A data processing method, device, apparatus, storage medium, and product

Abstract

Embodiments of the present application disclose a data processing method, device and equipment, storage medium and product. The method comprises: obtaining attribute information of a merchant and an object, and interaction data between the merchant and the object; constructing a relationship network based on the attribute information of the merchant and the object, and the interaction data between the merchant and the object, each merchant and each object corresponding to a node in the relationship network, an edge between a target merchant node and a target object node being used to represent an interaction relationship between the target merchant and the target object, and the flow loss probability of each object being predicted from at least two dimensions through the interaction data and the relationship network. It can be seen that the relationship network is first constructed to mine the positional relationship between the merchant and the object, and then the flow loss probability of each object is predicted from two dimensions of interaction and position, so as to improve the accuracy of the prediction result.

Description

Technical Field

[0001] This application relates to the field of computer technology, specifically to a data processing method, a data processing apparatus, a computer device, a computer-readable storage medium, and a data processing product. Background Technology

[0002] With advancements in technological research, a massive amount of business has shifted from offline to online. Common online businesses include product transactions, online bill payments, and online education. In scenarios involving product transactions, merchants, in addition to considering product prices, also need to maintain their customers (potential clients) to ensure a stable customer base. Customer maintenance methods include targeted recovery strategies to prevent customer churn. Research has found that the probability of potential client churn is one of the main bases for developing client recovery strategies, and how to accurately predict the probability of client churn is currently a hot research topic. Summary of the Invention

[0003] This application provides a data processing method, apparatus, device, computer-readable storage medium, and product that can more accurately predict the probability of object loss.

[0004] On one hand, embodiments of this application provide a data processing method, including:

[0005] Obtain attribute information of merchants and objects, as well as interaction data between merchants and objects;

[0006] Based on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, a relationship network is constructed. Each merchant and each object corresponds to a node in the relationship network. The edge between the target merchant node and the target object node is used to represent the interaction relationship between the target merchant and the target object.

[0007] By leveraging interaction data and relationship networks, we can predict the churn probability of each object from both interaction and location dimensions.

[0008] On one hand, embodiments of this application provide a data processing apparatus, which includes:

[0009] The acquisition unit is used to acquire attribute information of merchants and objects, as well as interaction data between merchants and objects;

[0010] The processing unit is used to construct a relationship network based on the attribute information of merchants and objects, as well as the interaction data between merchants and objects. Each merchant and each object corresponds to a node in the relationship network, and the edge between the target merchant node and the target object node is used to represent the interaction relationship between the target merchant and the target object.

[0011] And it is used to predict the churn probability of individual objects from the interaction dimension and the location dimension by using interaction data and relationship networks.

[0012] In one implementation, the processing unit is configured to construct a relationship network based on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, specifically for:

[0013] M+N nodes are generated using the attribute information of M merchants and N objects. Each node is configured using the attribute information of a merchant or an object. M and N are both positive integers.

[0014] If the interaction data indicates that there is an interaction between the i-th merchant and the j-th object, then establish an edge between the node of the i-th merchant and the node of the j-th object to obtain a relationship network, where i is a positive integer less than or equal to M and j is a positive integer less than or equal to N.

[0015] In one embodiment, the processing unit is further configured to:

[0016] By performing low-dimensional space mapping on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, we can obtain the embedded representation of merchants and objects.

[0017] Based on the embedded representation of the i-th merchant and the embedded representation of the j-th object, configure the weight of the edge connection between the node of the i-th merchant and the node of the j-th object;

[0018] The weight value is proportional to the strength of the target interaction relationship, which is the interaction relationship between the i-th merchant and the j-th object.

[0019] In one implementation, the processing unit is configured to perform low-dimensional space mapping on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, to obtain embedded representations of merchants and objects, specifically for:

[0020] Feature extraction is performed on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, to obtain the feature information of merchants and objects;

[0021] The feature information of merchants and objects is encoded separately to obtain the embedded representation of merchants and objects; or, the feature information of merchants and objects is jointly encoded to obtain the embedded representation of merchants and objects.

[0022] In one implementation, the embedded representations of merchants and objects are obtained through encoding by an encoder, the training process of which includes:

[0023] The encoder to be trained is invoked to jointly encode the feature information of the sample objects and sample merchants to obtain the embedded representations of the sample objects and sample merchants.

[0024] The decoder to be trained is invoked to jointly decode the embedded representations of the sample objects and sample merchants to obtain the decoding results of the sample objects and sample merchants.

[0025] Based on the difference between the decoding result and the feature information, the encoder and decoder to be trained are jointly trained to obtain the trained encoder.

[0026] In one embodiment, the processing unit is further configured to:

[0027] If the churn probability of the target object is greater than the probability threshold, then the churn probability of the target object, the embedded representation of the target object and the characteristic information of the merchant are used to determine the target object recovery strategy.

[0028] Among them, the target object's embedding representation is obtained by mapping the target object's attribute information and associated interaction data in a low-dimensional space; the merchant's feature information is obtained by extracting features from the merchant's attribute information and associated interaction data; the strength of the recovery strategy is proportional to the target object's churn probability.

[0029] In one implementation, the strategy for winning back the target is determined through a decision model, and the optimization process of the decision model includes:

[0030] The first recovery strategy is used to recover the objects in the first set, and the object churn rate of the first set after the first recovery strategy is implemented is calculated.

[0031] The second recovery strategy is used to recover the objects in the second set, and the churn rate of the objects in the second set after the second recovery strategy is implemented is statistically analyzed. The churn probability of the objects in the second set is matched with the churn probability of the objects in the first set.

[0032] If the churn rate of the first set of objects is less than that of the second set of objects, then the decision model to be optimized is optimized based on the embedded representation of the objects in the first set and the first recovery strategy.

[0033] If the churn rate of the first set of objects is greater than that of the second set of objects, then the decision model to be optimized is optimized based on the embedded representation of the objects in the second set and the second recovery strategy.

[0034] In one implementation, the processing unit is configured to determine a recovery strategy for the target object based on the churn probability of the target object, the embedded representation of the target object, and the characteristic information of the merchant, specifically configured to:

[0035] Based on the embedded representation of the target object, analyze the importance of the target object;

[0036] The strength of the strategy is determined based on the importance of the target object and the probability of churn of the target object. The strength of the strategy is directly proportional to the importance of the target object.

[0037] Based on the merchant's characteristic information, determine the candidate recovery strategies for the merchant;

[0038] Based on strategy strength, a recovery strategy for the target object is selected from the candidate recovery strategies; the strength of the recovery strategy is matched with the overall strategy strength.

[0039] In one implementation, the processing unit is configured to predict the churn probability of each object from the interaction dimension and the location dimension using interaction data and relationship networks, specifically configured to:

[0040] The interaction dimension evaluation index of the target object is calculated through interaction data. The interaction dimension evaluation index includes at least one of the following: the activity level of the target object within a preset time period, and the loyalty of the target object.

[0041] The location dimension evaluation index of the target object is calculated through the relation network. The location dimension evaluation index includes at least one of the following: degree centrality of the target object's nodes, proximity centrality of the target object's nodes, and betweenness centrality of the target object's nodes.

[0042] The probability prediction model is invoked to predict the churn probability of the target object based on the interaction dimension evaluation index and the location dimension evaluation index of the target object.

[0043] In one implementation, the processing unit is configured to calculate the interaction dimension evaluation index of the target object through interaction data, specifically for:

[0044] The first and second time points are obtained through interactive data. The first time point is the time when the target object first interacts with the merchant within a preset time period, and the second time point is the time when the target object last interacts with the merchant within a preset time period.

[0045] Calculate the difference between the second time point and the first time point to obtain the first duration;

[0046] The ratio between the first duration and the second duration is calculated to obtain the loyalty of the target object. The second duration is the duration corresponding to the preset time period.

[0047] In one implementation, the merchant's attribute information includes product information for P products, where P is a positive integer; the processing unit is used to construct a relationship network based on the attribute information of the merchant and the object, as well as the interaction data between the merchant and the object, specifically for:

[0048] M+N+P nodes are generated using the attribute information of M merchants, the attribute information of N objects, and the product information of P products. Each node is configured using the attribute information of a merchant, the attribute information of an object, or the product information of a product. M and N are both positive integers.

[0049] If the interaction data indicates that there is an interaction between the i-th merchant and the j-th object, then establish an edge between the node of the i-th merchant and the node of the j-th object to obtain a relationship network, where i is a positive integer less than or equal to M and j is a positive integer less than or equal to N.

[0050] If the k-th product is associated with the i-th merchant, then establish an edge between the node of the i-th merchant and the node of the k-th product to obtain a relational network, where k is a positive integer less than or equal to P.

