Method and system for maximizing influence of dynamic network structure based on learning automaton

Abstract

The present invention provides a method and system for maximizing the influence of a dynamic network structure based on a learning automaton, comprising: 0 ,G 1 ,…} is mapped into a short-term stable non-stationary environment consisting of a set of stable environments {E 0 ,E 1 ,…}, initialize the seed node set and current influence; based on the learning automaton LA algorithm, in the environment E 0 Explore and get the initial network G 0 Based on the adaptive sliding window algorithm, two consecutive network topology structures G 0 →G 1 Perform change detection; between the old and new network G 0 →G 1 The historical information of the interaction between the learning automaton and the environment is transferred to obtain the seed node set S in the new network * The present invention transforms the influence maximization problem into an optimization learning problem by learning an automaton algorithm without using network structure information and without limiting a specific propagation model, thereby achieving rapid selection of an influential seed node set.

Description

Technical Field

[0001] The present invention relates to the technical field of learning automata, and in particular to a method and system for maximizing the influence of a dynamic network structure based on learning automata. Background Art

[0002] The study of the influence maximization problem has good application prospects in online public opinion guidance, online rumor refutation, online marketing, and advertising. Influence maximization has become an important research direction in the field of Internet information dissemination. Since the Internet is a complex network that changes dynamically, the connection relationship between nodes is constantly changing, which leads to changes in the influence of each node. Therefore, it is necessary to update the initial seed node set of information dissemination in time when the dissemination relationship between nodes changes, so as to ensure the stability and effectiveness of solving the influence maximization problem in a dynamic network.

[0003] Therefore, the influence maximization method in the prior art (the Chinese patent with application number CN202210169168.X discloses the "social network influence maximization node identification method") provides a social network influence maximization node identification method and system, including: for a given social network, according to the number of ternary closure structures of each node, calculate the ternary closure weighted degree value of each node; according to the ternary closure weighted degree values ​​of all neighboring nodes of each node, calculate the weighted influence expected value of each node; based on the weighted influence expected value of each node, arrange all nodes in descending order; according to the weighted influence expected value arranged in descending order, identify the influence maximized node. The present invention solves the problem that the time complexity of the existing method for updating the expected value of node influence is too high in the process of identifying nodes with maximized influence in social networks, and the method of selecting nodes with maximized influence is too localized, and the identification result is not accurate enough. Although it can obtain a set of influential nodes in a static network, it cannot quickly update the set of seed nodes in a dynamic network structure.

[0004] Therefore, it is necessary to propose a new technical solution to improve the above technical problems. Summary of the invention

[0005] In view of the defects in the prior art, the purpose of the present invention is to provide a method and system for maximizing the influence of a dynamic network structure based on a learning automaton.

[0006] According to a method for maximizing the influence of a dynamic network structure based on a learning automaton provided by the present invention, the method comprises the following steps:

[0007] Step S1: transform the dynamic network structure {G 0 , G 1, ...} is mapped into a short-term stable non-stationary environment consisting of a set of stable environments {E 0 , E 1 ,…}, initialize the seed node set and current influence;

[0008] Step S2: Based on the learning automaton LA algorithm, in the environment E 0 Explore and get the initial network G 0 The set of seed nodes on ;

[0009] Step S3: Based on the adaptive sliding window algorithm, two consecutive network topology structures G 0 →G 1 Perform change detection;

[0010] Step S4: In the new and old networks G 0 →G 1 The historical information of the interaction between the learning automaton and the environment is transferred to obtain the seed node set S in the new network * .

[0011] Preferably, step S1 comprises the following steps:

[0012] Step S1.1: Network G i With stable environment i One-to-one correspondence, and satisfying the above mapping relationship between G(V, E) and E(A, B, C), where for the network G, V is the set of network participants, i.e., nodes, and E is the set of connections between nodes, i.e., edges; for the environment E, A is the set of behaviors, B is the set of feedbacks, and C is the reward expectation matrix;

[0013] Step S1.2: Update of network snapshot G i →G i+1 That is, the change of random environment E i →E i+1 .

