Resource recovery channel selection method and device, storage medium and electronic equipment

Abstract

The invention discloses a resource recovery channel selection method and device, a storage medium and electronic equipment. The method relates to the field of artificial intelligence, and comprises the steps of generating a transaction node graph according to transaction information of a target object, the transaction node graph being used for indicating a transaction relationship between the target object and a candidate object; determining a first feature of the target object based on the transaction node graph, the first feature being used for indicating information of the target object; calculating a matching parameter corresponding to at least one recovery channel based on a second feature corresponding to at least one to-be-selected recovery channel and the first feature; and under the condition that the matching parameters greater than the preset value exist, determining the recovery channel corresponding to the maximum matching parameter as a target recovery channel. Through the method and the device, the problem of relatively low efficiency of a method for determining a recovery channel of fund resources in related technologies is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular, to a resource recovery channel selection method and device, a storage medium and an electronic device. BACKGROUND

[0002] At present, in the process of a user (a financial resource user) licensed by a third party institution and a merchant (a service provider) conducting a transaction, the user can make the third party institution (a financial institution) pay the financial resource for a short period of time free of charge by virtue of the user's own evaluation level. The user can return the financial resource paid by the third party object for a predetermined period of time within a predetermined period of time. If the user does not return the above-mentioned financial resource within the predetermined period of time, or the third party institution believes that the user may not be able to return the above-mentioned financial resource within the predetermined period of time, the third party institution needs to use a predetermined method to urge the user to return (collect) the above-mentioned financial resource.

[0003] With the increase in the number of users licensed by the third party institution, the number of people that the third party institution needs to collect also increases, and the collection methods and ways of the third party institution are different for different user situations. However, in the prior art, it is usually necessary to manually determine whether the user needs to be collected, and manually determine what collection method is used for the user. This process is confirmed by manual experience, which is subjective and inefficient.

[0004] In view of the problem of low efficiency of the method for determining the recovery channel of the financial resource in the related art, an effective solution has not been proposed. SUMMARY

[0005] The main purpose of the present application is to provide a resource recovery channel selection method and device, a storage medium and an electronic device to solve the problem of low efficiency of the method for determining the recovery channel of the financial resource in the related art.

[0006] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a resource recovery channel selection method is provided. The method comprises: generating a transaction node graph according to transaction information of a target object, wherein the transaction node graph is used to indicate a transaction relationship between the target object and a candidate object; determining a first feature of the target object based on the transaction node graph, wherein the first feature is used to indicate information of the target object; calculating a matching parameter corresponding to each of at least one recovery channel based on a second feature corresponding to each of the at least one recovery channel and the first feature; and in the case where there is a matching parameter greater than a predetermined value, determining the recovery channel corresponding to the greatest matching parameter as a target recovery channel.

[0007] To achieve the above object, according to another aspect of the present application, a resource recycling channel selection device is provided. The device comprises: a node graph generation unit configured to generate a transaction node graph according to transaction information of a target object, wherein the transaction node graph is used to indicate a transaction relationship between the target object and candidate objects; a feature extraction unit configured to determine a first feature of the target object based on the transaction node graph, wherein the first feature is used to indicate information of the target object; a matching parameter calculation unit configured to calculate a matching parameter corresponding to each of at least one recycling channel to be selected based on a second feature corresponding to each of the at least one recycling channel and the first feature; and a channel determination unit configured to determine a recycling channel corresponding to the largest matching parameter as a target recycling channel if there is a matching parameter larger than a preset value.

[0008] Optionally, the feature extraction unit comprises: a first feature determination module configured to determine a target feature of the target object and a candidate feature corresponding to each of the at least one candidate object; a first mean value calculation module configured to calculate a mean value of the candidate feature corresponding to each of the candidate objects having a neighboring relationship with the target object to obtain a candidate feature mean value; a first feature fusion module configured to fuse the candidate feature mean value and the target feature to obtain a fusion feature, and multiply the fusion feature by a first weight matrix to obtain a weighted feature, wherein the first weight matrix is used to change the dimension of the feature; and a first regularization module configured to determine a ratio between the weighted feature and a length of the weighted feature as the first feature.

[0009] Optionally, the first feature determination module is further configured to determine an i-th candidate feature corresponding to each of the at least one candidate object based on a reference feature of a reference object in the transaction node graph, wherein the reference object is an object having a transaction relationship with the at least one candidate object, and i is an integer greater than 1; the first mean value calculation module is further configured to calculate a mean value of the i-th candidate feature corresponding to each of the candidate objects having a neighboring relationship with the target object to obtain an i-th candidate feature mean value; the first feature fusion module is further configured to fuse an (i-1)-th first feature and the i-th candidate feature mean value to obtain an i-th fusion feature, and multiply the i-th fusion feature by the first weight matrix to obtain an i-th weighted feature; and the first regularization module is further configured to determine a ratio between the i-th weighted feature and a length of the i-th weighted feature as an i-th first feature; and in a case where i is equal to a preset number of rounds, the i-th first feature is determined as the first feature of the target object.

[0010] Optionally, the matching parameter calculation unit comprises: a first traversal module, configured to traverse the M recycling channels to be selected, and perform the following operations on the nth recycling channel: perform feature fusion on the nth second feature corresponding to the nth recycling channel and the first feature to obtain an nth matching feature, wherein M is an integer greater than 0, and n is an integer greater than 0 and less than or equal to M; multiply the nth matching feature by a second weight matrix and add a first bias to obtain an nth matching value; activate the nth matching value to obtain an nth matching parameter; and in the case where n is equal to M, determine the respective matching parameters of the M recycling channels.

[0011] Optionally, the first traversal module is further configured to: determine an nth channel feature corresponding to the nth recycling channel based on channel information of the nth recycling channel; and multiply the nth channel feature by a third weight matrix and add a second bias to obtain an nth second feature.

[0012] Optionally, the resource recycling channel selection apparatus further comprises: a model parameter training unit, configured to traverse Q sample objects, and determine a sample feature of a pth sample object based on a sample node graph of the pth sample object in the Q sample objects, wherein Q is an integer greater than 0, and p is an integer greater than 0 and less than or equal to Q; determine a pth sample recycling channel and a pth matching parameter of the pth sample object based on the pth sample feature and a to-be-trained first weight matrix, a to-be-trained second weight matrix, a to-be-trained first bias, a to-be-trained third weight matrix, and a to-be-trained second bias; determine a pth resource recycling value and a pth resource input value corresponding to the sample recycling channel of the pth sample object; calculate a pth first difference value between the pth resource recycling value and the pth resource input value, and a pth first summation value between the pth resource recycling value and the pth resource input value; determine a pth indication value as a ratio between a product of the pth matching parameter and the pth first difference value and the pth first summation value; update the to-be-trained first weight matrix, the to-be-trained second weight matrix, the to-be-trained first bias, the to-be-trained third weight matrix, and the to-be-trained second bias based on the pth indication value; and in the case where a sum of Q indication values is minimum, determine the current to-be-trained first weight matrix, the to-be-trained second weight matrix, the to-be-trained first bias, the to-be-trained third weight matrix, and the to-be-trained second bias as the first weight matrix, the second weight matrix, the first bias, the third weight matrix, and the second bias.

