A network representation method and system based on multidimensional heterogeneous resource summary views

Abstract

This invention discloses a network representation method and system based on a multidimensional heterogeneous resource summary view, belonging to the field of communication network technology. The method includes: acquiring network topology, device, and link status information; constructing a global original view; initializing each node and link as a supernode and superedge respectively to form an initial summary view; generating candidate supernode pairs; calculating the combined merging benefit that balances storage compression and path quality distortion; and iteratively performing supernode merging under attribute purity constraints until the summary view size meets a preset target; during the merging process, aggregating and updating the multidimensional attributes of superedges; and retaining only superedges that reduce overall storage overhead based on the minimum description length principle. This invention reduces the network state size while maintaining effective representation of multidimensional resource characteristics and possesses good dynamic adaptability, making it suitable for efficient modeling and analysis in large-scale complex network environments.

Description

Technical Field

[0001] This invention belongs to the field of communication network technology, specifically relating to a network representation method and system based on a multidimensional heterogeneous resource summary view. Background Technology

[0002] As communication networks continue to expand, the number of network nodes, link size, and resource status dimensions are constantly increasing, resulting in network states characterized by large scale, complex structure, and numerous attributes. Against this backdrop, how to efficiently model and compress the representation of large-scale network states has become a key issue in improving the efficiency of network analysis and management.

[0003] To reduce the storage overhead and computational complexity of complex network states, graph summarization techniques are widely used in network representation. This technique compresses the original network topology and its associated attributes, constructing a smaller summary view while preserving key structural features and attribute information as much as possible, thereby improving the efficiency of subsequent processing. Based on whether the original graph can be completely reconstructed, graph summarization methods are generally divided into two categories: lossless graph summarization and lossy graph summarization.

[0004] In lossless graph summarization methods, existing research mainly achieves graph structure compression through node clustering, topology reconstruction, and encoding optimization. While these methods maintain the integrity of the original information, their compression efficiency is limited and cannot meet the demands for high-efficiency processing in large-scale network environments. In contrast, lossy graph summarization methods, by discarding some detailed information, have significant advantages in compression efficiency and computational performance, making them more suitable for large-scale complex network scenarios.

[0005] However, most existing lossy graph summarization methods are designed for unweighted graphs or single-attribute graphs, or only model node attributes, lacking the ability to uniformly characterize multi-dimensional heterogeneous resources, and especially failing to effectively integrate the relationship between link attributes and node attributes. In real-world networks, node resource states and link states jointly determine the overall characteristics of the network. If only single-dimensional information is considered, the summarization view may be insufficient in representing the original network, reducing its effectiveness in applications such as network analysis, resource scheduling, and performance evaluation.

[0006] Therefore, there is an urgent need to propose a network representation method that can comprehensively represent multidimensional heterogeneous resource information, while reducing the scale of network states and improving the accuracy of expressing the original network structure and attribute characteristics, so as to meet the needs of efficient analysis and processing in large-scale complex network environments. Summary of the Invention

[0007] This invention addresses the shortcomings of existing technologies that only model node attributes, lacking a unified capacity to characterize multidimensional heterogeneous resources, and particularly failing to effectively integrate the relationships between link attributes and node attributes. This results in insufficient precision in representing the original network structure and attribute features, making it unsuitable for efficient analysis and processing in large-scale complex network environments. The invention provides a network representation method based on a multidimensional heterogeneous resource summary view. This method acquires information on network topology, device and link states, constructs a global original view, and initializes each node and link as a supernode and superedge, respectively, forming an initial summary view. Candidate supernode pairs are generated, and the combined merging benefit, balancing storage compression gains and path quality distortion, is calculated. Under attribute purity constraints, supernode merging is iteratively performed until the summary view size meets a preset target. This invention reduces the network state size while maintaining effective representation of multidimensional resource features and possesses good dynamic adaptability, making it suitable for efficient modeling and analysis in large-scale complex network environments.

