which in turn causes the service information recommended to the user to be inconsistent with the user's current actual situation, resulting in a poor user experience.
[0051] In order to solve the above technical problems, Figure 1 FIG. 1 is an application environment diagram of a service recommendation method in an embodiment. Figure 1The service recommendation system includes a user terminal 110 and a server 120. The user terminal 110 and the server 120 are connected via a network. The user terminal 110 may be a desktop user terminal. The server 120 may be implemented as an independent server or a server cluster consisting of multiple servers.
[0052] Below, the service recommendation method provided by the embodiment of the present application will be introduced and explained in detail through several specific embodiments.
[0053] like Figure 2 As shown, in one embodiment, a service recommendation method is provided. This embodiment mainly uses the method applied to a computer device as an example. The computer device can be the above-mentioned Figure 1 The server 120 in.
[0054] Reference Figure 2 , the service recommendation method specifically includes the following steps:
[0055] S11. Respond to the service recommendation instruction.
[0056] In this embodiment, the server receives and responds to the service recommendation instruction, wherein the service recommendation instruction may be sent by the user, and the user may directly input the relevant instruction into the server, or the user may send the service recommendation instruction to the server through the terminal, or the server may automatically generate the service recommendation instruction when receiving the user's characteristic data, including payment characteristic data, from the user terminal. After receiving and responding to the service recommendation instruction, the server begins to execute the following steps. In this embodiment, the manner in which the server receives and responds to the service recommendation instruction is not specifically limited.
[0057] S12. According to the service recommendation instruction, a first payment feature data set of the target user in multiple preset time periods is obtained, where the first payment feature data set includes multiple first payment feature data.
[0058] In this embodiment, the server obtains the first payment characteristic data of the target user in multiple preset time periods from the database of the user terminal, and all the first payment characteristic data obtained constitute a first payment characteristic data set. The first payment characteristic data refers to the relevant data when the user performs actions such as game recharge or commodity purchase.
[0059] In this embodiment, the first payment characteristic data is illustrated by the relevant data of the user's payment behavior for game recharge. For example, the first payment characteristic data may include the user's total recharge amount, the user's recharge times, the user's maximum recharge amount, etc.
[0060] In this embodiment, the time from when the user starts using the terminal to the current moment is defined as the registration duration, wherein the server can select multiple preset time periods from the registration duration according to the service recommendation instruction, or the server can randomly select time periods from the registration duration as preset time periods, and the number of selected preset time periods is specified by the service recommendation instruction. Finally, the server obtains the first payment feature data in multiple preset time periods from the database of the user terminal. For example, if the user starts using the terminal on January 1, 2020, then the preset time periods can be February 1, 2020, March 1, 2020, and May 1, 2020. Therefore, the first payment feature data of the target user obtained in multiple preset time periods can be the specific values of the total recharge amount of the user, the number of user recharges, and the maximum recharge amount of the user within three days of February 1, 2020, March 1, 2020, and May 1, 2020.
[0061] In one embodiment, according to the service recommendation instruction, obtaining a first payment feature data set of a target user in a plurality of preset time periods includes:
[0062] According to the service recommendation instruction, a first payment feature data set of the target user in multiple preset time periods is obtained from each target application, wherein each target application is a target application logged in by the target user through the same user information.
[0063] In this embodiment, the server first obtains the user information of the user logging into the first target application from the database of the user terminal according to the service recommendation instruction, and then obtains the second target application installed in the user terminal and matching the login information from the user terminal according to the user information. The first target application and the second target application are both applications installed in the user terminal, and the first target application and the second target application are target applications logged in by the target user through the same user information. Among them, the first target application can be a target application randomly selected by the server from all target applications installed in the user terminal, or a target application selected by the server according to the service recommendation instruction, and the second target application refers to other target applications installed in the user terminal except the first target application.
[0064] In this embodiment, for the method of logging in with the same user information, the second target application and the first target application can be games in the same game platform, that is, after logging in to the game platform with one game platform account, different game applications can be opened in the game platform, for example, the first target application is the game "God of War Awakening", and the second target application is the game "King City Heroes", and both games are logged in through the platform of Sanqi Online Games; it can also be that when the user registers the first target application and the second target application, the customized login account is the same, for example, the first target application is the game "Cloud City Song", and the second target application is the game "Douluo Continent", and the customized login account when registering these two games is: sanqi123; it can also be that when the user registers the first target application and the second target application, the account of the associated external application is the same, for example, the first target application is the game "Eternal Era", and the second target application is the game "Sword of Archangel", and both games use mobile phone numbers for registration and login. In this embodiment, there is no specific limitation on the way in which the login information of the second target application matches that of the first target application.
