User value classification method and device based on double-layer sub-classification model

Abstract

The embodiment of the present application discloses a user value classification method and device based on a two-layer sub-classification model, the method comprising: obtaining user feature data and object feature data, associating the user feature data with the object feature data to obtain a first feature data matrix and a second feature data matrix; inputting the user feature data into the first sub-classification model, classifying the user value level to obtain a user level classification, and setting a corresponding second sub-classification model for each level classification; using the first feature data matrix to train the second sub-classification model corresponding to each level classification to obtain a user group model corresponding to each level classification; predicting relative user value based on the user level classification, the second feature data matrix, and the user group model corresponding to each level classification. The present application classifies the user value level to obtain a user level classification, thereby improving the accuracy of predicting user value.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of Internet technology, and specifically to a user value classification method and device based on a double-layer sub-classification model. Background Art

[0002] With the rapid development of artificial intelligence and machine learning, machine learning models are being used for business analysis in a variety of business scenarios. For example, in many application scenarios, users need to be classified and identified, such as identifying the risk level of users and distinguishing the groups to which users belong. To this end, it is often necessary to train user classification models to perform business-related user identification and classification. Although the user classification system based on large-scale collaborative filtering algorithms can make up for the shortcomings of manual strategies to a certain extent through automation, the accuracy of its algorithm in predicting user value still needs to be improved. Summary of the invention

[0003] In view of the above problems, the present application proposes a user value classification method, apparatus, computing device and computer storage medium based on a two-layer sub-classification model to solve the following problems: the existing user value prediction method has low prediction accuracy.

[0004] According to one aspect of an embodiment of the present application, a user value classification method based on a two-layer sub-classification model is provided, comprising:

[0005] Acquire user feature data and object feature data, and associate the user feature data with the object feature data to obtain a first feature data matrix and a second feature data matrix;

[0006] Inputting user feature data into the first sub-classification model, classifying the user value level using the first sub-classification model to obtain a user level classification, and setting a corresponding second sub-classification model for each level classification in the user level classification;

[0007] Using the first feature data matrix, the second sub-classification model corresponding to each hierarchical classification is trained to obtain a user group model corresponding to each hierarchical classification;

[0008] The relative user value is predicted based on the user level classification, the second feature data matrix and the user group model corresponding to each level classification in the user level classification; wherein the relative user value is the user value of the user relative to the object to be predicted.

[0009] Furthermore, obtaining object feature data further includes:

[0010] The original object feature data of each object in multiple dimensions is obtained, and the original object feature data is linearly encoded to obtain the object feature data.

[0011] Furthermore, obtaining user characteristic data further includes:

[0012] The original user characteristic data of each user is obtained, and the original user characteristic data is supplemented and denoised to obtain the user characteristic data.

[0013] Furthermore, associating the user feature data with the object feature data to obtain a first feature data matrix and a second feature data matrix further includes:

[0014] For each user's user feature data, determine the objects that the user has interacted with according to the historical object interaction data in the user feature data, establish an association relationship between the user feature data of the user and the object feature data of the objects that the user has interacted with, and obtain a first feature data matrix;

[0015] For the user feature data of each user, an association relationship is established between the user feature data of the user and the object feature data of the object to be predicted, so as to obtain a second feature data matrix.

[0016] Furthermore, the second sub-classification model corresponding to each hierarchical classification is trained using the first feature data matrix to obtain a user group model corresponding to each hierarchical classification, further comprising:

[0017] For each hierarchical classification, partial feature data corresponding to the users in the hierarchical classification is extracted from the first feature data matrix, and the second sub-classification model corresponding to the hierarchical classification is trained using the partial feature data to obtain a user group model corresponding to the hierarchical classification.

[0018] Furthermore, predicting the relative user value according to the user level classification, the second feature data matrix, and the user group model corresponding to each level classification in the user level classification further includes:

[0019] According to the user level classification, determine the users included in each level classification;

[0020] For each hierarchical classification, the second feature data matrix of the users included in the hierarchical classification is input into the user group model corresponding to the hierarchical classification for prediction, so as to obtain the user value prediction result of the users included in the hierarchical classification relative to the object to be predicted;

[0021] Summarize the user value prediction results of users included in each level classification relative to the object to be predicted.

[0022] Further, the first sub-classification model and the second sub-classification model are models obtained based on a regression model.