[0008] According to one aspect of this specification, a network representation method based on a multidimensional heterogeneous resource summary view is provided, comprising:

[0009] Step 1: Collect information on network topology, devices, and link status to construct a global original view of multidimensional heterogeneous resources;

[0010] Step 2: Initialize each node and each link in the global original view as a super node and a super edge, respectively, and construct an initial summary view based on all super nodes and super edges;

[0011] Step 3: Based on the initial summary view, a candidate supernode pair set is generated using a wedge sampling-based candidate generation strategy;

[0012] Step 4: Based on the preset attribute purity constraint, the candidate supernode pair set is filtered to obtain candidate supernode pairs whose attribute purity is greater than or equal to the set attribute purity threshold; the new supernode is the supernode formed by merging the two supernodes in the supernode pair.

[0013] Step 5: Calculate the combined benefit of each candidate supernode pair after screening; the combined benefit is the weighted sum of storage compression benefit and normalized path quality distortion penalty;

[0014] Step 6: Merge the two supernodes in the candidate supernode pair with the highest overall benefit, update the node mapping relationship, aggregate the superedge attributes based on the number of nodes, and determine whether to retain the superedge. Iterate through steps 3 to 6 until the number of supernodes in the summary view is reduced to the preset target number.

[0015] As a further technical solution, step 3 includes:

[0016] Calculate the degree of each supernode in the initial summary view, and calculate the wedge number of the central supernode and the total wedge number of the entire graph based on the degree;

[0017] Based on the preset sampling budget, the number of samples is allocated according to the proportion of the wedge number of each central supernode to the total number of wedges in the entire graph. The number of samples is denoted as [sample number]. ;

[0018] Based on the neighbor set of each central supernode, extract Each pair of supernodes is counted, and the occurrence count of each pair of supernodes is accumulated.

[0019] Based on the occurrence count, candidate merge objects corresponding to each supernode are determined, and the candidate merge objects are sorted and filtered according to the occurrence count from high to low.

[0020] Based on the candidate merged objects after sorting and filtering, a two-way mutual selection verification is performed, and the supernode pairs that pass the two-way mutual selection verification are included in the candidate supernode pair set.

[0021] As a further technical solution, the purity of the attribute is defined as the ratio of the number of nodes corresponding to the device type with the highest proportion inside the supernode to the total number of nodes inside the supernode.

[0022] As a further technical solution, the expression for the storage compression benefit is:

[0023]

[0024] in, Indicates the benefits of storage compression. Indicates a pair of supernodes. Indicates the new local storage overhead. This indicates the local storage overhead.

[0025] As a further technical solution, the expression for the normalized path quality distortion penalty is:

[0026]

[0027] in, This represents the penalty for normalized path quality distortion. This indicates the local maximum path quality difference. Indicates a pair of supernodes. Represents the set of candidate supernode pairs. It represents the smallest positive number that is preset.

[0028] As a further technical solution, the hyperedge attributes are aggregated in a weighted manner based on the number of nodes, including:

[0029] Based on any neighboring supernode of the merged new supernode, obtain the three-dimensional path quality vectors between the two merged supernodes and their neighboring supernodes before the merger.

[0030] The number of original nodes contained in each of the two merged supernodes is used as the weight to perform a weighted average on the two 3D path quality vectors, resulting in the 3D path quality vector between the new supernode and its neighboring supernodes.

[0031] As a further technical solution, determining whether to retain the hyperedge includes:

[0032] Calculate the total cost of creating a hyperedge, and calculate the direct storage cost of not creating a hyperedge;

[0033] If the total overhead is less than or equal to the direct storage overhead, then the hyperedge is retained;

[0034] If the total overhead is greater than the direct storage overhead, the hyperedge is not retained.

[0035] As a further technical solution, the expression for updating the node mapping relationship is:

[0036]

[0037] in, Indicates the new supernode. Both represent supernodes. Represents a set of supernode pairs. Indicates from Delete and , This indicates an update to the assignment relationship.

[0038] According to one aspect of this specification, a network representation system based on a multidimensional heterogeneous resource summary view is provided, comprising:

[0039] The first processing module is used to collect information on network topology, devices and link status, and to build a global original view of multi-dimensional heterogeneous resources.