[0051] In one implementation, the processing unit is configured to predict the churn probability of each object from the interaction dimension and the location dimension using interaction data and relationship networks, specifically configured to:

[0052] The interaction dimension evaluation index of the target object is calculated through interaction data. The interaction dimension evaluation index includes at least one of the following: the activity level of the target object within a preset time period, and the loyalty of the target object.

[0053] The position dimension evaluation index of each node is calculated through the relation network. The position dimension evaluation index includes at least one of the following: node degree, node clustering coefficient, node betweenness centrality, node degree centrality, and node proximity centrality.

[0054] The probability prediction model is invoked to predict the churn probability of the target object based on the interaction dimension evaluation index of the target object and the location dimension evaluation index of each node.

[0055] Accordingly, this application provides a computer device comprising:

[0056] Memory, which stores computer programs;

[0057] A processor is used to load computer programs to implement the above data processing methods.

[0058] Accordingly, this application provides a computer-readable storage medium storing a computer program adapted to be loaded by a processor and executed by the above-described data processing method.

[0059] Accordingly, this application provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the aforementioned data processing method.

[0060] In this embodiment, attribute information of merchants and objects, as well as interaction data between them, are obtained. Based on this attribute information and interaction data, a relationship network is constructed. Each merchant and each object corresponds to a node in the relationship network. The edges between target merchant nodes and target object nodes represent the interaction relationship between them. Through the interaction data and the relationship network, the churn probability of each object is predicted from at least two dimensions. Therefore, by first constructing a relationship network to mine the positional relationships between merchants and objects, and then comprehensively predicting the churn probability of each object from both interaction and positional dimensions, the accuracy of the prediction results is improved. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 A data processing scenario diagram provided in an embodiment of this application;

[0063] Figure 2 A flowchart illustrating a data processing method provided in an embodiment of this application;

[0064] Figure 3 A flowchart illustrating another data processing method provided in this application embodiment;

[0065] Figure 4a A decision support page diagram provided for an embodiment of this application;

[0066] Figure 4b Another decision support page diagram provided for embodiments of this application;

[0067] Figure 4c A schematic diagram of a message push window provided in an embodiment of this application;

[0068] Figure 4d An example of a promotional push page provided in this application embodiment;

[0069] Figure 4e A merchant operation page diagram provided in this application embodiment;

[0070] Figure 5 This is a schematic diagram of the structure of a data processing device provided in an embodiment of this application;

[0071] Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0072] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0073] This application relates to embedded representations and relational networks, which are briefly introduced below.

[0074] Embedded representation: used to map high-dimensional data to a low-dimensional space, allowing the data to retain its original structure and relationships in the low-dimensional space. In this application, embedded representation is used to represent the feature information of merchants and objects in a low dimension for better analysis and processing.

[0075] Relationship network: Consists of nodes and edges connecting them. In this application, a node represents a merchant or an object, and the edge connecting the target merchant node and the target object node indicates the interaction relationship between them. Specifically, the edge may be weightless (indicating whether there is an interaction between the target merchant and the target object) or may carry weight (further indicating the strength of the relationship between them).

[0076] Based on the above-mentioned embedded representation and relational network, embodiments of this application provide a data processing scheme that can more accurately predict the churn probability of objects. Figure 1 A data processing scenario diagram provided for an embodiment of this application, such as Figure 1As shown, the data processing scenario provided in this application includes a terminal device 101, a server 102, and a computer device 103. The data processing solution provided in this application can be executed by the server 102. Specifically, the computer device can be a terminal device or a server. The terminal device can include, but is not limited to, smartphones (such as Android phones, iOS phones, etc.), tablet computers, portable personal computers, mobile internet devices (MIDs), smart voice interaction devices, smart home appliances, vehicle terminals, aircraft, wearable devices, etc. This application embodiment does not limit this. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. This application embodiment does not limit this.

[0077] It should be noted that, Figure 1 The number of terminal devices, servers, and computer devices is for illustrative purposes only and does not constitute an actual limitation of this application. Terminal device 101, server 102, and computer device 103 can be connected via wired or wireless means, and this application does not impose any limitations on this.

[0078] The general flow of the data processing solution provided in this application is as follows:

[0079] (1) Server 102 obtains the attribute information of merchants and objects, as well as the interaction data between merchants and objects. The attribute information of merchants includes, but is not limited to: merchant name, main products, store rating, and license certificates. In addition, the attribute information of merchants may further include product information (such as product name, product price, sales volume, etc.). The attribute information of objects includes, but is not limited to: basic information of objects, membership status, and region. The interaction data between merchants and objects may include, but is not limited to: dwell time, collection history, like history, and evaluation information.

[0080] (2) Server 102 constructs a relationship network based on the attribute information of merchants and objects, as well as the interaction data between merchants and objects; wherein each merchant and each object corresponds to a node in the relationship network, and the edge between the target merchant node and the target object node is used to represent the interaction relationship between the target merchant and the target object. In one embodiment, server 102 can configure different nodes based on the attribute information of merchants and objects, and configure the edges between each node according to whether there is interaction between the merchants and objects. Further, server 102 can also perform low-dimensional space mapping on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, to obtain the embedded representation of merchants and objects, and configure the weight of the edges based on the embedded representation of merchants and objects. In another embodiment, the attribute information of the merchant includes the product information of P products, where P is a positive integer. Server 102 can configure different nodes based on the merchant's attribute information, the object's attribute information, and the product's product information (that is, based on the previous implementation method, a node corresponding to the product is added), and configure the connection edges between each node according to the interaction relationship between the merchant and the object, and the association relationship between the merchant and the product (such as whether the merchant sells the product).

[0081] (3) Server 102 predicts the churn probability of each object from the interaction dimension and location dimension through interaction data and relationship network. The churn probability of an object is used to indicate the likelihood of the object churning. Specifically, it can be the churn probability of a merchant, the churn probability of a merchant platform (containing multiple merchants), or the churn probability of a product. In one implementation, on the one hand, server 102 calculates the interaction dimension evaluation index of the target object through interaction data; wherein, the interaction dimension evaluation index includes at least one of the following: the activity level of the target object within a preset time period, and the loyalty of the target object. On the other hand, server 102 calculates the location dimension evaluation index of the target object through relationship network; wherein, the location dimension evaluation index includes at least one of the following: the degree centrality of the target object's nodes, the proximity centrality of the target object's nodes, and the betweenness centrality of the target object's nodes. After obtaining the interaction dimension evaluation index and the location dimension evaluation index of the target object, server 102 calls the probability prediction model to predict the churn probability of the target object based on the interaction dimension evaluation index and the location dimension evaluation index of the target object.

[0082] It should be noted that the data processing solution provided in this application involves collecting attribute information and behavioral data of the objects being processed. When applying the various embodiments in this application to specific products or technologies, relevant data collection can only be carried out after obtaining permission or consent from the relevant objects, and the collection, use, and processing of relevant data must comply with the relevant laws, regulations, and standards of the relevant countries and regions.

[0083] In this embodiment, attribute information of merchants and objects, as well as interaction data between them, are obtained. Based on this attribute information and interaction data, a relationship network is constructed. Each merchant and each object corresponds to a node in the relationship network. The edges between target merchant nodes and target object nodes represent the interaction relationship between them. Through the interaction data and the relationship network, the churn probability of each object is predicted from at least two dimensions. Therefore, by first constructing a relationship network to mine the positional relationships between merchants and objects, and then comprehensively predicting the churn probability of each object from both interaction and positional dimensions, the accuracy of the prediction results is improved.

[0084] Based on the above data processing scheme, this application proposes a more detailed data processing method. The data processing method proposed in this application will be described in detail below with reference to the accompanying drawings.

[0085] Please see Figure 2 , Figure 2 A flowchart illustrating a data processing method provided in this application embodiment, which can be executed by a computer device; for example, by... Figure 1 The computer device 102 shown performs the operation. (As...) Figure 2 As shown, the data processing method may include the following steps S201-S203:

[0086] S201. Obtain the attribute information of merchants and objects, as well as the interaction data between merchants and objects.

[0087] Attribute information is used to reflect the characteristics of merchants and objects. Different merchants may have partially identical attribute information (e.g., store rating, main products, etc.); similarly, different objects may also have partially identical attribute information (e.g., membership status, etc.). Specifically, object attribute information includes, but is not limited to: basic object information, membership status, and region. Merchant attribute information includes, but is not limited to: merchant name, main products, store rating, and licenses. In one embodiment, merchant attribute information may further include product information (e.g., product name, product price, sales volume, etc.).

[0088] Interaction data between merchants and objects can reflect both the existence of a connection between them and the strength of that relationship. For example, the more interactions between a merchant and an object within a preset timeframe, the stronger the relationship (stickiness, loyalty) between them. Interaction data can include, but is not limited to, dwell time, collection history, likes, and reviews. A target merchant can be any merchant, and a target object can be any object.

