Multi-role embedding representation method and system based on self-network split

Abstract

The application provides a multi-role embedding representation method and system based on self-network splitting, which comprises the following steps: constructing a multi-role network by performing self-network splitting and reconstruction on an enterprise network dataset; capturing enterprise node walk sequences from the local and global perspectives of the network through quantum walk to obtain the structural information of the network; comparing the walk sequences to generate a node role identification matrix of the enterprise network; and fusing the role representation of the enterprise node with the neighborhood feature information of the enterprise node to obtain a multi-role embedding representation of the enterprise node. The application realizes multi-role identification and embedding representation of enterprise nodes in the network based on self-network splitting, thereby fundamentally solving the problems of single role identification of enterprise nodes in the enterprise network and low accuracy of multi-role embedding representation of enterprise nodes in the prior art.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of complex network data mining, and particularly relates to a multi-role embedding representation method, system and device based on self-network splitting. BACKGROUND

[0002] Networks are ubiquitous, full of complexity and diversity. Enterprise network is one of the common networks, in which each node represents an enterprise and the edge represents the relationship between two nodes. In order to better understand and analyze enterprise network data, network embedding technology emerges as the times require and becomes an important technology and method for downstream tasks such as node classification, network clustering and link prediction.

[0003] Among the nodes of the enterprise network, the node set with similar structural characteristics or the same mode is called a role, and specifically, the role includes suppliers, distributors and customers, etc. The network constructed by the role nodes is called a role network. Most of the existing methods are to map the nodes in the network to a low-dimensional vector space to realize one-to-one correspondence between the nodes and the vectors.

[0004] Role network embedding aims to project similar nodes of the role in the network to a closer low-dimensional vector space and become an effective tool for network analysis. However, the existing multi-role representation method is usually limited to single role identification of enterprise nodes in the enterprise network, ignoring the different functions of enterprise nodes in the enterprise network. Identifying different roles of enterprise nodes in the network can better understand the structure and function of the enterprise network, reveal hidden patterns and rules, find key nodes and paths, design more effective network management and control strategies, and predict the behavior and evolution of the network. That is, the existing multi-role embedding representation method only identifies single role of enterprise nodes in the enterprise network, and the accuracy of multi-role embedding representation of the enterprise network is low.

[0005] Therefore, how to realize multi-role identification and multi-role embedding representation of enterprise nodes in the enterprise network and improve the accuracy of multi-role embedding representation is a problem to be solved in the field. SUMMARY

[0006] The present application aims at the defects of the prior art, and provides a multi-role embedding representation method and system based on self-network splitting.

[0007] In order to achieve the above object, the present application adopts the following technical scheme:

[0008] The present application provides a multi-role embedding representation method based on self-network splitting, characterized in that it comprises the following steps:

[0009] S1, splitting and reconstructing an enterprise network dataset to construct a multi-role network; specifically comprising:

[0010] S11, performing self-network splitting based on the enterprise network dataset; wherein in the enterprise network dataset, nodes represent different enterprises, and edges represent the relationship between enterprise nodes;

[0011] S12, reconstructing the network based on the split nodes to form a multi-role network; wherein in the multi-role network, nodes represent different enterprises, edges represent the relationship between enterprise nodes, and the characteristics of nodes include enterprise size and business type, and the roles of nodes include supplier, distributor and customer;

[0012] S2, capturing enterprise node walk sequences through quantum walk based on the multi-role network; specifically comprising:

[0013] S21, obtaining a probability distribution matrix under the current walk evolution number in the global walk process; generating global role representation and global quantum walk sequences of enterprise nodes based on the probability distribution matrix;

[0014] S22, generating local role representation and local quantum walk sequences of enterprise nodes based on biased quantum walk;

[0015] S3, comparing the walk sequences to generate role identification; specifically comprising:

[0016] S31, splicing the global quantum walk sequence and the local quantum walk sequence to obtain a complete walk sequence of the enterprise node;

[0017] S32, determining a similarity function of the complete walk sequences of two enterprise nodes;

[0018] S33, performing sequence alignment on all nodes in the entire network based on the similarity function to construct a similarity matrix;

[0019] S34, obtaining a node role identification matrix of the enterprise network according to the similarity matrix;

[0020] S4, fusing the role representation of the enterprise node with the enterprise node neighborhood feature information to obtain a multi-role embedding representation of the enterprise node; specifically comprising:

[0021] S41, splicing the global role representation and the local role representation of the enterprise node to obtain a complete role representation fusing global and local role information;

[0022] S42, calculating all node features and evaluation points by using a feature function to obtain enterprise node neighborhood feature information, and fusing features by using the probability distribution matrix obtained by quantum walk to obtain a feature representation of the enterprise node;

[0023] S43, fusing the complete role representation and the feature representation of the enterprise node to obtain a multi-role embedding representation of the enterprise node.