[0023] According to another aspect of an embodiment of the present application, a user value classification device based on a double-layer sub-classification model is provided, comprising:

[0024] An association module, used to obtain user feature data and object feature data, and associate the user feature data with the object feature data to obtain a first feature data matrix and a second feature data matrix;

[0025] A user classification module, used to input user feature data into a first sub-classification model, classify the user value level using the first sub-classification model to obtain a user level classification, and set a corresponding second sub-classification model for each level classification in the user level classification;

[0026] A model training module, used to train the second sub-classification model corresponding to each hierarchical classification using the first feature data matrix to obtain a user group model corresponding to each hierarchical classification;

[0027] The prediction module is used to predict the relative user value based on the user level classification, the second feature data matrix and the user group model corresponding to each level classification in the user level classification; wherein the relative user value is the user value of the user relative to the object to be predicted.

[0028] According to another aspect of the embodiment of the present application, there is provided a computing device, including: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0029] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned user value classification method based on the two-layer sub-classification model.

[0030] According to another aspect of the embodiments of the present application, a computer storage medium is provided, in which at least one executable instruction is stored, and the executable instruction enables a processor to perform operations corresponding to the above-mentioned user value classification method based on a two-layer sub-classification model.

[0031] According to a user value classification method and device based on a two-layer sub-classification model provided in an embodiment of the present application, by associating user feature data and object feature data, the obtained first feature data matrix and second feature data matrix can well meet the cold start function requirements, so as to quickly perform model training and model prediction, and effectively improve the model training and prediction efficiency; and, a classification architecture including a two-layer sub-classification model is constructed, the user value level is classified by the first sub-classification model to obtain the user level classification, and the second sub-classification model corresponding to each level classification is trained using the first feature data matrix, thereby achieving precise parameter adjustment of the second sub-classification model for each level classification, so that the user group model corresponding to each level classification obtained through training has a high prediction accuracy and can accurately predict the user value.

[0032] The above description is only an overview of the technical solution of the embodiment of the present application. In order to more clearly understand the technical means of the embodiment of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the embodiment of the present application more obvious and easy to understand, the specific implementation method of the embodiment of the present application is listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only used for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the embodiments of the present application. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0034] Figure 1 A schematic diagram of a process of a user value classification method based on a double-layer sub-classification model according to an embodiment of the present application is shown;

[0035] Figure 2 A schematic diagram of a process of a user value classification method based on a double-layer sub-classification model according to another embodiment of the present application is shown;

[0036] Figure 3 A schematic diagram of the PS-SMART model structure according to another embodiment of the present application is shown;

[0037] Figure 4a A schematic diagram of a model training process according to another embodiment of the present application is shown;

[0038] Figure 4b A schematic diagram of a model prediction process according to another embodiment of the present application is shown;

[0039] Figure 5 A structural block diagram of a user value classification device based on a double-layer sub-classification model according to an embodiment of the present application is shown;

[0040] Figure 6 A schematic diagram of the structure of a computing device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0041] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0042] First, the terms involved in one or more embodiments of the present application are explained.

[0043] Gradient Boosting Decision Tree (GBDT): An iterative decision tree algorithm consisting of multiple decision trees. The conclusions of all trees are accumulated as the final answer. GBDT can be used for commodity price regression prediction, commodity multi-classification, etc.

[0044] Parameter-server Scalable Multiple Additive Regression Tree (PS-SMART): An iterative implementation of GBDT, in which the parameter server PS (Parameter Server) is committed to solving large-scale offline and online training tasks, and SMART (Scalable Multiple Additive Regression Tree) is an iterative algorithm implemented by GBDT based on PS.

[0045] Figure 1 A schematic diagram of a user value classification method based on a double-layer sub-classification model according to an embodiment of the present application is shown. Figure 1 As shown, the method includes the following steps S101 to S104:

[0046] Step S101: acquiring user feature data and object feature data, and associating the user feature data with the object feature data to obtain a first feature data matrix and a second feature data matrix.

[0047] Among them, the user feature data of each user and the object feature data of each object can be obtained from the database, etc. The object can be a game, a store, a book, a commodity, etc., which is not limited here. Considering that the original object feature data of each object may have inconsistent encoding, in order to facilitate model training and prediction, the original object feature data of each object can be encoded and processed uniformly to obtain object feature data that can be used for model training and prediction. Specifically, the original object feature data of each object in multiple dimensions can be obtained, and the original object feature data can be linearly encoded by label encoding (LabelEncoder) technology to obtain object feature data, so as to make its classification rules consistent.

[0048] Specifically, user feature data may include one or more of the following data: user device type data, user time zone data, user basic data, historical object interaction data; object feature data may include one or more of the following data: object category data, object label data, user magnitude data, manufacturer data, operator data. Those skilled in the art may set user feature data and object feature data to include other data according to actual needs, which is not specifically limited here.

