Recommendation Method and Device for Financial Products, Electronic Device, and Storage Medium

Abstract

The present invention discloses a method and apparatus for recommending financial products, an electronic device, and a storage medium, relating to the field of artificial intelligence. The method includes: extracting a target feature set of a target user, calculating a user cluster to which the target user belongs based on the feature values of each feature in the target feature set, and when the user cluster belongs to a preset cluster set, querying matrix nodes with a relevant value one greater than or equal to a first preset threshold based on a first type of association matrix to obtain a first node set, so as to determine a target product attribute set corresponding to the target feature set of the target user, extracting product attributes in the target product attribute set, customizing a target financial product based on the extracted product attributes, and recommending the target financial product to the target user. The present invention solves the technical problem in the related art that customized financial products cannot be recommended for new users without consumption history data, resulting in a reduced experience for new users.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular, to a method and device for recommending financial products, an electronic device, and a storage medium. Background Art

[0002] With the development of computer technology and the rapid growth of users' consumption ability of financial products, how financial institutions can fully rely on customer information big data, and make full use of existing customers' asset data, credit data, transaction data, etc., predict and mine the consumption ability of financial products of customer groups through computing technology, and effectively guide the further innovation of financial products is an important means to achieve the sustainable development of financial business.

[0003] In related technologies, a recommendation system is an advanced business intelligence platform based on massive data mining to help institutions provide fully personalized decision-making support and information services for their customers' shopping. The recommendation system can recommend information, products, etc. that customers are interested in to customers according to customers' information needs, interests, etc. Compared with a search engine, the recommendation system conducts personalized calculations by studying customers' interest preferences, discovers customers' interest points by the system, and thus guides customers to discover their own product and information needs.

[0004] Currently, in the field of financial products, the potential correlation between existing customers and financial products can be inferred through a recommendation system and related algorithms, and recommendation results of financial products can be provided to customers. However, there are different problems in the recommendation system and related algorithms. For example, problems such as user sparsity, cold start, and large data modeling difficulty limit the actual application scope of the recommendation system, which is only limited to a few industries such as e-commerce. Moreover, the recommendation system itself is limited to the internal matching relationship between existing users and financial products, and does not have the ability to automatically model and produce financial products. When the number of new users with demand in the financial product market increases rapidly, due to the blank of the financial product consumption history data of new users themselves, the existing recommendation system is affected by the data cold start problem and cannot play a role for this part of users in a timely manner. It can only perform matching prediction calculations for the existing customer group and product system, and does not have the ability to guide the product innovation direction through exploratory calculations. Therefore, financial institutions need more effective computing technology means to accurately measure and model the financial product needs of new users to achieve rapid production of customized financial products.

[0005] To solve the above problems, no effective solution has been proposed yet. Summary of the Invention

[0006] An embodiment of the present invention provides a method and apparatus for recommending financial products, an electronic device, and a storage medium, so as to at least solve the technical problem in the related art that customized financial products cannot be recommended for new users without consumption history data, resulting in a reduced experience for new users.

[0007] According to one aspect of the embodiments of the present invention, a method for recommending financial products is provided, including: extracting a target feature set of a target user, where the target feature set is obtained by screening a feature set composed of all features of the target user using a preset target feature library; calculating the user cluster to which the target user belongs based on the feature values of each feature in the target feature set; in the case where the user cluster belongs to a preset cluster set, querying matrix nodes with a correlation value one greater than or equal to a first preset threshold based on a first type of association matrix, obtaining a first node set, where the first type of association matrix is an association matrix corresponding to the preset cluster set, and the matrix nodes represent the association relationship between each user cluster in the preset cluster set and the product attributes of financial products, and each matrix node corresponds to the correlation value one; combining the first node set to determine a target product attribute set corresponding to the target feature set of the target user; extracting the product attributes in the target product attribute set, and customizing a target financial product based on the extracted product attributes, and recommending the target financial product to the target user.

[0008] Optionally, after calculating the user cluster to which the target user belongs based on the feature values of each feature in the target feature set, the recommendation method further includes: in the case where the user cluster does not belong to the preset cluster set, determining a second type of association matrix based on the association relationship between each user feature in the target feature set and the product attributes, where each matrix node in the second type of association matrix is a node represented by the user feature and the product attribute, and the node value corresponding to each matrix node represents the correlation value two between the user feature and the product attribute; sorting the correlation value two of all the matrix nodes in the second type of association matrix to obtain a first sorting result; based on the first sorting result, selecting matrix nodes with a correlation value two greater than or equal to a second preset threshold; combining the product attributes corresponding to the matrix nodes into a target product attribute set.

[0009] Optionally, before extracting the target feature set of the target user, the recommendation method further includes: obtaining historical consumption data of financial products and multiple historical user features within a historical time period; combining the historical user features in any pairwise combination to obtain multiple user feature combinations; establishing a third type of association matrix between the user feature combinations and the product attributes, where each matrix node in the third type of association matrix refers to the node formed between the user feature combination and the product attribute; based on the historical consumption data, calculating the matrix node value of each matrix node in the third type of association matrix to obtain a first matrix node value distribution diagram, where the matrix node value is the consumption transaction frequency of the user feature combination and the product attribute indicated by the matrix node.

[0010] Optionally, after calculating the matrix node value of each matrix node in the third type of association matrix based on the historical consumption data to obtain a first matrix node value distribution diagram, the recommendation method further includes: sorting the matrix node values based on the first matrix node value distribution diagram to obtain a second sorting result; based on the second sorting result, extracting the matrix node values greater than or equal to a third preset threshold, and combining the extracted matrix node values to obtain a first numerical queue.

[0011] Optionally, after obtaining the first numerical queue, the recommendation method further includes: preprocessing the first numerical queue to obtain a processed first numerical queue; calculating the combined value of the user feature combination based on the matrix node value in the processed first numerical queue; calculating the total feature value of each user feature based on the combined value; sorting the total feature values to determine the target features whose total feature values are greater than or equal to a fourth preset threshold; and forming the target feature library with the target features.

[0012] Optionally, after forming the target feature library with the target features, the recommendation method further includes: determining a sliding window corresponding to each user cluster based on all the features in the target feature library; randomly selecting a center point based on the sliding window, and sliding the center point within the sliding window until the highest density point is found, and representing the highest density point as the new center point; after repeatedly selecting new center points for the sliding window, if the positions of the selected center points are within a certain fixed area within the target selection time period, it is determined that the center point selection stability reaches a preset stable condition; determining the center point corresponding to each user cluster to obtain multiple cluster center points; and obtaining a cluster set based on the multiple cluster center points.

[0013] Optionally, after obtaining the clustering set based on the multiple clustering center points, the recommendation method further includes: obtaining a two-dimensional matrix between the clustering set and financial products based on the clustering set and the historical consumption data, where the matrix node value corresponding to the matrix node in the two-dimensional matrix is the trading volume.

[0014] Optionally, before extracting the target feature set of the target user, the recommendation method further includes: obtaining the historical consumption data of financial products within a historical time period; establishing a fourth type of association matrix between historical user features and the product attributes; calculating the matrix node value of each matrix node in the fourth type of association matrix based on the historical consumption data to obtain a second matrix node value distribution map, where the matrix node value is the consumption transaction frequency of the user features and product attributes indicated by the matrix node; sorting the matrix node values based on the second matrix node value distribution map to obtain a third sorting result; extracting the matrix node values greater than or equal to a fifth preset threshold based on the third sorting result, and combining the extracted matrix node values to obtain a second numerical queue.

