Consumption financial commodity recommendation method and system based on graph clustering of granular ball calculation

Abstract

The invention belongs to the field of machine learning and graph data mining, and relates to a consumption financial commodity recommendation method based on graph clustering of granular ball calculation. The method comprises the steps of obtaining a financial transaction network graph data set and performing information extraction on the financial transaction network graph data set to obtain graph data G; carrying out granular ball calculation on the graph data G through a self-adaptive splitting mechanism, and finally generating a plurality of granular balls; edges are added between the pellets according to the adjacency relation between the pellets, and a granulation graph is generated; constructing an objective function of a particle-sphere similarity matrix; solving an objective function of the granular-spherical similarity matrix to obtain the granular-spherical similarity matrix; performing symmetric non-negative on the particle-ball similarity matrix, and performing spectral clustering on the symmetric non-negative particle-ball similarity matrix to obtain a commodity clustering result; according to the method, transaction graph data are preliminarily covered and represented by high-quality granular balls generated through an adaptive splitting mechanism, fine-grained graph nodes are replaced by coarse-grained granular balls to perform a clustering task, global and local features are captured through multi-granularity representation, and thus rapid and accurate node clustering is realized.

Description

Technical Field

[0001] The present invention belongs to the field of machine learning and graph data mining, and relates to a consumer finance product recommendation method and system based on graph clustering using granular sphere computing. Background Art

[0002] In the consumer sector, transaction networks are an important type of data structure that is widely used in customer behavior analysis and product recommendations. In transaction networks, customers are nodes and transaction relationships are edges. These networks usually have complex structures and a large number of interactions between nodes. For such complex graph structures, traditional K-means and hierarchical clustering methods are difficult to effectively process the global and local structural information of nodes. Therefore, graph clustering technology has gradually become one of the mainstream methods for product recommendations in transaction networks.

[0003] Currently, there are mainly the following technologies:

[0004] Clustering based on graph partitioning: The transaction network is divided into multiple subgraphs through a graph partitioning algorithm, each of which represents a customer group. Common graph partitioning methods include spectral clustering and community detection. These methods are usually implemented by optimizing the cutting of the graph, such as minimizing the number of edges across groups and maximizing the closeness of customer transaction relationships within a group.

[0005] Clustering based on graph embedding: The graph embedding method embeds customer nodes in the transaction network into a low-dimensional vector space and then clusters these embedded vectors. This method can compress the high-dimensional network structure into a low-dimensional space, thereby capturing the complex relationship between nodes, and is suitable for customer segmentation and behavior prediction.

[0006] Clustering based on graph neural network (GNN): Graph neural network can learn the representation of each customer by aggregating the neighborhood information of nodes and perform clustering based on these representations. GNN model can capture the complex structure in the transaction network, especially when dealing with large-scale networks.

[0007] Clustering based on hybrid methods: Some studies combine multiple technical means, such as graph embedding combined with time series analysis, or hybrid methods of GNN and community detection. By integrating information from multiple dimensions, these methods can more comprehensively cluster customers in the consumer network.

[0008] In summary, the graph clustering methods in the existing transaction network product recommendation methods lack multi-granularity analysis and have high computational complexity. They usually operate at a single granularity, lack hierarchical modeling capabilities, and may ignore certain global information. At the same time, due to their fine-grained input, when facing complex structured consumer data, high-order convolution or attention calculations will have large memory overhead and computational complexity, making it difficult to process and significantly increasing the computational cost. Summary of the invention

[0009] In order to solve the above-mentioned prior art problems, the present invention adopts a consumer finance product recommendation method based on graph clustering of granular sphere computing, which includes:

[0010] S1. Obtain a financial transaction network graph data set, extract information from the financial transaction network graph data set, and obtain extracted graph data G; the financial transaction network graph data includes commodities, commodity information, and the relationship between commodities;

[0011] S2. Perform granular calculation on the graph data G through an adaptive splitting mechanism to generate multiple granular balls; each granular ball includes multiple commodities with similar structures;

[0012] S3, add edges between the granules according to the adjacency relationship between the granules, generate a granulation graph, calculate the adjacency matrix and characteristic matrix of the granulation graph, and obtain the granulation graph G′=(V GB , E GB , X GB , A GB ), where V GB is the node set of the granular graph, each node represents a sphere, X GB is the characteristic matrix of the granulation graph, E GB is the edge set of the granulated graph, A GB is the adjacency matrix of the granulated graph;

[0013] S4, constructing an objective function of the particle-sphere similarity matrix; solving the objective function of the particle-sphere similarity matrix to obtain the particle-sphere similarity matrix;

[0014] S5, symmetric non-negative particle-sphere similarity matrix, and spectral clustering is performed on the symmetric non-negative particle-sphere similarity matrix to obtain the product clustering result;

[0015] S6. Recommend products based on the product clustering results.

