Object saving method, related device, equipment and storage medium

Abstract

The invention discloses an object saving method, a related device, equipment and a storage medium. According to the method provided by the invention, an object data cluster comprising object data of multiple objects can be input into an object loss prediction model based on a random forest algorithm for classification, a corresponding classification label is output for each object, then a first label indicating to be lost in the classification labels is determined, and a collaborative filtering algorithm is adopted, so that the object loss prediction is realized. And determining a recommended commodity according to the first purchase record of the first object corresponding to the first label, and pushing retention information about the recommended commodity to the first object. Through the above mode, the prediction accuracy is improved through the random forest algorithm, and then personalized recommendation is performed according to the commodities purchased and recorded by the user, so that the attraction to the user and the retention information of the recommended commodities are improved, the purchase desire of the user is improved, the users who purchase the recommended commodities cannot be further lost, and the user retention is improved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to an object retention method, related devices, equipment and storage medium. Background Art

[0002] With the development of Internet technology, application platforms that realize various functions have emerged. For application platforms, in various stages of their life cycle, there will often be varying degrees of user loss, which will have an adverse impact on the corresponding application platform. Take online games as an example. Online games generally refer to electronic games that are played by multiple people through computer networks. The life cycle of online games includes growth period, stable period and decline period, etc. In each period, there will often be varying degrees of game user loss. The loss of game users will not only reduce the number of consumers, but also affect the internal environment of the entire game platform, such as causing some servers to become "ghost servers". Therefore, it is necessary to predict the loss of users on the application platform.

[0003] Currently, after predicting the users who are about to churn, incentive measures are used to retain them. However, the incentive effect of such incentive measures on the users who are about to churn is unstable and easily affects the retention effect. Summary of the invention

[0004] The embodiments of the present application provide an object retention method, related devices, equipment and storage medium, which are used to first improve the prediction accuracy through a random forest algorithm, and then make personalized recommendations based on the products in the user's purchase record to increase the attractiveness to the user, as well as retention information for the recommended products, thereby increasing the user's desire to buy. Users who purchase the recommended products will not be further lost, thereby improving user retention.

[0005] In view of this, the present application provides, on one hand, a method for retaining an object, comprising:

[0006] Acquire an object data cluster, the object data cluster including object data of a plurality of objects;

[0007] Classifying the object data clusters by an object loss prediction model based on a random forest algorithm to output classification labels, the classification labels corresponding to the plurality of objects one by one, the classification labels including a first label, the first label indicating a subject to be lost;

[0008] determining a first object among the plurality of objects according to the first tag;

[0009] Using a collaborative filtering algorithm, determine a recommended product based on the first purchase record of the first object;

[0010] A first retention information is pushed to the first object according to the recommended product.

[0011] Another aspect of the present application provides a user retention device, comprising:

[0012] An acquisition unit, configured to acquire an object data cluster, wherein the object data cluster includes object data of a plurality of objects;

[0013] A classification unit, configured to classify the object data clusters by using an object loss prediction model based on a random forest algorithm to output classification labels, wherein the classification labels correspond to the plurality of objects one by one, and the classification labels include a first label, and the first label indicates a target to be lost;

[0014] a determining unit, configured to determine a first object among the multiple objects according to the first tag, and to determine a recommended product according to a first purchase record of the first object by using a collaborative filtering algorithm;

[0015] The push unit is used to push the first retention information to the first object according to the recommended product.

[0016] In a possible design, in another implementation of another aspect of the embodiment of the present application, the determining unit is specifically configured to:

[0017] determining a plurality of object similarities according to the first purchase record and a plurality of second purchase records, each second purchase record corresponding to a second object;

[0018] The recommended products are determined according to the products in the third purchase record, where the third purchase record is a purchase record whose object similarity in the plurality of second purchase records reaches a first preset threshold.

[0019] In a possible design, in another implementation of another aspect of the embodiment of the present application, the determining unit is specifically configured to:

[0020] determining a plurality of commodity similarities based on a first commodity and a plurality of second commodities in a first purchase record, wherein the first purchase record does not include the plurality of second commodities;

[0021] The recommended product is determined according to the third product, where the third product is a product among the plurality of second products whose product similarity reaches a second preset threshold.

[0022] In a possible design, in another implementation of another aspect of the embodiment of the present application, the pushing unit is specifically used for:

[0023] A price reduction of the recommended product is notified to the first subject.

[0024] In a possible design, in another implementation of another aspect of the embodiment of the present application, the acquisition unit is further used to:

[0025] Get object sample data;

[0026] The device also includes an extraction unit, which is specifically used for:

[0027] Extract multi-dimensional features from object sample data;

[0028] The acquisition unit is also used to:

[0029] Get sample labels corresponding to multi-dimensional features;

[0030] The device also includes a training unit, which is specifically used for:

[0031] Train the object churn prediction model based on multi-dimensional features and sample labels.

[0032] In a possible design, in another implementation of another aspect of the embodiment of the present application, the determining unit is further configured to:

[0033] The sample label is determined according to the data cutoff time in the multi-dimensional features, and the sample label indicates the retained object or the lost object.

[0034] In a possible design, in another implementation of another aspect of the embodiment of the present application, the acquisition unit is further used to:

[0035] Get a sample label, where the sample label indicates a retained object or a lost object with a loss reason.

[0036] In a possible design, in another implementation of another aspect of the embodiment of the present application, the training unit is specifically used for:

[0037] Divide the multi-dimensional features and sample labels into multiple data sets, including a training set;

[0038] Performing a grid search on the training set to determine a first hyperparameter combination, where the first hyperparameter combination is a hyperparameter combination that meets preset requirements;

[0039] The training set is trained according to the first hyperparameter combination to obtain an object churn prediction model.

[0040] In a possible design, in another implementation of another aspect of the embodiment of the present application, the multiple data sets further include a test set, and the apparatus further includes a verification unit, which is specifically used to:

[0041] The object churn prediction model is verified through the test set.

[0042] In a possible design, in another implementation of another aspect of the embodiment of the present application, the determining unit is further configured to:

[0043] One of the multiple data sets is determined as a test set in turn, and the data sets except the test set in the multiple subsets are used as training sets.

[0044] In a possible design, in another implementation of another aspect of the embodiment of the present application, the extraction unit is specifically used for:

[0045] Perform data preprocessing on the object sample data to obtain sample features;

[0046] Select sample features according to the L1 regularization selection method to obtain multi-dimensional features; or,

[0047] Select sample features according to the information gain selection method to obtain multi-dimensional features; or,

[0048] The sample features are selected according to the Gini coefficient selection method to obtain multi-dimensional features.

[0049] In a possible design, in another implementation of another aspect of the embodiment of the present application, the device further includes a dimension reduction unit, and the dimension reduction unit is specifically used to:

[0050] Perform dimensionality reduction on multi-dimensional features.