[0015] Optionally, after obtaining the two-dimensional matrix between the clustering set and the products, the recommendation method further includes: decomposing the financial products according to the product attributes based on the two-dimensional matrix between the clustering set and the financial products to obtain a first type of association matrix between user clusters and product attributes; calculating the matrix node value of each matrix node in the first type of association matrix based on the second numerical queue, the user features in the user cluster, and the product attributes, where the matrix node value is characterized as a correlation value one.

[0016] According to another aspect of the embodiments of the present invention, there is also provided a recommendation device for financial products, which is characterized in that it includes: an extraction unit for extracting a target feature set of a target user, where the target feature set is obtained by screening a feature set composed of all features of the target user using a preset target feature library; a calculation unit for calculating the user cluster to which the target user belongs based on the feature values of each feature in the target feature set; a query unit for, when the user cluster belongs to a preset cluster set, querying matrix nodes whose relevant value one is greater than or equal to a first preset threshold based on a first type of association matrix, to obtain a first node set, where the first type of association matrix is an association matrix corresponding to the preset cluster set, the matrix nodes represent the association relationship between each user cluster in the preset cluster set and the product attributes of financial products, and each matrix node corresponds to the relevant value one; a determination unit for determining a target product attribute set corresponding to the target feature set of the target user in combination with the first node set; a recommendation unit for extracting the product attributes in the target product attribute set, customizing a target financial product based on the extracted product attributes, and recommending the target financial product to the target user.

[0017] Optionally, the recommendation device further includes: a first determination module for, after calculating the user cluster to which the target user belongs based on the feature values of each feature in the target feature set, when the user cluster does not belong to the preset cluster set, determining a second type of association matrix based on the association relationship between each user feature in the target feature set and the product attributes, where each matrix node in the second type of association matrix is a node represented by the user feature and the product attribute, and the node value corresponding to each matrix node represents the relevant value two between the user feature and the product attribute; a first sorting module for sorting the relevant value two of all the matrix nodes in the second type of association matrix to obtain a first sorting result; a first selection module for selecting, based on the first sorting result, matrix nodes whose relevant value two is greater than or equal to a second preset threshold; a first combination module for combining the product attributes corresponding to the matrix nodes into a target product attribute set.

[0018] Optionally, the recommendation device further includes: a first acquisition module, configured to acquire historical consumption data of financial products and multiple historical user characteristics within a historical time period before extracting the target feature set of the target user; a second combination module, configured to arbitrarily combine the historical user characteristics in pairs to obtain multiple user characteristic combinations; a first establishment module, configured to establish a third type of association matrix between the user characteristic combinations and the product attributes, where each matrix node in the third type of association matrix refers to a node formed between the user characteristic combination and the product attribute; a first calculation module, configured to calculate a matrix node value of each matrix node in the third type of association matrix based on the historical consumption data to obtain a first matrix node value distribution map, where the matrix node value is the consumption transaction frequency of the user characteristic combination and the product attribute indicated by the matrix node.

[0019] Optionally, the recommendation device further includes: a second sorting module, configured to sort the matrix node values based on the first matrix node value distribution map after calculating the matrix node value of each matrix node in the third type of association matrix based on the historical consumption data to obtain a second sorting result; a first extraction module, configured to extract matrix node values greater than or equal to a third preset threshold based on the second sorting result, and combine the extracted matrix node values to obtain a first numerical queue.

[0020] Optionally, the recommendation device further includes: a first processing module, configured to preprocess the first numerical queue after obtaining the first numerical queue to obtain a preprocessed first numerical queue; a second calculation module, configured to calculate a combined value of the user characteristic combinations based on the matrix node values in the preprocessed first numerical queue; a third calculation module, configured to calculate a total feature value of each user characteristic based on the combined value; a third sorting module, configured to sort the total feature values to determine target features whose total feature values are greater than or equal to a fourth preset threshold; a third combination module, configured to form the target features into the target feature library.

[0021] Optionally, the recommendation device further includes: a second determination module, configured to, after forming the target feature library with the target features, determine a sliding window corresponding to each user cluster based on all the features in the target feature library; a second selection module, configured to randomly select a center point based on the sliding window, and slide the center point within the sliding window until the highest density point is found, and characterize the highest density point as a new center point; a third determination module, configured to, after repeatedly selecting new center points for the sliding window, if the positions of the selected center points are all within a certain fixed area during the target selection time period, determine that the stability of the center point selection reaches a preset stability condition; a fourth determination module, configured to determine the center points corresponding to each user cluster to obtain a plurality of cluster center points; a first output module, configured to obtain a cluster set based on the plurality of cluster center points.

[0022] Optionally, the recommendation device further includes: a second output module, configured to, after obtaining a cluster set based on the plurality of cluster center points, obtain a two-dimensional matrix between the cluster set and financial products based on the cluster set and the historical consumption data, where the matrix node value corresponding to the matrix node in the two-dimensional matrix is the trading volume.

[0023] Optionally, the recommendation device further includes: a second acquisition module, configured to acquire the historical consumption data of financial products during a historical time period before extracting the target feature set of the target user; a second establishment module, configured to establish a fourth type of association matrix between the historical user features and the product attributes; a fourth calculation module, configured to calculate the matrix node value of each matrix node in the fourth type of association matrix based on the historical consumption data to obtain a second matrix node value distribution map, where the matrix node value is the consumption transaction frequency of the user features and product attributes indicated by the matrix node; a fourth sorting module, configured to sort the matrix node values based on the second matrix node value distribution map to obtain a third sorting result; a second extraction module, configured to extract the matrix node values greater than or equal to a fifth preset threshold based on the third sorting result, and combine the extracted matrix node values to obtain a second numerical queue.

[0024] Optionally, the recommendation device further includes: a first decomposition module, configured to, after obtaining the two-dimensional matrix between the cluster set and the products, decompose the financial products according to the product attributes based on the two-dimensional matrix between the cluster set and the financial products to obtain a first type of association matrix between the user clusters and the product attributes; a fifth calculation module, configured to calculate the matrix node value of each matrix node in the first type of association matrix based on the second numerical queue, the user features in the user cluster, and the product attributes, where the matrix node value is characterized as a correlation value one.

[0025] According to another aspect of the embodiments of the present invention, there is also provided a computer-readable storage medium, which includes a stored computer program. When the computer program runs, it controls the device where the computer-readable storage medium is located to execute the above-mentioned financial product recommendation method.

[0026] According to another aspect of the embodiments of the present invention, there is also provided an electronic device, including one or more processors and a memory. The memory is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the above-mentioned financial product recommendation method.