[0013] In the embodiment of the present application, a transaction node graph is generated according to transaction information of a target object, wherein the transaction node graph is used to indicate a transaction relationship between the target object and a candidate object; a first feature of the target object is determined based on the transaction node graph, wherein the first feature is used to indicate information of the target object; a matching parameter corresponding to each of the at least one recycling channel is calculated based on a second feature corresponding to each of the at least one recycling channel and the first feature; and in a case where the matching parameter is greater than a preset value, the recycling channel corresponding to the largest matching parameter is determined as a target recycling channel. After the target object is determined, a transaction node graph from which a transaction feature of the target object can be extracted can be determined based on transaction information of the target object, the first feature of the target object can be extracted through the transaction node graph, and the first feature can indicate a transaction between the target object and the candidate object. The second feature of the recycling channel to be selected can be extracted, and a matching parameter between the target object and the recycling channel can be calculated based on the first feature and the second feature, so as to determine which channel is suitable for the target object. Finally, whether to recycle the target object is determined according to the size of the matching parameter, and in a case where the recycling is determined, which recycling channel is used for the target object is determined. Thus, the problem that the method for determining a recycling channel of a fund resource in the related art has low efficiency is solved. BRIEF DESCRIPTION OF DRAWINGS

[0014] The accompanying drawings, which form a part of the present application, are intended to provide further understanding of the present application, and are used to interpret the application and its specification, and do not constitute improper limitations to the present application. In the drawings:

[0015] Figure 1 Fig. 1 shows a hardware structure block diagram of a computer terminal for implementing a method for selecting a resource recycling channel;

[0016] Figure 2 Fig. 2 is a flowchart of a method for selecting a resource recycling channel according to an embodiment of the present application;

[0017] Figure 3 Fig. 3 is a schematic diagram of a method for selecting a resource recycling channel according to an embodiment of the present application;

[0018] Figure 4 Fig. 4 is a schematic diagram of another method for selecting a resource recycling channel according to an embodiment of the present application;

[0019] Figure 5 Fig. 5 is a schematic diagram of a device for selecting a resource recycling channel according to an embodiment of the present application;

[0020] Figure 6 Fig. 6 is a structure block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] In the following, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort should fall into the scope of protection of the present application.

[0022] It should be noted that the terms "first", "second" and the like in the description and claims of the present application and the above drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in other than the order illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a list of steps or units need not be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to such processes, methods, products or devices.

[0023] It should be noted that the information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for display, analyzed data, etc.) involved in the present application are information and data authorized by the user or authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of related data comply with relevant laws, regulations and standards, necessary security measures are taken, do not violate public order and good customs, and provide corresponding operation portal for user to choose authorization or refusal. For example, the system and related users or institutions are provided with an interface to provide the user with a corresponding operation portal for the user to choose to agree or refuse the automatic decision result; if the user chooses to refuse, the expert decision process is entered.

[0024] Embodiment 1

[0025] According to the embodiments of the present application, a method for selecting a resource recycling channel is also provided. It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in a different order.

[0026] The method provided by the embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1A hardware structure block diagram of a computer terminal (or mobile device) for implementing the selection method of resource recycling channels is shown. As shown in Figure 1 The computer terminal 10 (or mobile device) can include one or more processors 102 (the processor 102 can include but not limited to a microprocessor MCU or a programmable logic device FPGA processing device, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it can also include a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which can be included as one of the ports in the BUS bus), a network interface, a power supply and / or a camera. Those skilled in the art can understand that Figure 1 The structure shown is only schematic, which does not limit the structure of the above-mentioned electronic device. For example, the computer terminal 10 can include more or less components than those shown in Figure 1 or have a different configuration than that shown in Figure 1 .

[0027] It should be noted that the one or more processors 102 and / or other data processing circuits described above can be referred to herein as "data processing circuits" in general. The data processing circuit can be embodied in whole or in part as software, hardware, firmware or any other combination. In addition, the data processing circuit can be a single independent processing module, or any one of the other elements combined into the computer terminal 10 (or mobile device) in whole or in part. As referred to in the embodiments of the present application, the data processing circuit is a processor control (for example, the selection of the variable resistance terminal path connected to the interface).

[0028] The memory 104 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the selection method of resource recycling channels in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, implements the above-mentioned selection method of resource recycling channels. The memory 104 can include a high-speed random access memory, and can also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 can further include a memory remotely disposed with respect to the processor 102, which can be connected to the computer terminal 10 through a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0029] The transmission device 106 is configured to receive or send data via a network. The network can include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network interface controller (NIC) that can be connected to other network devices through a base station to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module configured to communicate with the Internet wirelessly.

[0030] The display can be a liquid crystal display (LCD) that is touch screen, for example, which can enable a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0031] In the above operating environment, the present application provides a method for selecting a resource recycling channel as shown in Figure 2 Figure 2 is a flowchart of the method for selecting a resource recycling channel according to Embodiment 1 of the present application.

[0032] In step S101, a transaction node graph is generated according to transaction information of a target object, wherein the transaction node graph is configured to indicate a transaction relationship between the target object and a candidate object.

[0033] It should be noted that the target object can refer to a customer permitted by a third institution, and the third institution will evaluate the target object to determine whether the target object needs to be recycled and through which channel to recycle the resource. The candidate object can refer to a merchant having a transaction relationship with the target object, and the merchant can be any party involved in the past consumption behavior of the target object. The transaction information can include consumption records of the target object in each merchant, such as transaction time, amount, location, and transaction frequency.

[0034] The transaction node graph can be a kind of graph data structure, which can be a bipartite graph. The transaction node graph is configured to indicate a transaction relationship between the target object (customer) and the candidate object (merchant). In the graph, the nodes represent the customer and the merchant, and the edges represent the transaction activities between the two, and the weight of the edge can be an index of the transaction intensity such as the transaction amount and the transaction frequency.

[0035] Figure 3 is a schematic diagram of the method for selecting a resource recycling channel according to the present application; and Figure 3 ​As shown, both the target object 302 and the reference object 306 can be customers licensed by a third-party institution, and the candidate object 304 can be an object having a transaction relationship with the target object 302 or the reference object 306. The objects having the transaction relationship are connected by lines between the nodes corresponding to the objects, and the connection lines (edges) can indicate the transaction relationship between the objects. Moreover, the customer nodes corresponding to the target object 302 and the reference object 306 and the merchant node corresponding to the candidate object 304 are two different subsets. Each node has a corresponding feature, which will be described in detail below, and will not be described here again.

[0036] In an optional embodiment, the system collects historical transaction data of the target object, analyzes the transaction data to construct a transaction node graph. The transaction node graph not only reflects the transaction behavior between the target object and the candidate object, but also contains the intensity information of the transaction, such as the transaction amount and frequency. By constructing the transaction node graph, it can be directly seen which candidate objects have transaction with the target object and the degree of the transaction, which is crucial for subsequent analysis of customer behavior and potential channel optimization.

[0037] In an optional example, it is assumed that the following data is available:

[0038] Customer A, age 30, working type is engineer, annual income is 80,000 yuan.

[0039] Merchant B belongs to the electronic product retail industry, has good business conditions, and the annual turnover is 500,000 yuan.

[0040] Customer A has consumed at merchant B for 5 times in the past 6 months, with a total amount of 20,000 yuan.

[0041] Customer A and merchant B can be regarded as two different types of nodes in the transaction node graph, namely customer nodes and merchant nodes. A edge is generated between customer A and merchant B, and the weight of the edge can be set to the total consumption amount or the number of times of customer A at merchant B, i.e. 20,000 yuan or 5 times, to represent the transaction intensity. The detailed feature information of customer A and merchant B is expanded into a vector and stored in the corresponding node. For example, the feature vector of customer A can be [30, engineer, 80000], and the feature vector of merchant B can be [electronic product retail, good, 500000]. These vectors can be further encoded into numerical form for subsequent processing. In the transaction node graph obtained finally, the transaction relationship between customer A and merchant B and their respective attribute information are contained.