[0040] The second processing module is used to initialize each node and each link in the global original view as a super node and a super edge, respectively, and to construct an initial summary view based on all super nodes and super edges;

[0041] The third processing module is used to generate a set of candidate supernode pairs based on the initial summary view and a wedge sampling-based candidate generation strategy.

[0042] The fourth processing module is used to filter the set of candidate supernode pairs based on a preset attribute purity constraint, and obtain candidate supernode pairs whose attribute purity is greater than or equal to a set attribute purity threshold; the new supernode is the supernode formed by merging two supernodes in the supernode pair.

[0043] The fifth processing module is used to calculate the combined comprehensive benefit of each candidate supernode pair after screening; the combined comprehensive benefit is a weighted sum of storage compression benefit and normalized path quality distortion penalty;

[0044] The sixth processing module is used to merge the two supernodes in the candidate supernode pair with the highest overall benefit, update the node mapping relationship, aggregate the superedge attributes according to the number of nodes, and determine whether the superedge should be retained. Iteratively execute the third processing module to the sixth processing module until the number of supernodes in the summary view is reduced to the preset target number.

[0045] According to one aspect of the present invention, an electronic device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the network representation method based on a multidimensional heterogeneous resource summary view.

[0046] Compared with the prior art, the beneficial effects of the present invention are:

[0047] 1. This invention constructs a multi-dimensional heterogeneous resource summary view, compressing the original large-scale network topology into a compact structure composed of a finite number of supernodes, thus achieving efficient representation of network structure and resource status. During the summary construction process, a superedge retention mechanism based on the minimum description length principle is introduced, retaining superedges only when their introduction reduces overall storage overhead. This avoids generating redundant or low-information-content connections, effectively reducing the storage size of the summary view and improving the efficiency of network status representation.

[0048] 2. In the process of constructing the summary view, this invention considers both node attributes and link attributes. By introducing supernode attribute purity constraints and a distortion control mechanism based on path quality differences, the merged supernodes maintain high consistency in multidimensional resource features. This effectively reduces the expression deviation of the original network structure and attribute information in the summarization process, and improves the approximation accuracy and structure preservation capability of the summary view of the original network.

[0049] 3. This invention can uniformly model multi-dimensional heterogeneous resources such as bandwidth, latency, and reliability, and retain multi-dimensional attribute information in the same summary view. Based on this feature, the weights of each dimension can be flexibly adjusted according to different application requirements, realizing differentiated analysis and optimization for different business scenarios, and improving applicability and flexibility in complex network environments. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 A flowchart of a network representation method based on a multidimensional heterogeneous resource summary view provided in an embodiment of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] like Figure 1 As shown in the figure, this invention proposes a network representation method based on a multidimensional heterogeneous resource summary view, including: Step 1, collecting network topology, device and link status information to construct a global original view of multidimensional heterogeneous resources (including node attributes and link multidimensional attributes); Step 2, initializing each node and each link in the global original view as an independent supernode and superedge, respectively, to form an initial summary view; Step 3, using a wedge sampling-based candidate generation strategy, generating a set of candidate supernode pairs for merging from the supernodes of the current summary view; Step 4, based on preset attribute purity constraints... Step 5: For each selected candidate supernode pair, the combined merging benefit is calculated. The combined merging benefit is the weighted sum of storage compression benefit and normalized path quality distortion penalty. Step 6: The candidate supernode pair with the highest combined merging benefit is selected for merging. The node mapping relationship is updated, the superedge attributes are aggregated according to the number of nodes, and it is determined whether the superedge is retained. Steps 3 to 6 are iteratively executed until the number of supernodes in the summary view is reduced to the preset target number.