[0024] Further, in step S11, based on the enterprise network dataset, self-network splitting is performed, specifically:

[0025] The enterprise network dataset is G=(V, E), the nodes represent different enterprises, and the edges represent the relationship between enterprise nodes; for All nodes in the graph are traversed, and the following steps are performed for each node:

[0026] The neighbor nodes of node v are defined as Disconnect the edges between the target node v and all one-hop neighbor nodes around it, and the edge set at this moment is

[0027] In the network after disconnecting the edges, calculate the connected subgraphs Where C i represents the node v i Through the subgraph set after disconnecting the edges with the surrounding neighbors, represents the node v i The s i th subgraph in the generated subgraph set, s i is the number of connected subgraphs;

[0028] Each of the subnets can represent a specific role feature of the center node; the center node is split into s i nodes, each of which is assigned to a respective subnet; that is, for each connected subgraph create a new node so that the new node set Let Finally, the node set

[0029] In the node splitting process, the number of roles of the network is obtained

[0030] Further, S12, based on the split nodes, reconstructs the network to form a multi-role network, specifically including:

[0031] External reconstruction: connecting each split node with the nodes V on the original graph G; specifically: for a target node v i , traverse all its edges (v i , u) ∈ E; for the connected subgraph set C i after the disconnected edges, if u exists in the subgraph , mark it as u j ; update the new edge set where is the new node generated under the corresponding subgraph i v ; thus, the externally reconstructed network is G′ = (V′, E″);

[0032] Internal reconstruction: for the externally reconstructed network G′, the following operations are performed: traverse all nodes v i in the original network G, find the split node set V′ i where For all nodes in the set V′ i , connect each other to build new edges where 1≤j≤k≤s i ; update the edge set to form a multi-role network.

[0033] Further, in step S21, during the global walk process, a probability distribution matrix under the current walk evolution number is obtained; based on the probability distribution matrix, a global role representation and a global quantum walk sequence of the enterprise node are generated, specifically:

[0034]

[0035] Among them, d(v) represents the node degree of enterprise node v, A represents the adjacency matrix of the graph, Sum() is the row summation function, k g represents the number of global quantum walk evolutions; Indicates that enterprise node v is in the kth g The evolution sequence representation obtained under the quantum walk evolution number; Indicates that enterprise node v is in the kth g -1 quantum walk evolution times to obtain the evolution sequence representation; Indicates that enterprise node w is in the kth g -1 quantum walk evolution times to obtain the evolution sequence representation; Indicated by The kth row of the splicing g The evolution sequence matrix obtained under the quantum walk evolution times; M0(v,·) represents the initialization vector obtained by calculating the degree of different nodes v; M0 represents the initialization matrix spliced ​​by row according to M0(v,·); z global It represents the global role of the enterprise node, which is expressed as the sum of the probability matrix under global walk;

[0036] z global The i-th row represents the enterprise node v i The global walk sequence of nodes obtained by summing the probability matrix is ​​z global The walk sequence of each enterprise node is arranged horizontally to obtain the final global quantum walk sequence q = {q(v1), q(v2), ..., q(v n )}.

[0037] Furthermore, in step S22, based on the biased quantum walk, a local role representation and a local quantum walk sequence of the enterprise node are generated. Specifically, the global role information is obtained through quantum walk, and the quantum walk is biased to emphasize the local structure of the enterprise node. ratio The initial probability of each enterprise node is biased, and the local role representation z of the node is obtained by calculating the role proximity local ,

[0038]

[0039] Among them, M'0(v,·) represents the bias probability p ratio The initialization sequence is obtained by calculating the degree of different nodes v; M'0 represents the initialization matrix spliced ​​by row according to M'0(v,·); z local It is expressed as the sum of the probability matrix under local walk, k l represents the number of local quantum walk evolutions; M′1 represents the probability matrix obtained under one evolution of the local quantum walk; representing local quantum walk k l probability matrix obtained under sub-evolution;

[0040] z local The i-th row in the matrix represents the enterprise node v i The node local walk sequence obtained by summing the probability matrix, z local The walk sequence of each node in the matrix is transversely arranged and spliced by row to obtain the final local quantum walk sequence q' = {q'(v1), q'(v2),..., q'(v n )}.

[0041] Further, step S3, comparing the walk sequence, generating the role identification; specifically including:

[0042] S31, splicing the global quantum walk sequence and the local quantum walk sequence to obtain the complete walk sequence of the enterprise node, specifically: Q = concat(q, q'), wherein Q(v i ) represents the walk sequence of node v i ;

[0043] S32, determining the similarity function of the complete walk sequence of two enterprise nodes, specifically:

[0044]

[0045] S33, based on the similarity function, performing sequence comparison on all nodes in the entire network, and constructing a similarity matrix; wherein the similarity matrix is identified as M R ;

[0046] S34, obtaining the node role identification matrix of the enterprise network according to the similarity matrix; specifically including:

[0047] Set the initial threshold R threshold = 0.5, if R similarity (v i ,v j ) ≥ R threshold , then v i , v j are marked as the same role;

[0048] According to the similarity matrix M R , all nodes are marked to obtain the role matrix R M at this time;

[0049] Set the parameter Δ = 0.05, and count the number of roles rn in R M , if rn ≤ numR, then R threshold = R threshold+ Δ, if rn≥ numR, let R threshold = R threshold - Δ; calculate the value of |numR-rn| under each parameter; repeat the step of judging the size relationship between R similarity (v i , v j ) and R threshold under the new threshold, if R similarity (v i , v j )≥ R threshold , mark v i , v j as the same role, and update the role matrix R M at this time; at this time, R M is:

[0050] x∈[1,rn]

[0051] wherein r x represents the xth role;

[0052] When R threshold = min(|numR-rn|), the role matrix R M obtained by comparing the threshold at this moment is the node role identification matrix obtained by the final network.

