Financial user behavior analysis method and system based on multi-feature large model

Abstract

The invention discloses a financial user behavior analysis method and system based on a multi-feature large model, and belongs to the field of financial user data analysis and processing. According to the method, financial feature vectors of a user are constructed by collecting historical financial data (including transaction unit price, quantity, financial product total price, transaction rate and the like) of the user, and the cosine similarity of any two feature vectors is calculated to perform mean clustering, so that an initial financial category is obtained. Furthermore, through a multi-threshold segmentation and factor analysis method, the initial category is optimized, independent factors and information contents thereof are extracted, a separation necessity value and a separation value are calculated, and finally a plurality of user categories are divided and financial user portraits are constructed. According to the method, a multi-dimensional data source and various data modeling technologies can be effectively fused, the refined description capability of user behaviors is improved, and accurate recommendation and differentiated services in the financial field are supported. The system comprises a data acquisition module, a clustering analysis module, a clustering optimization module, a category subdivision module, a factor analysis module, a category division module and a user portrait construction module, and is suitable for financial user behavior analysis, marketing and personalized recommendation.

Description

Technical Field

[0001] The present invention belongs to the field of financial user data analysis and processing, and more specifically, relates to a financial user behavior analysis method and system based on a multi-feature large model. Background Art

[0002] The big data multi-model method refers to combining multiple different types of data models or processing methods when dealing with large-scale data, and designing a more targeted solution for the diversity and complexity of the data. By comprehensively using different data processing tools and technologies, this method can more effectively cope with heterogeneous data environments, improve the accuracy and depth of analysis, and is widely used in fields such as user behavior analysis, marketing, and personalized recommendation.

[0003] A user profile is an abstract description constructed based on user characteristics, behaviors, and preference information, used to depict the commonalities and characteristics of the target user group. In digital marketing, market research, and user experience design, it is an important means to understand user needs and formulate targeted strategies. However, traditional user profile methods often rely on a single model or a single dimension (such as clustering analysis), and the description of user behavior is relatively rough, failing to fully reflect the diverse needs and personalized characteristics of different user groups. This shortcoming is particularly obvious in the user profile behavior analysis in the financial field. Relying solely on the general analysis of financial data, it is difficult to achieve accurate profiling of different users, thereby restricting the differential service capabilities of the financial industry in analyzing user behavior. Summary of the Invention

[0004] To solve the above problems, the present invention provides a financial user behavior analysis method and system based on a multi-feature large model.

[0005] In a first aspect, the present invention provides a financial user behavior analysis method based on a multi-feature large model, including the following steps:

[0006] Collect historical financial data of users, where the historical financial data includes transaction unit price, quantity, total price of financial products, and transaction rate, and construct a financial feature vector of the user based on the historical financial data;

[0007] Calculate the cosine similarity between any two financial feature vectors to obtain a first similarity, and perform mean clustering based on the first similarity to obtain multiple initial financial categories;

[0008] For each initial financial category, determine its clustering center, marked as the first financial feature vector; for the financial feature vectors other than the clustering center in each initial financial category, mark them as the second financial feature vectors, and calculate the cosine similarity between each second financial feature vector and the clustering center to obtain a second similarity;

[0009] Perform multi-threshold segmentation based on the second similarity, and reclassify the users in each initial financial category into multiple secondary financial categories;

[0010] For each secondary financial category, use factor analysis to extract independent factors and their information content, and calculate the separation necessity value and separation demand value for each secondary financial category based on the information content of the independent factors;

[0011] Calculate the separation value for each secondary financial category according to the separation necessity value and separation demand value;

[0012] Divide multiple user categories according to the separation value, and determine the financial behavior portrait of the user based on the user category.