[0014] Preferably, step S2 comprises the following steps:

[0015] Step S2.1: At time t, the learning automaton selects behavior a(t) according to its behavior selection strategy;

[0016] Step S2.2: Take S∪{s(t)} as the initial node in the network G according to the information propagation model Perform an information propagation to obtain the number of activated nodes |V active (t)|;

[0017] Step S2.3: Calculate the environmental feedback b(t) as follows, where |V| is the total number of nodes in the network and E[S} is the expected value of the influence of the node set S in the network:

[0018]

[0019] Step S2.4: the learning automaton updates its internal state according to its state update strategy;

[0020] Step S2.5: Repeat the process of steps S2.1 to S2.4 until |S|=k.

[0021] Preferably, step S3 comprises the following steps:

[0022] Step S3.1: Let i = 1, 2, ..., k, and the node set {s 1 ,s 2 ,…,s i} and network G 1 In the information diffusion model 1 interaction is performed under the condition of using the number of activated nodes and E[{s 1 ,s 2 ,…,s i}] Calculate the marginal benefit of the normalized propagation range and record it in sequence N i，1 , when the sequence length satisfies the following When, only the latest |N is kept i，0 | feedback values, where N i，0 For the node set {s 1 ,s 2 ,…,s i}In the network G 0 The sequence of feedback obtained during the interaction;

[0023] Step S3.2: Calculate the boundary value ∈ using the adaptive sliding window ASWIN algorithm cut (i), given the confidence parameter δ and the harmonic mean n h :

[0024]

[0025]

[0026] Step S3.3: If sequence N i，0 and N i，1 The mean difference is greater than ∈ cut (i) If the network environment changes, it starts to reconnect to the new network G. t+1 Find a new set of seed nodes.

[0027] Preferably, step S4 comprises the following steps:

[0028] Step S4.1: Let i = 1, 2, ..., k, the node set {s 1,…,s i} and the new network G 1 In the same information diffusion model Next interaction μ|N i，0 | times, where μ is the proportional coefficient, using the number of activated nodes and E[{s 1 ,s 2 ,…,s i}] Calculate the marginal benefit of the normalized propagation range and record it in the feedback sequence N i，1 ;

[0029] Step S4.2: Given the new and old sequences N i，0 and N i，1 , using the adaptive sliding window ASWIN algorithm to calculate the boundary value ∈ cut (i);

[0030] Step S4.3: In the new and old networks G t →G t+1 The historical information of the learning automaton is transferred between (1-ω i ) to retain each element in the historical feedback sequence, where Δ i For sequence N i，0 and N i，1 The mean difference between

[0031] Step S4.4: Using the feedback sequence after migration, the learning automaton LA algorithm is used to learn the i-th seed node and converge to the node And add it to the new seed node set

[0032] Step S4.5: If The historical information migration ends, and the subsequent rounds of seed node learning continue until |S * |=k; otherwise repeat the above process until |S * |=k.

[0033] The present invention also provides a system for maximizing the influence of a dynamic network structure based on a learning automaton, the system comprising the following modules:

[0034] Module M1: Dynamic network structure {G 0 , G 1 , ...} is mapped into a short-term stable non-stationary environment consisting of a set of stable environments {E 0 , E 1 ,…}, initialize the seed node set and current influence;

[0035] Module M2: Based on the learning automaton LA algorithm, in the environment E 0Explore and get the initial network G 0 The set of seed nodes on ;

[0036] Module M3: Based on the adaptive sliding window algorithm, two consecutive network topologies G 0 →G 1 Perform change detection;

[0037] Module M4: In old and new networks G 0 →G 1 The historical information of the interaction between the learning automaton and the environment is transferred to obtain the seed node set S in the new network * .

[0038] Preferably, the module M1 includes the following modules:

[0039] Module M1.1: Network G i With stable environment i One-to-one correspondence, and satisfying the above mapping relationship between G(V, E) and E(A, B, C), where for the network G, V is the set of network participants, i.e., nodes, and E is the set of connections between nodes, i.e., edges; for the environment E, A is the set of behaviors, B is the set of feedbacks, and C is the reward expectation matrix;

[0040] Module M1.2: Network snapshot update G i →G i+1 That is, the change of random environment E i →E i+1 .

