Optimal cognitive sub-graph query method based on time sequence wave neural network

Abstract

The invention provides a time sequence wave neural network (TSWNN), which is used for solving an optimal cognitive sub-graph query problem (MCSQR), and is characterized in that the optimal cognitive sub-graph query problem (MCSQR) is solved by using a time sequence wave neural network (TSWNN), so that the optimal cognitive sub-graph query problem (MCSQR) can be solved by using the time sequence wave neural network (TSWNN), and the optimal cognitive sub-graph query problem (MCSQR) can be solved by using the time sequence wave neural network (TSWNN). The MCSQP aims to query a path from a starting point to an end point in a specific time in a network, optimize node cognition degree accumulation and realize query of an optimal cognition subgraph. The proposed TSWNN is a novel neural network model, has time dependence and transmission time limitation, and is used for simulating and analyzing the process of information spreading and node activation in the network. All neurons of the method are in parallel calculation, and the model does not need any training. Each neuron in the TSWNN is composed of six parts: an input unit, a wave receiver, a neuron state memory, a wave generator, a wave transmitter and an output unit. The optimal cognitive sub-graph needs to find a network from a starting point to an end point, a network where time sequence waves arrive within given time is obtained, then the final cognitive degree is calculated, and the neural network deals with the optimal cognitive sub-graph problem by relying on the time sequence waves. Experimental results show that the method provided by the invention can solve the globally optimal solution of the problem.

Description

Technical Field

[0001] The present invention relates to the technical field of subgraph query, and in particular to an optimal cognitive subgraph query method based on a temporal wave neural network. Background Art

[0002] Subgraph query is a problem in the field of graph theory. It is widely used in social networks, protein networks, knowledge graph construction, and recommendation systems. The optimal cognitive subgraph query problem is to find multiple propagation paths from the starting point to the end point in the process of information propagation in a static social network for a given time period to merge and form a subgraph with the best overall cognition. Its goal is to find all paths from the starting point to the end point within a limited time and merge them into a network to maximize the cognition of the end point. The Ullmann algorithm is one of the most widely used subgraph query algorithms. It uses a depth-first search strategy to enumerate subgraphs that match the query graph q in the data graph G. Researchers proposed the VF2 algorithm, which improved the existing algorithm and introduced a new heuristic pruning strategy based on the backtracking search method, which greatly improved the processing performance of the graph. After that, many improved algorithms were used to prune by using the path and structural information around the vertex, thereby strengthening the constraints and reducing the number of backtracking times in the subgraph matching process. However, these methods are widely used in static networks. In real networks, the dynamic changes of data and the evolution of user behavior in social networks make previous algorithms face challenges in dealing with the optimal cognitive subgraph query problem. There is a significant difference between the optimal cognitive subgraph query and the traditional subgraph query. The traditional subgraph query aims to find isomorphic subgraphs of a given query graph in a static network, while the optimal cognitive subgraph query focuses not only on the topological structure of the subgraph, but also on the propagation path of information between nodes and the nodes' awareness of the information. However, since the current subgraph query algorithm focuses on finding subgraphs that are isomorphic to the given query graph in a static network, it fails to consider the propagation path of information between nodes and the nodes' awareness of the information, and there are still certain defects in practical applications. Summary of the invention

[0003] The present invention provides an innovative optimal cognitive subgraph query method for a time-series wave neural network, which solves the shortcomings of traditional subgraph query methods in terms of message transmission by time-series waves. Based on the principles of neural networks and existing research results, the present invention conducts in-depth research on the time-series wave propagation mechanism, neuronal activity patterns, and neural network topological structures, and successfully constructs a complex time-series wave neural network model that can accurately simulate the propagation and interaction of time-series waves between neurons. The present invention designs a set of efficient subgraph query algorithms, combines time-series wave feature extraction and optimization strategy application, and realizes fast and accurate query of the optimal cognitive subgraph in the neural network. Specifically, the implementation of this method can be divided into the following key steps:

[0004] Step 1: Based on the modeling of node states and behaviors, a time-series wave neuron model is constructed using mathematical modeling techniques to describe the dynamic process of information transmission between neurons, providing a basis for the modeling of time-series wave neural networks.