[0042] In step S102, a first feature of the target object is determined based on the transaction node graph, wherein the first feature is used to indicate information of the target object.

[0043] It should be noted that the first feature can refer to the key attributes and information extracted from the transaction node graph for describing the target object (i.e., the customer). These features can include the customer's basic information, historical transaction records, transaction frequency with different merchants, transaction amount distribution, etc., which together constitute the comprehensive portrait of the customer in terms of transactions.

[0044] In optional embodiments, the customer transaction behavior patterns and correlations contained in the transaction node graph can be deeply mined using a graph convolutional neural network or other suitable algorithms for processing graph-structured data. Through multi-layer feature extraction and transformation, a first feature vector that integrates multiple dimensions such as customer personal attributes, transaction habits, and social network influence can be generated. This feature vector not only contains the customer's static information but also reflects their dynamic behavior and status in the transaction network, thereby providing data support for personalized strategy formulation.

[0045] Step S103, calculating a matching parameter corresponding to each of the at least one recycling channel based on the first feature and the second feature corresponding to each of the at least one recycling channel;

[0046] It should be noted that the at least one recycling channel to be selected can refer to multiple channels available for selection in the resource recycling process, such as phone notification, SMS notification, email reminder, and home visit, etc., and it is necessary to determine which one or which ones are most suitable for the target object. The second feature can be a feature related to the recycling channel, which can include the cost of the channel, the historical resource recycling success rate, and the customer group preference covered by the channel, etc. The matching parameter can be a reflection of the mutual matching degree between the customer features and the recycling channel features, and is an important indicator for evaluating whether the channel is effective for a particular customer.

[0047] In optional embodiments, based on the feature vectors of the customers and the channels, the model can automatically identify the most effective way for each customer by calculating the similarity or compatibility between the vectors. The calculation of the matching parameter is not just a simple numerical comparison, but also includes complex model operations and summaries of past experiences, aiming to achieve personalized and efficient strategies.

[0048] Step S104, in the case where the matching parameter is greater than a preset value, determining the recycling channel corresponding to the largest matching parameter as the target recycling channel.

[0049] It should be noted that the preset value is a set threshold value for screening customers who are worth taking resource recycling actions. Only when the calculated matching parameter is greater than this preset value, the target object is considered to be recycled resources, and the best one from all possible channels is further selected.

[0050] The target recovery channel can be, among all available recovery channels, the one that is considered the most likely to succeed and the most cost-effective for a particular customer based on the model's calculation results.

[0051] In an optional embodiment, after the model matches a customer with a recovery channel, if there is at least one channel whose matching parameter exceeds a pre-set threshold, the model will select those channels with the highest scores as the preferred resource recovery method for this customer. Only when the matching parameter is high enough to indicate that a particular channel is very effective for the customer, will this channel be selected for resource recovery.

[0052] In an optional example, assume there is a group of customers, among which the transaction relationship graph of customer A and the channel information graph have been processed and characterized by the model. When selecting a channel for customer A, the model obtains the following matching parameters for the channels: phone notification: 0.45; letter notification: 0.72; SMS notification: 0.61; lawyer letter notification: 0.83.

[0053] Here, the pre-set value is set to 0.5. Therefore, phone notification (0.45) is not considered an effective way because it does not exceed the pre-set value. The matching parameters of the remaining three channels (letter notification, SMS notification, and lawyer letter notification) all exceed 0.5, and they are considered effective ways. Among the three effective channels, the matching parameter of the lawyer letter notification is the highest, reaching 0.83. The lawyer letter notification can be selected as the target recovery channel for customer A because it has the highest score among all channels, indicating that it is expected to be the best for customer A, and it also passes the cost-effectiveness consideration.

[0054] In an optional embodiment, determining the first feature of the target object based on the transaction node graph comprises: determining a target feature of the target object and a candidate feature corresponding to each of at least one candidate object; performing mean value calculation on the candidate feature corresponding to each candidate object having an adjacency relationship with the target object to obtain a candidate feature mean value; performing feature fusion on the candidate feature mean value and the target feature to obtain a fusion feature, and multiplying the fusion feature by a first weight matrix to obtain a weighted feature, wherein the first weight matrix is used to change the dimension of the feature; and determining a ratio between the weighted feature and a length of the weighted feature as the first feature.

[0055] It should be noted that the target feature can represent specific attributes of the target object, which can include personal basic information, consumption habits, repayment records, etc. The candidate feature can be the attribute information of the candidate object having a transaction relationship with the target object, which can cover merchant types, geographic locations, customer evaluations, etc., reflecting the characteristics of the target object's interactive objects in the transaction network.

[0056] Mean calculation refers to the sum of a set of values ​​and then dividing by the number of values. It is often used for data simplification or feature fusion to obtain a generalized value. Feature fusion can be a step in machine learning that involves combining multiple features into a single composite feature, often to enhance the expressiveness of the model or reduce feature redundancy. The first weight matrix is ​​used to adjust the dimension of the feature vector or the weight matrix. Through matrix multiplication operations, the combination or scale of input features can be changed to adapt to the subsequent processing requirements of the model.

[0057] In an optional embodiment, the basic features of the target object and candidate objects are extracted from the transaction node graph. Then, the features of all candidate objects that have traded with the target object (adjacency relationship) are averaged to obtain a summarized candidate feature mean. This step takes into account the impact of environmental factors on the target object. Next, the target feature is fused with the candidate feature mean to construct a comprehensive feature that reflects the overall information of the target object and its environment. This comprehensive feature is then multiplied by the first weight matrix to adjust the dimension or importance of the feature to obtain a weighted feature. Finally, the standardized first feature is obtained by calculating the ratio of the weighted feature to its own length, ensuring that the size of the feature vector is not affected by the absolute value, so that the model can make consistent decisions across inputs of different scales.

[0058] Considering the interaction between the target object and its environment, it is necessary to analyze the features of all candidate objects that have transacted with it and calculate the mean of these features to reflect the overall properties of its transaction network. The target features are fused with the mean of the candidate features to form a more comprehensive feature description. This step helps capture changes in the target object due to the influence of the environment. The fused features are then adjusted using a first weight matrix, perhaps by reducing their dimensionality or assigning different weights to different attributes, so that the model can more effectively understand and utilize these features.

[0059] In an optional embodiment, after determining the ratio between the fused feature and the length of the fused feature as the first feature, it includes: determining the i-th candidate feature corresponding to at least one candidate object based on the reference feature of the reference object in the transaction node graph, wherein the reference object is an object that has a transaction relationship with at least one candidate object, and i is an integer greater than 1; calculating the mean of the i-th candidate feature corresponding to each candidate object that has an adjacency relationship with the target object to obtain the i-th candidate feature mean; fusing the i-1-th first feature with the i-th candidate feature mean to obtain the i-th fused feature, and multiplying the i-th fused feature with the first weight matrix to obtain the i-th weighted feature; determining the ratio between the i-th weighted feature and the length of the i-th weighted feature as the i-th first feature; when i is equal to the preset number of rounds, determining the i-th first feature as the first feature of the target object.

[0060] It should be noted that the reference features can be features extracted from objects that have a transaction relationship with the candidate object, which reflect the interaction mode and environmental influence of the target object in the transaction network. In graph data, the existence of an edge between two nodes is called an adjacency relationship, meaning they are directly connected or influence each other.