[0049] Among them, user device type data is used to characterize the device type used by the user; user time zone data is used to represent the earth time zone where the user is located; user basic data may include basic data such as user gender and age; historical object interaction data may include interaction data such as historical consumption data and historical usage data of the user for the object. Object category data refers to the category to which the object belongs; object label data refers to the label data set for the object; user magnitude data refers to the number of users owned by the object, for example, user magnitude data can be measured in units of 100,000, 1 million or 10 million; manufacturer data is used to represent the issuing manufacturer of the object or the manufacturer of the object; operator data is used to represent the operator of the object.

[0050] After obtaining the user feature data and the object feature data, the user feature data is associated with the object feature data to obtain a first feature data matrix and a second feature data matrix. The first feature data matrix is ​​different from the second feature data matrix, and the first feature data matrix is ​​used in the model training process, and the second feature data matrix is ​​used in the model prediction process. Specifically, the first feature data matrix refers to a data matrix obtained by associating the user feature data of the user with the object feature data of the object with which the user has interacted, and the second feature data matrix refers to a data matrix obtained by associating the user feature data of the user with the object feature data of the object to be predicted. The first feature data matrix and the second feature data matrix can well meet the cold start function requirements, so as to quickly perform model training and model prediction, and effectively improve the efficiency of model training and prediction.

[0051] Step S102: Input the user feature data into the first sub-classification model, use the first sub-classification model to classify the user value level to obtain the user level classification, and set a corresponding second sub-classification model for each level classification in the user level classification.

[0052] In this step, the user value level can be pre-classified by the first sub-classification model to obtain the user level classification, and a corresponding second sub-classification model can be set for each level classification in the user level classification; wherein the user level classification can specifically include multiple level classifications. Specifically, each level classification records the users belonging to the level classification, and the user level classification can clearly know which users are included in each level classification.

[0053] For example, the first sub-classification model can be used to classify the user value level of each user from high to low into three hierarchical categories based on user characteristic data, and three corresponding second sub-classification models are set for the three hierarchical categories, that is, the hierarchical categories and the second sub-classification models correspond one to one.

[0054] Step S103: using the first feature data matrix to train the second sub-classification model corresponding to each hierarchical classification, to obtain a user group model corresponding to each hierarchical classification.

[0055] Specifically, for each hierarchical classification, partial feature data corresponding to the users in the hierarchical classification is extracted from the first feature data matrix, and the second sub-classification model corresponding to the hierarchical classification is trained using the partial feature data. After multiple iterative trainings, the user group model corresponding to the hierarchical classification is finally obtained.

[0056] Compared with setting a second sub-classification model for all users for training, this embodiment trains the user group model by setting a corresponding second sub-classification model for each level classification, thereby achieving precise parameter adjustment of the second sub-classification model for each level classification, so that the user group models corresponding to each level classification obtained through training have higher prediction accuracy.

[0057] Step S104: predicting the relative user value according to the user level classification, the second feature data matrix and the user group model corresponding to each level classification in the user level classification.

[0058] Among them, the relative user value is the user value of the user relative to the object to be predicted. In this step, according to each level classification in the user level classification, the corresponding user group model is activated; the corresponding part of the feature data of the users included in each level classification in the second feature data matrix is ​​input into the user group model corresponding to the level classification for prediction, so as to obtain the user value prediction result of the corresponding user relative to the object to be predicted, and realize the accurate prediction of user value.

[0059] According to a user value classification method based on a two-layer sub-classification model provided in an embodiment of the present application, by associating user feature data and object feature data, the obtained first feature data matrix and second feature data matrix can well meet the cold start function requirements, so as to quickly perform model training and model prediction, and effectively improve the model training and prediction efficiency; and, a classification architecture including a two-layer sub-classification model is constructed, the user value level is classified by the first sub-classification model to obtain the user level classification, and the second sub-classification model corresponding to each level classification is trained using the first feature data matrix, thereby achieving precise parameter adjustment of the second sub-classification model for each level classification, so that the user group model corresponding to each level classification obtained through training has a high prediction accuracy and can accurately predict the user value.

[0060] Figure 2 A schematic diagram of a process of a user value classification method based on a double-layer sub-classification model according to another embodiment of the present application is shown. Figure 2 As shown, the method includes the following steps S201 to S209:

[0061] Step S201: Acquire user characteristic data and object characteristic data.

