OODA heterogeneous network link prediction method, device, equipment and medium

Abstract

The invention discloses an OODA heterogeneous network link prediction method, device and equipment and a medium, and relates to the technical field of OODA heterogeneous network link pre-processing, and the method comprises the steps: employing a CROTS algorithm based on an OODA heterogeneous network, and constructing a teacher-student model; and mapping nodes in the OODA heterogeneous network to a low-dimensional vector space, predicting the OODA heterogeneous network through a teacher model by adopting a probability propagation algorithm to obtain a prediction result, and training a student model based on the prediction result to perform link prediction. According to the method, the learning thought of teachers and students is introduced into OODA heterogeneous network link prediction, so that the link prediction speed is greatly improved. The teacher model effectively avoids the problem that the prediction time of a high-accuracy link prediction algorithm is long, and meanwhile, the high accuracy is kept; after the student model completes the teacher-student learning process, OODA heterogeneous network link prediction is carried out according to the topological characteristics, so that high accuracy can be kept while the prediction speed is high.

Description

Technical Field

[0001] The present application relates to the technical field of OODA heterogeneous network link prediction, and more specifically, to a OODA heterogeneous network link prediction method, device, equipment and medium. Background Art

[0002] Due to the rapid changes in information, the network topology of OODA heterogeneous networks is often missing. Link prediction is a complementary method to network topology, so there is an urgent need for a link prediction method suitable for OODA heterogeneous networks. The current link prediction methods are mainly divided into three categories:

[0003] 1. Traditional link prediction methods: This link prediction method is mainly based on link similarity. For example, the local similarity indicator common neighbor (CN) index calculates the possibility of link existence based on the number of common neighbors between two nodes; the global similarity indicator Katz index considers all paths between network nodes. This method has certain advantages in speed, but is often not very accurate.

[0004] 2. Methods based on network representation learning: This method mainly uses graph embedding or node embedding methods to represent the network for further link prediction. For example, the metapath2vec model combines the Skip-Gram model with meta-path-based random walks to construct heterogeneous neighborhoods of nodes, and then uses heterogeneous jump models as network representation methods for node embedding. LINE uses breadth-first search (BFS) random walks to sample nodes in the graph to build node embedding vectors, which further calculate node similarity for link prediction. However, this method is mainly used for homogeneous networks, has low accuracy in heterogeneous networks, and takes a long time.

[0005] 3. Machine learning-based methods: Traditional machine learning algorithms mainly regard link prediction as a supervised learning task, and use classification algorithms to classify existing or possible links and non-existent links in the network to achieve the purpose of link prediction. However, with the breakthrough of machine learning technology, many scholars have begun to use graph embedding technology combined with classifiers to solve the link prediction problem of complex networks. For example, NGCF designed a neural network method to recursively implement graph embedding, thereby constructing a recommendation system framework based on graph neural network. This method has certain advantages in accuracy, but because it requires a large amount of data for pre-training, it will have the problem of insufficient fitting in small networks.

[0006] Therefore, the existing link prediction algorithms are not tightly integrated with OODA heterogeneous networks. Conducting link prediction research on OODA heterogeneous networks and accurately, reasonably and quickly predicting the links that may appear in the network is an objective need to deeply understand the internal mechanism of system network evolution and explore potential links in the network. It is a practical need to promote the development of system link prediction capabilities in the information age and an urgent need to break through the technical bottleneck of OODA heterogeneous network link prediction research. Summary of the invention

[0007] In response to at least one defect or improvement need in the prior art, the present invention provides an OODA heterogeneous network link prediction method, device, equipment and medium to address the defects of the prior art. By analyzing and utilizing the characteristic information of nodes and links in the network, a specific quantitative evaluation is performed on the similarity or connection possibility between possible node pairs that are not yet directly connected in the network, and possible links between nodes are predicted, thereby achieving accurate, reasonable and fast link prediction in the OODA heterogeneous network.

[0008] To achieve the above object, according to a first aspect of the present invention, a method for link prediction in an OODA heterogeneous network is provided, characterized in that it includes:

[0009] Based on OODA heterogeneous network, the CROTS algorithm is used to build a teacher-student model;

[0010] Map the nodes in the OODA heterogeneous network to a low-dimensional vector space, and use the probability propagation algorithm through the teacher model to predict the OODA heterogeneous network and obtain the prediction results;

[0011] A student model is trained based on the prediction results to perform link prediction.

