A fine-grained financial management style recognition method and device, electronic equipment and medium

Abstract

The application discloses a fine-grained financial style recognition method and device, electronic equipment and medium; the method comprises the following steps: dividing the transaction data of the current customer into M equal-length data segments; wherein M is a natural number greater than 1; constructing a corresponding frequency diagram containing P fund category attributes based on the M equal-length data segments; wherein P is a natural number greater than 1; inputting the frequency diagram into a pre-trained financial style recognition model to obtain the fine-grained financial style of the current user during the corresponding period through the financial style recognition model; wherein the financial style recognition model comprises a graph attention layer, a flattening layer, a full connection layer and a classifier. The embodiment of the application can predict the financial style of the customer in a fine-grained manner, thereby providing more timely and more personalized services for the customer.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a fine-grained financial style recognition method, device, electronic device, and medium. Background Technology

[0002] With the sustained and stable growth of the national economy and national income, simple savings can no longer meet the needs of residents for currency appreciation and preservation. Commercial banks' personal wealth management services are attracting increasing attention. Investment style is a complex combination of a client's investment behavior and habits during wealth management transactions. As a long-term research topic in the field of wealth management behavior analysis, investment style identification is an issue that cannot be ignored. Understanding clients' wealth management behavior is of significant practical importance for the sound development of the banking wealth management market and for improving clients' investment efficiency. For example, financial professionals can rely on the analysis of clients' historical wealth management habits to identify suitable investment projects and tailor effective long-term investment strategies for clients.

[0003] In recent years, with the rapid development of mobile internet and IoT technologies, the generation of massive amounts of transaction data from bank wealth management products has provided important data resources for the analysis of wealth management behavior. Based on this, almost all previous studies on wealth management style identification have emphasized the entire customer granularity, which may lead to the identified wealth management styles failing to accurately reflect the complex wealth management behaviors that change over time.

[0004] Currently, traditional methods for identifying financial styles still primarily rely on questionnaires to directly determine a client's investment style. This method has several limitations, including high survey costs, heavy burden on respondents, and potentially unreliable results. More seriously, a client's financial style can change over time due to various factors such as income, occupation, and social experience. Summary of the Invention

[0005] This application provides a fine-grained financial style identification method, device, electronic device, and medium, which can predict customers' financial style in a fine-grained manner, thereby providing customers with more timely and personalized services.

[0006] In a first aspect, embodiments of this application provide a fine-grained financial style identification method, the method comprising:

[0007] Divide the current customer's transaction data into M equal-length data segments; where M is a natural number greater than 1.

[0008] Based on the M equal-length data segments, construct a corresponding frequency map containing P fund category attributes; where P is a natural number greater than 1;

[0009] The frequency graph is input into a pre-trained financial style recognition model, and the fine-grained financial style of the current user during the corresponding period is obtained through the financial style recognition model; wherein, the financial style recognition model includes: a graph attention layer, a flattening layer, a fully connected layer and a classifier.

[0010] Secondly, embodiments of this application also provide a fine-grained financial style recognition device, the device comprising: a segmentation module, a construction module, and a recognition module; wherein,

[0011] The segmentation module is used to divide the current customer's transaction data into M data segments of equal length; where M is a natural number greater than 1.

[0012] The construction module is used to construct a frequency map containing P fund category attributes based on the M equal-length data segments; where P is a natural number greater than 1.

[0013] The recognition module is used to input the frequency graph into a pre-trained financial style recognition model, and obtain the fine-grained financial style of the current user during the corresponding period through the financial style recognition model; wherein, the financial style recognition model includes: a graph attention layer, a flattening layer, a fully connected layer and a classifier.

[0014] Thirdly, embodiments of this application provide an electronic device, including:

[0015] One or more processors;

[0016] Memory, used to store one or more programs.

[0017] When the one or more programs are executed by the one or more processors, the one or more processors implement the fine-grained financial style recognition method described in any embodiment of this application.

[0018] Fourthly, embodiments of this application provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the fine-grained financial style recognition method described in any embodiment of this application.

