Multi-task propagation network inference method and device, equipment and readable storage medium

Abstract

The application provides a multitask propagation network inference method, device and equipment and a readable storage medium. In the application, based on a similar information propagation network, firstly, a statistical method is used to reveal the potential influence relationship between nodes in the propagation network, and then the influence relationship between nodes in the similar propagation network is used to iteratively correct the existing influence relationship in the propagation network until the propagation network no longer changes with the number of iterations, that is, the final influence relationship graph is determined. By mining the relationship information in the similar information propagation network, the application uses the information to increase the inference accuracy of the finally obtained propagation network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information propagation, and in particular to a multi-task propagation network inference method, device and equipment and readable storage medium. BACKGROUND

[0002] Information propagation is a basic process of social network and an indispensable link in social activities. Analyzing the information propagation mode in social network can guide our work in real life. For example, disease propagation is a common mode of information propagation. Mastering the flow direction of disease propagation can help us do a good job in disease prevention and control and reduce losses. In analyzing the mode of information propagation, the main purpose is to construct an information propagation network according to the recorded historical information propagation data and further infer the flow direction of information propagation in the next stage to help researchers better promote or prevent information propagation.

[0003] Existing propagation network inference methods need infection time information in the propagation process or node-based infection state data for propagation network inference. The propagation network inference method based on node-based infection state data is more and more common. They use statistical methods to reveal the potential influence relationship between nodes, and then calculate the influence degree between nodes to correct the existing influence relationship, so as to determine the final influence relationship graph. However, the mainstream node-based infection state data propagation network inference method can only infer a single network and cannot utilize the relationship in networks with different time periods but similar structures. For example, disease propagation, most people have relatively fixed social circles and activity ranges. For disease propagation in different time periods in the same area, there will be a certain degree of similarity in the propagation path. In disease propagation inference, the effective information contained in these similar networks cannot be utilized. There is effective relationship information in similar information propagation networks, so new methods are needed to mine the potential information in similar networks and enhance the propagation network inference ability. SUMMARY

[0004] To solve the above technical problems, the main purpose of the present application is to provide a multi-task propagation network inference method, device and equipment and readable storage medium.

[0005] In the first aspect, the present application provides a multi-task propagation network inference method, which comprises:

[0006] Step 1: for directed graphs G={G1, G2,..., G k} in the directed graph G r , each node v r in G r，i is considered in turn, and the remaining each node v r，j ∈v r \v r，i} is calculated vr，i The probability of the parent node:

[0007]

[0008] Where r∈{1,2,...,k}, k≥1 represents the number of directed graphs that make up the directed graph set G. Each directed graph in the directed graph set G consists of nodes and directed edges, and all directed graphs have the same number of nodes. X r，i x r，j ∈{0,1} is the infection state variable, and β represents the number of transmission processes; and These represent the nodes v after the l-th propagation process. r，i and v r，j The infection status; I(.) represents an indicator function, which takes the value 1 when (.) is true and takes the value 0 when it is false;

[0009] Step 2: For the directed graph G in the directed graph set G... r Where r∈{1, 2, ..., k}, calculate the value of each node v in the directed graph in turn. r，i and v r，j The activation Pearson correlation coefficient (APCC) between the two r，i v r，j ):

[0010]

[0011] Where β represents the number of propagation processes; and These represent the nodes v after the l-th propagation process. r，i and v r，j The infection status; I(.) represents an indicator function, which takes the value 1 when (.) is true and takes the value 0 when it is false;

[0012] Step 3: Initialize the directed graph G r G r ={V r E r W r}, r∈{1, 2, ..., k}, V r Let E represent the n nodes that make up the network. r G represents r The set of directed edges in the array is initially empty. For each node v r，i The choice is v r，i The set F consists of the p nodes with the highest parent node probability. r，i In E rAdd the corresponding directed edges and set the weights on the directed edges to the activation Pearson correlation coefficient between the two nodes;

[0013] Step 4: For W r The weighted data in the dataset is clustered using the K-means method, with the number of clusters set to 2. For the two groups of elements after clustering, assuming t r To find the maximum value in the smaller set of elements, set G... r Medium weight less than t r Delete the edge of G. r The graph notation formed by the remaining edges in the graph is as follows:

[0014] Step 5: [Regarding...] Perform iterations on the graph after T iterations. Each node In the figure Found set of parent nodes At the same time The infection status of each node after the end of the l-th propagation process is recorded in the set. For the set of parent nodes Perform all combinations, and calculate g for each combination. r Fraction, take g r The node with the maximum score Fractions, corresponding combinations are nodes set of parent nodes The node score is calculated as follows:

[0015] in, express Nodes in The node score, For nodes The set of parent nodes after T iterations Represents a set The number of elements in the set; k≥1 represents the number of directed graphs that make up the directed graph set G; α is an adjustment parameter; representing N iab X appeared in all infection results r，i =b and The number of times b takes values ​​{0, 1}, and is a variable. In the summation operation, N is used as the traversal variable. ia =N ia0 +N ia1 ; Represents graph G s Does V exist in it? s，i To V s，jFor each edge G, if it exists, its value is 1; otherwise, its value is 0. s With G r Both belong to the directed graph set G, G s ∈G\{G r};V s，i It is G s The i-th node, V s，j It is G s The j-th node;

[0016] Step 6: According to the diagram The set of parent nodes of each node in the graph is updated. Connecting nodes With parent node set Directed edges between nodes, removing other nodes that point to them. The directed edges, the updated graph notation is If any graph exists and If there are differences, increment the iteration count T by 1 and return to step 5; if all and If they are the same, output the directed graph set obtained through iteration.

