Trading combination recommendation method, device, equipment, storage medium and program product

Abstract

The present application provides a transaction combination recommendation method, apparatus, device, storage medium, and program product, relating to the field of financial technology or other related fields. The method comprises: obtaining user feature information of a target user, and a current session sequence converted based on the target user's current transaction data, inputting the current session sequence into a first prediction model, causing the first prediction model to abstract the current session sequence into a sequence feature vector based on an attention mechanism, and generating a first recommendation result based on the sequence feature vector, further obtaining a second recommendation result based on the first recommendation result and the user feature information, and then determining a target transaction combination to be recommended to the target user based on the second recommendation result. The method of the present application provides users with accurate and personalized transaction recommendation services.

Description

Technical Field

[0001] The present application relates to the field of financial technology or other related fields, and in particular to a transaction combination recommendation method, device, equipment, storage medium and program product. Background Art

[0002] With the rapid development of technology and the digital transformation of the financial industry, bank self-service terminal transaction systems have become an essential component of modern financial services. Amidst the reform of branch operations, traditional over-the-counter transactions are gradually shifting to self-service terminals. Currently, branch users are aging and facing a growing number of products and services. Given the limited availability of self-service terminals, helping users quickly find the transaction combination that best suits their needs has become a key area for improving the efficiency of self-service terminals.

[0003] In related technologies, when users conduct business transactions on self-service terminals, the SR-GNN algorithm is used to represent the user's current transaction data in the form of a graph structure. This algorithm then captures the spatial proximity and correlation between different transactions within the graph to generate recommended transactions. When the user completes the transaction on the self-service terminal, the recommended transactions are automatically displayed on the terminal interface. However, this approach cannot provide accurate and personalized transaction recommendation services. Summary of the Invention

[0004] The present application provides a transaction combination recommendation method, device, equipment, storage medium and program product to solve the technical problem that related technologies cannot provide accurate and personalized transaction recommendation services.

[0005] In a first aspect, the present application provides a transaction combination recommendation method, comprising: obtaining a current session sequence and user feature information of a target user, wherein the current session sequence is obtained by converting the current transaction data of the target user; inputting the current session sequence into a first prediction model, wherein the first prediction model abstracts the current session sequence into a sequence feature vector based on an attention mechanism, and generates a first recommendation result based on the sequence feature vector; obtaining a second recommendation result based on the first recommendation result and the user feature information; and determining a target transaction combination to be recommended to the target user based on the second recommendation result.

[0006] In a second aspect, the present application provides a trading combination recommendation device, comprising:

[0007] An acquisition module is used to acquire the current session sequence and user feature information of the target user, where the current session sequence is obtained by converting the current transaction data of the target user;

[0008] A generation module, configured to input the current conversation sequence into a first prediction model, wherein the first prediction model abstracts the current conversation sequence into a sequence feature vector based on an attention mechanism, and generates a first recommendation result based on the sequence feature vector;

[0009] A prediction module, configured to obtain a second recommendation result based on the first recommendation result and user feature information;

[0010] The determination module is used to determine the target transaction combination recommended to the target user based on the second recommendation result.

[0011] In a third aspect, the present application provides an electronic device comprising a processor and a memory communicatively connected to the processor; the memory stores computer-executable instructions; and the processor executes the computer-executable instructions stored in the memory to implement the transaction combination recommendation method provided in the first aspect above.

[0012] In a fourth aspect, the present application provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by a processor, they are used to implement the transaction combination recommendation method provided in the first aspect above.

[0013] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the transaction combination recommendation method provided in the first aspect above.

[0014] The transaction combination recommendation method, apparatus, device, storage medium, and program product provided in the present application obtain user feature information of a target user and a current session sequence converted based on the target user's current transaction data, input the current session sequence into a first prediction model, and enable the first prediction model to abstract the current session sequence into a sequence feature vector based on an attention mechanism, and generate a first recommendation result based on the sequence feature vector. A second recommendation result is further obtained based on the first recommendation result and the user feature information, and then a target transaction combination to be recommended to the target user is determined based on the second recommendation result. The present application predicts the current session sequence of the target user based on transaction sequence features through the first prediction model, and integrates user feature information into the first recommendation result predicted based on transaction sequence features to achieve prediction based on user features, thereby providing accurate and personalized transaction recommendation services to the target user. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0016] Figure 1 A schematic diagram of the transaction functions of the self-service terminal provided in an embodiment of the present application;

[0017] Figure 2 Schematic diagram of the process of the transaction combination recommendation method provided in the embodiment of the application Figure 1 ;

[0018] Figure 3 A schematic diagram of the structure of the first prediction model provided in an embodiment of the present application;

[0019] Figure 4 A schematic diagram of the adjacency relationship of the session graph provided in an embodiment of the present application;

[0020] Figure 5 Schematic diagram of the process of the transaction combination recommendation method provided in the embodiment of the application Figure 2 ;