[0016] The particle-sphere calculation for the graph data G includes:

[0017] S21. Treat all commodity nodes in the graph data G as a sphere and perform Split, get Initial sphere Where f is the index of the initial sphere, and n is the number of commodity nodes in the graph data G;

[0018] S22, according to the principle of proximity, each initial ball gb f Perform binary fission to obtain each initial sphere gb f The two balls gb 1，f 、gb 2，f ;

[0019] S23, judge the two balls gb 1，f 、gb 2，f Is the sum of the densities lower than the corresponding parent ball gb f The density of the parent ball is gb. f Otherwise, return to step S22 and calculate the two sub-balls gb according to the principle of proximity. 1，f 、gb 2，f Perform binary fission.

[0020] Treat all commodity nodes in the graph data G as a sphere and The split includes: taking the product of the commodity node degree and the shortest path distance between commodity nodes as the index of the center point, The commodity node is taken as the initial center point, and the remaining commodity nodes are divided into the initial center point according to the shortest path principle. Initial sphere gb f .

[0021] The binary split of the initial sphere according to the proximity principle includes: selecting two commodity nodes with the farthest path distance in the initial sphere as temporary center points, calculating the shortest path distance between the remaining commodity nodes in the initial sphere and the temporary center point, and adding the remaining commodity nodes in the initial sphere to the temporary center point closest to them to obtain two sub-spheres.

[0022] The calculation process of the density of the sphere includes: selecting two commodity nodes with the farthest path distance in the sphere, taking half of the path distance between the two commodity nodes as the radius of the sphere, dividing the number of commodity nodes in the sphere by the radius of the sphere, and obtaining the density of the sphere.

[0023] Granular ball GB in granulation diagram i The characteristic matrix Among them, x k represents the i-th ball GB i The feature corresponding to the inner commodity node k, |GB i | represents the number of commodity nodes in the sphere.

[0024] Use graph convolution to process the granular graph and obtain attribute information For the adjacency matrix A GBPerform power calculation to obtain second-order neighbor information based on the attribute information in the granulation graph The objective function of constructing the particle-sphere similarity matrix with second-order neighbor information.

[0025] For the adjacency matrix A GB The power calculation includes: calculating the adjacency matrix A GB The square of A 2 GB , the adjacency matrix A GB and A 2 GB Add together to get the second-order neighbor information f(A GB ).

[0026] The objective function of the particle-sphere similarity matrix is:

[0027]

[0028] Among them, S GB is the particle-sphere similarity matrix, α is a trade-off parameter, and the superscript T is the transpose. represents the square of the Frobenius norm, f(A GB ) is the second-order neighbor information.

[0029] On the other hand, the present invention adopts a consumer finance product recommendation system, comprising:

[0030] An input module for obtaining financial transaction network graph data;

[0031] A particle sphere generation module, used to generate multiple particles according to the financial transaction network graph data;

[0032] A granulation map generation module, used for generating a granulation map according to a plurality of granules;

[0033] A particle-sphere similarity matrix calculation module is used to calculate the particle-sphere similarity matrix according to the granulation graph;

[0034] Clustering module, used to obtain commodity clustering results based on the particle-sphere similarity matrix;

[0035] The product recommendation module is used to make product recommendations based on product clustering results.

[0036] Beneficial effects:

[0037] 1. Inspired by the human brain's "large-scale first" cognitive mechanism and the granular computing splitting mechanism, the present invention uses granules of different sizes to cover and represent the data space of the financial transaction network graph, and has good data representation capabilities; 2. The present invention uses the characteristic matrix of granules instead of the characteristic matrix of graph nodes for clustering, and reconstructs the graph similarity matrix in a coarse-grained manner, which enhances the adaptability to complex topological structures. The number of granules is much less than the number of graph nodes, which greatly reduces the matrix scale, reduces the computing cost and improves the algorithm speed; 3. The present invention introduces the granule center density to control its splitting to ensure the similarity of the structure within the granule, thereby capturing a more advanced community structure and improving the accuracy of product recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of a flow chart of a transaction network commodity recommendation method based on granular sphere calculation provided by an embodiment of the present invention;

[0039] Figure 2 A schematic diagram of a detailed generation process of a granulation map provided in an embodiment of the present invention;