[0041] Preferably, the module M2 includes the following modules:

[0042] Module M2.1: At time t, the learning automaton selects behavior a(t) according to its behavior selection strategy;

[0043] Module M2.2: Take S∪{s(t)} as the initial node in the network G according to the information propagation model Perform an information propagation to obtain the number of activated nodes |V active (t)|;

[0044] Module M2.3: Calculate the environmental feedback b(t) as follows, where |V| is the total number of nodes in the network and E[S} is the expected value of the influence of the node set S in the network:

[0045]

[0046] Module M2.4: The learning automaton updates its internal state according to its state update strategy;

[0047] Module M2.5: Repeat the process of module M2.1-module M2.4 until |S|=k.

[0048] Preferably, the module M3 includes the following modules:

[0049] Module M3.1: Let i = 1, 2, ..., k, with the node set {s 1 ,s 2 ,…,s i} and network G 1 In the information diffusion model 1 interaction is performed under the condition of using the number of activated nodes and E[{s 1 ,s 2 ,…,s i}] Calculate the marginal benefit of the normalized propagation range and record it in sequence N i，1 , when the sequence length satisfies the following When, only the latest |N is kept i，0 | feedback values, where N i，0 For the node set {s 1 ,s 2 ,…,s i}In the network G 0 The sequence of feedback obtained during the interaction;

[0050] Module M3.2: Calculate boundary value ∈ using adaptive sliding window ASWIN algorithm cut (i), given the confidence parameter δ and the harmonic mean n h :

[0051]

[0052]

[0053] Module M3.3: If the sequence N i，0 and N i，1 The mean difference is greater than ∈ cut (i) If the network environment changes, it starts to reconnect to the new network G. t+1 Find a new set of seed nodes.

[0054] Preferably, the module M4 includes the following modules:

[0055] Module M4.1: Let i = 1, 2, ..., k, the node set {s 1 ,…,s i} and the new network G 1 In the same information diffusion model Next interaction μ|N i，0 | times, where μ is the proportional coefficient, using the number of activated nodes and E[{s 1 ,s2 ,…,s i}] Calculate the marginal benefit of the normalized propagation range and record it in the feedback sequence N i，1 ;

[0056] Module M4.2: Given the new and old sequences N i，0 and N i，1 , using the adaptive sliding window ASWIN algorithm to calculate the boundary value ∈ cut (i);

[0057] Module M4.3: In old and new networks G t →G t+1 The historical information of the learning automaton is transferred between (1-ω i ) to retain each element in the historical feedback sequence, where Δ i For sequence N i，0 and N i，1 The mean difference between

[0058] Module M4.4: Using the feedback sequence after migration, the learning automaton LA algorithm is used to learn the i-th seed node and converge to the node And add it to the new seed node set

[0059] Module M4.5: If The historical information migration ends, and the subsequent rounds of seed node learning continue until |S * |=k; otherwise repeat the above process until |S * |=k.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] 1. The present invention transforms the influence maximization problem into an optimization learning problem by learning an automaton algorithm without using network structure information and without limiting a specific propagation model, thereby achieving rapid selection of an influential seed node set;

[0062] 2. The present invention uses an adaptive sliding window algorithm to analyze the feedback sequence obtained by the interaction between the learning automaton and the new and old networks, detect the degree of network change, and adaptively migrate the historical information learned by the learning automaton in the old network to the new network, thereby quickly updating the set of nodes with the greatest influence in the dynamic network. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0064] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0065] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0066] Embodiment 1:

[0067] According to the present invention, a method for maximizing the influence of a dynamic network structure based on a learning automaton is provided, and the method comprises the following steps:

[0068] Step S1: transform the dynamic network structure {G 0 , G 1 , ...} is mapped into a short-term stable non-stationary environment consisting of a set of stable environments {E 0 , E 1 ,…}, initialize the seed node set and current influence;

[0069] Step S1.1: Network G i With stable environment i One-to-one correspondence, and satisfying the above mapping relationship between G(V, E) and E(A, B, C), where for the network G, V is the set of network participants, i.e., nodes, and E is the set of connections between nodes, i.e., edges; for the environment E, A is the set of behaviors, B is the set of feedbacks, and C is the reward expectation matrix;

[0070] Step S1.2: Update of network snapshot G i →G i+1 That is, the change of random environment E i →E i+1 .