[0021] Figure 6 A schematic diagram of the flow chart of the training method of the first prediction model and the second preset model provided in the embodiment of the present application;

[0022] Figure 7 A schematic diagram of the structure of the joint training of the first prediction model and the second prediction model provided in an embodiment of the present application;

[0023] Figure 8 A schematic diagram of the structure of a transaction combination recommendation device provided in an embodiment of the present application;

[0024] Figure 9 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0025] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0026] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0027] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all 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 relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0028] In addition, this application involves big data analysis of user information (including but not limited to personal biometrics, identity data, consumption data, asset data, electronic terminal operation data, etc.), and the use of artificial intelligence technology for automated decision-making, and a technical solution for making decisions that have a significant impact on personal rights and interests based on the results of automated decision-making. The application provides users with corresponding operation entrances for users to choose to agree or reject the results of automated decision-making; if the user chooses to reject, the expert decision-making process will be entered.

[0029] It should be noted that the transaction combination recommendation method, device, equipment, storage medium and product provided in this application can be used in the field of financial technology, and can also be used in any field other than the field of financial technology. This application does not limit the application field of the transaction combination recommendation method, device, equipment, storage medium and product.

[0030] In related technologies, self-service terminals typically provide related recommended transactions or cross-selling suggestions when a user completes a business transaction. This practice helps banks or other institutions increase cross-selling rates while providing users with more options and services. Currently, common recommendation algorithms for user-predicted recommended transactions include traditional recurrent neural networks (RNNs), Markov algorithms, and SR-GNN algorithms. However, these recommendation algorithms have the following problems:

[0031] 1) Traditional RNNs are prone to gradient vanishing or gradient exploding problems when processing long sequences, making it difficult to capture long-term temporal dependencies;

[0032] 2) Markov algorithms usually only consider limited historical information and cannot capture long-term dependencies. Markov models have limitations for continuous state spaces or high-dimensional data.

[0033] 3) The SR-GNN algorithm represents the user's current transaction data in the form of a graph structure. Compared with traditional RNN and Markov algorithms, the SR-GNN algorithm can fully consider the spatial proximity and correlation between each transaction. However, the SR-GNN model only has sequence features and is not combined with user feature data, resulting in insufficient personalized recommendation capabilities.

[0034] Based on the technical problems existing in the related art, the embodiment of the present application predicts the current session sequence of the target user based on the transaction sequence characteristics through a first prediction model, and integrates user feature information into the first recommendation result based on the transaction sequence feature prediction to achieve user feature-based prediction, so as to provide the target user with accurate and personalized transaction recommendation services.

[0035] The following first describes in detail the application scenarios of the embodiments of the present application.

[0036] An applicable scenario of the transaction combination recommendation method, device, equipment, storage medium and program product provided in the embodiments of the present application is to provide personalized transaction recommendation services to users at the end of business transactions when business transactions are conducted based on self-service terminals.

[0037] Figure 1 This is a diagram of the functional structure of the self-service terminal provided in the embodiment of this application. Figure 1 As shown, when a business transaction is conducted based on a self-service terminal, there are transactions of different levels in the self-service terminal, such as primary transaction, secondary transaction and tertiary transaction.

[0038] like Figure 1 As shown in , for example, the first-level transaction can be financial management, account management, etc.; the second-level transaction can be password management and personal financial management, etc.; the third-level transaction can be my financial products and personal financial management details, etc.

[0039] It is understandable that there is a certain correlation between primary transactions, secondary transactions, and tertiary transactions. For example, a primary transaction can include multiple secondary transactions, and each secondary transaction can include multiple tertiary transactions.

[0040] It can be understood that the above is only an application scenario applicable to the embodiment of the present application. The embodiment of the present application does not limit the specific application scenario, which can be determined according to actual application requirements.

[0041] The transaction combination recommendation method provided in this application is intended to solve the above technical problems in the prior art.

[0042] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0043] Figure 2 Schematic diagram of the process of the transaction combination recommendation method provided in the embodiment of the application Figure 1 .like Figure 2 As shown, the specific implementation of the transaction combination recommendation method may include the following steps:

[0044] S201 , obtaining a current conversation sequence and user characteristic information of a target user, wherein the current conversation sequence is obtained by converting current transaction data of the target user.

[0045] Exemplarily, the current transaction data of the target user may be business transaction data corresponding to a business transaction completed by the target user via a self-service terminal.

[0046] For example, a business transaction may be as described above. Figure 1 The three-level transaction shown in .

[0047] It is understood that when completing a business transaction, the target user needs to click on multiple transactions of different functional levels on the self-service terminal in sequence. That is, the target user's current transaction data includes the multiple business transactions clicked by the target user in sequence. The functional levels of these business transactions may be the same or different.

[0048] In a possible implementation, multiple business transactions included in the current transaction data of the target user are arranged in sequence according to the order in which they were clicked by the target user to obtain a current session sequence corresponding to the current transaction data.