[0040] Figure 3 This is a comparison diagram before and after granulation provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] like Figure 1 As shown, the consumer finance product recommendation method based on graph clustering based on granular sphere computing in the present invention includes:

[0043] S1. Obtain a financial transaction network graph dataset, extract information from the financial transaction network graph dataset, and obtain the extracted graph data G = (V, E, X, A), where V = {v 1 , v 2 , ..., v n} is the set of commodity nodes, X={x 1 , x 2 , ..., x n} is the feature matrix corresponding to the commodity node, n is the number of commodity nodes, E is the edge set, A={a ij}∈R n×n is the adjacency matrix, if the product node v i , v j When there is an edge between them, that is, (vi , v j )∈E, then a ij =1, otherwise a ij =0.

[0044] In one embodiment, commodity nodes include: loan products, credit cards, financial products, etc.; an edge represents a common purchase or use relationship between two commodities. For example, if a customer often chooses two commodities at the same time and often purchases a financial product after applying for a certain credit card, an edge is established between the two commodities; the characteristics of each commodity are composed of its descriptive information, such as rate of return, risk level, fee structure, target customer group, etc.

[0045] S2. Perform sphere calculation on the graph data G through an adaptive splitting mechanism to capture high-order neighbor information, aggregate nodes with similar structures into spheres, and finally generate multiple spheres; each sphere includes multiple commodity nodes; each sphere can be regarded as a subgroup that aggregates similar commodities.

[0046] like Figure 2 As shown, the particle-sphere calculation for the graph data G includes:

[0047] The first stage of initial sphere division: all commodity nodes in the graph data G are regarded as a sphere and Split, get Initial spheres, that is, the initial sphere set;

[0048] Specifically, the product of the commodity node degree and the shortest path between commodity nodes is used as the index of the center point in the hope of obtaining a uniform distribution; the front node with the largest index is The commodity node is taken as the initial center point, and the remaining commodity nodes are divided into the initial center point according to the shortest path principle. Initial sphere gb f .

[0049] Two-stage quality-controlled sphere splitting: each initial sphere is subjected to cyclic binary splitting according to the proximity principle and sphere splitting density gain to obtain the final sphere set;

[0050] Since the present invention divides the sphere into two according to the shortest path from the node to the center, there may be a situation where the shortest path from a node to two centers is the same. In this case, the node will be divided into the center point that first participates in the judgment, which may cause the gap between the two sub-spheres of the sphere to be too large. In order to avoid this situation as much as possible, the present invention selects the two farthest commodity nodes as temporary center points for sphere division, so that the nodes will be added more evenly according to the shortest path.

[0051] Specifically, the two commodity nodes with the farthest distance from each other in the initial sphere are selected as temporary center points, the distances between the remaining commodity nodes in the initial sphere and the temporary center points are calculated, and the remaining commodity nodes in the initial sphere are added to the temporary center points closest to them to obtain two child spheres. It is determined whether the sum of the densities of the two child spheres is lower than the density of the parent sphere, that is, the density of the initial sphere. If so, the final sphere is obtained; otherwise, the two child spheres are subjected to binary fission respectively, and the binary fission process is repeated until the sum of the densities of the child spheres is lower than the density of the parent sphere, and the splitting cycle stops to obtain the final set of spheres.

[0052] The calculation process of the density of the sphere includes:

[0053] Calculate the sphere radius Where d(i, j) represents the shortest path distance between commodity node i and commodity node j in the sphere. It means taking the maximum value among all commodity node pairs (i, j).

[0054] Calculate the density of the pellet Among them, |GB| represents the number of commodity nodes in a single sphere. The larger the density, the tighter the connection between the commodity nodes in the sphere and the higher the quality of the sphere.

[0055] S3, adding edges between the granules according to the adjacency relationship between the granules, generating a granulation graph; calculating the adjacency matrix and characteristic matrix of the granulation graph, obtaining the granulation graph G′=(V GB , E GB , X GB , A GB );

[0056] Among them, V GB = {GB 1 , GB 2 , ..., GB N} is a set of sphere nodes generated by the graph dataset, and each node represents a sphere;

[0057] is the feature matrix corresponding to the particle ball, N is the number of particle ball nodes in the granulation graph, d is the dimension of the feature of each particle ball, Among them, x k represents the i-th ball GB i The feature corresponding to the inner commodity node k, | GB i | indicates a sphere GB i The number of product nodes in the group.