[0071] Step S2: Based on the learning automaton LA algorithm, in the environment E 0 Explore and get the initial network G 0 The set of seed nodes on ;

[0072] Step S2.1: At time t, the learning automaton selects behavior a(t) according to its behavior selection strategy;

[0073] Step S2.2: Take S∪{s(t)} as the initial node in the network G according to the information propagation model Perform an information propagation to obtain the number of activated nodes |V active (t)|;

[0074] Step S2.3: Calculate the environmental feedback b(t) as follows, where |V| is the total number of nodes in the network and E[S} is the expected value of the influence of the node set S in the network:

[0075]

[0076] Step S2.4: the learning automaton updates its internal state according to its state update strategy;

[0077] Step S2.5: Repeat the process of steps S2.1 to S2.4 until |S|=k.

[0078] Step S3: Based on the adaptive sliding window algorithm, two consecutive network topology structures G 0 →G 1 Perform change detection;

[0079] Step S3.1: Let i = 1, 2, ..., k, and the node set {s 1 ,s 2 ,…,s i} and network G 1 In the information diffusion model 1 interaction is performed under the condition of using the number of activated nodes and E[{s 1 ,s 2 ,…,s i}] Calculate the marginal benefit of the normalized propagation range and record it in sequence N i，1 , when the sequence length satisfies the following When, only the latest |N is kept i，0 | feedback values, where N i，0 For the node set {s 1 ,s 2 ,…,s i}In the network G 0 The sequence of feedback obtained during the interaction;

[0080] Step S3.2: Calculate the boundary value ∈ using the adaptive sliding window ASWIN algorithm cut (i), given the confidence parameter δ and the harmonic mean n h :

[0081]

[0082]

[0083] Step S3.3: If sequence N i，0 and N i，1 The mean difference is greater than ∈ cut(i) If the network environment changes, it starts to reconnect to the new network G. t+1 Find a new set of seed nodes.

[0084] Step S4: In the new and old networks G 0 →G 1 The historical information of the interaction between the learning automaton and the environment is transferred to obtain the seed node set S in the new network * ;

[0085] Step S4.1: Let i = 1, 2, ..., k, the node set {s 1 ,…,s i} and the new network G 1 In the same information diffusion model Next interaction μ|N i，0 | times, where μ is the proportional coefficient, using the number of activated nodes and E[{s 1 ,s 2 ,…,s i}] Calculate the marginal benefit of the normalized propagation range and record it in the feedback sequence N i，1 ;

[0086] Step S4.2: Given the new and old sequences N i，0 and N i，1 , using the adaptive sliding window ASWIN algorithm to calculate the boundary value ∈ cut (i);

[0087] Step S4.3: In the new and old networks G t →G t+1 The historical information of the learning automaton is transferred between (1-ω i ) to retain each element in the historical feedback sequence, where Δ i For sequence N i，0 and N i，1 The mean difference between

[0088] Step S4.4: Using the feedback sequence after migration, the learning automaton LA algorithm is used to learn the i-th seed node and converge to the node And add it to the new seed node set

[0089] Step S4.5: If The historical information migration ends, and the subsequent rounds of seed node learning continue until |S * |=k; otherwise repeat the above process until |S * |=k.

[0090] Embodiment 2:

[0091] Embodiment 2 is a preferred example of Embodiment 1, and is used to illustrate the present invention in more detail.

[0092] The present invention also provides a dynamic network structure based on learning automaton influence maximization system, the system includes the following modules:

[0093] Module M1: Dynamic network structure {G 0 , G 1 , ...} is mapped into a short-term stable non-stationary environment consisting of a set of stable environments {E 0 , E 1 ,…}, initialize the seed node set and current influence;

[0094] Module M1.1: Network G i With stable environment i One-to-one correspondence, and satisfying the above mapping relationship between G(V, E) and E(A, B, C), where for the network G, V is the set of network participants, i.e., nodes, and E is the set of connections between nodes, i.e., edges; for the environment E, A is the set of behaviors, B is the set of feedbacks, and C is the reward expectation matrix;

[0095] Module M1.2: Network snapshot update G i →G i+1 That is, the change of random environment E i →E i+1 .