[0049] Exemplarily, when converting the current transaction data, the tertiary transactions contained therein are converted into corresponding secondary transactions based on the mapping relationship between the tertiary transaction identity (ID) and the secondary transaction identity ID.

[0050] For example, user characteristics may include personal financial assets, age, region, and gender of the target user.

[0051] Illustratively, the user characteristic information may be a user characteristic including a weight coefficient, such as a weight coefficient of personal financial assets being 0.3.

[0052] Exemplarily, the weight coefficients of different user characteristics are determined based on empirical data and the age of the target user, and this application does not impose any limitation on this.

[0053] S202: Input the current conversation sequence into a first prediction model. The first prediction model abstracts the current conversation sequence into a sequence feature vector based on an attention mechanism, and generates a first recommendation result based on the sequence feature vector.

[0054] Exemplarily, the first prediction model may be an SR-GNN model.

[0055] First, combine Figure 3 The first prediction model will be described in detail.

[0056] Figure 3 This is a schematic diagram of the structure of the first prediction model provided in the embodiment of the present application. Figure 3 As shown, the first prediction model includes four parts: constructing a conversation graph 31, abstracting the self-service terminal transaction into a vector representation 32, abstracting the self-service terminal transaction conversation into a vector representation based on the self-attention network 33, and generating recommendation results 34.

[0057] 1) Constructing a conversation graph: Given a conversation sequence S, construct a conversation graph:

[0058] G s =(V s ,ε s )

[0059] Among them, V s represents the number of nodes in the session graph, ε s Represents the number of directed edges.

[0060] For example, in the constructed session graph, each node can be represented by v i Indicates that each node represents a transaction in the conversation sequence S. The connection between two adjacent nodes is a directed edge. That is, in the conversation sequence S, there is a directed edge between two transactions clicked by the user in sequence. The directed edge can be expressed as (v i-1 ,v i ).

[0061] For example, Figure 3 In the session graph 310 shown in FIG, the session sequence includes (v1, v2, v4, v3), (v2, v5, v6, v7), and (v5, v4, v3, v6), etc., and the arrows represent directed edges consisting of two transactions clicked in sequence.

[0062] 2) Abstract the self-service terminal transaction into a vector representation:

[0063] Graph Neural Networks (GNN) are used to learn a unified representation of all nodes in the above session graph. The main calculation propagation formula is:

[0064]

[0065] Among them, the connection matrix A is used in formula (1) s Integrate information from adjacent nodes and connect the matrix A s Determines how nodes in the session graph are connected. s Two adjacency matrices and Spliced ​​together, A s,i Represents the node v s,i respectively and The two corresponding columns.

[0066] Formula (2) to Formula (5) are similar to the calculation process of the gated recurrent unit (GRU), where Formula (2) Indicates the control of forgetting information, formula (3) Indicates the control of newly generated information, in formula (4) Indicates the decision of which historical information to generate new information from, Represents the newly generated information, in formula (5) Indicates which historical information to choose to forget. Indicates which new information to remember. Indicates the final updated node status.

[0067] Figure 4 This is a schematic diagram of the adjacency relationship of the session graph provided in the embodiment of the present application. Figure 4 As shown, the corresponding session sequence in the session graph is (v1, v2, v3, v2, v4) where the connection matrix A s Two adjacency matrices and Spliced ​​together.

[0068] For example, the connection matrix A s ∈R n×2n .

[0069] 3) Based on the self-attention network, the self-service terminal transaction session is abstracted into a vector representation:

[0070] For each transaction in a self-service terminal, an attention mechanism is used to combine global and current preferences to represent each transaction. The current preference can be represented by a local embedding, which represents the last node in the transaction sequence. For example, the current preference is used to emphasize the influence of the last clicked transaction.

[0071] For example, the current preference can be expressed by the following formula:

[0072] S1=v n

[0073] Among them, S1 represents the current preference.

[0074] Global preferences are captured through a self-attention network, which can be represented by a global embedding. The attention network assigns a weight coefficient to each transaction in a session, which represents the importance of each transaction to the entire session.

[0075] For example, the weight coefficient can be expressed by the following formula:

[0076] a i =q T σ(W1v n +W2v i +c)

[0077] Among them, a i Represents the weight coefficient, q, W1 and W2 control the weight.

[0078] For example, the global preference can be expressed by the following formula:

[0079]

[0080] Combine the current preference S1 with the global preference S g Splice together to get the final session.

[0081] For example, the final session can be expressed by the following formula:

[0082] S h =W3[S1;S g ]

[0083] Among them, S h Indicates the final session.

[0084] 4) Generate recommendation results:

[0085] The final session S h The dot product is performed with each transaction to calculate the recommendation score of each transaction, and the probability of each transaction is obtained through the softmax function.

[0086] For example, the probability of each transaction can be expressed by the following formula:

[0087]

[0088] Illustratively, the first recommendation result may include multiple recommended transactions, wherein each recommended transaction has a different recommendation score.

[0089] S203: Obtain a second recommendation result based on the first recommendation result and the user characteristic information.