[0012] As described in the OODA heterogeneous network link prediction method, the prediction of the OODA heterogeneous network by using the probability propagation algorithm through the teacher model to obtain the prediction result includes:

[0013] The Meta-Probs algorithm is used to construct the teacher model and obtain the node topology structure characteristics;

[0014] Based on the node topology structure characteristics, the degree, H-index and density (DHC) theorem are used to represent the network and obtain the node embedding vector;

[0015] Based on the DHC theorem, the convergence process of the H index sequence is used as the node embedding vector;

[0016] The cosine distance between node embedding vectors is used to quantify the similarity between nodes and used as edge weight to control the resource diffusion of the ProbS framework to obtain the link weight matrix.

[0017] Link prediction is performed according to the link weight matrix to obtain a prediction result.

[0018] As described in the OODA heterogeneous network link prediction method, the student model is trained based on the prediction result to perform link prediction, including:

[0019] Construct the adjacency matrix of the OODA heterogeneous network;

[0020] constructing a similarity function based on the adjacency matrix;

[0021] A ranking result is determined based on the similarity function to obtain a predicted link.

[0022] As described in the OODA heterogeneous network link prediction method, the similarity function is constructed based on the adjacency matrix, including:

[0023] Initializing optimal parameter values ​​and corresponding maximum predicted link numbers based on the adjacency matrix;

[0024] Analyze the prediction results of the teacher model based on the best parameter value and the maximum number of predicted links to obtain the updated best parameter value;

[0025] A similarity function is determined based on the updated optimal parameter value.

[0026] As described in the OODA heterogeneous network link prediction method, the similarity function is specifically:

[0027]

[0028] Among them, S ij represents the similarity between node i and node j, CN and CN_L3 represent the second-order path and third-order path indicators of node i and node j respectively; the α value is obtained through model learning; T(i) and T(j) represent the types of node i and node j respectively.

[0029] As described in the OODA heterogeneous network link prediction method, the adjacency matrix includes a second-order path matrix and a third-order path matrix.

[0030] As described in the OODA heterogeneous network link prediction method, the CROTS algorithm is used based on the OODA heterogeneous network model to construct a teacher-student model, including:

[0031] Conduct network modeling for OODA heterogeneous network and build OODA heterogeneous network model;

[0032] A teacher-student model is constructed based on the OODA heterogeneous network model.

[0033] According to the second aspect of the present invention, there is also provided an OODA heterogeneous network link prediction device, comprising: a construction unit, used to construct a teacher-student model based on the OODA heterogeneous network model using the CROTS algorithm; a first processing unit, used to map the nodes in the OODA heterogeneous network to a low-dimensional vector space, and predict the OODA heterogeneous network using a probability propagation algorithm through the teacher model to obtain a prediction result; a second processing unit, used to train a student model based on the prediction result to perform link prediction.

[0034] According to a third aspect of the present invention, there is also provided a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned OODA heterogeneous network link prediction method when running.

[0035] According to a fourth aspect of the present invention, there is also provided an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the OODA heterogeneous network link prediction method through the computer program.

[0036] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0037] (1) The present invention provides a OODA heterogeneous network link prediction method, which successfully introduces the idea of ​​teacher-student learning into the OODA heterogeneous network link prediction by targeting the problem of insufficient computing power of devices in reality and high real-time requirements. A teacher-student model is constructed by using the CROTS algorithm based on the OODA heterogeneous network model; the nodes in the OODA heterogeneous network are mapped to a low-dimensional vector space, and the OODA heterogeneous network is predicted by the teacher model using a probability propagation algorithm to obtain a prediction result, and then the student model is trained based on the prediction result to perform link prediction. The teacher model uses prior knowledge for learning, which effectively avoids the problem of long prediction time of high-accuracy link prediction algorithms, while maintaining high accuracy; after completing the teacher-student learning process, the student model performs OODA heterogeneous network link prediction based on topological features, so that it can have a faster prediction speed while retaining high accuracy, meeting the real-time requirements of the OODA heterogeneous network, and providing a guarantee for individuals or organizations to grasp the situation and make correct commands in a shorter time.

