Network node semantic information mining method, device, storage medium and electronic device

Abstract

The present invention provides a method, device, storage medium, and electronic device for mining semantic information of network nodes, and relates to the technical field of network node semantic information mining. The technical solution of the present invention constructs a weighted undirected network between enterprises based on enterprise text data, and utilizes a rule-based event mining algorithm and a time decay function to construct a negative event mapping function to mine negative event information of enterprises. An improved PageRank algorithm and the negative event information are then used to mine the node semantic information of the weighted undirected network. The present invention not only avoids the influence of human subjectivity in the process of mining semantic information of network nodes, but also utilizes computers and corresponding algorithms to replace manual operations, avoiding the problems of low efficiency and high cost in data processing, making the results of network node semantic information mining more objective, accurate, true, and effective.

Description

Technical Field

[0001] The present invention relates to the technical field of network node semantic information mining, and in particular to a network node semantic information mining method, device, storage medium and electronic equipment. Background Art

[0002] Network node influence generally refers to the impact a node has on other nodes through the relationships between them. However, this relationship is structural and does not include the semantic information implicit in the nodes themselves. Therefore, simply mining the structural influence of network nodes cannot fully reflect the influence of network nodes, that is, it cannot obtain complete semantic information about network nodes. For example, the greater the influence of a company (which can be considered a node) in a corporate network, the more resources or risks it will spread to other companies. Resources or risks are two different types of semantic information. Therefore, scientifically and rationally mining the semantic information of network nodes can help quantify the influence of network nodes and provide a basis for scientific evaluation of network nodes.

[0003] Since the semantic information of network nodes is an abstract concept without any material carrier, its mining and quantification are difficult. The existing technology mainly mines the semantic information of network nodes through manual judgment.

[0004] However, this manual method is time-consuming and labor-intensive, inefficient, and easily affected by human subjectivity in mining semantic information of network nodes, resulting in non-objective and inaccurate mining results. Summary of the Invention

[0005] (1) Technical problems solved

[0006] In response to the deficiencies of the existing technology, the present invention provides a method, device, storage medium and electronic device for mining semantic information of network nodes, which solve the problems of low efficiency and inaccurate mining results in the existing network node semantic information mining technology.

[0007] (2) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] In a first aspect, the present invention provides a method for mining semantic information of network nodes, the method comprising:

[0010] Obtain enterprise text data;

[0011] Based on the enterprise text data, a weighted undirected network among enterprises is constructed, and a negative event mapping function is constructed using a rule-based event mining algorithm and a time decay function to mine negative event information of enterprises;

[0012] The semantic information of network nodes of the weighted undirected network is mined using the PageRank algorithm with an improved state transition matrix and restart vector and negative event information.

[0013] Preferably, the enterprise text data includes: enterprise directors, supervisors and senior management text data, enterprise shareholders text data, and judicial document text data;

[0014] The method of constructing a weighted undirected network among enterprises based on the enterprise text data and constructing a negative event mapping function using a rule-based event mining algorithm and a time decay function to mine negative event information of enterprises includes:

[0015] Construct a weighted undirected network among enterprises based on the text data of directors, supervisors and senior managers and shareholders;

[0016] Based on the text data of judicial documents, a rule-based event mining algorithm is used to obtain corporate negative event data. Then, based on the corporate negative event data, a negative event mapping function is constructed using a time decay function to mine corporate negative event information.

[0017] Preferably, the improved state transition matrix and restart vector include:

[0018] State transition matrix M:

[0019] M=D -1 A

[0020] A=(a i,j ) n×n

[0021] D=diag(D1,...,D n )

[0022]

[0023] Among them, A=(a i,j ) n×n represents the weighted adjacency matrix of a weighted undirected network; n represents the number of vertices; a i,j is the weight of vertices i and j, a i,j =0 means there is no connection between the two nodes, and the weight is 0; D = diag(D1,...,D n ) represents the diagonal matrix corresponding to the degree of each node, express;

[0024] Restart vector based on negative events

[0025]

[0026] in, is the restart vector; is the negative event vector.

[0027] Preferably, the negative event mapping function is:

[0028]

[0029] in, represents the score of the enterprise on the negative event; γ is the decay constant; f represents the enterprise; t is the difference between the year when the negative event occurred and the year when the loan was applied for; m a It is the type of negative event activity; Indicates year t m a The total number of negative events for enterprise f in the negative event type.