[0090] In a possible implementation, the first recommendation result is fused with user feature information to obtain a second recommendation result.

[0091] For example, the second recommendation result may include multiple different recommended transactions. The recommended transactions in the second recommendation result may be partially the same as those in the first recommendation result, or may be completely different.

[0092] S204: Determine a target transaction combination to be recommended to the target user based on the second recommendation result.

[0093] Illustratively, the target transaction combination may include multiple different recommended transactions belonging to secondary functional transactions.

[0094] For example, the number of recommended transactions in the target transaction combination may be 3-4. The transaction combination recommendation method provided in the embodiment of the present application does not limit the number of recommended transactions in the target transaction combination, and can be determined according to actual application requirements.

[0095] For example, when a target user completes a business transaction based on a self-service terminal, the target transaction combination is automatically displayed on the user interface of the self-service terminal to automatically recommend personalized transactions to the target user.

[0096] In an embodiment of the present application, user feature information of a target user is obtained, as well as a current conversation sequence converted from the target user's current transaction data. The current conversation sequence is input into a first prediction model, so that the first prediction model abstracts the current conversation sequence into a sequence feature vector based on an attention mechanism, and generates a first recommendation result based on the sequence feature vector. A second recommendation result is further obtained based on the first recommendation result and the user feature information, and then a target transaction combination recommended to the target user is determined based on the second recommendation result. In an embodiment of the present application, the current conversation sequence of the target user is predicted based on transaction sequence features by the first prediction model, and user feature information is integrated into the first recommendation result predicted based on transaction sequence features to achieve a prediction based on user features, thereby providing the target user with accurate and personalized transaction recommendation services.

[0097] Optionally, a possible implementation method of obtaining the second recommendation result based on the first recommendation result and user feature information in step S203 may be: inputting the first recommendation result and user feature information into a second prediction model to perform recommendation transaction prediction, and obtaining a second recommendation result output by the second prediction model.

[0098] Exemplarily, the second prediction model may be any one of a k-nearest neighbor (KNN) classification algorithm, a random forest algorithm, and a support vector machine.

[0099] The embodiment of the present application integrates user feature information through the second prediction model to achieve personalized transaction recommendations.

[0100] Optionally, a possible implementation method of inputting the first recommendation result and user characteristic information into the second prediction model for recommendation transaction prediction may be: filtering the first recommendation result to obtain a third recommendation result, the third recommendation result containing at least two first recommended transactions; inputting the third recommendation result and user characteristic information into the second prediction model for recommendation transaction prediction.

[0101] In one possible implementation, the multiple recommended transactions included in the first recommendation result are arranged in descending order according to the recommendation score corresponding to each recommended transaction, and at least two recommended transactions with higher recommendation scores are selected as the third recommendation result.

[0102] The following combination Figure 5 The specific implementation method of step S204 of determining the target transaction combination recommended to the target user based on the second recommendation result is described in detail.

[0103] Figure 5 Schematic diagram of the process of the transaction combination recommendation method provided in the embodiment of the application Figure 2 .like Figure 5 As shown, according to the second recommendation result, a specific implementation method of determining the target transaction combination recommended to the target user may include the following steps:

[0104] S501 : Selecting a second recommended transaction with the highest recommendation probability from a plurality of second recommended transactions included in the second recommendation result as a target recommended transaction.

[0105] Illustratively, the second recommendation result includes multiple different second recommended transactions and a recommendation probability corresponding to each second recommended transaction.

[0106] S502: Determine whether the at least two first recommended transactions include a target recommended transaction.

[0107] Exemplarily, it is determined whether the target recommended transaction is the same as any first recommended transaction in at least two first recommended transactions. If any first recommended transaction is the same as the target recommended transaction, it means that the target recommended transaction is included in at least two first recommended transactions; if no first recommended transaction is the same as the target recommended transaction, it means that the target recommended transaction is not included in at least two first recommended transactions.

[0108] In this step, when it is determined that the target recommended transaction is included in the at least two first recommended transactions, step S503 is executed; when it is determined that the target recommended transaction is not included in the at least two first recommended transactions, step S504 is executed.

[0109] S503 , selecting a second recommended transaction with the second highest recommendation probability from the multiple second recommended transactions included in the second recommendation result as a target recommended transaction, and repeating the steps to determine whether the target recommended transaction is included in at least two first recommended transactions.

[0110] It can be understood that when selecting the target recommended transaction from the second recommendation results, the recommendation probability corresponding to each second recommended transaction included in the second recommendation results is selected in descending order, and the selected second recommended transaction that is not included in at least two first recommended transactions is determined as the target recommended transaction.

[0111] Exemplarily, when the second recommended transaction corresponding to the maximum recommendation probability is included in at least two first recommended transactions, the second recommended transaction ranked second in descending order of recommendation probability is selected from the second recommendation results as the target recommended transaction, and it is determined whether the current target recommended transaction is included in at least two first recommended transactions. Further, when it is included in at least two first recommended transactions, the second recommended transaction ranked third in descending order of recommendation probability is selected from the second recommendation results as the target recommended transaction, and so on, until the selected target recommended transaction is not included in at least two first recommended transactions.