[0027] In the present disclosure, a target feature set of a target user is extracted. Based on the feature values of each feature in the target feature set, the user cluster to which the target user belongs is calculated. When the user cluster belongs to a preset cluster set, matrix nodes with a correlation value one greater than or equal to a first preset threshold are queried based on a first type of association matrix to obtain a first node set. Combining the first node set, a target product attribute set corresponding to the target feature set of the target user is determined. Product attributes in the target product attribute set are extracted, and a target financial product is customized based on the extracted product attributes, and the target financial product is recommended to the target user. In the present application, through the association relationship (i.e., the first type of association matrix) between the user cluster and the product attributes of the financial product, the correlation value between the user cluster to which the user belongs and the product attributes can be calculated, so as to obtain a strong association product attribute set of the user (i.e., the target product attribute set). Arbitrarily selecting product attributes in the product attribute set for product customization can realize customizing personalized financial products for users to meet the financial product needs of users, and further solve the technical problem in the related art that it is impossible to recommend customized financial products to new users without consumption history data, resulting in a reduced experience for new users. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0029] Figure 1 is a flowchart of an optional financial product recommendation method according to an embodiment of the present invention;

[0030] Figure 2 is a flowchart of an optional financial product customization production according to an embodiment of the present invention;

[0031] Figure 3 is a schematic diagram of an optional financial product recommendation device according to an embodiment of the present invention;

[0032] Figure 4 It is a hardware structure block diagram of an electronic device (or mobile device) for a financial product recommendation method according to an embodiment of the present invention. Detailed implementation manners

[0033] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0034] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those clearly listed steps or units, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0035] For the convenience of those skilled in the art to understand the present invention, some terms or nouns involved in each embodiment of the present invention are explained below:

[0036] Mean shift clustering algorithm: It is an algorithm based on a sliding window and is used to find dense regions of data points.

[0037] It should be noted that the financial product recommendation method and its device in the present disclosure can be used in the field of artificial intelligence when recommending financial products, and can also be used in any field other than the field of artificial intelligence when recommending financial products. The application field of the financial product recommendation method and its device in the present disclosure is not limited.

[0038] 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 for display, data for analysis, etc.) involved in the present disclosure are all information and data authorized by the user or fully authorized by all parties.

[0039] The following embodiments of the present invention can be applied to various systems / applications / devices for recommending financial products, and recommend suitable financial products according to the characteristics of new users. The financial products involved include, but are not limited to: loan products, fund products, etc. Since financial products have the characteristics that their product attributes can be quantified and the dimension set is relatively fixed, and they have the conditions for realizing automated and customized production of products. Therefore, the present invention can rely on and establish a user big data model and a financial product model, deeply extract user behavior data and financial product attributes, establish a feature value model library, and realize the automated and customized production of financial products through machine learning means, and can realize the function of meeting the financial product needs of new users, and can repeatedly strengthen and improve this function during the machine learning process.

[0040] The present invention will be described in detail below in conjunction with each embodiment.

[0041] Embodiment 1

[0042] According to an embodiment of the present invention, there is provided an embodiment of a method for recommending financial products. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0043] Figure 1 is a flowchart of an optional method for recommending financial products according to an embodiment of the present invention, as Figure 1 shown, the method includes the following steps:

[0044] Step S101, extract the target feature set of the target user, where the target feature set is obtained by screening the feature set composed of all the features of the target user using a pre-set target feature library.

[0045] Step S102, calculate the user cluster to which the target user belongs based on the feature values of each feature in the target feature set.

[0046] Step S103, when the user cluster belongs to a preset cluster set, query matrix nodes whose relevant value one is greater than or equal to a first preset threshold based on a first type of association matrix, and obtain a first node set, where the first type of association matrix is an association matrix corresponding to the preset cluster set, and the matrix nodes represent the association relationship between each user cluster in the preset cluster set and the product attributes of financial products, and each matrix node corresponds to a relevant value one.

[0047] Step S104, in combination with the first node set, determine the target product attribute set corresponding to the target feature set of the target user.

[0048] Step S105: Extract the product attributes from the target product attribute set, customize the target financial product based on the extracted product attributes, and recommend the target financial product to the target user.

[0049] Through the above steps, the target feature set of the target user can be extracted. Based on the feature values of each feature in the target feature set, the user cluster to which the target user belongs is calculated. When the user cluster belongs to the preset cluster set, the matrix nodes with the relevant value one greater than or equal to the first preset threshold are queried based on the first type of association matrix to obtain the first node set. Combining the first node set, the target product attribute set corresponding to the target feature set of the target user is determined. The product attributes in the target product attribute set are extracted, and the target financial product is customized based on the extracted product attributes, and the target financial product is recommended to the target user. In the embodiment of the present invention, through the association relationship (i.e., the first type of association matrix) between the user cluster and the product attributes of the financial product, the correlation value between the user cluster to which the user belongs and the product attributes can be calculated, so as to obtain the strongly associated product attribute set of the user (i.e., the target product attribute set). Arbitrarily selecting the product attributes in the product attribute set for product customization can realize customizing personalized financial products for users to meet the financial product needs of users, and further solve the technical problem in the related art that customized financial products cannot be recommended for new users without consumption history data, resulting in a reduced experience for new users.

[0050] The embodiments of the present invention will be described in detail below in combination with the above steps.

[0051] In the embodiment of the present invention, the user basic data from multiple sources such as the user information big data platform, the central bank, and external companies can be extracted and integrated. By parsing the user basic data, user features are obtained. Refining and decomposing the user features can form a two-dimensional matrix of user features. For example, the user features can include: general attributes (age, occupation, education level, etc.), financial attributes (total assets, total loan amount, average daily asset balance, average monthly income, average monthly overdraft, etc.), credit attributes (credit score, anti-fraud score, risk level, etc.), etc. The user features can be adjusted according to the actual situation and maintained by business personnel in the user feature library. Since the number of users of financial institutions is in the hundreds of millions, modeling user features of this magnitude can be processed by a distributed computing platform. The user data is divided into different computing nodes and data storages with partitioned distribution through a specific dimension (such as the household registration area), and then a separate data analysis cluster is responsible for extracting and integrating the data to obtain the two-dimensional matrix of user features.

[0052] In this embodiment, the existing financial product data of financial institutions can be pre-integrated. By parsing the financial product data, product attributes are obtained, and the product attributes are refined and decomposed to form a two-dimensional matrix of product attributes. For example, product attributes can be extracted from financial products according to specific dimensions. The product attributes can include: loan carriers (debit cards, credit cards, online loans, etc.), loan forms (cash loans, consumer loans, special auto loans, etc.), loan interest rates (e.g., 3%, 10%, 15%, etc.), repayment periods (months, half years, years, etc.), repayment methods (equal principal, equal principal and interest, variable amount, etc.), loan terms (6, 12, 24, 48, etc.). The product attributes can be flexibly expanded according to the actual situation, and business personnel can maintain them in the product attribute library.

[0053] In this embodiment, user feature modeling (i.e., to obtain a two-dimensional matrix of user features) and financial product attribute modeling (i.e., to obtain a two-dimensional matrix of product attributes) can be repeatedly optimized and iterated through machine learning. Eventually, the higher the market feedback degree of the effectiveness of financial product modeling and production, the higher the effectiveness of user feature and financial product attribute screening. Therefore, in order to improve the effectiveness of modeling, the set of user features and financial product attributes can be continuously adjusted until the optimal goal is achieved.

[0054] Optionally, before extracting the target feature set of the target user, the recommendation method further includes: obtaining the historical consumption data of financial products and multiple historical user features within a historical time period; combining the historical user features in any pairwise manner to obtain multiple user feature combinations; establishing a third type of association matrix between the user feature combinations and the product attributes, where each matrix node in the third type of association matrix refers to the node formed between the user feature combination and the product attribute; based on the historical consumption data, calculating the matrix node value of each matrix node in the third type of association matrix to obtain a first matrix node value distribution diagram, where the matrix node value is the consumption transaction frequency of the user feature combination and the product attribute indicated by the matrix node.