[0038] (2) This application uses a highly accurate link prediction algorithm as a teacher model and a faster link prediction algorithm as a student model, and complements the advantages of the two through teacher-student learning. In order to promote the teacher-student learning process and ensure the robustness of the student model, the similarity calculation index is improved by combining the second-order path with the third-order path. The current mainstream network structure similarity calculation index is the CN index, but the prediction accuracy of this index will drop sharply in OODA heterogeneous networks with high sparsity (i.e., there are few or even no second-order neighbors). Therefore, this application adds the third-order path to the similarity calculation index, which to a certain extent alleviates the situation where there are too few second-order paths and accurate link prediction cannot be performed, thereby improving the robustness of prediction when the system environment changes. At the same time, according to the characteristics of OODA heterogeneous networks, such as decision makers are often in a central position, the similarity of some node links is appropriately modified so that the similarity calculation index can be more in line with the characteristics of the OODA heterogeneous network model. Since this indicator cannot interact with the teacher model, this application adopts the method of optimal parameter value learning, that is, using the learning process to obtain the optimal parameter value to balance the second-order path and third-order path indicators, so that the student model can obtain the most suitable parameter value for the current OODA heterogeneous network after learning, so as to achieve the purpose of accurate and rapid prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 A flowchart of an optional OODA heterogeneous network link prediction method provided in an embodiment of the present application;

[0041] Figure 2 A schematic diagram of the structure of an optional OODA heterogeneous network link prediction device provided in an embodiment of the present application;

[0042] Figure 3 A schematic diagram of the structure of the network topology of an optional OODA heterogeneous network provided in an embodiment of the present application.

[0043] Figure 4 A schematic diagram of the structure of an optional link prediction algorithm framework provided in an embodiment of the present application.

[0044] Figure 5 A schematic diagram of the structure of an optional teacher model framework provided in an embodiment of the present application.

[0045] Figure 6A schematic diagram of the structure of an optional electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0047] The terms "first", "second", "third", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices.

[0048] The technology involved in the present invention is briefly introduced below.

[0049] The Observe-Orient-Decide-Act (OODA) loop proposed by Colonel John Boyd of the U.S. Air Force is widely used in complex systems in the military, business and other fields. It covers the entire process from observing the environment (Observe), determining the goal (Orient), making decisions (Decide) and taking actions (Act), which can provide effective decisions for individuals or organizations in the system and guide actions reasonably. These systems are composed of entities that perform different functions. When abstracting these systems, we can regard these entities as nodes in the network and the links between entities as links on the network. In this way, the system can be transformed into a heterogeneous network, that is, a network composed of different types of nodes and links.

[0050] Since these systems follow the decision-making and execution patterns in the OODA loop and can be abstracted as heterogeneous networks, these heterogeneous networks with OODA loop characteristics are defined as OODA heterogeneous networks. Their network structure not only reflects the complexity of the system, but also shows the dynamic interaction and decision-making process between the various parts of the system.

[0051] In fact, OODA heterogeneous networks exist in a wide range. For example, in the police system, the handling of cases follows the logic of the OODA cycle: first observe the scene of the case and collect relevant information; then locate the nature of the case and key clues based on the information. Then make decisions and assign tasks; finally, take action to solve the case. Similarly, in the enterprise system, opportunity management also follows the process of observing market dynamics: first observe the market dynamic environment; then judge the direction of market demand; then formulate business strategies; finally implement corresponding action plans. In-depth research on the heterogeneous network structure of OODA is not only of great significance for solving police cases and gaining insight into the enterprise market, so the research on the OODA heterogeneous network model will have a far-reaching impact on many fields and has important academic value and practical significance.

[0052] In OODA heterogeneous networks, each node carries a specific function, and the links between nodes are the key to understanding the network structure and dynamic characteristics. However, in practical applications, due to the dynamic changes of information, the node structure in the network is often missing, such as hidden criminal clues or encrypted business secrets. The hiddenness or lack of these potential information not only affects our comprehensive understanding of the network structure, but also hinders our insight into the potential relationships between nodes. With the emergence of a large number of OODA heterogeneous network scenarios in real life, it is very important to obtain the complete topological structure of the OODA heterogeneous network. Link prediction can help the police grasp the case as soon as possible and help companies seize marketing opportunities.