[0112] S504: Determine the target recommended transaction and at least two first recommended transactions as a target transaction combination.

[0113] In an embodiment of the present application, the second recommended transaction with the highest recommendation probability is selected as the target recommended transaction from the multiple second recommended transactions included in the second recommendation result, and it is determined whether the target recommended transaction is included in at least two first recommended transactions. If it is included, the second recommended transaction with the second highest recommendation probability is selected as the target recommended transaction from the multiple second recommended transactions included in the second recommendation result. The steps are repeated to determine whether the target recommended transaction is included in at least two first recommended transactions. If it is not included, the target recommended transaction and at least two first recommended transactions are determined to be a target transaction combination. In an embodiment of the present application, at least two first recommended transactions with higher recommendation scores in the first recommendation result obtained by the first prediction model based on the conversation sequence features and the second recommended transaction with higher recommendation probability and different from the first recommendation in the second recommendation result obtained by the second prediction model by fusing user features are used as the target transaction combination, so as to achieve personalized recommendation services while ensuring the accuracy of the recommended transactions.

[0114] Optionally, the user characteristic information is obtained by: obtaining an identity identifier of a target user; determining a customer group corresponding to the target user based on the identity identifier; and obtaining user characteristic information corresponding to the customer group based on the customer group.

[0115] To better integrate user characteristics and implement personalized transaction portfolio recommendations for users, one possible implementation method is to group users based on their age, thereby obtaining customer groups of different age groups. Furthermore, for each customer group, different weight coefficients are added to each user characteristic within the customer group, such as financial assets and gender, to obtain user characteristic information.

[0116] For example, the customer groups may include teenagers, young adults, middle-aged people, middle-aged and elderly people.

[0117] For example, users aged between 0 and 7 can be classified as teenagers, users aged between 7 and 18 can be classified as young people, users aged between 18 and 46 can be classified as middle-aged, users aged between 46 and 60 can be classified as middle-aged and elderly, and users aged over 60 can be classified as elderly.

[0118] For example, user feature information can be expressed by the following formula:

[0119]

[0120] Among them, α 1,1 etc. represent different weight coefficients, and X1 etc. represent user features.

[0121] The following combination Figure 6 The training methods of the first prediction model and the second prediction model in the transaction combination recommendation method provided in the embodiment of the present application are described in detail.

[0122] Figure 6 Schematic diagram of the flow of the first prediction model and the second preset model training method provided in the embodiment of the present application. Figure 6 As shown, the implementation method of the training method of the first prediction model and the second preset model specifically includes the following steps:

[0123] S601, obtaining the user's historical transaction data and user characteristic information.

[0124] Exemplarily, historical transaction data of the user is obtained from a transaction database based on self-service terminal transactions.

[0125] For example, the historical transaction data can be Figure 1 Detailed transaction data for level 3 transactions and above as shown in .

[0126] The method of obtaining user feature information is similar to the above and will not be repeated here.

[0127] S602: Preprocess the historical transaction data to obtain multiple session sequences. The preprocessing includes screening and conversion.

[0128] Exemplarily, the screening process may be to filter out transaction data corresponding to each user in the historical transaction data, data in which the user uses the self-service terminal only once for a business transaction, and transaction data in which the cumulative number of clicks is less than 5.

[0129] Exemplarily, the conversion process may be to convert each transaction data of the user into corresponding secondary transactions first according to the mapping relationship between the tertiary transaction ID and the secondary transaction ID, and then convert the transaction data into a session sequence according to the order of the business transactions clicked in the transaction data.

[0130] Table 1 shows the data preprocessing results provided in the examples of this application.

[0131] Table 1

[0132] Serial number Session Sequence Transaction Tags User ID financial assets gender age 0 [1,2,3,4,5] 1 0 3 2.0 6 1 [1,2,3,4] 5 0 3 2.0 6 2 [1,2,3] 4 0 3 2.0 6 3 [1,2] 3 0 3 2.0 6 4 [1] 2 0 3 2.0 6

[0133] As shown in Table 1, the transaction data includes the user's age, gender, financial assets, user ID, transaction tags, and session sequence.

[0134] As shown in Table 1, transaction tags have certain associations. For example, the transaction tag associated with transaction tag 1 is transaction tag 2, the transaction tag associated with transaction tag 2 is transaction tag 3, the transaction tag associated with transaction tag 3 is transaction tag 4, and so on. That is, if the transaction tag corresponding to the business transaction of the previous click in the session sequence is transaction tag 1, then the transaction tag of the next click in the session sequence is transaction tag 2. If the transaction tag corresponding to the business transaction of the previous click in the session sequence is transaction tag 2, then the transaction tag of the next click in the session sequence is transaction tag 3, and so on.