[0019] This application proposes a fine-grained financial style identification method, apparatus, electronic device, and medium. First, the current customer's transaction data is divided into M equal-length data segments. Then, a frequency graph containing P fund category attributes is constructed based on the M equal-length data segments. The frequency graph is then input into a pre-trained financial style identification model to obtain the current user's fine-grained financial style for the corresponding period. The financial style identification model includes a graph attention layer, a flattening layer, a fully connected layer, and a classifier. In other words, the technical solution of this application can use the financial style identification model to identify the current customer's financial style in a fine-grained manner, thereby obtaining the current customer's fine-grained financial style without requiring financial professionals to analyze the customer's historical financial behavior habits. It is not limited by constraints such as survey costs, respondent burden, and unreliable survey results. In contrast, existing financial style identification methods still mainly rely on questionnaires to directly determine the customer's financial style, which is usually subject to various constraints and cannot accurately predict the user's financial style. Therefore, compared with the prior art, the fine-grained financial style identification method, device, electronic device and medium proposed in this application can predict customers' financial style in a fine-grained manner, thereby providing customers with more timely and personalized services; moreover, the technical solution of this application is simple and convenient to implement, easy to popularize, and has a wider range of applications. Attached Figure Description

[0020] Figure 1 A schematic diagram of the first process of the fine-grained financial style recognition method provided in this application embodiment;

[0021] Figure 2 A schematic diagram of the second process of the fine-grained financial style recognition method provided in the embodiments of this application;

[0022] Figure 3 A schematic diagram of the third process of the fine-grained financial style recognition method provided in the embodiments of this application;

[0023] Figure 4 A schematic diagram of the structure of the fine-grained financial style recognition device provided in the embodiments of this application;

[0024] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0025] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the application and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present application, not the entire structure.

[0026] Example 1

[0027] Figure 1 This is a first flowchart illustrating the fine-grained financial style recognition method provided in this application. This method can be executed by a fine-grained financial style recognition device or electronic device, which can be implemented in software and / or hardware, and can be integrated into any smart device with network communication capabilities. Figure 1 As shown, the fine-grained financial style identification method may include the following steps:

[0028] S101. Divide the current customer's transaction data into M data segments of equal length; where M is a natural number greater than 1.

[0029] In this step, the electronic device can divide the current customer's transaction data into M equal-length data segments; where M is a natural number greater than 1. This application, while protecting customer security and privacy information (such as customer name, ID number, and address), aims to identify a customer's investment style within a short-term investment transaction period (i.e., fine-grained). Based on universally applicable and anonymized fund transaction data, it divides the customer's transaction data into a series of continuous and equal-length data segments (i.e., data samples) using a data segmentation method. Each segment has a length of L, measured in time, such as one month. Then, a small portion of the sample set is manually labeled with the corresponding investment style type.

[0030] S102. Construct a frequency map containing P fund category attributes based on M data segments of equal length; where P is a natural number greater than 1.

[0031] In this step, the electronic device can construct a frequency graph containing P fund category attributes based on M data segments of equal length; where P is a natural number greater than 1. The fine-grained financial behavior representation in this application embodiment is the foundation of fine-grained financial style recognition. This application designs an effective representation form (i.e., a frequency graph) to quantify the complex and ever-changing financial behavior information of customers, achieving a coarse correlation between fund transaction data and financial behavior, and characterizing the dynamic financial behavior of customers. The specific construction process includes transaction status calculation and frequency graph construction.

[0032] Investment style, as a semantic description and analysis of investment behavior, refers to the complex combination of a client's investment behavior and habits during investment transactions. This application defines investment style within a short-term investment transaction period (i.e., fine-grained), for example, one month. Investment styles are divided into five levels according to risk tolerance, from weakest to strongest: conservative, cautious, moderate, aggressive, and volatile. It is worth noting that in reality, a client's overall investment style is a growing vector composed of these five investment styles that evolves over time. For example, if a client invests in a volatile manner most of the time (i.e., most elements in the vector are volatile), then the client can be considered aggressive. Therefore, the fine-grained investment style recognition algorithm proposed in this application can be easily extended to solve previous investment behavior analysis problems.

[0033] S103. Input the frequency graph into the pre-trained financial style recognition model to obtain the fine-grained financial style of the current user during the corresponding period through the financial style recognition model; wherein, the financial style recognition model includes: graph attention layer, flattening layer, fully connected layer and classifier.