[0053] Further, S4, the role representation of the enterprise node is fused with the neighborhood feature information of the enterprise node to obtain a multi-role embedding representation of the enterprise node; the specific formula is:

[0054] S41, the global role representation and the local role representation of the enterprise node are spliced to obtain a complete role representation fused with global and local role information; specifically: z′=concat{z local , z global};

[0055] S42, all node features and evaluation points are calculated by using a feature function to obtain neighborhood feature information of the enterprise node, and the feature representation of the enterprise node is obtained by feature fusion through the probability distribution matrix obtained by quantum walk, specifically:

[0056]

[0057] wherein, x w is a feature vector of a target node w; represents the probability value of the enterprise node v walking to the enterprise node w in the probability matrix; the parameter θ represents the evaluation point value, and θ∈[-6, 6] is set;

[0058] S43, the complete role representation and the feature representation of the enterprise node are fused to obtain a multi-role embedding representation of the enterprise node, specifically: Z = concat{z', z feature}.

[0059] The application further provides a multi-role embedding representation system based on self-network splitting, which executes the multi-role embedding representation method based on self-network splitting and comprises a multi-role network construction module, a node walk sequence acquisition module, a role identification generation module and a multi-role embedding representation construction module.

[0060] The multi-role network construction module splits and reconstructs an enterprise network dataset to construct a multi-role network.

[0061] The node walk sequence acquisition module acquires enterprise node walk sequences through quantum walk based on the multi-role network.

[0062] The role identification generation module generates role identifications by comparing the walk sequences.

[0063] The multi-role embedding representation construction module fuses the role representation of an enterprise node with the neighborhood feature information of the enterprise node to obtain a multi-role embedding representation of the enterprise node.

[0064] The application further provides a computer device, which comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to realize the method.

[0065] Compared with the prior art, the application has the following beneficial effects:

[0066] The multi-role embedding representation method based on self-network splitting of the application splits and reconstructs an enterprise network dataset to construct a multi-role network, acquires network structure information by capturing enterprise node walk sequences through quantum walk from the local and global perspectives of the network based on the multi-role network, generates a node role identification matrix of the enterprise network by comparing the walk sequences, fuses the role representation of an enterprise node with the neighborhood feature information of the enterprise node to obtain a multi-role embedding representation of the enterprise node, realizes multi-role identification of enterprise nodes in the network, and improves the accuracy of the multi-role embedding representation of the enterprise nodes in the network. BRIEF DESCRIPTION OF DRAWINGS

[0067] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative effort.

[0068] Figure 1 A method for multi-role embedding representation based on self-network splitting provided by an embodiment of the present application.

[0069] Figure 2 A framework diagram of a method for multi-role embedding representation based on self-network splitting provided by an embodiment of the present application.

[0070] Figure 3 A diagram of network node splitting provided by an embodiment of the present application, wherein (a) is an original enterprise network, (b) is a diagram of self-network partitioning of an example enterprise node, (c) is an example enterprise node disconnecting edges around it, (d) is a node community divided according to a connected subgraph, (e) is a corresponding number of nodes generated according to the number of partitions, and (f) is an enterprise node divided into corresponding communities according to the splitting.

[0071] Figure 4 A diagram of an internal and external edge process of a split node provided by an embodiment of the present application.

[0072] Figure 5 A diagram of generating a final network embedding representation in combination with a feature function provided by an embodiment of the present application.

[0073] Figure 6 A system diagram of a method for multi-role embedding representation based on self-network splitting provided by an embodiment of the present application. DETAILED DESCRIPTION

[0074] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0075] It should be understood that when used in the specification and the appended claims, the terms “comprise” and “include” indicate the presence of described features, integers, steps, operations, elements, and / or components, but do not exclude one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0076] The present application will be further described below with reference to the drawings and specific embodiments, but is not limited to the embodiments.

[0077] It is also to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. As used in this specification and the appended claims, the singular forms "a," "an" and "the" include plural referents unless the context clearly dictates otherwise.

[0078] It is further to be understood that the term "and / or," as used herein in the specification and in the claims, is to be taken as an open-ended transition term indicating that the listed items are associated with one another in any combination and that the combination is inclusive of any permutation of the listed items.

[0079] The present application proposes a multi-role embedding representation method based on self-network splitting. As shown in the figure, the multi-role embedding representation method based on self-network splitting includes the following steps S1-S4. The multi-role embedding representation method based on self-network splitting framework diagram is shown in the figure. Figure 1 Figure 2

[0080] S1, split and reconstruct the enterprise network dataset, and construct a multi-role network; specifically including:

[0081] S11, based on the enterprise network dataset, perform self-network splitting, as shown in the figure. Specifically: Figure 3

[0082] The enterprise network dataset is G=(V, E), where the nodes represent different enterprises, and the edges represent the relationship between enterprise nodes. Specifically, the edges represent the membership relationship or business dealings between enterprise nodes, which can be in the form of purchase, sale, cooperation, etc.