[0089] In one implementation, the computer device can obtain attribute information of merchants and objects, as well as interaction data between merchants and objects, from channels such as the merchant's database, log files, and third-party data providers. This application does not limit this.

[0090] S202. Construct a relationship network based on the attribute information of merchants and objects, as well as the interaction data between merchants and objects.

[0091] Each merchant and each object corresponds to a node in the relationship network. Different merchants correspond to different nodes, different objects correspond to different nodes, and the nodes corresponding to merchants and objects are also different. The edges between the target merchant node and the target object node are used to represent the interaction relationship between the target merchant and the target object. The target merchant is any one of the merchants, and the target object is any one of the objects.

[0092] This method can be applied to situations where a merchant has multiple objects, in which the relationship network includes only a node corresponding to the merchant; it can also be applied to situations where multiple merchants have multiple objects, where the merchants can be related (e.g., belonging to the same platform, the same operator, etc.) or independent of each other. For related merchants, edges can be established between the corresponding nodes. In other words, the data processing method provided in this application can be used to predict the churn probability of a single merchant or to predict the churn probability of multiple merchants.

[0093] In one implementation, a computer device generates M+N nodes based on the attribute information of M merchants and N objects. Each node is configured using the attribute information of a merchant or an object, where M and N are both positive integers. The computer device then generates edges between the M merchant nodes and the N object nodes based on the interaction data between the merchants and objects, thus obtaining a relationship network. Specifically, if the interaction data indicates an interaction between the i-th merchant and the j-th object (such as browsing, liking, sharing, saving, purchasing, etc.), then an edge is established between the node of the i-th merchant and the node of the j-th object, where i is a positive integer less than or equal to M, and j is a positive integer less than or equal to N. Conversely, if there is no interaction between the i-th merchant and the j-th object, then the node of the i-th merchant and the node of the j-th object are not directly connected.

[0094] In one embodiment, after obtaining the relationship network based on the above steps, the computer device can further perform low-dimensional space mapping on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, to obtain embedded representations of merchants and objects. Then, the computer device further configures the weights of each edge in the relationship network based on the embedded representations of merchants and objects. Specifically, the computer device configures the weights of the edges between the nodes of the i-th merchant and the j-th object based on the embedded representation of the i-th merchant and the embedded representation of the j-th object; wherein the weight value is proportional to the strength of the target interaction relationship, which is the interaction relationship between the i-th merchant and the j-th object.

[0095] In another implementation, the merchant's attribute information includes product information for P products, where P is a positive integer. The computer device generates M+N+P nodes using the attribute information of M merchants, the attribute information of N objects, and the product information of P products. Each node is configured using the attribute information of a merchant, an object, or a product, where M and N are both positive integers. Next, the computer device generates edges between M merchant nodes and N object nodes based on the interaction data between merchants and objects, and generates edges between M merchant nodes and P product nodes based on the association between products and merchants (e.g., whether the products sold by the merchant include the target product), thus obtaining a relationship network. Specifically, if the interaction data indicates an interaction between the i-th merchant and the j-th object, the computer device establishes an edge between the node of the i-th merchant and the node of the j-th object, where i is a positive integer less than or equal to M, and j is a positive integer less than or equal to N. If the k-th product is associated with the i-th merchant, the computer device establishes an edge connecting the node of the i-th merchant and the node of the k-th product, where k is a positive integer less than or equal to P. It is understood that the computer device can further configure the weights of each edge based on the embedded representations of merchants, objects, and products, which will not be elaborated upon here.

[0096] Optionally, the computer equipment can also establish connections between merchant nodes based on relationships between merchants (such as belonging to a platform or a business operator), establish connections between product nodes based on relationships between products (such as belonging to a category or a merchant), and establish connections between object nodes based on relationships between objects (such as sharing relationships, fan relationships, or friend relationships).

[0097] S203. Predict the churn probability of each object from the interaction dimension and the location dimension by using interaction data and relationship network.

[0098] The churn probability of an object indicates the likelihood of that object being churned. Specifically, it can be the churn probability for a single merchant, a merchant platform (containing multiple merchants), or a product. Predicting the churn probability of various objects using computer devices through interaction data and relationship networks, from the dimensions of interaction and location, can be understood as: comprehensively predicting the churn probability of various objects based on both interaction data and relationship networks, considering both interaction and location dimensions.

[0099] In one implementation, on the one hand, the computer device calculates the interaction dimension evaluation index of the target object through interaction data; wherein the interaction dimension evaluation index includes at least one of the following: the activity level of the target object within a preset time period, and the loyalty level of the target object. On the other hand, the computer device calculates the location dimension evaluation index of the target object through a relationship network; wherein the location dimension evaluation index includes at least one of the following: the degree centrality of the target object's nodes, the proximity centrality of the target object's nodes, and the betweenness centrality of the target object's nodes. After obtaining the interaction dimension evaluation index and the location dimension evaluation index of the target object, the computer device calls a probability prediction model (or evaluation function) to predict the churn probability of the target object based on the interaction dimension evaluation index and the location dimension evaluation index.

[0100] In another implementation, the relationship network includes nodes corresponding to the product. On one hand, the computer device calculates interaction dimension evaluation metrics for the target object using interaction data; these metrics include at least one of the following: the target object's activity level within a preset time period, and the target object's loyalty. On the other hand, the computer device calculates location dimension evaluation metrics for each node (considering the positional relationships of all nodes, not just the node corresponding to the object) through the relationship network; these metrics include at least one of the following: node degree, node clustering coefficient, node betweenness, node degree centrality, node proximity centrality, and node betweenness centrality. After obtaining the interaction dimension evaluation metrics for the target object and the location dimension evaluation metrics for each node, the computer device invokes a probability prediction model (or evaluation function) to predict the churn probability of the target object based on these metrics.

[0101] Optionally, computer devices can issue warnings to merchants based on the churn probability of each object, in order to reduce the actual churn rate of the object.

[0102] In this embodiment, attribute information of merchants and objects, as well as interaction data between them, are obtained. Based on this attribute information and interaction data, a relationship network is constructed. Each merchant and each object corresponds to a node in the relationship network. The edges between target merchant nodes and target object nodes represent the interaction relationship between them. Through the interaction data and the relationship network, the churn probability of each object is predicted from at least two dimensions. Therefore, by first constructing a relationship network to mine the positional relationships between merchants and objects, and then comprehensively predicting the churn probability of each object from both interaction and positional dimensions, the accuracy of the prediction results is improved.

[0103] Please see Figure 3 , Figure 3 A flowchart illustrating another data processing method provided in this application embodiment, which can be executed by a computer device; for example, by... Figure 1 The server 102 shown is executing.

[0104] like Figure 3 As shown, the data processing method may include the following steps S301-S306:

[0105] S301. Obtain the attribute information of merchants and objects, as well as the interaction data between merchants and objects.

[0106] For a detailed implementation of step S301, please refer to Figure 2 The implementation method of step S201 will not be described in detail here.

[0107] S302. Construct a relationship network based on the attribute information of merchants and objects, as well as the interaction data between merchants and objects.

[0108] In one implementation, the computer device generates M+N nodes using attribute information from M merchants and N objects. Each node is configured using attribute information from either a merchant or an object, where M and N are both positive integers. The computer device then generates edges between the M merchant nodes and the N object nodes based on the interaction data between the merchants and objects, thus obtaining a relationship network.

[0109] Furthermore, the computer device performs low-dimensional space mapping on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, to obtain embedded representations of merchants and objects.

[0110] In one embodiment, the computer device extracts features from the attribute information of merchants and objects, as well as the interaction data between merchants and objects, to obtain feature information of merchants and objects. After obtaining the feature information of merchants and objects, the computer device encodes the feature information of merchants and objects respectively to obtain embedded representations of merchants and objects. Specifically, this can be represented as:

[0111] e x =f e (x)

[0112] Among them, f e Here is the mapping function for the encoder, where x represents the feature information of the merchant or the feature information of the object, and e... x This is the embedded representation of the input data x.

[0113] In another embodiment, the computer device extracts features from the attribute information of merchants and objects, as well as the interaction data between merchants and objects, to obtain feature information of merchants and objects. After obtaining the feature information of merchants and objects, the computer device performs joint encoding on the feature information of merchants and objects to obtain the embedded representation of merchants and objects. Joint encoding can be specifically understood as encoding the joint feature information of merchants and objects, and the mutual influence between merchant feature information and object feature information is additionally considered during the encoding process. Specifically, it can be represented as follows:

[0114] (e c ,e u )=f e_joint (c,u)

[0115] Among them, f e_joint Let (c,u) be the mapping function of the encoder, (e) be the joint feature information of the merchant and the object, and (e) be the mapping function of the encoder. c ,e u This represents the joint embedding of merchants and objects. It can be understood that (e...) c ,e u ) in e c and e u It can be extracted and used individually. Furthermore, it can also represent the joint feature information of merchants and objects in a one-to-many manner; for example, (c,u1…u ... n ) or (c1…c m Compared to encoding the feature information of merchants and objects separately, the embedded representation of merchants and objects obtained through joint encoding can better reflect the relationship between merchants and objects.

