Credit card collection method, device and equipment, medium and product

Abstract

The invention discloses a credit card collection method and device, equipment, a medium and a product, and relates to the technical field of artificial intelligence, and the method comprises the steps: determining a target user having an association relationship with a to-be-collected user, carrying out the feature splicing according to a target user feature vector of the target user and a to-be-collected user feature vector of the to-be-collected user, and obtaining a to-be-collected user feature vector of the to-be-collected user; determining a user feature optimization vector; determining a verbal skill vector optimization matrix according to the verbal skill vector matrix of the candidate collection verbal skill and the word vector of the candidate collection word, and determining a channel feature optimization vector according to the channel feature vector of the candidate collection channel; and according to the user feature optimization vector, the verbal skill vector optimization matrix and the channel feature optimization vector, determining a target collection verbal skill and a target collection channel for collecting the to-be-collected user. The effect of intelligently and automatically selecting the collection channel and the collection verbal skill which are most suitable for the to-be-collected user is achieved, the cost of human input is reduced, and the efficiency and the success rate of credit card collection are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a credit card collection method, device, equipment, medium and product. Background Art

[0002] With the rapid economic growth of today's society, the consumption model based on personal credit has gradually been accepted by everyone, and more and more people have begun to use credit cards as a means of daily payment. Although this payment method can meet people's consumption needs and effectively improve consumption power, it also brings some financial risks and risks to cardholders and card issuers, which puts forward certain requirements on the risk control ability of card issuers. Specifically, according to the different stages of using credit cards, risk control can be divided into three stages: credit limit approval, abnormal transaction detection and credit card collection. Only by strictly and effectively controlling the risks in these three stages can a good credit card system be created.

[0003] Most existing credit card collection technologies can only simply filter out a list of users who need to be collected, but lack automatic selection of collection words and collection channels. This means that collection words and collection channels still need to be manually configured to collect debts from users, resulting in excessive labor costs and affecting the efficiency and success rate of credit card collection. Summary of the invention

[0004] The present invention provides a credit card collection method, device, equipment, medium and product to solve the problem that the existing credit card collection technology relies on manual configuration of collection words and collection channels, resulting in excessive labor cost investment and low efficiency and success rate of credit card collection.

[0005] According to one aspect of the present invention, a credit card collection method is provided, the method comprising:

[0006] Determine the user to be collected corresponding to the overdue credit card, and at least one candidate collection speech and at least one candidate collection channel associated with the user to be collected;

[0007] Determine a target user associated with the user to be collected from other users, and perform feature splicing based on the target user feature vector corresponding to the target user and the user feature vector to be collected corresponding to the user to be collected, to determine a user feature optimization vector corresponding to the user to be collected;

[0008] According to the speech vector matrix corresponding to the candidate collection speech and the word vector corresponding to the candidate collection word, the speech vector optimization matrix corresponding to the candidate collection speech is determined, and according to the channel feature vector corresponding to the candidate collection channel, the channel feature optimization vector corresponding to the candidate collection channel is determined; wherein the candidate collection word is the word included in the candidate collection speech;

[0009] According to the user characteristic optimization vector, the speech vector optimization matrix and the channel characteristic optimization vector, a target collection speech is determined from the candidate collection speech, and a target collection channel is determined from the candidate collection channels, and according to the target collection speech and the target collection channel, collection is performed on the users to be collected.

[0010] According to another aspect of the present invention, a credit card collection device is provided, the device comprising:

[0011] A collection information determination module, used to determine the user to be collected corresponding to the overdue credit card, and at least one candidate collection speech and at least one candidate collection channel associated with the user to be collected;

[0012] A first vector optimization module is used to determine a target user associated with the user to be collected from other users, and perform feature splicing based on a target user feature vector corresponding to the target user and a user feature vector to be collected corresponding to the user to be collected, to determine a user feature optimization vector corresponding to the user to be collected;

[0013] A second vector optimization module is used to determine the speech vector optimization matrix corresponding to the candidate collection speech based on the speech vector matrix corresponding to the candidate collection speech and the word vector corresponding to the candidate collection word, and determine the channel feature optimization vector corresponding to the candidate collection channel based on the channel feature vector corresponding to the candidate collection channel; wherein the candidate collection word is the word included in the candidate collection speech;

[0014] The collection module is used to determine the target collection speech from the candidate collection speech and the target collection channel from the candidate collection channels according to the user characteristic optimization vector, the speech vector optimization matrix and the channel characteristic optimization vector, and to collect the debts from the users to be collected according to the target collection speech and the target collection channel.

[0015] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0016] at least one processor; and

[0017] a memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the credit card collection method described in any one of the present invention.

[0019] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to implement the credit card collection method described in any one of the present invention when executed by a processor.

[0020] According to another aspect of the present invention, there is provided a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the credit card collection method according to any one of the present invention is implemented.

[0021] The present invention determines the user feature optimization vector corresponding to the user to be collected, the speech vector optimization matrix corresponding to the candidate collection speech, and the channel feature optimization vector corresponding to the candidate collection channel, and determines the target collection speech from the candidate collection speech and the target collection channel from the candidate collection channels according to the user feature optimization vector, the speech vector optimization matrix and the channel feature optimization vector, and collects debts from the user to be collected according to the target collection speech and the target collection channel, thereby realizing the effect of intelligently and automatically selecting the collection channel and the collection speech that are most suitable for the user to be collected, without the need to manually configure the collection speech and the collection channel for collection, thereby reducing the cost of manpower investment and improving the efficiency and success rate of credit card collection.

[0022] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 A flowchart of a credit card collection method provided in Embodiment 1 of the present invention;

[0025] Figure 2 A flowchart of a credit card collection method provided in Embodiment 2 of the present invention;

[0026] Figure 3A flowchart of a credit card collection method provided in Embodiment 3 of the present invention;

[0027] Figure 4 A flowchart of a credit card collection method provided in Embodiment 4 of the present invention;

[0028] Figure 5 A flowchart of a credit card collection method provided in Embodiment 5 of the present invention;

[0029] Figure 6 A schematic diagram of the structure of a credit card collection device provided in Embodiment 6 of the present invention;

[0030] Figure 7 The present invention is a schematic diagram of the structure of an electronic device for implementing the credit card collection method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0032] It should be noted that the terms "candidate", "other", "target", "first", "second", "third", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0033] Embodiment 1

[0034] Figure 1 A flowchart of a credit card collection method provided in Example 1 of the present invention. This embodiment can be used to intelligently and automatically select the most suitable collection channels and collection words for the user to be collected, and is used to collect credit card debts from the user to be collected. The method can be executed by a credit card collection device, and the credit card collection device can be implemented in the form of hardware and / or software.

