A Search Method for Key Members of Suspicious Organizations in the Public Security Information Network

Abstract

The present invention relates to a method for searching for backbone members of suspicious organizations in a public security information network. First, a random walk subgraph of potential suspicious persons is obtained based on a group of clue suspicious persons. Then, the random walk transfer probability between the potential suspicious persons is calculated based on the support of the edge. When each clue suspicious person in the group of clue suspicious persons reaches a steady state and generates a steady-state probability vector, all the steady-state probability vectors are combined into a matrix, which is transposed and segmented to generate the eigenvectors corresponding to each potential suspicious person in the subgraph. All eigenvectors are weightedly sorted based on the vector mean and the similarity between vectors, and the potential suspicious persons corresponding to the top k vectors are selected and output as backbone members of the suspicious organization. The method of the present invention generates a random walk subgraph of potential suspicious persons based on the group of clue suspicious persons, and obtains an approximate solution of the backbone members of the suspicious organization by random walk calculation within the subgraph. The purpose is to greatly improve the query efficiency while ensuring a high query accuracy rate.

Description

Technical Field

[0001] The present invention belongs to the field of search technology, and in particular relates to a method for searching backbone members of suspicious organizations in a public security information network. Background Art

[0002] The core of searching for key members of suspicious organizations on public security information networks is to perform community key node queries, that is, to find key members (key nodes) of the suspicious organization (community) to which a suspicious individual belongs within the public security information network. Two general approaches are involved: the first uses a nested community query method, which uses an input node to query the community where the input node resides, and then uses a further community query method within that community to find the closest community and return it as the key node. This method suffers from excessive time consumption; the second uses a Top-k method based on Personal PageRank, which suffers from low query accuracy and may not return strong connections between the query nodes. Summary of the Invention

[0003] Based on the above-mentioned shortcomings and deficiencies in the prior art, the purpose of the present invention is to solve the following technical problem: the traditional nested community query method uses an input node to query the community where the input node is located, and then continues to use the community query method within the community to find the closest community and return it as the key node, resulting in excessive time overhead.

[0004] In order to achieve the above-mentioned object of the invention, the present invention adopts the following technical solutions:

[0005] A method for searching for key members of a suspicious organization in a public security information network comprises the following steps:

[0006] S1, obtaining a public safety information network dataset including natural person data and relationships between natural persons, and constructing a public safety information network, wherein the public safety information network comprises nodes and edges, wherein the nodes represent natural persons in the public safety information network dataset, and the edges represent relationships between natural persons in the public safety information network dataset;

[0007] S2, obtaining a number of suspicious persons and a bounded number of hops, extending the number of suspicious persons to other natural persons according to the obtained hops, and expanding a local subgraph as a random walk subgraph of potential suspicious persons;

[0008] S3, calculate the number of common neighbors of every two potential suspicious persons in the random walk subgraph of potential suspicious persons as the edge support,

[0009] Calculate the average support of the representative node of the potential suspicious person based on the support of each edge.

[0010] Calculate the random walk transfer probability between each pair of potential suspicious persons. When the number of potential suspicious persons is n, we get an n*n random walk transfer matrix P, where the value of the i-th row and j-th column in P is P ij , represents the random walk transfer probability of potential suspicious person i moving to j;

[0011] S4, on the potential suspicious person random walk subgraph, each of the plurality of clue suspicious persons undergoes a random walk with recurrence until a steady state is reached, and each clue suspicious person generates a steady-state probability vector after the random walk;

[0012] S5, generating a feature vector for each potential suspicious person on the potential suspicious person random walk subgraph based on all steady-state probability vectors, wherein the feature vector is composed of steady-state probabilities from each clue suspicious person to the potential suspicious person;

[0013] S6, obtaining the top-k number of core members expected to be returned, sorting the feature vectors according to the mean of the eigenvalues of the feature vectors and the similarity between the feature vectors, and returning the top-k potentially suspicious persons as the core members of the suspicious organization or gang.

[0014] As a preferred solution, in step S3, the calculation formulas for the average support and the random walk transition probability are respectively:

[0015]

[0016] Among them, the Ave i is the average support of potential suspicious person i, sup(e ix ) is the edge e ix The support of represents the number of common neighbors of potential suspicious persons i and x, N(i) is the set of neighbors of potential suspicious person i, P ij is the random walk transfer probability between potential suspicious persons i and j. When the number of potential suspicious persons is n, we get an n*n random walk transfer matrix P, where the value of the i-th row and j-th column in P is P ij , represents the random walk transfer probability of potential suspicious person i moving to j;

[0017] As a preferred solution, in step S4, the random walk calculation formula is:

[0018] r i =cPr i +(1-c)e i i=1,2,3……m;

[0019] Wherein, c is the re-entry probability, P is the transition probability matrix, ri is the steady-state probability vector, which is a one-dimensional vector that represents the probability that the clue suspicious person i is transferred to other potential suspicious persons after the wandering reaches a steady state, e i is the initial one-dimensional vector, the i-th position is 1, and the rest are 0. In the case of m clues and suspicious persons, we finally get r1, r2, r3...r m .