[0017] Secondly, the present invention also provides a multi-task propagation network inference apparatus, the multi-task propagation network inference apparatus comprising:

[0018] The first computation module is used for: for a directed graph set G = {G1, G2, ..., G...} k The directed graph G in} r Consider G in turn r Each node v in r，i Calculate the remaining node v r，j ∈v r \v r，i} is v r，i The probability of the parent node:

[0019]

[0020] Where r∈{1,2,...,k}, k≥1 represents the number of directed graphs that make up the directed graph set G. Each directed graph in the directed graph set G consists of nodes and directed edges, and all directed graphs have the same number of nodes. X r，i X r，j ∈{0,1} is the infection state variable, and β represents the number of transmission processes; and These represent the nodes v after the l-th propagation process. r，i and v r，jan infection state of a node v

[0021] a second calculation module, configured to: for a directed graph G r in the set of directed graphs G r，i , calculate an activation Pearson Correlation Coefficient (APCC) between nodes v r，j and v r，i in each directed graph G r，j :

[0022]

[0023] wherein β represents a number of propagation processes; and respectively represent an infection state of a node v r，i and v r，j after the lth propagation process; I(.) represents an indicator function, which takes a value of 1 when (.) is true, and takes a value of 0 when (.) is false;

[0024] an initialization module, configured to: initialize a directed graph G r , wherein G r = {V r , E r , W r}, r ∈ {1, 2,..., k}, V r represents n nodes constituting a network, E r represents a set of directed edges in G r , which is initially an empty set, for each node v r，i , select p nodes with the largest parent node probability to form a set F r，i , add corresponding directed edges in E r，i , and set a weight on the directed edge as an activation Pearson Correlation Coefficient between the two nodes;

[0025] a clustering module, configured to: cluster weight data in W r using a K-means method, and set a clustering number as 2, for two groups of elements after clustering, assuming t r is a maximum value in the smaller group of elements, remove edges with a weight less than t r in G r , and form a graph composed of remaining edges in G r as G

[0026] an iteration module, configured to: iterate G , and obtain a graph G after T iterations Each node In the figure Found set of parent nodes At the same time The infection status of each node after the end of the l-th propagation process is recorded in the set. For the set of parent nodes Perform all combinations, and calculate g for each combination. r Fraction, take g r The node with the maximum score Fractions, corresponding combinations are nodes set of parent nodes The node score is calculated as follows:

[0027] in, express Nodes in The node score, For nodes The set of parent nodes after T iterations Represents a set The number of elements in the set; k≥1 represents the number of directed graphs that make up the directed graph set G; α is an adjustment parameter; representing N iab X appeared in all infection results r，i =b and The number of times b takes values ​​{0, 1}, and is a variable. In the summation operation, N is used as the traversal variable. ia =N ia0 +N ia1 ; Represents graph G s Does V exist in it? s，i To V s，j For each edge G, if it exists, its value is 1; otherwise, its value is 0. s With G r Both belong to the directed graph set G, G s ∈G\{G r};V s，i It is G s The i-th node, V s，j It is G s The j-th node;

[0028] The determination module is used to: determine the result based on the graph. The set of parent nodes of each node in the graph is updated. Connecting nodes With parent node set Directed edges between nodes, removing other nodes that point to them. the updated graph is denoted as If there exists any graph different from , the iteration number T is added by 1, and the action of returning to the iteration module is performed; if all are the same as , the directed graph set obtained through iteration is output.

[0029] In a third aspect, the present application further provides a multi-task propagation network inference device, comprising a processor, a memory, and a multi-task propagation network inference program stored in the memory and executable by the processor, wherein the multi-task propagation network inference program, when executed by the processor, implements the steps of the multi-task propagation network inference method as described above.

[0030] In a fourth aspect, the present application further provides a readable storage medium, wherein the readable storage medium stores a multi-task propagation network inference program, and the multi-task propagation network inference program, when executed by a processor, implements the steps of the multi-task propagation network inference method as described above.