[0096] Module M2: Based on the learning automaton LA algorithm, in the environment E 0 Explore and get the initial network G 0 The set of seed nodes on ;

[0097] Module M2.1: At time t, the learning automaton selects behavior a(t) according to its behavior selection strategy;

[0098] Module M2.2: Take S∪{s(t)} as the initial node in the network G according to the information propagation model Perform an information propagation to obtain the number of activated nodes |V active (t)|;

[0099] Module M2.3: Calculate the environmental feedback b(t) as follows, where |V| is the total number of nodes in the network and E[S} is the expected value of the influence of the node set S in the network:

[0100]

[0101] Module M2.4: The learning automaton updates its internal state according to its state update strategy;

[0102] Module M2.5: Repeat the process of module M2.1-module M2.4 until |S|=k.

[0103] Module M3: Based on the adaptive sliding window algorithm, two consecutive network topologies G 0 →G 1 Perform change detection;

[0104] Module M3.1: Let i = 1, 2, ..., k, with the node set {s 1 ,s 2 ,…,s i} and network G 1 In the information diffusion model 1 interaction is performed under the condition of using the number of activated nodes and E[{s 1 ,s 2 ,…,s i}] Calculate the marginal benefit of the normalized propagation range and record it in sequence N i，1 , when the sequence length satisfies the following When, only the latest |N is kept i，0 | feedback values, where N i，0 For the node set {s 1 ,s 2 ,…,s i}In the network G 0 The sequence of feedback obtained during the interaction;

[0105] Module M3.2: Calculate boundary value ∈ using adaptive sliding window ASWIN algorithm cut (i), given the confidence parameter δ and the harmonic mean n h :

[0106]

[0107]

[0108] Module M3.3: If the sequence N i，0 and N i，1 The mean difference is greater than ∈ cut (i) If the network environment changes, it starts to reconnect to the new network G. t+1 Find a new set of seed nodes.

[0109] Module M4: In old and new networks G 0 →G 1 The historical information of the interaction between the learning automaton and the environment is transferred to obtain the seed node set S in the new network * ;

[0110] Module M4.1: Let i = 1, 2, ..., k, the node set {s 1,…,s i} and the new network G 1 In the same information diffusion model Next interaction μ|N i，0 | times, where μ is the proportional coefficient, using the number of activated nodes and E[{s 1 ,s 2 ,…,s i}] Calculate the marginal benefit of the normalized propagation range and record it in the feedback sequence N i，1 ;

[0111] Module M4.2: Given the new and old sequences N i，0 and N i，1 , using the adaptive sliding window ASWIN algorithm to calculate the boundary value ∈ cut (i);

[0112] Module M4.3: In old and new networks G t →G t+1 The historical information of the learning automaton is transferred between (1-ω i ) to retain each element in the historical feedback sequence, where Δ i For sequence N i，0 and N i，1 The mean difference between

[0113] Module M4.4: Using the feedback sequence after migration, the learning automaton LA algorithm is used to learn the i-th seed node and converge to the node And add it to the new seed node set

[0114] Module M4.5: If The historical information migration ends, and the subsequent rounds of seed node learning continue until |S * |=k; otherwise repeat the above process until |S * |=k.

[0115] Embodiment 3:

[0116] Embodiment 3 is a preferred example of Embodiment 1, and is used to illustrate the present invention in more detail.

[0117] The present invention proposes a method for maximizing influence based on learning automata in a dynamic network structure. The method can use the learning automata algorithm to quickly find a set of seed nodes with the maximum influence benefit in a huge network node, and when the network topology changes, use an adaptive sliding window algorithm to determine the degree of change of the influence of the seed nodes and update the seed nodes in a timely manner.