[0051] In a possible design, in another implementation of another aspect of the embodiment of the present application, the extraction unit is further used to:

[0052] extracting a feature vector from the historical behavior data of the first object;

[0053] The push unit is also used for:

[0054] The second retention information corresponding to the feature vector is pushed to the first object.

[0055] In a possible design, in another implementation of another aspect of the embodiment of the present application, the pushing unit is further used for:

[0056] The second retention information corresponding to the loss reason of the first object is pushed to the first object.

[0057] Another aspect of the present application provides a computer device, comprising:

[0058] memories, transceivers, processors, and bus systems;

[0059] Wherein, the memory is used to store programs;

[0060] The processor is used to execute the program in the memory, including executing the above-mentioned methods;

[0061] The bus system is used to connect the memory and the processor so that the memory and the processor can communicate with each other.

[0062] Another aspect of the present application provides a computer-readable storage medium, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer is enabled to execute the above-mentioned methods.

[0063] Another aspect of the present application provides a computer program product or a computer program, the computer program product or the computer program includes computer instructions, the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided by the above aspects.

[0064] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0065] In an embodiment of the present application, an object retention method is provided. The method can input an object data cluster including object data of multiple objects into an object loss prediction model based on a random forest algorithm for classification, output a corresponding classification label for each object, and then determine the first label in the classification label indicating the object to be lost, and use a collaborative filtering algorithm to determine the recommended product based on the first purchase record of the first object corresponding to the first label, and then push retention information about the recommended product to the first object. In the above manner, the prediction accuracy is first improved by the random forest algorithm, and then personalized recommendations are made based on the products in the user's purchase record to improve the attractiveness to users, and the retention information for the recommended product can improve the user's desire to buy, and users who purchase the recommended product will not be further lost, thereby improving user retention. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a schematic diagram of the architecture of an object retention system in an embodiment of the present application;

[0067] Figure 2 A schematic diagram of a process flow of a method for retaining an object in an embodiment of the present application;

[0068] Figure 3 A schematic diagram of a classification of a random forest in an embodiment of the present application;

[0069] Figure 4 This is a schematic diagram of the structure of an object retention device in an embodiment of the present application;

[0070] Figure 5 It is a structural schematic diagram of a computer device in an embodiment of the present application. DETAILED DESCRIPTION

[0071] The embodiments of the present application provide an object retention method, related devices, equipment and storage medium, which are used to first improve the prediction accuracy through a random forest algorithm, and then make personalized recommendations based on the products in the user's purchase record to increase the attractiveness to the user, as well as retention information for the recommended products, thereby increasing the user's desire to buy. Users who purchase the recommended products will not be further lost, thereby improving user retention.

[0072] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "corresponding to" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0073] User data is an important type of data in business operations. In combination with the specific business needs of the enterprise, targeted processing and analysis of user data is of great guiding significance, and it can play a huge role in retaining users who are about to lose products. Currently, after predicting the users who are about to lose, incentives are used to retain the users who are about to lose. However, the incentive effect of this incentive on lost users is unstable and easily affects the retention effect. In the embodiments of the present application, user data can also be called object data. In view of this, the present application provides an object retention method that can effectively solve the above problems.

[0074] The object retention method provided in the embodiment of the present application can predict the probability of loss of specific users in the future cycle of the business and retain them for businesses such as games, social networking, and e-commerce. As an optional example, the user retention probability prediction method provided in the embodiment of the present application can predict the probability of retention of specific users in the future cycle of the game business for the game business. Figure 1 The schematic diagram of the structure of the object retention system provided by the embodiment of the present invention is shown. Figure 1 The system may include: a game database 10 , a retention server 11 and a user device 12 .

[0075] The game database 10 is a database for storing game data of users of the game service. The game data may store various types of data related to users in the game service. Specifically, as an example, the game database may store the following data for each user:

[0076] User basic data: gender, age, education level, etc.

[0077] User login data: login duration, number of logins, number of login days, login frequency, etc.;

[0078] User recharge data: recharge amount, purchase history, account balance, number of recharges, average revenue per user (ARPU), etc.

[0079] User performance data: character level, equipment information, team formation frequency, number of monsters killed, etc.

[0080] In the embodiment of the present invention, the game database 10 can also be used to record the user churn data and active user data of the game in each historical period. A period can refer to one day, one week, one month, one quarter, etc., or other preset time ranges, such as five days, ten days, two months, etc.

[0081] The retention server 11 is a service device used in the embodiment of the present application to predict the probability of user churn in the future cycle of the game service and perform corresponding retention operations. Its functions can be implemented by the game server or by a separately set server. Among them, the game server is a service device that provides game services to users, and can exchange with the user's game terminal to realize functions such as game login, exit, and game logic processing.

[0082] The user device 12 is a mobile phone, a portable music player, a personal digital assistant, a dedicated messaging device, a portable gaming device, or other user terminals of various types, and may be installed with various applications, such as instant messaging applications, third-party payment applications, video viewing applications, shopping applications, etc. The user device 12 may use the instant messaging application to communicate with friends, relatives, or colleagues in real life or work, may use the third-party payment application to make electronic payments, may use the shopping application to make online purchases, etc. Figure 1 The user device 12 is taken as an example of a gaming device or a device capable of loading a game.

[0083] In the embodiment of the present application, the retention server 11 can cooperate with the game database 10 and the user device 12 to predict the churn probability of any specific user of the game service in the future period and perform a retention operation. Among them, the game database 10 can provide data support for the prediction of the user's churn probability in the future period, and the retention server 11 can predict the user's churn probability in the future period of the game service based on the user data in the game database 10, and then perform a corresponding retention operation on the user device 12 of the user predicted to churn.

[0084] It should be noted that Figure 1 The game database, retention server and user merger shown are only an example of a system architecture for a game-type service; in any type of service, the embodiment of the present application can provide a system architecture of a service database, a retention server and a user device to achieve user churn probability prediction and retention. The service database can be a database storing user service data ( Figure 1 The game database shown is only a form of business database in the context of game business); the retention server can be a business server that provides business services (such as Figure 1 The retention server shown in the figure can be implemented by a game server), or it can be a service device specially set up for predicting the probability of user churn and retaining users. The user device can be a device that generates business data for users (such as Figure 1 The user device shown may be implemented as a gaming device).

[0085] Obviously, Figure 1 The game database shown may also be a data storage unit arranged in the game server, that is, the service database may be a data storage unit in the service server.

[0086] Correspondingly, the retention server can call the object data clusters in the game database, and classify the object data clusters through the object churn prediction model based on the random forest algorithm to output classification labels, the classification labels correspond to multiple objects one by one, the classification labels include a first label, the first label indicates the objects to be churned; determine the first object among the multiple objects according to the first label; use a collaborative filtering algorithm to determine the recommended product according to the first purchase record of the first object; push retention information to the user device of the first object according to the recommended product to increase the attractiveness and purchasing desire to the user and improve user retention.