[0031] In the present application, based on similar information propagation networks, the potential influence relationship between nodes in the propagation network is first revealed by using statistical methods, and then the existing influence relationship in the propagation network is iteratively corrected by using the influence relationship between nodes in the similar propagation network until the propagation network no longer changes with the iteration number, i.e., the final influence relationship graph is determined. The present application increases the inference accuracy of the final obtained propagation network by mining the relationship information in similar information propagation networks. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 FIG. 1 is a flowchart of an embodiment of the multi-task propagation network inference method of the present application;

[0033] Figure 2 FIG. 4 is a result graph of the F value corresponding to the influence relationship graph constructed by the embodiment of the present application on the artificial network set generated by the LFR algorithm with different numbers of nodes;

[0034] Figure 3 FIG. 5 is a result graph of the F value corresponding to the influence relationship graph constructed by the embodiment of the present application on the artificial network set generated by the LFR algorithm with different similarity degrees;

[0035] Figure 4 FIG. 6 is a result graph of the F value corresponding to the influence relationship graph constructed by the embodiment of the present application on the network set generated based on the real networks NetSci and DUNF with different numbers of networks;

[0036] Figure 5is a result graph of the F value corresponding to the influence relationship graph constructed by the different numbers of propagation trajectory data used by the network set generated by the real network NetSci and DUNF according to an embodiment of the present application;

[0037] Figure 6 A hardware structure schematic diagram of a multi-task propagation network inference device involved in an embodiment of the present application is shown.

[0038] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0039] It should be understood that the specific embodiments described herein are merely intended to explain the present application, and are not intended to limit the present application.

[0040] In a first aspect, an embodiment of the present application provides a multi-task propagation network inference method.

[0041] In an embodiment, with reference to Figure 1 , Figure 1 A flowchart of an embodiment of the multi-task propagation network inference method of the present application is shown. As shown in Figure 1 , the multi-task propagation network inference method comprises:

[0042] Step 1: for a directed graph G k in a directed graph set G={G1, G2,..., G r , each node v r in G r，i is considered in turn, and the probability that each remaining node v r，j ∈v r \{v r，i} is a parent node of v r，i is calculated:

[0043]

[0044] wherein r∈{1, 2,..., k}, k≥1 represents the number of directed graphs constituting the directed graph set G, each directed graph in the directed graph set G is composed of nodes and directed edges, and the number of nodes of all directed graphs is the same, X r，i , X r，j ∈{0, 1} is an infection state variable, and β represents the number of propagation processes; and respectively represent the infection state of node v r，i and v r，j after the end of the lth propagation process; I(.) represents an indicator function, which takes the value 1 when (.) is true, and takes the value 0 when (.) is not true;

[0045] Step 2: for a directed graph Gr where r e {1, 2,..., k}, the activation Pearson correlation coefficient APCC(V r，i , V r，j ) between each pair of nodes v r，i and v r，j in each directed graph is calculated in turn:

[0046]

[0047] where β represents the number of propagation processes; and respectively represent the infection state of nodes v r，i and v r，j after the lth propagation process; I(.) represents an indicator function, which takes the value 1 when (.) is true, and takes the value 0 when (.) is false;

[0048] Step 3: initialize the directed graph G r , where G r = {V r , E r , W r}, r e {1, 2,..., k}, V r represents the n nodes that make up the network, E r represents the set of directed edges in G r , which is initially empty, for each node v r，i , select the p nodes with the largest parent node probability to form the set F r，i , add the corresponding directed edges in E r，i , and set the weight of the directed edge to the activation Pearson correlation coefficient between the two nodes;

[0049] Step 4: use the K-means method to cluster the weight data in W r , set the cluster number to 2, for the two groups of elements after clustering, assume t r is the maximum value in the smaller group of elements, delete the edges in G r with weights less than t r , and form a graph G r from the remaining edges in G r , denoted as

[0050] Step 5: iterate on G , for each node v in the graph G after T iterations, find the parent node set P of v in G , and update G The infection status of each node after the end of the l-th propagation process is recorded in the set. For the set of parent nodes Perform all combinations, and calculate g for each combination. r Fraction, take g r The node with the maximum score Fractions, corresponding combinations are nodes set of parent nodes The node score is calculated as follows:

[0051] in, express Nodes in The node score, For nodes The set of parent nodes after T iterations Represents a set The number of elements in the set; k≥1 represents the number of directed graphs that make up the directed graph set G; α is an adjustment parameter; representing N iab X appeared in all infection results r，i =b and The number of times b takes values ​​{0, 1}, and is a variable. In the summation operation, N is used as the traversal variable. ia =N ia0 +N ia1 ; Represents graph G s Does V exist in it? s，i To V s，j For each edge G, if it exists, its value is 1; otherwise, its value is 0. s With G r Both belong to the directed graph set G, G s ∈G\{G r};V s，i It is G s The i-th node, V s，j It is G s The j-th node;

[0052] Step 6: According to the diagram The set of parent nodes of each node in the graph is updated. Connecting nodes With parent node set Directed edges between nodes, removing other nodes that point to them. The directed edges, the updated graph notation is If any graph exists and If there are differences, increment the iteration count T by 1 and return to step 5; if all With the same, the output iteration of the directed graph set.