[0053] Link prediction aims to use known network structure information to judge and predict the possibility of potential links between nodes, thereby improving the network structure and revealing potential relationships. At present, mainstream link prediction methods have shown certain effects in complex network analysis, such as Deepwalk, Node2vec, LINE, SDNE based on graph embedding methods, and neural network-based methods such as GCN, GAT, NGCF. However, these algorithms are mainly designed for homogeneous networks, while the complexity of heterogeneous network structures and relationships greatly limits the application of these algorithms.

[0054] Although some studies have proposed link prediction algorithms for heterogeneous networks, such as Meta AIProbs and metapath2vec, they still have certain limitations when applied to OODA heterogeneous networks, including slow prediction speed, which makes it difficult to meet the time-sensitive requirements of OODA heterogeneous networks (such as business competition, where business opportunities are fleeting); they do not have the function of processing prior knowledge and are not suitable for OODA heterogeneous networks with prior knowledge characteristics (such as in police networks, where the police have a fingerprint database of all citizens).

[0055] Therefore, the existing link prediction algorithms are not tightly integrated with OODA heterogeneous networks, and further research is still needed on the link prediction methods of OODA heterogeneous networks. This patent intends to introduce the teacher-student model (TSFNLP) based on the idea of ​​the CROTS algorithm, taking into account the high precision and efficiency of link prediction, and providing a link prediction calculation model that conforms to the structural characteristics of the OODA heterogeneous network. First, the OODA heterogeneous network is networked and modeled, and then the teacher-student model (TSFNLP) is proposed based on the CROTS algorithm to map the nodes in the OODA heterogeneous network to a low-dimensional vector space. The teacher model uses a probability propagation algorithm to predict the OODA heterogeneous network, and trains the student model through the prediction results. Finally, the student model is used to perform link prediction to solve the link prediction problem in the OODA heterogeneous network.

[0056] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0057] According to one aspect of an embodiment of the present application, a OODA heterogeneous network link prediction method is provided. Figure 1 is a flow chart of an optional OODA heterogeneous network link prediction method provided in an embodiment of the present application, such as Figure 1 As shown, the process of the method may include the following steps (STEP, referred to as S):

[0058] S102, based on OODA heterogeneous network, adopts CROTS algorithm to build a teacher-student model;

[0059] S104, mapping nodes in the OODA heterogeneous network to a low-dimensional vector space, and predicting the OODA heterogeneous network using a probability propagation algorithm through a teacher model to obtain a prediction result;

[0060] S106, training a student model based on the prediction result to perform link prediction.

[0061] In an optional embodiment, the method of using the CROTS algorithm based on the OODA heterogeneous network model to construct a teacher-student model includes:

[0062] Conduct network modeling for OODA heterogeneous network and build OODA heterogeneous network model;

[0063] A teacher-student model is constructed based on the OODA heterogeneous network model.

[0064] In the network modeling of OODA heterogeneous network, based on OODA loop, OODA heterogeneous network is a heterogeneous complex network formed by abstracting network entities into nodes and abstracting the information, material, energy and other exchanges between entities into links. The entities are abstracted into three types of nodes: OO, D and A, which are described as follows:

[0065] 1.OO Node (Sensor): As a sensor node, the main responsibility of the OO node is to receive observable phenomena or information from network entities. These phenomena can be signals, data or events in the physical world, or state changes in other network nodes. The OO node integrates the received information and sends it to the D node for subsequent analysis and decision making.

[0066] 2. D nodes (decision makers): As information judgment and decision nodes, D nodes play a core role in heterogeneous networks. They receive information from OO nodes and conduct in-depth analysis and processing of this information to understand the status and changes of the network. Based on this information, D nodes decide the actions of other nodes, including how to respond to external events, how to adjust the network structure, or how to optimize network performance.

[0067] 3. A Node (Influencer): As a participant, A Node receives instructions or commands from D Node and interacts with other nodes based on these instructions. The actions of A Node can have a direct impact on the network, including changing the network state, transmitting information, or performing specific tasks. They act as bridges and links in the network, ensuring that the decisions of D Node can be effectively executed.