[0035] like Figure 1 As shown, the method includes:

[0036] S101. Determine the user to be collected corresponding to the overdue credit card, as well as at least one candidate collection speech and at least one candidate collection channel associated with the user to be collected.

[0037] Among them, overdue credit cards refer to the situation where the credit card user fails to repay the amount of his consumption within the repayment date specified by the card issuer, or the repayment amount is lower than the minimum repayment amount. Users to be collected refer to the cardholders corresponding to overdue credit cards, that is, the cardholders who need credit card collection.

[0038] Collection scripts refer to a series of language skills and strategies used in the credit card collection process, which are designed to effectively communicate with the users to be collected and encourage them to repay their credit card debts as soon as possible. It can be understood that candidate collection scripts refer to pre-configured collection scripts that can be selected during the collection stage for users to be collected.

[0039] Collection channels refer to the methods and means used by card issuers in the credit card collection stage, including but not limited to telephone collection, SMS collection or email collection, etc. It is understood that candidate collection channels refer to pre-configured collection channels that can be selected during the collection stage for users to be collected.

[0040] In one implementation, a credit card with overdue payment behavior is determined as an overdue credit card, and based on the holding relationship between the credit card and the user (cardholder), the user corresponding to the overdue credit card is determined as a user to be collected. For example, assuming that there is a holding relationship between credit card A and user B, and assuming that credit card A is determined to be an overdue credit card, user B is determined to be a user to be collected based on the holding relationship.

[0041] Optional collection words and collection channels are pre-configured for each user, and the association between the user and the collection words and collection channels is established. After the user to be collected is determined, the association between the user and the collection words and collection channels is used to determine the collection words associated with the user to be collected as candidate collection words, and the collection channels associated with the user to be collected as candidate collection channels.

[0042] For example, it is assumed that the optional collection words for user C are pre-configured to include collection words 1, collection words 2, and collection words 3, and the optional collection channels include collection channel 1 and collection channel 2. Assuming that user C is determined to be a user to be collected, then based on the association between the user and the collection words and collection channels, collection words 1, collection words 2, and collection words 3 are determined as candidate collection words, and collection channel 1 and collection channel 2 are determined as candidate collection channels.

[0043] S102: Determine a target user associated with the user to be collected from other users, and perform feature splicing based on the target user feature vector corresponding to the target user and the user feature vector to be collected corresponding to the user to be collected, to determine a user feature optimization vector corresponding to the user to be collected.

[0044] Among them, other users refer to users other than the user to be collected, which may be cardholders or merchants, etc. The association relationship reflects whether there is a direct or indirect transaction relationship between the user to be collected and other users. For example, assuming that there is a direct transaction relationship between other user 1 and user to be collected 2, it means that there is an association relationship between other user 1 and user to be collected 2; for another example, assuming that there is a direct transaction relationship between other user 1 and other user 2, and there is a direct transaction relationship between other user 2 and user to be collected 2, that is, there is an indirect transaction relationship between other user 1 and user to be collected 2, it means that there is an association relationship between other user 1 and user to be collected 2.

[0045] The user feature vector is a vector representation of the user's basic feature information, including but not limited to age, job type, income, merchant category and business conditions. It is understandable that the target user feature vector is the user feature vector corresponding to the target user; the pending collection user feature vector is the user feature vector corresponding to the pending collection user. It is understandable that the collected feature information is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good customs, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0046] In one embodiment, the direct and indirect transaction relationships between users are determined, and other users who have direct or indirect transaction relationships with the user to be collected are selected as target users. Further, target feature information corresponding to the target user is determined from pre-collected feature information, and vector conversion is performed based on the target feature information to generate a target user feature vector; and feature information to be collected corresponding to the user to be collected is determined from pre-collected feature information, and vector conversion is performed based on the feature information to be collected to generate a user feature vector to be collected.

[0047] The target user feature vectors corresponding to each target user are averaged, and the average user feature vector is determined based on the average calculation result; feature splicing is performed based on the average user feature vector and the feature vector of the user to be collected, and the user feature optimization vector corresponding to the user to be collected is determined based on the feature splicing result. Since the user feature optimization vector is obtained by splicing the target user feature vector and the feature vector of the user to be collected, the user feature optimization vector includes not only the feature expression of the user to be collected, but also the feature expression of the target user, thereby improving the feature richness of the user feature optimization vector.

[0048] S103. Determine the speech vector optimization matrix corresponding to the candidate collection speech based on the speech vector matrix corresponding to the candidate collection speech and the word vector corresponding to the candidate collection words, and determine the channel feature optimization vector corresponding to the candidate collection channel based on the channel feature vector corresponding to the candidate collection channel.

[0049] The word vector matrix refers to the feature vector matrix expression of the candidate collection words, and the word vector refers to the feature vector expression of the candidate collection words. The candidate collection words are the words included in the candidate collection words. The channel feature vector refers to the feature vector expression of the candidate collection channel.

[0050] In one embodiment, for determining the speech vector optimization matrix: a self-attention mechanism is processed according to the speech vector matrix to generate a first self-attention processing result, and the first self-attention processing result is input into the first feedforward network to reduce the dimension of the speech vector matrix and generate a reduced dimension matrix. A self-attention mechanism is processed according to the reduced dimension matrix and the word vector to generate a second self-attention processing result, and the second self-attention processing result is input into the second feedforward network to reduce the dimension of the word vector and generate a reduced dimension vector. Furthermore, mean pooling is performed according to each reduced dimension vector, and the speech vector optimization matrix corresponding to the candidate collection speech is determined according to the mean pooling result.