[0116] In one implementation, the embedding representations of merchants and objects are obtained through encoding by an encoder. The training process of the independent encoder includes: calling the encoder to be trained to encode the feature information of the sample object or sample merchant to obtain the embedding representation of the sample object or sample merchant; calling the decoder to be trained to decode the embedding representation of the sample object or sample merchant to obtain the decoding result of the sample object or sample merchant; and jointly training the encoder and decoder to be trained based on the difference between the decoding result and the feature information to obtain the trained encoder. Specifically, this can be represented as:

[0117] min||xf d (f e (x))||

[0118] Among them, f e f is the mapping function of the encoder. d Here, x is the mapping function of the decoder, x is the feature information of the merchant or the feature information of the object, and |||| is the norm, used to measure the magnitude of the reconstruction error.

[0119] The training process of the joint encoder includes: calling the encoder to be trained to jointly encode the feature information of the sample object and the sample merchant, obtaining the embedding representation of the sample object and the sample merchant. Calling the decoder to be trained to jointly decode the embedding representation of the sample object and the sample merchant, obtaining the decoding result of the sample object and the sample merchant. Based on the difference between the decoding result and the feature information, the encoder and decoder to be trained are jointly trained to obtain the trained encoder. Specifically, this can be represented as:

[0120] min||(c,u)-f d_joint (f e_joint (c,u))||

[0121] Among them, f e_joint f is the mapping function of the encoder. d_joint Let (c,u) be the mapping function of the decoder, (c,u) be the joint feature information of the merchant and the object, and |||| be the norm, which is used to measure the magnitude of the reconstruction error.

[0122] After obtaining the embedded representations of merchants and objects, the computer device configures the weights of the edges between the nodes of the i-th merchant and the j-th object based on the embedded representations of the i-th merchant and the j-th object. The weights are proportional to the strength of the target interaction relationship, which is the interaction relationship between the i-th merchant and the j-th object. In one implementation, the computer device determines the distance between the i-th merchant and the j-th object in the low-dimensional space based on the embedded representations of the i-th merchant and the j-th object, and configures the weights of the edges between the nodes of the i-th merchant and the j-th object based on this distance and a distance mapping function. Specifically, this can be expressed as: w = f(d); where w is the weight, f(d) is the distance mapping function, and d is the distance between the i-th merchant and the j-th object in the low-dimensional space.

[0123] S303. Calculate the interaction dimension evaluation index of the target object through interaction data.

[0124] The interaction dimension evaluation metrics include at least one of the following: the target audience's activity level within a preset time period, and the target audience's loyalty. The target audience's activity level within the preset time period can be expressed as:

[0125]

[0126] Where A(u) represents the activity level of the target object within the preset time period, and N(u) represents the number of interactions of the target object within the preset time period. The number of interactions can be the number of interactions between the target object and one merchant, or the number of interactions between the target object and M merchants, and can be adjusted according to the required metrics. T represents the duration of the preset time period.

[0127] In one embodiment, the computer device acquires a first time point and a second time point through interactive data. The first time point is the time when the target object first interacts with a merchant (which can be one merchant or M merchants, depending on the required metrics) within a preset time period. The second time point is the time when the target object last interacts with a merchant within the preset time period. Then, the difference between the second time point and the first time point is calculated to obtain a first duration, and the ratio between the first duration and the second duration is calculated to obtain the target object's loyalty; where the second duration corresponds to the preset time period. The target object's loyalty can be expressed as:

[0128]

[0129] Where L(u) represents the loyalty of the target object. last (u) represents the second time point, which is the time point when the target object last interacted with the merchant within the preset time period. T first (u) represents the first time point, which is the time when the target object first interacts with the merchant within the preset time period. T represents the second duration, which is the duration corresponding to the preset time period.

[0130] S304. Calculate location dimension evaluation indicators through relationship networks.

[0131] The location dimension evaluation metrics include at least one of the following: node degree, node clustering coefficient, node betweenness centrality, node degree centrality, and node proximity centrality. The node degree can be expressed as:

[0132]

[0133] Among them, A vu It is the adjacency matrix of a relational network, where v and u are any two distinct nodes in the network, and A vu =1 indicates that there is an edge between node v and node u, A vu=0 indicates that there is no edge between node v and node u, where V is the set of nodes in the relational network.

[0134] The clustering coefficient of a node can be expressed as:

[0135]

[0136] Among them, E v Let d(v) represent the number of edges between node v and all connected nodes, d(v) be the degree of node v, and C(v) be the clustering coefficient of node v.

[0137] The betweenness centrality of a node can be expressed as:

[0138]

[0139] Where, σ st σ represents the number of shortest paths from node s to node t. st B(v) represents the number of nodes v in the shortest path from node s to node t, and B(v) is the betweenness centrality of node v.

[0140] The degree centrality of a node can be expressed as:

[0141]

[0142] Where d(u) represents the degree of node u (such as the node corresponding to the target object), H represents the number of nodes in the relational network, and D(u) is the degree centrality of node u.

[0143] The proximity centrality of a node can be expressed as:

[0144]

[0145] Where d(u,v) represents the shortest path length between node u and node v (such as the node corresponding to the target object and the node corresponding to the target merchant), H represents the number of nodes in the relationship network, and C(u) is the proximity centrality of node u.

[0146] S305. Invoke the probability prediction model, based on the interaction dimension evaluation index and location dimension evaluation index of the target object, to predict the churn probability of the target object.

[0147] In one implementation, the interaction dimension evaluation indicators for the target object include the target object's activity level A(u) and loyalty level L(u) within a preset time period; the location dimension evaluation indicators for the target object include the degree centrality D(u), proximity centrality C(u), and betweenness centrality B(u) of the target object's nodes. The prediction formula for the churn probability of the target object can be expressed as:

[0148] P(u)=f(A(u),L(u),D(u),C(u),B(u))

[0149] Where f is the probability prediction model (or evaluation function), which can be implemented using machine learning algorithms or statistical models such as logistic regression, support vector machine, and random forest. P(u) is the churn probability of the target object.

[0150] S306. If the churn probability of the target object is greater than the probability threshold, then based on the churn probability of the target object, the embedded representation of the target object and the characteristic information of the merchant, determine the target object recovery strategy.

[0151] The probability threshold can be a preset value or dynamically adjusted by the merchant according to actual needs. The target object's embedding representation is obtained by mapping the target object's attribute information and associated interaction data in a low-dimensional space. The merchant's feature information is obtained by extracting features from the merchant's attribute information and associated interaction data. The strength of the recovery strategy is directly proportional to the target object's churn probability. Specific recovery strategies may include, but are not limited to: sending promotional information, providing coupons, and distributing gifts. There can be one or more recovery strategies; combining multiple strategies can make the recovery strategy more targeted and further improve the success rate.

[0152] In one implementation, the formula for determining the recovery strategy for the target object can be expressed as:

[0153] S(u)=g(P(u),e u c)

[0154] Where g is the decision model (or decision support function), P(u) is the churn probability of the target object, and e u Let c be the embedded representation of the target object, and c be the feature information of the merchant (which can be replaced by the feature information of the product).

[0155] In one embodiment, the optimization process of the decision model includes: the computer device employing a first recovery strategy to recover objects in a first set, and calculating the churn rate (or retention rate) of the objects in the first set after implementing the first recovery strategy. The computer device employing a second recovery strategy to recover objects in a second set, and calculating the churn rate (or retention rate) of the objects in the second set after implementing the second recovery strategy; wherein the churn probability of objects in the second set matches the churn probability of objects in the first set; for example, both the first set and the second set include 10 objects, and the churn probability of the x-th object in the first set is the same as the churn probability of the x-th object in the second set, where x is an integer in the range [1, 10].

[0156] Furthermore, if the churn rate of the first set of objects is less than that of the second set of objects, the computer device optimizes the decision model to be optimized based on the embedded representation of the objects in the first set and the first recovery strategy. Conversely, if the churn rate of the first set of objects is greater than that of the second set of objects, the computer device optimizes the decision model to be optimized based on the embedded representation of the objects in the second set and the second recovery strategy. Optionally, if the churn rate of the first set of objects is equal to that of the second set of objects, the computer device randomly selects a recovery strategy and a corresponding set to optimize the decision model to be optimized; additionally, the computer device can add a third recovery strategy and a corresponding set for further comparison.