[0087] To implement the above solution, an embodiment of the present application provides an object retention method, which is described below from the perspective of a retention server.

[0088] See also Figure 2 ,like Figure 2 The figure is a flow chart of a method for retaining an object provided in an embodiment of the present application, the method comprising:

[0089] Step 201: Obtain an object data cluster, where the object data cluster includes object data of multiple objects.

[0090] In one or more embodiments, the object data cluster may be specifically a cluster of usage data of various objects on various software application products, such as games, video players, or text browsers. The embodiment of the present application may be based on a software application product, such as a game, to obtain game data (object data) of a game usage object, and the game data of all objects may be collectively referred to as an object data cluster. Among them, since the user establishes an account, all kinds of information and activities under the product account will form object data and be recorded, including personal information, friend information, product usage records, consumption data, customer service records, etc.

[0091] Step 202: Classify the object data clusters using an object loss prediction model based on a random forest algorithm to output classification labels, where the classification labels correspond one-to-one to a plurality of objects, and the classification labels include a first label, where the first label indicates a subject to be lost.

[0092] In one or more embodiments, the object data includes multiple features, and these features can be used to evaluate the state of the object. For example, in the embodiment of the present application, an object loss prediction model based on a random forest algorithm can be used. The random forest algorithm model includes multiple decision trees, each decision tree includes multiple nodes, and each node identifies a classification on an object data feature. After the object data cluster is input into the decision tree model, the classification label to which each object data belongs is determined based on the classification results output by all decision trees. Among them, each object data has its corresponding classification label, and the classification label can indicate that the object corresponding to the object data will continue to remain, or the object corresponding to the object data will be lost. Among them, the embodiment of the present application can refer to the classification label indicating that the object will be lost as the first label.

[0093] The following is an explanation of the decision tree in the random forest. A decision tree contains a root node, several internal nodes, and several leaf nodes; the leaf node corresponds to the decision result (i.e., classification), and each other node corresponds to an attribute test; the sample set contained in each node is divided into child nodes according to the result of the attribute test, that is, an attribute is selected for splitting on each internal node, and each fork corresponds to an attribute value; the root node contains the entire set of samples, and the path from the root node to each leaf node corresponds to a judgment test sequence. For details, please refer to Figure 3 As shown, taking the decision tree as a binary tree as an example, assuming that the classification result has only two categories, and the feature judgment of each node is 1, the classification of the object data is classification 1, and if there is a feature judgment of 2, it means that the classification of the object data is classification 2, then Figure 3 Only the paths indicated by bold lines point to category 1.

[0094] The object loss model in the embodiment of the present application may also be an XGBoost model or a neural network model.

[0095] Specifically, for the XGBoost model, the object data cluster can be input into the XGBoost model. The XGBoost model can perform an initial screening of the object data cluster, eliminate irrelevant or redundant features, retain the top several features in each object data, and finally output the corresponding classification label.

[0096] Specifically, for the neural network model, the object data cluster is input into the input layer included in the neural network model, and each object data outputs a first feature vector, and then the first feature vector is input into at least one hidden layer to obtain a second feature vector. Finally, the second feature vector is input into the output layer, wherein the output layer can use a softmax operation to obtain a probability distribution, and the classification label to which the object data belongs can be determined based on the probability distribution.

[0097] The embodiments of the present application may also integrate multiple or all models of a random forest model, an XGBoost model, or a neural network model, and perform weighted voting on the results determined by each model to obtain a final result, which is not limited here.

[0098] Step 203: Determine a first object among multiple objects according to the first tag.

[0099] In one or more embodiments, after determining the first label, object data whose prediction result is the first label can be correspondingly determined, wherein the object data can carry an object identifier, and the object identifier indicates the first object that generates the object data using a software application product, and the first object can be one or more users, that is, multiple first labels correspond to multiple first objects.

[0100] Step 204: Using a collaborative filtering algorithm, determine a recommended product based on the first purchase record of the first object.

[0101] First of all, the collaborative filtering algorithm is a process of analyzing the interests of each user, determining the similarity between the user or product and the interest, and then screening the products whose similarity with the interest reaches a threshold. The type of the product is related to the software application. When the software application is a game, the product can be a game prop; when the software application is a video player, the product can be a video; when the software application is a text browser, the product can be an article or document.

[0102] In one or more embodiments, after determining the first object, consumption data can be determined from the object data of the first object. The consumption data of the first object can be called a first purchase record. The embodiment of the present application can analyze the products with high interest similarity to the first object according to the collaborative filtering algorithm based on the products in the first purchase record as recommended products. The recommended products are products that the first object is interested in, which can increase the attractiveness to the first object.

[0103] Step 205: Pushing first retention information to the first object based on the recommended product.

[0104] In one or more embodiments, after determining the recommended product, even if the first object is interested in the recommended product, it is possible that he or she does not purchase it due to cost-effectiveness reasons. The embodiment of the present application can push a first retention message to the first object, and the first retention message can indicate that preferential measures are provided for the recommended product to increase the first object's desire to purchase the recommended product.

[0105] In the embodiment of the present application, an object data cluster including object data of multiple objects is input into an object loss prediction model based on a random forest algorithm for classification, and a corresponding classification label is output for each object. Then, the first label indicating the loss in the classification label is determined, and a collaborative filtering algorithm is used to determine the recommended product according to the first purchase record of the first object corresponding to the first label, and then the retention information about the recommended product is pushed to the first object. In the above manner, the prediction accuracy is first improved by the random forest algorithm, and then personalized recommendations are made based on the products in the user's purchase record to improve the attractiveness to the user, and the retention information for the recommended product is used to improve the user's desire to buy, and the users who purchase the recommended product will not be further lost, thereby improving user retention.

[0106] Optionally, in the above Figure 2 On the basis of the corresponding embodiments, in another optional embodiment provided by the embodiment of the present application, determining the recommended product according to the first purchase record of the first object includes:

[0107] determining a plurality of object similarities according to the first purchase record and a plurality of second purchase records, each second purchase record corresponding to a second object;

[0108] The recommended products are determined according to the products in the third purchase record, where the third purchase record is a purchase record whose object similarity in the plurality of second purchase records reaches a first preset threshold.