[0061] In an optional implementation, in a multi-layer graph convolutional neural network, each layer performs feature fusion and update according to the attributes of the target object and the attributes of its adjacent objects in the transaction network, where i is the number of layers. First, the model determines the candidate features based on the reference features of the reference objects in the transaction network. Then, the average value of the features of the adjacent candidate objects is calculated, which is used as the basis for the fusion of the features of the target object in the previous layer (the first feature of the i-1th layer) to obtain the i-th fusion feature. Next, the fusion feature is multiplied by the weight matrix to obtain the weighted feature, and finally the length ratio of the weighted feature is calculated and used as the first feature of the i-th layer. When the preset number of rounds is reached, the first feature value calculated in the last round becomes the final first feature of the target object.

[0062] Under the deep learning framework, especially in the graph convolutional neural network for processing graph data, the construction of the representation layer of the customer transaction relationship graph is a multi-step process aimed at gradually enhancing the feature representation of each customer node to better reflect its status and behavior in the transaction network. The model first extracts candidate features from the direct neighbor nodes of the target object, and then calculates the average value of these features to form the candidate feature mean. In each round of feature extraction, the model fuses the features of the target object in the previous layer with the candidate feature mean of the current round to generate a fusion feature, which is multiplied by the first weight matrix to obtain a weighted feature. By calculating the length ratio of the weighted feature, a standardized first feature value can be obtained, which gradually converges to the final first feature reflecting the overall transaction behavior and attributes of the target object after multiple iterations. This method can capture and utilize deep-level association information in the transaction network to provide more rich inputs for the reinforcement learning model to optimize the resource recovery strategy.

[0063] In an optional implementation, the extraction of the first feature of the target object can use a multi-layer graph convolutional neural network, and each layer of the graph convolutional neural network can be composed of an input layer, a hidden layer, and an output layer. In each layer, the following operations can be performed:

[0064] First, an aggregation operation is performed:

[0065]

[0066] wherein, The candidate feature u outputted in the k-1 layer graph convolutional neural network can be represented as follows: u is a candidate object belonging to N(v), and N(v) can be a candidate object having a transaction relationship with the target object. AVG can represent an average value operation on a plurality of candidate features. Finally, the above formula can obtain the candidate feature mean value of the current layer.

[0067] It should be noted that the candidate feature in each layer can be obtained by using the graph convolutional neural network alone, and the feature of the reference object can be added to the candidate feature to comprehensively consider the features of the target object, the reference object and the candidate object.

[0068] Then, the candidate feature mean value can be subjected to a pooling dimension reduction operation, which can be as follows:

[0069]

[0070] CONCAT can be a concatenation operation on the elements therein to obtain a fusion feature, W k is a first weight matrix, multiplied by the fusion feature to obtain a weighted feature.

[0071] Then, the weighted feature can be subjected to a regularization operation:

[0072]

[0073] where ||·|| represents the two-norm of a vector. The first feature corresponding to the kth layer is obtained through the above calculation.

[0074] Through the above embodiments of the present application, each iteration of the graph convolutional neural network layer is deepening and refining the feature representation of the target object, and the finally determined first feature will more comprehensively reflect the behavior pattern and condition of the target object in the transaction network, providing a solid data foundation for subsequent strategy optimization.

[0075] In the optional embodiment, the matching parameter corresponding to each of the at least one recycling channel is calculated based on the first feature and the second feature corresponding to each of the at least one recycling channel to be selected, including: traversing the M recycling channels to be selected, and performing the following operations on the nth recycling channel: performing feature fusion on the nth second feature corresponding to the nth recycling channel and the first feature to obtain an nth matching feature, wherein M is an integer greater than 0, and n is an integer greater than 0 and less than or equal to M; multiplying the nth matching feature by a second weight matrix and adding a first bias to obtain an nth matching value; activating the nth matching value to obtain an nth matching parameter; and in the case where n is equal to M, determining the matching parameter corresponding to each of the M recycling channels.

[0076] It should be noted that the M recycling channels to be selected can refer to multiple recycling methods or ways available for selection, such as telephone notification, short message reminder, email notification, etc., and M represents the total number of these recycling channels, and M is a positive integer.

[0077] The matching feature can be the result of feature fusion, which is a vector representing the potential matching degree or relevance of the customer to a specific recycling channel. The second weight matrix can be used in the conversion process of matching features to matching values in the neural network model, and this matrix contains the weight parameters of the model, which are used to adjust the importance or influence of different features. The first bias can be a value added after the second weight matrix is multiplied in the neural network model, which is used to adjust the output baseline of the model to ensure the flexibility and adaptability of the model.

[0078] The matching value can be an intermediate result obtained by multiplying the matching feature by the second weight matrix and adding the first bias, which reflects the matching degree of the customer to the specific recycling channel under preliminary evaluation. In the neural network, a nonlinear activation function (such as ReLU, sigmoid, etc.) is usually used to process the matching value, with the purpose of introducing a nonlinear relationship to enable the model to learn more complex matching patterns.

[0079] It should be noted that in order to quantify the expected effect of each recycling channel on the target object resource recycling, the system needs to perform a series of calculations. First, based on the first feature of the target customer and the second feature of each recycling channel, the system generates matching features through feature fusion, which is equivalent to concentrating the relevant information between the customer and the channel into a vector. Then, the system uses the pre-trained second weight matrix and the first bias to convert the matching features of each channel to generate matching values, which is similar to the model's preliminary evaluation of information. Then, the matching value is nonlinearly processed by the activation function to obtain the final matching parameter, which reflects the matching degree between the customer and the channel. Finally, when all M recycling channels are processed, the system will determine the respective matching parameters of each channel, providing a basis for subsequent decision-making, i.e., selecting those channels with higher matching parameters to recycle resources from the customer, in order to achieve higher effectiveness and lower cost.

[0080] In an optional implementation, for a determined recycling channel, the matching parameter between it and the target object can be calculated one by one. The first feature and the second feature can be input into a fully connected layer, and in the fully connected layer, the first feature and the second feature can be fused by the cascade operation described above to obtain the matching feature.

[0081] After obtaining the matching features, complex transformations can be performed on the matching features to learn the optimal matching relationship between the target object and the recycling channel. A plurality of fully connected layers can be set, and for each fully connected layer, the following can be performed:

[0082]

[0083] wherein, and are the input and output of the current fully connected layer respectively, the matrix W k and the vector represent the second weight matrix and the first bias of the layer respectively, the second weight matrix and the first bias in each layer can be different, and the setting of the specific matrix and bias will be described in the training process below, and thus will not be described here. The input of each layer can be the output of the previous layer, and for the first layer, the input can be the matching features described above.

[0084] After passing through a plurality of fully connected layers, the obtained matching value can be activated, and a sigmoid function can be used, for example:

[0085]

[0086] After the obtained matching value is activated by the function described above, the matching parameter corresponding to the current recycling channel can be obtained.

[0087] Through the above embodiments of the present application, the matching parameter is calculated from the feature information of the target object and the channel through feature fusion, weight matrix and activation function, which provides data support for subsequent selection. Through this quantitative method, more accurate and efficient decisions can be made, while reducing labor costs and improving resource recycling effectiveness.

[0088] In an optional embodiment, before the n-th second feature corresponding to the n-th recycling channel is fused with the first feature, the following is included: determining an n-th channel feature corresponding to the n-th recycling channel based on channel information of the n-th recycling channel; and multiplying the n-th channel feature with a third weight matrix and adding a second bias to obtain the n-th second feature.

[0089] In an optional embodiment, a preliminary channel feature vector (the nth channel feature) is first calculated based on the detailed information of the nth recycling channel, such as cost, historical success rate, etc. Then, the preliminary feature vector is multiplied by the third weight matrix, which aims to adjust the importance of different channel features, so that the model can more accurately evaluate the effectiveness of the channel. After the multiplication operation, the model is further added with the second bias, which can ensure that valuable information can be captured even in the case of low feature value, avoiding information loss. The final nth second feature is fused with the first feature of the customer, preparing for the subsequent matching parameter calculation.