[0005] Step 2: Based on the neural network theory, the Temporal Sequential Wave Neural Network (TSWNN) model is constructed to describe the propagation and interaction process of the temporal wave in the network, initialize various parameters of TSWNN, set the network's running time limit T, and ensure that the model can fully demonstrate its dynamic behavior within the specified time;

[0006] Step 3: Initialize all neurons in the time-series wave neural network based on the initial state of the node, activate the initial neurons, set the initial time t = 0, provide a unified starting point for the network operation, and ensure the smooth progress of the information transmission process;

[0007] Step 4: Based on the neural network theory and mathematical model, combined with the dynamic changes between neurons, update the information of each neuron in TSWNN, carefully adjust the activity state of neurons in the network, effectively activate neurons and promote the generation and transmission of timing waves, ensure the normal generation and transmission of timing waves in the network, and realize efficient flow and processing of information;

[0008] Step 5: Combine the propagation rules of time-series waves and the characteristics of information interaction in the network to deeply mine and analyze the network structure and output the optimal cognitive subgraph.

[0009] Step 1: Based on the state and behavior modeling of the nodes, a time-series wave neuron model is constructed. The specific steps are as follows:

[0010] Step 1.1: Design an input device, which is used to receive the timing waves from its predecessor neurons;

[0011] Step 1.2: Design a wave receiver. The function of the wave receiver is to receive the wave information and decode it, including: i , w ij ,y ij and t j ;

[0012] Step 1.3, neuron state memory, determines whether the neuron is activated, stores the neuron state, and stores the neuron recognition degree according to the information of the received wave, which is represented by y j and c j express;

[0013] Step 1.4: Design a wave generator. When a neuron is activated, it generates a timing wave. The generated wave is transmitted to the unactivated successor neurons and consists of two parts:jk ,y jk ;

[0014] Step 1.5: Design a wave transmitter. The transmitter sends waves to unactivated successor neurons and encodes the waves. When the successor neuron is not activated, the wave is sent; if the successor node is activated, the wave is not sent. This can be regarded as the reverse process of the wave transmitter. It contains two parts: jk ,y jk ;

[0015] Step 1.6: Design the wave output device. The wave output device is mainly used to send the new wave to its corresponding successor neuron.

[0016] Step 2: Construct a Temporal Sequential Wave Neural Network (TSWNN) and initialize various parameters of TSWNN. The specific steps are as follows:

[0017] Step 2.1: Construct a time-series wave neural network model based on the network topology of the problem, where each node corresponds to a neuron in TSWNN, each arc corresponds to an activity in TSWNN, and a random transmission time and cognition degree are generated for each arc (the edge connecting two neurons);

[0018] Step 2.2: Set the initial neuron S and the end neuron D;

[0019] Step 2.3: Set the departure time of the initial neuron to t, where t is the earliest departure time from the initial neuron;

[0020] Step 2.4: Set the neural network time limit T;

[0021] Step 2.5: Set the time step Δt of the neural network iteration.

[0022] Step 3: Initialize all neurons in the timing wave neural network and activate the initial neurons. The specific steps are as follows:

[0023] Step 3.1: For the initial neuron S, let

[0024] Step 3.2: For non-initial neurons, let

[0025] Step 3.3: Set the initial time t=0.

[0026] Step 4: Update the state information of each neuron on TSWNN, activate the neuron and generate waves. The specific steps are as follows:

[0027] Step 4.1: Determine whether t is less than T. If t<T, perform the following operations; otherwise, perform step 5.

[0028] Step 4.2: For each unactivated neuron i∈V on TSWNN, after receiving the first wave wni, the neuron state yi, neuron transmission time ti and cognitive degree ci, node path Pi are modified, wave wij is generated and sent;

[0029] w ni = l ni ,y ni =c ni .

[0030]

[0031] t j =t i +l ij

[0032] c j =c i *y ij +c j

[0033] P j =<P i ,v j >

[0034] Step 4.3: For the activated neuron j∈V on TSWNN, generate wave wjk to determine whether the successor node is activated. If not, send a wave to change the activation time Y. m ; If the node receives the wave, modify the node awareness cj, transmission time tj, and node path Pj;

[0035] Y m = l jm

[0036] w jk = l jk

[0037]

[0038] c j =c i *y ij +c j

[0039] P ij =<P i , v j >

[0040] Step 4.4, superimpose the timer t, t = t + Δt;

[0041] Step 4.5: Determine again whether t is less than T. If t<T, execute steps 4.2-4.4; otherwise, execute step 5 and the search ends.