[0157] In another implementation, on one hand, the computer device analyzes the importance of the target object based on its embedded representation. For example, the computer device can determine the target object's position in a low-dimensional space based on its embedded representation, and determine the target object's importance based on its position in the low-dimensional space (such as its distance from the target merchant or its distance from a high-value object). Further, the computer device determines the strategy strength based on the target object's importance and its churn probability, where the strategy strength is proportional to the target object's importance (and its churn probability). On the other hand, the computer device determines candidate recovery strategies for the merchant using the merchant's characteristic information; for example, candidate recovery strategies for merchant A include strategies 1-5; and for merchant B, they include strategies 3-10. After determining the strategy strength and the merchant's candidate recovery strategies, the computer device selects a recovery strategy for the target object from the candidate strategies based on the strategy strength; wherein the strength of the recovery strategy matches the overall strategy strength.

[0158] Figure 4a This is a decision support page diagram provided in an embodiment of this application. This decision support page is displayed on the merchant's computer device based on the churn probability of the object after the above-described data processing method is executed. Figure 4a As shown, the decision support page 401 includes the churn probability of the target object 4011, basic information and historical interaction data of the target object 4012 (such as total purchase amount, last purchase time, number of likes, dwell time, number of favorites, etc.), and recommended recovery strategies 4013. Merchants can trigger the execution of the recommended recovery strategy 4013 through the confirmation button 4014. In addition, merchants can also adjust the recommended recovery strategy 4013 by triggering the adjustment button 4015, or select multiple objects for batch processing through the batch selection button 4016. When multiple objects are selected, 4012 can display the average attribute values ​​of multiple objects, or display the relevant information of multiple objects separately, and the churn probability 4011 is displayed similarly.

[0159] Figure 4b Another decision support page diagram provided for an embodiment of this application. For example... Figure 4b As shown, in Figure 4a In addition, the decision support page 401 can also display the time when the target object last triggered the execution of the recovery strategy, as well as the specific recovery strategy (displayed in area 4016), to assist merchants in making decisions.

[0160] Figure 4c This is a schematic diagram of a message push window provided in an embodiment of this application. Figure 4c As shown, when a merchant implements a recovery strategy and pushes a message to the target audience, the target audience's terminal device displays a message push window 402, which shows the message pushed by the merchant (such as an activity notification, product link, etc.).

[0161] Figure 4d This is an example of a promotional push page provided in an embodiment of this application. Figure 4d As shown, when a merchant implements a customer retention strategy and pushes a discount to the target audience, the target audience's terminal device displays the discount push page 403, which shows the discount information 4031. In addition, the discount push page 403 may also display the merchant's identifier 4032, as well as the entry (link) 4033 for the corresponding product or promotional activity, etc.

[0162] Figure 4e This is a screenshot of a merchant operation page provided as an embodiment of this application. Figure 4e As shown, on the merchant operation page 404, the merchant can update the probability threshold using the threshold update control 4041. The merchant can configure optional candidate recovery strategies (such as adding, deleting, or changing candidate recovery strategies) and adjust the priority of each recovery strategy (such as the priority of multiple candidate recovery strategies of the same strength) using the strategy editing control 4042.

[0163] The following section uses an e-commerce platform as an example to further illustrate the data processing method provided in this application.

[0164] (1) Computer equipment performs data collection and preprocessing. This specifically includes:

[0165] Computer devices collect relevant data from merchants and objects, including object attribute information (such as basic information, membership status, region, etc.), transaction records (such as purchase time, purchase amount, purchased goods, etc.), behavioral data (such as browsing history, collection history, evaluation history, etc.), as well as merchant basic information (such as store name, main products, store rating, etc.), product information (such as product name, price, sales volume, etc.), marketing activities, etc.

[0166] Computer equipment preprocesses the collected data, including data cleaning, noise reduction, and normalization. For example, it removes outliers and erroneous values, eliminates noise, and converts the data to a uniform scale.

[0167] (2) Computer devices embed merchants and objects into their representations. Specifically, this includes:

[0168] Computer devices employ autoencoders to embed feature information of merchants and objects. Using the feature information of merchants or objects as input data, the autoencoder is trained to obtain a low-dimensional embedding representation of the merchants or objects. To better capture the relationship between merchants and objects, the computer device can also use joint encoding, using joint feature information of merchants and objects as input data, and training the autoencoder to obtain a joint low-dimensional embedding representation of the merchants and objects.

[0169] (3) Computer devices construct relational networks. This specifically includes:

[0170] Computer equipment generates nodes in a relational network based on the attribute information of merchants and objects, and establishes edges between these nodes based on the interaction relationships between merchants and objects, thus obtaining the relational network. The interaction relationships between merchants and objects can be defined by the target behaviors of the objects (such as purchasing behavior, liking behavior, etc.); that is, if there is a target behavior between an object and a merchant, then an interaction relationship is determined to exist between them. Furthermore, the computer equipment can analyze the relational network (such as degree, clustering coefficient, betweenness, etc.) to obtain evaluation indicators of the node's positional dimension.

[0171] (4) The probability of churn for users predicted by computer equipment. Specifically, this includes:

[0172] On one hand, computer equipment calculates interaction-dimensional evaluation metrics based on interaction data between users and merchants. For example, it calculates the number of purchases and the amount spent within a certain period to assess user activity; it calculates the time interval between the user's first purchase and the most recent purchase to assess user loyalty. On the other hand, computer equipment calculates location-dimensional evaluation metrics based on the user's position and connections within the relationship network. For example, it calculates the user's degree centrality, proximity centrality, and betweenness centrality to assess the user's importance within the network. Finally, the computer equipment comprehensively considers both interaction-dimensional and location-dimensional evaluation metrics to predict the user's churn probability.

[0173] (5) Computer equipment provides early warning and decision support. This includes:

[0174] If the probability of a customer churns exceeds a certain threshold, the computer system issues an alert, reminding the merchant to take appropriate measures to retain the customer. For example, when the probability of a customer churns is greater than 0.5, the computer system sends an alert to the merchant, reminding them to pay attention to the customer.

[0175] Furthermore, computer devices can provide merchants with personalized decision support based on the embedded representation and churn probability of objects. For example, for high-value objects (such as those with large purchase amounts and high purchase frequency), merchants can provide exclusive promotional activities and dedicated services; for low-value objects (such as those with small purchase amounts and low purchase frequency), merchants can use methods such as email marketing and SMS push notifications to retain them.

[0176] In addition, in practical applications, computer equipment can also update and optimize at least one of steps (1)-(5) based on the recovery effect of the recovery strategy.

[0177] The embodiments of this application are as follows: Figure 2 Building upon the previous implementation, embedding representation technology is used to map the feature information of merchants and objects in a low dimension, enabling a more comprehensive reflection of the object's behavior and characteristics. Relationship networks can be used to uncover potential relationships between merchants and objects, reducing reliance on labeled data. Real-time updates to the interaction relationships between merchants and objects allow for dynamic assessment of object churn. Specifically, by updating the interaction behavior between merchants and objects in real time, the probability of object churn can be predicted promptly, providing merchants with more accurate early warnings and personalized decision support, further improving object retention rates.

[0178] The methods of the embodiments of this application have been described in detail above. In order to facilitate better implementation of the above solutions of the embodiments of this application, the apparatus of the embodiments of this application is provided below.

[0179] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of a data processing device provided in an embodiment of this application. Figure 5 The data processing device shown can be mounted in a computer device, which can specifically be a terminal device or a server. Figure 5 The data processing apparatus shown can be used to perform the above. Figure 2 and Figure 3 Some or all of the functionality described in the method embodiments. Please refer to [link / reference]. Figure 5 The data processing device includes:

[0180] Acquisition unit 501 is used to acquire attribute information of merchants and objects, as well as interaction data between merchants and objects;

[0181] The processing unit 502 is used to construct a relationship network based on the attribute information of merchants and objects, as well as the interaction data between merchants and objects. Each merchant and each object corresponds to a node in the relationship network, and the edge between the target merchant node and the target object node is used to represent the interaction relationship between the target merchant and the target object.

[0182] And it is used to predict the churn probability of individual objects from the interaction dimension and the location dimension by using interaction data and relationship networks.

[0183] In one implementation, the processing unit 502 is configured to construct a relationship network based on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, specifically for:

[0184] M+N nodes are generated using the attribute information of M merchants and N objects. Each node is configured using the attribute information of a merchant or an object. M and N are both positive integers.

[0185] If the interaction data indicates that there is an interaction between the i-th merchant and the j-th object, then establish an edge between the node of the i-th merchant and the node of the j-th object to obtain a relationship network, where i is a positive integer less than or equal to M and j is a positive integer less than or equal to N.