[0053] In this embodiment, 20 network sets are used, as shown in Table 1, wherein the network sets NetSet1-NetSet5 are artificial networks with different node numbers generated by the LFR algorithm, NetSet6-NetSet 10 are artificial networks with different network similarity generated by the LFR algorithm, NetSet 11 -NetSet 15 are network sets with different network numbers generated based on the real network NetSci, NetSet 16 -NetSet 20 are network sets with different network numbers generated based on the real network DUNF. The propagation data generation process of each network is as follows: assuming that the influence degree between each node in the network is the same, 15% of the nodes in the test network are randomly selected as initial "infected" points each time, and propagation simulation is performed according to the IC model, and after the propagation simulation is completed, the infection state of each node is recorded as a propagation trajectory data, and multiple data are generated as observation data sets for the experiment. The generation process of the network set is as follows: first, the benchmark network with the corresponding parameters is generated by the LFR algorithm, and 1 / 3 of the edges of the benchmark network are randomly removed to obtain a sub-network, and then the sub-network set can be obtained; the network set is generated based on the real network NetSci and DUNF by the same method. This method can ensure that the network similarity is the same and the network set parameters are the same as the benchmark network. The measurement of the network similarity uses the Vertex / Edge overlap formula:

[0054]

[0055]

[0056] Table 1

[0057] Referring to Figure 2 , Figure 2 is the result graph of the F value corresponding to the influence relationship graph constructed by the embodiment of the present application on the artificial network set with different node numbers generated by the LFR algorithm. Figure 2 The accuracy (the higher the F value, the more accurate) of the running results of the scheme in this embodiment on the network set with different scales is shown in the graph. The accuracy of the scheme in this embodiment is higher on the five different network sets in the graph, and the accuracy of the scheme in this embodiment has an upward trend as the network size increases.

[0058] Referring to Figure 3 , Figure 3is a result graph of F value corresponding to the influence relationship graph constructed by the embodiment of the application on the artificial network set of different similarity generated by the LFR algorithm. Figure 3 The accuracy rate of the running result of the embodiment of the application on the different network sets is shown in the figure, the embodiment of the application has a higher accuracy rate in the five network sets in the figure, and the accuracy rate has an upward trend with the increase of the similarity.

[0059] Referring to Figure 4 , Figure 4 is a result graph of F value corresponding to the influence relationship graph constructed by the embodiment of the application on the network set generated based on the real network NetSci and DUNF with different numbers of networks. Figure 4 The accuracy rate of the running result of the embodiment of the application on the real network set with different numbers of networks is shown in the figure, the embodiment of the application has a higher accuracy rate in the five different network sets in the figure, and the accuracy rate of the embodiment of the application has an upward trend with the increase of the number of networks.

[0060] Referring to Figure 5 , Figure 5 is a result graph of F value corresponding to the influence relationship graph constructed by the embodiment of the application on the network set generated based on the real network NetSci and DUNF using different numbers of propagation trajectory data. Figure 5 The accuracy rate of the running result of the embodiment of the application on the real network set with different numbers of propagation trajectories is shown in the figure, the embodiment of the application has a higher accuracy rate in the five different network sets in the figure, and the accuracy rate of the embodiment of the application has an upward trend with the increase of the number of propagation trajectories.

[0061] In the embodiment, based on the similar information propagation network, first, the statistical method is used to reveal the potential influence relationship between the nodes in the propagation network, and then the influence relationship between the nodes in the similar propagation network is used to iteratively correct the existing influence relationship in the propagation network, until the propagation network no longer changes with the number of iterations, that is, the final influence relationship graph is determined. The embodiment mines the relationship information in the similar information propagation network, and uses the information to increase the inference accuracy of the final propagation network.

[0062] Further, in an embodiment, p is an integer in (1, n).

[0063] Further, in an embodiment, a is a real number in (0, 1].

[0064] Further, in an embodiment, before step 1, the following rules are set:

[0065] The rth (r = 1, 2,..., k) directed graph in the directed graph set G is represented as G r = (V r , E r , W r ); Vr = {v r，1 r，2 ,..., v r，n} represents n nodes constituting the rth directed graph; E r is a set of directed edges of G r , wherein if one node in G r can affect another node in G r , there is a directed edge from the one node to the another node on G r ; W r is a set of weights corresponding to the directed edges; F r = {v r，i | (v r，i , v r，j ) ∈ E r，j} is a set of parent nodes of node v r，i in G r ; any node v r，i has two infection states, i.e., infected and uninfected, and the parent nodes of node v r，i have |F | infection states, wherein |F r，i | represents the number of elements in set F r，i ; X r，i is an infection state variable of node v r，i , is an infection state variable of set F r，i ; variable X r，i has a value of 0 or 1, X r，i ∈ {0, 1}, 0 represents uninfected, and 1 represents infected; variable has a value of , and π ria is the ath value of variable ; set D = {D1, D2,..., D k} is an infection state result of directed graph set G, wherein G r corresponding infection state result is D r = {D r，1 , D r，2 ,..., D r，β} records the result after β propagation processes in G r end, wherein represents the infection state of n nodes in G r after the lth propagation process ends, represents the infection state of node v r，i in G r after the lth propagation process ends, represents that node v r，i is infected after the lth propagation process ends,​ This represents node v after the l-th propagation. r，i Not infected.