[0083] For each node v in the graph, the following steps are performed:

[0084] The definition of the neighbor nodes of node v is Disconnect the edges between the target node v and all one-hop neighbor nodes around it, and the edge set at this moment is

[0085] In the network after disconnecting the edges, calculate the connected subgraphs where C i represents the node v i Through the subgraph set after disconnecting the edges with the surrounding neighbors, represents the node v i the s i th subgraph in the generated subgraph set, s i is the number of connected subgraphs;

[0086] Each of the subnets can represent a specific role feature of the center node; the center node is split into s​​​​i nodes, each node is assigned to its own subnet; that is, for each connected subgraph Create a new node Make the new node set make Finally, we get a set of nodes

[0087] During the node splitting process, the number of roles in the network is obtained

[0088] S12. Based on the split nodes, the network is reconstructed to form a multi-role network;

[0089] In the multi-role network, nodes represent different enterprises, and edges represent the relationship between enterprise nodes. Furthermore, edges represent the affiliation or business dealings between enterprise nodes, which can be in the form of purchase, sale, cooperation, etc.

[0090] Node characteristics include attributes or information about each node. These include enterprise size (e.g., number of employees, annual revenue, market share), business type (e.g., types of products or services provided), and roles. Nodes can be suppliers, distributors, customers, and more. A single enterprise can be both a supplier and a customer simultaneously. This multi-role nature creates complex interactions within the network.

[0091] Step S12 specifically includes:

[0092] External reconstruction: split each node Connect the edge with the node V on the original graph G; specifically: for the target node v i , traverse all its edges (v i ,u)∈E; for the connected subgraph set C after disconnecting the edges i , if u exists in the subgraph In the j ; Update the new edge set in For node v i In the corresponding subgraph The new nodes generated under the above conditions; thus, the network after external reconstruction is G′=(V′, E″).

[0093] In a specific embodiment, if Figure 4 As shown in the figure, the three multi-role nodes generated by enterprise node 4 are numbered as 4'. At the same time as these three split nodes are generated, the three sub-networks {v1, v2, v3}, {v5, v6, v7}, {v8, v9, v1}, {v ... 10Disconnect with the original node v4, at this time the v4 node on the graph does not have any edge. In the external reconstruction step, the method will connect v4' with the nodes in the three sub-networks respectively, wherein

[0094] Internal reconstruction: after external reconstruction of the network G', the following operations are performed: traversing all nodes v in the original network G i , find the split node set V' i , wherein For all nodes in the set V' i , connect each other to build new edges , wherein 1≤j≤k≤s i ; update the edge set Form a multi-role network.

[0095] After generating new split multi-role nodes, in the external reconstruction step, only the multi-role nodes and the "old nodes" on the original graph are connected to build new edges, but the multi-role nodes split from the same original node are still disconnected components. In order to keep the structural integrity of the entire input graph, new edges need to be built for these points. For Figure 3 (e), since v4' is generated from the original node v4, there is a strong correlation between them, so the application performs internal reconstruction operation on the network G' after external reconstruction.

[0096] S2, based on the multi-role network, capturing enterprise node walk sequence by quantum walk; specifically comprising:

[0097] S21, in the global walk process, obtain the probability distribution matrix under the current walk evolution number; based on the probability distribution matrix, generate the global role representation and global quantum walk sequence of the enterprise node.

[0098] In the quantum walk process, in the global walk process, the probability distribution matrix under the current walk evolution number All nodes participate in the evolution process of quantum walk, and the probability distribution matrix of the graph The final global quantum walk sequence q is obtained by multiple evolutions through the Grover operator U, that is It contains the role structure information of the node to the entire network, represented as z global ;

[0099] Step S21, specifically:

[0100]

[0101] Among them, d(v) represents the node degree of enterprise node v, A represents the adjacency matrix of the graph, Sum() is the row summation function, k g represents the number of global quantum walk evolutions; Indicates that enterprise node v is in the kth g The evolution sequence representation obtained under the quantum walk evolution number; Indicates that enterprise node v is in the kth g -1 quantum walk evolution times to obtain the evolution sequence representation; Indicates that enterprise node w is in the kth g -1 quantum walk evolution times to obtain the evolution sequence representation; Indicated by The kth row of the splicing g The evolution sequence matrix obtained under the quantum walk evolution times; M0(v,·) represents the initialization vector obtained by calculating the degree of different nodes v; M0 represents the initialization matrix spliced ​​by row according to M0(v,·); z global It represents the global role of the enterprise node, which is expressed as the sum of the probability matrix under global walk;

[0102] When the initial state probability of all nodes is the same, all nodes evolve simultaneously on the entire network, and the model is distributed by the probability of quantum walk. To simulate the global evolution characteristics, to represent the role relevance of nodes in the global network, and to generate the global role representation z global .

[0103] z global The i-th row represents the enterprise node v i The global walk sequence of nodes obtained by summing the probability matrix is ​​z global The walk sequence of each enterprise node is arranged horizontally to obtain the final global quantum walk sequence q = {q(v1), q(v2), ..., q(v n )}.