[0051] In another embodiment, the channel feature optimization vector is determined as follows: the channel feature vector is input into the target fully connected layer, and the channel feature vector is weightedly calculated using the target weight value and the target bias value of the target fully connected layer to generate a weighted feature vector; the weighted feature vector is processed using the target activation function of the target fully connected layer to generate a channel feature optimization vector corresponding to the candidate collection channel.

[0052] S104. Determine the target collection speech from the candidate collection speech and the target collection channel from the candidate collection channels according to the user feature optimization vector, the speech vector optimization matrix and the channel feature optimization vector, and collect debts from the users to be collected according to the target collection speech and the target collection channel.

[0053] In one embodiment, the user feature optimization vector is combined with each speech vector optimization matrix and each channel feature optimization vector to determine at least one group of candidate feature combination vectors. Each candidate feature combination vector is input into a pre-trained collection success rate prediction model to predict the collection success rate corresponding to each candidate feature combination vector. Finally, the candidate collection speech corresponding to the speech vector optimization matrix included in the candidate feature combination vector with the highest collection success rate is selected as the target collection speech, and the candidate collection channel corresponding to the channel feature optimization vector is selected as the target collection channel. Furthermore, according to the target collection speech and the target collection channel, the collection users are collected.

[0054] The embodiment of the present invention determines the user feature optimization vector corresponding to the user to be collected, the speech vector optimization matrix corresponding to the candidate collection speech, and the channel feature optimization vector corresponding to the candidate collection channel, and determines the target collection speech from the candidate collection speech and the target collection channel from the candidate collection channels according to the user feature optimization vector, the speech vector optimization matrix and the channel feature optimization vector, and collects the debt from the user to be collected according to the target collection speech and the target collection channel, thereby realizing the effect of intelligently and automatically selecting the collection channel and the collection speech that are most suitable for the user to be collected, without the need to manually configure the collection speech and the collection channel for collection, thereby reducing the cost of manpower investment and improving the efficiency and success rate of credit card collection.

[0055] Embodiment 2

[0056] Figure 2 This is a flow chart of a credit card collection method provided by Embodiment 2 of the present invention. This embodiment further optimizes and expands the above embodiment and can be combined with the above optional implementations. Figure 2 As shown, the method includes:

[0057] S201. Determine the user to be collected corresponding to the overdue credit card, as well as at least one candidate collection speech and at least one candidate collection channel associated with the user to be collected.

[0058] S202: Determine in the target bipartite graph the node corresponding to the user to be collected as the first user node, and determine the nodes corresponding to other users as the second user nodes.

[0059] The target bipartite graph consists of a first user node and a second user node. User nodes in the bipartite graph are classified into two categories according to their types. The edges between user nodes only exist between user nodes of different types, and there are no edges between user nodes of the same type. For example, user nodes can be divided into two types: customer nodes and merchant nodes. Edges only exist between user nodes of different types. An edge exists if and only if the customer node has made a purchase with a credit card at the merchant node.

[0060] S203: Based on the first user node, random walk sampling is performed on the second user node to determine the second user node having an association relationship with the first user node as the third user node, and other users corresponding to the third user node are used as target users having an association relationship with the user to be collected.

[0061] In one embodiment, a random walk algorithm is used to perform random walk sampling on each second user node based on the first user node, and a third user node associated with the first user node is determined from each second user node according to the random walk sampling result. Further, according to the relationship between the user node and the user, other users corresponding to the third user node are determined as target users associated with the user to be collected.

[0062] The node corresponding to the user to be collected is determined as the first user node in the target bipartite graph, and the nodes corresponding to other users are determined as the second user node; wherein the target bipartite graph is composed of the first user node and the second user node; random walk sampling is performed on the second user node based on the first user node to determine the second user node having an association relationship with the first user node as the third user node; other users corresponding to the third user node are used as target users having an association relationship with the user to be collected, thereby realizing the use of graph network technology to determine the target users having an association relationship with the user to be collected, thereby ensuring the efficiency and accuracy of the target user determination.

[0063] S204, performing mean calculation according to the vector elements respectively contained in each target user feature vector, and determining the mean user feature vector according to the mean calculation result; performing feature splicing according to the mean user feature vector and the feature vector of the user to be collected, and determining the spliced ​​user feature vector according to the feature splicing result.

[0064] In one implementation, the target bipartite graph is input into a multi-layer graph convolutional neural network, and the following calculations are performed using each hidden layer in the multi-layer graph convolutional neural network:

[0065]

[0066] in, represents the mean user feature vector calculated by the kth hidden layer, N(v) represents the set of target users, and u represents each target user. It represents the target user feature vector calculated by the target user u in the k-1th hidden layer. AVG represents the operation of finding the average of each vector element in the vector.

[0067] Furthermore, the following calculations are performed using each hidden layer in the multi-layer graph convolutional neural network:

[0068]

[0069] Among them, σ(·) represents a nonlinear activation function, which is used to increase the learning ability of the model. k Represents the linear mapping matrix corresponding to the k-th layer, which is used for dimensionality reduction. represents the concatenated user feature vector calculated by the kth hidden layer, represents the concatenated user feature vector calculated by the k-1th hidden layer, It represents the mean user feature vector calculated by the kth hidden layer. CONCAT(a,b) represents the concatenation of feature vectors a and b.

[0070] S205: Determine a target binary norm corresponding to the concatenated user feature vector, and determine a user feature optimization vector corresponding to the user to be collected based on a ratio between the concatenated user feature vector and the target binary norm.

[0071] In one implementation, the following calculation is performed using each hidden layer in the multi-layer graph convolutional neural network to obtain the user feature optimization vector calculated by the kth hidden layer:

[0072]

[0073] Among them, ||·|| represents the target binary norm corresponding to the concatenated user feature vector.

[0074] By performing mean calculation according to the vector elements respectively contained in each target user feature vector and determining the mean user feature vector according to the mean calculation result, the effect of performing aggregation operation on the target user feature vector is achieved, and the balance of feature expression of the mean user feature vector is ensured.