[0030] In a second aspect, the present invention provides a network node semantic information mining device, the device comprising:

[0031] Data acquisition module, used to obtain enterprise text data;

[0032] A data processing module is used to construct a weighted undirected network among enterprises based on the enterprise text data, and to construct a negative event mapping function using a rule-based event mining algorithm and a time decay function to mine negative event information of enterprises;

[0033] The network node semantic information acquisition module is used to mine the network node semantic information of the weighted undirected network by using the PageRank algorithm with improved state transition matrix and restart vector and negative event information.

[0034] Preferably, the enterprise text data includes: enterprise directors, supervisors and senior management text data, enterprise shareholders text data, and judicial document text data;

[0035] The method of constructing a weighted undirected network among enterprises based on the enterprise text data, constructing a negative event mapping function using a rule-based event mining algorithm and a time decay function, and mining negative event information of enterprises includes:

[0036] Construct a weighted undirected network among enterprises based on the text data of directors, supervisors and senior managers and shareholders;

[0037] Based on the text data of judicial documents, a rule-based event mining algorithm is used to obtain corporate negative event data. Then, based on the corporate negative event data, a negative event mapping function is constructed using a time decay function to mine corporate negative event information.

[0038] Preferably, the improved state transition matrix and restart vector include:

[0039] State transition matrix M:

[0040] M=D -1 A

[0041] A=(a i,j ) n×n

[0042] D=diag(D1,...,D n )

[0043]

[0044] Among them, A=(a i,j ) n×n represents the weighted adjacency matrix of a weighted undirected network; n represents the number of vertices; a i,j is the weight of vertices i and j, a i,j =0 means there is no connection between the two nodes, and the weight is 0; D = diag(D1,...,D n ) represents the diagonal matrix corresponding to the degree of each node, express;

[0045] Restart vector based on negative events

[0046]

[0047] in, is the restart vector; is the negative event vector.

[0048] Preferably, the negative event mapping function is:

[0049]

[0050] in, represents the score of the enterprise on the negative event; γ is the decay constant; f represents the enterprise; t is the difference between the year when the negative event occurred and the year when the loan was applied for; m a It is the type of negative event activity; Indicates year t m a The total number of negative events for enterprise f in the negative event type.

[0051] In a third aspect, the present invention provides a computer-readable storage medium storing a computer program for named entity recognition, wherein the computer program enables a computer to execute the network node semantic information mining method as described above.

[0052] In a fourth aspect, the present invention provides an electronic device, comprising:

[0053] one or more processors;

[0054] Memory; and

[0055] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including instructions for executing the network node semantic information mining method as described above.

[0056] (3) Beneficial effects

[0057] The present invention provides a method, device, storage medium, and electronic device for mining semantic information of network nodes. Compared with the prior art, the present invention has the following advantages:

[0058] The present invention proposes a method, device, storage medium, and electronic device for mining network node semantic information. This method constructs a weighted undirected network between enterprises based on enterprise text data, uses a rule-based event mining algorithm and a time decay function to construct a negative event mapping function to mine negative event information from enterprises, and then uses an improved PageRank algorithm and this negative event information to mine the node semantic information in the weighted undirected network. This method not only avoids the influence of human subjectivity in the network node semantic information mining process, but also utilizes computers and corresponding algorithms to replace manual operations, avoiding the problems of low efficiency and high cost in data processing. This method makes the network node semantic information mining results more objective, accurate, authentic, and effective. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0060] Figure 1 This is an overall flow chart of a method for mining semantic information of network nodes according to an embodiment of the present invention;

[0061] Figure 2 is a flow chart of a rule-based event mining algorithm in an embodiment of the present invention;

[0062] Figure 3 (a) A diagram of the connections between enterprises;

[0063] Figure 3 (b) is a weighted undirected network graph among enterprises. DETAILED DESCRIPTION

[0064] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0065] The embodiments of the present application solve the problems of low efficiency and inaccurate mining results in existing network node semantic information mining technologies by providing a network node semantic information mining method, device, storage medium and electronic device, and achieve the purpose of efficiently and accurately mining network node semantic information.