[0066] Secondly, embodiments of the present invention also provide a multi-task propagation network inference device.

[0067] In one embodiment, the multi-task propagation network inference apparatus includes:

[0068] The first calculation module 10 is used for: for a directed graph set G = {G1, G2, ..., G...} k The directed graph G in} r Consider G in turn r Each node v in r，i Calculate the remaining node v r，j ∈v r \v r，i} is v r，i The probability of the parent node:

[0069]

[0070] Where r∈{1,2,...,k}, k≥1 represents the number of directed graphs that make up the directed graph set G. Each directed graph in the directed graph set G consists of nodes and directed edges, and all directed graphs have the same number of nodes. X r，i X r，j ∈{0,1} is the infection state variable, and β represents the number of transmission processes; and These represent the nodes v after the l-th propagation process. r，i and v r，j The infection status; I(.) represents an indicator function, which takes the value 1 when (.) is true and takes the value 0 when it is false;

[0071] The second calculation module 20 is used for: calculating the directed graph G in the directed graph set G. r Where r∈{1, 2, ..., k}, calculate the value of each node v in the directed graph in turn. r，i and v r，j The activation Pearson correlation coefficient (APCC) between the two r，i V r，j ):

[0072]

[0073] Where β represents the number of propagation processes; and These represent the nodes v after the l-th propagation process. r，i and v r，jThe infection status; I(.) represents an indicator function, which takes the value 1 when (.) is true and takes the value 0 when it is false;

[0074] Initialization module 30 is used to: initialize the directed graph G r G r ={V r E r W r}, r∈{1, 2, ..., k}, V r Let E represent the n nodes that make up the network. r G represents r The set of directed edges in the array is initially empty. For each node v r，i The choice is v r，i The set F consists of the p nodes with the highest parent node probability. r，i In E r Add the corresponding directed edges and set the weights on the directed edges to the activation Pearson correlation coefficient between the two nodes;

[0075] Clustering module 40 is used for: clustering W r The weighted data in the dataset is clustered using the K-means method, with the number of clusters set to 2. For the two groups of elements after clustering, assuming t r To find the maximum value in the smaller set of elements, set G... r Medium weight less than t r Delete the edge of G. r The graph notation formed by the remaining edges in the middle is:

[0076] Iteration module 50, used for: Perform iterations on the graph after T iterations. Each node In the figure Found set of parent nodes At the same time The infection status of each node after the end of the l-th propagation process is recorded in the set. For the set of parent nodes Perform all combinations, and calculate g for each combination. r Fraction, take g r The node with the maximum score Fractions, corresponding combinations are nodes set of parent nodes The node score is calculated as follows:

[0077] in, express Nodes in The node score, For nodes The set of parent nodes after T iterations Represents a set The number of elements in the set; k≥1 represents the number of directed graphs that make up the directed graph set G; α is an adjustment parameter; representing N iab X appeared in all infection results r，i =b and The number of times b takes values ​​{0, 1}, and is a variable. In the summation operation, N is used as the traversal variable. ia =N ia0 +N ia1 ; Represents graph G s Does V exist in it? s，i To V s，j For each edge G, if it exists, its value is 1; otherwise, its value is 0. s With G r Both belong to the directed graph set G, G s ∈G\{G r};V s，i It is G s The i-th node, V s，j It is G s The j-th node;

[0078] Decision module 60 is used to: determine the result based on the diagram. The set of parent nodes of each node in the graph is updated. Connecting nodes With parent node set Directed edges between nodes, removing other nodes that point to them. The directed edges, the updated graph notation is If any graph exists and If there are differences, the iteration count T is incremented by 1, and the action of the iteration module is returned; if all and If they are the same, output the directed graph set obtained through iteration.

[0079] Furthermore, in one embodiment, p∈(1,n), where p is an integer.

[0080] Furthermore, in one embodiment, α∈(0,1], where α is a real number.