[0068] In OODA heterogeneous networks, information flows between entities are the core mechanism of network operation, and these information flows are usually abstracted as directed edges between nodes. The traditional OODA heterogeneous network model assumes that OO nodes only act as information providers and do not directly participate in operations, thus excluding direct links from OO nodes to A nodes. However, with the development of wireless sensor network technology, this assumption is no longer completely applicable. Wireless sensor networks enable OO nodes (sensors) to directly control A nodes (influencers) through wireless connections, thereby improving the response speed and flexibility of the entire network system. This direct control link is particularly important in scenarios with high real-time requirements, such as automatic control, intelligent environmental monitoring, etc. At the same time, considering the position of A nodes in the network information flow, they are usually located at the "nerve endings" and are responsible for performing specific operations or influencing other entities. Therefore, this paper assumes that there are no direct links between A nodes in OODA heterogeneous networks.

[0069] Based on the above analysis, this paper abstracts and models the links in the OODA heterogeneous network. The links are divided into 8 categories, as shown in Table 1:

[0070]

[0071]

[0072] Model construction based on OODA heterogeneous network, Figure 2 This is an example diagram of the OODA heterogeneous network topology. Figure 2 The left figure shows an example network topology of the OODA heterogeneous network, and the right figure shows three representative nodes and the links between them: solid lines represent currently existing links, and dashed lines represent possible links based on the OODA heterogeneous network link prediction.

[0073] In the construction of the link prediction algorithm framework based on the CROTS algorithm, the CROTS (Cross-domain Teacher-Student Learning for Passive Domain Adaptive Semantic Segmentation) algorithm has significant application value in the field of image learning. The core idea is to introduce a teacher model to generate pseudo labels and provide learning goals for the student model. At the same time, the prediction results of the student model also provide feedback for the optimization of the teacher model. The performance of the teacher model is evaluated by comparing the difference between the prediction results of the student model and the actual labels, and further optimized accordingly. This interaction and feedback mechanism between teachers and students enables the CROTS algorithm to gradually improve its performance in the continuous learning process. The algorithm takes into account the pseudo labels of class imbalance and the limitations of the minority classes that the target model easily forgets. It adopts a rare class patch mining strategy to alleviate the imbalance phenomenon. The CROTS algorithm process is as follows Figure 3 shown.

[0074] The link prediction algorithm framework proposed based on the CROTS algorithm is as follows Figure 4 As shown in the figure. First, the nodes in the OODA heterogeneous network are mapped to a low-dimensional vector space through a deterministic and interpretable network representation method. Then the teacher model uses the probability propagation algorithm to predict the OODA heterogeneous network, and trains the student model through the prediction results. Finally, the student model uses the topological characteristics of the OODA heterogeneous network nodes to perform link prediction to achieve the desired effect.

[0075] In an optional embodiment, the OODA heterogeneous network link prediction results are generated based on the teacher model, and the Meta-Probs algorithm can be used as the teacher model algorithm. In the Meta Probs algorithm, the OODA heterogeneous network uses a network representation method based on the degree, H-index and density (DHC) theorem to further obtain the node embedding vector. The centrality of a node refers to the importance of the connection between the node and other nodes in the network structure or function. Node degree, H-index and core number are both key indicators in the DHC theorem and commonly used indicators for judging the centrality of nodes.

[0076] Node degree refers to the number of neighbors of a node. An example of the definition of node degree is:

[0077]

[0078] Among them, Γ (i) Indicates v i The set of neighbor nodes of a. ij is the adjacency matrix A n×n The elements in .

[0079] The H-index is the maximum value of at least h neighbors with degree not less than h in the neighbor node set. Define node v i H-index:

[0080] h i =H(d1,d2,d3,…,d j ) (3-2)

[0081] Define the H-index of order n:

[0082]

[0083] The coreness of a node describes its central position in the network. The coreness is calculated by k-shell decomposition.

[0084] According to the DHC theorem, in the process of network decomposition, the degree, H-index and core number of the node correspond to the initial state, intermediate state and stable state respectively. Any node in a complex network will converge to the core number within a finite step length. The convergence step length of this sequence can reflect the size of the node's influence, that is, the more important the node, the slower its H-index decays and the larger the convergence step length.

[0085] Using node v in the Meta-Probs algorithm i The convergence vector of is used as the node embedding vector, and the vector dimension is the finite step length s of its convergence. The node embedding vector thus formed can well judge the centrality of the node.