[0109] In one or more embodiments, the method of determining the recommended product based on the collaborative filtering algorithm can be determined by user similarity, that is, obtaining the first purchase record of the first object and multiple second purchase records of multiple second objects saved in advance, the second object is an object different from any first object, and the embodiment of the present application can determine the object similarity between the first object and any other object by determining the similarity between the purchase records of the first object and other objects. The higher the object similarity, the more similar the purchase interest between the first object and the second object. When the object similarity between the first purchase record and the second purchase record reaches a first preset threshold, such as 80%, 85%, 90% or 95%, the second purchase record can be called a third purchase record. At this time, it can be considered that the purchase interest of the second object corresponding to the third purchase record is similar to the purchase interest of the first object. It can be considered that the product not included in the first purchase record in the third purchase record may also be the demand of the first object, and the product not included in the first purchase record in the third purchase record can be used as a recommended product. The embodiment of the present application can also only select the purchase record with the highest object similarity with the first purchase record among multiple second purchase records as the third purchase record, which is not limited here.

[0110] In the embodiment of the present application, cosine similarity may be used to calculate object similarity, and the similarity between objects i and j may be calculated using the following formula:

[0111]

[0112] Among them, r ik represents the rating of object i for product k. If object i does not rate the product, then r ik =0. jk represents the rating of object j for product k. If object j does not rate the product, then r jk =0.

[0113] In the embodiment of the present application, a method for recommending goods based on an object is provided. Through the above implementation, goods purchased by objects with high object similarity can be recommended to the first object, thereby increasing the attractiveness to the object.

[0114] Optionally, in the above Figure 2 On the basis of the corresponding embodiments, in another optional embodiment provided by the embodiment of the present application, determining the recommended product according to the first purchase record of the first object includes:

[0115] determining a plurality of commodity similarities based on a first commodity and a plurality of second commodities in a first purchase record, wherein the first purchase record does not include the plurality of second commodities;

[0116] The recommended product is determined according to the third product, where the third product is a product among the plurality of second products whose product similarity reaches a second preset threshold.

[0117] In one or more embodiments, the method for determining recommended products based on the collaborative filtering algorithm can be determined by product similarity, that is, obtaining the first product in the first purchase record of the first object, and then obtaining any second product that is different from the first product in the first purchase record. The embodiment of the present application determines the first object's interest in the second product by determining the product similarity between the first product and the second product. The higher the product similarity, the higher the interest of the first object in the second product. When the product similarity between the first product and the second product reaches a second preset threshold, such as 80%, 85%, 90% or 95%, the second product can be called the third product. At this time, it can be considered that the third product is most likely to be of interest to the first object, and the third product can be used as a recommended product. The embodiment of the present application can also only select the product with the highest product similarity with the first product among multiple second products as the third product, which is not limited here.

[0118] In the embodiment of the present application, a commodity recommendation method based on commodities is provided. Through the above implementation, commodities with high commodity similarity to the first commodity in the first purchase record can be recommended to the first object, thereby increasing the attractiveness to the object.

[0119] Optionally, in the above Figure 2 On the basis of the corresponding embodiments, in another optional embodiment provided by the embodiment of the present application, the retention information indicates that the price of the recommended product is reduced, and pushing the retention information to the first object according to the recommendation information includes:

[0120] A price reduction of the recommended product is notified to the first subject.

[0121] In one or more embodiments, after determining the recommended product, in order to further attract the first subject to purchase the product, the price of the recommended product can be reduced, and the first subject is notified that the price of the recommended product can be reduced. The method of notifying the first subject of the price reduction of the recommended product can be to provide the first subject with a coupon, discount code, etc. for the recommended product.

[0122] In the embodiment of the present application, a method for retaining an object based on recommended goods is provided. Through the above implementation, the first object can be notified of the preferential measures for the recommended goods, thereby increasing the first object's desire to buy.

[0123] Optionally, in the above Figure 2 On the basis of the corresponding embodiments, in another optional embodiment provided by the embodiment of the present application, before classifying the object data clusters by the object loss prediction model based on the random forest algorithm to output the classification labels, the method further includes:

[0124] Get object sample data;

[0125] Extract multi-dimensional features from object sample data;

[0126] Get sample labels corresponding to multi-dimensional features;

[0127] Train the object churn prediction model based on multi-dimensional features and sample labels.

[0128] In one or more embodiments, before using the object churn prediction model to classify the object data, it is necessary to first train the object churn prediction model, and the object sample data may be obtained first. The user usage status corresponding to the object sample data has been determined, that is, the object sample data has been labeled, and the label indicates the user retention status of the object sample data. The label of each object sample data may be called a sample label. Based on the existence of redundant data in the object sample data, in order to reduce the model training time, feature extraction may be performed on the object sample data to extract multidimensional features with high correlation with the user retention status. All features in the multidimensional features are combined to correspond to the labeled sample labels. The embodiment of the present application may train the multidimensional features and the corresponding sample labels based on the random forest algorithm, and then the object churn prediction model may be obtained.

[0129] In an embodiment of the present application, a training method for an object loss prediction model is provided. Through the above implementation, the object loss prediction model is trained based on multi-dimensional features and sample labels, which can save training time and computing power.

[0130] Optionally, in the above Figure 2 On the basis of the corresponding embodiments, in another optional embodiment provided by the embodiment of the present application, obtaining sample labels corresponding to the multi-dimensional features includes:

[0131] The sample label is determined according to the data cutoff time in the multi-dimensional features, and the sample label indicates the retained object or the lost object.

[0132] In one or more embodiments, the data cutoff time is the cutoff time for the user to use the software application product. For example, if the user logs in 0 times within a continuous time, the continuous time can be called the cutoff time. The embodiment of the present application can set a preset time. When the cutoff time exceeds the preset time, the sample label corresponding to the multi-dimensional feature is a loss object. When the cutoff time does not exceed the preset time, the sample label corresponding to the multi-dimensional feature can be determined as a retention object.

[0133] In the embodiment of the present application, a method for determining sample labels is provided. Through the above implementation, sample labels are determined based on the data cutoff time in the multi-dimensional features. Users who have not logged in for a long time are obviously lost users, which can improve the accuracy of sample labels.

[0134] Optionally, in the above Figure 2 On the basis of the corresponding embodiments, in another optional embodiment provided by the embodiment of the present application, obtaining sample labels corresponding to the multi-dimensional features includes:

[0135] Get a sample label, where the sample label indicates a retained object or a lost object with a loss reason.

[0136] In one or more embodiments, in addition to classifying the user retention status into retained objects and lost objects, the sample labels may also classify lost objects into multiple categories according to the loss reasons, wherein the loss reasons may be obtained through object data analysis, or may be determined through questionnaires or background checks. For example, assuming that 0 represents a retained object, 1 may represent a lost object with reason 1, 2 may represent a lost object with reason 2, 3 may represent a lost object with reason 3, and so on.

[0137] In the embodiment of the present application, a method for determining a sample label is provided. Through the above implementation, the churn objects are divided into multiple categories according to the churn reasons, and retention measures can be accurately configured based on the churn reasons to improve the success rate of retention.