[0065] In this embodiment, the server obtains the first payment feature data of the target user in multiple preset time periods from each target application installed in the user terminal, and all the obtained first payment feature data constitute a first payment feature data set.
[0066] In this embodiment, by obtaining a payment feature data set from a target application to which a user logs in using the same user information, the obtained payment feature data can more accurately reflect the user's payment habits, thereby improving the accuracy of payment level predictions and recommending more accurate service information to the user.
[0067] S13. Update the initial weight of each first payment feature data set according to the interval between each preset time period and the current moment to obtain the preset weight of each first payment feature data set, wherein the interval is negatively correlated with the preset weight.
[0068] In order to fully consider the impact of time on paid feature data, it is necessary to set corresponding weights according to the different times of paid feature data sets.
[0069] In this embodiment, the longer the time interval between the preset period and the current moment, that is, the longer the time interval between the preset period and the current moment, the smaller the influence of the first payment feature data in the preset period on the payment level prediction. Therefore, the weight set for the payment feature data in the preset period with a longer time interval from the current moment should be smaller. For example, the preset weight of the first payment feature data set in the preset period on February 1, 2020, that is, the preset weight set for the first payment feature data of the total recharge amount of the user, the number of recharges of the user, and the maximum recharge amount of the user is 1; the preset weight of the first payment feature data set in the preset period on March 1, 2020, that is, the preset weight set for the first payment feature data of the total recharge amount of the user, the number of recharges of the user, and the maximum recharge amount of the user is 2; the preset weight of the first payment feature data set in the preset period on May 1, 2020, that is, the preset weight set for the first payment feature data of the total recharge amount of the user, the number of recharges of the user, and the maximum recharge amount of the user is 4.
[0070] In one embodiment, before determining the interval between each preset time period and the current time, the method further includes:
[0071] A second payment feature data set of each non-target user in a preset time period is extracted, wherein the second payment feature data set includes a plurality of second payment feature data.
[0072] A data sequence is formed according to the order of the first payment feature data set in each second payment feature data set within the corresponding preset time period, so as to obtain the initial weight of the first payment feature data set within the corresponding preset time period according to the data sequence.
[0073] In this embodiment, the second payment characteristic data refers to the payment characteristic data of non-target users within a preset time period, that is, the second payment characteristic data and the first payment characteristic data obtain the same relevant data within the same preset time period, and the difference is that the second payment characteristic data and the first payment characteristic data come from different users. Among them, after the server obtains the second payment characteristic data from the database of the user terminal, the second payment characteristic data constitutes the second payment characteristic data set. For example, the specific values of the total recharge amount, the number of user recharges, and the maximum recharge amount of users other than the target user within three days of February 1, 2020, March 1, 2020, and May 1, 2020 are obtained, and the above second payment characteristic data constitutes the second payment characteristic data set.
[0074] In this embodiment, the first payment feature data and each second payment feature data, that is, the first payment feature data of the user of the payment level to be predicted in the same preset time period and the second payment feature data of each non-target user are sorted, for example, the first payment feature data of the target user in the preset time period of February 1, 2020 and the second payment feature data of each non-target user are sorted to obtain a data sequence. Specifically, the first payment feature data set includes the total recharge amount of the target user, the number of user recharges, and the maximum recharge amount of the user. For the total recharge amount of the user, the first payment feature data is 500 yuan, and the second payment feature data includes the payment feature data of three non-target users, which are 5000, 100, and 800 respectively; therefore, the data sequence obtained after sorting is 5000, 800, 500, and 100. The first payment feature data "the total recharge amount of the user is 500 yuan" ranks third in the sequence, and the initial weight can be set in the form of adding 1 from low to high weights, so the initial weight of the data sequence is 4, 3, 2, 1, and the initial weight corresponding to the first payment feature data is 2. For the number of times the user recharges, the first payment feature data is 2 times, and the second payment feature data includes the payment feature data of three non-target users, which are 3, 3, and 5 respectively; therefore, the data sequence obtained after sorting is 5, 3, 3, and 2. The first payment feature data "the number of times the user recharges is 2 times" ranks fourth in the sequence, and the initial weight can be set in the form of weight plus 1 from low to high, so the initial weight of the data sequence is 4, 3, 2, 1, and the initial weight corresponding to the first payment feature data is 1. For the maximum recharge amount of the user, the first payment feature data is 400 yuan, and the second payment feature data includes the payment feature data of three non-target users, which are 2000, 50, and 300 respectively; therefore, the data sequence obtained after sorting is 2000, 400, 300, and 50. The first payment feature data "the maximum recharge amount of the user is 400 yuan" ranks second in the sequence, and the initial weight can be set in the form of weight plus 1 from low to high, so the initial weight of the data sequence is 4, 3, 2, 1, and the initial weight corresponding to the first payment feature data is 3.