[0086] Construct node similarity matrix S using node embedding vectors n×n , element s ij ∈S n×n Represents a node pair (v i ,v j ) is a similarity measure of the node pair (v i ,v j ) ij The -cosine distance of its corresponding node embedding vector:

[0087]

[0088] Where x is v i The node embedding vector of y is v j The node embedding vector of .

[0089] Normalize the matrix elements:

[0090]

[0091] Among them, min is S n×n The smallest element in, max is S n×n The largest element in .

[0092] In order to consider the impact of information flow in OODA heterogeneous networks and the difference in the influence of each node,

[0093] The algorithm uses the cosine distance of node embedding vectors as edge weights under the Probs framework.

[0094] Definition ij is the edge weight between node i and node j:

[0095]

[0096] The resource diffusion process of the ProbS framework is controlled by edge weights, including forward propagation and reverse diffusion.

[0097] According to the link construction of OODA heterogeneous network, in forward propagation, we use edge weight as the carrying resource according to and The two meta-paths are propagated without back-propagation. After the resource propagation of all decision nodes and sensor nodes is completed, the reverse diffusion process of resources between sensor nodes and response nodes begins. In the reverse diffusion process, sensor nodes and response nodes pass resources along the meta-paths OO→D←A and A→OO to the nodes connected to them.

[0098] Finally, according to the type and quantity of resources carried by the decision nodes, the relationship weight matrix of the decision nodes is established, and based on the matrix, it is predicted whether there is an unobserved link between the two nodes.

[0099] In summary, the teacher model studies the topological structure characteristics of nodes based on the Meta-Probs algorithm, and uses the convergence process of the H-index sequence as the node embedding vector based on the DHC theorem; the vector obtained by this network representation method has the advantages of determinable dimension and interpretable elements.

[0100] Then, the cosine distance between node embedding vectors is used to quantify the similarity between nodes, and used as edge weight to control the resource diffusion process of the ProbS framework.

[0101] Finally, link prediction is performed based on the ranking of the scores (or setting a threshold). The algorithm flow is shown in Figure 5.

[0102] In training the student model based on the prediction results of the teacher model and generating link prediction results, the student model is committed to achieving relatively accurate link prediction in the shortest possible time. To achieve this goal, this study uses a link prediction algorithm based on the characteristics of the network topology in the student model. In addition, to further improve the accuracy of link prediction, we deeply analyze the output results of the teacher model and continuously adjust the similarity calculation formula of the student model based on the analysis results.

[0103] Through this process, we successfully determined the corresponding optimal parameter values ​​for the corresponding OODA heterogeneous network and applied them to the link prediction of the student model.

[0104] The connections between nodes on a real network are localized, so using the local topological information between nodes for prediction will achieve better results. Currently, the representative indicators include the common neighbor indicator (CN indicator).

[31] , Resource Allocation Index (RA Index)

[33] and Adamic-Adar indicator (AA indicator)

[32] Given a network G = (V, E, T), where V is the set of nodes, E is the set of edges, and T is the set of node or edge types. ij ] N×N represents the adjacency matrix of G. If there is a link between node i and node j, then a ij =1, otherwise a ij = 0. The calculation formula of CN index is as follows:

[0105]

[0106] in is the set of neighbor nodes of node i, is the set of neighbor nodes of node j.

[0107] However, the second-order path method shows obvious limitations when dealing with sparse networks, and tends to over-prefer certain specific connections in dense network structures. In order to overcome this challenge and specifically target the structural characteristics of protein networks, scholars such as KOVK CS [8] A similarity calculation strategy based on third-order paths is proposed. This algorithm significantly improves the prediction accuracy of potential interaction relationships in protein networks.

[0108] Inspired by this study, ZHOU et al.

[60] further integrated the three key indicators of CN, AA and RA into the framework of the third-order path and constructed a new local similarity algorithm. The research data showed that in some network scenarios, the prediction method using the third-order path showed higher prediction accuracy than the second-order path method, thus proving its superior performance. Similar to the second-order path indicator, Martínez et al.

[14] Three third-order path similarity indices, CN_L3, AA_L3 and RA_L3, are proposed. In this paper, the CN_L3 index is used to represent the third-order path similarity, which is defined as follows:

[0109]

[0110] in is the set of neighbor nodes of node i, is the set of neighbor nodes of node j, a xy are the elements in the adjacency matrix.