[0138] Optionally, in the above Figure 2 On the basis of the corresponding embodiments, in another optional embodiment provided by the embodiment of the present application, training the object loss prediction model according to the multi-dimensional features and sample labels includes:

[0139] Divide the multi-dimensional features and sample labels into multiple data sets, including a training set;

[0140] Performing a grid search on the training set to determine a first hyperparameter combination, where the first hyperparameter combination is a hyperparameter combination that meets preset requirements;

[0141] The training set is trained according to the first hyperparameter combination to obtain an object churn prediction model.

[0142] In one or more embodiments, the data set of multi-dimensional features and corresponding sample labels is divided into multiple data sets, and some or all of the multiple data sets can be selected as training sets for training. A grid search is performed on the training set to find a hyperparameter combination that meets the preset requirements as the first hyperparameter combination. Among them, the hyperparameters may include the number of decision trees, the maximum depth, the minimum number of leaf nodes, etc., and the first hyperparameter combination can be used to train the entire data set to obtain the object loss prediction model.

[0143] In an embodiment of the present application, a training method for an object loss prediction model is provided. By selecting a first hyperparameter combination that meets preset requirements for a training set as the initial training parameter of the object loss prediction model, the learning performance and effect of the model can be improved.

[0144] Optionally, in the above Figure 2 On the basis of the corresponding embodiments, in another optional embodiment provided by the embodiment of the present application, the multiple data sets also include a test set, and the method further includes:

[0145] The object churn prediction model is verified through the test set.

[0146] In one or more embodiments, when selecting some data sets from multiple data sets as training sets, data sets other than the training sets may also be selected as test sets, and the test sets may be used to test the object churn prediction model to verify the accuracy, recall rate, F1 value and other indicators of the object churn prediction model. The ratio of the training set to the test set in the multiple data sets may be 7:3, 8:2 or 9:1, etc., which is not limited here.

[0147] In the embodiment of the present application, a method for testing a subject churn prediction model is provided. Through the above implementation, the subject churn prediction model is verified by dividing the test set from the sample data. Since the test set and the training set are different, the verification accuracy can be improved.

[0148] Optionally, in the above Figure 2 On the basis of the corresponding embodiments, in another optional embodiment provided by the embodiment of the present application, the method further includes:

[0149] One of the multiple data sets is determined as a test set in turn, and the data sets except the test set in the multiple subsets are used as training sets.

[0150] In one or more embodiments, a cross-validation algorithm may be used to evaluate the object churn prediction model to ensure the robustness of the model. For multiple data sets, each data set may be used as a test set in turn, and the remaining data sets may be used as training sets for model training, and then the test set may be used to verify the model. Assuming that the multiple data sets are k data sets, the average value of the k verification results may be calculated as the final model accuracy, recall rate, F1 value and other indicators.

[0151] In the embodiment of the present application, a test method of an object loss prediction model is provided. Through the above implementation, the accuracy of the verification can be improved through a cross-validation algorithm.

[0152] Optionally, in the above Figure 2 On the basis of the corresponding embodiments, in another optional embodiment provided by the embodiment of the present application, extracting multi-dimensional features from the object sample data includes:

[0153] Perform data preprocessing on the object sample data to obtain sample features;

[0154] Select sample features according to the L1 regularization selection method to obtain multi-dimensional features; or,

[0155] Select sample features according to the information gain selection method to obtain multi-dimensional features; or,

[0156] The sample features are selected according to the Gini coefficient selection method to obtain multi-dimensional features.

[0157] In one or more embodiments, during the data preprocessing process, the raw data needs to be cleaned, feature encoded, and feature normalized so as to be converted into a format that can be used by a machine learning algorithm to obtain sample features. The embodiment of the present application can also perform feature selection on sample features, that is, select the most critical features for object loss prediction from all possible features. Among them, the embodiment of the present application can use the L1 regularization selection method, the information gain selection method, or the Gini index selection method for feature selection to screen multi-dimensional features from sample features. The embodiment of the present application can also use some or all of the L1 regularization selection method, the information gain selection method, or the Gini index selection method to perform feature extraction at the same time, and only weighted voting is required on the extraction results.

[0158] Among them, L1 regularization is a regularization term added to the loss function, which is used to constrain the sum of the absolute values ​​of the model parameters. During the training process, L1 regularization can make the weights of some features become 0, thereby realizing feature selection. In the embodiment of the present application, a suitable linear model can be selected, and an L1 regularization term can be added to the loss function, and the loss function can be minimized and the model parameters can be updated through optimization algorithms such as gradient descent. The trained model is then used to extract features from the sample features. The features with a weight of 0 are the features that are screened out, and the features with non-0 weights are the extracted valid features.

[0159] Among them, the formula used for L1 regularization can be shown as follows:

[0160]

[0161] w is the model parameter, b is the bias term, x i is the feature vector of the i-th sample, y i is the label of the sample ((0 means loss, 1 means no loss) or (0 means no loss, 1 means loss due to reason 1, 2 means loss due to reason 2, etc.)), and C is the regularization coefficient. T Represents the transpose of the matrix w.

[0162] Information gain is an entropy-based feature selection method, mainly used in decision tree algorithms. By calculating the information gain of each feature to the data set, the importance of the feature can be measured. The greater the information gain, the greater the contribution of the feature to the classification task. The embodiment of the present application can calculate the entropy of the entire sample feature according to the category distribution of the object sample data, and for each feature, calculate the conditional entropy of the sample feature under the feature. Information gain is equal to the entropy of the sample feature minus the conditional entropy of the feature. By calculating the information gain for each feature, the most representative feature subset can be selected as the multidimensional feature according to the information gain value.

[0163] The Gini coefficient is an indicator to measure the purity of a data set, and is mainly used for feature selection in classification tasks. The smaller the Gini coefficient, the higher the purity of the data set. By calculating the Gini coefficient of each feature to the data set, the importance of the feature can be measured. The embodiment of the present application can calculate the Gini coefficient of the entire sample feature according to the category distribution of the sample feature, and for each feature, calculate the conditional Gini coefficient of the sample feature under the feature. The Gini index is equal to the conditional Gini coefficient of the feature minus the Gini coefficient of the sample feature. Calculate the Gini index for each feature, and then select the most representative feature subset as the multidimensional feature according to the Gini index value.

[0164] The data preprocessing in the embodiment of the present application may include data cleaning, feature encoding and feature normalization, wherein:

[0165] Data cleaning refers to the removal of duplicates, outlier processing, missing value filling and other operations on the original data to ensure data quality. In the present invention, the following steps can be used for data cleaning:

[0166] Deduplication: If there are duplicate order numbers in the user's purchase records, they need to be deleted.

[0167] Outlier processing: If there are extreme values ​​in the user activity data, outlier processing is required. You can use a boxplot to detect outliers and set the outliers as the upper and lower boundaries of the feature.

[0168] Missing value filling: If there are missing values ​​in the user information data, missing values ​​need to be filled. You can use the mean, median, or mode to fill. If there is a correlation between the feature to be filled and other features, you can use regression, interpolation, and other methods to fill.