[0135] S603: For each conversation sequence in the multiple conversation sequences, input the conversation sequence into the first prediction model to perform recommended transaction prediction, obtain a first predicted transaction output by the first prediction model, and based on a first loss function, obtain a first loss value between the first predicted transaction and the true label, and adjust the model parameters of the first prediction model according to the first loss value.

[0136] The first prediction model is similar to the above one and will not be described in detail here.

[0137] Exemplarily, each conversation sequence corresponds to a conversation graph, and the first loss function of each conversation graph may be the cross entropy between the predicted value and the true value.

[0138] Exemplarily, the first loss function can be expressed by the following formula:

[0139]

[0140] in, represents cross entropy, y i represents the real transaction corresponding to the i-th session sequence, represents the predicted transaction corresponding to the i session sequence.

[0141] S604: Input the first predicted transaction and user feature information into the second prediction model to perform recommended transaction prediction, obtain a second predicted transaction output by the second prediction model, and based on a second loss function, obtain a second loss value between the second predicted transaction and the true label, and adjust the model parameters of the second prediction model according to the second loss value.

[0142] It can be understood that in the transaction combination recommendation method provided in the embodiment of the present application, the first prediction model and the second prediction model are obtained by joint training.

[0143] Figure 7 This is a schematic diagram of the structure of the joint training of the first prediction model and the second prediction model provided in the embodiment of the present application. Figure 7 As shown, the first prediction model 71 is trained by the conversation sequence obtained based on the user's historical transaction data, and the first predicted transaction and user feature information output by the first prediction model 71 are input into the second prediction model 72 to train the second prediction model 72.

[0144] In a possible implementation, the top three predicted transactions with the highest prediction scores in the first predicted transaction and user feature information may be concatenated and input into the second prediction model to train the second prediction model.

[0145] Exemplarily, the second loss function is calculated based on the second predicted transaction and the actual transaction labels of the transactions in the conversation sequence. The actual transaction labels are similar to the transaction labels shown in Table 1 above and are not described here in detail.

[0146] The second prediction model is similar to the above one and will not be described in detail here.

[0147] Compared to related technologies, during model training, user characteristics are often screened for a fixed number of characteristics, such as gender, age, and region. However, different groups of people place different importance on these characteristics. In the transaction combination recommendation method provided in the embodiments of the present application, by grouping users, adding appropriate weights to corresponding user characteristics for different customer groups, and then training the second prediction model based on the obtained user characteristic information, the prediction accuracy of the model can be improved.

[0148] In an embodiment of the present application, by obtaining a user's historical transaction data and user feature information, filtering and converting the historical transaction data, a plurality of conversation sequences are obtained. For each conversation sequence in the plurality of conversation sequences, the conversation sequence is input into a first prediction model for transaction recommendation prediction, a first predicted transaction output by the first prediction model is obtained, and based on a first loss function, a first loss value between the first predicted transaction and the true label is obtained, and the model parameters of the first prediction model are adjusted according to the first loss value; the first predicted transaction and user feature information are input into a second prediction model for transaction recommendation prediction, a second predicted transaction output by the second prediction model is obtained, and based on a second loss function, a second loss value between the second predicted transaction and the true label is obtained, and the model parameters of the second prediction model are adjusted according to the second loss value. In an embodiment of the present application, by using historical transaction data to train the first prediction model, the accuracy of the first prediction model for predicting recommended transactions based on sequence features is improved, and the output results of the first prediction model are further integrated with user feature information to train the second prediction model, thereby improving the second prediction model's predictive ability for personalized transaction recommendations.

[0149] In summary, in the transaction combination recommendation method provided in the embodiment of the present application, on the one hand, the first prediction model can be used to extract implicit features between transactions from complex historical self-service terminal transaction data, realize prediction based on transaction sequence, better capture user combination transaction needs, and provide accurate transaction recommendation services; on the other hand, the second prediction model is used to integrate user feature information related to the user's self-service transaction terminal to realize personalized combination recommendation for self-service terminal transactions. In summary, it can improve the user experience and recognition of business transactions based on self-service terminals, reduce the limitations of the use of self-service terminals, and thus improve the utilization rate of self-service terminals, and promote the development of financial services towards digitalization and intelligence.

[0150] The following are device embodiments of the present application, which can be used to implement the method embodiments of the present application. For details not disclosed in the device embodiments of the present application, please refer to the method embodiments of the present application.

[0151] Figure 8 This is a schematic diagram of the structure of the transaction combination recommendation device provided in the embodiment of the present application. Figure 8 As shown, the transaction combination recommendation device 80 includes an acquisition module 810 , a generation module 820 , a prediction module 830 and a determination module 840 .

[0152] The acquisition module 810 is used to acquire the current session sequence and user feature information of the target user. The current session sequence is obtained by converting the current transaction data of the target user.