[0075] By performing feature splicing based on the mean user feature vector and the feature vector of the user to be collected, and determining the spliced ​​user feature vector based on the feature splicing result, since the spliced ​​user feature vector is obtained by splicing the mean user feature vector and the feature vector of the user to be collected, the spliced ​​user feature vector includes not only the feature expression of the user to be collected, but also the feature expression of the target user, thereby improving the feature richness of the spliced ​​user feature vector.

[0076] By determining the target binary norm corresponding to the spliced ​​user feature vector and determining the user feature optimization vector corresponding to the user to be collected based on the ratio between the spliced ​​user feature vector and the target binary norm, the effect of regularization processing is achieved and the robustness of the user feature optimization vector is ensured.

[0077] S206. Determine the speech vector optimization matrix corresponding to the candidate collection speech based on the speech vector matrix corresponding to the candidate collection speech and the word vector corresponding to the candidate collection words, and determine the channel feature optimization vector corresponding to the candidate collection channel based on the channel feature vector corresponding to the candidate collection channel.

[0078] Among them, the candidate collection words are the words included in the candidate collection scripts.

[0079] S207. Determine the target collection speech from the candidate collection speech and the target collection channel from the candidate collection channels based on the user feature optimization vector, the speech vector optimization matrix and the channel feature optimization vector, and collect debts from the users to be collected based on the target collection speech and the target collection channel.

[0080] Embodiment 3

[0081] Figure 3 This is a flowchart of a credit card collection method provided by Embodiment 3 of the present invention. This embodiment further optimizes and expands the above embodiment and can be combined with the above optional implementations. Figure 3 As shown, the method includes:

[0082] S301. Determine the user to be collected corresponding to the overdue credit card, as well as at least one candidate collection speech and at least one candidate collection channel associated with the user to be collected.

[0083] S302: Determine a target user associated with the user to be collected from other users, and perform feature splicing based on the target user feature vector corresponding to the target user and the user feature vector to be collected corresponding to the user to be collected, to determine a user feature optimization vector corresponding to the user to be collected.

[0084] S303, performing self-attention mechanism processing according to the discourse vector matrix and the target dictionary length to generate a first self-attention processing result; inputting the first self-attention processing result into the first feedforward network to reduce the dimension of the discourse vector matrix to generate a reduced dimension matrix.

[0085] Among them, the target dictionary length is the dictionary length corresponding to the target dictionary. The target dictionary is used to segment the dialogue vector matrix and determine the word vector. The self-attention mechanism is inspired by the retrieval system. For a set of key-value pairs, according to the relationship between the query condition and the key, a set of weights of different sizes are calculated and applied to the set of values. It is often used in neural network models to filter out irrelevant information and focus on more useful parts.

[0086] In one embodiment, the speech vector matrix is ​​input into a language model, which is composed of an encoder and a decoder. The encoder is responsible for learning the structural pattern of the input sequence, while the decoder is responsible for combining the structural pattern with the current input of the sequence to obtain a predicted output value.

[0087] The encoder is used to process the self-attention mechanism according to the speech vector matrix and the target dictionary length to generate the first self-attention processing result, including the following methods:

[0088]

[0089] in, represents the discourse vector matrix, d represents the target dictionary length, and softmax(·) is a normalization function that normalizes the elements in each row vector of the matrix input to the interval between 0 and 1 and makes the sum of all elements of the row vector equal to 1.

[0090] The encoder is used to input the first self-attention processing result into the first feedforward network to reduce the dimension of the dialogue vector matrix and generate a reduced dimension matrix, including the following methods:

[0091]

[0092] Among them, W1, and W2, These are all trainable model parameters. This calculated low-dimensional vector FFN encode (X) is the dimension reduction matrix finally output by the encoder.

[0093] S304, performing self-attention mechanism processing according to the dimensionality reduction matrix, word vector and target dictionary length to generate a second self-attention processing result, and inputting the second self-attention processing result into the second feedforward network to reduce the dimensionality of the word vector and generate a reduced dimensionality vector.

[0094] In one embodiment, the decoder is used to perform a self-attention mechanism according to the dimension reduction matrix, the word vector, and the target dictionary length to generate a second self-attention processing result, including the following methods:

[0095]

[0096] Among them, X (i，·) represents the i-th word vector, and FFN(X) represents the dimensionality reduction matrix.

[0097] The decoder is used to input the second self-attention processing result into the second feedforward network to reduce the dimension of the word vector and generate a reduced-dimensional vector. The process can refer to the processing method of the first feedforward network and will not be repeated here.

[0098] S305. Perform mean pooling on each dimensionality reduction vector, and determine a speech vector optimization matrix corresponding to the candidate collection speech based on the mean pooling result.

[0099] In one implementation, a language model is used to perform mean pooling on each dimensionality reduction vector, and a speech vector optimization matrix corresponding to the candidate collection speech is determined according to the mean pooling result, including the following methods:

[0100]

[0101] Wherein, D(l) represents all candidate collection words included in the candidate collection words.

[0102] By processing the self-attention mechanism according to the speech vector matrix and the target dictionary length, the first self-attention processing result is generated, which is conducive to filtering out irrelevant information in the speech vector matrix and extracting useful information in the speech vector matrix.

[0103] By inputting the first self-attention processing result into the first feedforward network, the dimensionality of the speech vector matrix is ​​reduced to generate a reduced dimensionality matrix, thereby mapping the speech vector matrix to a new low-dimensional vector space, which is beneficial to reducing the amount of data in the speech vector matrix.

[0104] By processing the self-attention mechanism according to the dimensionality reduction matrix, word vector and target dictionary length, a second self-attention processing result is generated, which is helpful to filter out irrelevant information in the word vector and extract useful information in the word vector.

[0105] By inputting the second self-attention processing result into the second feedforward network, the word vector is reduced in dimension to generate a reduced-dimensional vector, thereby mapping the word vector to a new low-dimensional vector space, which is beneficial to reducing the amount of data of the word vector.

[0106] By performing mean pooling on each dimensionality reduction vector and determining the optimized matrix of the speech vector corresponding to the candidate collection speech based on the mean pooling result, it is helpful to reduce the amount of calculation and prevent overfitting.