[0118] The present invention provides a method for maximizing the influence of a dynamic network structure based on a learning automaton, comprising the following steps:

[0119] Step 1: Transform the dynamic network structure {G 0 , G 1 , ...} is mapped into a short-term stable non-stationary environment consisting of a set of stable environments {E 0 , E 1 ,…}, initialize the seed node set and current influence;

[0120] Step 2: Based on the learning automaton LA algorithm, in the environment E 0 Explore and get the initial network G 0 The set of seed nodes on ;

[0121] Step 3: Based on the adaptive sliding window algorithm, two consecutive network topologies G 0 →G 1 Perform change detection;

[0122] Step 4: Connect the old and new networks (G 0 →G 1 ) to transfer the historical information of the interaction between the learning automaton and the environment, and obtain the seed node set S in the new network * .

[0123] The step 1 comprises the following steps:

[0124] Step 1.1: Network G i With stable environment i One-to-one correspondence, and satisfying the above mapping relationship between G(V, E) and E(A, B, C), where for the network G, V is the set of network participants, i.e., nodes, and E is the set of connections between nodes, i.e., edges; for the environment E, A is the set of behaviors, B is the set of feedbacks, and C is the reward expectation matrix;

[0125] Step 1.2: Update of network snapshot G i →G i+1 That is, the change of random environment E i →E i+1 .

[0126] The step 2 comprises the following steps:

[0127] Step 2.1: At time t, the learning automaton selects behavior a(t) according to its behavior selection strategy;

[0128] Step 2.2: Take S∪{s(t)} as the initial node and follow the information propagation model in the network G Perform an information propagation to obtain the number of activated nodes |V active (t)|;

[0129] Step 2.3: Calculate the environmental feedback b(t) as follows, where |V| is the total number of nodes in the network and E[S} is the expected value of the influence of the node set S in the network:

[0130]

[0131] Step 2.4: The learning automaton updates its internal state according to its state update strategy;

[0132] Step 2.5: Repeat steps 2.1-2.4 until |S|=k.

[0133] The step 3 comprises the following steps:

[0134] Step 3.1: Let i = 1, 2, ..., k, and the node set {s 1 ,s 2 ,…,s i} and network G 1 In the information diffusion model 1 interaction is performed under the condition of using the number of activated nodes and E[{s 1 ,s 2 ,…,s i}] Calculate the marginal benefit of the normalized propagation range and record it in sequence N i，1 , when the sequence length satisfies the following When, only the latest |N is kept i，0 | feedback values, where N i，0 For the node set {s 1 ,s 2 ,…,s i}In the network G 0 The sequence of feedback obtained during the interaction;

[0135] Step 3.2: Calculate the boundary value ∈ using the adaptive sliding window ASWIN algorithm cut (i), given the confidence parameter δ and the harmonic mean n h :

[0136]

[0137]

[0138] Step 3.3: If the sequence N i，0 and N i，1 The mean difference is greater than ∈ cut (i) If the network environment changes, it starts to reconnect to the new network G. t+1 Find a new set of seed nodes.

[0139] The step 4 comprises the following steps:

[0140] Step 4.1: Let i = 1, 2, ..., k, the node set {s 1 ,…,s i} and the new network G 1 In the same information diffusion model Next interaction μ|N i，0 | times, where μ is the proportional coefficient, using the number of activated nodes and E[{s 1 ,s 2 ,…,s i}] Calculate the marginal benefit of the normalized propagation range and record it in the feedback sequence N i，1 ;

[0141] Step 4.2: Given the new and old sequences N i，0 and N i，1 , using the adaptive sliding window ASWIN algorithm to calculate the boundary value ∈ cut (i);

[0142] Step 4.3: In the old and new networks (G t →G t+1 ) to transfer the historical information of the learning automaton, with (1-ω i ) to retain each element in the historical feedback sequence, where Δ i For sequence N i，0 and N i，1 The mean difference between

[0143] Step 4.4: Using the feedback sequence after migration, the learning automaton LA algorithm is used to learn the i-th seed node and converge to the node And add it to the new seed node set

[0144] Step 4.5: If The historical information migration ends, and the subsequent rounds of seed node learning continue until |S * |=k; otherwise repeat the above process until |S * |=k.

[0145] The present invention transforms the influence maximization problem into an optimization learning problem by learning an automaton algorithm without utilizing network structure information and without limiting a specific propagation model, thereby achieving rapid selection of an influential seed node set.