[0034] In this step, the electronic device can input the frequency map into a pre-trained financial style recognition model to obtain the fine-grained financial style of the current user during the corresponding period. The financial style recognition model includes a graph attention layer, a flattening layer, a fully connected layer, and a classifier.

[0035] Preferably, the financial style recognition model in this embodiment can be a graph attention network. As a graph representation learning technique, the graph attention network effectively captures the relationships between nodes and their neighbors from complex graph structure data through an attention mechanism. The attention mechanism allows the learning process to focus on the parts of the graph structure data more relevant to the specific task, avoiding interference from noisy parts of the graph structure data, thereby helping the model make better decisions. Graph attention networks have the following two characteristics: 1) The graph attention layer can perform parallel computation on the output features of all edges and all nodes, and does not require complex matrix operations (such as feature decomposition) during the computation process, thus making the network computationally efficient. 2) By assigning different weights to different neighbor nodes, the graph attention layer can be applied to graph nodes of different degrees and does not require prior knowledge of the graph's structure; therefore, this network is directly applicable to inductive learning problems.

[0036] The fine-grained investment style identification method proposed in this application first divides the current customer's transaction data into M equal-length data segments; then, it constructs a frequency graph containing P fund category attributes based on the M equal-length data segments; finally, it inputs the frequency graph into a pre-trained investment style identification model, and obtains the fine-grained investment style of the current user for the corresponding period through the investment style identification model. The investment style identification model includes a graph attention layer, a flattening layer, a fully connected layer, and a classifier. In other words, the technical solution of this application can use the investment style identification model to identify the current customer's investment style in a fine-grained manner, thereby obtaining the current customer's fine-grained investment style without requiring financial professionals to rely on the customer's historical investment behavior habits for analysis, and is not limited by constraints such as survey costs, respondent burden, and unreliable survey results. In contrast, in existing technologies, traditional investment style identification methods still mainly rely on questionnaires to directly determine the customer's investment style, which is usually subject to various constraints and cannot accurately predict the user's investment style. Therefore, compared with the prior art, the financial style identification method proposed in this application can predict customers' financial style in a fine-grained manner, thereby providing customers with more timely and personalized services; moreover, the technical solution of this application is simple and convenient to implement, easy to popularize, and has a wider range of applications.

[0037] Example 2

[0038] Figure 2 This is a schematic diagram of the second process of the fine-grained financial style recognition method provided in this application embodiment. Further optimizations and extensions are possible based on the above technical solution, and it can be combined with the various optional implementation methods described above. For example... Figure 2 As shown, the fine-grained financial style identification method may include the following steps:

[0039] S201. Divide the current customer's transaction data into M data segments of equal length; where M is a natural number greater than 1.

[0040] S202. Calculate the corresponding current customer's transaction status sequence based on M data segments of equal length.

[0041] S203. Construct a frequency graph containing P types of fund category attributes based on the current user's transaction status sequence.

[0042] In this step, the electronic device can construct a frequency graph containing P fund category attributes based on the current user's transaction state sequence. Specifically, the electronic device can first calculate the corresponding current customer's transaction state sequence based on M equal-length data segments; then, it can construct a frequency graph containing P fund category attributes based on the current user's transaction state sequence.

[0043] Furthermore, when calculating the corresponding current customer's transaction status sequence, the electronic device can first extract the transaction records for each period from each of the M equal-length data segments; then, it can calculate the current customer's transaction status based on the transaction records for each period within each data segment; and based on the above calculation of transaction status, the corresponding customer's transaction status sequence can be easily obtained for each data segment. When calculating the current customer's transaction status sequence, the electronic device can, based on transaction data preprocessing, calculate the average return ar of the i-th fund type for the transaction records within each period (e.g., one day) of each data segment (e.g., one month). i :ar i =r ij ×p ij ; Where, r ij p represents the rate of return of the j-th fund product belonging to the i-th fund type; ij This represents the amount m held by the j-th fund product. ij The total amount held by the i-th type of fund, s i The weighting; since there are four types of funds: money market funds, bond funds, mixed funds, and equity funds, this application uses an 8-tuple t. k = <ar k1 ,ar k2 ,ar k3 ,ar k4 ,s k1 ,s k2 ,s k3 ,s k4 The value > represents the fund trading state corresponding to day k. It should be noted that the eight physical quantities corresponding to the fund trading state are continuous, resulting in many different but semantically overlapping trading states. Therefore, this application employs an equal-frequency discretization method to limit the number of trading states. Based on the calculation of trading states, for each data segment of a trading period, a corresponding sequence of trading states can be easily obtained.