[0055] In the embodiment of the present invention, while introducing user features, the historical consumption data of financial products of users within a historical time period (e.g., the previous year) can be statistically analyzed (i.e., obtaining the historical consumption data of financial products and multiple historical user features within a historical time period). On this basis, the data correlation relationship between user features and product attributes can be calculated, and (user feature - product attribute) combinations with effective correlation degrees can be extracted. Specifically, it can:

[0056] Randomly combine the user features (i.e., combine the historical user features in any pairwise manner to obtain multiple user feature combinations), and then calculate with the product attributes to form an association matrix of user feature combination - product attribute {Z1(X1,X2,Y1),...,Z n (X m,X n ,Y n )} (i.e., establishing a third-type association matrix between user feature combinations and product attributes, where each matrix node in the third-type association matrix refers to a node formed between a user feature combination and a product attribute), where Z1,...,Z n is the name of the incidence matrix, X1,X2,...,X m ,X n represents user features, Y1,...,Y n Represents product attributes. After that, the matrix node value of each matrix node in the third type of association matrix can be calculated based on historical consumption data to obtain the first matrix node value distribution map, where the matrix node can be represented by O(X m ,X n ,Y n ) indicates that the matrix node value is user feature X m ,X n and product attribute Y n The anchored consumption transaction frequency (i.e., the consumption transaction frequency of the matrix node value being the user feature combination and product attribute indicated by the matrix node).

[0057] Optionally, after calculating the matrix node value of each matrix node in the third type of association matrix based on historical consumption data to obtain a first matrix node value distribution map, the recommendation method also includes: sorting the matrix node values ​​based on the first matrix node value distribution map to obtain a second sorting result; based on the second sorting result, extracting matrix node values ​​greater than or equal to a third preset threshold, and combining the extracted matrix node values ​​to obtain a first value queue.

[0058] In the embodiment of the present invention, according to O(X m ,X n ,Y n ) distribution (i.e., the first matrix node value distribution diagram), calculate the user feature X m ,X n and product attribute Y n The data distribution correlation P′(X m ,X n ,Y n )(for example, by calculating the correlation between the two through linear regression), and then sorting the P' values ​​(i.e. sorting the matrix node values ​​to obtain a second sorting result), generating a numerical queue Q'(P') = [P1', P2', P3'...] (i.e., a first numerical queue), wherein P1'>P2'>...>P n ',P n' is a preset valid domain value (i.e., the third preset threshold, such as 30%) (i.e., based on the second sorting result, matrix node values greater than or equal to the third preset threshold are extracted, and the extracted matrix node values are combined to obtain the first numerical queue).

[0059] Optionally, after obtaining the first numerical queue, the recommendation method further includes: preprocessing the first numerical queue to obtain a preprocessed first numerical queue; calculating a combined value of the user feature combination based on the matrix node values in the preprocessed first numerical queue; calculating a total feature value of each user feature based on the combined value; sorting the total feature values to determine target features whose total feature values are greater than or equal to a fourth preset threshold; and forming a target feature library with the target features.

[0060] In the embodiment of the present invention, the first numerical queue Q' can be preprocessed (for example, performing secondary data denoising processing), and user features that have an effective correlation with product attributes are extracted and stored in a library to obtain an available feature library (i.e., the target feature library), specifically as follows:

[0061] Extract all matrix node values P′ in the first numerical queue Q', filter all user feature libraries, and calculate the combined value of the user feature combination according to the following formula (for example, F(X1,X2), F(X1,X n ))), and calculate the total feature value of each user feature based on the combined value (for example, ∑F(X1), ∑F(X n )):

[0062] F(X1,X2) = P′(X1,X2,Y1) + P′(X1,X2,Y2) +... + P′(X1,X2,Y n );

[0063] F(X1,X3) = P′(X1,X3,Y1) + P′(X1,X3,Y2) +... + P′(X1,X3,Y n ); ......

[0065] F(X1,X n ) = P′(X1,X n ,Y1) + P′(X1,X n ,Y2) +... + P′(X1,X n ,Y n ); ......

[0067] ∑F(X1) = F(X1,X2) + F(X1,X3) +... + F(X1,X n );

[0068] F(X2,X1) = P′(X2,X1,Y1) + P′(X2,X1,Y2) +... + P′(X2,X1,Y n );

[0069] F(X2,X3) = P′(X2,X3,Y1) + P′(X2,X3,Y2) +... + P′(X2,X3,Y n ); ......

[0071] F(X2,X n ) = P′(X2,X n ,Y1) + P′(X2,X n ,Y2) +... + P′(X2,X n ,Y n );

[0072] ∑F(X2) = F(X2,X1) + F(X2,X3) +... + F(X2,X n ); ......

[0074] ∑F(X n ) = F(X n ,X1) + F(X n ,X2) +... + F(X n ,X m ).

[0075] Sort the total feature value ∑F(X n ), preset the user feature availability threshold μ (i.e., the fourth preset threshold), remove the user feature X for which ∑F(X n ) < μ, and retain the remaining user feature X to obtain the set S(X) = {X1,..., X n ) as the available feature library (i.e., determine the target features whose total feature value is greater than or equal to the fourth preset threshold, and form the target feature library with the target features).

[0076] Optionally, after forming the target feature library with the target features, the recommendation method further includes: determining a sliding window corresponding to each user cluster based on all the features in the target feature library; randomly selecting a center point based on the sliding window and sliding the center point within the sliding window until the highest density point is found, and representing the highest density point as the new center point; after repeatedly selecting new center points for the sliding window, if the positions of the selected center points are all within a certain fixed area during the target selection time period, then determine that the center point selection stability reaches the preset stable condition; determining the center points corresponding to each user cluster to obtain multiple cluster center points; and obtaining a cluster set based on the multiple cluster center points.

[0077] In an embodiment of the present invention, after obtaining the target feature library, the mean shift clustering algorithm can be used to obtain a clustering set, specifically as follows:

[0078] Based on the user features in S(X), a sliding window is determined (i.e., based on all the features in the target feature library, a sliding window corresponding to each user cluster is determined), where the radius of the sliding window A center point c is randomly selected and gradually slides towards a higher density space within the space defined by S(X) until the highest density window is found as a new center point (i.e., the center point slides within the sliding window until the highest density point is found, and the highest density point is characterized as the new center point). Specifically, in each iteration, the sliding window moves towards a higher density region by moving the center point towards the mean of the points within the window (i.e., the average distance from each point within the sliding window to the center point). The density within the sliding window is proportional to the number of points inside it. Therefore, by moving towards the mean of the points within the window, it gradually moves towards a region with a higher point density.

[0079] After repeating the above process multiple times, the center point remains stable (i.e., if the position of the selected center point is within a certain fixed region during the target selection time period, it is determined that the stability of the center point selection reaches the preset stable condition), and multiple new center points c' are obtained (determining the center points corresponding to each user cluster, and obtaining multiple clustering center points), that is, the clustering analysis of the user group is completed, and the user clustering set S(C) is obtained (based on multiple clustering center points, the clustering set is obtained).

[0080] Optionally, after obtaining the clustering set based on multiple clustering center points, the recommendation method further includes: based on the clustering set and historical consumption data, obtaining a two-dimensional matrix between the clustering set and financial products, where the matrix node value corresponding to the matrix node in the two-dimensional matrix is the trading volume.

[0081] In an embodiment of the present invention, the user clustering set S(C) and historical consumption data can be associated to obtain a two-dimensional matrix Z(C, D) between the clustering set and financial products, where the matrix node value corresponding to the matrix node in the two-dimensional matrix is the trading volume T, C represents the user cluster, and D represents the financial product.