[0066] The technical solution in the embodiments of the present application is to solve the above technical problems, and the overall idea is as follows:

[0067] In order to mine the semantic information of enterprise network nodes and solve the problems of low efficiency and non-objective mining results of existing network node semantic information mining technologies, the present invention first constructs a weighted undirected network between enterprises based on enterprise text data, and mines enterprise negative event information based on enterprise text data using a rule-based event mining algorithm and a time decay function. Then, the state transition matrix and restart vector of the PageRank algorithm are improved. Finally, the semantic information of network nodes is obtained based on the improved PageRank algorithm and the negative event information.

[0068] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0069] The PageRank algorithm is a typical method for mining structural semantic information. When mining network structural semantic information, the PageRank centrality represents the central score of the node's neighborhood and the function of random jumps to another node in the network. It can be mathematically expressed as:

[0070]

[0071] in, is a vector containing node centrality scores, α is the restart probability (default value is 0.85), M is the state transition matrix (the probability matrix of a node randomly walking to other nodes), is the restart vector (a random jump to another node). For an unweighted directed graph, the state transition matrix M is usually defined as M = D -1A, where D is the diagonal matrix corresponding to the degree of each node, and A is the adjacency matrix of the graph. However, the original PageRank algorithm can only be applied to directed unweighted networks and is not suitable for undirected weighted networks. Therefore, the original PageRank algorithm needs to be improved so that it can be applied to undirected weighted networks. In addition, the original PageRank algorithm is oriented towards mining semantic information of network structure and does not include semantic information of nodes. In order to more completely mine the semantic information of enterprise network nodes, the original PageRank algorithm needs to be improved.

[0072] Example 1:

[0073] In the first aspect, the embodiment of the present invention first proposes a method for mining semantic information of network nodes. Figure 1 , the method comprising:

[0074] S1. Obtain enterprise text data;

[0075] S2. Based on the above enterprise text data, a weighted undirected network among enterprises is constructed, and a negative event mapping function is constructed using a rule-based event mining algorithm and a time decay function to mine negative event information of enterprises;

[0076] S3. Using the PageRank algorithm with improved state transition matrix and restart vector and negative event information, the semantic information of network nodes in the weighted undirected network is mined.

[0077] As can be seen, the network node semantic information mining method, device, storage medium, and electronic device proposed in this invention construct a weighted undirected network between enterprises based on enterprise text data, and utilize a rule-based event mining algorithm and a time decay function to construct a negative event mapping function to mine negative event information from enterprises. The improved PageRank algorithm and the negative event information are then used to mine the node semantic information in the weighted undirected network. This invention not only avoids the influence of human subjectivity in the network node semantic information mining process, but also utilizes computers and corresponding algorithms to replace manual operations, avoiding the problems of low efficiency and high cost in data processing, making the network node semantic information mining results more objective, accurate, authentic, and effective.

[0078] In the above method of the embodiment of the present invention, in order to mine the complete semantic information of enterprise network nodes, it is necessary to obtain enterprise text data for constructing inter-enterprise networks and summarizing negative events related to enterprises. In this case, a preferred processing method is that the enterprise text data includes: enterprise directors, supervisors, and senior management text data, enterprise shareholders text data, and judicial document text data;

[0079] The method of constructing a weighted undirected network among enterprises based on the enterprise text data and constructing a negative event mapping function using a rule-based event mining algorithm and a time decay function to mine negative event information of enterprises includes:

[0080] Based on the text data of corporate directors, supervisors and senior managers and corporate shareholders, a weighted undirected network between enterprises is constructed; based on the text data of judicial documents, a rule-based event mining algorithm is used to obtain corporate negative event data, and then based on the said corporate negative event data, a negative event mapping function is constructed using a time decay function to mine corporate negative event information.

[0081] In addition, in order to adapt to the characteristics that the connection between enterprise network nodes is bidirectional and the strength of the connection is related to the number of shared nodes, and to enable the PageRank algorithm to favor nodes with more negative events during random walks, so that the algorithm technically conforms to the embedding rules of node semantic information, it is necessary to improve the state transition matrix and restart vector in the original PageRank algorithm. At this time, a better approach is to improve the state transition matrix and restart vector including:

[0082] State transition matrix M:

[0083] M=D -1 A

[0084] A=(a i,j ) n×n

[0085] D=diag(D1,...,D n )

[0086]

[0087] Among them, A=(a i,j ) n×n represents the weighted adjacency matrix of a weighted undirected network; n represents the number of vertices; a i,j is the weight of vertices i and j, a i,j =0 means there is no connection between the two nodes, and the weight is 0; D = diag(D1,...,D n ) represents the diagonal matrix corresponding to the degree of each node, express;

[0088] Restart vector based on negative events

[0089]

[0090] in, is the restart vector; is the negative event vector.