[0044] Based on the established transaction state sequence, this application constructs a frequency graph to reflect the correlation between transaction states, thereby quantifying fine-grained financial behavior information of clients. A frequency graph (f-Graph) measures the frequency of transitions between different transaction states for a specific client during short-term trading. In the f-Graph, nodes represent transaction states, and edges represent the frequency of changes between different transaction states. Specifically, given a segment of fund trading data, the corresponding daily transaction state t can first be calculated. Then, based on the obtained transaction state sequence, the frequency of transitions between different transaction states within that transaction data segment (data sample T) can be statistically analyzed.

[0045] S204. Input the frequency graph into the pre-trained financial style recognition model to obtain the fine-grained financial style of the current user during the corresponding period through the financial style recognition model; wherein, the financial style recognition model includes: graph attention layer, flattening layer, fully connected layer and classifier.

[0046] In this step, the electronic device can input the frequency map into a pre-trained financial style recognition model to obtain the fine-grained financial style of the current user during the corresponding period. The financial style recognition model includes a graph attention layer, a flattening layer, a fully connected layer, and a classifier. The design of each of these network layers and the classifier is described in detail below.

[0047] Graph Attention Layer: This layer primarily extracts high-level and abstract node feature representations by learning the relationships between nodes and their neighboring nodes. Its input is a set of node features from the data sample, and its output is a corresponding new set of node features. For a single data sample, this application uses A... f ∈R N×N This represents the adjacency matrix corresponding to the probabilistic graph (f-Graph); where N represents the number of nodes (i.e., the number of transaction states). Since the adjacency matrix directly reflects the feature relationship between each node and its neighboring nodes, this application uses the adjacency matrix corresponding to the frequency graph as the node feature set of the data sample (i.e., the input of the graph attention layer) for feature extraction. The calculation process of the graph attention layer mainly includes two steps: 1) calculation and normalization of attention coefficients; 2) obtaining new node representations through linear combinations of node features.

[0048] For ease of description, use Represents the set of node features, where, The feature of a node is represented by the weights of the relationships between node i and its neighboring nodes. To transform the input node features into a higher-level node feature representation, a shared weight matrix W1∈R needs to be applied to each node first. F×N The attention coefficient between a node and its neighboring nodes can be calculated using the following formula: Among them, e ij ∈R signifies the importance of the features of neighboring node j relative to node i; α represents a shared attention mechanism used to map high-dimensional features to a real number. To facilitate comparison of attention coefficients between different nodes and their neighbors, the attention coefficients can be normalized using the softmax function. The normalization process is as follows: Where, N i This represents the one-hop neighborhood of node i, including itself. In the experiments, the attention mechanism α is a single-layer feedforward neural network that uses a weight vector. The calculation involves the use of the LeakyReLU nonlinear activation function during normalization. Therefore, the complete expansion of the above formula is as follows: in,· T and || represent the transpose and concatenation operations of matrices, respectively.

[0049] Based on the normalized attention coefficients, they need to be linearly combined with the corresponding node features to obtain a new feature representation for each node. A new feature representation for a node can be expressed as: Where W2∈R F×N The weight matrix representing the learned value; N i This represents the one-hop neighborhood of node i, including itself. To ensure the graph attention layer can extract node features completely and stably, a multi-head attention method is needed. Specifically, the above formula is calculated k times independently and repeatedly, then summed and averaged to obtain the final output: a new feature representation of a node. It can be formalized as: in, w represents the normalized attention coefficient between node i and its neighbor j, calculated according to the above formula in the kth iteration. k ∈R F×N Let represent the weight matrix for the k-th learning iteration. Similarly, the normalized attention coefficients of the remaining N-1 nodes and their corresponding neighbor nodes can be calculated using the formula above. Then, the new features of these N-1 nodes can be obtained using the formula above. Finally, the new feature representations of all nodes are aggregated (i.e., ...). () to represent the data sample.