[0020] As a preferred solution, step S5 specifically includes:

[0021] S5.1, when there are m clues of suspicious persons, the number of potential suspicious persons is n, and each clue of suspicious persons reaches a steady state after random walk, and obtains an n*1 steady-state probability vector r i , combine the steady-state probabilities obtained from all clues and suspicious persons into the matrix R n*m ;

[0022] S5.2, the matrix R n*m Transpose to obtain the matrix After segmentation, we obtain n m*1 feature vectors a i , each eigenvector represents the steady-state probability of all clue suspicious persons i wandering to the potential suspicious person.

[0023] As a preferred solution, step S6 specifically includes:

[0024] S6.1, obtain the top-k number of backbone members expected to be returned and calculate the mean of the eigenvalues of each eigenvector

[0025] Among them, a ix is the eigenvalue, representing the steady-state probability of the clue suspicious person x wandering to the potential suspicious person i;

[0026] S6.2, calculate the distance between each two eigenvectors according to the following formula,

[0027]

[0028] The similarity is calculated according to the following formula:

[0029]

[0030] S6.3, for each potential suspicious person’s feature vector, select the k-1 most similar feature vectors and calculate the average similarity to obtain s(a i ), s(α i The smaller the value of ), the smaller the average distance between the selected k-1 eigenvectors and the eigenvector is, that is, the closer the k eigenvectors are with the eigenvector as the center.

[0031] S6.4, Calculate the score for each potentially suspicious person

[0032] According to the data size of the public security information network data set, α and β are determined, and α and β are mean(a i )and The weight coefficient of score(i) is sorted and the top-k potentially suspicious persons are returned as the backbone members of the suspicious organization.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] The method of the present invention can quickly query the backbone members of the suspicious organization while ensuring high accuracy.

[0035] First, there must be a close connection between the suspicious clues and the core members of the suspicious organization. The random walk subgraph of potential suspicious clues is obtained by extending the search scope based on several suspicious clues, which greatly reduces the time cost and effectively solves the problem of excessive time cost.

[0036] Secondly, the random walk method is adopted instead of the traditional community query method, which avoids calculating the data structure representing the complete community on the graph. The possibility of potential suspicious persons being core members of the suspicious organization is judged according to the probability generated by random walk. The idea of trading precision for time is used to avoid the large amount of time consumed in calculating the data structure representing the complete community.

[0037] This method can ensure that the operation speed is 10 times faster than the traditional method while maintaining 90% accuracy.

[0038] Supplement: Which steps significantly help reduce time complexity? Explain how to promote the reduction of time complexity. It is best to compare the effect with traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flow chart of a method for searching for key members of a suspicious organization in a public security information network according to an embodiment of the present invention;

[0040] Figure 2 This is an example flow chart of a method for searching for key members of a suspicious organization on a public security information network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0041] To more clearly illustrate the embodiments of the present invention, specific embodiments of the present invention will be described below with reference to the accompanying drawings. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings and other embodiments can be obtained based on these drawings without inventive efforts.

[0042] This embodiment provides a method for searching for key members of a suspicious organization in a public security information network. The flowchart is as follows: Figure 1 As shown, in this embodiment, the search for key members of a suspicious organization on a public security information network is taken as an example, and the flowchart for further explanation is as follows. Figure 2 shown.

[0043] First, step S1 is performed to obtain a public security information network data set including natural person data and the relationship between natural persons, and a public security information network is constructed. The public security information network contains nodes and edges, wherein the nodes represent the natural persons in the public security information network data set, and the edges represent the relationship between natural persons in the public security information network data set. Then step S2 is performed to obtain a number of clue suspicious persons and a bound number of hops, and a local subgraph is expanded as a random walk subgraph of potential suspicious persons by extending the number of hops obtained from the several clue suspicious persons to other people. The local subgraph can be obtained by selecting the neighbors within two hops of each clue suspicious person from the several clue suspicious persons, and the union of all neighbors is the potential suspicious person in the local subgraph. The relationship between the potential suspicious persons in the subgraph is filled in according to the public security information network to generate a random walk subgraph of potential suspicious persons; in this example, the several clue suspicious persons are {q1,q2……q m}, for each of the clues, suspicious persons q i , obtain the neighbor set Q within two hops of the clue suspicious person i , find the union of all neighbor sets to obtain all potential suspicious persons, and fill in the edges for these potential suspicious persons to generate a random walk subgraph.