[0107] S306. Determine a channel feature optimization vector corresponding to the candidate collection channel according to the channel feature vector corresponding to the candidate collection channel.

[0108] Among them, the candidate collection words are the words included in the candidate collection scripts.

[0109] S307. Determine the target collection speech from the candidate collection speech and the target collection channel from the candidate collection channels according to the user feature optimization vector, the speech vector optimization matrix and the channel feature optimization vector, and collect debts from the users to be collected according to the target collection speech and the target collection channel.

[0110] Embodiment 4

[0111] Figure 4 This is a flowchart of a credit card collection method provided by Embodiment 4 of the present invention. This embodiment further optimizes and expands the above embodiment and can be combined with the above optional implementations. Figure 4 As shown, the method includes:

[0112] S401. Determine the user to be collected corresponding to the overdue credit card, as well as at least one candidate collection speech and at least one candidate collection channel associated with the user to be collected.

[0113] S402: Determine a target user associated with the user to be collected from other users, and perform feature splicing based on the target user feature vector corresponding to the target user and the user feature vector to be collected corresponding to the user to be collected, to determine a user feature optimization vector corresponding to the user to be collected.

[0114] S403. Determine an optimized matrix of speech vectors corresponding to the candidate collection speech based on the speech vector matrix corresponding to the candidate collection speech and the word vectors corresponding to the candidate collection words.

[0115] S404: Input the channel feature vector into the target fully connected layer, and perform weighted calculation on the channel feature vector using the target weight value and the target bias value of the target fully connected layer to generate a weighted feature vector.

[0116] S405. Process the weighted feature vector using the target activation function of the target fully connected layer to generate a channel feature optimization vector corresponding to the candidate collection channel.

[0117] Among them, the type of the target fully connected layer is a single-layer fully connected layer.

[0118] In one embodiment, the channel feature optimization vector is determined in the following manner:

[0119]

[0120] Among them, the matrix W and the vector They represent the target weight value and target bias value of the target fully connected layer, which are trainable model parameters. σ(·) represents the target activation function, which is used to increase the learning ability of the model. The optional function is ReLu.

[0121] By inputting the channel feature vector into the target fully connected layer, the target weight value and target bias value of the target fully connected layer are used to perform weighted calculation on the channel feature vector to generate a weighted feature vector; the weighted feature vector is processed using the target activation function of the target fully connected layer to generate a channel feature optimization vector corresponding to the candidate collection channel, which is beneficial to reduce the influence of the feature position on the channel feature optimization vector and ensure the robustness of the channel feature optimization vector.

[0122] S406. Determine the target collection speech from the candidate collection speech and the target collection channel from the candidate collection channels according to the user feature optimization vector, the speech vector optimization matrix and the channel feature optimization vector, and collect debts from the users to be collected according to the target collection speech and the target collection channel.

[0123] Embodiment 5

[0124] Figure 5 This is a flowchart of a credit card collection method provided in Example 5 of the present invention. This embodiment further optimizes and expands the above-mentioned embodiments of "determining the target collection speech from the candidate collection speech according to the user feature optimization vector, the speech vector optimization matrix and the channel feature optimization vector, and determining the target collection channel from the candidate collection channels", and can be combined with the above-mentioned optional implementation methods. Figure 5 As shown, the method includes:

[0125] S501. Determine at least one group of candidate feature combination vectors according to the combination results of the user feature optimization vector, each speech vector optimization matrix and each channel feature optimization vector.

[0126] Exemplarily, it is assumed that the speech vector optimization matrix includes speech vector optimization matrix 1 and speech vector optimization matrix 2; the channel feature optimization vector includes channel feature optimization vector 1 and channel feature optimization vector 2. Then the candidate feature combination vectors include <user feature optimization vector, speech vector optimization matrix 1, channel feature optimization vector 1>, <user feature optimization vector, speech vector optimization matrix 1, channel feature optimization vector 2>, <user feature optimization vector, speech vector optimization matrix 2, channel feature optimization vector 1> and <user feature optimization vector, speech vector optimization matrix 2, channel feature optimization vector 2>.

[0127] S502. Input the candidate feature combination vectors into the collection success rate prediction model to determine the collection success rate corresponding to each candidate feature combination vector, and determine the target feature combination vector from each candidate feature combination vector according to the ranking results of the collection success rate corresponding to each candidate feature combination vector.

[0128] Among them, the collection success rate prediction model is a fully connected layer model.

[0129] In one embodiment, the candidate feature combination vectors are input into the collection success rate prediction model, the collection success rate is predicted using the collection success rate prediction model, and the collection success rates corresponding to the candidate feature combination vectors are output. Furthermore, the candidate feature combination vector with the largest collection success rate is used as the target feature combination vector.

[0130] For example, assuming that the collection success rate prediction model outputs candidate feature combination vector 1, candidate feature combination vector 2, and candidate feature combination vector 3, which correspond to collection success rates of 90%, 80%, and 70%, respectively, candidate feature combination vector 1 is used as the target feature combination vector.

[0131] S503: Use the speech vector optimization matrix included in the target feature combination vector as the speech vector optimization matrix to be selected, and use the channel feature optimization vector included in the target feature combination vector as the channel feature optimization vector to be selected.

[0132] S504. The candidate collection speech corresponding to the selected speech vector optimization matrix is ​​used as the target collection speech, and the candidate collection channel corresponding to the selected channel feature optimization vector is used as the target collection channel.

[0133] Exemplarily, assuming that the target feature combination vector includes speech vector optimization matrix 1 and channel feature optimization vector 2, speech vector optimization matrix 1 is used as the speech vector optimization matrix to be selected, and channel feature optimization vector 2 is used as the channel feature optimization vector to be selected. Assuming that the candidate collection speech corresponding to the speech vector optimization matrix to be selected is "candidate collection speech A", and the candidate collection channel corresponding to the channel feature optimization vector to be selected is "candidate collection channel B", "candidate collection speech A" is used as the target collection speech, and "candidate collection channel B" is used as the target collection channel.