[0082] As shown in Table 1, it is an optional association analysis table of user clustering and financial products in this embodiment. In this embodiment, through a distributed computing platform, all users can be clustered and analyzed according to the big data clustering algorithm, and then the association degree (such as the number of transaction pens) between user features and financial products is sorted to extract matching relationships with a higher association strength.

[0083] Table 1

[0084] Financial Product 1 Financial Product 2 Financial Product 3 Financial Product 4 User Cluster 1 100 transactions 65 transactions ……… ……… User Cluster 2 0 transactions 150 transactions ……… ……… User Cluster 3 30 transactions 5 transactions ……… ……… User Cluster 4 200 transactions 300 transactions ……… ……… ……… ……… ……… ……… ………

[0085] Optionally, before extracting the target feature set of the target user, the recommendation method further includes: obtaining historical consumption data of financial products within a historical time period; establishing a fourth type of association matrix between historical user features and product attributes; calculating a matrix node value for each matrix node in the fourth type of association matrix based on the historical consumption data to obtain a second matrix node value distribution diagram, where the matrix node value is the consumption transaction frequency of the user feature and product attribute indicated by the matrix node; sorting the matrix node values based on the second matrix node value distribution diagram to obtain a third sorting result; and extracting matrix node values greater than or equal to a fifth preset threshold based on the third sorting result, and combining the extracted matrix node values to obtain a second numerical queue.

[0086] In an embodiment of the present invention, while introducing user features, historical consumption data of financial products of a user within a historical time period (for example, the previous year) can be statistically obtained (i.e., obtaining historical consumption data of financial products within a historical time period). On this basis, the data correlation relationship between user features and product attributes can be calculated, and (user feature - product attribute) combinations with effective correlation degrees can be extracted. Specifically,

[0087] The user features and financial products are split by attribute according to dimensions to form multiple association matrices of user features - product attributes {Z1(X1, Y1),..., Z n (X n , Y n )} (i.e., establishing a fourth type of association matrix between historical user features and product attributes), where Z1,..., Z n are the names of the association matrices, X1,..., X n represent user features, and Y1,..., Y n represent product attributes. Then, based on the historical consumption data, a matrix node value for each matrix node in the fourth type of association matrix can be calculated to obtain a second matrix node value distribution diagram, where the matrix node can be represented by O(X n , Y n ), and the matrix node value is the consumption transaction frequency anchored by the user feature X n and the product attribute Y n (i.e., the matrix node value is the consumption transaction frequency of the user feature and product attribute indicated by the matrix node). The data scatter correlation degree P(X n , Y n ) between the user feature X n and the product attribute Y n can be calculated based on the distribution of O(X n , Y n)(For example, calculate the correlation between the two through linear regression), and then sort the P values (that is, sort the matrix node values to obtain the third sorting result), generating a numerical queue Q(P) = [P1, P2, P3...](that is, the second numerical queue), where, P1 > P2 >... > P n , P n is a preset valid domain value (that is, the fifth preset threshold, such as 30%)(that is, based on the third sorting result, extract the matrix node values greater than or equal to the fifth preset threshold, and combine the extracted matrix node values to obtain the second numerical queue).

[0088] Optionally, after obtaining the two-dimensional matrix between the clustering set and the products, the recommendation method further includes: based on the two-dimensional matrix between the clustering set and the financial products, decompose the financial products according to the product attributes to obtain the first type of association matrix of user clustering and product attributes; based on the second numerical queue, the user characteristics in the user clustering, and the product attributes, calculate the matrix node values of each matrix node in the first type of association matrix, where the matrix node value is characterized as the correlation value one.

[0089] In the embodiment of the present invention, according to the two-dimensional matrix Z(C, D) between the clustering set and the financial products, decompose the financial products according to the product attributes, calculate the association matrix Z(C, Y)(that is, the first type of association matrix) of user clustering and product attributes, and according to the user characteristics {X1, X2,..., X n} in the user clustering C and the product attribute Y, look up the P value in the second numerical queue, and calculate the matrix node values of each matrix node in the first type of association matrix (the matrix node value is characterized as the correlation value one) P(C, Y n ) = P(X1, Y n ) +... + P(X n , Y n ).

[0090] As shown in Table 2, it is another optional association analysis table of user clustering and product attributes in this embodiment, which can sort the correlation degree between user clustering and product attributes, and extract the matching relationships with a correlation degree higher than the preset domain value.

[0091] Table 2

[0092] Product Attribute 1 Product Attribute 2 Product Attribute 3 Product Attribute 4 User Cluster 1 Relevance 40% Relevance 95% ……… ……… User Cluster 2 Relevance 0% Relevance 60% ……… ……… User Cluster 3 Relevance 55% Relevance 5% ……… ……… User Cluster 4 Relevance 35% Relevance 65% ……… ……… ……… ……… ……… ……… ………

[0093] Step S101, extract the target feature set of the target user, where the target feature set is obtained by screening the feature set composed of all the features of the target user using a preset target feature library.

[0094] In an embodiment of the present invention, a feature queue {X1, X2,..., X m} of the user can be extracted according to all the data accessible to the target user, and an available feature set s(X) of the user (i.e., the target feature set, which is obtained by filtering the feature set composed of all the features of the target user using a preset target feature library) can be filtered out from the available feature library S(X) (i.e., the target feature library). If the available data of the user is incomplete, s(X) is a subset of S(X).

[0095] Step S102: Calculate the user cluster to which the target user belongs based on the feature values of each feature in the target feature set.

[0096] In an embodiment of the present invention, the user cluster C to which the user belongs can be calculated according to the data drop point intervals of the elements in the user available feature set s(X) (i.e., the feature values of each feature in the target feature set).

[0097] Step S103: When the user cluster belongs to a preset cluster set, query matrix nodes whose associated value one is greater than or equal to a first preset threshold based on a first type of association matrix, to obtain a first node set, where the first type of association matrix is an association matrix corresponding to the preset cluster set, and the matrix nodes represent the association relationships between each user cluster in the preset cluster set and the product attributes of financial products, and each matrix node corresponds to an associated value one.

[0098] In an embodiment of the present invention, if the user belongs to the user cluster set S(C) (i.e., when the user cluster belongs to the preset cluster set), filter the top N% of the product feature Y values according to the P values of the matrix nodes of the association matrix Z(C, Y) (i.e., the first type of association matrix) between the user cluster and the product attributes, to obtain a strong association product attribute set S{Y1, Y2,..., Y n} (i.e., query matrix nodes whose associated value one is greater than or equal to a first preset threshold based on the first type of association matrix, to obtain a first node set, and the product attributes corresponding to each matrix node in the first node set form the strong association product attribute set), where the first type of association matrix is an association matrix corresponding to the preset cluster set S(C), the matrix nodes represent the association relationships between each user cluster in the preset cluster set and the product attributes of financial products, and each matrix node corresponds to an associated value one (i.e., the P value).

[0099] Optionally, after calculating the user cluster to which the target user belongs based on the feature values of each feature in the target feature set, the recommendation method further includes: when the user cluster does not belong to the preset cluster set, determining a second type of association matrix based on the association relationship between each user feature in the target feature set and the product attributes, where each matrix node in the second type of association matrix is a node represented by a user feature and a product attribute, and the node value corresponding to each matrix node represents the correlation value two between the user feature and the product attribute; sorting the correlation values two of all matrix nodes in the second type of association matrix to obtain a first sorting result; based on the first sorting result, selecting matrix nodes whose correlation value two is greater than or equal to a second preset threshold; and combining the product attributes corresponding to the matrix nodes into a target product attribute set.