[0044] Then, step S3 is performed to calculate the number of common neighbors of every two potential suspicious persons in the random walk subgraph of potential suspicious persons as the edge support, and the average support of the representative node of the potential suspicious person is calculated based on the support of each edge, and the random walk transition probability between each pair of potential suspicious persons is calculated; in this example, for any edge, the edge support sup(e ij )=|N(i)∩N(j)|, and calculate the average support of potential suspicious persons through the support of edges Then calculate the random walk transfer probability between each pair of potential suspicious persons

[0045] After calculating the random walk transfer probability between each pair of potential suspicious persons, when the number of potential suspicious persons is n, an n*n random walk transfer matrix P is obtained, where the value of the i-th row and j-th column in P is P ij , represents the random walk transfer probability of potential suspicious person i walking to j; obtain the random walk transfer matrix P, and proceed to step S4. On the potential suspicious person random walk subgraph, each of the clue suspicious persons undergoes a random walk with recurrence until a steady state is reached. Each clue suspicious person will generate a steady-state probability vector after a random walk; in this example, through the random walk formula r i =cPr i +(1-c)e i , calculate the steady-state probability vector r after each clue suspicious person reaches a steady state i , r i The h-th row in represents the probability that the clue suspicious person visits the h-th potential suspicious person.

[0046] After the steady-state probability is calculated, step S5 is performed to generate a feature vector for each potential suspicious person on the potential suspicious person random walk subgraph based on all steady-state probability vectors. The feature vector is composed of the steady-state probability of each clue suspicious person to the potential suspicious person;

[0047] Furthermore, the process of generating the feature vector can be:

[0048] S5.1, when there are m clues of suspicious persons, the number of potential suspicious persons is n, and each clue of suspicious persons reaches a steady state after random walk, and obtains an n*1 steady-state probability vector r i , combine the steady-state probabilities obtained from all clues and suspicious persons into the matrix R n*m In this example, for a case where there are m clues of suspicious persons and n potential suspicious persons, m n*1 steady-state probability vectors are obtained by random walk, and the matrix R is obtained by piecing together. n*m ;

[0049] S5.2, the matrix R n*m Transpose to obtain the matrix After segmentation, we obtain n m*1 feature vectors a i , each eigenvector represents the steady-state probability of all clue suspicious persons i wandering to the potential suspicious person; in this example, after the transposed matrix is divided, n m*1 eigenvectors are obtained, each vector represents the probability of m clue suspicious persons visiting the potential suspicious person.

[0050] After all feature vectors are generated, step S6 can be performed to return the top-k core members of the suspicious organization.

[0051] Specifically, step S6 includes:

[0052] S6.1, obtain the top-k number of backbone members expected to be returned and calculate the mean of the eigenvalues of each vector In this example, the mean value represents the average probability of m clue suspicious persons visiting the potential suspicious person. The greater the probability, the closer the potential suspicious person is to the m clue suspicious persons.

[0053] S6.2, calculate the distance between each two eigenvectors Calculating similarity In this example, a higher similarity between the two feature vectors indicates a higher similarity between the relationship between the potential suspicious person 1 and the m clue suspicious persons and the relationship between the suspicious person 2 and the m clue suspicious persons;

[0054] S6.3, for each potential suspicious person’s feature vector, select the k-1 most similar feature vectors and calculate the average similarity to obtain s(a i ), s(a i The smaller the value of ), the smaller the average distance between the selected k-1 eigenvectors and the eigenvector is, that is, the closer the k eigenvectors are with the eigenvector as the center.

[0055] S6.4, Calculate the score for each potentially suspicious person

[0056] According to the data size of the public security information network data set, α and β are determined, and α and β are mean(a i )and The weight coefficient is . Sorting score(i), returning the top-k potential suspicious persons as the core members of the suspicious organization; in this example, weighting the mean and similarity is considered, returning the top-k potential suspicious persons who are closely connected to the m clue suspicious persons and have high similarity with each other as the core members of the suspicious organization.

[0057] The above method realizes the search method for key members of suspicious organizations in public security information networks by calculating the steady-state probability of the clue suspicious person group to other potential suspicious persons and the similarity between the potential suspicious persons with respect to all clue suspicious persons, and then sorting them, which greatly improves the query efficiency while ensuring high accuracy.

[0058] It should be noted that the above is only a detailed description of the preferred embodiments and principles of the present invention. For ordinary technicians in this field, based on the ideas provided by the present invention, there will be changes in the specific implementation methods, and these changes should also be regarded as the scope of protection of the present invention.