[0091] In practice, in order to summarize negative events and perform quantitative calculations on them so that data acquisition is more in line with the actual situation, a negative event mapping function can be constructed to accurately quantify and quantitatively calculate dynamic negative event information. In this case, a better processing method is that the negative event mapping function is:

[0092]

[0093] in, represents the score of the enterprise on the negative event; γ is the decay constant; f represents the enterprise; t is the difference between the year when the negative event occurred and the year when the loan was applied for; m a It is the type of negative event activity; Indicates year t m a The total number of negative events for enterprise f in the negative event type.

[0094] The network node semantic information mining method proposed in the present invention is applicable to network node semantic information mining in many complex networks, such as author collaboration networks, enterprise networks, etc. The implementation process of an embodiment of the present invention is described in detail below using the enterprise network node semantic information mining process as an example.

[0095] See also Figure 1 ,The specific process of enterprise network node semantic ,information mining method includes:

[0096] S1. Obtain enterprise text data.

[0097] First, we used Python-based crawler technology to parse enterprise-related website data from websites like the National Enterprise Credit Information Publicity System, Qichacha, and Tianyancha. We then crawled text data related to the companies and their networks. This primarily included company director, supervisor, and senior management data, including company personnel relationships, shareholder information, and legal documents.

[0098] S2. Based on the above enterprise text data, a weighted undirected network between enterprises is constructed, and a negative event mapping function is constructed using a rule-based event mining algorithm and a time decay function to mine negative event information of enterprises.

[0099] Specifically, a weighted undirected network between enterprises is constructed based on the text data of corporate directors, supervisors and senior managers and corporate shareholders; based on the text data of judicial documents, a rule-based event mining algorithm is used to obtain corporate negative event data; based on the above-mentioned corporate negative event data, a negative event mapping function is constructed using the time decay function to mine corporate negative event information.

[0100] Since the basis of the connection between current corporate networks is the shareholders or directors of different companies, the same shareholder or director can serve as a channel for the connection between two companies. This connection channel is bidirectional, that is, there is no obvious directionality, but this connection will increase as the number of common shareholders or directors increases. In order to adapt to this connection between corporate networks, an undirected weighted network between companies can be constructed. When constructing a weighted undirected network between companies based on the text data of corporate directors, supervisors and senior managers and corporate shareholder text data, the information of the company's existing shareholders, directors, supervisors and senior managers is analyzed based on the corporate directors, supervisors and senior managers text data and shareholder information text data containing corporate personnel relationships, and the interlocking of directors, supervisors and senior managers and shareholder interlocking are used as links to construct two undirected weighted networks between companies. Specifically, see Figure 3 , define enterprises as nodes. If two enterprises share the same person, that is, a person has the same position in both enterprises, then there is an edge between the two enterprises, and the number of shared individuals is used as the weight of the edge. Figure 3 As shown in (a), let ABCDE represent five different companies and P1, P2, P3, and P4 represent four different people. Then, the inter-company network can be obtained by the connections between different companies and different people. For example, if company A and company B share two identical people, P1 and P2, then the weight between A and B is 2. Figure 3 (a) Simplified to Figure 3 (b) Figure 3 (b) is the undirected weighted network graph among enterprises.

[0101] See also Figure 2 When using a rule-based event mining algorithm to obtain negative event data for an enterprise based on judicial document text data, a rule set is set, such as case number, time, cause of action, parties involved, etc.; and a selection rule is set: whether there is an empty slot. Whether there is an empty slot indicates the judgment of the extracted event result, that is, whether the position of elements such as case number, time, cause of action, parties involved, etc. is blank. If there is an empty slot, that is, the corresponding rule set element is blank, then the rule set is updated and re-extracted; if there is no empty slot, that is, the rule set element exists, then the output is a judicial event. Using a rule-based event mining algorithm, relevant negative event text data is extracted from judicial documents, and then the negative event text data is divided into five categories according to negative events in tax credit rating, supplier (or customer) relationship, employee relationship, product, and environmental protection.