[0146] The present invention analyzes the feedback sequence obtained by the interaction between the learning automaton and the new and old networks through an adaptive sliding window algorithm, detects the degree of network change and adaptively migrates the historical information learned by the learning automaton in the old network to the new network, thereby quickly updating the set of nodes with the greatest influence in the dynamic network.

[0147] Those skilled in the art may understand this embodiment as a more specific description of Embodiment 1 and Embodiment 2.

[0148] Those skilled in the art know that, in addition to realizing the system and its various devices, modules, and units provided by the present invention in a purely computer-readable program code, it is entirely possible to realize the same functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for realizing various functions can also be regarded as structures within the hardware component; the devices, modules, and units for realizing various functions can also be regarded as both software modules for realizing the method and structures within the hardware component.

[0149] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A method for maximizing the influence of a dynamic network structure based on a learning automaton, characterized in that: The method comprises the following steps: Step S1: Map the dynamic network structure {G0, G1, ...} into a short-term stable non-stationary environment {E0, E1, ...} consisting of a set of stable environments, and initialize the seed node set and current influence; Step S2: Based on the learning automaton LA algorithm, explore in the environment E0 to obtain the seed node set on the initial network G0; Step S3: Based on the adaptive sliding window algorithm, change detection is performed on two consecutive network topology structures G0→G1; Step S4: Migrate the historical information of the interaction between the learning automaton and the environment between the new and old networks G0→G1 to obtain the seed node set S* in the new network; The step S1 comprises the following steps: Step S1.1: Network G i With stable environment i One-to-one correspondence, and satisfies the mapping relationship between G(V, E) and E(A, B, C), where for the network G, V is the set of network participants, i.e., nodes, and E is the set of connections between nodes, i.e., edges; for the environment E, A is the set of behaviors, B is the set of feedbacks, and C is the reward expectation matrix; Step S1.2: Update of network snapshot G i →G i+1 That is, the change of random environment E i →E i+1 ; The step S2 comprises the following steps: Step S2.1: At time t, the learning automaton selects behavior a(t) according to its behavior selection strategy; Step S2.2: Take S ∪{s(t)} as the initial node and follow the information propagation model in the network G Perform an information propagation to obtain the number of activated nodes |V active (t)|; Step S2.3: Calculate the environmental feedback b(t) as follows, where |V| is the total number of nodes in the network and E[S} is the expected value of the influence of the node set S in the network: Step S2.4: the learning automaton updates its internal state according to its state update strategy; Step S2.5: Repeat the process of steps S2.1 to S2.4 until |S|=k; The step S3 comprises the following steps: Step S3.1: Let i = 1, 2, ..., k, and the node set {s1, s2, ..., s i } and network G1 in the information propagation model 1 interaction is performed under the same condition, using the number of activated nodes and E[{s1, s2, …, s i }] Calculate the marginal benefit of the normalized propagation range and record it in sequence N i，1 , when the sequence length satisfies the following When, only the latest |N is kept i，0 | feedback values, where N i，0 is a node set {s1, s2, ..., s i }The feedback sequence obtained when interacting with network G0; Step S3.2: Calculate the boundary value ∈ using the adaptive sliding window ASWIN algorithm cut (i), given the confidence parameter δ and the harmonic mean n h : Step S3.3: If sequence N i，0 and N i，1 The mean difference is greater than ∈ cut (i) If the network environment changes, it starts to reconnect to the new network G. t+1 Find a new set of seed nodes on The step S4 comprises the following steps: Step S4.1: Let i = 1, 2, ..., k, the node set {s1, ..., s i } and the new network G1 in the same information propagation model Next interaction μ|N i，0 | times, where μ is the proportional coefficient, using the number of activated nodes and E[{s1, s2, …, s i }] Calculate the marginal benefit of the normalized propagation range and record it in the feedback sequence N i，1 ; Step S4.2: Given the new and old sequences N i，0 and N i，1 , using the adaptive sliding window ASWIN algorithm to calculate the boundary value ∈ cut (i); Step S4.3: In the new and old networks G t →G t+1 The historical information of the learning automaton is transferred between (1-ω i ) to retain each element in the historical feedback sequence, where △ i For sequence N i，0 and N i，1 The mean difference between Step S4.4: Using the feedback sequence after migration, the learning automaton LA algorithm is used to learn the i-th seed node and converge to the node And add it to the new seed node set Step S4.5: If The historical information migration ends and the learning of subsequent rounds of seed nodes continues until |S*|=k; otherwise, the above process is repeated until |S*|=k.