[0013] In a second aspect, the present invention provides a financial user behavior analysis system based on a multi-feature large model, including:

[0014] A data acquisition module for collecting historical financial data of users, where the historical financial data includes transaction unit price, quantity, total price of financial products, and transaction rate, and constructing a financial feature vector of the user based on the historical financial data;

[0015] A clustering analysis module for calculating the cosine similarity between any two financial feature vectors to obtain the first similarity, and performing mean clustering based on the first similarity to obtain multiple initial financial categories;

[0016] A clustering optimization module for determining the clustering center of each initial financial category, marked as the first financial feature vector, and marking the financial feature vectors other than the clustering center in each initial financial category as the second financial feature vector, and calculating the cosine similarity between each second financial feature vector and the clustering center to obtain the second similarity;

[0017] A category subdivision module for performing multi-threshold segmentation based on the second similarity, and reclassifying the users in each initial financial category into multiple secondary financial categories;

[0018] A factor analysis module for using factor analysis to extract independent factors and their information content for each secondary financial category, and calculating the separation necessity value and separation demand value for each secondary financial category based on the information content of the independent factors;

[0019] A category division module for calculating the separation value of each secondary financial category according to the separation necessity value and separation demand value, and dividing multiple user categories according to the separation value;

[0020] A user portrait construction module for determining the financial behavior portrait of the user based on the user category.

[0021] Advantages of the present invention:

[0022] While clustering general financial order data, the present invention updates the cluster categories by combining the common and specific features of different clusters, thereby obtaining the cluster category information after secondary clustering, so that the obtained user portraits can meet the special needs of different financial customer groups, in order to achieve the purpose of accurate recommendation. Description of the Drawings

[0023] Figure 1 It is a flowchart of a financial user behavior analysis method based on a multi-feature large model according to an embodiment of the present application. Detailed Embodiment

[0024] The following further describes this embodiment with reference to the drawings. The overall flowchart is shown in Figure 1 .

[0025] Step 1: Collect historical financial data of users, where the historical financial data includes transaction unit price, quantity, total price of financial products, and transaction rate, and construct a financial feature vector of the user;

[0026] Step 2: Calculate the cosine similarity between any two financial feature vectors, perform mean calculation, and obtain multiple first financial categories;

[0027] Exemplarily, taking 1 - cosine similarity as the distance metric index, applying the k - means clustering algorithm to classify the financial feature vectors, and obtaining several initial financial categories. The clustering result ensures that the financial feature vectors within the same category have high similarity, while the similarity between financial feature vectors in different categories is significantly reduced.

[0028] Step 3: Use the feature centroid of all financial feature vectors in the first category (i.e., the initial financial category) as the clustering center of this category, and mark it as the financial feature center vector (i.e., the first financial feature vector).

[0029] Exemplarily, take the financial feature vector corresponding to the feature dimension centroid feature value as the clustering center of this first category, and thus obtain the clustering center of each first category.

[0030] Step 4: Mark all financial feature vectors except the financial feature center vector in each first category as the second financial feature vectors, and calculate the cosine similarity between each second financial feature vector and the financial feature center vector. The obtained result is defined as the second similarity.

[0031] Exemplarily, there are multiple second similarities in each second category.

[0032] Step 5: Perform multi - threshold segmentation according to the second similarity to obtain multiple second categories (i.e., secondary financial categories).

[0033] Exemplarily, based on the second similarity, a histogram is constructed and a local peak in the histogram is identified. The second similarity is segmented at multiple levels using the peak as a threshold basis to generate multiple subdivided categories (second categories). Each second category corresponds to a peak, and the maximum frequency of the same second similarity in the histogram is the peak. By comparing the similarity between each second category and the target category, the second category with the largest peak is marked as the target category. In the second category division, the similarity between the second financial feature vector in each second category and the financial feature vector of the cluster center of the corresponding first category (initial clustering result) is further calculated. By evaluating the separation degree of each second category, the necessity of its independence from the first category to which it belongs is judged. The separation degree measures the difference between the second category and the target category. The smaller the difference, the higher the separation degree required for separation, indicating that the personalized characteristics of the category are more significant. When the separation degree of the second category reaches a higher threshold, the category is separately divided into an independent category to more accurately reflect its uniqueness.

[0034] Furthermore, for example, assume that the first category A contains two second categories a1 and a2, where the similarity between a1 and the target category is 0.9, and the similarity between a2 and the target category is 0.8. Since the separation degree of a1 is higher than that of a2, a1 is preferentially separated from the first category to form an independent second category. In addition, based on the unique factor analysis, the similarity between the unique factor of each second category and the unique factor of the target category is calculated to evaluate the information content of the unique factor of this category. The higher the information content of the unique factor, the greater the necessity of separation, indicating that the category has more significant differentiation characteristics. This approach helps to enhance the personalized characteristics of the user group and provide support for subsequent precise recommendations.