Claims

1. A method for searching key members of suspicious organizations in a public security information network, characterized in that: The following steps are involved: S1, obtaining a public safety information network dataset including natural person data and relationships between natural persons, and constructing a public safety information network, wherein the public safety information network comprises nodes and edges, wherein the nodes represent natural persons in the public safety information network dataset, and the edges represent relationships between natural persons in the public safety information network dataset; S2, obtaining several clues of suspicious persons and obtaining a bound of hops, extending the clues of suspicious persons to other natural persons according to the obtained hops, and expanding a local subgraph as a random walk subgraph of potential suspicious persons; S3, calculate the number of common neighbors of every two potential suspicious persons in the random walk subgraph of potential suspicious persons as the edge support, Calculate the average support of the representative node of the potential suspicious person based on the support of each edge. Calculate the random walk transfer probability between each pair of potential suspicious persons. When the number of potential suspicious persons is n, we get an n*n random walk transfer matrix P, where the value of the i-th row and j-th column in P is P ij , represents the random walk transfer probability of potential suspicious person i moving to j; S4, on the potential suspicious person random walk subgraph, each of the plurality of clue suspicious persons undergoes a random walk with recurrence based on the random walk transfer matrix P until a steady state is reached. Each clue suspicious person generates a steady-state probability vector after the random walk; S5, generating a feature vector for each potential suspicious person on the potential suspicious person random walk subgraph based on all steady-state probability vectors, wherein the feature vector is composed of steady-state probabilities from each clue suspicious person to the potential suspicious person; S6, obtaining the top-k number of core members expected to be returned, sorting the feature vectors according to the mean of the eigenvalues of the feature vectors and the similarity between the feature vectors, and returning the top-k potentially suspicious persons as the core members of the suspicious organization or gang.

2. A method for searching for key members of a suspicious organization in a public security information network according to claim 1, characterized in that: In step S3, the calculation formulas for the average support and the random walk transition probability are respectively: Among them, the Ave i is the average support of potential suspicious person i, sup(e ix ) is the edge e ix The support of represents the number of common neighbors of potential suspicious persons i and x, N(i) is the set of neighbors of potential suspicious person i, P ij is the random walk transfer probability between potential suspicious persons i and j.

3. A method for searching for key members of a suspicious organization in a public security information network according to claim 1, characterized in that: In step S4, the calculation formula of the random walk is: r i =cPr i +(1-c)e i i=1,2,3……m; Wherein, c is the re-entry probability, P is the transition probability matrix, r i is the steady-state probability vector, which is a one-dimensional vector that represents the probability that the clue suspicious person i is transferred to other potential suspicious persons after the wandering reaches a steady state, e i is the initial one-dimensional vector, the i-th position is 1, and the rest are 0; in the case of m clues and suspicious persons, we finally get r1, r2, r3...r m .

4. A method for searching for key members of a suspicious organization in a public security information network according to claim 1, characterized in that: The step S5 specifically includes the following steps: S5.1, when there are m clues of suspicious persons, the number of potential suspicious persons is n, and each clue of suspicious persons reaches a steady state after random walk, and obtains an n*1 steady-state probability vector r i , combine the steady-state probabilities obtained from all the clues and suspicious persons into the matrix R n*m ; S5.2, the matrix R n*m Transpose to obtain the matrix After segmentation, we obtain n m*1 feature vectors a i , each eigenvector represents the steady-state probability of all clue suspicious persons i wandering to the potential suspicious person.

5. A method for searching for key members of a suspicious organization in a public security information network according to claim 1, characterized in that: The step S6 specifically includes the following steps: S6.1, obtain the top-k number of backbone members expected to be returned and calculate the mean of the eigenvalues of each eigenvector Among them, a ix is the eigenvalue, representing the steady-state probability of the clue suspicious person x wandering to the potential suspicious person i; S6.2, calculate the distance between each two eigenvectors according to the following formula, The similarity is calculated according to the following formula: S6.3, for each potential suspicious person’s feature vector, select the k-1 most similar feature vectors and calculate the average similarity to obtain s(a i ), s(a i The smaller the value of ), the smaller the average distance between the selected k-1 eigenvectors and the eigenvector is, that is, the closer the k eigenvectors are with the eigenvector as the center. S6.4, Calculate the score for each potentially suspicious person According to the data size of the public security information network data set, α and β are determined, and α and β are mean(a i )and The weight coefficient of score(i) is calculated; score(i) is sorted and the top-k potentially suspicious persons are returned as the backbone members of the suspicious organization.

Images