[0102] Based on the above-mentioned enterprise negative event data, a negative event mapping function is constructed using a time decay function. When mining enterprise negative event information, the above-mentioned negative event data is used to summarize the negative events of the enterprise at different times based on a time decay function. This time decay function is used to construct a negative event mapping function to establish a mapping relationship between the above-mentioned negative event data and the enterprise's score on negative events. The negative event mapping function can be expressed as follows:

[0103]

[0104] in, represents the score of the enterprise on the negative event; γ is the decay constant; f represents the enterprise; t is the difference between the year when the negative event occurred and the year when the loan was applied for; m a It is the type of negative event activity; Indicates year t m a The total number of negative events for enterprise f in the negative event type.

[0105] S3. Using the PageRank algorithm with improved state transition matrix and restart vector and negative event information, the semantic information of network nodes in the weighted undirected network is mined.

[0106] The original PageRank algorithm can only be applied to unweighted directed networks, not weighted undirected networks. In order to mine the network structure semantic information of weighted undirected networks, the original PageRank algorithm needs to be improved. Specifically, first, for the above-mentioned weighted undirected network, the state transition matrix M is improved. Define a new state transition matrix M:

[0107] M=D -1 A

[0108] A=(a i,j ) n×n

[0109] D=diag(D1,...,D n )

[0110]

[0111] Among them, A=(a i,j ) n×n As the weighted adjacency matrix of a weighted undirected network, where n is the number of vertices, a i,j is the weight of vertices i and j (if two nodes are not connected, the weight is zero), in addition, define D = diag (D1, ..., D n ), Indicates a iThe sum of the weights of the connections between a node and other nodes. D is a diagonal matrix corresponding to the degree of each node, representing the sum of the weights of the connections between each node and other nodes in the network. Finally, a random walk can be viewed as a transition from any vertex to an adjacent vertex based on the weight of the edge.

[0112] Secondly, the original PageRank algorithm only quantifies the structural relationship between nodes, that is, it only reflects the structural semantic information of network nodes, and does not capture the semantic information of single nodes, which makes the semantic information mining of enterprise network nodes defective. Therefore, it is necessary to restart the vector. Improvements are made to use the improved restart vector To characterize the impact of negative events on the semantic information of nodes, so as to measure the influence of network structure while also quantifying the influence of the node's own characteristics. For example, if a node with high network structure influence has more negative events, the impact of the negative event on other nodes in the network will be greater. When making specific improvements, a new restart vector based on negative events is defined.

[0113]

[0114] in, is the restart vector; is the negative event vector; is the restart vector based on negative events, i.e., the restart vector based on negative events Restart vector and negative event vectors The element-wise product of .

[0115] Through the above operations, the state transition matrix M of the original PageRank algorithm is improved, and a new restart vector based on negative events is defined The improved PageRank algorithm is obtained, and then the improved PageRank algorithm is used to mine the complete enterprise network node semantic information of the above weighted undirected network. The improved PageRank algorithm can be expressed as:

[0116]

[0117] in, A vector representing the semantic information of each node in the network; α is the restart probability (the default value is 0.85); D is a diagonal matrix corresponding to the degree of each node, representing the sum of the weights of the connections between each node and other nodes in the network; A = (a i,j ) n×n As the weighted adjacency matrix of a weighted undirected network, where n is the number of vertices, a i,jis the weight of vertices i and j (if the two nodes are not connected, the weight is zero); is the restart vector; is the negative event vector.

[0118] The company's score on negative events As a negative event vector Substitute it into the improved PageRank algorithm above to mine the complete enterprise network node semantic information of the above weighted undirected network. Specifically, The larger the value of the corresponding node in the vector is, the greater the negative impact of the node is.

[0119] At this point, the entire process of the network node semantic information mining method of the present invention is completed.