[0035] Step 6: Use factor analysis on the second category to obtain the information content of each independent factor, and calculate the separation necessity value and separation requirement value of each second category based on the information content of the independent factor.

[0036] Exemplarily, a feature matrix is constructed with the financial feature vectors in each second category, and factor analysis is applied to this matrix to extract its common factors and unique factors, which are respectively denoted as the common factor vector and unique factor vector of the financial feature matrix. The common factor and unique factor of the target category are respectively defined as the target category common factor and target category unique factor. By calculating the cosine similarity between any two common factors, the common factor combination with the largest similarity is selected and marked as the common factor pair. The corresponding two groups of financial feature vector matrices are then marked as the financial feature matrix pair. Factor analysis is used again on the financial feature matrix pair to extract its unique factors, obtaining the set of unique factors of the financial feature matrix pair. Then, based on the relationship between the unique factors and the common factor pair, the cosine similarity between each unique factor and any one of the common factors in the common factor pair is calculated, and the result is normalized to obtain the contribution degree information of each unique factor, which is used as the feature information content of the unique factor. Further, according to the feature information of the unique factors and common factors, the separation necessity value of each second category is calculated. At the same time, by calculating the cosine similarity between the common factor of the second category and the common factor of the target category, the separation demand value of each second category is obtained. The separation necessity value and separation demand value together reflect the difference between this category and the target category, providing a scientific basis for optimizing the user portrait classification.

[0037] Step 7: Calculate the separability value of the second category according to the separation necessity value and the separation requirement value.

[0038] The separability value satisfies the following relational expression: where p represents the separability value of the second category, f represents the separation requirement value, and g represents the separation necessity value of the second category.

[0039] Step 8: Divide multiple user categories according to the separability value, and determine the financial behavior portrait of the user based on the user category.

[0040] In response to the separability value of the second category being greater than the preset threshold, it is divided into the first category, and financial service recommendation and promotion strategies are formulated based on the user category.

[0041] Exemplarily, the preset threshold is 0.7. By calculation, the separability of each second category in each first category can be obtained. Those second categories with a value greater than 0.7 are divided into the first category, resulting in two user categories, and then financial service recommendation and promotion strategies are formulated according to the user category.

[0042] Based on the same concept as the above method, the embodiment of the present application also provides a financial user behavior analysis system based on a multi-feature large model, including:

[0043] A data collection module for collecting historical financial data of users, where the historical financial data includes transaction unit price, quantity, total price of financial products, and transaction rate, and constructing a financial feature vector of the user based on the historical financial data;

[0044] A clustering analysis module for calculating the cosine similarity between any two financial feature vectors to obtain a first similarity, and performing mean clustering based on the first similarity to obtain multiple initial financial categories;

[0045] A clustering optimization module for determining the clustering center of each initial financial category, marked as the first financial feature vector, and for the financial feature vectors other than the clustering center in each initial financial category, marked as the second financial feature vector, calculating the cosine similarity between each second financial feature vector and the clustering center to obtain a second similarity;

[0046] A category subdivision module for performing multi-threshold segmentation based on the second similarity to reclassify the users in each initial financial category into multiple secondary financial categories;

[0047] A factor analysis module for each secondary financial category, using factor analysis to extract independent factors and their information content, and calculating the separation necessity value and separation demand value of each secondary financial category based on the information content of the independent factors;

[0048] A category division module for calculating the separability value of each secondary financial category according to the separation necessity value and separation demand value, and dividing multiple user categories according to the separability value;

[0049] A user portrait construction module for determining the financial behavior portrait of the user based on the user category.

[0050] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a computer program. The financial user behavior analysis program based on the multi-feature large model can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the storage medium of the financial user behavior analysis program based on the multi-feature large model can be a magnetic disk, an optical disk, a read-only memory, or a random access memory, etc.

[0051] The above-disclosed are only the preferred embodiments of the present application. Of course, the scope of the rights of the present application cannot be limited by this. Therefore, equivalent changes made according to the claims of the present application still fall within the scope covered by the present application.