[0169] Feature encoding refers to converting non-numeric features into numerical features for processing by machine learning algorithms. In the present invention, the following two feature encoding methods can be used:

[0170] One-hot encoding: One-hot encoding is to split a discrete feature into multiple binary features. For example, if the user's gender is "male" or "female", it can be converted into two features "male or female" and "female or not" using one-hot encoding.

[0171] Label encoding: Label encoding refers to representing a discrete feature with a set of integers. For example, if there are 30 provinces that a user belongs to, its label can be encoded as 0 to 29.

[0172] Feature normalization refers to converting the value ranges of different features into the same interval to avoid certain features from having too much influence on the model prediction results. In the present invention, the following two feature normalization methods can be used:

[0173] Min-max normalization: Min-max normalization refers to scaling the features to the interval [0,1]. The specific formula is as follows:

[0174]

[0175] Among them, x is the original feature and x′ is the normalized feature.

[0176] Z-score normalization: Z-score normalization refers to scaling the features to a normal distribution interval with a mean of 0 and a variance of 1. The specific formula is as follows:

[0177]

[0178] Among them, x is the original feature, μ and σ are the mean and standard deviation of the feature respectively, and x′ is the normalized feature.

[0179] In the embodiment of the present application, a multi-dimensional feature extraction method is provided. Based on the above implementation, feature extraction is performed through multiple feature selection methods, which can improve the feature selection speed and improve the prediction accuracy of the trained model.

[0180] Optionally, in the above Figure 2 On the basis of the corresponding embodiments, in another optional embodiment provided by the embodiment of the present application, before obtaining the sample labels corresponding to the multi-dimensional features, the method further includes:

[0181] Perform dimensionality reduction on multi-dimensional features.

[0182] In one or more embodiments, feature dimensionality reduction refers to compressing high-dimensional feature vectors into low-dimensional space to reduce computational complexity and improve model generalization ability. The embodiment of the present application can compress the high dimension of multi-dimensional features into low-dimensional space. Among them, the embodiment of the present application can use the principal component analysis (PCA) algorithm to perform feature dimensionality reduction on multi-dimensional features. The PCA algorithm can convert the original data into a new set of orthogonal features, retaining most of the information in the original data while removing redundant and weakly correlated information. Specifically, the multi-dimensional features can be mean normalized, and then the covariance matrix of the multi-dimensional features is calculated, and the covariance matrix is ​​subjected to eigenvalue decomposition to obtain eigenvalues ​​and eigenvectors. The eigenvalues ​​are sorted by size, and the first k eigenvectors are selected as the new feature space, then the multi-dimensional features can be projected into the new feature space to obtain the eigenvector after dimensionality reduction.

[0183] In the embodiment of the present application, a method for processing multi-dimensional features is provided. Based on the above implementation, by reducing the dimension of multi-dimensional features, the complexity of calculation is greatly reduced, and the error caused by redundant information is reduced.

[0184] Optionally, in the above Figure 2 On the basis of the corresponding embodiments, in another optional embodiment provided by the embodiment of the present application, the method further includes:

[0185] extracting a feature vector from the historical behavior data of the first object;

[0186] The second retention information corresponding to the feature vector is pushed to the first object.

[0187] In one or more embodiments, the historical behavior data may be included in the object data of the first object. The historical behavior data may be determined from the object data of the first object first, and then a feature vector may be extracted from the historical behavior data, that is, representative features are extracted from the historical behavior data as feature vectors, and then the feature vectors are matched with corresponding second retention information and pushed to the first object. The second retention information may be an incentive prompt for the historical behavior data. Specifically, according to the behavioral characteristics of different users, corresponding retention rules are defined and a rule base is established. The user historical data is preprocessed, feature encoded and feature normalized to extract representative feature vectors. The feature vector is matched with the rule base to find the retention rule applicable to the user, and then the corresponding second retention information is automatically pushed according to the matched retention rule. Exemplarily, if the user does not log in for a long time after registration, a welcome email may be sent to the user or incentives such as point rewards may be provided; if the user frequently initiates a refund application, after-sales service may be strengthened or a better product experience may be provided and notified to the user.

[0188] In the embodiment of the present application, a retention method based on historical behavior data is provided. Through the above implementation, the historical behavior data is matched with the corresponding second retention information, and the retention method for the first object can be accurately found to improve the retention effect.

[0189] Optionally, in the above Figure 2 On the basis of the corresponding embodiments, in another optional embodiment provided by the embodiment of the present application, the method further includes:

[0190] The second retention information corresponding to the loss reason of the first object is pushed to the first object.

[0191] In one or more embodiments, the churn reason can be determined by the first label output by the object churn prediction model, that is, the first label carries the churn reason, and the churn reason is matched with the corresponding second retention information and pushed to the first object. The second retention information can be an incentive prompt for the churn reason. Specifically, corresponding retention rules are defined according to the churn reasons of different users and a rule base is established. The churn reason is matched with the rule base to find the retention rule applicable to the first object, and then the corresponding second retention information is automatically pushed according to the matched retention rule. Exemplarily, if the user leaves because the product price is too high, they can be notified and provided with promotional activities such as coupons or discount codes; if the user leaves because the product function is insufficient, they can be notified of updated content and displayed with new functions or improved interfaces.

[0192] In the embodiment of the present application, a retention method based on the loss reason is provided. Through the above implementation, the loss reason is matched with the corresponding second retention information, and the retention method for the first object can be accurately found to improve the retention effect.

[0193] The object retention device in this application is described in detail below. Figure 4 , Figure 4 This is a schematic diagram of an embodiment of an object retention device in an embodiment of the present application. The object retention device 40 includes:

[0194] An acquisition unit 401 is used to acquire an object data cluster, where the object data cluster includes object data of multiple objects;

[0195] A classification unit 402 is used to classify the object data cluster by using an object loss prediction model based on a random forest algorithm to output a classification label, wherein the classification label corresponds to a plurality of objects one by one, and the classification label includes a first label, and the first label indicates a subject to be lost;

[0196] A determining unit 403 is configured to determine a first object among the multiple objects according to the first tag, and to determine a recommended product according to a first purchase record of the first object by using a collaborative filtering algorithm;

[0197] The push unit 404 is configured to push the first retention information to the first object according to the recommended product.

[0198] In the embodiment of the present application, a device for retaining an object is provided. The above device is used to first improve the prediction accuracy through the random forest algorithm, and then personalized recommendations are made based on the products purchased by the user to improve the attractiveness to the user, and retention information for the recommended products is provided to increase the user's desire to buy, and users who purchase the recommended products will not be further lost, thereby improving user retention.