[0090] In an optional embodiment, for each recycling channel, the model generates a channel feature vector based on the channel information, which captures the basic attributes and effectiveness indicators of the channel. Then, the vector is multiplied by the third weight matrix, which assigns different importance weights to different channel features according to the model training, to reflect the actual effectiveness of the channel in the resource recycling process. The result of the multiplication operation is added with the second bias, which can be regarded as an adjustment mechanism of the model to optimize the expression of the second feature, making it closer to the real needs of decision-making. The final second feature is fused with the first feature vector of the customer to form a feature set that integrates customer information and recycling channel information, providing comprehensive input information for subsequent matching parameter calculation and helping the model make more accurate strategy selection.

[0091] In an optional embodiment, the extraction of the second feature can be realized by using a single-layer fully connected layer, such as:

[0092]

[0093] wherein the vector may be the channel information of the cth recycling channel, W may be the third weight matrix, and the vector may be the second bias.

[0094] Through the above embodiments of the present application, by adjusting and optimizing the second feature of each recycling channel and combining the first feature of the target object, more accurate strategy decisions can be generated, the balance between cost and benefit can be achieved, and the efficiency and effectiveness of the entire resource recycling process can be improved.

[0095] In an optional embodiment, before determining the first feature of the target object based on the transaction node graph, the method comprises: traversing the Q sample objects, and determining a sample feature of a pth sample object based on a sample node graph of the pth sample object in the Q sample objects, where Q is an integer greater than 0, and p is an integer greater than 0 and less than or equal to Q; determining a pth sample recycling channel and a pth matching parameter of the pth sample object based on the pth sample feature and the to-be-trained first weight matrix, the to-be-trained second weight matrix, the to-be-trained first bias, the to-be-trained third weight matrix, and the to-be-trained second bias; determining a pth resource recycling value and a pth resource input value corresponding to the sample recycling channel of the pth sample object; calculating a pth first difference value between the pth resource recycling value and the pth resource input value, and a pth first summation value between the pth resource recycling value and the pth resource input value; determining a pth indication value as a ratio between a product of the pth matching parameter and the pth first difference value and the pth first summation value; updating the to-be-trained first weight matrix, the to-be-trained second weight matrix, the to-be-trained first bias, the to-be-trained third weight matrix, and the to-be-trained second bias based on the pth indication value; and determining the current to-be-trained first weight matrix, to-be-trained second weight matrix, to-be-trained first bias, to-be-trained third weight matrix, and to-be-trained second bias as the first weight matrix, the second weight matrix, the first bias, the third weight matrix, and the second bias when the sum of the Q indication values is minimum.

[0096] It should be noted that the sample object refers to a customer used for training the model, and the data of the customer is used to adjust the model parameters to optimize the strategy. The resource recycling value can refer to the actual recycling of resources through a certain channel, which is a key indicator for measuring the effect of resource recycling. The resource input value can be the cost required to implement a certain recycling channel, including manpower, material resources, financial resources, and other expenditures, which is used to evaluate the economic benefits of resource recycling. The first difference value is the gap between the resource recycling value and the resource input value, which reflects the net income of the recycling channel. The first summation value is the sum of the resource recycling value and the resource input value, which is an indicator for comprehensively considering the benefits and costs.

[0097] Figure 4 is a schematic diagram of another method for selecting a resource recycling channel according to an embodiment of the present application; the above method can be integrated in a model.

[0098] In an optional embodiment, a graph convolutional neural network (GCN) can be used to process the customer's relationship graph (transaction node graph), which can capture complex relationship patterns from the graph, extract features from the node's characteristics (such as customer attributes, merchant attributes) and edge information (such as transaction frequency, transaction amount) through multi-layer convolution operations. Each layer of GCN updates the node's feature vector, and the final output is a high-order feature vector that can reflect the customer's position and influence in the transaction network.

[0099] Specifically, the input layer receives the attribute features of the customer node and the features of the connected merchant nodes, which are usually in the form of pre-processed vectors. The hidden layer performs multi-layer graph convolution operations, and each layer of graph convolution updates the current node's feature vector based on the features of the neighbor nodes, which helps the model learn the relationship and interaction between nodes. In addition, each layer of graph convolution may also include a nonlinear activation function (such as ReLU) to introduce nonlinear relationships and enhance the model's representation ability. The output layer produces the final representation features of the customer transaction relationship graph (transaction node graph), which will be used for subsequent decision-making and channel selection.

[0100] For recycling channel information feature extraction, the cost and effect information of recycling channels can be processed through a fully connected layer to convert these information into vector form, facilitating the fusion and comparison with customer features. The input layer can receive the attribute information of each recycling channel, usually including cost and effect quantitative indicators. The fully connected layer can process the input data through linear transformation and nonlinear activation function to generate the representation features of the recycling channel. The weight matrix and bias vector of the fully connected layer are trainable model parameters, and through optimizing these parameters, the model can learn the importance of different channel features.

[0101] Then is the fully connected layer for feature fusion and decision learning, its input layer receives the representation features of the customer transaction relationship graph (first feature) and the representation features of the recycling channel (second feature). The hidden layer is stacked by multiple fully connected layers, which is responsible for complex transformation of input features and learning the best matching pattern between customers and recycling channels. Each fully connected layer has its own weight matrix and bias vector, and through training, these parameters will be optimized to better fuse customer and channel features and predict recycling effect.

[0102] At the input end of the fully connected layer, the model will concatenate the customer features and channel features to form a comprehensive feature vector, which contains the customer's historical behavior and the attribute information of the recycling channel, providing a comprehensive decision basis for the model. Through the linear transformation of the weight matrix to the concatenated feature vector, plus the bias vector, an adjusted feature representation is obtained, which can capture the interaction between different features. Then use the activation function such as ReLU to increase the non-linear expression ability of the model, which helps the model to learn more complex decision boundaries.

[0103] Finally, by combining the output of the fully connected layer with a sigmoid function or similar activation function, the output corresponds to the matching parameters of each recycling channel. These parameters reflect the model's belief in the possibility and effectiveness of a particular channel for resource recycling for a particular customer.

[0104] It should be noted that the above is a description of the specific structure of the model, and the model can be trained before use.

[0105] In an optional implementation, the model training process can be as follows:

[0106] Step S201, data preparation and preprocessing;

[0107] The transaction records between the fund resource users (sample objects) collected from the database of the third-party institution and different merchants include transaction amount, time, location, etc. The collected transaction data is converted into the form of a two-part attribute graph, with the left side representing the sample objects and the right side representing the merchants, and the edges representing the transactions between the sample objects and the merchants, while each node carries its attribute features. Organize the relevant information of various recycling channels (such as phone, SMS, email, etc.), including channel cost, historical effect, etc., to build a channel information graph. Clean the data to remove outliers and missing values, and then standardize the attribute values to the same range to eliminate the scale difference between features, which is beneficial to model training.

[0108] Step S202, feature extraction and representation;

[0109] Use the graph convolutional neural network (GCN) to extract features from the customer nodes in the two-part attribute graph, which can learn the customer's behavior patterns and feature vectors related to resource recycling from the customer-merchant transaction relationship. Use the fully connected layer to process the data in the channel information graph, converting the attribute information of the channel into a vector form for comparison and fusion with customer features.