[0054] Step 1: By periodically collecting network topology information, device information, and link status information, construct a multi-dimensional global original view of heterogeneous resources, denoted as a graph. ,in, A set of nodes represents heterogeneous forwarding devices in the network, with each node... It has one node attribute, denoted as This invention uses functional support levels as the basis for classifying heterogeneous forwarding devices, categorizing them into three types: high-performance core switches, standard aggregation switches, and lightweight access devices. Each node... The node attributes are taken from a discrete set of attributes. In this embodiment, These correspond to the three device types mentioned above. E is the set of edges, representing the physical links between devices, i.e. , and All represent nodes. This invention focuses on the comprehensive impact of edge attributes—remaining bandwidth, latency, and reliability—on path quality, with each link associated with a three-dimensional feature vector. ,in This indicates the available bandwidth characteristic of the edge. This indicates the time delay characteristic of the edge. This indicates the reliability characteristics of the edge. A represents the graph. The adjacency matrix, if the nodes With nodes If there is a link between them, then ;otherwise .

[0055] Step 2: Set the global original view Each node in Initialize as an independent supernode That is, the set of supernodes at this time Number of supernodes At the same time, each link in the global original view will be... Initialize with a corresponding hyperedge and construct an initial summary view. During the initialization phase, each original node corresponds to an independent supernode, which is a one-to-one relationship. As the nodes are iterated and merged, multiple original nodes can be attributed to the same supernode, meaning that the mapping from original nodes to supernodes is many-to-one.

[0056] Step 3: To quickly filter out candidate pairs with potentially high merging potential from a massive number of node pairs, a wedge-based candidate generation strategy is adopted. The wedge number of each supernode is calculated, and the sampling budget is allocated proportionally. Through bidirectional mutual selection verification, candidate supernode pairs with high structural similarity are selected, significantly reducing the size of the candidate supernode pair set. In an undirected graph, if two supernodes... and Sharing a large number of common neighbors (i.e., constituting a large number of...) wedge structure centered This indicates that their network locations and connection patterns are highly similar, making them potential high-quality merge targets.

[0057] Specifically, embodiments of the present invention provide the operation steps of a candidate generation strategy based on wedge sampling:

[0058] Step 3.1, Calculate the current summary view Each supernode degree And calculate the wedge number centered on it, i.e., the wedge number of the central supernode. Total number of wedges in the entire map ;in The calculation of combinations, from If you choose any two options without considering their order, how many different ways are there to do so?

[0059] Step 3.2, according to the preset sampling budget (Sampling budget refers to the maximum allowed total number of samples, for example) The number of samples is allocated according to the proportion of the wedge number of each central supernode, as expressed in the following expression:

[0060] (1)

[0061] in, Indicates that the integer part is taken, and W represents the total number of wedges in the entire graph.

[0062] Step 3.3, for each central supernode Extracted from its neighbors' concentrated no-replacement sites An unordered pair And accumulate the number of times the pair appears. .

[0063] Step 3.4: Perform Top-R filtering on the candidate lists of each supernode and execute two-way mutual selection verification, only when... Simultaneously appearing and Only in the Top-R list will Include in the set of candidate supernode pairs The Top-R filtering refers to determining the candidate merge objects for each supernode based on the occurrence frequency, sorting the candidate merge objects from high to low according to their occurrence frequency, and retaining the top R candidate merge objects.

[0064] Step 4: To ensure the homogeneity of supernodes in the summary view, this embodiment of the invention constrains the attribute purity of supernodes. For any supernode... It contains one or more raw nodes, each belonging to a different device type. Assume there are a total of device types in the network. species (in this embodiment) (i.e., high-performance core switches, standard aggregation switches, and lightweight access devices) are used to count the supernode. Belongs to the first The number of nodes of each device type is denoted as Then the supernode Purity of properties Defined as the ratio of the number of nodes corresponding to the most prevalent device type within the supernode to the total number of nodes within the supernode, the calculation formula is:

[0065] (2)

[0066] in, Supernode The total number of nodes included. Supernode Belongs to the first The number of nodes for each type of device =| | for supernode The total number of nodes included. This indicates taking the maximum value.