[0062] The object feature data includes one or more of the following data: object category data, object label data, user level data, manufacturer data, and operator data. In this embodiment, taking the object as an example of a game, obtaining the object feature data is to obtain the object feature data of the relevant game, that is, the game feature data; wherein the object category data may include data such as game category, game play, game elements, game recharge, etc.; the object label data may include tags set for the game, such as two-player battle tags, etc.; the user level data is the number of people playing the game, generally measured in hundreds of thousands, millions, or tens of millions; the manufacturer data is the game manufacturer; the operator data is the game operator; wherein the game category may include role-playing game category or strategy game category, etc.; the game play may include chess and card gameplay, battle gameplay, etc.; the game elements may include ancient style elements, two-dimensional elements, modern elements, and technological elements, etc.; the game recharge data is used to indicate whether the game needs to be recharged.

[0063] Among them, the user characteristic data includes one or more of the following data: user device type data, user time zone data, user basic data, and historical object interaction data. Taking the user characteristic data of the game as an example, the user device type data is the type of device used by the user when participating in the corresponding game; the user time zone data is the earth time zone of the user's location; the user basic data may include basic data such as the user's gender and age; the historical object interaction data may include the user's historical consumption data for each game, the user's historical login data for each game, etc.; for example, the user's historical consumption data for each game may include the user's recharge amount, recharge time, etc. for each game; the user's historical login data for each game may include: the user's historical login time for each game per day, the game data with the longest login time of the user in a single day, the game data with the longest login time of the user in the past three days, or the game with the longest login time of the user in the past seven days, etc.

[0064] Considering that the original object feature data of objects in multiple dimensions may have the problem of inconsistent classification, and thus cannot be directly input into the model for training or application, in order to associate user feature data with object feature data, after obtaining the original object feature data of each object in multiple dimensions, the original object feature data can be linearly encoded through LabelEncoder technology to obtain object feature data. LabelEncoder technology can convert the text corresponding to the original object feature data into a numerical value of 0 or 1; specifically, taking games as an example, the original object feature data of games, such as game categories, game tags, game operators, game channels and other game feature data, can be linearly encoded to make their classification rules consistent.

[0065] For the processing of user feature data, after obtaining the original user feature data of each user, the original user feature data can be supplemented and denoised by filling missing values, deleting null values, etc. to obtain user feature data. User feature data can be processed through the mapping (Map) and reduction (Reduce) model, and the data of the data detail layer (DataWareHouse Detail, DWD) and the data middle layer (Data WareHouse Middle, DWM) are aggregated into the data service layer (Data WareHouseServce, DWS) that can be used for model training.

[0066] Step S202: for each user's user characteristic data, determine the objects that the user has interacted with according to the historical object interaction data in the user characteristic data.

[0067] In this step, for each user's user feature data, the objects that the user has interacted with are determined based on the user's historical consumption data on each object, the user's historical login data on each object, etc., and an association relationship is established between the user feature data of the user and the object feature data of the objects that the user has interacted with, to obtain a first feature data matrix; for example, the games that the user has consumed are determined based on the user's historical consumption data on each game.

[0068] Step S203: establishing an association relationship between the user feature data of the user and the object feature data of the objects with which the user has interacted, and obtaining a first feature data matrix.

[0069] In this step, taking the object as an example of a game, an association relationship between the user feature data of the user and the object feature data of the game that the user has interacted with historically can be established to obtain a first feature data matrix.

[0070] Step S204: establishing an association relationship between the user feature data of the user and the object feature data of the object to be predicted, and obtaining a second feature data matrix.

[0071] In this step, an association relationship is established between the user feature data of the user and the object feature data of the game to be predicted, and a second feature data matrix is ​​obtained; for example, the game to be predicted can be a game that the user has not interacted with in the past and is newly launched on the platform. For different application scenarios, those skilled in the art can select the object to be predicted according to actual needs, which is not limited here.

[0072] Step S205: Input the user feature data into the first sub-classification model, use the first sub-classification model to classify the user value level to obtain a user level classification, and set a corresponding second sub-classification model for each level classification in the user level classification.

[0073] In an optional implementation, the first sub-classification model and the second sub-classification model are models obtained based on a regression model. The regression model may include a GBDT model, a PS-SMART model, or a linear regression model.

[0074] Specifically, the PS-SMART model supports tasks with tens of billions of samples or even hundreds of thousands of features, and can provide prediction data for a large number of users at the same time, thus making up for the shortcomings of human decision-making in this regard. Figure 3 FIG. 4 shows a schematic diagram of a PS-SMART model structure according to another embodiment of the present application. Figure 3 As shown in the figure, the PS-SMART master node distributes tasks to sub-nodes on different machines. Each sub-node is iteratively trained in turn by sampling data, and finally returns the GBDT model parameters after n iterations of tuning.