[0050] Flattening layer: In the flattening layer, a new feature is represented by A. f It is "flattened" into a long column vector (i.e.) ).

[0051] Fully connected layer: To map the feature vectors to the sample label space, a fully connected layer is constructed with 5 neurons (i.e., m∈R). 5×1 This corresponds to five different investment style categories for fine-grained investment style identification, and is defined as follows: Where W3∈R 5×NF The weight matrix representing the learned value; b4∈R 5×1 Let m represent the bias column vector for learning. In m, each neuron is used to calculate the probability that the sample belongs to the corresponding financial style category.

[0052] Classifier: For the input transaction data sample T, this application uses a softmax classifier applied to the extracted feature vector (i.e., m) to obtain the k-th investment style category (i.e., c). kThe probability that a character belongs to the true label (i.e.) ), which is defined as follows: k = 1, 2, ..., 5; where m (k) This represents the output of the k-th neuron in the fully connected layer for that sample. Furthermore, the final prediction result (i.e....) The category of investment style corresponding to the highest probability is represented as: k = 1, 2, ..., 5.

[0053] The optimization objective of the financial style recognition model based on graph attention networks is to calculate the classification loss of the training samples using cross-entropy, which can be formalized as: Where, N l Indicates the number of training samples; Let the i-th training sample belong to the k-th financial style category c. k The predicted probability; The k-th item of the one-hot encoding is assigned a true label to the i-th training sample. Finally, to obtain better weights for the graph attention network, this application uses the backpropagation algorithm to iteratively update the network parameters.

[0054] The fine-grained investment style identification method proposed in this application first divides the current customer's transaction data into M equal-length data segments; then, it constructs a frequency graph containing P fund category attributes based on the M equal-length data segments; finally, it inputs the frequency graph into a pre-trained investment style identification model, and obtains the fine-grained investment style of the current user for the corresponding period through the investment style identification model. The investment style identification model includes a graph attention layer, a flattening layer, a fully connected layer, and a classifier. In other words, the technical solution of this application can use the investment style identification model to identify the current customer's investment style in a fine-grained manner, thereby obtaining the current customer's fine-grained investment style without requiring financial professionals to rely on the customer's historical investment behavior habits for analysis, and is not limited by constraints such as survey costs, respondent burden, and unreliable survey results. In contrast, in existing technologies, traditional investment style identification methods still mainly rely on questionnaires to directly determine the customer's investment style, which is usually subject to various constraints and cannot accurately predict the user's investment style. Therefore, compared with the prior art, the fine-grained financial style identification method proposed in this application can predict customers' financial style in a fine-grained manner, thereby providing customers with more timely and personalized services; moreover, the technical solution of this application is simple and convenient to implement, easy to popularize, and has a wider range of applications.

[0055] Example 3

[0056] Figure 3This is a schematic diagram of the third process of the fine-grained financial style recognition method provided in this application embodiment. Further optimizations and extensions can be made based on the above technical solution, and it can be combined with the various optional implementation methods described above. For example... Figure 3 As shown, the fine-grained financial style identification method may include the following steps:

[0057] S301. If the financial style recognition model does not meet the pre-set first convergence condition, the historical transaction data of different users will be divided into X equal-length data segments, and a corresponding frequency diagram can be constructed according to the above method; where X is a natural number greater than 1.

[0058] S302. Randomly select a certain number of labeled training samples from X equal-length data segments as the training set; train the financial style recognition model using the training set based on the frequency map; repeat the above operation until the financial style recognition model satisfies the first convergence condition.

[0059] S303. If the financial style recognition model does not meet the pre-set second convergence condition, select the above-mentioned labeled data samples and a certain amount of unlabeled data samples from X data segments of equal length as the training set; use pseudo-label semi-supervised learning technology to train the financial style recognition model using the training set; repeat the above operation until the financial style recognition model meets the second convergence condition.