[0042] Step 5: Calculate the endpoint awareness and merge the endpoint paths into a network. The specific steps are as follows:

[0043] Step 5.1: At t = T (neural network update stops), output the terminal path network, which is a set of terminal path neuron pairs;

[0044] Step 5.2: Output the awareness cd of the path.

[0045] Advantages and beneficial effects of the present invention

[0046] The present invention proposes an innovative method of time-series wave neural network, which is a new type of neural network with time-series neurons. It uses parallel computing to improve the computing rate and can efficiently complete the analysis and query of complex network data. Unlike traditional neural networks, it does not require any training and can be directly applied to practical scenarios. The social network is regarded as a time-series wave neural network topology structure, and each node is regarded as a time-series wave neuron, which can accurately capture the information propagation path between nodes and the node's awareness of information, and improve the accuracy and reliability of the minimum cognitive subgraph query. The present invention has significant advantages in solving the optimal cognitive subgraph query problem. Compared with existing methods, this method can quickly solve the optimal solution to the problem, and has important application significance in the fields of social networks, protein networks, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a workflow diagram of solving the optimal subgraph query problem of the present invention;

[0048] Figure 2 It is a network structure diagram;

[0049] Figure 3 It is a neuron structure diagram of the timing wave neural network;

[0050] Figure 4 This is a structural diagram of a time series wave neural network. DETAILED DESCRIPTION

[0051] The present invention is described in detail below with reference to the accompanying drawings and examples.

[0052] An optimal cognitive subgraph query method based on temporal wave neural network. The overall implementation process of solving the optimal cognitive subgraph is as follows: Figure 1 As shown; for the five-node, seven-arc network, the optimal cognitive subgraph is solved. The initial node is A, the destination node is D, and the network structure is as follows Figure 2As shown, the time-series wave parameters of each edge in the network are shown in Table 1. The specific implementation plan includes the following content and steps:

[0053] Table 1 Time-series wave parameters of each edge in the network

[0054]

[0055]

[0056] Step 1: Construct time-dependent wave neurons, where each neuron consists of six parts: an input device, a wave receiver, a neuron state memory, a wave generator, a wave transmitter, and an output device. The structure of the neuron is as Figure 3 shown.

[0057] Step 2: Construct a time-series wave neural network and initialize the parameters of the TSWNN. The specific steps are as follows:

[0058] Step 2.1: Construct a time-series wave neural network model according to the network topology. This TSWNN model consists of five time-series wave neurons and seven edges to form a neural network model. At the same time, specify the initial neuron A, the target neuron D, and the limited time T. The structure of the TSWNN is as Figure 4 shown;

[0059] Step 2.2: Let t = 0, T = 10, and Δt = 1.

[0060] Step 3: Initialize all neurons in the TSWNN and activate the initial neuron A. For the time-series wave neuron A, let Pa = φ, ya = 1, ca = 1; for the time-series wave neuron B, let Pb = φ, yb = 0, cb = 0; for the time-series wave neuron C, let Pc = φ, yc = 0, cc = 0; for the time-series wave neuron D, let Pd = φ, yd = 0, cd = 0; for the time-series wave neuron E, let Pe = φ, ye = 0, ce = 0.

[0061] Step 4: Update the state information of each neuron on the TSWNN, activate the neurons and generate waves. The specific steps are as follows:

[0062] Step 4.1: Since t < T = 10 at t = 0, each neuron needs to be updated;

[0063] Step 4.2: For neuron A, set wab = 2, wad = 3, cab = 0.3, cad = 0.4, and Pa = {A}; for neuron B, set Y B= 2; For neuron D, set Y D = 3; Since t < Y B , t < Y D , so neurons B and D are not activated and cannot generate new waves, Δt + 1;

[0064] Step 4.3: Let t = 1. At this time, t < T = 10, so each neuron needs to be updated continuously;

[0065] Step 4.4: Since t < Y B , t < Y D , so neurons B and D are not activated and cannot generate new waves, Δt + 1;

[0066] Step 4.5: Let t = 2. At this time, t < T = 10, so each neuron needs to be updated continuously;