[0075] In this embodiment, considering that the user feature data may reach tens of millions to hundreds of millions of dimensions, the PS-SMART model can be used to train the first sub-classification model and the second sub-classification model. Figure 4a A schematic diagram of a model training process according to another embodiment of the present application is shown. Figure 4a As shown, by inputting user feature data into the first sub-classification model, the main node distributes tasks to the sea of ​​quantum nodes existing in different machines, and each sub-node is iteratively trained in sequence by sampling user feature data, pre-classifying the user's own value, and outputting the user level classification; taking the game as an example, the user feature data is associated with the game feature data of the game that the user has interacted with to obtain the first feature data matrix. By selecting the PS-SMART classification model as the first sub-classification model, users can be divided into low-value user level, medium-value user level, and high-value user level; for different user levels, the corresponding second sub-classification model is set.

[0076] Step S206: For each hierarchical classification, extract partial feature data corresponding to the users in the hierarchical classification from the first feature data matrix, use the partial feature data to train the second sub-classification model corresponding to the hierarchical classification, and obtain the user group model corresponding to the hierarchical classification.

[0077] In order to improve the prediction accuracy, the second sub-classification model is taken as the PS-SMART model as an example, Figure 4a As shown, for each hierarchical classification, partial feature data corresponding to the users in the hierarchical classification is extracted from the first feature data matrix, a second sub-classification model corresponding to the hierarchical classification is established, and the second sub-classification model is trained using partial feature data to obtain a user group model corresponding to the hierarchical classification. The PS-SMART classification model may include at least one classification tree. Figure 4a Schematically shows a case where a PS-SMART classification model including two classification trees is used as the first sub-classification model and the second sub-classification model.

[0078] For example, the PS-SMART classification model is selected as the first sub-classification model, and users are divided into low-value user level, medium-value user level and high-value user level. Part of the feature data corresponding to the users in the low-value user level is extracted from the first feature data matrix, and the data is input into the second sub-classification model corresponding to the low-value user level for training, and a low-value user group model is obtained after training. Part of the feature data corresponding to the users in the medium-value user level is extracted from the first feature data matrix, and the data is input into the second sub-classification model corresponding to the medium-value user level for training, and a medium-value user group model is obtained after training. Part of the feature data corresponding to the users in the high-value user level is extracted from the first feature data matrix, and the data is input into the second sub-classification model corresponding to the high-value user level for training, and a high-value user group model is obtained after training.

[0079] Step S207: According to the user level classification, determine the users included in each level classification.

[0080] Step S208: For each hierarchical classification, the second feature data matrix of the users included in the hierarchical classification is input into the user group model corresponding to the hierarchical classification for prediction, so as to obtain the user value prediction result of the users included in the hierarchical classification relative to the object to be predicted.

[0081] Step S209: Summarize the user value prediction results of the users included in each level classification relative to the object to be predicted.

[0082] Figure 4b A schematic diagram of a model prediction process according to another embodiment of the present application is shown. Figure 4bAs shown; through the above step 206, the user group model corresponding to each level classification is trained, and the user value of the user to the object to be predicted is predicted using multiple user group models. Specifically, taking the object as a game as an example, the user feature data is associated with the game feature data of the game to be predicted to obtain a second feature data matrix. According to the user level classification, the users included in each level classification are determined. For each level classification, the second feature data matrix of the users included in the level classification is input into the user group model corresponding to the level classification for prediction, and the user value prediction results of the users included in the level classification relative to the game to be predicted are obtained. The user value prediction results of the users included in each level classification relative to the game to be predicted are summarized, so as to obtain the user value of each user for the game to be predicted; for example, for the low-value user level, the part of the feature data corresponding to the users included in the level classification in the second feature data matrix is ​​input into the low-value user group model corresponding to the level classification for prediction, and the user value prediction results of the users included in the level classification relative to the game to be predicted are obtained; for the medium-value user level, For the user level, part of the feature data corresponding to the users included in the level classification in the second feature data matrix is ​​input into the medium-value user group model corresponding to the level classification for prediction, so as to obtain the user value prediction result of the users included in the level classification relative to the game to be predicted; for the high-value user level, part of the feature data corresponding to the users included in the level classification in the second feature data matrix is ​​input into the high-value user group model corresponding to the level classification for prediction, so as to obtain the user value prediction result of the users included in the level classification relative to the game to be predicted, and the user value prediction results of the users included in the three levels classification relative to the game to be predicted are summarized, and the user value of each user relative to the game to be predicted, that is, the relative user value, is determined according to the user value prediction results of different levels classification.