[0081] Furthermore, in one embodiment, the multi-task propagation network inference apparatus further includes a setting module, used for:

[0082] Set the following rules:

[0083] The rth (r = 1, 2,..., k) directed graph in the directed graph set G is denoted as G r = (V r , E r , W r ); V r = {v r，1 , v r，2 ,..., v r，n} represents n nodes constituting the rth directed graph; E r is a directed edge set of G r , wherein if a node in G r can affect another node in G r , there is a directed edge from the former to the latter in G r ; W r is a weight set corresponding to the directed edge; the parent node set of a node v r，i in G r，i is F r，j = {v r，j |(v r，i , v r ) ∈ E r，i}; any node v r，i has two infection states, namely infected and uninfected, and the parent nodes of the v r，i node have infection states, wherein |F r，i | represents the number of elements in the set F r，i ; X r，i is an infection state variable of the node v r，i , r，i is an infection state variable of the set F r，i ; the variable X ria has a value of 0 or 1, X k ∈ {0, 1}, 0 represents uninfected, and 1 represents infected; the variable has values, π r is the ath value of the variable , the set D = {D1, D2,..., D r} is an infection state result of the directed graph set G, wherein the corresponding infection state result of G r，1 is D r，2 = {D r，β , D r ,..., D r} records the result after the βth propagation process in G r is completed, wherein represents the infection state of n nodes in G r，i after the lth propagation process is completed, denotes that node v is infected after the lth propagation process r the infection state of node v r，i , denotes that node v is infected after the lth propagation process r，i , denotes that node v is not infected after the lth propagation process r，i .

[0084] The functions of each module in the multi-task propagation network inference device correspond to the steps in the multi-task propagation network inference method, and the functions and implementation processes will not be repeated here.

[0085] In a third aspect, an embodiment of the present application provides a multi-task propagation network inference device, which can be a personal computer (PC), a notebook computer, a server, or other devices with data processing functions.

[0086] Referring to Figure 6 , Figure 6 is a schematic diagram of the hardware structure of the multi-task propagation network inference device involved in the embodiment of the present application. In the embodiment of the present application, the multi-task propagation network inference device can include a processor 1001 (such as a central processing unit (CPU)), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to realize the connection and communication between these components; the user interface 1003 can include a display screen (Display) and an input unit such as a keyboard (Keyboard); the network interface 1004 can optionally include a standard wired interface, a wireless interface (such as a wireless fidelity (WIreless-FIdelity, WI-FI) interface); the memory 1005 can be a high-speed random access memory (RAM), and can also be a stable memory (non-volatile memory) such as a disk memory; the memory 1005 can optionally also be a storage device independent of the aforementioned processor 1001. Those skilled in the art can understand that the hardware structure shown in the figure is not a limitation on the present application, and can include more or fewer components than shown in the figure, or combine certain components, or different component arrangements. Figure 6 The hardware structure shown in the figure is not a limitation on the present application, and can include more or fewer components than shown in the figure, or combine certain components, or different component arrangements.

[0087] Referring to Figure 6 , Figure 6The memory 1005 as a computer storage medium can include an operating system, a network communication module, a user interface module, and a multi-task propagation network inference program. The processor 1001 can invoke the multi-task propagation network inference program stored in the memory 1005 and execute the multi-task propagation network inference method provided by the embodiments of the present application.

[0088] In a fourth aspect, the embodiments of the present application also provide a readable storage medium.

[0089] The readable storage medium of the present application stores a multi-task propagation network inference program, wherein the multi-task propagation network inference program is executed by a processor to implement the steps of the multi-task propagation network inference method as described above.

[0090] The method implemented when the multi-task propagation network inference program is executed can refer to each embodiment of the multi-task propagation network inference method of the present application, which will not be described here.

[0091] It should be noted that in this paper, the term "include", "contain" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or includes elements inherent to such process, method, article or system. Without more limitations, the element defined by the statement "including a" does not exclude the presence of another identical element in the process, method, article or system including the element.

[0092] The above-mentioned embodiment numbers of the present application are only for description, not representing the advantages and disadvantages of the embodiments.

[0093] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be realized by means of software and necessary general hardware platforms, of course, they can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application or the parts that contribute to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a plurality of instructions to make a terminal device execute the methods described in the embodiments of the present application.