[0186] In one embodiment, the processing unit 502 is further configured to:

[0187] By performing low-dimensional space mapping on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, we can obtain the embedded representation of merchants and objects.

[0188] Based on the embedded representation of the i-th merchant and the embedded representation of the j-th object, configure the weight of the edge connection between the node of the i-th merchant and the node of the j-th object;

[0189] The weight value is proportional to the strength of the target interaction relationship, which is the interaction relationship between the i-th merchant and the j-th object.

[0190] In one implementation, the processing unit 502 is configured to perform low-dimensional space mapping on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, to obtain embedded representations of merchants and objects, specifically for:

[0191] Feature extraction is performed on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, to obtain the feature information of merchants and objects;

[0192] The feature information of merchants and objects is encoded separately to obtain the embedded representation of merchants and objects; or, the feature information of merchants and objects is jointly encoded to obtain the embedded representation of merchants and objects.

[0193] In one implementation, the embedded representations of merchants and objects are obtained through encoding by an encoder, the training process of which includes:

[0194] The encoder to be trained is invoked to jointly encode the feature information of the sample objects and sample merchants to obtain the embedded representations of the sample objects and sample merchants.

[0195] The decoder to be trained is invoked to jointly decode the embedded representations of the sample objects and sample merchants to obtain the decoding results of the sample objects and sample merchants.

[0196] Based on the difference between the decoding result and the feature information, the encoder and decoder to be trained are jointly trained to obtain the trained encoder.

[0197] In one embodiment, the processing unit 502 is further configured to:

[0198] If the churn probability of the target object is greater than the probability threshold, then the churn probability of the target object, the embedded representation of the target object and the characteristic information of the merchant are used to determine the target object recovery strategy.

[0199] Among them, the target object's embedding representation is obtained by mapping the target object's attribute information and associated interaction data in a low-dimensional space; the merchant's feature information is obtained by extracting features from the merchant's attribute information and associated interaction data; the strength of the recovery strategy is proportional to the target object's churn probability.

[0200] In one implementation, the strategy for winning back the target is determined through a decision model, and the optimization process of the decision model includes:

[0201] The first recovery strategy is used to recover the objects in the first set, and the object churn rate of the first set after the first recovery strategy is implemented is calculated.

[0202] The second recovery strategy is used to recover the objects in the second set, and the churn rate of the objects in the second set after the second recovery strategy is implemented is statistically analyzed. The churn probability of the objects in the second set is matched with the churn probability of the objects in the first set.

[0203] If the churn rate of the first set of objects is less than that of the second set of objects, then the decision model to be optimized is optimized based on the embedded representation of the objects in the first set and the first recovery strategy.

[0204] If the churn rate of the first set of objects is greater than that of the second set of objects, then the decision model to be optimized is optimized based on the embedded representation of the objects in the second set and the second recovery strategy.

[0205] In one implementation, the processing unit 502 is configured to determine a recovery strategy for the target object based on the churn probability of the target object, the embedded representation of the target object, and the characteristic information of the merchant, specifically configured to:

[0206] Based on the embedded representation of the target object, analyze the importance of the target object;

[0207] The strength of the strategy is determined based on the importance of the target object and the probability of churn of the target object. The strength of the strategy is directly proportional to the importance of the target object.

[0208] Based on the merchant's characteristic information, determine the candidate recovery strategies for the merchant;

[0209] Based on strategy strength, a recovery strategy for the target object is selected from the candidate recovery strategies; the strength of the recovery strategy is matched with the overall strategy strength.

[0210] In one implementation, the processing unit 502 is configured to predict the churn probability of each object from the interaction dimension and the location dimension using interaction data and relationship networks, specifically configured to:

[0211] The interaction dimension evaluation index of the target object is calculated through interaction data. The interaction dimension evaluation index includes at least one of the following: the activity level of the target object within a preset time period, and the loyalty of the target object.

[0212] The location dimension evaluation index of the target object is calculated through the relation network. The location dimension evaluation index includes at least one of the following: degree centrality of the target object's nodes, proximity centrality of the target object's nodes, and betweenness centrality of the target object's nodes.

[0213] The probability prediction model is invoked to predict the churn probability of the target object based on the interaction dimension evaluation index and the location dimension evaluation index of the target object.

[0214] In one implementation, the processing unit 502 is configured to calculate the interaction dimension evaluation index of the target object through interaction data, specifically for:

[0215] The first and second time points are obtained through interactive data. The first time point is the time when the target object first interacts with the merchant within a preset time period, and the second time point is the time when the target object last interacts with the merchant within a preset time period.

[0216] Calculate the difference between the second time point and the first time point to obtain the first duration;

[0217] The ratio between the first duration and the second duration is calculated to obtain the loyalty of the target object. The second duration is the duration corresponding to the preset time period.

[0218] In one implementation, the merchant's attribute information includes product information for P products, where P is a positive integer; the processing unit 502 is used to construct a relationship network based on the attribute information of the merchant and the object, as well as the interaction data between the merchant and the object, specifically for:

[0219] M+N+P nodes are generated using the attribute information of M merchants, the attribute information of N objects, and the product information of P products. Each node is configured using the attribute information of a merchant, the attribute information of an object, or the product information of a product. M and N are both positive integers.

[0220] If the interaction data indicates that there is an interaction between the i-th merchant and the j-th object, then establish an edge between the node of the i-th merchant and the node of the j-th object to obtain a relationship network, where i is a positive integer less than or equal to M and j is a positive integer less than or equal to N.

[0221] If the k-th product is associated with the i-th merchant, then establish an edge between the node of the i-th merchant and the node of the k-th product to obtain a relational network, where k is a positive integer less than or equal to P.

[0222] In one implementation, the processing unit 502 is configured to predict the churn probability of each object from the interaction dimension and the location dimension using interaction data and relationship networks, specifically configured to:

[0223] The interaction dimension evaluation index of the target object is calculated through interaction data. The interaction dimension evaluation index includes at least one of the following: the activity level of the target object within a preset time period, and the loyalty of the target object.

[0224] The position dimension evaluation index of each node is calculated through the relation network. The position dimension evaluation index includes at least one of the following: node degree, node clustering coefficient, node betweenness centrality, node degree centrality, and node proximity centrality.

[0225] The probability prediction model is invoked to predict the churn probability of the target object based on the interaction dimension evaluation index of the target object and the location dimension evaluation index of each node.

[0226] According to one embodiment of this application, Figure 2 and Figure 3 The data processing method shown may involve some steps that can be derived from... Figure 5 The data processing apparatus shown is executed by each unit within it. For example, Figure 2 Step S201 shown can be performed by Figure 5 The acquisition unit 501 shown is executed, and steps S202 and S203 can be performed by... Figure 5 The processing unit 502 shown executes the operation; Figure 3 Step S301 shown can be performed by Figure 5 The acquisition unit 501 shown is executed, and steps S302-S306 can be performed by... Figure 5 The processing unit 502 shown is executed. Figure 5 The data processing apparatus shown can be constructed by combining each unit individually or entirely into one or more other units, or one or more of the units can be further divided into multiple functionally smaller units. This achieves the same operation without affecting the technical effects of the embodiments of this application. The above-mentioned units are based on logical function division. In practical applications, the function of one unit can be implemented by multiple units, or the function of multiple units can be implemented by one unit. In other embodiments of this application, the data processing apparatus may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented collaboratively by multiple units.

[0227] According to another embodiment of this application, a general-purpose computing device, such as a computer device including processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM), can perform operations such as... Figure 2 and Figure 3 The computer program (including program code) for each step involved in the corresponding method shown, to construct such... Figure 5 The data processing apparatus shown herein, and the data processing method for implementing the embodiments of this application, are described. A computer program may be recorded on, for example, a computer-readable recording medium, loaded onto the aforementioned computing device via the computer-readable recording medium, and executed therein.

[0228] Based on the same inventive concept, the principle and beneficial effects of the data processing device provided in the embodiments of this application in solving the problem are similar to the principle and beneficial effects of the data processing method in the embodiments of this application in solving the problem. For the sake of brevity, the principle and beneficial effects of the method implementation can be referred to.

[0229] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. The computer device may be a terminal device or a server. Figure 6As shown, the computer device includes at least a processor 601, a communication interface 602, and a memory 603. The processor 601, communication interface 602, and memory 603 can be connected via a bus or other means. The processor 601 (or Central Processing Unit, CPU) is the computing and control core of the computer device. It can parse various instructions within the computer device and process various data. For example, the CPU can parse power-on / off commands issued by an object to the computer device and control the computer device to perform power-on / off operations; it can also transmit various interactive data between internal structures of the computer device, and so on. The communication interface 602 may optionally include a standard wired interface or a wireless interface (such as Wi-Fi, mobile communication interface, etc.), and under the control of the processor 601, it can be used to send and receive data; the communication interface 602 can also be used for data transmission and interaction within the computer device. The memory 603 is the storage device in the computer device, used to store programs and data. It can be understood that the memory 603 here can include the computer device's built-in memory, or it can include extended memory supported by the computer device. The memory 603 provides storage space for storing the operating system of the computer device, which may include, but is not limited to, Android, iOS, Windows Phone, etc. This application does not limit this.