[0083] According to a user value classification method based on a two-layer sub-classification model provided in an embodiment of the present application, by associating user feature data and object feature data, the obtained first feature data matrix and second feature data matrix can well meet the cold start function requirements, so as to quickly perform model training and model prediction, and effectively improve the model training and prediction efficiency; and, a classification architecture including a two-layer sub-classification model is constructed, the user value level is classified by the first sub-classification model to obtain the user level classification, and the second sub-classification model corresponding to each level classification is trained using the first feature data matrix, so as to achieve precise parameter adjustment of the second sub-classification model of each level classification, so that the user group model corresponding to each level classification obtained through training has high prediction accuracy and can accurately predict the user value; this method can iteratively train massive user feature data at the same time by adopting a regression model, thereby improving the accuracy and speed of training the user group model and avoiding errors and delays in human decision-making.

[0084] Figure 5 FIG. 4 shows a structural block diagram of a user value classification device based on a double-layer sub-classification model according to an embodiment of the present application. Figure 5 As shown, the device includes: an association module 501, a user classification module 502, a model training module 503 and a prediction module 504:

[0085] The association module 501 is used to obtain user feature data and object feature data, and associate the user feature data with the object feature data to obtain a first feature data matrix and a second feature data matrix.

[0086] In an optional implementation, the association module 501 is further used to: obtain original object feature data of each object in multiple dimensions, and perform linear encoding on the original object feature data to obtain object feature data.

[0087] In an optional implementation, the association module 501 is further used to: obtain original user characteristic data of each user, and perform completion and denoising processing on the original user characteristic data to obtain user characteristic data.

[0088] In an optional implementation, user feature data includes one or more of the following data: user device type data, user time zone data, user basic data, and historical object interaction data; object feature data includes one or more of the following data: object category data, object label data, user level data, manufacturer data, and operator data.

[0089] In an optional embodiment, the association module 501 is further used to: for each user's user feature data, determine the objects with which the user has interacted based on the historical object interaction data in the user feature data, establish an association relationship between the user feature data of the user and the object feature data of the objects with which the user has interacted, and obtain a first feature data matrix; for each user's user feature data, establish an association relationship between the user feature data of the user and the object feature data of the object to be predicted, and obtain a second feature data matrix.

[0090] The user classification module 502 is used to input user feature data into the first sub-classification model, classify the user value level using the first sub-classification model to obtain a user level classification, and set a corresponding second sub-classification model for each level classification in the user level classification.

[0091] In an optional implementation, the first sub-classification model and the second sub-classification model are models obtained based on a regression model.

[0092] In an optional embodiment, the regression model includes a GBDT model, a PS-SMART model, or a linear regression model.

[0093] The model training module 503 is used to train the second sub-classification model corresponding to each hierarchical classification using the first feature data matrix to obtain a user group model corresponding to each hierarchical classification.

[0094] In an optional embodiment, the model training module 503 is further used to: for each hierarchical classification, extract partial feature data corresponding to the users in the hierarchical classification from the first feature data matrix, use the partial feature data to train the second sub-classification model corresponding to the hierarchical classification, and obtain the user group model corresponding to the hierarchical classification.

[0095] The prediction module 504 is used to predict the relative user value based on the user level classification, the second feature data matrix and the user group model corresponding to each level classification in the user level classification; wherein the relative user value is the user value of the user relative to the object to be predicted.

[0096] In an optional embodiment, the prediction module 504 is further used to: determine the users included in each hierarchical classification based on the user hierarchical classification; for each hierarchical classification, input the second feature data matrix of the users included in the hierarchical classification into the user group model corresponding to the hierarchical classification for prediction, and obtain the user value prediction results of the users included in the hierarchical classification relative to the object to be predicted; summarize the user value prediction results of the users included in each hierarchical classification relative to the object to be predicted.

[0097] The description of each module above refers to the corresponding description in the method embodiment and will not be repeated here.