[0111] In order to make the similarity calculation formula adaptable to sparse networks and perform well in dense networks, this paper uses a combination of third-order paths and second-order paths to quantify similarity:

[0112] S ij =CN+α×CN_L3 (4-3)

[0113] Where S ij It represents the similarity between node i and node j, CN and CN_L3 represent the second-order path and third-order path indicators of node i and node j respectively. The α value is obtained through model learning.

[0114] Considering the heterogeneity of OODA heterogeneous networks and analyzing the characteristics of OODA heterogeneous networks, it can be found that the D node is located at the central node of the network, so the link similarity between the D node is often greater than that between other nodes. At the same time, since the A node is located at the nerve endings of the network, there is no possibility of links between the A nodes.

[0115] Taking all factors into consideration, we adjust the similarity formula:

[0116]

[0117] Where T(i) and T(j) represent the types of node i and node j respectively.

[0118] The student model needs to make fine adjustments and optimizations to the similarity formula with the help of the teacher-student learning mechanism to ensure that it can more accurately adapt to the characteristics of the OODA heterogeneous network and achieve better link prediction results.

[0119] In this learning process, the student model makes an in-depth analysis of the prediction results of the teacher model and adjusts the parameters and weights of its similarity calculation formula in a targeted manner. In this way, the student model can gradually improve the accuracy and stability of its link prediction, and ultimately achieve the goal of efficient and accurate link prediction in OODA heterogeneous networks.

[0120] According to another aspect of the embodiment of the present application, a macro model construction device for implementing the above-mentioned OODA heterogeneous network link prediction method is also provided. The device may include:

[0121] A construction unit for constructing a teacher-student model based on the OODA heterogeneous network model using the CROTS algorithm;

[0122] A first processing unit is used to map nodes in the OODA heterogeneous network to a low-dimensional vector space, and predict the OODA heterogeneous network using a probability propagation algorithm through a teacher model to obtain a prediction result;

[0123] The second processing unit is used to train a student model based on the prediction result to perform link prediction.

[0124] It should be noted here that the examples and scenarios implemented by the above-mentioned modules and corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned embodiments. It should be noted that the above-mentioned modules as part of the device can run in a hardware environment and can be implemented by software or hardware, wherein the hardware environment includes a network environment.

[0125] According to another aspect of the embodiments of the present application, a storage medium is also provided. Optionally, in this embodiment, the storage medium can be used to execute the program code of any of the above OODA heterogeneous network link prediction methods in the embodiments of the present application.

[0126] Optionally, in this embodiment, the storage medium is configured to store program codes for executing the following steps:

[0127] S1, obtaining S parameter data of the target system and constructing an initial macro model using a rational function form in the frequency domain, wherein the rational function form is a partial expansion;

[0128] S1, based on OODA heterogeneous network, adopts CROTS algorithm to build a teacher-student model;

[0129] S2, maps the nodes in the OODA heterogeneous network to a low-dimensional vector space, and uses the probability propagation algorithm to predict the OODA heterogeneous network through the teacher model to obtain the prediction results;

[0130] S3, training a student model based on the prediction result to perform link prediction.

[0131] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, which will not be described in detail in this embodiment.

[0132] Among them, computer-readable storage media may include, but are not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives and magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0133] According to another aspect of an embodiment of the present application, an electronic device for implementing the above-mentioned OODA heterogeneous network link prediction method is also provided. The electronic device may be a server, a terminal, or a combination thereof.

[0134] Figure 6 is a schematic diagram of the structure of an optional electronic device according to an embodiment of the present application, such as Figure 6 As shown, it includes a processor 602, a communication interface 604, a memory 606 and a communication bus 608, wherein the processor 602, the communication interface 604, and the memory 606 communicate with each other through the communication bus 608, wherein,

[0135] Memory 606, used to store computer programs;

[0136] The processor 602 is used to implement the following steps when executing the computer program stored in the memory 606:

[0137] S1, based on OODA heterogeneous network, adopts CROTS algorithm to build a teacher-student model.

[0138] S2, maps the nodes in the OODA heterogeneous network to a low-dimensional vector space, and uses the probability propagation algorithm through the teacher model to predict the OODA heterogeneous network and obtain the prediction results.

[0139] S3, training a student model based on the prediction result to perform link prediction.