[0060] In a specific embodiment of this application, to reduce the burden of manual labeling while obtaining a classifier with better recognition performance, this application employs pseudo-label semi-supervised learning technology on the basis of a financial style recognition model based on graph representation learning. This involves assigning pseudo-labels to unlabeled data samples to participate in the training of the graph attention network, and encouraging the class probability distribution of predicted samples to approximate the one-hot encoding form, thereby achieving low-density separation between different categories and improving the model's recognition performance. This financial style recognition algorithm based on semi-supervised learning includes pseudo-label labeling and semi-supervised training. Specific details are as follows: For a given unlabeled training sample, a corresponding pseudo-label can be assigned. More specifically, with the help of the graph attention network, the unlabeled training sample can be represented as a vector. Then, the current classifier assigns a label (i.e., a pseudo-label) to the sample. During semi-supervised training, the classifier is continuously updated iteratively. Simultaneously, the pseudo-labels of the unlabeled training samples are relabeled by the new classifier. This encourages the class probability distribution of predicted samples to approximate the one-hot encoding form, achieving low-density separation between different categories, and thus enabling the model to obtain better recognition performance. Specifically, the k-th term of the one-hot encoding form of the sample pseudo-label can be calculated using the following formula: in, This indicates that the input data sample T belongs to the k′th financial style category (i.e., c). k′The predicted probability of ).

[0061] Unlike graph representation learning, semi-supervised learning trains the model using both labeled and pseudo-labeled training samples. During training, a classification loss is needed to distinguish between pseudo-labeled and labeled training samples. Therefore, this application defines the overall classification loss as a weighted sum of the two classification losses, which can be formalized as follows: in, and Let represent the predicted probabilities that the i-th labeled training sample and the j-th unlabeled training sample belong to the k-th financial style category, respectively. and N represents the k-th term of the one-hot encoding form of the label of the i-th labeled training sample and the j-th unlabeled training sample, respectively; l and N u The numbers of labeled and unlabeled training samples represent the total number of training samples, respectively, and their values ​​affect the recognition performance of semi-supervised learning. α is a weighting coefficient used to control the contribution of unlabeled training samples to the overall classification loss. It is worth noting that the hyperparameter α can also have a significant impact on the performance of semi-supervised learning, and its value should not be too small or too large. Too small an α will lead to insufficient utilization of information from unlabeled training samples; conversely, too large an α may introduce excessive noise, thereby affecting the model's recognition performance. Since the pseudo-labels of unlabeled samples can be improved through iterative training, this application introduces a scheduling function α(itr), which cleverly increases the value of α according to the number of iterations (itr) to mitigate this negative impact. Its definition is as follows: Here, α0, T1, and T2 are the three hyperparameters of the model. Analyzing these three hyperparameters can provide a deeper understanding of the training process of unlabeled training samples participating in semi-supervised learning. Specifically, in the initial stage of model training (i.e., when the number of iterations itr is less than T1), the information from the unlabeled training samples is not utilized by the model; then, as the number of iterations increases, α gradually increases; finally, when the number of iterations reaches a certain threshold (i.e., T2), α stops increasing and remains unchanged in subsequent training processes (i.e., α0).

[0062] S304. Divide the current customer's transaction data into M data segments of equal length; where M is a natural number greater than 1.

[0063] S305. Calculate the current transaction status of the corresponding customer based on M data segments of equal length; based on the transaction status, for each data segment, obtain the corresponding customer transaction status sequence.

[0064] S306. Construct a frequency graph containing P types of fund category attributes based on the current user's transaction status sequence.

[0065] S307. Input the frequency graph into the pre-trained financial style recognition model, and obtain the fine-grained financial style of the current user for the corresponding period through the financial style recognition model; wherein, the financial style recognition model includes: graph attention layer, flattening layer, fully connected layer and classifier.

[0066] This application proposes a fine-grained bank wealth management style recognition method based on graph representation learning and semi-supervised learning, which enables real-time automatic prediction of customers' fine-grained bank wealth management behavior and style. This method can play a significant role in promoting the sound development of the bank wealth management market and improving customers' investment efficiency. It has the following advantages: (1) The fund transaction data used in this application is anonymized, which can effectively protect certain security and privacy information of customers (such as customer name, ID number, etc.); (2) This application solves the semantic gap between low-quality coarse transaction data and advanced fine-grained wealth management behavior features. Specifically, the original data is first converted into graph data through fine-grained wealth management behavior expression to effectively quantify customers' fine-grained wealth management behavior information. Then, implicit and discriminative features related to complex wealth management behavior are extracted through graph representation learning model; (3) This application reduces the burden of manual labeling in model training. Specifically, a semi-supervised learning framework is introduced to make full use of unlabeled samples to further improve the recognition performance of the model; (4) This application has good scalability. On the one hand, the data used is universal and can be extended to anonymized financial transaction data of other products (such as stocks, bonds, etc.); on the other hand, the method proposed in this application can also be used to solve other complex behavior and style recognition tasks in real life (such as driving style).