[0104] S22. Based on the biased quantum walk, generate the local role representation and local quantum walk sequence of the enterprise node.

[0105] Step S22, based on the biased quantum walk, generates the local role representation and local quantum walk sequence of the enterprise node, specifically: obtain global role information through quantum walk, bias the quantum walk to emphasize the local structure of the enterprise node, and use the bias probability p ratio The initial probability of each enterprise node is biased, and the local role representation z of the node is obtained by calculating the role proximity local ,

[0106]

[0107] wherein M'0(v,·) represents the initial probability of the node v ratio The initialization sequence obtained according to the degree of different nodes v; M'0 represents the initialization matrix spliced according to M'0(v,·) by row; z local is expressed as the summation result of the probability matrix under local walk, k l represents the number of local quantum walk evolution times; M'1 represents the probability matrix obtained under one evolution of local quantum walk; represents the probability matrix obtained under k l evolution of local quantum walk;

[0108] The initial probability of each node is biased by the bias probability p ratio , so that the enterprise node obtains a higher probability proportion, and the local role representation z local of the enterprise node is obtained through role proximity calculation.

[0109] z local The i-th row in z i represents the local walk sequence of the node obtained after summation of the probability matrix, and z local The final local quantum walk sequence q' = {q'(v1), q'(v2),..., q'(v n )}

[0110] As shown in FIG. Figure 2 , the biased quantum walk part in the framework, if the initial state probability of a certain node is higher, the evolution of quantum walk has a bias, emphasizing the proximity of the local structure of the node, and the probability distribution obtained is considered to be the correlation of the node and the neighboring nodes, so the node has a higher walk probability in the neighborhood range and a smaller walk probability outside the neighborhood range. The biased quantum walk emphasizes the local structure of the node, learns the role proximity of the node, and thus generates the local role representation z local and the local quantum walk sequence q'.

[0111] S3, comparing the walk sequences to generate role identification; specifically comprising:

[0112] S31, splicing the global quantum walk sequence and the local quantum walk sequence to obtain a complete walk sequence of the enterprise node;

[0113] S32, determining a similarity function of the complete walk sequences of two enterprise nodes;

[0114] S33, sequence alignment is performed on all nodes in the entire network based on the similarity function, and a similarity matrix is constructed;

[0115] S34, a node role identification matrix of the enterprise network is obtained according to the similarity matrix.

[0116] Step S3, alignment of the walk sequence and generation of the role identification; specifically including:

[0117] S31, the global quantum walk sequence and the local quantum walk sequence are spliced to obtain a complete walk sequence of the enterprise node, specifically: Q = concat (q, q'), wherein Q (v i ) represents the walk sequence of node v i ;

[0118] S32, a similarity function of the complete walk sequence of two enterprise nodes is determined, specifically:

[0119]

[0120] S33, sequence alignment is performed on all nodes in the entire network based on the similarity function, and a similarity matrix is constructed; wherein, the similarity matrix is identified as M R ;

[0121] S34, a node role identification matrix of the enterprise network is obtained according to the similarity matrix. Specifically including:

[0122] Set the initial threshold R threshold = 0.5, if R similarity (v i , v j ) ≥ R threshold , then v i , v j are marked as the same role;

[0123] According to the similarity matrix M R , all nodes are marked to obtain the role matrix R M at this time;

[0124] Set the parameter Δ = 0.05, and count the number of roles rn in R M . If rn ≤ numR, then R threshold = R threshold + Δ, if rn ≥ numR, then R threshold = R threshold - Δ; calculate the value of |numR-rn| under each parameter; repeat the step of judging R similarity (v i ,v j ) and the new threshold R thresholdThe size relationship, if R similarity (v i , v j )≥R threshold , then v i , v j Mark as the same role and update the role matrix R at this time M ; At this time R M for:

[0125] x∈[1,rn]

[0126] where r x Represents the xth role;

[0127] When R threshold =min(|numR-rn|), the role matrix R obtained by comparing the threshold at this moment M The node role identification matrix obtained for the final network.

[0128] The above steps are used to obtain the preliminary role assignment of nodes in the entire network.

[0129] like Figure 2 As shown in the final network diagram of the framework, each node v in the split multi-role network i Has its own character attributes r(v i ). Finally, the role attributes of all split nodes in the multi-role network are merged, that is, merge get The set(·) function removes duplicate roles from the set. After completing these steps, we get the multi-role labels R(v1), R(v2), ..., R(v n ).

[0130] S4. The role representation of the enterprise node is integrated with the neighborhood feature information of the enterprise node to obtain the multi-role embedding representation of the enterprise node, such as Figure 5 As shown; specifically including:

[0131] S41, concatenating the global role representation and the local role representation of the enterprise node to obtain a complete role representation that integrates global and local role information;

[0132] S42. Calculate all node features and evaluation points using a characteristic function to obtain neighborhood feature information of the enterprise node, and perform feature fusion on the probability distribution matrix obtained by quantum walking to obtain a feature representation of the enterprise node.