[0199] Optionally, in the above Figure 4 On the basis of the corresponding embodiment, in another embodiment of the object retention device 40 provided in the embodiment of the present application,

[0200] The determining unit 403 is specifically used for:

[0201] determining a plurality of object similarities according to the first purchase record and a plurality of second purchase records, each second purchase record corresponding to a second object;

[0202] The recommended products are determined according to the products in the third purchase record, where the third purchase record is a purchase record whose object similarity in the plurality of second purchase records reaches a first preset threshold.

[0203] In an embodiment of the present application, a device for retaining an object is provided. By using the device, a commodity purchased by an object with high object similarity can be recommended to a first object, thereby increasing the attractiveness to the object.

[0204] Optionally, in the above Figure 4On the basis of the corresponding embodiment, in another embodiment of the object retention device 40 provided in the embodiment of the present application,

[0205] The determining unit 403 is specifically used for:

[0206] determining a plurality of commodity similarities based on a first commodity and a plurality of second commodities in a first purchase record, wherein the first purchase record does not include the plurality of second commodities;

[0207] The recommended product is determined according to the third product, where the third product is a product among the plurality of second products whose product similarity reaches a second preset threshold.

[0208] In an embodiment of the present application, a device for retaining an object is provided. By using the device, a commodity with a high similarity to the first commodity in the first purchase record can be recommended to the first object, thereby increasing the attractiveness to the object.

[0209] Optionally, in the above Figure 4 On the basis of the corresponding embodiment, in another embodiment of the object retention device 40 provided in the embodiment of the present application,

[0210] The push unit 404 is specifically used for:

[0211] A price reduction of the recommended product is notified to the first subject.

[0212] In an embodiment of the present application, a subject retention device is provided. By using the above device, the first subject can be notified of preferential measures for recommended products to increase the first subject's desire to buy.

[0213] Optionally, in the above Figure 4 On the basis of the corresponding embodiment, in another embodiment of the object retention device 40 provided in the embodiment of the present application,

[0214] The acquisition unit 401 is further used for:

[0215] Get object sample data;

[0216] The device further includes an extraction unit 405, which is specifically configured to:

[0217] Extract multi-dimensional features from object sample data;

[0218] The acquisition unit 401 is further used for:

[0219] Get sample labels corresponding to multi-dimensional features;

[0220] The apparatus further includes a training unit 406, which is specifically configured to:

[0221] Train the object churn prediction model based on multi-dimensional features and sample labels.

[0222] In an embodiment of the present application, a subject retention device is provided. By using the device, a subject loss prediction model is trained based on multi-dimensional features and sample labels, which can save training time and computing power.

[0223] Optionally, in the above Figure 4 On the basis of the corresponding embodiment, in another embodiment of the object retention device 40 provided in the embodiment of the present application,

[0224] The determining unit 403 is further configured to:

[0225] The sample label is determined according to the data cutoff time in the multi-dimensional features, and the sample label indicates the retained object or the lost object.

[0226] In an embodiment of the present application, a device for retaining an object is provided. By using the device, sample labels are determined based on the data cutoff time in the multi-dimensional features. If a user does not log in for a long time, it is obvious that the user has been lost, which can improve the accuracy of the sample label.

[0227] Optionally, in the above Figure 4 On the basis of the corresponding embodiment, in another embodiment of the object retention device 40 provided in the embodiment of the present application,

[0228] The acquisition unit 401 is further used for:

[0229] Get a sample label, where the sample label indicates a retained object or a lost object with a loss reason.

[0230] In an embodiment of the present application, a device for retaining an object is provided. By using the device, the lost objects are divided into multiple categories according to the loss reasons, and retention measures can be accurately configured based on the loss reasons to improve the retention success rate.

[0231] Optionally, in the above Figure 4 On the basis of the corresponding embodiment, in another embodiment of the object retention device 40 provided in the embodiment of the present application,

[0232] The training unit 406 is specifically used for:

[0233] Divide the multi-dimensional features and sample labels into multiple data sets, including a training set;

[0234] Performing a grid search on the training set to determine a first hyperparameter combination, where the first hyperparameter combination is a hyperparameter combination that meets preset requirements;

[0235] The training set is trained according to the first hyperparameter combination to obtain an object churn prediction model.

[0236] In an embodiment of the present application, a subject retention device is provided. By using the above device, the performance and effect of model learning can be improved by selecting a first hyperparameter combination that meets preset requirements from a training set as the initial training parameters of the subject loss prediction model.

[0237] Optionally, in the above Figure 4 On the basis of the corresponding embodiment, in another embodiment of the object retention device 40 provided in the embodiment of the present application,

[0238] The multiple data sets also include a test set, and the device also includes a verification unit 407, which is specifically used to:

[0239] The object churn prediction model is verified through the test set.

[0240] In an embodiment of the present application, a subject retention device is provided. The above device is used to verify the subject loss prediction model by dividing the test set from the sample data. Since the test set and the training set are different, the verification accuracy can be improved.

[0241] Optionally, in the above Figure 4 On the basis of the corresponding embodiment, in another embodiment of the object retention device 40 provided in the embodiment of the present application,

[0242] The determining unit 403 is further configured to:

[0243] One of the multiple data sets is determined as a test set in turn, and the data sets except the test set in the multiple subsets are used as training sets.

[0244] In an embodiment of the present application, a subject retention device is provided. By using the device, the verification accuracy can be improved through a cross-validation algorithm.

[0245] Optionally, in the above Figure 4 On the basis of the corresponding embodiment, in another embodiment of the object retention device 40 provided in the embodiment of the present application,

[0246] The extraction unit 405 is specifically used for:

[0247] Perform data preprocessing on the object sample data to obtain sample features;

[0248] Select sample features according to the L1 regularization selection method to obtain multi-dimensional features; or,

[0249] Select sample features according to the information gain selection method to obtain multi-dimensional features; or,

[0250] The sample features are selected according to the Gini coefficient selection method to obtain multi-dimensional features.

[0251] In an embodiment of the present application, a device for retaining an object is provided. By using the device and performing feature extraction through multiple feature selection methods, the feature selection speed can be increased and the prediction accuracy of the trained model can be increased.

[0252] Optionally, in the above Figure 4 On the basis of the corresponding embodiment, in another embodiment of the object retention device 40 provided in the embodiment of the present application,

[0253] The device further includes a dimension reduction unit 408, which is specifically used for:

[0254] Perform dimensionality reduction on multi-dimensional features.

[0255] In an embodiment of the present application, a device for retaining an object is provided. By adopting the device, the complexity of calculation is greatly reduced by reducing the dimension of multi-dimensional features, and the error caused by redundant information is reduced.

[0256] Optionally, in the above Figure 4 On the basis of the corresponding embodiment, in another embodiment of the object retention device 40 provided in the embodiment of the present application,

[0257] The extraction unit 405 is further used for:

[0258] extracting a feature vector from the historical behavior data of the first object;

[0259] The push unit 404 is also used for:

[0260] The second retention information corresponding to the feature vector is pushed to the first object.