[0058] E GB is an edge set. As long as there is an edge between a pair of commodity nodes in two spheres, there is an edge between the spheres. Use the adjacency matrix A GB = {b ij}∈RN×N Indicates that if (GB i , GB j )∈E GB , then b ij =1, otherwise b ij =0.

[0059] Adjacency Matrix of Granular Graph

[0060] Original Graph Data and Granulated Graph Data are as follows Figure 3 shown.

[0061] S4, constructing an objective function of a particle-sphere similarity matrix according to the granulation graph; solving the objective function of the particle-sphere similarity matrix to obtain the particle-sphere similarity matrix;

[0062] Specifically, we first use graph convolution to aggregate the neighbor information in the granular graph and smooth the information in the granular graph, so that the attributes between adjacent granular nodes are more similar, and obtain the attribute information X′ GB ; The specific steps of graph convolution are as follows:

[0063] First, the adjacency matrix is ​​symmetric and normalized:

[0064]

[0065] Where I is the identity matrix, L GB is the number of filter layers, D GB Degree matrix of the granulated graph.

[0066] Then, the graph filter is used to transform the properties of the particle sphere nodes to filter out unnecessary high-frequency signals and retain the geometric features of the graph, where m is a positive integer and the formula is as follows:

[0067]

[0068] For data sets of different complexity, the required m is different. Preferably, m∈{1, 2, ..., 6}.

[0069] For the adjacency matrix A GB Perform power calculations, accumulate the second-order neighbor information, and use it as part of the similarity matrix: f(A GB )=A GB +A 2 GB ; Second-order neighbor information represents the indirect relationship between products and reveals potential purchase paths.

[0070] The particle similarity matrix records the similarity between each pair of particle nodes. The higher the similarity, the more likely the two particle nodes are to belong to the same group.

[0071] The objective function of the particle-sphere similarity matrix based on structure and features is as follows:

[0072]

[0073] Among them, S GB is the particle-sphere similarity matrix, α is a trade-off parameter that controls the amount of neighboring information to be retained, and the superscript T is the transpose. represents the square of the Frobenius norm, f(A GB ) is the second-order neighbor information.

[0074] By minimizing and And the particle-sphere similarity matrix S GB And the adjacency matrix A of the spherical graph GB The calculated second-order neighbor f(A GB ), optimize the retention of node feature representation, minimize information loss in clustering or other forms of feature transformation, and maintain the high-order structural characteristics of the graph. The adjacency matrix alone may not be able to fully simulate the pairwise relationship between particles because it is usually sparse, so the second-order neighbors are used.

[0075] The solved S GB as follows:

[0076]

[0077] S5. Perform symmetric non-negative on the particle-sphere similarity matrix, perform spectral clustering on the symmetric non-negative particle-sphere similarity matrix, and obtain the product clustering result.

[0078] Symmetric non-negative calculation of the particle-sphere similarity matrix includes:

[0079]

[0080] Among them, C GB is the particle-sphere similarity matrix after symmetric non-negative, || is the absolute value, and the superscript T is the transpose.

[0081] S6. Recommend products based on the product clustering results.

[0082] Recommend other products in the cluster of the products that the user has browsed and purchased to the user. If the user is interested in a certain product, recommend other products in the same cluster because they are similar in attributes. For example, if the user purchases "Short-term Financial Product A", you can recommend "Short-term Financial Product B" and "Short-term Financial Product C" in the same cluster.

[0083] In one embodiment, a consumer finance product recommendation system is used, including:

[0084] An input module for obtaining financial transaction network graph data;

[0085] A particle sphere generation module, used to generate multiple particles according to the financial transaction network graph data;

[0086] A granulation map generation module, used for generating a granulation map according to a plurality of granules;

[0087] A particle-sphere similarity matrix calculation module is used to calculate the particle-sphere similarity matrix according to the granulation graph;

[0088] Clustering module, used to obtain commodity clustering results based on the particle-sphere similarity matrix;

[0089] The product recommendation module is used to make product recommendations based on product clustering results.

[0090] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation modes of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. The consumer finance product recommendation method based on graph clustering of granular sphere computing is characterized by: include: S1. Obtain a financial transaction network graph dataset, extract information from the financial transaction network graph dataset, and obtain extracted graph data G; Financial transaction network graph data includes commodities, commodity information, and the relationships between commodities; S2. Perform granular calculation on the graph data G through an adaptive splitting mechanism to generate multiple granular balls; each granular ball includes multiple commodities; S3, add edges between the granules according to the adjacency relationship between the granules, and generate a granulation graph G′=(V GB ,E GB ,X GB ,A GB ), where V GB is the node set of the granular graph, each node represents a sphere, X GB is the characteristic matrix of the granulation graph, E GB is the edge set of the granulated graph, A GB is the adjacency matrix of the granulated graph; S4, constructing an objective function of a particle-sphere similarity matrix according to the granulation graph; solving the objective function of the particle-sphere similarity matrix to obtain the particle-sphere similarity matrix; S5, symmetric non-negative particle-sphere similarity matrix, and spectral clustering is performed on the symmetric non-negative particle-sphere similarity matrix to obtain the product clustering result; S6. Recommend products based on the product clustering results.