[0075] In this embodiment, after obtaining the initial weight of the first payment feature data of the target user in the preset time period based on the data sequence composed of the first payment feature data of the target user and the second payment feature data of the non-target user in the same preset time period, the initial weight of each first payment feature data set is updated according to the interval between each preset time period and the current moment, thereby obtaining the preset weight of each first payment feature data set. For example, in the preset time period of February 1, 2020, the first payment feature data set includes the first payment feature data of "the total recharge amount of the user is 500 yuan", the first payment feature data of "the number of recharges of the user is 2 times", and the first payment feature data of "the maximum recharge amount of the user is 400 yuan", and the corresponding initial weights are 2, 1, and 3 respectively. The initial weight of the corresponding first payment feature data set can be obtained by averaging, so the initial weight of the corresponding first payment feature data set is 2. However, the interval between the preset time period and the current moment is relatively large, so the initial weight of the first paid feature data set is reduced, that is, the initial weight of the first paid feature data set in the preset time period on February 1, 2020 is multiplied by 0.5, and the final preset weight of the first paid feature data set in the preset time period on February 1, 2020 is 1, that is, the preset weights of the first paid feature data "the total recharge amount of the user is 500 yuan", the first paid feature data "the number of user recharges is 2 times", and the first paid feature data "the maximum recharge amount of the user is 400 yuan" are 1, 0.5, and 1.5 respectively.
[0076] In this embodiment, by obtaining the payment feature data set of other users in a preset time period, a data sequence is formed with the payment feature data set of the target user, thereby obtaining the initial weight of the payment feature data set of the target user, so that the initial weight of the payment feature data set of the target user can be comprehensively set by considering the levels of payment features of other users, so that the predicted payment level of the target user can be adapted to the current payment situation, thereby improving the accuracy of the prediction, and recommending more accurate service information to the user.
[0077] S14: Input each first payment feature data set and each preset weight into a classification model, and output a payment level of the target user based on the classification model.
[0078] In this embodiment, the first payment feature data set of the target user for multiple preset time periods and the preset weights corresponding to the first payment feature data set of the target user for each preset time period are obtained from the database installed in the user terminal and input into the classification model. The classification model can be based on a long short-term memory neural network (Long-Short Term Memory, LSTM). By inputting the first payment feature data such as the total recharge amount of the target user in each preset time period, the number of user recharges, and the maximum recharge amount of the user into the LSTM model, the LSTM model calculates the corresponding probability when the target user is at different payment levels. For example, the payment level of the target user can be divided into high-payment level users, medium-payment level users, and low-payment level users. The probabilities corresponding to the three categories output by the LSTM model are 75%, 20%, and 5%, respectively. Therefore, it can be judged that the payment level of the target user is a high-payment level user.
[0079] S15. Send corresponding recommendation information to the target user according to the payment level.
[0080] In this embodiment, the server sends corresponding recommendation information according to the predicted payment level of the target user. For example, when the payment level of the target user is a high-paying user, the server sends the target user recommendation information such as advanced skin purchase, advanced treasure box purchase, and high-level forging purchase.
[0081] In one embodiment, the service recommendation method further includes:
[0082] Based on the data sequence, the payment level of each historical user is updated.
[0083] In this embodiment, for the total recharge amount of the user, the original data sequence of the corresponding second payment feature data is 5000, 800, 100. The initial weight can be set in the form of weight plus 1 from low to high, so the initial weight of the original data sequence is 3, 2, 1. When the first payment feature data "the total recharge amount of the user is 500 yuan" corresponding to the total recharge amount of the user is updated, the data sequence corresponding to the total recharge amount of the user is 5000, 800, 500, 100. Therefore, the initial weights corresponding to the new data sequence are 4, 3, 2, 1, and the new initial weights corresponding to the second payment feature data 5000, 800, 100 are 4, 3, 1.