[0067] During the construction of the summary view, any supernode The property purity constraint must be satisfied: (For example This constraint ensures a high degree of consistency in device type among forwarding devices within the same supernode. For each supernode pair in the candidate supernode pair set ( ), calculate the attribute purity of the new supernode after merging two supernodes in a supernode pair. Only retain those that meet the requirements. Candidate supernode pairs, among which This is a preset attribute purity threshold. During the summary view construction process, the optimal merging object is first selected from the candidate supernode pairs that satisfy the preset attribute purity constraint. If no candidate supernode pair satisfies the constraint, the attribute purity constraint is gradually relaxed to ensure the summary process can continue (for example, decreasing by 0.01 each time, and then checking if there is a candidate supernode pair that satisfies the attribute purity constraint; if not, it continues to decrease by 0.01 until there is a candidate supernode pair that satisfies the attribute purity constraint; relaxing the attribute purity constraint is to ensure the merging process can continue normally) in order to obtain a feasible merging scheme.

[0068] Step 5: For the filtered set of candidate supernode pairs Each supernode pair in The overall benefit of merging them into a new supernode is calculated, taking into account both storage compression benefits and path quality distortion. The specific operation is as follows:

[0069] Step 5.1, Calculate storage compression benefits .

[0070] This invention evaluates supernodes based on the Minimum Description Length (MDL) principle. and The change in the total number of bits required to describe the network state before and after the merge. Since the merge operation only affects... or The associated local neighborhood allows for efficient measurement of the storage compression benefits of the merge operation simply by calculating the storage overhead changes within that local neighborhood. For the supernode pairs to be merged... Define its set of associated supernode pairs For all that contain at least or One of the sets of supernode pairs (including self-loop pairs), namely:

[0071] (3)

[0072] in, and These are respectively represented in the summary view. China and and The set of neighboring supernodes connected by a hyperedge. If there is no current hyperedge between two supernodes but a physical link exists, the storage overhead is calculated directly based on the number of physical links and is not included in the overhead calculation for models with hyperedges.

[0073] For sets Each supernode pair in { }, perform the following steps:

[0074] 1. Get the supernode Number of original nodes included Supernode Number of original nodes included

[0075] 2. Calculate the theoretical maximum number of edges for this supernode. ,when hour, ;when hour, .

[0076] 3. Obtain the number of actual physical links for this supernode. ;

[0077] 4. Calculate the edge density parameters ,when When =0, set .

[0078] Under the MDL framework, the storage overhead required for this supernode pair consists of two parts: model overhead and data overhead.

[0079] If a hyperedge is created for this pair of supernodes, the model cost is... and data overhead They are respectively:

[0080] (4)

[0081] (5)

[0082] Among them, | | represents the total number of supernodes in the current summary view. This represents the maximum value of the actual number of links corresponding to all current hyperedges. This refers to the quantization precision of edge attribute values.

[0083] If a hyperedge is not created for this hypernode pair, the model cost will be... The overhead is 0. The data overhead is the number of bits required to directly store each link of the supernode to all its internal physical links:

[0084] (6)

[0085] in, This represents the total number of nodes in the original network.

[0086] In summary, a single supernode pair { The storage overhead of} is the smaller of the two options:

[0087] (7)

[0088] Before the merger, with Associated local storage overhead For set The sum of storage overhead for all supernode pairs in the network:

[0089] (8)

[0090] After the merger, the new supernode This process requires rebuilding its relationships with its neighbors. (Note: The last sentence appears to be incomplete and possibly contains errors. It's unclear what "merged" means.) The relevant set of supernode pairs is:

[0091] (9)

[0092] in, Indicates the new supernode The neighborhood group, Indicates the new supernode The neighbor set excludes those that are merged. , itself.

[0093] For each Its actual number of links Calculate the storage overhead according to the aforementioned aggregation rules and formulas, and then calculate the new local storage overhead after merging. for:

[0094] (10)

[0095] Merge supernode pairs The resulting storage compression benefits are defined as:

[0096] (11)

[0097] in, A higher value indicates a more significant benefit of the merge in terms of storage compression.

[0098] Step 5.2, calculate the normalized path quality distortion penalty. .