[0134] At least one group of candidate feature combination vectors is determined based on the combination results of the user feature optimization vector with each speech vector optimization matrix and each channel feature optimization vector; the candidate feature combination vectors are input into the collection success rate prediction model to determine the collection success rate corresponding to each candidate feature combination vector, and the target feature combination vector is determined from each candidate feature combination vector according to the ranking results of the collection success rate corresponding to each candidate feature combination vector; the speech vector optimization matrix included in the target feature combination vector is used as the speech vector optimization matrix to be selected, and the channel feature optimization vector included in the target feature combination vector is used as the channel feature optimization vector to be selected; the candidate collection speech corresponding to the speech vector optimization matrix to be selected is used as the target collection speech, and the candidate collection channel corresponding to the channel feature optimization vector to be selected is used as the target collection channel, so as to intelligently and accurately select the collection channel and collection speech that are most suitable for the user to be collected, so as to better improve the collection effect and success rate and ensure the user's card usage satisfaction.

[0135] Optionally, the collection success rate prediction model is generated in the following manner:

[0136] A. Obtain user feature sample vectors corresponding to sample users, speech vector sample matrices corresponding to sample collection speech, and channel feature sample vectors corresponding to sample collection channels, and determine at least one set of sample feature combination vectors based on the combination results of the user feature sample vectors with each speech vector sample matrix and each channel feature sample vector.

[0137] Among them, sample users, sample collection words and sample collection channels are pre-collected sample data. User feature sample vectors can refer to the method for determining user feature optimization vectors, word vector sample matrices can refer to the method for determining word vector optimization matrices, and channel feature sample vectors can refer to the method for determining channel feature optimization vectors, which will not be described in detail here.

[0138] B. Input the sample feature combination vector into the model to be trained to determine the predicted collection success rate corresponding to each sample feature combination vector.

[0139] C. Based on the actual collection success rate and predicted collection success rate annotated by each sample feature combination vector, the model to be trained is trained to generate a collection success rate prediction model.

[0140] In one embodiment, a preset loss function is used to determine a loss value based on an actual collection success rate and a predicted collection success rate, and a model to be trained is trained based on the loss value to generate a collection success rate prediction model.

[0141] Optionally, the preset loss function includes the following functions:

[0142]

[0143] Among them, y t Denotes the predicted collection success rate, y' t Indicates the actual collection success rate of the annotation.

[0144] Optional, predict the collection success rate y t The last layer output of the multi-layer fully connected structure in the training model can be Add the following sigmoid function as the output layer of the model:

[0145]

[0146] in, It can be determined using the following formula:

[0147]

[0148] Among them, W k and b k They represent the weight and bias of the fully connected layer respectively, which are trainable model parameters. is the input of the current fully connected layer.

[0149] Among them, if the collection of any of the sample collection words and any of the sample collection channels is successful on any of the sample users, the actual collection success rate of the marked target sample feature combination vector is 1; the target sample feature combination vector includes the speech vector sample matrix corresponding to the sample collection words, the channel feature sample vector corresponding to the sample collection channel, and the user feature sample vector corresponding to the sample user.

[0150] For example, assuming that sample collection speech 1 and sample collection channel 1 are used to successfully collect debts from sample user X, the target sample feature combination vector composed of the speech vector sample matrix corresponding to sample collection speech 1, the channel feature sample vector corresponding to sample collection channel 1, and the user feature sample vector corresponding to sample user X is marked with an actual collection success rate of 1.

[0151] By obtaining user feature sample vectors corresponding to sample users, speech vector sample matrices corresponding to sample collection speech, and channel feature sample vectors corresponding to sample collection channels, and based on the combination results of user feature sample vectors with each speech vector sample matrix and each channel feature sample vector, at least one set of sample feature combination vectors is determined; the sample feature combination vectors are input into the model to be trained to determine the predicted collection success rate corresponding to each sample feature combination vector; based on the actual collection success rate and the predicted collection success rate marked by each sample feature combination vector, the model to be trained is trained to generate a collection success rate prediction model, so that the trained collection success rate prediction model has the ability to predict the collection success rate, thereby ensuring that the credit card collection method can be smoothly and effectively executed.

[0152] Embodiment 6

[0153] Figure 6 This is a schematic diagram of the structure of a credit card collection device provided in Embodiment 6 of the present invention, which can be used to intelligently and automatically select the most suitable collection channel and collection speech for the user to be collected, and is used for collecting credit card debts from the user to be collected, such as Figure 6 As shown, the device comprises:

[0154] The collection information determination module 61 is used to determine the user to be collected corresponding to the overdue credit card, and at least one candidate collection speech and at least one candidate collection channel associated with the user to be collected;

[0155] The first vector optimization module 62 is used to determine a target user associated with the user to be collected from other users, and perform feature splicing based on the target user feature vector corresponding to the target user and the user feature vector to be collected corresponding to the user to be collected, to determine a user feature optimization vector corresponding to the user to be collected;

[0156] The second vector optimization module 63 is used to determine the speech vector optimization matrix corresponding to the candidate collection speech according to the speech vector matrix corresponding to the candidate collection speech and the word vector corresponding to the candidate collection word, and determine the channel feature optimization vector corresponding to the candidate collection channel according to the channel feature vector corresponding to the candidate collection channel; wherein the candidate collection word is the word included in the candidate collection speech;

[0157] The collection module 64 is used to determine the target collection speech from the candidate collection speech according to the user characteristic optimization vector, the speech vector optimization matrix and the channel characteristic optimization vector, and determine the target collection channel from the candidate collection channels, and collect the debts from the users to be collected according to the target collection speech and the target collection channel.

[0158] Optionally, the first vector optimization module 62 is specifically configured to:

[0159] Determine in a target bipartite graph the node corresponding to the user to be collected as the first user node, and determine the node corresponding to the other user as the second user node; wherein the target bipartite graph is composed of the first user node and the second user node;

[0160] Performing random walk sampling on the second user node based on the first user node to determine the second user node having an association relationship with the first user node as a third user node;

[0161] The other user corresponding to the third user node is used as a target user having an associated relationship with the user to be collected.