[0133] S43, fuse the complete role representation of the enterprise node and the feature representation to obtain a multi-role embedding representation of the enterprise node.

[0134] In the real network, different distributions of the enterprise node neighborhood features can have different effects on the enterprise node. Therefore, considering the complexity of the node neighborhood feature distribution, the application does not fuse the feature information in a simple linear aggregation manner, but uses a feature function based on a graph to describe the neighborhood feature distribution of the enterprise node, and uses the quantum walk result as the connection strength to weight the feature function to obtain feature information highlighting the structure of the node.

[0135] S4, fuse the role representation of the enterprise node with the enterprise node neighborhood feature information to obtain a multi-role embedding representation of the enterprise node; the specific formula is:

[0136] S41, splice the global role representation of the enterprise node and the local role representation to obtain a complete role representation that fuses global and local role information; specifically: z' = concat{z local , z global};

[0137] S42, calculate all node features and evaluation points using a feature function to obtain enterprise node neighborhood feature information, and fuse the features using the probability distribution matrix obtained by quantum walk to obtain a feature representation of the enterprise node, specifically:

[0138]

[0139] Where x w is the feature vector of the target node w; represents the probability value of the enterprise node v walking to the enterprise node w in the probability matrix; the parameter θ represents the evaluation point value, and θ is set to [-6, 6];

[0140] Probability distribution matrix in quantum walk is used to represent the correlation strength between enterprise nodes in the feature function, and the greater the distribution probability value, the greater the correlation strength between enterprise nodes. The application uses a quantum walk weighted feature function Fuse the enterprise node role information and the enterprise node neighborhood feature information to obtain a more optimal and richer node feature representation z feature .

[0141] S43, fuse the complete role representation of the enterprise node and the feature representation to obtain a multi-role embedding representation of the enterprise node, specifically: Z = concat{z', z feature}.

[0142] i.e., the final multi-role embedding representation is finally obtained by fusing the role and structure information together.

[0143] The present application provides a multi-role network representation method of enterprise network nodes based on self-network splitting and quantum walk, which realizes multi-role identification and embedding representation of enterprise nodes in the network. Specifically, first, the enterprise nodes are divided into different sub-networks by traversing the entire network, and then the enterprise nodes are split to generate multi-role nodes according to the different sub-networks of each node. Then, the role structure and other information of the network are obtained by using the superposition of quantum walk. According to the similarity of the walk sequences of different enterprise nodes, the role attribute information of the enterprise nodes in the network is identified. Then, the model fuses the neighborhood information of the nodes by using a weighted feature function, so as to obtain more comprehensive role features. Finally, the model designs a variational autoencoder to reduce information noise, and forms a multi-role embedding representation of the enterprise network based on the multi-role identification of the enterprise network. The method provided by the present application realizes multi-role identification of enterprise nodes in the network, improves the accuracy of the multi-role embedding representation of the enterprise nodes in the network, and further helps to effectively manage the supply chain, optimize business cooperation and improve the competitiveness of enterprises.

[0144] Figure 6 The present application provides a multi-role embedding representation system based on self-network splitting. As shown in Figure 6 the multi-role embedding representation system based on self-network splitting includes a multi-role network construction module, a node walk sequence acquisition module, a role identification generation module, and a multi-role embedding representation construction module.

[0145] The multi-role network construction module splits and reconstructs the enterprise network dataset to construct a multi-role network.

[0146] The node walk sequence acquisition module captures the enterprise node walk sequence based on the multi-role network through quantum walk.

[0147] The role identification generation module compares the walk sequences to generate role identification.

[0148] The multi-role embedding representation construction module fuses the role representation of the enterprise nodes with the neighborhood feature information of the enterprise nodes to obtain a multi-role embedding representation of the enterprise nodes.

[0149] The above multi-role embedding representation system based on self-network splitting can be realized in the form of a computer program, which can run on a computer device.

[0150] The computer device includes a processor, a memory, and a network interface connected through a system bus, wherein the memory can include a non-volatile storage medium and an internal memory.

[0151] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions which, when executed, cause the processor to perform a multi-role embedding representation method based on self-network splitting.

[0152] The processor is configured to provide computing and control capabilities to support the operation of the entire computer device.

[0153] The memory provides an environment for the operation of the computer program in the non-volatile storage medium, and the computer program, when executed by the processor, causes the processor to perform a multi-role embedding representation method based on self-network splitting.

[0154] The network interface is configured to communicate with other devices over a network. Those skilled in the art can understand that the above computer device structure is only part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. A specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0155] The processor is configured to run a computer program stored in the memory, and the program implements the multi-role embedding representation method based on self-network splitting of the first embodiment.

[0156] It should be understood that in the embodiments of the present application, the processor can be a central processing unit (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0157] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the above-mentioned embodiments.

[0158] The application further provides a storage medium. The storage medium can be a computer readable storage medium. The storage medium stores a computer program, and the computer program, when executed by a processor, causes the processor to perform the method of the self-network splitting based multi-role embedding representation method according to the first embodiment.

[0159] The storage medium can be a U disk, a mobile hard disk, a read-only memory (ROM), a magnetic disk or an optical disk, or various computer readable storage media that can store program codes.