[0099] To control path quality distortion caused by node merging during the summarization process, this embodiment of the invention defines a local maximum path quality difference, denoted as . This metric is used to evaluate the supernode. and The maximum path quality deviation caused within the local neighborhood when merging into a new supernode. Its calculation formula is as follows:

[0100] (12)

[0101] in, and These represent summary views. Mid and Super Nodes and supernode The set of adjacent supernodes. The first term of the formula measures the distance to each neighbor before and after the merge. The path quality difference; the latter three measures the quality difference of self-loops and mutual paths before and after the merger. The smaller the value, the less the disturbance to the local and even global path quality caused by merging the node.

[0102] In formula (12), In summary view In the middle, node and Between no more than The optimal synthesized path quality under jump constraints. This invention defines its specific form as follows:

[0103] (13)

[0104] in, Represents a node and The path between them , , This represents the weighting coefficients corresponding to the three dimensions of bandwidth (bw), delay (delay), and reliability (rel), and satisfies the following conditions: + + =1. , , Paths The quality function in three dimensions is defined as follows:

[0105] (14)

[0106] (15)

[0107] (16)

[0108] in, This represents the bandwidth dimension, specifically the bottleneck bandwidth of the path, reflecting the end-to-end capacity to handle maximum traffic. This represents the latency dimension, which is the reciprocal of the cumulative path latency. The smaller the latency, the larger this value. The reliability dimension represents the product of the reliability of each link segment in the path, reflecting the probability of successful end-to-end transmission. It should be noted that before calculating the weighted sum, the path quality function of each dimension needs to be normalized and mapped to the same order of magnitude (e.g., [0,1]) to avoid the value of one dimension dominating the weighted result due to differences in units.

[0109] Will Normalization on the set of candidate supernode pairs yields a normalized path quality distortion penalty. :

[0110] (17)

[0111] in, It is a preset, extremely small positive number used to avoid division by zero errors in normalization calculations.

[0112] Step 5.3, calculate the consolidated comprehensive income.

[0113] For candidate supernode pairs that satisfy the attribute purity constraint, calculate the supernode... The combined comprehensive income is expressed as follows:

[0114] (18)

[0115] in, Indicates the benefits of storage compression. This represents the penalty for normalized path quality distortion. (For example ) is a weighting parameter used to weigh storage efficiency against path quality. When When the value is large, the supernode merge selection tends to choose merge pairs with higher compression benefits; when When the value is small, the supernode merging selection tends to choose merging pairs that have less disturbance to path quality.

[0116] Step 6: Select the pair of supernodes with the highest overall benefit and merge them to create a new supernode. The process involves updating node mappings, weighting and aggregating hyperedge attributes by node count, determining which hyperedges to retain based on MDL principles, and iterating until the summary view size is reduced to a preset target value. In each iteration, the set of candidate hypernode pairs is... Select the pair of supernodes that satisfy the attribute purity constraint and have the highest merge score. Perform the merge. Update the summary view by following these steps. Structure and attribute information:

[0117] Step 6.1, Merging supernodes and updating node mapping.

[0118] Create a new supernode Update the set of supernode pairs in the current summary view to:

[0119] (19)

[0120] in, Indicates from Delete and , This indicates an update to the assignment relationship.

[0121] At the same time, maintain a mapping relationship from the original node to the supernode in the summary view. For all those belonging to or original node Update its mapping to .

[0122] Step 6.2, aggregation and reconstruction of hyperedge information.

[0123] The merge operation alters the set of supernodes, requiring a reconstruction of the set. All relevant superedges and their statistics. Record the supernodes in the initial summary view. and The hyperedge between them is { , The number of actual physical links contained within it is} Similarly, define , , as well as .

[0124] For new supernodes With supernode hyperedge between { , The corresponding actual number of links is updated as follows:

[0125] (20)

[0126] For new supernodes Self-loop { , The actual number of links after merging is updated as follows:

[0127] (twenty one)

[0128] Step 6.3, aggregation calculation of hyperedge attributes.