[0100] In an embodiment of the present invention, if the user does not belong to the user cluster set S(C) (i.e., when the user cluster does not belong to the preset cluster set), then according to the user feature set s(X) (i.e., the target feature set), an association matrix z(X, Y) between the user feature and the product attribute is calculated (i.e., determining a second type of association matrix based on the association relationship between each user feature in the target feature set and the product attribute), where ZЭz(X, Y) indicates that the association matrix z(X, Y) belongs to one of the multiple association matrices Z of user feature - product attribute. Each matrix node in the second type of association matrix (i.e., the association matrix z(X, Y)) is a node represented by a user feature and a product attribute, and the node value corresponding to each matrix node represents the correlation value two between the user feature and the product attribute (i.e., the data dispersion correlation degree P(o) between the user feature and the product attribute). Sorting is performed according to the P(o) value (i.e., sorting the correlation values two of all matrix nodes in the second type of association matrix to obtain a first sorting result), and all nodes with P(o) >= valid domain value (e.g., 50%) are selected (i.e., based on the first sorting result, selecting matrix nodes whose correlation value two is greater than or equal to a second preset threshold), and the product attributes corresponding to the nodes are obtained. On this basis, the strong association product attribute set S{Y1, Y2,..., Y n}(i.e., combining the product attributes corresponding to the matrix nodes into a target product attribute set, and this target product attribute set is the strong association product attribute set).

[0101] Step S104: Determine the target product attribute set corresponding to the target feature set of the target user in combination with the first node set.

[0102] In an embodiment of the present invention, the product attributes corresponding to each matrix node in the first node set form a strong association product attribute set (i.e., the target product attribute set corresponding to the target feature set of the target user can be determined)

[0103] Step S105: Extract the product attributes from the target product attribute set, customize the target financial product based on the extracted product attributes, and recommend the target financial product to the target user.

[0104] In the embodiment of the present invention, the product attributes in the target product attribute set are extracted, and a random combination algorithm is used for these product attributes to generate multiple product attribute combinations (Y m +...+Y n ), which is the customized product model. Then, the business personnel can randomly select N% of the product models from the customized product model pool, and on this basis, generate customized financial products (that is, customize the target financial product based on the extracted product attributes), and recommend the customized financial products (that is, the target financial products) to the target users.

[0105] As shown in Table 3, it is an optional financial product model table in this embodiment. In this embodiment, based on the extracted effective user feature library and product attribute library, data combination analysis is carried out, potential better combination relationships are calculated, the product model set is sorted according to the effectiveness, new product models are automatically generated, and the characteristics of the customer groups they face can be calculated. And this embodiment supports inputting external constraint conditions, such as interest rate restrictions and credit score restrictions, to perform boundary constraints on the new product models.

[0106] Table 3

[0107]

[0108]

[0109] In the embodiment of the present invention, by combining various technical means such as big data analysis, distributed computing, data modeling, and machine learning algorithms, it is possible to achieve automated and customized production of financial products, solve the contradiction between the rapidly growing demand for financial products and the supply capacity of financial products, provide financial institutions with the ability to quickly produce customized financial products, and achieve the following beneficial effects:

[0110] (1) Based on the user information big data platform, and flexible introduction of big data from the central bank and external companies is possible. User data is three-dimensional, multi-dimensional, and holographic, and the available user feature library can be continuously optimized through machine learning means. This embodiment relies on the distributed computing platform and has strong big data analysis capabilities, fully ensuring the real-time performance, accuracy, and fullness of user feature modeling.

[0111] (2) It can meet the rapid customization requirements of financial products for financial institution users. Relying on the rapid modeling and clustering analysis capabilities of user data, accurately locate the strong correlation relationship between users and product attributes, and quickly build a customized financial product model through machine learning algorithms to achieve automated production of financial products.

[0112] (3) While implementing automated financial product customization, this embodiment provides a recommendation function. For problems such as user data cold start and data sparsity existing in the existing recommendation systems, this embodiment can well solve these problems, and calculate the financial product requirements for new users without consumption history data and recommend customized financial products for new users.

[0113] Embodiment Two

[0114] Figure 2 is a flowchart of an optional customized production of financial products according to an embodiment of the present invention, as Figure 2 shown, and includes the following steps:

[0115] (1) Extract and integrate the user basic data from multiple sources such as the customer information big data platform, the People's Bank of China, and external companies, and perform user feature modeling;

[0116] (2) Integrate the existing financial product data of financial institutions, extract product attributes according to specific dimensions, and perform product attribute modeling;

[0117] (3) Obtain the financial product consumption history data, analyze the correlation between user features and financial product attributes, and extract the combinations of user features and financial product attributes with effective correlation;

[0118] (4) Based on the correlation between user features and financial product attributes, extract the user features with effective correlation with financial product attributes into the database, and then use the effective user feature database as a measurement factor to obtain a user clustering set with significant distinctiveness through a clustering algorithm;

[0119] (5) Perform correlation analysis through the user clustering set and the financial product consumption history data, and calculate the two-dimensional matrix of user clustering and financial products;

[0120] (5) According to the two-dimensional matrix of user clustering and financial products, decompose the financial products according to product attributes, and calculate the correlation matrix of user clustering and financial product attributes;

[0121] (6) According to the user feature data of new users and some existing users, and the correlation matrix of user clustering and financial product attributes, perform product attribute combination operations to perform financial product customization modeling;

[0122] (7) Recommend and consume the customized new financial products, and use the financial product consumption data generated by the consumption in the next analysis of the correlation between user features and financial product attributes.

[0123] In the embodiments of the present invention, it is possible to achieve automated and customized production of financial products, which can solve the contradiction between the rapidly growing demand for financial products and the supply capacity of financial products, provide financial institutions with the ability to quickly produce customized financial products, thereby realizing the function of meeting the financial product needs of new users, calculating the financial product needs of new users, and recommending customized financial products to new users.

[0124] Embodiment III

[0125] A financial product recommendation device provided in this embodiment includes multiple implementation units, and each implementation unit corresponds to each implementation step in Embodiment I above.

[0126] Figure 3 It is a schematic diagram of an optional financial product recommendation device according to an embodiment of the present invention. As Figure 3 shown, the recommendation device may include: an extraction unit 30, a calculation unit 31, a query unit 32, a determination unit 33, and a recommendation unit 34, where

[0127] The extraction unit 30 is used to extract a target feature set of a target user, where the target feature set is obtained by screening the feature set composed of all features of the target user using a pre-set target feature library;

[0128] The calculation unit 31 is used to calculate the user cluster to which the target user belongs based on the feature values of each feature in the target feature set;

[0129] The query unit 32 is used to, when the user cluster belongs to a preset cluster set, query matrix nodes whose relevant value one is greater than or equal to a first preset threshold based on a first type of association matrix, and obtain a first node set, where the first type of association matrix is an association matrix corresponding to the preset cluster set, the matrix nodes represent the association relationships between each user cluster in the preset cluster set and the product attributes of financial products, and each matrix node corresponds to a relevant value one;

[0130] The determination unit 33 is used to determine a target product attribute set corresponding to the target feature set of the target user in combination with the first node set;

[0131] The recommendation unit 34 is used to extract the product attributes in the target product attribute set, customize a target financial product based on the extracted product attributes, and recommend the target financial product to the target user.