[0120] Example 2:

[0121] In a second aspect, the present invention provides a network node semantic information mining device, the device comprising:

[0122] Data acquisition module, used to obtain enterprise text data;

[0123] A data processing module is used to construct a weighted undirected network among enterprises based on the enterprise text data, and to construct a negative event mapping function using a rule-based event mining algorithm and a time decay function to mine negative event information of enterprises;

[0124] The network node semantic information acquisition module mines the network node semantic information of the weighted undirected network by using the PageRank algorithm of the improved state transition matrix and restart vector and negative event information.

[0125] Optionally, the enterprise text data includes: enterprise directors, supervisors and senior management text data, enterprise shareholders text data, and judicial document text data;

[0126] The method of constructing a weighted undirected network among enterprises based on the enterprise text data, constructing a negative event mapping function using a rule-based event mining algorithm and a time decay function, and mining negative event information of enterprises includes:

[0127] Construct a weighted undirected network among enterprises based on the text data of directors, supervisors and senior managers and shareholders;

[0128] Based on the text data of judicial documents, a rule-based event mining algorithm is used to obtain corporate negative event data. Then, based on the corporate negative event data, a negative event mapping function is constructed using a time decay function to mine corporate negative event information.

[0129] Optionally, the improved state transition matrix and restart vector include:

[0130] State transition matrix M:

[0131] M=D -1 A

[0132] A=(a i,j ) n×n

[0133] D=diag(D1,...,D n )

[0134]

[0135] Among them, A=(a i,j ) n×n represents the weighted adjacency matrix of a weighted undirected network; n represents the number of vertices; a i,j is the weight of vertices i and j, a i,j =0 means there is no connection between the two nodes, and the weight is 0; D = diag(D1,...,D n ) represents the diagonal matrix corresponding to the degree of each node, express;

[0136] Restart vector based on negative events

[0137]

[0138] in, is the restart vector; is the negative event vector.

[0139] Optionally, the negative event mapping function is:

[0140]

[0141] in, represents the score of the enterprise on the negative event; γ is the decay constant; f represents the enterprise; t is the difference between the year when the negative event occurred and the year when the loan was applied for; m a It is the type of negative event activity; Indicates year t m a The total number of negative events for enterprise f in the negative event type.

[0142] Example 3:

[0143] In a third aspect, the present invention provides a computer-readable storage medium storing a computer program for mining semantic information of network nodes, wherein the computer program causes a computer to execute the following steps:

[0144] Obtain enterprise text data;

[0145] Based on the enterprise text data, a weighted undirected network among enterprises is constructed, and a negative event mapping function is constructed using a rule-based event mining algorithm and a time decay function to mine negative event information of enterprises;

[0146] The semantic information of network nodes of the weighted undirected network is mined using the PageRank algorithm with improved state transition matrix and restart vector and negative event information.

[0147] Optionally, the enterprise text data includes: enterprise directors, supervisors and senior management text data, enterprise shareholders text data, and judicial document text data;

[0148] The method of constructing a weighted undirected network among enterprises based on the enterprise text data and constructing a negative event mapping function using a rule-based event mining algorithm and a time decay function to mine negative event information of enterprises includes:

[0149] Construct a weighted undirected network among enterprises based on the text data of directors, supervisors and senior managers and shareholders;

[0150] Based on the text data of judicial documents, a rule-based event mining algorithm is used to obtain corporate negative event data. Then, based on the corporate negative event data, a negative event mapping function is constructed using a time decay function to mine corporate negative event information.

[0151] Optionally, the improved state transition matrix and restart vector include:

[0152] State transition matrix M:

[0153] M=D -1 A

[0154] A=(a i,j ) n×n

[0155] D=diag(D1,...,D n )

[0156]

[0157] Among them, A=(a i,j ) n×n represents the weighted adjacency matrix of a weighted undirected network; n represents the number of vertices; a i,j is the weight of vertices i and j, a i,j =0 means there is no connection between the two nodes, and the weight is 0; D = diag(D1,...,D n ) represents the diagonal matrix corresponding to the degree of each node, express;

[0158] Restart vector based on negative events

[0159]

[0160] in, is the restart vector; is the negative event vector.

[0161] Optionally, the negative event mapping function is:

[0162]

[0163] in, represents the score of the enterprise on the negative event; γ is the decay constant; f represents the enterprise; t is the difference between the year when the negative event occurred and the year when the loan was applied for; m a It is the type of negative event activity; Indicates year t m a The total number of negative events for enterprise f in the negative event type.