[0084] For the number of times a user recharges, the original data sequence of the corresponding second payment feature data is 5, 3, 3. The initial weight can be set in the form of adding 1 from low to high weights, so the initial weights of the original data sequence are 3, 1, 1. When the data sequence of the first payment feature data "the total recharge amount of the user is 2 times" corresponding to the number of times a user recharges is updated, the data sequence corresponding to the number of times a user recharges is 5, 3, 3, 2. Therefore, the initial weights corresponding to the new data sequence are 4, 2, 2, 1, and the new initial weights corresponding to the second payment feature data 5, 3, 3 are 4, 2, 2.
[0085] For the highest recharge amount of the user, the original data sequence of the corresponding second payment feature data is 2000, 300, 50. The initial weight can be set in the form of weight plus 1 from low to high, so the initial weight of the original data sequence is 3, 2, 1. When the first payment feature data "the highest recharge amount of the user is 400 yuan" corresponding to the number of user recharges is updated, the data sequence corresponding to the highest recharge amount of the user is 2000, 400, 300, 50. Therefore, the initial weights corresponding to the new data sequence are 4, 3, 2, 1, and the new initial weights corresponding to the second payment feature data 2000, 300, 50 are 4, 2, 1.
[0086] Then, the initial weight of the second payment feature data set is adjusted according to the interval between the preset time period and the current time to obtain the preset weight, and then the payment level of each historical user is updated according to the preset weight, for example, whether it is a high-paying level user.
[0087] In this embodiment, the payment level of historical users is updated by forming a data sequence based on the payment feature data set of other users in a preset time period and the payment feature data set of the target user, so that the users included in each payment level are updated, thereby improving the accuracy of the prediction of the payment level of subsequent users, and further improving the accuracy of service information recommendation.
[0088] Furthermore, the service recommendation method further includes:
[0089] The associated data of the target user is stored in the training set, wherein the associated data includes each first payment feature data set, a preset weight of each first payment feature data set, and a payment level of the target user.
[0090] In this embodiment, after the payment level prediction of user A whose payment level is to be predicted is completed, for user B whose payment level is to be predicted next, the first payment feature data set used for payment level prediction of user A whose payment level is to be predicted, the preset weights of each first payment feature data set, and the payment level of user A obtained are changed to the payment feature data set of historical users, the preset weights of each payment feature data set, and the payment level. Therefore, the associated data of the above target users are stored in the training set, thereby ensuring real-time update training of the training set and gradually improving the accuracy of user classification.
[0091] In one embodiment, storing the associated data of the target user into the training set includes:
[0092] According to the small class samples in the training set, corresponding first paid feature data is extracted from the first paid feature data set to perform data amplification on the small class samples.
[0093] Since too little data in the training set will affect the training effect of the classification model, in this embodiment, by classifying the small category samples in the training set and extracting the corresponding paid feature data from the paid feature data set to perform data augmentation, the training set is increased accordingly to ensure the updated training effect of the training set and improve the accuracy of user classification.
[0094] In one embodiment, outputting the payment level of the target user based on the classification model includes:
[0095] According to the preset weights of the first payment feature data sets, each first payment feature data set is weighted by a classification model, so as to obtain the payment level of the target user according to the weighted result.
[0096] In this embodiment, the payment level is predicted by inputting the first payment feature data set of the target user into the classification model to obtain the probability that each first payment feature data is a different payment level, and the obtained probabilities are weighted by preset weights to finally determine the payment level of the target user. For example, the first payment feature data includes the user's total recharge amount of 500 yuan, the user's recharge times of 2 times, and the user's maximum recharge amount of 400 yuan. The corresponding preset weights are 1, 0.5, and 1.5. The above first payment feature data is input into the classification model to predict the payment level. The result of the first payment feature data "the user's total recharge amount is 500 yuan" is that the probability of a high payment level user is 50%, the probability of a medium payment level user is 25%, and the probability of a low payment level user is 25%; the result of the first payment feature data "the user's recharge times is 2 times" is that the probability of a high payment level user is 20%, the probability of a medium payment level user is 20%, and the probability of a low payment level user is 60%; the result of the first payment feature data "the user's maximum recharge amount is 400 yuan" is that the probability of a high payment level user is 80%, the probability of a medium payment level user is 20%, and the probability of a low payment level user is 0%. Therefore, after weighted calculation, the value of the user being a high-paying level user is 1.8, the value of the user being a medium-paying level user is 0.65, and the value of the user being a low-paying level user is 0.55. Therefore, the user's payment level is a high-paying level user.
[0097] In this embodiment, a weighted calculation method is used to improve the accuracy of the payment level prediction result obtained after the payment feature data set is input into the classification model.