[0067] Step 4.6: For neuron B, set yb = 1, tb = 2, cbd = 0.2, cbe = 0.6, cbc = 0.7, wbd = 2, wbe = 3, wbc = 4, Pb = {<A, B>}, calculate cb = ca * cab; For neuron C, set Y C = tb + wbc, for neuron D, because Y D = 3 < tb + wbd, so Y D = 3; For neuron E, set Y E = tb + wbe; Since t < Y D , t < Y C , t < Y E So neurons D, C, and E are not activated and cannot generate new waves, Δt + 1;

[0068] Step 4.7: Let t = 3. At this time, t < T = 10, so each neuron needs to be updated continuously;

[0069] Step 4.8: For neuron D, set yd = 1, td = 3, cde = 0.8, wde = 6, Pd = {<A, D>}, calculate cd = ca * cad; For neuron E, because Y E = tb + wbd < td + wd, so Y E = tb + wbd; Since t < Y C , t < Y E So neurons C and E are not activated and cannot generate new waves, Δt + 1;

[0070] Step 4.9: Let t = 4. At this time, t < T = 10, so each neuron needs to be updated continuously;

[0071] Step 4.10: For neuron D, set td = 3, cde = 0.8, wde = 6, Pd = {<A, D>, <B, D>, <A, B>}, calculate cd = ca * ca + d; For neuron E, since Y E = tb + wb d < t d +, so Y E = tb + wbd; Since t < Y C , t < Y E Therefore, neurons C and E are not activated and cannot generate new waves, Δt + 1;

[0072] Step 4.11: Let t = 5. At this time, t < T = 10, so each neuron needs to be updated continuously;

[0073] Step 4.12: For neuron E, set ye = 1, te = 5, Pe = {<A, B>, <B, E>}, calculate ce = cb * cbe; Since t < Y C Therefore, neuron C is not activated and cannot generate new waves, Δt + 1;

[0074] Step 4.13: Let t = 6. At this time, t < T = 10, so the neurons need to be updated continuously;

[0075] Step 4.14: For neuron C, set yc = 1, tc = 6, cce = 0.5, wce = 5, Pd = {<A, D>}, calculate cc = cb * cbc + cc; There are still waves being transmitted, Δt + 1;

[0076] Step 4.15: Let t = 7. At this time, t < T = 10. Since the wave arrival time is greater than t, so Δt + 1;

[0077] Step 4.16: Let t = 8. At this time, t < T = 10. Since the wave arrival time is greater than t, so Δt + 1;

[0078] Step 4.17: Let t = 9. At this time, t < T = 10. Since the wave arrival time is equal to t, the neurons need to be updated continuously;

[0079] Step 4.18: For neuron E, Pe = {<A, D> <D, E>} Merge the path of E at t = 5 with the path at t = 9, and set Pe = {<A, B> <A, D> <B, E> <D, E>}, calculate ce = cd * cde + ce; There are still waves being transmitted, Δt + 1;

[0080] Step 4.19: Let t = 10. At this time, t = T = 10. Since the wave arrival time is equal to t, the neurons need to be updated continuously;

[0081] Step 4.20: For neuron E, Pe = {<A,B> <B,D> <D,E>}. Merge the path of E at t = 4 with the path at t = 9, and set Pe = {<A,B> <A,D> <B,E> <D,E>}. Pe={<A,B><A,D><B,D><B,E><D,E>}, calculate ce=cd*cde+ce; there is still a wave transmitting, Δt+1;

[0082] Step 4.21, let t=11, at this time t>T=10, stop updating neurons.

[0083] Step 5: Output the optimal cognitive subgraph. The specific steps are as follows:

[0084] Step 5.1: Output the optimal cognitive subgraph sequence <A,B> <A,D> <B,D> <B,E> <D,E>;

[0085] Step 5.2: Output the awareness cd of the path.

[0086] The above description is only a preferred example of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of the claims of the present invention.

Claims

1. An optimal cognitive subgraph query method based on temporal wave neural network, characterized in that: The following steps are involved: Step 1: Model the state and behavior of the node and construct a time-series wave neuron model; Step 2: Based on the neural network theory, construct the TSWNN neural network model; Step 3: Initialize all neurons in the time series wave neural network based on the initial state of the nodes; Step 4: Update the information of each neuron in TSWNN based on neural network theory and mathematical models and the dynamic changes between neurons; Step 5: Combine the propagation rules of time-series waves and the characteristics of information interaction in the network to deeply mine and analyze the network structure and output the optimal cognitive subgraph.