[0164] Embodiment 4:

[0165] In a fourth aspect, the present invention provides an electronic device, comprising:

[0166] one or more processors;

[0167] Memory; and

[0168] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs comprising steps for performing the following steps:

[0169] Obtain enterprise text data;

[0170] Based on the enterprise text data, a weighted undirected network among enterprises is constructed, and a negative event mapping function is constructed using a rule-based event mining algorithm and a time decay function to mine negative event information of enterprises;

[0171] The semantic information of network nodes of the weighted undirected network is mined using the PageRank algorithm with an improved state transition matrix and restart vector and negative event information.

[0172] Optionally, the enterprise text data includes: enterprise directors, supervisors and senior management text data, enterprise shareholders text data, and judicial document text data;

[0173] The method of constructing a weighted undirected network among enterprises based on the enterprise text data and constructing a negative event mapping function using a rule-based event mining algorithm and a time decay function to mine negative event information of enterprises includes:

[0174] Construct a weighted undirected network among enterprises based on the text data of directors, supervisors and senior managers and shareholders;

[0175] Based on the text data of judicial documents, a rule-based event mining algorithm is used to obtain corporate negative event data. Then, based on the corporate negative event data, a negative event mapping function is constructed using a time decay function to mine corporate negative event information.

[0176] Optionally, the improved state transition matrix and restart vector include:

[0177] State transition matrix M:

[0178] M=D -1 A

[0179] A=(a i,j ) n×n

[0180] D=diag(D1,...,D n )

[0181]

[0182] Among them, A=(a i,j ) n×n represents the weighted adjacency matrix of a weighted undirected network; n represents the number of vertices; a i,j is the weight of vertices i and j, a i,j =0 means there is no connection between the two nodes, and the weight is 0; D = diag(D1,...,D n ) represents the diagonal matrix corresponding to the degree of each node, express;

[0183] Restart vector based on negative events

[0184]

[0185] in, is the restart vector; is the negative event vector.

[0186] Optionally, the negative event mapping function is:

[0187]

[0188] in, represents the score of the enterprise on the negative event; γ is the decay constant; f represents the enterprise; t is the difference between the year when the negative event occurred and the year when the loan was applied for; m a It is the type of negative event activity; Indicates year t m a The total number of negative events for enterprise f in the negative event type.

[0189] In summary, compared with the existing technology, the present invention has the following beneficial effects:

[0190] 1. The present invention proposes a method, device, storage medium, and electronic device for mining semantic information of network nodes. This method constructs a weighted undirected network between enterprises based on enterprise text data, and utilizes a rule-based event mining algorithm and a time decay function to construct a negative event mapping function to mine negative event information about enterprises. It then utilizes an improved PageRank algorithm and the negative event information to mine the node semantic information of the weighted undirected network. This invention not only prevents the network node semantic information mining process from being affected by human subjectivity, but also utilizes computers and corresponding algorithms to replace manual operations, thus avoiding the problems of low efficiency and high cost in data processing, making the network node semantic information mining results more objective, accurate, authentic, and effective.

[0191] 2. The present invention improves the state transition matrix M in the original PageRank algorithm to be suitable for mining semantic information of undirected and weighted network structures. This conforms to the characteristics that the connection between enterprise network nodes is bidirectional and the strength of the connection is related to the number of shared nodes. Then, the restart vector is improved to a restart vector based on negative events, so that the improved PageRank algorithm can be biased towards nodes with more negative events when performing random walks. This makes the algorithm technically consistent with the embedding law of node semantic information, and ultimately the mined network node semantic information is more objective, accurate, real and effective;

[0192] 3. By constructing a negative event mapping function based on a time decay function, the present invention can aggregate negative events and perform quantitative calculations on them. This process simulates the law that the impact of negative events decays over time, making data acquisition more realistic. It can accurately quantify and quantitatively calculate dynamic negative event information, making the results of network node semantic information mining more realistic, accurate, and effective.

[0193] 4. The present invention uses a rule-based event mining algorithm to mine negative events of network nodes. It uses rules defined according to expert experience, which is more in line with the needs of data mining and can make the negative event mining results more accurate, explainable and traceable.