[0153] A generation module 820 is configured to input the current conversation sequence into a first prediction model, which abstracts the current conversation sequence into a sequence feature vector based on an attention mechanism and generates a first recommendation result based on the sequence feature vector;

[0154] Prediction module 830, configured to obtain a second recommendation result based on the first recommendation result and user characteristic information;

[0155] The determination module 840 is configured to determine a target transaction combination to be recommended to the target user based on the second recommendation result.

[0156] In a possible implementation, the prediction module 830 is specifically configured to input the first recommendation result and user feature information into a second prediction model to perform recommendation transaction prediction, and obtain a second recommendation result output by the second prediction model.

[0157] In one possible implementation, the prediction module 830 is further used to: filter the first recommendation result to obtain a third recommendation result, where the third recommendation result includes at least two first recommended transactions; and input the third recommendation result and user feature information into the second prediction model to perform recommended transaction prediction.

[0158] In one possible implementation, the determination module 840 is specifically used to: select the second recommended transaction with the highest recommendation probability as the target recommended transaction from the multiple second recommended transactions included in the second recommendation result; determine whether the target recommended transaction is included in at least two first recommended transactions; if included, select the second recommended transaction with the second highest recommendation probability as the target recommended transaction from the multiple second recommended transactions included in the second recommendation result, and repeat the steps to determine whether the target recommended transaction is included in at least two first recommended transactions; if not included, determine the target recommended transaction and at least two first recommended transactions as the target transaction combination.

[0159] In a possible implementation, the user characteristic information is obtained by: obtaining an identity identifier of a target user; determining a customer group corresponding to the target user based on the identity identifier; and obtaining user characteristic information corresponding to the customer group based on the customer group.

[0160] In one possible implementation, the first prediction model and the second preset model are trained in the following manner: obtaining the user's historical transaction data and user feature information; preprocessing the historical transaction data to obtain multiple conversation sequences, the preprocessing including screening and conversion; for each conversation sequence in the multiple conversation sequences, inputting the conversation sequence into the first prediction model for recommended transaction prediction, obtaining a first predicted transaction output by the first prediction model, and based on the first loss function, obtaining a first loss value between the first predicted transaction and the true label, and adjusting the model parameters of the first prediction model according to the first loss value; inputting the first predicted transaction and user feature information into the second prediction model for recommended transaction prediction, obtaining a second predicted transaction output by the second prediction model, and based on the second loss function, obtaining a second loss value between the second predicted transaction and the true label, and adjusting the model parameters of the second prediction model according to the second loss value.

[0161] The transaction combination recommendation device provided in this embodiment can be used to execute the method steps of the above method embodiment. The specific implementation method and technical effects are similar and will not be repeated here.

[0162] Figure 9 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present application. Figure 9 As shown, the electronic device 90 includes: at least one processor 910, a memory 920, a communication interface 930, and a system bus 940. The memory 920 and the communication interface 930 are connected to the processor 910 via the system bus 940 and communicate with each other. The memory 920 is used to store instructions, the communication interface 930 is used to communicate with other devices, and the processor 910 is used to call instructions in the memory to execute the method steps provided in the above method embodiment. The specific implementation methods and technical effects are similar and will not be repeated here.

[0163] Should Figure 9 The system bus 940 mentioned in the figure can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The system bus 940 can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, the figure shows only one thick line, but this does not mean that there is only one bus or only one type of bus.

[0164] The communication interface 930 is used to implement communication between the database access apparatus and other devices (such as clients, read-write libraries, and read-only libraries).

[0165] The memory 920 may include a random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0166] The processor 910 can be a general-purpose processor, including a central processing unit, a network processor (NP), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, and discrete hardware components.

[0167] An embodiment of the present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the method steps in the above-mentioned method embodiment. The specific implementation method and technical effect are similar and will not be repeated here.

[0168] The present application also provides a computer program product comprising computer-executable instructions. When the computer-executable instructions are executed, the method steps in the above method embodiment are implemented. The specific implementation methods and technical effects are similar and will not be described in detail here.

[0169] It should be noted that for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all optional embodiments, and the actions and modules involved are not necessarily required by this application.

[0170] It should be further noted that, although the various steps in the flowchart are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps may be performed in other orders. Moreover, at least a portion of the steps in the flowchart may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily performed at the same time, but may be performed at different times. The execution order of these sub-steps or stages is not necessarily to be performed in sequence, but may be performed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0171] It should be understood that the above-described device embodiments are merely illustrative, and the device of the present application may also be implemented in other ways. For example, the division of units / modules in the above-described embodiments is merely a logical functional division, and actual implementations may employ other division methods. For example, multiple units, modules, or components may be combined or integrated into another system, or some features may be omitted or not implemented.

[0172] In addition, unless otherwise specified, the functional units / modules in the various embodiments of the present application may be integrated into a single unit / module, each unit / module may exist physically separately, or two or more units / modules may be integrated together. The aforementioned integrated units / modules may be implemented in the form of hardware or software program modules.