[0110] Step S203, model construction and initialization;

[0111] Design the model architecture, including graph convolutional layers, fully connected layers, and output layers. Ensure that the model can accept customer and channel representations as input and output matching parameters. Initialize model parameters, such as the weight matrix and bias vector, typically by random initialization to begin the training process. Set training hyperparameters, such as the learning rate, number of training epochs, and batch size, as these parameters affect the model's learning speed and convergence.

[0112] Step S204: model training and parameter optimization;

[0113] For each sample object, repeat the following steps: fuse the customer features of the sample object and the channel features of each candidate channel into matching features through a cascade operation. Multiply the fused matching features with the second weight matrix, and add the first bias to obtain the matching value. Use an activation function (such as sigmoid) to transform the matching value to obtain the final matching parameter, which reflects the fit between the channel and the customer. Record the results of the p-th sample object through the n-th recovery channel, including the resource recovery value (actual amount of debt recovered) and the resource input value (cost). Calculate the indicator value based on the resource recovery value, resource input value and matching parameters. The indicator value is a comprehensive evaluation indicator that reflects the gap between the resource recovery channel selection predicted by the model and the actual recovery effect.

[0114] Using the Adam optimization algorithm, the model's training parameters, including the first weight matrix, second weight matrix, first bias matrix, third weight matrix, and second bias matrix, are updated based on the calculated indicator values. The goal is to minimize the sum of the indicator values ​​of all sample objects through parameter adjustment, thereby optimizing the overall performance of the entire decision model.

[0115] Step S205, verification and evaluation;

[0116] Cross-validation techniques were used to assess the model's generalization and stability to avoid overfitting. Key metrics such as the model's prediction accuracy and recall were calculated to assess the model's performance in predicting channel effectiveness and cost-effectiveness.

[0117] Step S206: Model saving and deployment.

[0118] After training is complete, the model parameters are saved for subsequent use and tuning. The trained model is then deployed to the production environment for use by third-party organizations in actual resource recycling decisions.

[0119] Through the above embodiments of the present application, the key steps in the model training process can be seen, including data preparation, feature extraction, preliminary prediction, effect evaluation, indicator value calculation and parameter updating. This whole process ensures that the model can continuously optimize its strategy based on the historical data of resource recovery, so as to achieve the purpose of effectively recovering arrears and controlling costs in future resource recovery.

[0120] In the embodiment of the present application, a transaction node graph is generated according to the transaction information of the target object, wherein the transaction node graph is used to indicate the transaction relationship between the target object and the candidate object; a first feature of the target object is determined based on the transaction node graph, wherein the first feature is used to indicate the information of the target object; a matching parameter corresponding to each of the at least one recovery channel is calculated based on the second feature corresponding to each of the at least one recovery channel and the first feature; in the case where there is a matching parameter greater than a preset value, the recovery channel corresponding to the largest matching parameter is determined as the target recovery channel. After the target object is determined, the transaction node graph from which the transaction feature of the target object can be extracted can be determined based on the transaction information of the target object, the first feature of the target object can be extracted through the transaction node graph, and the first feature can indicate the transaction between the target object and the candidate object. The second feature of the recovery channel to be selected can be extracted, and the matching parameter between the target object and the recovery channel can be calculated based on the first feature and the second feature to determine which channel is suitable for the target object. Finally, whether to recover the resource of the target object is determined according to the size of the matching parameter, and in the case of recovery, which recovery channel is used for the target object is determined. Thus, the problem of low efficiency in the related art that the method of determining the recovery channel of the fund resource exists is solved.

[0121] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0122] Embodiment 2

[0123] The embodiment of the present application also provides a resource recovery channel selection device. It should be noted that the resource recovery channel selection device of the embodiment of the present application can be used to execute the resource recovery channel selection method provided by the embodiment of the present application. The resource recovery channel selection device provided by the embodiment of the present application is introduced as follows.

[0124] According to the embodiment of the present application, a device for implementing the above-mentioned resource recovery channel selection method is also provided, as shown in Figure 5 The device comprises:

[0125] The node graph generation unit 502 is configured to generate a transaction node graph according to the transaction information of the target object, where the transaction node graph is used to indicate the transaction relationship between the target object and the candidate objects.

[0126] The feature extraction unit 504 is configured to determine a first feature of the target object based on the transaction node graph, where the first feature is used to indicate the information of the target object.

[0127] The matching parameter calculation unit 506 is configured to calculate a matching parameter corresponding to each of the at least one recycling channel based on the first feature and the second feature corresponding to each of the at least one recycling channel.

[0128] The channel determination unit 508 is configured to determine the recycling channel corresponding to the maximum matching parameter as the target recycling channel when the matching parameter is greater than a preset value.

[0129] The resource recycling channel selection device provided by the embodiment of the present application determines a transaction data set according to an enterprise number, where the transaction data set is used to indicate the information of transactions corresponding to the enterprise number; extracts the transaction data set to obtain a reference transaction feature set, and determines a target transaction feature set from the reference transaction feature set based on the enterprise number, where the transaction feature is used to indicate the feature of the transaction data set; inputs the target transaction feature set into an amount prediction model to obtain a compensation amount corresponding to the enterprise number, and determines a selection proportion of the resource recycling channel according to the compensation amount and the transaction data set, where the amount prediction model is matched with the enterprise number. For different enterprises, the transaction data of the enterprise can be determined according to the enterprise number, the transaction feature can be determined, and the target transaction feature corresponding to the enterprise number can be selected from the transaction feature, so that the compensation amount corresponding to the enterprise and the selection proportion of the resource recycling channel can be accurately calculated. Thus, the technical problem that the accuracy of the processing proportion is low due to manual determination of the processing proportion is solved.

[0130] Optionally, the feature extraction unit 504 includes: a first feature determination module configured to determine a target feature of the target object and candidate features corresponding to the at least one candidate object; a first mean value calculation module configured to calculate the mean value of the candidate features corresponding to each candidate object having an adjacent relationship with the target object to obtain a candidate feature mean value; a first feature fusion module configured to fuse the candidate feature mean value and the target feature to obtain a fusion feature, and multiply the fusion feature by a first weight matrix to obtain a weighted feature, where the first weight matrix is used to change the dimension of the feature; and a first regularization module configured to determine the ratio between the weighted feature and the length of the weighted feature as the first feature.

[0131] Optionally, the first feature determination module is further configured to determine an i-th candidate feature corresponding to each of the at least one candidate object based on a reference feature of a reference object in the transaction node graph, where the reference object is an object having a transaction relationship with the at least one candidate object, i is an integer greater than 1; the first mean value calculation module is further configured to calculate a mean value of the i-th candidate feature corresponding to each of the candidate objects having an adjacency relationship with the target object to obtain an i-th candidate feature mean value; the first feature fusion module is further configured to fuse the i-1-th first feature and the i-th candidate feature mean value to obtain an i-th fusion feature, and multiply the i-th fusion feature by a first weight matrix to obtain an i-th weighted feature; the first regularization module is further configured to determine a ratio between the i-th weighted feature and a length of the i-th weighted feature as the i-th first feature; and in a case where i is equal to a preset round number, the i-th first feature is determined as the first feature of the target object.

[0132] Optionally, the matching parameter calculation unit 506 includes a first traversal module configured to traverse the M recycling channels to be selected, and perform the following operations on the n-th recycling channel: fuse an n-th second feature corresponding to the n-th recycling channel with the first feature to obtain an n-th matching feature, where M is an integer greater than 0, and n is an integer greater than 0 and less than or equal to M; multiply the n-th matching feature by a second weight matrix and add a first bias to obtain an n-th matching value; activate the n-th matching value to obtain an n-th matching parameter; and in a case where n is equal to M, determine the matching parameters corresponding to the M recycling channels.