[0162] Optionally, the first vector optimization module 62 is further configured to:

[0163] Performing mean calculation according to the vector elements respectively included in each of the target user feature vectors, and determining a mean user feature vector according to the mean calculation result;

[0164] Performing feature splicing according to the mean user feature vector and the to-be-collected user feature vector, and determining a spliced ​​user feature vector according to the feature splicing result;

[0165] The target binary norm corresponding to the concatenated user feature vector is determined, and the user feature optimization vector corresponding to the user to be collected is determined according to the ratio between the concatenated user feature vector and the target binary norm.

[0166] Optionally, the second vector optimization module 63 is specifically configured to:

[0167] Processing the self-attention mechanism according to the speech vector matrix and the target dictionary length to generate a first self-attention processing result; wherein the target dictionary length is the dictionary length corresponding to the target dictionary, and the target dictionary is used to segment the speech vector matrix to determine the word vector;

[0168] Inputting the first self-attention processing result into the first feedforward network to reduce the dimension of the speech vector matrix and generate a reduced-dimensional matrix;

[0169] Performing a self-attention mechanism processing according to the dimension reduction matrix, the word vector and the target dictionary length to generate a second self-attention processing result, and inputting the second self-attention processing result into a second feedforward network to reduce the dimension of the word vector and generate a reduced dimension vector;

[0170] Mean pooling is performed based on each of the dimensionality reduction vectors, and an optimization matrix of speech vectors corresponding to the candidate debt collection speech is determined based on the mean pooling result.

[0171] Optionally, the second vector optimization module 63 is further configured to:

[0172] Inputting the channel feature vector into the target fully connected layer, performing weighted calculation on the channel feature vector using the target weight value and the target bias value of the target fully connected layer, and generating a weighted feature vector;

[0173] The weighted feature vector is processed using the target activation function of the target fully connected layer to generate a channel feature optimization vector corresponding to the candidate collection channel.

[0174] Optionally, the collection module 64 is specifically configured to:

[0175] Determine at least one set of candidate feature combination vectors according to the combination results of the user feature optimization vector, each of the speech vector optimization matrices and each of the channel feature optimization vectors;

[0176] The candidate feature combination vectors are input into a collection success rate prediction model to determine the collection success rates corresponding to the candidate feature combination vectors, and a target feature combination vector is determined from the candidate feature combination vectors according to the ranking results of the collection success rates corresponding to the candidate feature combination vectors;

[0177] The speech vector optimization matrix included in the target feature combination vector is used as the speech vector optimization matrix to be selected, and the channel feature optimization vector included in the target feature combination vector is used as the channel feature optimization vector to be selected;

[0178] The candidate collection speech corresponding to the selected speech vector optimization matrix is ​​used as the target collection speech, and the candidate collection channel corresponding to the selected channel feature optimization vector is used as the target collection channel.

[0179] Optionally, the collection success rate prediction model is generated in the following manner:

[0180] Obtain user feature sample vectors corresponding to sample users, speech vector sample matrices corresponding to sample collection speech, and channel feature sample vectors corresponding to sample collection channels, and determine at least one set of sample feature combination vectors based on the combination results of the user feature sample vectors with each of the speech vector sample matrices and each of the channel feature sample vectors;

[0181] Inputting the sample feature combination vectors into the model to be trained to determine the predicted collection success rate corresponding to each sample feature combination vector;

[0182] According to the actual collection success rate and the predicted collection success rate marked by each of the sample feature combination vectors, the model to be trained is trained to generate the collection success rate prediction model;

[0183] Among them, if the collection of any of the sample collection words and any of the sample collection channels is successful on any of the sample users, the actual collection success rate of the marked target sample feature combination vector is 1; the target sample feature combination vector includes the speech vector sample matrix corresponding to the sample collection words, the channel feature sample vector corresponding to the sample collection channel, and the user feature sample vector corresponding to the sample user.

[0184] The credit card collection device provided in the embodiment of the present invention can execute the credit card collection method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0185] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0186] Embodiment 7

[0187] Figure 7A schematic diagram of an electronic device 70 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0188] like Figure 7 As shown, the electronic device 70 includes at least one processor 71, and a memory connected to the at least one processor 71, such as a read-only memory (ROM) 72, a random access memory (RAM) 73, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 71 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 72 or the computer program loaded from the storage unit 78 to the random access memory (RAM) 73. In the RAM 73, various programs and data required for the operation of the electronic device 70 can also be stored. The processor 71, the ROM 72, and the RAM 73 are connected to each other via a bus 74. An input / output (I / O) interface 75 is also connected to the bus 74.

[0189] A number of components in the electronic device 70 are connected to the I / O interface 75, including: an input unit 76, such as a keyboard, a mouse, etc.; an output unit 77, such as various types of displays, speakers, etc.; a storage unit 78, such as a disk, an optical disk, etc.; and a communication unit 79, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 79 allows the electronic device 70 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0190] The processor 71 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 71 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 71 executes the various methods and processes described above, such as a credit card collection method.

[0191] In some embodiments, the credit card collection method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 78. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 70 via the ROM 72 and / or the communication unit 79. When the computer program is loaded into the RAM 73 and executed by the processor 71, one or more steps of the credit card collection method described above may be performed. Alternatively, in other embodiments, the processor 71 may be configured to execute the credit card collection method in any other appropriate manner (e.g., by means of firmware).

[0192] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0193] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0194] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0195] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0196] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0197] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0198] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0199] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A credit card collection method, characterized in that: The method comprises: Determine the user to be collected corresponding to the overdue credit card, and at least one candidate collection speech and at least one candidate collection channel associated with the user to be collected; Determine a target user associated with the user to be collected from other users, and perform feature splicing based on the target user feature vector corresponding to the target user and the user feature vector to be collected corresponding to the user to be collected, to determine a user feature optimization vector corresponding to the user to be collected; According to the speech vector matrix corresponding to the candidate collection speech and the word vector corresponding to the candidate collection word, the speech vector optimization matrix corresponding to the candidate collection speech is determined, and according to the channel feature vector corresponding to the candidate collection channel, the channel feature optimization vector corresponding to the candidate collection channel is determined; wherein the candidate collection word is the word included in the candidate collection speech; According to the user characteristic optimization vector, the speech vector optimization matrix and the channel characteristic optimization vector, a target collection speech is determined from the candidate collection speech, and a target collection channel is determined from the candidate collection channels, and according to the target collection speech and the target collection channel, collection is performed on the users to be collected.