[0098] In one embodiment, a plurality of preset time periods constitute a continuous time sequence.
[0099] In this embodiment, the multiple preset time periods selected from the registration time can be three consecutive days from February 1 to February 3, 2020, and each day is regarded as a preset time period. When the preset time period is a continuous time series, the interval time between each preset time period and the current moment will not differ much, so that the process of modifying the initial weight to obtain the preset weight is basically the same, so that the obtained payment level can better reflect the user's continuous payment situation, improve the accuracy of the payment level prediction, and thus improve the accuracy of service recommendations.
[0100] In the above embodiment, by obtaining the payment feature data set of the user in multiple preset time periods, there is no need to use long-term payment features to predict the payment level, and then the preset weight of the payment feature data set is obtained according to the interval length, so that the obtained payment level can fully reflect the payment characteristics of each time period, and at the same time, the service information recommended based on the payment level can better fit the user's current actual situation, thereby improving the user experience.
[0101] In one embodiment, Figure 3 As shown, a service recommendation device is provided, comprising:
[0102] The instruction response module 101 is used to respond to the service recommendation instruction.
[0103] The data acquisition module 102 is used to acquire a first payment feature data set of a target user in a plurality of preset time periods according to the service recommendation instruction, wherein the first payment feature data set includes a plurality of first payment feature data.
[0104] The weight updating module 103 is used to update the initial weight of each of the first payment feature data sets according to the interval length between each of the preset time periods and the current moment, and obtain the preset weight of each of the first payment feature data sets, wherein the interval length is negatively correlated with the preset weight.
[0105] The level classification module 104 is used to input each of the first payment feature data sets and each of the preset weights into a classification model, and output the payment level of the target user based on the classification model.
[0106] The service recommendation module 105 is used to send corresponding recommendation information to the target user according to the payment level.
[0107] In one embodiment, a plurality of preset time periods constitute a continuous time sequence.
[0108] In one embodiment, the data acquisition module 102 is further used to:
[0109] According to the service recommendation instruction, a first payment feature data set of the target user in multiple preset time periods is obtained from each target application, wherein each target application is a target application logged in by the target user through the same user information.
[0110] In one embodiment, the weight updating module 103 is further configured to:
[0111] Before the interval between each preset time period and the current moment, a second payment feature data set of each non-target user in the preset time period is extracted, wherein the second payment feature data set includes a plurality of second payment feature data.
[0112] A data sequence is formed according to the order of the first payment feature data set in each second payment feature data set within the corresponding preset time period, so as to obtain the initial weight of the first payment feature data set within the corresponding preset time period according to the data sequence.
[0113] In one embodiment, the weight updating module 103 is further configured to:
[0114] Based on the data sequence, the payment level of each historical user is updated.
[0115] In one embodiment, the level classification module 104 is further configured to:
[0116] According to the preset weights of the first payment feature data sets, each first payment feature data set is weighted by a classification model, so as to obtain the payment level of the target user according to the weighted result.
[0117] In another embodiment, Figure 4 As shown, a service recommendation device is provided, comprising:
[0118] The instruction response module 101 is used to respond to the service recommendation instruction.
[0119] The data acquisition module 102 is used to acquire a first payment feature data set of a target user in a plurality of preset time periods according to the service recommendation instruction, wherein the first payment feature data set includes a plurality of first payment feature data.
[0120] The weight updating module 103 is used to update the initial weight of each of the first payment feature data sets according to the interval length between each of the preset time periods and the current moment, and obtain the preset weight of each of the first payment feature data sets, wherein the interval length is negatively correlated with the preset weight.
[0121] The level classification module 104 is used to input each of the first payment feature data sets and each of the preset weights into a classification model, and output the payment level of the target user based on the classification model.
[0122] The service recommendation module 105 is used to send corresponding recommendation information to the target user according to the payment level.
[0123] The data storage module 106 is used to store the associated data of the target user into the training set, wherein the associated data includes each first payment feature data set, the preset weight of each first payment feature data set, and the payment level of the target user. According to the small class samples in the training set, the corresponding first payment feature data is extracted from the first payment feature data set to perform data amplification of the small class samples.
[0124] In one embodiment, a computer device is provided, such as Figure 5As shown, the computer device includes a processor, a memory, a network interface, an input device and a display screen connected via a system bus. The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program. When the computer program is executed by the processor, the processor may implement the service recommendation method. The internal memory may also store a computer program. When the computer program is executed by the processor, the processor may implement the service recommendation method. Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0125] In one embodiment, the service recommendation device provided by the present application can be implemented in the form of a computer program. Figure 5 The computer device shown in the figure is run. The memory of the computer device can store various program modules constituting the service recommendation device. The computer program composed of various program modules enables the processor to execute the steps of the service recommendation method of various embodiments of the present application described in this specification.