[0067] The fine-grained investment style identification method proposed in this application first divides the current customer's transaction data into M equal-length data segments; then, it constructs a frequency graph containing P fund category attributes based on the M equal-length data segments; finally, it inputs the frequency graph into a pre-trained investment style identification model, and obtains the fine-grained investment style of the current user for the corresponding period through the investment style identification model. The investment style identification model includes a graph attention layer, a flattening layer, a fully connected layer, and a classifier. In other words, the technical solution of this application can use the investment style identification model to identify the current customer's investment style in a fine-grained manner, thereby obtaining the current customer's fine-grained investment style without requiring financial professionals to rely on the customer's historical investment behavior habits for analysis, and is not limited by constraints such as survey costs, respondent burden, and unreliable survey results. In contrast, in existing technologies, traditional investment style identification methods still mainly rely on questionnaires to directly determine the customer's investment style, which is usually subject to various constraints and cannot accurately predict the user's investment style. Therefore, compared with the prior art, the fine-grained financial style identification method proposed in this application can predict customers' financial style in a fine-grained manner, thereby providing customers with more timely and personalized services; moreover, the technical solution of this application is simple and convenient to implement, easy to popularize, and has a wider range of applications.

[0068] Example 4

[0069] Figure 4 This is a schematic diagram of the structure of the fine-grained financial style recognition device provided in an embodiment of this application. Figure 4 As shown, the fine-grained financial style recognition device includes: a segmentation module 401, a construction module 402, and a recognition module 403; wherein,

[0070] The segmentation module 401 is used to divide the current customer's transaction data into M data segments of equal length; where M is a natural number greater than 1.

[0071] The construction module 402 is used to construct a frequency map containing P fund category attributes based on the M equal-length data segments; where P is a natural number greater than 1.

[0072] The recognition module 403 is used to input the frequency map into a pre-trained financial style recognition model, and obtain the fine-grained financial style of the current user during the corresponding period through the financial style recognition model; wherein, the financial style recognition model includes: a graph attention layer, a flattening layer, a fully connected layer and a classifier.

[0073] The fine-grained financial style recognition device described above can execute the method provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the fine-grained financial style recognition method provided in any embodiment of this application.

[0074] Example 5

[0075] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 5 A block diagram of an exemplary electronic device suitable for implementing embodiments of the present application is shown. Figure 5 The electronic device 12 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0076] like Figure 5 As shown, the electronic device 12 is represented in the form of a general-purpose computing device. The components of the electronic device 12 may include, but are not limited to: one or more processors or processing units 16, system memory 28, and bus 18 connecting different system components (including system memory 28 and processing unit 16).

[0077] Bus 18 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0078] Electronic device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by electronic device 12, including volatile and non-volatile media, removable and non-removable media.

[0079] System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. Electronic device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 may be used to read and write non-removable, non-volatile magnetic media (… Figure 5 Not shown; usually referred to as a "hard drive"). Although Figure 5Not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 18 via one or more data media interfaces. Memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of this application.

[0080] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in memory 28. Such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 42 typically perform the functions and / or methods described in the embodiments of this application.

[0081] Electronic device 12 can also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), and with one or more devices that enable a user to interact with electronic device 12, and / or with any device that enables electronic device 12 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed via input / output (I / O) interface 22. Furthermore, electronic device 12 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 20. As shown, network adapter 20 communicates with other modules of electronic device 12 via bus 18. It should be understood that, although... Figure 5 As not shown, other hardware and / or software modules may be used in conjunction with electronic device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0082] The processing unit 16 executes various functional applications and data processing by running programs stored in the system memory 28, such as implementing the fine-grained financial style recognition method provided in the embodiments of this application.

[0083] Example 6

[0084] This application provides a computer storage medium.