2. The method according to claim 1, characterized in that The step of determining a target user having an associated relationship with the user to be collected from other users includes: Determine in a target bipartite graph the node corresponding to the user to be collected as the first user node, and determine the node corresponding to the other user as the second user node; wherein the target bipartite graph is composed of the first user node and the second user node; Performing random walk sampling on the second user node based on the first user node to determine the second user node having an association relationship with the first user node as a third user node; The other user corresponding to the third user node is used as a target user having an associated relationship with the user to be collected.

3. The method according to claim 1, characterized in that The step of performing feature splicing based on the target user feature vector corresponding to the target user and the to-be-collected user feature vector corresponding to the to-be-collected user to determine the user feature optimization vector corresponding to the to-be-collected user includes: Performing mean calculation according to the vector elements respectively included in each of the target user feature vectors, and determining a mean user feature vector according to the mean calculation result; Performing feature splicing according to the mean user feature vector and the to-be-collected user feature vector, and determining a spliced ​​user feature vector according to the feature splicing result; The target binary norm corresponding to the concatenated user feature vector is determined, and the user feature optimization vector corresponding to the user to be collected is determined according to the ratio between the concatenated user feature vector and the target binary norm.

4. The method according to claim 1, characterized in that The step of determining the optimized speech vector matrix corresponding to the candidate collection speech based on the speech vector matrix corresponding to the candidate collection speech and the word vector corresponding to the candidate collection word comprises: Processing the self-attention mechanism according to the speech vector matrix and the target dictionary length to generate a first self-attention processing result; wherein the target dictionary length is the dictionary length corresponding to the target dictionary, and the target dictionary is used to segment the speech vector matrix to determine the word vector; Inputting the first self-attention processing result into the first feedforward network to reduce the dimension of the speech vector matrix and generate a reduced-dimensional matrix; Performing a self-attention mechanism processing according to the dimension reduction matrix, the word vector and the target dictionary length to generate a second self-attention processing result, and inputting the second self-attention processing result into a second feedforward network to reduce the dimension of the word vector and generate a reduced dimension vector; Mean pooling is performed based on each of the dimensionality reduction vectors, and an optimization matrix of speech vectors corresponding to the candidate debt collection speech is determined based on the mean pooling result.

5. The method according to claim 1, characterized in that The determining, according to the channel feature vector corresponding to the candidate collection channel, the channel feature optimization vector corresponding to the candidate collection channel comprises: Inputting the channel feature vector into the target fully connected layer, performing weighted calculation on the channel feature vector using the target weight value and the target bias value of the target fully connected layer, and generating a weighted feature vector; The weighted feature vector is processed using the target activation function of the target fully connected layer to generate a channel feature optimization vector corresponding to the candidate collection channel.

6. The method according to claim 1, characterized in that The step of determining a target collection speech from the candidate collection speeches and determining a target collection channel from the candidate collection channels according to the user feature optimization vector, the speech vector optimization matrix, and the channel feature optimization vector includes: Determine at least one set of candidate feature combination vectors according to the combination results of the user feature optimization vector, each of the speech vector optimization matrices and each of the channel feature optimization vectors; The candidate feature combination vectors are input into a collection success rate prediction model to determine the collection success rates corresponding to the candidate feature combination vectors, and a target feature combination vector is determined from the candidate feature combination vectors according to the ranking results of the collection success rates corresponding to the candidate feature combination vectors; The speech vector optimization matrix included in the target feature combination vector is used as the speech vector optimization matrix to be selected, and the channel feature optimization vector included in the target feature combination vector is used as the channel feature optimization vector to be selected; The candidate collection speech corresponding to the selected speech vector optimization matrix is ​​used as the target collection speech, and the candidate collection channel corresponding to the selected channel feature optimization vector is used as the target collection channel.

7. The method according to claim 6, characterized in that The collection success rate prediction model is generated in the following way: Obtain user feature sample vectors corresponding to sample users, speech vector sample matrices corresponding to sample collection speech, and channel feature sample vectors corresponding to sample collection channels, and determine at least one set of sample feature combination vectors based on the combination results of the user feature sample vectors with each of the speech vector sample matrices and each of the channel feature sample vectors; Inputting the sample feature combination vectors into the model to be trained to determine the predicted collection success rate corresponding to each sample feature combination vector; According to the actual collection success rate and the predicted collection success rate marked by each of the sample feature combination vectors, the model to be trained is trained to generate the collection success rate prediction model; Among them, if the collection of any of the sample collection words and any of the sample collection channels is successful on any of the sample users, the actual collection success rate of the marked target sample feature combination vector is 1; the target sample feature combination vector includes the speech vector sample matrix corresponding to the sample collection words, the channel feature sample vector corresponding to the sample collection channel, and the user feature sample vector corresponding to the sample user.

8. A credit card collection device, characterized in that: The device comprises: A collection information determination module, used to determine the user to be collected corresponding to the overdue credit card, and at least one candidate collection speech and at least one candidate collection channel associated with the user to be collected; A first vector optimization module is used to determine a target user associated with the user to be collected from other users, and perform feature splicing based on a target user feature vector corresponding to the target user and a user feature vector to be collected corresponding to the user to be collected, to determine a user feature optimization vector corresponding to the user to be collected; A second vector optimization module is used to determine the speech vector optimization matrix corresponding to the candidate collection speech based on the speech vector matrix corresponding to the candidate collection speech and the word vector corresponding to the candidate collection word, and determine the channel feature optimization vector corresponding to the candidate collection channel based on the channel feature vector corresponding to the candidate collection channel; wherein the candidate collection word is the word included in the candidate collection speech; The collection module is used to determine the target collection speech from the candidate collection speech and the target collection channel from the candidate collection channels according to the user characteristic optimization vector, the speech vector optimization matrix and the channel characteristic optimization vector, and to collect the debts from the users to be collected according to the target collection speech and the target collection channel.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the credit card collection method described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to execute the credit card collection method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the credit card collection method according to any one of claims 1 to 7.

Images