[0140] Optionally, the communication bus may be a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The communication bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6The communication interface is used for communication between the electronic device and other devices.

[0141] The memory may include RAM, or may include non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0142] The above-mentioned processor can be a general-purpose processor, which can include but not be limited to: CPU (Central Processing Unit), NP (Network Processor), etc.; it can also be DSP (Digital Signal Processing), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0143] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, and this embodiment will not be described in detail here.

[0144] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0145] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0146] In the several embodiments provided in the present application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of the units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interfaces, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0147] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0148] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0149] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a memory, including a number of instructions to enable a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned memory includes: U disk, read-only memory (ROM), random access memory (RAM), mobile hard disk, disk or optical disk and other media that can store program code.

[0150] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable memory, which can include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0151] The above is only an exemplary embodiment of the present disclosure, and the scope of the present disclosure cannot be limited thereto. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the specification and practicing the disclosure here, those skilled in the art will easily think of the implementation scheme of the present disclosure. This application is intended to cover any modification, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the technical field not recorded in the present disclosure. The description and examples are regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

[0152] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0153] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A OODA heterogeneous network link prediction method, characterized in that: include: Based on the OODA heterogeneous network model, the CROTS algorithm is used to build a teacher-student model; Based on mapping the nodes in the OODA heterogeneous network to a low-dimensional vector space, the probability propagation algorithm is used to predict the OODA heterogeneous network through the teacher model to obtain the prediction results; A student model is trained based on the prediction results to perform link prediction.

2. The OODA heterogeneous network link prediction method according to claim 1, characterized in that: The method of using the probability propagation algorithm to predict the OODA heterogeneous network through the teacher model to obtain the prediction results includes: The Meta-Probs algorithm is used to construct the teacher model and obtain the node topology structure characteristics; Based on the node topology structure characteristics, the degree, H-index and density (DHC) theorem are used to represent the network and obtain the node embedding vector; Based on the DHC theorem, the convergence process of the H index sequence is used as the node embedding vector; The cosine distance between node embedding vectors is used to quantify the similarity between nodes and used as edge weight to control the resource diffusion of the ProbS framework to obtain the link weight matrix. Link prediction is performed according to the link weight matrix to obtain a prediction result.

3. The OODA heterogeneous network link prediction method according to claim 1, characterized in that: The training of the student model based on the prediction result to perform link prediction includes: Construct the adjacency matrix of the OODA heterogeneous network; constructing a similarity function based on the adjacency matrix; A ranking result is determined based on the similarity function to obtain a predicted link.

4. The OODA heterogeneous network link prediction method according to claim 3, characterized in that: The constructing a similarity function based on the adjacency matrix comprises: Initializing optimal parameter values ​​and corresponding maximum predicted link numbers based on the adjacency matrix; Analyze the prediction results of the teacher model based on the best parameter value and the maximum number of predicted links to obtain the updated best parameter value; A similarity function is determined based on the updated optimal parameter value.

5. The OODA heterogeneous network link prediction method according to claim 3, characterized in that: The similarity function is specifically: Among them, S ij represents the similarity between node i and node j, CN and CN_L3 represent the second-order path and third-order path indicators of node i and node j respectively; the α value is obtained through model learning; T(i) and T(j) represent the types of node i and node j respectively.

6. The OODA heterogeneous network link prediction method according to claim 3, characterized in that: The adjacency matrix includes a second-order path matrix and a third-order path matrix.

7. The OODA heterogeneous network link prediction method according to claim 1, characterized in that: The CROTS algorithm is used in the OODA heterogeneous network model to construct a teacher-student model, including: Conduct network modeling for OODA heterogeneous network and build OODA heterogeneous network model; A teacher-student model is constructed based on the OODA heterogeneous network model.

8. An OODA heterogeneous network link prediction device, characterized in that: include: A construction unit for constructing a teacher-student model based on the OODA heterogeneous network model using the CROTS algorithm; A first processing unit is used to map nodes in the OODA heterogeneous network to a low-dimensional vector space, and predict the OODA heterogeneous network using a probability propagation algorithm through a teacher model to obtain a prediction result; The second processing unit is used to train a student model based on the prediction result to perform link prediction.

9. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to execute the method according to any one of claims 1 to 7 through the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program executes the method according to any one of claims 1 to 7 when executed.

Images