[0094] The above is only the preferred embodiment of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent flow transformation made by using the content of the present application specification and drawings, or directly or indirectly applied to other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A multitask propagation network inference method, characterized in that, The multi-task propagation network inference method comprises: Step 1: For each directed graph G k in the set of directed graphs G = {G1, G2, …, G r} r , consider each node v r,i in G r,j in turn, and compute the probability that each node v r \ {v r,i} is a parent of v r,i . where r e {1, 2, …, k}, k > 1 represents the number of directed graphs constituting the directed graph set G, each directed graph in the directed graph set G is composed of nodes and directed edges, and the number of nodes of all directed graphs is the same, X r,i , r,j e {0, 1} is an infection state variable, and β represents the number of propagation processes; and respectively represent the infection states of nodes v r,i and v r,j after the lth propagation process ends; I(.) represents an indicator function, which takes the value 1 when (.) is true, and takes the value 0 when (.) is not true. Step 2: For each directed graph G in the set of directed graphs G r where r e {1,2,...,k}, compute the activation Pearson Correlation Coefficient APCC(v r,i ,v r,j ) between nodes v r,i and v r,j in each directed graph in turn: where β denotes the number of propagation processes; and denote the infection status of node v r,i and v r,j after the lth propagation process, respectively; I(.) denotes an indicator function that takes value 1 when (.) is true and 0 otherwise. Step 3: Initialize the directed graph G r G r ={V r E r W r }, r∈{1,2,…,k}, V r Let E represent the n nodes that make up the network. r G represents r The set of directed edges in the array is initially empty. For each node v r,i The choice is v r,i The set F consists of the p nodes with the highest parent node probability. r,i In E r Add the corresponding directed edges and set the weights on the directed edges to the activation Pearson correlation coefficient between the two nodes; Step 4: Cluster the weight data in W using K-means method, set the cluster number as 2, for the two clusters, assume t is the maximum value in the smaller cluster, delete the edges in G whose weight is less than t, the graph composed of the remaining edges in G is denoted as G r . r r r r ​​​​ Step 5: [Regarding...] Perform iterations on the graph after T iterations. Each node In the figure Found set of parent nodes At the same time The infection status of each node after the end of the l-th propagation process is recorded in the set. For the set of parent nodes Perform all combinations, and calculate g for each combination. r Fraction, take g r The node with the maximum score Fractions, corresponding combinations are nodes set of parent nodes The node score is calculated as follows: wherein, denotes the node score of a node in the set of nodes, is the set of parent nodes of a node after T iterations, denotes the set of nodes; k≥1 denotes the number of directed graphs constituting the set G of directed graphs; a is a tuning parameter; denotes the number of nodes in the set N iab X appears in all infection results r,i = b and the number of times b takes the values {0,1}, which is a variable, in the summation operation of N ia = N ia0 + N ia1 ; denotes whether there is an edge from V s,i to V s,j in the graph G s , if there is an edge the value is 1, if there is no edge the value is 0, G s and G r belong to the set G of directed graphs, G s ∈ G \ {G r}; V s,i is the i-th node of G s , V s,j is the j-th node of G s ; Step 6: According to the diagram The set of parent nodes of each node in the graph is updated. Connecting nodes With parent node set Directed edges between nodes, removing other nodes that point to them. The directed edges, the updated graph notation is If any graph exists and If there are differences, increment the iteration count T by 1 and return to step 5; if all and If they are the same, output the directed graph set obtained through iteration.

2. The multitask propagation network inference method of claim 1, wherein, p∈(1,n), p is an integer.

3. The multitask propagation network inference method of claim 1, wherein, α∈(0,1], α is a real number.

4. The multitask propagation network inference method of claim 1, wherein, Before step 1, the following rules are set: The rth(r=1, 2, …, k) directed graph in the directed graph set G is denoted as G r = (V r , E r , W r ) ; V r = {v r,1 , v r,2 , …, v r,n} represents n nodes constituting the rth directed graph; E r is a directed edge set of G r , wherein if a node in G r can affect another node in G r , there is a directed edge from the former to the latter in G r ; W r is a weight set corresponding to the directed edge; the parent node set of a node v r in G r,i is F r,i = {v r,j | (v r,j , v r,i ) ∈ E r} ; any node v r,i has two infection states, i.e., infected and uninfected, and the parent nodes of the node v r,i have infection states, wherein |F r,i | represents the number of elements in the set F r,i ; X r,i is an infection state variable of the node v r,i , is an infection state variable of the set F r,i ; the variable X r,i has a value of 0 or 1, X r,i ∈ {0, 1}, 0 indicating uninfected and 1 indicating infected; the variable has values, π ria is the ath value of the variable , the set D = {D1, D2, …, D k} is an infection state result of the directed graph set G, wherein the corresponding infection state result of G r is D r = {D r,1 , D r,2 , …, D r,β} records the result after the βth propagation process in G r ends, wherein represents the infection state of n nodes in G r after the lth propagation process ends, G represents the end of the l-th propagation process. r Middle node v r,i The infection status, This indicates that node v has completed the l-th propagation process. r,i Infected This represents node v after the l-th propagation. r,i Not infected.