[0133] Optionally, the first traversal module is further configured to determine an n-th channel feature corresponding to the n-th recycling channel based on channel information of the n-th recycling channel; and multiply the n-th channel feature by a third weight matrix and add a second bias to obtain the n-th second feature.

[0134] Optionally, the above-mentioned resource recovery channel selection device also includes: a model parameter training unit, which is used to traverse the Q sample objects and determine the sample feature of the p-th sample object based on the sample node graph of the p-th sample object in the Q sample objects, wherein Q is an integer greater than 0, and p is an integer greater than 0 and less than or equal to Q; determine the p-th sample recovery channel and the p-th matching parameter of the p-th sample object based on the p-th sample feature and the first weight matrix to be trained, the second weight matrix to be trained, the first bias to be trained, the third weight matrix to be trained, and the second bias to be trained; determine the p-th resource recovery value and the p-th resource input value corresponding to the sample recovery channel of the p-th sample object; calculate the p-th resource recovery value and the p-th resource input value. the pth first difference between the source input values ​​and the pth first summation value between the pth resource recovery value and the pth resource input value; the ratio of the product between the pth matching parameter and the pth first difference to the pth first summation value is determined as the pth indication value; based on the pth indication value, the first weight matrix to be trained, the second weight matrix to be trained, the first bias to be trained, the third weight matrix to be trained, and the second bias to be trained are updated; when the sum of the Q indication values ​​is the smallest, the current first weight matrix to be trained, the second weight matrix to be trained, the first bias to be trained, the third weight matrix to be trained, and the second bias to be trained are determined as the first weight matrix, the second weight matrix, the first bias, the third weight matrix, and the second bias.

[0135] It should be noted that the node graph generation unit 502 to the channel determination unit 508 correspond to steps S101 to S104 in Example 1. The examples and application scenarios implemented by the two modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Example 1. It should be noted that the above-mentioned modules or units can be hardware components or software components stored in a memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above-mentioned modules can also be run as part of the device in the computer terminal 10 provided in Example 1.

[0136] Example 3

[0137] An embodiment of the present application may provide an electronic device, Figure 6 This is a structural block diagram of an electronic device according to an embodiment of the present application. Figure 6 As shown, the electronic device may include: one or more ( Figure 6 Only one is shown) processor 1002, memory 1004, storage controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0138] The memory can be used to store software programs and modules, such as program instructions / modules corresponding to the method and device in the embodiments of the present application. The processor executes various functions and data processing by running the software programs and modules stored in the memory, that is, implements the above method. The memory can include a high-speed random access memory, and can also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some examples, the memory can further include a memory remotely arranged with respect to the processor, which can be connected to the terminal through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0139] The processor can call information and applications stored in the memory through the transmission device to perform the following steps:

[0140] S1, generating a transaction node graph according to transaction information of a target object, wherein the transaction node graph is used to indicate a transaction relationship between the target object and a candidate object;

[0141] S2, determining a first feature of the target object based on the transaction node graph, wherein the first feature is used to indicate information of the target object;

[0142] S3, calculating a matching parameter corresponding to each of the at least one recycling channel based on a second feature corresponding to each of the at least one recycling channel and the first feature;

[0143] S4, in the case where the matching parameter is greater than a preset value, determining a recycling channel corresponding to the largest matching parameter as a target recycling channel.

[0144] The processor can also call information and applications stored in the memory through the transmission device to perform the following steps: determining a target feature of the target object and a candidate feature corresponding to each of the at least one candidate object; performing mean value calculation on the candidate feature corresponding to each candidate object having an adjacency relationship with the target object to obtain a candidate feature mean value; performing feature fusion on the candidate feature mean value and the target feature to obtain a fusion feature, and multiplying the fusion feature by a first weight matrix to obtain a weighted feature, wherein the first weight matrix is used to change the dimension of the feature; determining a ratio between the weighted feature and a length of the weighted feature as the first feature.

[0145] The processor can further call information and application programs stored in the memory through the transmission device to perform the following steps: determining the i-th candidate feature corresponding to each of the at least one candidate object based on the reference feature of the reference object in the transaction node graph, wherein the reference object is an object having a transaction relationship with the at least one candidate object, and i is an integer greater than 1; performing mean value calculation on the i-th candidate feature corresponding to each candidate object having an adjacent relationship with the target object to obtain an i-th candidate feature mean value; fusing the i-1st first feature and the i-th candidate feature mean value to obtain an i-th fusion feature, and multiplying the i-th fusion feature and the first weight matrix to obtain an i-th weighted feature; determining the ratio between the i-th weighted feature and the length of the i-th weighted feature as the i-th first feature; and in the case that i is equal to a preset number of rounds, determining the i-th first feature as the first feature of the target object.

[0146] The processor can further call information and application programs stored in the memory through the transmission device to perform the following steps: traversing the M recycling channels to be selected, and performing the following operations on the n-th recycling channel: fusing the n-th second feature corresponding to the n-th recycling channel and the first feature to obtain an n-th matching feature, wherein M is an integer greater than 0, and n is an integer greater than 0 and less than or equal to M; multiplying the n-th matching feature by the second weight matrix and adding the first bias to obtain an n-th matching value; activating the n-th matching value to obtain an n-th matching parameter; and in the case that n is equal to M, determining the matching parameters corresponding to the M recycling channels respectively.

[0147] The processor can further call information and application programs stored in the memory through the transmission device to perform the following steps: determining the n-th channel feature corresponding to the n-th recycling channel based on the channel information of the n-th recycling channel; multiplying the n-th channel feature by the third weight matrix and adding the second bias to obtain the n-th second feature.

[0148] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: traversing the Q sample objects, and determining a sample feature of a p th sample object in the Q sample objects based on a sample node graph of the p th sample object, where Q is an integer greater than 0, and p is an integer greater than 0 and less than or equal to Q; determining a p th sample recycling channel and a p th matching parameter of the p th sample object based on the p th sample feature and a to-be-trained first weight matrix, a to-be-trained second weight matrix, a to-be-trained first bias, a to-be-trained third weight matrix, and a to-be-trained second bias; determining a p th resource recycling value and a p th resource input value corresponding to a sample recycling channel of the p th sample object; calculating a p th first difference value between the p th resource recycling value and the p th resource input value and a p th first summation value between the p th resource recycling value and the p th resource input value; determining a p th indication value as a ratio between a product of the p th matching parameter and the p th first difference value and the p th first summation value; updating the to-be-trained first weight matrix, the to-be-trained second weight matrix, the to-be-trained first bias, the to-be-trained third weight matrix, and the to-be-trained second bias based on the p th indication value; and determining the current to-be-trained first weight matrix, the to-be-trained second weight matrix, the to-be-trained first bias, the to-be-trained third weight matrix, and the to-be-trained second bias as the first weight matrix, the second weight matrix, the first bias, the third weight matrix, and the second bias when the sum of the Q indication values is minimum.

[0149] In the embodiments of the present application, a transaction node graph is generated according to transaction information of a target object, where the transaction node graph is used to indicate a transaction relationship between the target object and a candidate object; a first feature of the target object is determined based on the transaction node graph, where the first feature is used to indicate information of the target object; a matching parameter corresponding to each of the at least one recycling channel is calculated based on the first feature and a second feature corresponding to each of the at least one recycling channel; and in a case where there is a matching parameter greater than a preset value, a recycling channel corresponding to the greatest matching parameter is determined as a target recycling channel. After the target object is determined, a transaction node graph from which a transaction feature of the target object can be extracted can be determined based on the transaction information of the target object, the first feature of the target object can be extracted through the transaction node graph, and the first feature can indicate a transaction between the target object and the candidate object. The second feature of the recycling channel to be selected can be extracted, and the matching parameter between the target object and the recycling channel can be calculated based on the first feature and the second feature to determine which channel is suitable for the target object. Finally, whether to recycle the resource of the target object is determined according to the size of the matching parameter, and in a case where it is determined to recycle the resource, which recycling channel is used for the target object is determined. Thus, the problem that the method for determining the recycling channel of the fund resource in the related art has low efficiency is solved.