[0132] The above-mentioned recommendation device can extract the target feature set of the target user through the extraction unit 30, calculate the user cluster to which the target user belongs based on the feature values of each feature in the target feature set through the calculation unit 31, and query in the case that the user cluster belongs to the preset cluster set through the query unit 32. Matrix nodes with relevant value one greater than or equal to the first preset threshold are retrieved from the first type of association matrix to obtain the first node set. The determination unit 33 combines the first node set to determine the target product attribute set corresponding to the target feature set of the target user. The recommendation unit 34 extracts the product attributes in the target product attribute set, customizes the target financial product based on the extracted product attributes, and recommends the target financial product to the target user. In the embodiment of the present invention, the correlation value between the user cluster to which the user belongs and the product attribute can be calculated through the association relationship (i.e., the first type of association matrix) between the user cluster and the product attribute of the financial product, so as to obtain the strong association product attribute set of the user (i.e., the target product attribute set). Arbitrarily selecting the product attributes in the product attribute set for product customization can realize customizing personalized financial products for users to meet the financial product needs of users, and further solves the technical problem in the related art that it is impossible to recommend customized financial products for new users without consumption history data, resulting in a reduced experience for new users.

[0133] Optionally, the recommendation device further includes: a first determination module, configured to, after calculating the user cluster to which the target user belongs based on the feature values of each feature in the target feature set, determine a second type of association matrix based on the association relationship between each user feature in the target feature set and the product attribute in the case that the user cluster does not belong to the preset cluster set, where each matrix node in the second type of association matrix is a node represented by the user feature and the product attribute, and the node value corresponding to each matrix node represents the second correlation value between the user feature and the product attribute; a first sorting module, configured to sort the second correlation values of all matrix nodes in the second type of association matrix to obtain a first sorting result; a first selection module, configured to select matrix nodes with a second correlation value greater than or equal to a second preset threshold based on the first sorting result; and a first combination module, configured to combine the product attributes corresponding to the matrix nodes into a target product attribute set.

[0134] Optionally, the recommendation device further includes: a first acquisition module, configured to acquire historical consumption data of financial products and multiple historical user features within a historical time period before extracting the target feature set of the target user; a second combination module, configured to arbitrarily combine the historical user features in pairs to obtain multiple user feature combinations; a first establishment module, configured to establish a third type of association matrix between the user feature combinations and the product attributes, where each matrix node in the third type of association matrix refers to a node formed between the user feature combination and the product attribute; a first calculation module, configured to calculate the matrix node value of each matrix node in the third type of association matrix based on the historical consumption data to obtain a first matrix node value distribution diagram, where the matrix node value is the consumption transaction frequency of the user feature combination and the product attribute indicated by the matrix node.

[0135] Optionally, the recommendation device further includes: a second sorting module, configured to sort the matrix node values based on the first matrix node value distribution diagram after calculating the matrix node value of each matrix node in the third type of association matrix based on the historical consumption data to obtain a second sorting result; a first extraction module, configured to extract the matrix node values greater than or equal to a third preset threshold based on the second sorting result and combine the extracted matrix node values to obtain a first numerical queue.

[0136] Optionally, the recommendation device further includes: a first processing module, configured to preprocess the first numerical queue after obtaining the first numerical queue to obtain a preprocessed first numerical queue; a second calculation module, configured to calculate the combined value of the user feature combinations based on the matrix node values in the preprocessed first numerical queue; a third calculation module, configured to calculate the total feature value of each user feature based on the combined value; a third sorting module, configured to sort the total feature values to determine the target features whose total feature values are greater than or equal to a fourth preset threshold; a third combination module, configured to form a target feature library with the target features.

[0137] Optionally, the recommendation device further includes: a second determination module, configured to determine a sliding window corresponding to each user cluster based on all the features in the target feature library after forming the target feature library with the target features; a second selection module, configured to randomly select a center point based on the sliding window and slide the center point within the sliding window until the highest density point is found, and represent the highest density point as the new center point; a third determination module, configured to, after selecting a new center point for the sliding window multiple times, if the positions of the selected center points are all within a certain fixed area within the target selection time period, determine that the stability of the center point selection reaches a preset stable condition; a fourth determination module, configured to determine the center points corresponding to each user cluster to obtain multiple clustering center points; a first output module, configured to obtain a clustering set based on the multiple clustering center points.

[0138] Optionally, the recommendation device further includes: a second output module, configured to, after obtaining a clustering set based on a plurality of clustering center points, obtain a two-dimensional matrix between the clustering set and financial products based on the clustering set and historical consumption data, where a matrix node value corresponding to a matrix node in the two-dimensional matrix is a trading volume.

[0139] Optionally, the recommendation device further includes: a second acquisition module, configured to obtain historical consumption data of financial products within a historical time period before extracting a target feature set of a target user; a second establishment module, configured to establish a fourth type of association matrix between historical user features and product attributes; a fourth calculation module, configured to calculate a matrix node value of each matrix node in the fourth type of association matrix based on the historical consumption data to obtain a second matrix node value distribution map, where the matrix node value is a consumption transaction frequency of the user features and product attributes indicated by the matrix node; a fourth sorting module, configured to sort the matrix node values based on the second matrix node value distribution map to obtain a third sorting result; a second extraction module, configured to extract matrix node values greater than or equal to a fifth preset threshold based on the third sorting result, and combine the extracted matrix node values to obtain a second numerical queue.

[0140] Optionally, the recommendation device further includes: a first decomposition module, configured to, after obtaining a two-dimensional matrix between a clustering set and products, decompose financial products according to product attributes based on the two-dimensional matrix between the clustering set and financial products to obtain a first type of association matrix between user clusters and product attributes; a fifth calculation module, configured to calculate a matrix node value of each matrix node in the first type of association matrix based on the second numerical queue, user features in the user cluster, and product attributes, where the matrix node value is characterized as a correlation value one.

[0141] The above-mentioned recommendation device may further include a processor and a memory. The above-mentioned extraction unit 30, calculation unit 31, query unit 32, determination unit 33, recommendation unit 34, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to implement corresponding functions.

[0142] The above-mentioned processor includes a kernel, and the kernel retrieves the corresponding program units from the memory. The kernel can be set to one or more, and by adjusting the kernel parameters, customize target financial products based on the extracted product attributes and recommend the target financial products to the target user.

[0143] The above-mentioned memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of, for example, read-only memory (ROM) or flash memory (flash RAM), and the memory includes at least one storage chip.

[0144] The present application also provides a computer program product which, when executed on a data processing device, is adapted to execute a program initialized with the following method steps: extracting a target feature set of a target user, calculating a user cluster to which the target user belongs based on the feature values of each feature in the target feature set, in the case where the user cluster belongs to a preset cluster set, querying matrix nodes whose relevant value one is greater than or equal to a first preset threshold based on a first type of association matrix to obtain a first node set, combining the first node set to determine a target product attribute set corresponding to the target feature set of the target user, extracting product attributes in the target product attribute set, customizing a target financial product based on the extracted product attributes, and recommending the target financial product to the target user.

[0145] According to another aspect of the embodiments of the present invention, there is also provided a computer-readable storage medium. The computer-readable storage medium includes a stored computer program, wherein when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the above-mentioned financial product recommendation method.

[0146] According to another aspect of the embodiments of the present invention, there is also provided an electronic device, including one or more processors and a memory. The memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned financial product recommendation method.