[0085] The computer-readable storage medium of this application embodiment can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be—but is not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0086] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0087] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including—but not limited to—wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0088] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0089] Note that the above are merely preferred embodiments and the technical principles employed in this application. Those skilled in the art will understand that this application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of this application, the scope of which is determined by the scope of the appended claims.

Claims

1. A fine-grained financial style recognition method, characterized in that, The method comprises: dividing transaction data of a current customer into M equal-length data segments; wherein M is a natural number greater than 1; calculating a transaction state sequence of the current customer corresponding to the M equal-length data segments; constructing a frequency graph containing P fund category attributes according to the transaction state sequence of the current customer; wherein P is a natural number greater than 1, the frequency graph is used to measure the conversion frequency between different transaction states in the transaction process of the customer, the nodes of the frequency graph represent transaction states, and the edges of the frequency graph represent the frequency of changes between different transaction states; inputting the frequency graph into a pre-trained financial style recognition model to obtain a fine-grained financial style corresponding to the current customer during a period of time through the financial style recognition model; wherein the financial style recognition model comprises a graph attention layer, a flattening layer, a fully connected layer and a classifier; wherein the graph attention layer extracts node feature representation by learning the relationship between the node and its neighbor nodes, and the input is an adjacency matrix corresponding to the frequency graph.

2. The method of claim 1, wherein, The method further comprises: extracting transaction records in each period in each of the M equal-length data segments; calculating the transaction state of the current customer according to the transaction records in each period in each data segment; based on the transaction state, obtaining the transaction state sequence of the customer corresponding to each data segment.

3. The method of claim 1, wherein, Before inputting the frequency graph into the pre-trained financial style recognition model, the method further comprises: if the financial style recognition model does not meet a pre-set first convergence condition, dividing historical transaction data of different users into X equal-length data segments and constructing corresponding frequency graphs; wherein X is a natural number greater than 1; randomly selecting a certain amount of labeled training samples in the X equal-length data segments as a training set; training the financial style recognition model based on the frequency graph using the training set; and repeating the above operation until the financial style recognition model meets the first convergence condition.

4. The method of claim 3, wherein, The method further comprises: manually labeling a part of the data segments in the training set to obtain labeled data samples; training the financial style recognition model based on the frequency graph using the labeled data samples.

5. The method of claim 4, wherein, The method further comprises: if the financial style recognition model does not meet a pre-set second convergence condition, selecting the labeled data samples and a certain amount of unlabeled data samples in the X equal-length data segments as a training set; training the financial style recognition model using the training set by adopting a pseudo-label semi-supervised learning technology; and repeating the above operation until the financial style recognition model meets the second convergence condition.

6. The method of claim 5, wherein, The method further comprises: identifying the unlabeled data segments in the training set using the model during training to obtain pseudo-labeled data samples; The semi-supervised training is performed on the financial style recognition model using the labeled data sample and the pseudo-labeled data sample.

7. A fine-grained style identification device, characterized by, The device comprises a division module, a construction module and an identification module, wherein The division module is configured to divide the transaction data of the current customer into M equal-length data segments, wherein M is a natural number greater than 1. The construction module is configured to calculate the transaction state sequence of the current customer based on the M equal-length data segments. A frequency graph containing P fund category attributes is constructed according to the transaction state sequence of the current customer, wherein P is a natural number greater than 1, the frequency graph is used to measure the conversion frequency between different transaction states in the transaction process, the nodes of the frequency graph represent the transaction states, and the edges of the frequency graph represent the frequency of changes between different transaction states. The identification module is configured to input the frequency graph into a pre-trained financial style recognition model to obtain the fine-grained financial style corresponding to the current customer during a period of time through the financial style recognition model, wherein the financial style recognition model comprises a graph attention layer, a flattening layer, a fully connected layer and a classifier, wherein the graph attention layer learns the relationship between the nodes and their neighbor nodes to extract the node feature representation, and the input of the graph attention layer is an adjacency matrix corresponding to the frequency graph.

8. An electronic device, comprising: The device comprises: one or more processors; a memory for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the fine-grained financial style recognition method according to any one of claims 1 to 6.

9. A storage medium having stored thereon a computer program, characterized in that The program is executed by the processor to implement the fine-grained financial style recognition method according to any one of claims 1 to 6.

Images