[0150] Those skilled in the art can understand that, Figure 6The structure shown is only schematic, and the electronic device can also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a palm computer, a Mobile Internet Device (MID), a PAD, or the like. Figure 6 This does not limit the structure of the electronic device described above. For example, the electronic device can further include more or fewer components (such as a network interface, a display device, etc.) than those shown in the embodiment, or have a different configuration from that shown in the embodiment. Figure 6 Figure 6

[0151] Those of ordinary skill in the art can understand that all or part of the steps of the various methods in the above embodiments can be completed by a program instructing a terminal device-related hardware, and the program can be stored in a computer-readable storage medium, which can include a flash disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk, etc.

[0152] Embodiment 4

[0153] The embodiments of the present application also provide a storage medium. Optionally, in the embodiment, the storage medium can be used to store program codes executed by the resource recycling channel selection method provided in Embodiment 1.

[0154] Optionally, in the embodiment, the storage medium can be located in any one of a group of computer terminals in a computer network, or in any one of a group of mobile terminals.

[0155] The present application also provides a computer program product adapted to execute the resource recycling channel selection method when executed on a data processing device.

[0156] The above-mentioned serial numbers of the embodiments of the present application are only for description, and do not represent the advantages or disadvantages of the embodiments.

[0157] In the above embodiments of the present application, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0158] ​​In several embodiments provided in the present application, it should be understood that the disclosed technology can be implemented by other ways. Among them, the above-described device embodiments are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed units can be indirect coupling or communication connection through some interfaces, units or modules, and can be electrical or other forms.

[0159] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or they can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0160] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0161] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part of the prior art that contributes to the technical solutions or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0162] The above is only the preferred embodiment of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should be considered as the protection scope of the present application.

Claims

1. A method for selecting a resource recycling channel, characterized in that: include: Generating a transaction node graph based on the transaction information of the target object, wherein the transaction node graph is used to indicate the transaction relationship between the target object and the candidate object; determining a first feature of the target object based on the transaction node graph, wherein the first feature is used to indicate information of the target object; Calculating a matching parameter corresponding to each of the at least one recycling channel to be selected based on the second feature corresponding to each of the at least one recycling channel and the first feature; In the case that there is a matching parameter greater than a preset value, the recycling channel corresponding to the largest matching parameter is determined as the target recycling channel.

2. The method according to claim 1, characterized in that The determining the first feature of the target object based on the transaction node graph includes: determining a target feature of the target object and a candidate feature corresponding to each of at least one of the candidate objects; Calculate the mean of the candidate features corresponding to each candidate object that has an adjacency relationship with the target object to obtain the candidate feature mean; Performing feature fusion on the candidate feature mean and the target feature to obtain a fused feature, and multiplying the fused feature by a first weight matrix to obtain a weighted feature, wherein the first weight matrix is ​​used to change the dimension of the feature; A ratio between the weighted feature and the length of the weighted feature is determined as the first feature.

3. The method according to claim 2, characterized in that After determining the ratio between the fusion feature and the length of the fusion feature as the first feature, the method further includes: Determining an i-th candidate feature corresponding to each of at least one of the candidate objects based on a reference feature of a reference object in the transaction node graph, wherein the reference object is an object having a transaction relationship with the at least one candidate object, and i is an integer greater than 1; Calculate the mean of the i-th candidate feature corresponding to each of the candidate objects that have an adjacent relationship with the target object to obtain the i-th candidate feature mean; Fusing the i-1th first feature with the mean of the i-th candidate feature to obtain the i-th fused feature, and multiplying the i-th fused feature with the first weight matrix to obtain the i-th weighted feature; Determine the ratio between the i-th weighted feature and the length of the i-th weighted feature as the i-th first feature; When i is equal to a preset round number, the i-th first feature is determined as the first feature of the target object.

4. The method according to claim 1, wherein The calculating, based on the second feature corresponding to each of the at least one recycling channel to be selected and the first feature, to obtain the matching parameter corresponding to each of the at least one recycling channel includes: Traverse the M recycling channels to be selected, and perform the following operations on the nth recycling channel: Fusing the nth second feature corresponding to the nth recycling channel with the first feature to obtain an nth matching feature, where M is an integer greater than 0 and n is an integer greater than 0 and less than or equal to M; Multiplying the nth matching feature by the second weight matrix and adding the resultant to the first bias to obtain an nth matching value; Activating the nth matching value to obtain the nth matching parameter; When n is equal to M, the matching parameters corresponding to the M recycling channels are determined.

5. The method according to claim 4, characterized in that Before fusing the nth second feature corresponding to the nth recycling channel with the first feature, the method includes: Determining an nth channel feature corresponding to the nth recycling channel based on the channel information of the nth recycling channel; After multiplying the nth channel feature by the third weight matrix, the resultant is added to the second bias to obtain the nth second feature.

6. The method according to any one of claims 1 to 5, characterized in that Before determining the first feature of the target object based on the transaction node graph, the method includes: Traversing Q sample objects, and determining a sample feature of a p-th sample object among the Q sample objects based on a sample node graph of the p-th sample object, where Q is an integer greater than 0, and p is an integer greater than 0 and less than or equal to Q; Determining a p-th sample recovery channel and a p-th matching parameter of the p-th sample object based on the p-th sample feature and a first weight matrix to be trained, a second weight matrix to be trained, a first bias to be trained, a third weight matrix to be trained, and a second bias to be trained; Determining a p-th resource recovery value and a p-th resource input value corresponding to the sample recovery channel of the p-th sample object; Calculating a p-th first difference between the p-th resource recovery value and the p-th resource input value and a p-th first summation between the p-th resource recovery value and the p-th resource input value; Determine a p-th indication value as a ratio of a product of the p-th matching parameter and the p-th first difference value to the p-th first summation value; updating the first weight matrix to be trained, the second weight matrix to be trained, the first bias to be trained, the third weight matrix to be trained, and the second bias to be trained based on the pth indication value; When the sum of the Q indicator values ​​is minimized, the current first weight matrix to be trained, the second weight matrix to be trained, the first bias to be trained, the third weight matrix to be trained, and the second bias to be trained are determined as the first weight matrix, the second weight matrix, the first bias, the third weight matrix, and the second bias.

7. A device for selecting a resource recycling channel, characterized in that: include: a node graph generating unit, configured to generate a transaction node graph based on the transaction information of the target object, wherein the transaction node graph is used to indicate the transaction relationship between the target object and the candidate object; a feature extraction unit, configured to determine a first feature of the target object based on the transaction node graph, wherein the first feature is used to indicate information of the target object; a matching parameter calculation unit, configured to calculate a matching parameter corresponding to each of the at least one recycling channel to be selected based on the second feature corresponding to each of the at least one recycling channel and the first feature; The channel determination unit is configured to, when there is a matching parameter greater than a preset value, determine a recycling channel corresponding to the largest matching parameter as a target recycling channel.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the resource recovery channel selection method according to any one of claims 1 to 6.

9. An electronic device, characterized in that: include: a memory storing an executable program; A processor, configured to run the program, wherein the program executes the method according to any one of claims 1 to 6 when running.

10. A computer program product comprising computer instructions, characterized in that When the computer instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images