[0160] Those skilled in the art can understand that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been described in general terms in the above description. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0161] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of each unit is only a logical function division, and actual implementation can have another division manner. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0162] The steps in the method embodiments of the present application can be adjusted, combined and reduced in sequence according to actual needs. The units in the device embodiments of the present application can be combined, divided and reduced according to actual needs. In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0163] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a storage medium. Based on such understanding, the technical solutions of the present application essentially or say the parts that make contributions to the prior art, or the whole or part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions for causing a computer device (which can be a personal computer, a terminal or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application.

[0164] Note that the above merely describes preferred embodiments of the present application and the principles of the technology applied. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and that various obvious changes, modifications and substitutions can be made without departing from the scope of the present application. Therefore, although the present application has been described in detail through the above embodiments, the present application is not limited to the above embodiments, and can include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the claims.

Claims

1. A multi-role embedding representation method based on ego-network splitting, characterized by: Including steps: S1. Split and reconstruct the enterprise network dataset to build a multi-role network; Specifically include: S11. Performing self-network splitting based on the enterprise network dataset; wherein, in the enterprise network dataset, nodes represent different enterprises, and edges represent relationships between enterprise nodes; S12. Reconstructing the network based on the split nodes to form a multi-role network; wherein, in the multi-role network, nodes represent different enterprises, edges represent relationships between enterprise nodes, node characteristics include enterprise size and business type, and node roles include supplier, distributor, and customer; S2. Based on the multi-role network, capturing the enterprise node wandering sequence through quantum walks; specifically including: S21. During the global walk, obtain a probability distribution matrix under the current walk evolution number; based on the probability distribution matrix, generate a global role representation and a global quantum walk sequence of the enterprise node; S22. Based on the biased quantum walk, generate a local role representation and a local quantum walk sequence of the enterprise node; S3. Compare the walk sequences and generate role identifiers; specifically, the following steps are involved: S31, splicing the global quantum walk sequence and the local quantum walk sequence to obtain a complete walk sequence of the enterprise node; S32. Determine a similarity function of the complete walk sequences of two enterprise nodes; S33, based on the similarity function, performing sequence alignment on all nodes in the entire network to construct a similarity matrix; S34. Obtaining a node role identification matrix of the enterprise network based on the similarity matrix; S4. Fusion of the role representation of the enterprise node with the neighborhood feature information of the enterprise node to obtain a multi-role embedded representation of the enterprise node; specifically including: S41, concatenating the global role representation and the local role representation of the enterprise node to obtain a complete role representation that integrates global and local role information; S42. Calculate all node features and evaluation points using a characteristic function to obtain neighborhood feature information of the enterprise node, and perform feature fusion on the probability distribution matrix obtained by quantum walking to obtain a feature representation of the enterprise node. S43: Fusing the complete role representation and the feature representation of the enterprise node to obtain a multi-role embedded representation of the enterprise node.

2. The method according to claim 1, characterized in that Step S11: splitting the self-network based on the enterprise network dataset, specifically: The enterprise network dataset is G = (V, E), where nodes represent different enterprises and edges represent the relationships between enterprise nodes. Traverse all nodes in the graph and perform the following steps for each node: The neighbor nodes of node v are defined as Disconnect the edges between the target node v and all surrounding one-hop neighbor nodes. The edge set at this moment is Compute connected subgraphs in a disconnected network Among them C i Representative node v i By disconnecting the subgraph set from the surrounding neighbors, Representative node v i The sth subgraph in the generated subgraph set i subgraphs, s i is the number of connected subgraphs; Each of the subnetworks can represent a specific role characteristic of the central node; The central node splits into s i nodes, each node is assigned to its own subnet; that is, for each connected subgraph Create a new node Make the new node set make Finally, we get a node set During the node splitting process, the number of roles in the network is obtained 3. The method according to claim 2, characterized in that Step S12: Reconstruct the network based on the split nodes to form a multi-role network, specifically including: External reconstruction: split the nodes Connect the edge with the node V on the original graph G; specifically: for the target node v i , traverse all its edges (v i ,u)∈E; for the connected subgraph set C after disconnecting the edges i , if u exists in the subgraph In the j ; Update the new edge set in For node v i In the corresponding subgraph The new nodes generated under the above mentioned conditions; therefore, the network after external reconstruction is G′=(V,E″); Internal reconstruction: Perform the following operations on the externally reconstructed network G′: traverse all nodes v in the original network G i , find its split node set V′ i ,in For the set V′ i All nodes in the ,are connected to each other, building new edges where 1≤j≤k≤s i ; Update edge set Form a multi-role network.