[0230] This application embodiment also provides a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the processing system of the computer device. Furthermore, the storage space also stores computer programs suitable for loading and execution by the processor 601. It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device; optionally, it can also be at least one computer-readable storage medium located remotely from the aforementioned processor.

[0231] In one embodiment, processor 601 performs the following operations by running a computer program stored in memory 603:

[0232] Obtain attribute information of merchants and objects, as well as interaction data between merchants and objects;

[0233] Based on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, a relationship network is constructed. Each merchant and each object corresponds to a node in the relationship network. The edge between the target merchant node and the target object node is used to represent the interaction relationship between the target merchant and the target object.

[0234] By leveraging interaction data and relationship networks, we can predict the churn probability of each object from both interaction and location dimensions.

[0235] As an optional embodiment, the processor 601 constructs a relationship network based on the attribute information of merchants and objects, as well as the interaction data between merchants and objects. A specific embodiment of this implementation is as follows:

[0236] M+N nodes are generated using the attribute information of M merchants and N objects. Each node is configured using the attribute information of a merchant or an object. M and N are both positive integers.

[0237] If the interaction data indicates that there is an interaction between the i-th merchant and the j-th object, then establish an edge between the node of the i-th merchant and the node of the j-th object to obtain a relationship network, where i is a positive integer less than or equal to M and j is a positive integer less than or equal to N.

[0238] As an optional embodiment, the processor 601, by running a computer program in the memory 603, also performs the following operations:

[0239] By performing low-dimensional space mapping on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, we can obtain the embedded representation of merchants and objects.

[0240] Based on the embedded representation of the i-th merchant and the embedded representation of the j-th object, configure the weight of the edge connection between the node of the i-th merchant and the node of the j-th object;

[0241] The weight value is proportional to the strength of the target interaction relationship, which is the interaction relationship between the i-th merchant and the j-th object.

[0242] As an optional embodiment, the processor 601 performs low-dimensional space mapping on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, to obtain the embedded representation of merchants and objects. A specific embodiment of this is as follows:

[0243] Feature extraction is performed on the attribute information of merchants and objects, as well as the interaction data between merchants and objects, to obtain the feature information of merchants and objects;

[0244] The feature information of merchants and objects is encoded separately to obtain the embedded representation of merchants and objects; or, the feature information of merchants and objects is jointly encoded to obtain the embedded representation of merchants and objects.

[0245] As an optional implementation, the embedded representations of merchants and objects are obtained through encoding by an encoder, the training process of which includes:

[0246] The encoder to be trained is invoked to jointly encode the feature information of the sample objects and sample merchants to obtain the embedded representations of the sample objects and sample merchants.

[0247] The decoder to be trained is invoked to jointly decode the embedded representations of the sample objects and sample merchants to obtain the decoding results of the sample objects and sample merchants.

[0248] Based on the difference between the decoding result and the feature information, the encoder and decoder to be trained are jointly trained to obtain the trained encoder.

[0249] As an optional embodiment, the processor 601, by running a computer program in the memory 603, also performs the following operations:

[0250] If the churn probability of the target object is greater than the probability threshold, then the churn probability of the target object, the embedded representation of the target object and the characteristic information of the merchant are used to determine the target object recovery strategy.

[0251] Among them, the target object's embedding representation is obtained by mapping the target object's attribute information and associated interaction data in a low-dimensional space; the merchant's feature information is obtained by extracting features from the merchant's attribute information and associated interaction data; the strength of the recovery strategy is proportional to the target object's churn probability.

[0252] As an optional implementation, the strategy for winning back the target is determined through a decision model, and the optimization process of the decision model includes:

[0253] The first recovery strategy is used to recover the objects in the first set, and the object churn rate of the first set after the first recovery strategy is implemented is calculated.

[0254] The second recovery strategy is used to recover the objects in the second set, and the churn rate of the objects in the second set after the second recovery strategy is implemented is statistically analyzed. The churn probability of the objects in the second set is matched with the churn probability of the objects in the first set.

[0255] If the churn rate of the first set of objects is less than that of the second set of objects, then the decision model to be optimized is optimized based on the embedded representation of the objects in the first set and the first recovery strategy.

[0256] If the churn rate of the first set of objects is greater than that of the second set of objects, then the decision model to be optimized is optimized based on the embedded representation of the objects in the second set and the second recovery strategy.

[0257] As an optional embodiment, the processor 601 determines the target object's recovery strategy based on the target object's churn probability, the target object's embedded representation, and the merchant's feature information. A specific embodiment of this is as follows:

[0258] Based on the embedded representation of the target object, analyze the importance of the target object;

[0259] The strength of the strategy is determined based on the importance of the target object and the probability of churn of the target object. The strength of the strategy is directly proportional to the importance of the target object.

[0260] Based on the merchant's characteristic information, determine the candidate recovery strategies for the merchant;

[0261] Based on strategy strength, a recovery strategy for the target object is selected from the candidate recovery strategies; the strength of the recovery strategy is matched with the overall strategy strength.

[0262] As an optional embodiment, the processor 601 predicts the churn probability of each object from the interaction dimension and the location dimension through interaction data and relationship network.

[0263] The interaction dimension evaluation index of the target object is calculated through interaction data. The interaction dimension evaluation index includes at least one of the following: the activity level of the target object within a preset time period, and the loyalty of the target object.

[0264] The location dimension evaluation index of the target object is calculated through the relation network. The location dimension evaluation index includes at least one of the following: degree centrality of the target object's nodes, proximity centrality of the target object's nodes, and betweenness centrality of the target object's nodes.

[0265] The probability prediction model is invoked to predict the churn probability of the target object based on the interaction dimension evaluation index and the location dimension evaluation index of the target object.

[0266] As an optional embodiment, the processor 601 calculates the interaction dimension evaluation index of the target object through interaction data in the following specific embodiment:

[0267] The first and second time points are obtained through interactive data. The first time point is the time when the target object first interacts with the merchant within a preset time period, and the second time point is the time when the target object last interacts with the merchant within a preset time period.

[0268] Calculate the difference between the second time point and the first time point to obtain the first duration;

[0269] The ratio between the first duration and the second duration is calculated to obtain the loyalty of the target object. The second duration is the duration corresponding to the preset time period.

[0270] As an optional embodiment, the merchant's attribute information includes product information for P products, where P is a positive integer; the processor 601 constructs a relationship network based on the attribute information of the merchant and the object, as well as the interaction data between the merchant and the object. A specific embodiment of this is as follows:

[0271] M+N+P nodes are generated using the attribute information of M merchants, the attribute information of N objects, and the product information of P products. Each node is configured using the attribute information of a merchant, the attribute information of an object, or the product information of a product. M and N are both positive integers.

[0272] If the interaction data indicates that there is an interaction between the i-th merchant and the j-th object, then establish an edge between the node of the i-th merchant and the node of the j-th object to obtain a relationship network, where i is a positive integer less than or equal to M and j is a positive integer less than or equal to N.

[0273] If the k-th product is associated with the i-th merchant, then establish an edge between the node of the i-th merchant and the node of the k-th product to obtain a relational network, where k is a positive integer less than or equal to P.

[0274] As an optional embodiment, the processor 601 predicts the churn probability of each object from the interaction dimension and the location dimension through interaction data and relationship network.

[0275] The interaction dimension evaluation index of the target object is calculated through interaction data. The interaction dimension evaluation index includes at least one of the following: the activity level of the target object within a preset time period, and the loyalty of the target object.

[0276] The position dimension evaluation index of each node is calculated through the relation network. The position dimension evaluation index includes at least one of the following: node degree, node clustering coefficient, node betweenness centrality, node degree centrality, and node proximity centrality.

[0277] The probability prediction model is invoked to predict the churn probability of the target object based on the interaction dimension evaluation index of the target object and the location dimension evaluation index of each node.

[0278] Based on the same inventive concept, the principle and beneficial effects of the computer device provided in the embodiments of this application in solving the problem are similar to the principle and beneficial effects of the data processing method in the embodiments of this application in solving the problem. Please refer to the principle and beneficial effects of the implementation of the method. For the sake of brevity, they will not be repeated here.

[0279] This application also provides a computer-readable storage medium storing a computer program adapted to be loaded by a processor and to execute the data processing method described in the above method embodiments.