[0126] In one embodiment, an electronic device is provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the steps of the above service recommendation method when executing the program. The steps of the service recommendation method here may be the steps of the service recommendation method in the above embodiments.
[0127] In one embodiment, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer-executable instructions, wherein the computer-executable instructions are used to cause a computer to execute the steps of the above service recommendation method. The steps of the service recommendation method here may be the steps of the service recommendation method in the above embodiments.
[0128] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications are also considered to be within the scope of protection of the present application.
[0129] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.
Claims
1. A service recommendation method, applied to a server, characterized in that: include: Respond to service recommendation instructions; According to the service recommendation instruction, obtaining a first payment feature data set of the target user in multiple preset time periods, wherein the first payment feature data set includes multiple first payment feature data; Extracting a second payment feature data set of each non-target user in the preset time period, wherein the second payment feature data set includes a plurality of second payment feature data; forming a data sequence according to the order of the first payment feature data set in each of the second payment feature data sets in the corresponding preset time period, so as to obtain the initial weight of the first payment feature data set in the corresponding preset time period according to the data sequence; updating the initial weight of each of the first payment feature data sets according to the interval length between each of the preset time periods and the current moment, and obtaining the preset weight of each of the first payment feature data sets, wherein the interval length is negatively correlated with the preset weight; Inputting each of the first payment feature data sets and each of the preset weights into a classification model, and outputting the payment level of the target user based on the classification model; According to the payment level, corresponding recommendation information is sent to the target user.
2. The service recommendation method according to claim 1, characterized in that: The step of obtaining a first payment feature data set of a target user in a plurality of preset time periods according to the service recommendation instruction includes: According to the service recommendation instruction, the first payment feature data set of the target user in the multiple preset time periods is obtained from each target application, wherein each target application is a target application logged in by the target user through the same user information.
3. The service recommendation method according to claim 1, characterized in that: Also includes: The payment level of each historical user is updated according to the data sequence.
4. The service recommendation method according to claim 1, characterized in that: Also includes: The associated data of the target user is stored in a training set, wherein the associated data includes each of the first payment feature data sets, a preset weight of each of the first payment feature data sets, and a payment level of the target user.
5. The service recommendation method according to claim 4, characterized in that: The storing the associated data of the target user into the training set includes: According to the small class samples in the training set, the corresponding first payment feature data is extracted from the first payment feature data set to perform data amplification on the small class samples.
6. The service recommendation method according to claim 1 or 2, characterized in that: The outputting the payment level of the target user based on the classification model includes: According to the preset weight of each of the first payment feature data sets, each of the first payment feature data sets is weighted by the classification model to obtain the payment level of the target user according to the weighted result.
7. The service recommendation method according to claim 1, characterized in that: The plurality of preset time periods constitute a continuous time series.
8. A service recommendation device, characterized in that: include: An instruction response module, used to respond to a service recommendation instruction; A data acquisition module, configured to acquire a first payment feature data set of a target user in a plurality of preset time periods according to the service recommendation instruction, wherein the first payment feature data set includes a plurality of first payment feature data; A weight updating module is used to extract a second payment feature data set of each non-target user in the preset time period, wherein the second payment feature data set includes a plurality of second payment feature data; a data sequence is formed according to the order of the first payment feature data set in each of the second payment feature data sets in the corresponding preset time period, so as to obtain the initial weight of the first payment feature data set in the corresponding preset time period according to the data sequence; the initial weight of each of the first payment feature data sets is updated according to the interval length between each of the preset time periods and the current moment, so as to obtain the preset weight of each of the first payment feature data sets, wherein the interval length is negatively correlated with the preset weight; a level classification module, configured to input each of the first payment feature data sets and each of the preset weights into a classification model, and output a payment level of the target user based on the classification model; The service recommendation module is used to send corresponding recommendation information to the target user according to the payment level.
9. The service recommendation device according to claim 8, characterized in that: The data acquisition module is specifically used for: According to the service recommendation instruction, the first payment feature data set of the target user in the multiple preset time periods is obtained from each target application, wherein each target application is a target application logged in by the target user through the same user information.
10. An electronic device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the service recommendation method according to any one of claims 1 to 6 when executing the computer program.