[0173] If the integrated unit / module is implemented in hardware, the hardware may be digital circuits, analog circuits, etc. The physical implementation of the hardware structure includes, but is not limited to, transistors, memristors, etc. Unless otherwise specified, the processor may be any appropriate hardware processor, such as a CPU, GPU, FPGA, DSP, and ASIC. Unless otherwise specified, the storage unit may be any appropriate magnetic storage medium or magneto-optical storage medium, such as resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (EDRAM), high-bandwidth memory (HBM), hybrid memory cube (HMC), etc.

[0174] If the integrated unit / module is implemented in the form of a software program module and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a memory and includes a number of instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned memory includes various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0175] In the above embodiments, the description of each embodiment has its own emphasis. For parts not described in detail in a particular embodiment, please refer to the relevant description of other embodiments. The technical features of the above embodiments can be combined in any way. To keep the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0176] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present application that follow the general principles of the present application and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0177] It should be understood that the present application is not limited to the exact structure described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A trading combination recommendation method, characterized in that: include: Acquire a current session sequence and user feature information of a target user, wherein the current session sequence is obtained by converting current transaction data of the target user; Inputting the current conversation sequence into a first prediction model, the first prediction model abstracting the current conversation sequence into a sequence feature vector based on an attention mechanism, and generating a first recommendation result based on the sequence feature vector, wherein the first prediction model is an SR-GNN model; Obtaining a second recommendation result based on the first recommendation result and the user characteristic information; determining a target transaction combination to be recommended to the target user based on the second recommendation result; Obtaining a second recommendation result based on the first recommendation result and the user characteristic information includes: The first recommendation result and the user feature information are input into a second prediction model to perform recommendation transaction prediction, and a second recommendation result output by the second prediction model is obtained. The second prediction model is any one of a k-nearest neighbor classification algorithm, a random forest algorithm, and a support vector machine.

2. The transaction combination recommendation method according to claim 1, characterized in that: Inputting the first recommendation result and the user characteristic information into a second prediction model to perform recommendation transaction prediction includes: Filtering the first recommendation result to obtain a third recommendation result, wherein the third recommendation result includes at least two first recommended transactions; The third recommendation result and the user characteristic information are input into a second prediction model to perform recommendation transaction prediction.

3. The transaction combination recommendation method according to claim 2, characterized in that: Determining the target transaction combination recommended to the target user based on the second recommendation result includes: Selecting, from the plurality of second recommended transactions included in the second recommendation result, the second recommended transaction with the highest recommendation probability as the target recommended transaction; determining whether the at least two first recommended transactions include the target recommended transaction; If yes, then selecting the second recommended transaction with the second highest recommendation probability from the multiple second recommended transactions included in the second recommendation result as the target recommended transaction, and repeating the steps to determine whether the target recommended transaction is included in the at least two first recommended transactions; If not included, the target recommended transaction and the at least two first recommended transactions are determined to be the target transaction combination.

4. The transaction combination recommendation method according to any one of claims 1 to 3, characterized in that: The user characteristic information is obtained in the following manner: Obtain the target user's identity; Determining the customer group corresponding to the target user according to the identity identifier; According to the customer group, user characteristic information corresponding to the customer group is obtained.

5. The transaction combination recommendation method according to any one of claims 1 to 3, characterized in that: The first prediction model and the second prediction model are obtained by training in the following manner: Obtain users' historical transaction data and user feature information; Preprocessing the historical transaction data to obtain multiple session sequences, wherein the preprocessing includes screening and conversion; For each of the plurality of conversation sequences, input the conversation sequence into a first prediction model to perform transaction recommendation prediction, obtain a first predicted transaction output by the first prediction model, obtain a first loss value between the first predicted transaction and a true label based on a first loss function, and adjust model parameters of the first prediction model according to the first loss value; The first predicted transaction and the user feature information are input into a second prediction model to perform recommended transaction prediction, obtaining a second predicted transaction output by the second prediction model, and based on a second loss function, obtaining a second loss value between the second predicted transaction and the true label, and adjusting the model parameters of the second prediction model according to the second loss value.

6. A trading combination recommendation device, characterized in that: include: An acquisition module, configured to acquire a current session sequence and user feature information of a target user, wherein the current session sequence is obtained by converting current transaction data of the target user; a generation module, configured to input the current conversation sequence into a first prediction model, the first prediction model abstracting the current conversation sequence into a sequence feature vector based on an attention mechanism, and generating a first recommendation result based on the sequence feature vector, wherein the first prediction model is an SR-GNN model; a prediction module configured to obtain a second recommendation result based on the first recommendation result and the user characteristic information; the prediction module is specifically configured to input the first recommendation result and the user characteristic information into a second prediction model to perform recommendation transaction prediction, thereby obtaining a second recommendation result output by the second prediction model; the second prediction model being any one of a k-nearest neighbor classification algorithm, a random forest algorithm, and a support vector machine; A determination module is used to determine a target transaction combination to be recommended to the target user based on the second recommendation result.

7. An electronic device, characterized in that: include: a processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 5 when executed by a processor.

9. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 5 when executed by a processor.

Images