5. A multitask propagation network inference apparatus, comprising: The multi-task propagation network inference device comprises: The first computation module is used for: for a directed graph set G = {G1, G2, ..., G...} k The directed graph G in} r Consider G in turn r Each node v in r,i Calculate the remaining node v r,j ∈v r \{v r,i } is v r,i The probability of the parent node: where r e {1, 2, …, k}, k > 1 represents the number of directed graphs constituting the directed graph set G, each directed graph in the directed graph set G is composed of nodes and directed edges, and the number of nodes of all directed graphs is the same, X r,i , r,j e {0, 1} is an infection state variable, and β represents the number of propagation processes; and respectively represent the infection states of nodes v r,i and v r,j after the end of the lth propagation process; I(.) represents an indicator function, which takes the value 1 when (.) is true, and takes the value 0 when (.) is not true. The second calculation module is used for: calculating the directed graph G in the directed graph set G. r Where r∈{1,2,…,k}, calculate the value of each node v in the directed graph sequentially. r,i and v r,j The activation Pearson correlation coefficient (APCC) between the two r,i V r,j ): where β denotes the number of propagation processes; and denote the infection status of node v r,i and v r,j after the lth propagation process, respectively; I(.) denotes an indicator function that takes value 1 when (.) is true and 0 otherwise. The initialization module is used to initialize the directed graph G. r G r ={V r E r W r }, r∈{1,2,…,k}, V r Let E represent the n nodes that make up the network. r G represents r The set of directed edges in the array is initially empty. For each node v r,i The choice is v r,i The set F consists of the p nodes with the highest parent node probability. r,i In E r Add the corresponding directed edges and set the weights on the directed edges to the activation Pearson correlation coefficient between the two nodes; The clustering module is used for: clustering W r The weighted data in the dataset is clustered using the K-means method, with the number of clusters set to 2. For the two groups of elements after clustering, assuming t r To find the maximum value in the smaller set of elements, set G... r Medium weight less than t r Delete the edge of G. r The graph notation formed by the remaining edges in the middle is: The iteration module is used for: Perform iterations on the graph after T iterations. Each node In the figure Found set of parent nodes At the same time The infection status of each node after the end of the l-th propagation process is recorded in the set. For the set of parent nodes Perform all combinations, and calculate g for each combination. r Fraction, take g r The node with the maximum score Fractions, corresponding combinations are nodes set of parent nodes The node score is calculated as follows: wherein, denotes the node score of a node in the set of nodes, is the set of parent nodes of a node after T iterations, denotes the set of nodes; k≥1 denotes the number of directed graphs constituting the set G of directed graphs; a is a tuning parameter; denotes the number of elements in the set N iab the number of times X r,i = b and occurs, b takes values from {0,1}, which is a variable, in the summation operation of , N ia = N ia0 + N ia1 ; denotes whether there is an edge from V s to V s,i in the graph G s,j , if there is an edge, the value is 1, if there is no edge, the value is 0, G s and G r belong to the set G of directed graphs, G s ∈ G\{G r}; V s,i is the i-th node of G s , V s,j is the j-th node of G s ; The determination module is used to: determine the result based on the graph. The set of parent nodes of each node in the graph is updated. Connecting nodes With parent node set Directed edges between nodes, removing other nodes that point to them. The directed edges, the updated graph notation is If any graph exists and If there are differences, the iteration count T is incremented by 1, and the action of the iteration module is returned; if all and If they are the same, output the directed graph set obtained through iteration.

6. The multitask propagation network inference apparatus of claim 5, wherein, p∈(1,n), p is an integer.

7. The multitask propagation network inference apparatus of claim 5, wherein, α∈(0,1], α is a real number.

8. The multitask propagation network inference apparatus of claim 5, wherein, The multi-task propagation network inference device further comprises a setting module for: The following rules are set: The rth (r = 1, 2, …, k) directed graph in the directed graph set G is denoted as G r = (V r , E r , W r ) ; V r = {v r,1 ,v r,2 ,…,v r,n} represents n nodes constituting the rth directed graph; E r is a directed edge set of G r , wherein if a node in G r can affect another node in G r , there is a directed edge from the former to the latter on G r ; W r is a weight set corresponding to the directed edge; the parent node set of a node v r in G r,i is F r,i = {v r,j |(v r,j ,v r,i ) ∈ E r} ; any node v r,i has two infection states, i.e., infected and uninfected, and the parent node of the node v r,i has infection states, wherein |F r,i | represents the number of elements in the set F r,i ; X r,i is an infection state variable of the node v r,i , is an infection state variable of the set F r,i ; the variable X r,i has a value of 0 or 1, X r,i ∈ {0, 1}, 0 indicating uninfected and 1 indicating infected; the variable has values, π ria is the ath value of the variable , the set D = {D1, D2, …, D k} is an infection state result of the directed graph set G, wherein the corresponding infection state result of G r is D r = {D r,1 ,D r,2 ,…,D r,β}, which records the result after the βth propagation process in G r ends, wherein represents the infection state of the n nodes in G r after the lth propagation process ends, G represents the end of the l-th propagation process. r Middle node v r,i The infection status, This indicates that node v has completed the l-th propagation process. r,i Infected This represents node v after the l-th propagation. r,i Not infected.

9. A multitask propagation network inference device, comprising: The multi-task propagation network inference equipment comprises a processor, a memory, and a multi-task propagation network inference program stored on the memory and executable by the processor, wherein when the multi-task propagation network inference program is executed by the processor, the steps of the multi-task propagation network inference method according to any one of claims 1 to 4 are implemented.

10. A readable storage medium, characterized by, The readable storage medium stores a multi-task propagation network inference program, wherein when the multi-task propagation network inference program is executed by the processor, the steps of the multi-task propagation network inference method according to any one of claims 1 to 4 are implemented.

Images