[0261] In an embodiment of the present application, a device for retaining an object is provided. By using the device, historical behavior data is matched with corresponding second retention information, a method for retaining the first object can be accurately found, thereby improving the retention effect.

[0262] Optionally, in the above Figure 4 On the basis of the corresponding embodiment, in another embodiment of the object retention device 40 provided in the embodiment of the present application,

[0263] The push unit 404 is also used for:

[0264] The second retention information corresponding to the loss reason of the first object is pushed to the first object.

[0265] In an embodiment of the present application, a device for retaining an object is provided. By using the device, the loss reason is matched with the corresponding second retention information, and a method for retaining the first object can be accurately found, thereby improving the retention effect.

[0266] Figure 53 is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. The computer device 300 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 322 (for example, one or more processors) and memory 332, and one or more storage media 330 (for example, one or more mass storage devices) storing application programs 342 or data 344. Among them, the memory 332 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the computer device. Furthermore, the central processing unit 322 can be configured to communicate with the storage medium 330 to execute a series of instruction operations in the storage medium 330 on the computer device 300.

[0267] The computer device 300 may also include one or more power supplies 326, one or more wired or wireless network interfaces 350, one or more input and output interfaces 358, and / or one or more operating systems 341, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0268] The steps performed by the retention server in the above embodiment can be based on the Figure 5 The computer device structure shown.

[0269] A computer-readable storage medium is also provided in an embodiment of the present application, on which a computer program is stored. When the computer program is executed by a processor, the steps of the methods described in the above embodiments are implemented.

[0270] A computer program product is also provided in an embodiment of the present application, including a computer program, which, when executed by a processor, implements the steps of the methods described in the above embodiments.

[0271] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0272] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0273] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0274] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0275] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0276] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for retaining an object, characterized in that: include: Acquire an object data cluster, wherein the object data cluster includes object data of a plurality of objects; Classifying the object data clusters by an object loss prediction model based on a random forest algorithm to output classification labels, wherein the classification labels correspond one-to-one to the multiple objects, and the classification labels include a first label indicating a subject to be lost; determining a first object among the plurality of objects according to the first tag; Using a collaborative filtering algorithm, determine a recommended product according to the first purchase record of the first object; Pushing first retention information to the first object according to the recommended product.

2. The method according to claim 1, characterized in that Determining the recommended product according to the first purchase record of the first object includes: determining a plurality of object similarities according to the first purchase record and a plurality of second purchase records, each of the second purchase records corresponding to a second object; The recommended product is determined according to the product in a third purchase record, where the third purchase record is a purchase record whose object similarity in the plurality of second purchase records reaches a first preset threshold.

3. The method according to claim 1, characterized in that Determining the recommended product according to the first purchase record of the first object includes: determining a plurality of commodity similarities based on a first commodity and a plurality of second commodities in the first purchase record, wherein the first purchase record does not include the plurality of second commodities; The recommended product is determined according to a third product, where the third product is a product among the plurality of second products whose product similarity reaches a second preset threshold.

4. The method according to claim 1, characterized in that: The retaining information indicates that the price of the recommended product is reduced, and the pushing the retaining information to the first object according to the recommendation information includes: The first object is notified of a price reduction of the recommended product.

5. The method according to claim 1, characterized in that Before classifying the object data clusters by the object loss prediction model based on the random forest algorithm to output classification labels, the method further includes: Get object sample data; Extracting multi-dimensional features from the object sample data; Obtaining sample labels corresponding to the multi-dimensional features; The object churn prediction model is trained according to the multi-dimensional features and the sample labels.

6. The method according to claim 5, characterized in that The obtaining of sample labels corresponding to the multi-dimensional features includes: The sample label is determined according to the data cutoff time in the multi-dimensional feature, and the sample label indicates a retained object or a lost object.

7. The method according to claim 5, characterized in that The obtaining of sample labels corresponding to the multi-dimensional features includes: The sample label is obtained, where the sample label indicates a retained object or a lost object with a loss reason.

8. The method according to claim 5, characterized in that The training of the object churn prediction model according to the multi-dimensional features and the sample labels comprises: Dividing the multi-dimensional features and the sample labels into a plurality of data sets, wherein the plurality of data sets include a training set; Performing a grid search on the training set to determine a first hyperparameter combination, where the first hyperparameter combination is a hyperparameter combination that meets preset requirements; The training set is trained according to the first hyperparameter combination to obtain the object churn prediction model.

9. The method according to claim 8, characterized in that The plurality of data sets also include a test set, and the method further includes: The object churn prediction model is verified using the test set.

10. The method according to claim 9, characterized in that The method further comprises: One of the multiple data sets is sequentially determined as the test set, and the data sets in the multiple subsets except the test set are used as the training sets.

11. The method according to claim 5, characterized in that The extracting multi-dimensional features from the object sample data comprises: Performing data preprocessing on the object sample data to obtain sample features; Selecting the sample features according to an L1 regularization selection method to obtain the multi-dimensional features; or, Selecting the sample features according to the information gain selection method to obtain the multi-dimensional features; or, The sample features are selected according to the Gini coefficient selection method to obtain the multi-dimensional features.

12. The method according to claim 5 or 11, characterized in that: Before obtaining the label corresponding to the multi-dimensional feature, the method further includes: Performing dimension reduction on the multi-dimensional features.

13. The method according to claim 1, characterized in that The method further comprises: extracting a feature vector from the historical behavior data of the first object; Pushing second retention information corresponding to the feature vector to the first object.

14. The method according to claim 7, characterized in that The method further comprises: Second retention information corresponding to the loss reason of the first object is pushed to the first object.

15. A user retention device, characterized in that: include: An acquisition unit, configured to acquire an object data cluster, wherein the object data cluster includes object data of a plurality of objects; a classification unit, configured to classify the object data cluster by using an object loss prediction model based on a random forest algorithm to output classification labels, wherein the classification labels correspond one-to-one to the plurality of objects, and the classification labels include a first label indicating a subject to be lost; a determining unit, configured to determine a first object among the multiple objects according to the first tag, and to determine a recommended product according to a first purchase record of the first object by using a collaborative filtering algorithm; A push unit is used to push first retention information to the first object according to the recommended product.

16. A computer device, characterized in that: include: memories, transceivers, processors, and bus systems; Wherein, the memory is used to store programs; The processor is used to execute the program in the memory, including executing the method according to any one of claims 1 to 14; The bus system is used to connect the memory and the processor so that the memory and the processor can communicate with each other.

17. A computer-readable storage medium comprising instructions, which, when executed on a computer, causes the computer to perform the method according to any one of claims 1 to 14.

18. A computer program product, characterized in that When the computer program product is executed on a computer, the computer performs the method according to any one of claims 1 to 14.

Images