[0098] According to a user value classification device based on a double-layer sub-classification model provided in an embodiment of the present application, the device associates user feature data with object feature data, and the obtained first feature data matrix and second feature data matrix can well meet the cold start function requirements, so as to quickly perform model training and model prediction, and effectively improve the model training and prediction efficiency; and, a classification architecture including a double-layer sub-classification model is constructed, the user value level is classified by the first sub-classification model to obtain the user level classification, and the second sub-classification model corresponding to each level classification is trained using the first feature data matrix, so as to achieve precise parameter adjustment of the second sub-classification model for each level classification, so that the user group model corresponding to each level classification obtained through training has high prediction accuracy and can accurately predict the user value; the device can iteratively train massive user feature data at the same time by adopting a regression model, so as to improve the accuracy and speed of training the user group model and avoid errors and delays in human decision-making.

[0099] An embodiment of the present application also provides a non-volatile computer storage medium, which stores at least one executable instruction, and the executable instruction can execute the user value classification method based on the two-layer sub-classification model in any of the above method embodiments.

[0100] Figure 6 A schematic diagram of the structure of a computing device according to an embodiment of the present application is shown. The specific implementation of the embodiment of the present application does not limit the specific implementation of the computing device.

[0101] like Figure 6 As shown, the computing device may include: a processor (processor) 602 , a communication interface (Communications Interface) 604 , a memory (memory) 606 , and a communication bus 608 .

[0102] in:

[0103] The processor 602 , the communication interface 604 , and the memory 606 communicate with each other via a communication bus 608 .

[0104] The communication interface 604 is used to communicate with other devices such as clients or other servers.

[0105] The processor 602 is used to execute the program 610, and specifically can execute the relevant steps in the above-mentioned embodiment of the user value classification method based on the two-layer sub-classification model.

[0106] Specifically, the program 610 may include program codes, which include computer operation instructions.

[0107] The processor 602 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application. The one or more processors included in the computing device may be processors of the same type, such as one or more CPUs; or may be processors of different types, such as one or more CPUs and one or more ASICs.

[0108] The memory 606 is used to store the program 610. The memory 606 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0109] Program 610 can be specifically used to enable processor 602 to execute the user value classification method based on the two-layer sub-classification model in any of the above-mentioned method embodiments. The specific implementation of each step in program 610 can refer to the corresponding descriptions in the corresponding steps and units in the above-mentioned user value classification method embodiment based on the two-layer sub-classification model, which will not be repeated here. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described devices and modules can refer to the corresponding process description in the above-mentioned method embodiment, which will not be repeated here.

[0110] The algorithm and display provided herein are not inherently related to any specific computer, virtual system or other equipment. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious to construct the structure required for this type of system. In addition, the present application embodiment is not directed to any specific programming language yet. It should be understood that various programming languages ​​can be utilized to realize the content of the present application embodiment described herein, and the above description of specific languages ​​is to disclose the best mode of implementation of the present application embodiment.

[0111] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0112] Similarly, it should be understood that in order to streamline the present disclosure and help understand one or more of the various inventive aspects, in the above description of the exemplary embodiments of the present application, the various features of the embodiments of the present application are sometimes grouped together into a single embodiment, figure, or description thereof. However, the disclosed method should not be interpreted as reflecting the following intention: the claimed embodiments of the present application require more features than the features clearly stated in each claim. More specifically, as reflected in the claims below, the inventive aspects are less than all the features of the single embodiment disclosed above. Therefore, the claims following the specific embodiment are hereby explicitly incorporated into the specific embodiment, wherein each claim itself serves as a separate embodiment of the embodiment of the present application.

[0113] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and in addition they may be divided into a plurality of submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0114] In addition, those skilled in the art will appreciate that, although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the embodiments of the present application and form different embodiments. For example, in the claims below, any one of the claimed embodiments may be used in any combination.

[0115] The various component embodiments of the embodiments of the present application can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all functions of some or all components according to the embodiments of the present application. The embodiments of the present application can also be implemented as a device or apparatus program (e.g., computer program and computer program product) for executing part or all of the methods described herein. Such a program implementing the embodiments of the present application can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0116] It should be noted that the above embodiments illustrate rather than limit the embodiments of the present application, and that those skilled in the art may design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets should not be constructed as a limitation to the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of multiple such elements. The embodiments of the present application may be implemented by means of hardware including several different elements and by means of appropriately programmed computers. In a unit claim that lists several devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order. These words may be interpreted as names.