[0147] Figure 4 is a hardware structure block diagram of an electronic device (or mobile device) for a financial product recommendation method according to an embodiment of the present invention. As Figure 4 shown, the electronic device may include one or more (shown as 102a, 102b,..., 102n in the figure) processors 102 (the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA), and a memory 104 for storing data. In addition, it may further include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a keyboard, a power supply, and / or a camera. Those of ordinary skill in the art can understand that Figure 4 the structure shown is only schematic and does not limit the structure of the above-mentioned electronic device. For example, the electronic device may further include more or fewer components than Figure 4 shown, or have a different configuration from Figure 4 shown.

[0148] The above serial numbers of the embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0149] In the above embodiments of the present invention, the descriptions of the various embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0150] In several embodiments provided by the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only illustrative. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the units or modules can be in electrical or other forms.

[0151] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0152] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0153] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, 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. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: USB flash drives, read-only memories (ROMs), random access memories (RAMs), mobile hard disks, magnetic disks or optical disks and other various media that can store program codes.

[0154] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for recommending financial products, characterized in that, Including: Extracting a target feature set of a target user, where the target feature set is obtained by screening a feature set composed of all features of the target user using a preset target feature library; Calculating the user cluster to which the target user belongs based on the feature values of each feature in the target feature set; When the user cluster belongs to a preset cluster set, querying matrix nodes whose relevant value one is greater than or equal to a first preset threshold based on a first type of association matrix, to obtain a first node set, where the first type of association matrix is an association matrix corresponding to the preset cluster set, the matrix nodes represent the association relationship between each user cluster in the preset cluster set and the product attributes of financial products, the relevant value one refers to the matrix node value of each matrix node in the first type of association matrix calculated based on the historical consumption data of the financial product, the user features in the user cluster, and the product attributes, and each matrix node corresponds to the relevant value one; Combining the first node set to determine a target product attribute set corresponding to the target feature set of the target user; Extracting the product attributes in the target product attribute set, customizing a target financial product based on the extracted product attributes, and recommending the target financial product to the target user.

2. The recommendation method according to claim 1, wherein After calculating the user cluster to which the target user belongs based on the feature values of each feature in the target feature set, the recommendation method further includes: When the user cluster does not belong to the preset cluster set, determining a second type of association matrix based on the association relationship between each user feature in the target feature set and the product attributes, where each matrix node in the second type of association matrix is a node represented by the user feature and the product attribute, and the node value corresponding to each matrix node represents the relevant value two between the user feature and the product attribute; Sorting the relevant value two of all the matrix nodes in the second type of association matrix to obtain a first sorting result; Selecting matrix nodes whose relevant value two is greater than or equal to a second preset threshold based on the first sorting result; Combining the product attributes corresponding to the matrix nodes into a target product attribute set.

3. The recommendation method according to claim 1, wherein Before extracting the target feature set of the target user, the recommendation method further includes: Obtaining the historical consumption data of financial products and multiple historical user features within a historical time period; Arbitrarily combining the historical user features in pairs to obtain multiple user feature combinations; Establishing a third type of association matrix between the user feature combinations and the product attributes, where each matrix node in the third type of association matrix refers to a node formed by the user feature combination and the product attribute; Calculating the matrix node value of each matrix node in the third type of association matrix based on the historical consumption data to obtain a first matrix node value distribution map, where the matrix node value is the consumption transaction frequency of the user feature combination and the product attribute indicated by the matrix node.

4. The recommendation method according to claim 3, wherein After calculating the matrix node values of each matrix node in the third type of association matrix based on the historical consumption data and obtaining the first matrix node value distribution map, the recommendation method further includes: Sorting the matrix node values based on the first matrix node value distribution map to obtain a second sorting result; Extracting the matrix node values greater than or equal to a third preset threshold based on the second sorting result, and combining the extracted matrix node values to obtain a first numerical queue.

5. The recommendation method according to claim 4, wherein After obtaining the first numerical queue, the recommendation method further includes: Preprocessing the first numerical queue to obtain a processed first numerical queue; Calculating the combined value of the user feature combination based on the matrix node values in the processed first numerical queue; Calculating the total feature value of each user feature based on the combined value; Sorting the total feature values to determine the target features whose total feature values are greater than or equal to a fourth preset threshold; Forming the target feature library with the target features.

6. The recommendation method according to claim 5, characterized in that After forming the target feature library with the target features, the recommendation method further includes: Determining a sliding window corresponding to each user cluster based on all the features in the target feature library; Randomly selecting a center point based on the sliding window, and sliding the center point within the sliding window until the highest density point is found, and representing the highest density point as a new center point; After repeatedly selecting new center points for the sliding window, if the positions of the selected center points are within a certain fixed area during the target selection time period, it is determined that the center point selection stability reaches a preset stable condition; Determining the center points corresponding to each user cluster to obtain a plurality of cluster center points; Obtaining a cluster set based on the plurality of cluster center points.

7. The recommendation method according to claim 6, wherein After obtaining a cluster set based on the plurality of cluster center points, the recommendation method further includes: Obtaining a two-dimensional matrix between the cluster set and financial products based on the cluster set and the historical consumption data, where the matrix node value corresponding to the matrix node in the two-dimensional matrix is the trading volume.

8. The recommended method according to claim 1, characterized in that, Before extracting the target feature set of the target user, the recommendation method further includes: Obtaining the historical consumption data of financial products within a historical time period; Establishing a fourth type of association matrix between historical user features and the product attributes; Calculating the matrix node values of each matrix node in the fourth type of association matrix based on the historical consumption data to obtain a second matrix node value distribution map, where the matrix node value is the consumption transaction frequency of the user features and product attributes indicated by the matrix node; Sorting the matrix node values based on the second matrix node value distribution map to obtain a third sorting result; Extracting the matrix node values greater than or equal to a fifth preset threshold based on the third sorting result, and combining the extracted matrix node values to obtain a second numerical queue.

9. The recommendation method according to claim 8, wherein After obtaining the second numerical queue, the recommendation method further includes: Decomposing the financial products according to the product attributes based on the two-dimensional matrix between the cluster set and financial products to obtain a first type of association matrix between user clusters and product attributes; Based on the second numerical queue, the user characteristics in the user clustering, and the product attributes, calculate the matrix node values of each matrix node in the first type of association matrix, where the matrix node value is characterized as the correlation value one.

10. A recommendation device for a financial product, characterized in that, Including: An extraction unit for extracting a target feature set of a target user, where the target feature set is obtained by screening the feature set composed of all the features of the target user using a pre-set target feature library; A calculation unit for calculating the user clustering to which the target user belongs based on the feature values of each feature in the target feature set; A query unit for, when the user clustering belongs to a preset clustering set, querying matrix nodes in the first type of association matrix whose correlation value one is greater than or equal to a first preset threshold to obtain a first node set, where the first type of association matrix is an association matrix corresponding to the preset clustering set, the matrix node represents the association relationship between each user clustering in the preset clustering set and the product attributes of financial products, the correlation value one refers to the matrix node value of each matrix node in the first type of association matrix calculated based on the historical consumption data of the financial products, the user characteristics in the user clustering, and the product attributes, and each matrix node corresponds to the correlation value one; A determination unit for determining a target product attribute set corresponding to the target feature set of the target user in combination with the first node set; A recommendation unit for extracting the product attributes in the target product attribute set, customizing a target financial product based on the extracted product attributes, and recommending the target financial product to the target user.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, where, when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the financial product recommendation method according to any one of claims 1 to 9.

12. An electronic device, characterized in that, Including one or more processors and a memory, the memory is used to store one or more programs, where, when the one or more programs are executed by the one or more processors, the one or more processors implement the financial product recommendation method according to any one of claims 1 to 9.

Images