[0280] This application also provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the data processing method described above.

[0281] The steps in the method of this application embodiment can be adjusted, combined, or deleted according to actual needs.

[0282] The modules in the device of this application embodiment can be merged, divided, and deleted according to actual needs.

[0283] In the embodiments of this application, the term "module" or "unit" refers to a computer program or part of a computer program with a predetermined function, which works together with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0284] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc.

[0285] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Those skilled in the art will understand that all or part of the processes for implementing the above embodiments and equivalent variations made in accordance with the claims of this application are still within the scope of this application.

Claims

1. A data processing method, characterized in that, The method includes: Obtain the attribute information of the merchant and the object, as well as the interaction data between the merchant and the object; Based on the attribute information of the merchants and the objects, as well as the interaction data between the merchants and the objects, a relationship network is constructed. Each merchant and each object corresponds to a node in the relationship network. The edge between the target merchant node and the target object node is used to represent the interaction relationship between the target merchant and the target object. By using the interaction data and the relationship network, the churn probability of each object can be predicted from the interaction dimension and the location dimension.

2. The method as described in claim 1, characterized in that, The construction of a relationship network based on the attribute information of the merchants and the objects, as well as the interaction data between the merchants and the objects, includes: M+N nodes are generated using the attribute information of M merchants and N objects. Each node is configured using the attribute information of a merchant or an object. M and N are both positive integers. If the interaction data indicates that there is an interaction between the i-th merchant and the j-th object, then an edge is established between the node of the i-th merchant and the node of the j-th object to obtain a relationship network, where i is a positive integer less than or equal to M and j is a positive integer less than or equal to N.

3. The method as described in claim 2, characterized in that, The method further includes: The attribute information of the merchants and the objects, as well as the interaction data between the merchants and the objects, are mapped in a low-dimensional space to obtain the embedded representations of the merchants and the objects; Based on the embedded representation of the i-th merchant and the embedded representation of the j-th object, configure the weight of the edge connection between the node of the i-th merchant and the node of the j-th object; The weight value is proportional to the strength of the target interaction relationship, which is the interaction relationship between the i-th merchant and the j-th object.

4. The method as described in claim 3, characterized in that, The step of performing low-dimensional space mapping on the attribute information of the merchant and the object, as well as the interaction data between the merchant and the object, to obtain the embedded representation of the merchant and the object includes: Feature extraction is performed on the attribute information of the merchant and the object, as well as the interaction data between the merchant and the object, to obtain the feature information of the merchant and the object; The feature information of the merchant and the object are encoded separately to obtain the embedded representation of the merchant and the object; or, the feature information of the merchant and the object are jointly encoded to obtain the embedded representation of the merchant and the object.

5. The method as described in claim 3 or 4, characterized in that, The embedded representations of the merchants and the objects are obtained through encoding by an encoder, the training process of which includes: The encoder to be trained is invoked to jointly encode the feature information of the sample object and the sample merchant to obtain the embedded representation of the sample object and the sample merchant; The decoder to be trained is invoked to jointly decode the embedded representations of the sample object and the sample merchant to obtain the decoding results of the sample object and the sample merchant; Based on the difference between the decoding result and the feature information, the encoder and the decoder to be trained are jointly trained to obtain the trained encoder.

6. The method as described in claim 1, characterized in that, The method further includes: If the churn probability of the target object is greater than the probability threshold, then based on the churn probability of the target object, the embedded representation of the target object and the feature information of the merchant, a recovery strategy for the target object is determined. The embedded representation of the target object is obtained by low-dimensional spatial mapping of the target object's attribute information and associated interaction data; the merchant's feature information is obtained by feature extraction of the merchant's attribute information and associated interaction data; and the strength of the recovery strategy is proportional to the churn probability of the target object.

7. The method as described in claim 6, characterized in that, The strategy for winning back the target object is determined through a decision model, and the optimization process of the decision model includes: The first recovery strategy is used to recover the objects in the first set, and the churn rate of the objects in the first set after the first recovery strategy is executed is calculated. The second recovery strategy is used to recover the objects in the second set, and the churn rate of the objects in the second set after the second recovery strategy is executed is calculated. The churn probability of the objects in the second set is matched with the churn probability of the objects in the first set. If the churn rate of the first set of objects is less than that of the second set of objects, then the decision model to be optimized is optimized based on the embedded representation of the objects in the first set and the first recovery strategy. If the churn rate of the first set of objects is greater than that of the second set of objects, then the decision model to be optimized is optimized based on the embedded representation of the objects in the second set and the second recovery strategy.

8. The method as described in claim 6, characterized in that, The step of determining the recovery strategy for the target object based on the churn probability of the target object, the embedded representation of the target object, and the feature information of the merchant includes: Based on the embedded representation of the target object, the importance of the target object is analyzed; The strategy strength is determined based on the importance of the target object and the churn probability of the target object, and the strategy strength is proportional to the importance of the target object; Based on the merchant's characteristic information, candidate recovery strategies for the merchant are determined; Based on the strength of the strategy, a recovery strategy for the target object is selected from the candidate recovery strategies; the strength of the recovery strategy is matched with the strength of the strategy.

9. The method as described in claim 1, characterized in that, The step of predicting the churn probability of each object from the interaction dimension and the location dimension using the interaction data and the relationship network includes: The interaction dimension evaluation index of the target object is calculated based on the interaction data. The interaction dimension evaluation index includes at least one of the following: the activity level of the target object within a preset time period, and the loyalty of the target object. The location dimension evaluation index of the target object is calculated through the relation network. The location dimension evaluation index includes at least one of the following: degree centrality of the nodes of the target object, proximity centrality of the nodes of the target object, and betweenness centrality of the nodes of the target object. The probability prediction model is invoked to predict the churn probability of the target object based on the interaction dimension evaluation index and the location dimension evaluation index of the target object.

10. The method as described in claim 9, characterized in that, The calculation of the interaction dimension evaluation index of the target object through the interaction data includes: The first time point and the second time point are obtained through the interaction data. The first time point is the time point when the target object first interacts with the merchant within a preset time period, and the second time point is the time point when the target object last interacts with the merchant within the preset time period. Calculate the difference between the second time point and the first time point to obtain the first duration; The loyalty of the target object is obtained by calculating the ratio between the first duration and the second duration, where the second duration is the duration corresponding to the preset time period.

11. The method as described in claim 1, characterized in that, The merchant's attribute information includes product information for P products, where P is a positive integer; the construction of a relationship network based on the attribute information of the merchant and the object, as well as the interaction data between the merchant and the object, includes: M+N+P nodes are generated using the attribute information of M merchants, the attribute information of N objects, and the product information of P products. Each node is configured using the attribute information of a merchant, the attribute information of an object, or the product information of a product. M and N are both positive integers. If the interaction data indicates that there is an interaction between the i-th merchant and the j-th object, then establish an edge between the node of the i-th merchant and the node of the j-th object to obtain a relationship network, where i is a positive integer less than or equal to M and j is a positive integer less than or equal to N. If the k-th product is associated with the i-th merchant, then an edge is established between the node of the i-th merchant and the node of the k-th product to obtain a relational network, where k is a positive integer less than or equal to P.

12. The method as described in claim 11, characterized in that, The step of predicting the churn probability of each object from the interaction dimension and the location dimension using the interaction data and the relationship network includes: The interaction dimension evaluation index of the target object is calculated based on the interaction data. The interaction dimension evaluation index includes at least one of the following: the activity level of the target object within a preset time period, and the loyalty of the target object. The position dimension evaluation index of each node is calculated through the relation network. The position dimension evaluation index includes at least one of the following: node degree, node clustering coefficient, node betweenness centrality, node degree centrality, and node proximity centrality. The probability prediction model is invoked to predict the churn probability of the target object based on the interaction dimension evaluation index of the target object and the location dimension evaluation index of each node.

13. A data processing apparatus, characterized in that, The data processing device includes: The acquisition unit is used to acquire attribute information of merchants and objects, as well as interaction data between the merchants and objects; The processing unit is used to construct a relationship network based on the attribute information of the merchants and the objects, as well as the interaction data between the merchants and the objects. Each merchant and each object corresponds to a node in the relationship network, and the edge between the target merchant node and the target object node is used to represent the interaction relationship between the target merchant and the target object. And for predicting the churn probability of each object from the interaction dimension and the location dimension using the interaction data and the relationship network.

14. A computer device, characterized in that, include: A memory, wherein a computer program is stored; A processor for loading the computer program to implement the data processing method as described in any one of claims 1-12.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in any one of claims 1-12.

16. A computer program product, characterized in that, The computer program product includes a computer program adapted to be loaded by a processor and execute the data processing method as described in any one of claims 1-12.

Images