2. The method for recommending consumer finance products based on graph clustering using granular sphere computing according to claim 1 is characterized in that: The particle-sphere calculation for the graph data G includes: S21. Treat all commodity nodes in the graph data G as a sphere and perform Split, get Initial sphere Where f is the index of the initial sphere, and n is the number of commodity nodes in the graph data G; S22, according to the principle of proximity, each initial ball gb f Perform binary fission to obtain each initial particle gb f The two balls gb 1,f 、gb f2,f ; S23, judge the two balls gb 1,f 、gb 2,f Is the sum of the densities lower than the corresponding parent ball gb f The density of the parent ball is gb. f Otherwise, return to step S22 and calculate the two sub-balls gb according to the principle of proximity. 1,f 、gb 2,f Perform binary fission.

3. The method for recommending consumer finance products based on graph clustering using granular sphere computing according to claim 2 is characterized in that: Treat all commodity nodes in the graph data G as a sphere and The split includes: taking the product of the commodity node degree and the shortest path distance between commodity nodes as the index of the center point, The commodity nodes are taken as the initial center point, and the remaining commodity nodes are divided into the initial center point according to the shortest path principle. Initial sphere gb f .

4. The method for recommending consumer finance products based on graph clustering using granular sphere computing according to claim 2 is characterized in that: The binary split of the initial sphere according to the proximity principle includes: selecting the two farthest commodity nodes in the initial sphere as temporary center points, calculating the shortest path distance between the remaining commodity nodes in the initial sphere and the temporary center points, and adding the remaining commodity nodes in the initial sphere to the temporary center points closest to them to obtain two sub-spheres.

5. The method for recommending consumer finance products based on graph clustering using granular spherical computing according to claim 2 is characterized in that: The calculation process of the density of the sphere includes: selecting two commodity nodes with the farthest path distance in the sphere, taking half of the path distance between the two commodity nodes as the radius of the sphere, dividing the number of commodity nodes in the sphere by the radius of the sphere, and obtaining the density of the sphere.

6. The method for recommending consumer finance products based on graph clustering using granular spherical computing according to claim 1 is characterized in that: Granular ball GB in granulation diagram i The characteristic matrix Among them, x k represents the i-th ball GB i The feature corresponding to the inner commodity node k, |GB i | indicates a sphere GB i The number of product nodes in the group.

7. The method for recommending consumer finance products based on graph clustering using granular spherical computing according to claim 1 is characterized in that: Use graph convolution to process the granular graph and obtain attribute information For the adjacency matrix A GB Perform power calculation to obtain second-order neighbor information based on the attribute information in the granulation graph The objective function of constructing the particle-sphere similarity matrix with second-order neighbor information.

8. The method for recommending consumer finance products based on graph clustering using granular sphere computing according to claim 7 is characterized in that: For the adjacency matrix A GB The power calculation includes: calculating the adjacency matrix A GB The square of A 2 GB , the adjacency matrix A GB and A 2 2B Add together to get the second-order neighbor information f(A GB ).

9. The method for recommending consumer finance products based on graph clustering using granular sphere computing according to claim 7 is characterized in that: The objective function of the particle-sphere similarity matrix is: Among them, S GB is the particle-sphere similarity matrix, α is a trade-off parameter, and the superscript T is the transpose. represents the square of the Frobenius norm, f(A GB ) is the second-order neighbor information.

10. A system for recommending consumer finance products based on the graph clustering method based on granular sphere computing according to claims 1 to 9, characterized in that: include: An input module for obtaining financial transaction network graph data; A particle sphere generation module, used to generate multiple particles according to the financial transaction network graph data; A granulation map generation module, used for generating a granulation map according to a plurality of granules; A particle-sphere similarity matrix calculation module is used to calculate the particle-sphere similarity matrix according to the granulation graph; Clustering module, used to obtain commodity clustering results based on the particle-sphere similarity matrix; The product recommendation module is used to make product recommendations based on product clustering results.

Images