4. The method according to claim 3, characterized in that Step S21: During the global walk process, a probability distribution matrix is ​​obtained under the current walk evolution times; based on the probability distribution matrix, a global role representation of the enterprise node and a global quantum walk sequence are generated, specifically: Among them, d(v) represents the node degree of enterprise node v, A represents the adjacency matrix of the graph, Sum() is the row summation function, k g represents the number of global quantum walk evolutions; Indicates that enterprise node v is in the kth g The evolution sequence representation obtained under the quantum walk evolution number; Indicates that enterprise node v is in the kth g -1 quantum walk evolution times to obtain the evolution sequence representation; Indicates that enterprise node w is in the kth g -1 quantum walk evolution times to obtain the evolution sequence representation; Indicated by The kth row of the splicing g The evolution sequence matrix obtained under the quantum walk evolution times; M0(v,·) represents the initialization vector obtained by calculating the degree of different nodes v; M0 represents the initialization matrix spliced ​​by row according to M0(v,·); z global It represents the global role of the enterprise node, which is expressed as the sum of the probability matrix under global walk; z global The i-th row represents the enterprise node v i The global walk sequence of nodes obtained by summing the probability matrix is ​​z global The walk sequence of each enterprise node is arranged horizontally to obtain the final global quantum walk sequence q={q(v1),q(v2),...,q(v n )}.

5. The method according to claim 4, characterized in that Step S22, based on the biased quantum walk, generates the local role representation and local quantum walk sequence of the enterprise node, specifically: obtain global role information through quantum walk, bias the quantum walk to emphasize the local structure of the enterprise node, and use the bias probability p ratio The initial probability of each enterprise node is biased, and the local role representation z of the node is obtained by calculating the role proximity local , Among them, M'0(v,·) represents the bias probability p ratio The initialization sequence is obtained by calculating the degree of different nodes v; M'0 represents the initialization matrix spliced ​​by row according to M'0(v,·); z local It is expressed as the sum of the probability matrix under local walk, k l represents the number of local quantum walk evolutions; M′1 represents the probability matrix obtained under one evolution of the local quantum walk; represents the local quantum walk k l The probability matrix obtained under the second evolution; z local The i-th row represents the enterprise node v i The local walk sequence of the node obtained by summing the probability matrix is ​​z local The walk sequence of each node in the latitudinal arrangement is spliced ​​in rows to obtain the final local quantum walk sequence q′={q′(v1),q′(v2),...,q′(v n )}.

6. The method according to claim 5, characterized in that Step S3: Compare the walk sequences and generate a character identifier; specifically, the steps include: S31, concatenate the global quantum walk sequence and the local quantum walk sequence to obtain a complete walk sequence of the enterprise node, specifically: Q = concat (q, q'), where Q (v i ) represents node v i The wandering sequence; S32. Determine the similarity function of the complete walk sequences of the two enterprise nodes, specifically: S33, based on the similarity function, perform sequence alignment on all nodes in the entire network and construct a similarity matrix; wherein the similarity matrix is ​​identified as M R ; S34. Obtaining a node role identification matrix of the enterprise network based on the similarity matrix; specifically comprising: Set the initial threshold R threshold =0.5, if R similarity (v i , v j )≥R threshold , then v i , v j Marked as the same character; According to the similarity matrix M R Mark all nodes to get the role matrix R at this time M ; Set the parameter Δ = 0.05, and calculate the R M The number of roles in rn, if rn≤numR, then let R threshold =R threshold +Δ, if rn≥numR, then let R threshold =R threshold -Δ; calculate the value of |numR-rn| under each parameter; repeat the steps to determine R similarity (v i ,v j ) and R under the new threshold threshold The size relationship, if R similarity (v i ,v j )≥R threshold , then v i , v j Mark as the same role and update the role matrix R at this time M ; At this time R M for: x∈[1,rn] where r x Represents the xth role; When R threshold =min(|numR-rn|), the role matrix R obtained by comparing the threshold at this moment M The node role identification matrix obtained for the final network.

7. The method according to claim 6, characterized in that S4. The role representation of the enterprise node is integrated with the neighborhood feature information of the enterprise node to obtain the multi-role embedding representation of the enterprise node. The specific formula is: S41, concatenate the global role representation and the local role representation of the enterprise node to obtain a complete role representation that integrates global and local role information; specifically: z′=concat{z local ,z global }; S42. Calculate all node features and evaluation points using the characteristic function to obtain neighborhood feature information of the enterprise node, and perform feature fusion on the probability distribution matrix obtained by quantum walking to obtain a feature representation of the enterprise node, specifically: Among them, x w is the feature vector of the target node w; represents the probability value of enterprise node v wandering to enterprise node w in the probability matrix; parameter θ represents the evaluation point value, set θ∈[-6,6]; S43, fusing the complete role representation and the feature representation of the enterprise node to obtain a multi-role embedded representation of the enterprise node, specifically: Z = concat{z′,z feature }.

8. A multi-role embedding representation system based on ego-network splitting, characterized by: The multi-role embedding representation system executes the multi-role embedding representation method based on ego network splitting as claimed in claim 1, comprising: a multi-role network construction module, a node walk sequence acquisition module, a role identification generation module and a multi-role embedding representation construction module; The multi-role network construction module splits and reconstructs the enterprise network data set to construct a multi-role network; The node wandering sequence acquisition module captures the enterprise node wandering sequence through quantum wandering based on the multi-role network; The role identification generation module compares the wandering sequence and generates a role identification; The multi-role embedding representation construction module fuses the role representation of the enterprise node with the neighborhood feature information of the enterprise node to obtain the multi-role embedding representation of the enterprise node.

9. A computer device, characterized in that: The device includes a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

Images