Claims

1. A financial user behavior analysis method based on a multi-feature large model, characterized in that, It includes the following steps: Collect the user's historical financial data, where the historical financial data includes transaction unit price, quantity, total price of financial products, and transaction rate, and construct a financial feature vector of the user based on the historical financial data; Calculate the cosine similarity between any two financial feature vectors to obtain the first similarity, and perform mean clustering based on the first similarity to obtain multiple initial financial categories; For each initial financial category, determine its clustering center, marked as the first financial feature vector; for the financial feature vectors other than the clustering center in each initial financial category, mark them as the second financial feature vectors, and calculate the cosine similarity between each second financial feature vector and the clustering center to obtain the second similarity; Perform multi-threshold segmentation based on the second similarity, and reclassify the users in each initial financial category into multiple secondary financial categories; For each secondary financial category, use factor analysis to extract independent factors and their information content, and calculate the separation necessity value and separation demand value of each secondary financial category based on the information content of the independent factors; Calculate the separability value of each secondary financial category according to the separation necessity value and separation demand value; Divide multiple user categories according to the separability value, and determine the financial behavior portrait of the user based on the user categories.

2. The financial user behavior analysis method based on a multi-feature large model according to claim 1, wherein, The step of collecting the user's historical financial data further includes: Collect the user's static data, where the static data includes the user's age, gender, occupation, and income level; Combine the static data with the historical financial data to construct a more comprehensive financial feature vector.

3. A financial user behavior analysis method based on a multi-feature large model according to claim 1 or 2, characterized in that, In the step of calculating the cosine similarity, use 1 - cosine similarity as the distance metric index, and apply the k-means clustering algorithm to classify the financial feature vectors.

4. A financial user behavior analysis method based on a multi-feature large model according to claim 1, characterized in that, The step of multi-threshold segmentation includes: Construct a histogram based on the second similarity and identify the local peaks in the histogram; Use the peaks as the threshold basis to perform multi-level segmentation on the second similarity to generate multiple sub-categories.

5. The method for analyzing financial user behavior based on a multi-feature large model according to claim 1 or 4, characterized in that The steps of the factor analysis method include: Construct a feature matrix for the financial feature vectors in each secondary financial category; Apply factor analysis to the feature matrix to extract its common factors and unique factors; Calculate the cosine similarity between the unique factor and the common factor to obtain the contribution degree information of each unique factor.

6. The method for analyzing financial user behavior based on a multi-feature large model according to claim 1, characterized in that The calculation formula for the separability value is: ； Among them, represents the separability value of the secondary financial category, represents the separability demand value, represents the separability necessity value of the secondary financial category.

7. A financial user behavior analysis method based on a multi-feature large model according to claim 6, wherein, When the separability value is greater than the preset threshold, divide the corresponding secondary financial category into an independent user category.

8. A financial user behavior analysis method based on a multi-feature large model according to claim 1, characterized in that, The financial behavior portrait includes the user's consumption preferences, investment tendencies, and risk tolerance.

9. A financial user behavior analysis method based on a multi-feature large model according to claim 1 or 8, characterized in that The method further includes the step of formulating financial service recommendation and promotion strategies based on the financial behavior portrait.

10. A financial user behavior analysis system based on a multi-feature large model, characterized in that, It includes: A data collection module for collecting the user's historical financial data, where the historical financial data includes transaction unit price, quantity, total price of financial products, and transaction rate, and constructing a financial feature vector of the user based on the historical financial data; The clustering analysis module is used to calculate the cosine similarity between any two financial feature vectors to obtain the first similarity, and perform mean clustering based on the first similarity to obtain multiple initial financial categories; The clustering optimization module is used to determine the clustering center for each initial financial category, marked as the first financial feature vector, and mark the financial feature vectors other than the clustering center in each initial financial category as the second financial feature vectors, and calculate the cosine similarity between each second financial feature vector and the clustering center to obtain the second similarity; The category subdivision module is used to perform multi-threshold segmentation based on the second similarity to reclassify the users in each initial financial category into multiple secondary financial categories; The factor analysis module is used to, for each secondary financial category, extract independent factors and their information content using factor analysis, and calculate the separation necessity value and separation demand value for each secondary financial category based on the information content of the independent factors; The category division module is used to calculate the separation value for each secondary financial category according to the separation necessity value and separation demand value, and divide multiple user categories according to the separation value; The user portrait construction module is used to determine the financial behavior portrait of the user based on the user category.

Images