2. A dynamic network structure based on learning automata influence maximization system, characterized in that: The system includes the following modules: Module M1: Map the dynamic network structure {G0, G1, ...} into a short-term stable non-stationary environment {E0, E1, ...} consisting of a set of stable environments, initialize the seed node set and the current influence; Module M2: Based on the learning automaton LA algorithm, explore in the environment E0 to obtain the set of seed nodes on the initial network G0; Module M3: Based on the adaptive sliding window algorithm, change detection is performed on two consecutive network topology structures G0→G1; Module M4: Migrate the historical information of the interaction between the learning automaton and the environment between the new and old networks G0→G1 to obtain the seed node set S* in the new network; The module M1 includes the following modules: Module M1.1: Network G i With stable environment i One-to-one correspondence, and satisfies the mapping relationship between G(V, E) and E(A, B, C), where for the network G, V is the set of network participants, i.e., nodes, and E is the set of connections between nodes, i.e., edges; for the environment E, A is the set of behaviors, B is the set of feedbacks, and C is the reward expectation matrix; Module M1.2: Network snapshot update G i →G i+1 That is, the change of random environment E i →E i+1 ; The module M2 includes the following modules: Module M2.1: At time t, the learning automaton selects behavior a(t) according to its behavior selection strategy; Module M2.2: Take S ∪{s(t)} as the initial node in the network G according to the information propagation model Perform an information propagation to obtain the number of activated nodes |V active (t)|; Module M2.3: Calculate the environmental feedback b(t) as follows, where |V| is the total number of nodes in the network and E[S} is the expected value of the influence of the node set S in the network: Module M2.4: The learning automaton updates its internal state according to its state update strategy; Module M2.5: Repeat the process of module M2.1-module M2.4 until |S|=k; The module M3 includes the following modules: Module M3.1: Let i = 1, 2, ..., k, with the node set {s1, s2, ..., s i } and network G1 in the information propagation model 1 interaction is performed under the same condition, using the number of activated nodes and E[{s1, s2, …, s i }] Calculate the marginal benefit of the normalized propagation range and record it in sequence N i，1 , when the sequence length satisfies the following When, only the latest |N is kept i，0 | feedback values, where N i，0 is a node set {s1, s2, ..., s i }The feedback sequence obtained when interacting with network G0; Module M3.2: Calculate boundary value ∈ using adaptive sliding window ASWIN algorithm cut (i), given the confidence parameter δ and the harmonic mean n h : Module M3.3: If the sequence N i，0 and N i，1 The mean difference is greater than ∈ cut (i) If the network environment changes, it starts to reconnect to the new network G. t+1 Find a new set of seed nodes on The module M4 includes the following modules: Module M4.1: Let i = 1, 2, ..., k, the node set {s1, ..., s i } and the new network G1 in the same information propagation model Next interaction μ|N i，0 | times, where μ is the proportional coefficient, using the number of activated nodes and E[{s1, s2, …, s i }] Calculate the marginal benefit of the normalized propagation range and record it in the feedback sequence N i，1 ; Module M4.2: Given the new and old sequences N i，0 and N i，1 , using the adaptive sliding window ASWIN algorithm to calculate the boundary value ∈ cut (i); Module M4.3: In old and new networks G t →G t+1 The historical information of the learning automaton is transferred between (1-ω i ) to retain each element in the historical feedback sequence, where △ i For sequence N i，0 and N i，1 The mean difference between Module M4.4: Using the feedback sequence after migration, the learning automaton LA algorithm is used to learn the i-th seed node and converge to the node And add it to the new seed node set Module M4.5: If The historical information migration ends and the learning of subsequent rounds of seed nodes continues until |S*|=k; otherwise, the above process is repeated until |S*|=k.

Images