[0194] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0195] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for mining semantic information of network nodes, characterized in that: The method comprises: Obtain enterprise text data; Based on the enterprise text data, a weighted undirected network among enterprises is constructed, and a negative event mapping function is constructed using a rule-based event mining algorithm and a time decay function to mine negative event information of enterprises; Using the PageRank algorithm with an improved state transition matrix and restart vector and negative event information to mine the semantic information of the network nodes of the weighted undirected network; The state transition matrix M: M=D -1 A A=(a i,j ) n×n D=diag(D1,...,D n ) Among them, A=(a i,j ) n×n represents the weighted adjacency matrix of a weighted undirected network; n represents the number of vertices; a i,j is the weight of vertices i and j, a i,j =0 means there is no connection between the two nodes, and the weight is 0; D = diag(D1,...,D n ) represents the diagonal matrix corresponding to the degree of each node, Indicates a i The sum of the weights of the node's connections with other nodes; Restart vector based on negative events in, is the restart vector; is the negative event vector.

2. The method according to claim 1, wherein The enterprise text data includes: enterprise directors, supervisors and senior management text data, enterprise shareholders text data, and judicial document text data; The method of constructing a weighted undirected network among enterprises based on the enterprise text data and constructing a negative event mapping function using a rule-based event mining algorithm and a time decay function to mine negative event information of enterprises includes: Construct a weighted undirected network among enterprises based on the text data of directors, supervisors and senior managers and shareholders; Based on the text data of judicial documents, a rule-based event mining algorithm is used to obtain corporate negative event data. Then, based on the corporate negative event data, a negative event mapping function is constructed using a time decay function to mine corporate negative event information.

3. The method according to claim 2, wherein The negative event mapping function is: in, represents the score of the enterprise on the negative event; γ is the decay constant; f represents the enterprise; t is the difference between the year when the negative event occurred and the year when the loan was applied for; m a It is the type of negative event activity; Indicates year t m a The total number of negative events for enterprise f in the negative event type.

4. A network node semantic information mining device, characterized in that: The device comprises: Data acquisition module, used to obtain enterprise text data; A data processing module is used to construct a weighted undirected network among enterprises based on the enterprise text data, and to construct a negative event mapping function using a rule-based event mining algorithm and a time decay function to mine negative event information of enterprises; A network node semantic information acquisition module is used to mine the network node semantic information of the weighted undirected network using the PageRank algorithm with an improved state transition matrix and restart vector and negative event information; The state transition matrix M: M=D -1 A A=(a i,j ) n×n D=diag(D1,...,D n ) Among them, A=(a i,j ) n×n represents the weighted adjacency matrix of a weighted undirected network; n represents the number of vertices; a i,j is the weight of vertices i and j, a i,j =0 means there is no connection between the two nodes, and the weight is 0; D = diag(D1,...,D n ) represents the diagonal matrix corresponding to the degree of each node, Indicates a i The sum of the weights of the node's connections with other nodes; Restart vector based on negative events in, is the restart vector; is the negative event vector.

5. The device according to claim 4, characterized in that The enterprise text data includes: enterprise directors, supervisors and senior management text data, enterprise shareholders text data, and judicial document text data; The method of constructing a weighted undirected network among enterprises based on the enterprise text data, constructing a negative event mapping function using a rule-based event mining algorithm and a time decay function, and mining negative event information of enterprises includes: Construct a weighted undirected network among enterprises based on the text data of directors, supervisors and senior managers and shareholders; Based on the text data of judicial documents, a rule-based event mining algorithm is used to obtain corporate negative event data. Then, based on the corporate negative event data, a negative event mapping function is constructed using a time decay function to mine corporate negative event information.

6. The device according to claim 5, characterized in that The negative event mapping function is: in, represents the score of the enterprise on the negative event; γ is the decay constant; f represents the enterprise; t is the difference between the year when the negative event occurred and the year when the loan was applied for; m a It is the type of negative event activity; Indicates year t m a The total number of negative events for enterprise f in the negative event type.

7. A computer-readable storage medium, characterized in that It stores a computer program for named entity recognition, wherein the computer program enables a computer to execute the network node semantic information mining method according to any one of claims 1 to 3.

8. An electronic device, characterized in that: include: one or more processors; Memory; as well as One or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, the programs including instructions for executing the network node semantic information mining method according to any one of claims 1 to 3.

Images