Claims

1. A user value classification method based on a two-layer sub-classification model, comprising: Acquire user feature data and object feature data, and associate the user feature data with the object feature data to obtain a first feature data matrix and a second feature data matrix; wherein the first feature data matrix refers to a data matrix obtained by associating the user feature data of the user with the object feature data of the object with which the user has interacted, and the second feature data matrix refers to a data matrix obtained by associating the user feature data of the user with the object feature data of the object to be predicted; the object feature data includes one or more of the following data: object category data, object label data, user magnitude data, manufacturer data, and operator data; Inputting the user feature data into a first sub-classification model, classifying the user value level using the first sub-classification model to obtain a user level classification, and setting a corresponding second sub-classification model for each level classification in the user level classification; Using the first feature data matrix, training the second sub-classification model corresponding to each hierarchical classification to obtain a user group model corresponding to each hierarchical classification; Predicting the relative user value according to the user level classification, the second feature data matrix, and the user group model corresponding to each level classification in the user level classification; wherein the relative user value is the user value of the user relative to the object to be predicted; Wherein, associating the user characteristic data with the object characteristic data to obtain a first characteristic data matrix and a second characteristic data matrix further comprises: For each user's user feature data, determine the objects with which the user has interacted according to the historical object interaction data in the user feature data, establish an association relationship between the user feature data of the user and the object feature data of the objects with which the user has interacted, and obtain a first feature data matrix; For the user feature data of each user, an association relationship is established between the user feature data of the user and the object feature data of the object to be predicted, so as to obtain a second feature data matrix.

2. The method according to claim 1, wherein obtaining object feature data further comprises: Original object feature data of each object in multiple dimensions is obtained, and linear encoding is performed on the original object feature data to obtain the object feature data.

3. According to the method of claim 1, obtaining user characteristic data further comprises: The original user characteristic data of each user is obtained, and the original user characteristic data is supplemented and denoised to obtain the user characteristic data.

4. According to the method of claim 1, the step of using the first feature data matrix to train the second sub-classification model corresponding to each hierarchical classification to obtain the user group model corresponding to each hierarchical classification further comprises: For each hierarchical classification, partial feature data corresponding to users in the hierarchical classification is extracted from the first feature data matrix, and the second sub-classification model corresponding to the hierarchical classification is trained using the partial feature data to obtain a user group model corresponding to the hierarchical classification.

5. According to the method of any one of claims 1 to 4, the predicting of relative user value based on the user level classification, the second feature data matrix and the user group model corresponding to each level classification in the user level classification further comprises: According to the user level classification, determining the users included in each level classification; For each hierarchical classification, the second feature data matrix of the users included in the hierarchical classification is input into the user group model corresponding to the hierarchical classification for prediction, so as to obtain the user value prediction result of the users included in the hierarchical classification relative to the object to be predicted; Summarize the user value prediction results of users included in each level classification relative to the object to be predicted.

6. According to the method of claim 1, the first sub-classification model and the second sub-classification model are models obtained based on a regression model.

7. A user value classification device based on a double-layer sub-classification model, comprising: An association module is used to obtain user feature data and object feature data, and associate the user feature data with the object feature data to obtain a first feature data matrix and a second feature data matrix; wherein the first feature data matrix refers to a data matrix obtained by associating the user feature data of the user with the object feature data of the object with which the user has interacted, and the second feature data matrix refers to a data matrix obtained by associating the user feature data of the user with the object feature data of the object to be predicted; the object feature data includes one or more of the following data: object category data, object label data, user magnitude data, manufacturer data, and operator data; A user classification module, configured to input the user feature data into a first sub-classification model, classify the user value level using the first sub-classification model to obtain a user level classification, and set a corresponding second sub-classification model for each level classification in the user level classification; A model training module, used to train the second sub-classification model corresponding to each hierarchical classification using the first feature data matrix to obtain a user group model corresponding to each hierarchical classification; A prediction module, configured to predict relative user value according to the user level classification, the second feature data matrix, and a user group model corresponding to each level classification in the user level classification; wherein the relative user value is a user value of a user relative to an object to be predicted; Wherein, the association module is further used for: For each user's user feature data, determine the objects with which the user has interacted according to the historical object interaction data in the user feature data, establish an association relationship between the user feature data of the user and the object feature data of the objects with which the user has interacted, and obtain a first feature data matrix; For the user feature data of each user, an association relationship is established between the user feature data of the user and the object feature data of the object to be predicted, so as to obtain a second feature data matrix.

8. A computing device comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to a user value classification method based on a two-layer sub-classification model as described in any one of claims 1-6.

9. A computer storage medium, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to perform operations corresponding to a user value classification method based on a two-layer sub-classification model as described in any one of claims 1-6.

10. A computer program product, comprising at least one executable instruction, wherein the executable instruction enables a processor to perform operations corresponding to a user value classification method based on a two-layer sub-classification model as described in any one of claims 1 to 6.

Images