2. The optimal cognitive subgraph query method based on temporal wave neural network according to claim 1 is characterized in that: The specific steps of building the timing wave neuron model described in step 1 are as follows: Step 1.1: Design an input device, which is used to receive the timing waves from its predecessor neurons; Step 1.2: Design a wave receiver. The function of the wave receiver is to receive the wave information and decode it, including: i , w ij ,y ij and t j ; Step 1.3, neuron state memory, determines whether the neuron is activated, stores the neuron state, and stores the neuron recognition degree according to the information of the received wave, which is represented by y j and c j express; Step 1.4: Design a wave generator. When a neuron is activated, it generates a timing wave. The generated wave is transmitted to the unactivated successor neurons and consists of two parts: jk ,y jk ; Step 1.5: Design a wave transmitter. The transmitter sends waves to unactivated successor neurons and encodes the waves. When the successor neuron is not activated, the wave is sent; if the successor node is activated, the wave is not sent. This can be regarded as the reverse process of the wave transmitter. It contains two parts: jk ,y jk ; Step 1.6: Design the wave output device. The wave output device is mainly used to send the new wave to its corresponding successor neuron.

3. The optimal cognitive subgraph query method based on temporal wave neural network according to claim 1 is characterized in that: The specific steps of constructing the TSWNN neural network described in step 2 are as follows: Step 2.1: Construct a time-series wave neural network model based on the network topology of the problem, where each node corresponds to a neuron in TSWNN, each arc corresponds to an activity in TSWNN, and a random transmission time and cognition degree are generated for each arc (the edge connecting two neurons); Step 2.2, set the initial neuron S and the end neuron D; Step 2.3: Set the departure time of the initial neuron to t, where t is the earliest departure time from the initial neuron; Step 2.4: Set the neural network time limit T; Step 2.5: Set the time step Δt of the neural network iteration.

4. The optimal cognitive subgraph query method based on temporal wave neural network according to claim 1 is characterized in that: Initialize all neurons in the timing wave neural network described in step 3 and activate the initial neurons. The specific steps are as follows: Step 3.1: For the initial neuron S, let V i S =φ,V i P =φ,P i =φ,y i =1,c i =1; Step 3.2: For non-initial neurons, let V i S =φ,V i P =φ,P i =φ,y i =0,c i =0; Step 3.3: Set the initial time t=0.

5. The optimal cognitive subgraph query method based on temporal wave neural network according to claim 1 is characterized in that: Step 4 describes updating the state information of each neuron on the TSWNN, activating the neuron and generating waves. The specific steps are as follows: Step 4.1: Determine whether t is less than T. If t<T, perform the following operations; otherwise, perform step 5. Step 4.2: For each unactivated neuron i∈V on TSWNN, after receiving the first wave wni, the neuron state yi, neuron transmission time ti and cognitive degree ci, node path Pi are modified, wave wij is generated and sent; w ni =l ni ,y ni =c ni . t j =t i +l ij c j =c i *y ij +c j P j =<P i ,v j > Step 4.3: For the activated neuron j∈V on TSWNN, generate wave wjk to determine whether the successor node is activated. If not, send a wave to change the activation time Y. m ; If the node receives the wave, modify the node recognition c j , transmission time t j , node path P j ; Y m ＝l jm w jk ＝l jk c j =c i *y ij +c j P ij =<P i ,v j > Step 4.4, superimpose the timer t, t = t + Δt; Step 4.5: Determine again whether t is less than T. If t<T, execute steps 4.2-4.4; otherwise, execute step 5 and the search ends.

6. The optimal cognitive subgraph query method based on temporal wave neural network according to claim 1 is characterized in that: The calculation of the endpoint cognition described in step 5, merging the endpoint paths into a network, and outputting the optimal cognitive subgraph, the specific steps are as follows: Step 5.1: At t = T (neural network update stops), output the terminal path network, which is a set of terminal path neuron pairs; Step 5.2: Output the awareness cd of the path.

Images