[0129] To ensure the fidelity of the summary view for path quality queries, the path quality representation values ​​of the merged hyperedges need to be recalculated. In this embodiment of the invention, the new hypernode... With any neighbor supernode The hyperedge between them stores a three-dimensional path quality vector, denoted as:

[0130] (twenty two)

[0131] in, These represent the optimal single-dimensional path quality between two supernodes in terms of bandwidth, latency, and reliability, respectively, and their calculation methods are the same as described above. , , Consistent with the logic, in Each element independently takes its optimal value under the jump constraint.

[0132] For new supernodes With any neighbor supernode The 3D path quality vectors of the hyperedges between nodes are aggregated by a weighted average based on the number of nodes:

[0133] (twenty three)

[0134] Among them, supernode The number of original nodes included is Supernode The number of original nodes included is The aggregation operation is performed independently for each component of the vector, that is:

[0135] (twenty four)

[0136] (25)

[0137] (26)

[0138] For new supernodes The self-loop hyperedge, the 3D path mass vector of the self-loop is defined as:

[0139] (27)

[0140] The weighted average above is also applied independently to each component. The physical meaning of this formula is: the new supernode after merging... The expected path quality of any pair of original nodes in each dimension is equal to that before merging. , Each of their internal original node pairs and and The dimensional path quality of the original node pairs across supernodes is obtained by weighting the number of corresponding node pairs as weights.

[0141] Step 6.4: Determine whether to retain the super-edge.

[0142] The hyperedges obtained from aggregation are not retained unconditionally. This invention follows the principle of minimum description length, retaining a hyperedge in the summary view only if creating a hyperedge reduces the overall storage overhead.

[0143] Specifically, for each candidate hyperedge pair after aggregation { , (Including the case of self-loops), calculate the total cost of establishing the hyperedge according to the formulas described above. And the direct storage overhead of not building hyperedges ,like If a hyperedge is selected, it is retained in the summary view; otherwise, it is not retained, and its corresponding physical link information is encoded and stored separately. This judgment process is performed on the affected hypernode pairs after each round of merging to continuously maintain the optimal description efficiency of the summary view for a given size. The direct storage overhead refers to the overhead of directly encoding and storing the physical link information when no hyperedge is established.

[0144] Specifically, in this embodiment of the invention, steps 3 to 6 above are repeated (generating candidate supernode pairs, filtering supernode pairs that satisfy attribute purity constraints, calculating the comprehensive benefit of merging candidate supernode pairs, merging the supernode pairs with the highest benefits and updating the summary view), until the number of supernodes in the summary view is [number missing]. Reduce to target number (For example The final summary view The three-dimensional path quality directions of each hyperedge have been weighted and aggregated as described above, fully preserving the comprehensive path quality characteristics of the original network in terms of bandwidth, latency, and reliability.

[0145] The implementation of the various embodiments of the present invention is based on programmed processing through a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide a network representation system based on a multidimensional heterogeneous resource summary view. This system is used to execute a network representation method based on a multidimensional heterogeneous resource summary view from the above method embodiments.

[0146] The system includes: a first processing module for collecting information on network topology, devices, and link status, and constructing a global original view of multi-dimensional heterogeneous resources; a second processing module for initializing each node and link in the global original view as a supernode and a superedge, respectively, and constructing an initial summary view based on all supernodes and superedges; a third processing module for generating a set of candidate supernode pairs based on the initial summary view using a wedge sampling-based candidate generation strategy; a fourth processing module for filtering the set of candidate supernode pairs based on a preset attribute purity constraint, obtaining candidate supernode pairs whose merged supernode attribute purity meets a set threshold; a fifth processing module for calculating the merging comprehensive benefit of each candidate supernode pair obtained after filtering; the merging comprehensive benefit is a weighted sum of storage compression benefit and normalized path quality distortion penalty; and a sixth processing module for merging the candidate supernode pairs with the highest merging comprehensive benefit, updating node mapping relationships, aggregating superedge attributes, and determining whether to retain superedges, iteratively executing the third to sixth processing modules until the number of supernodes in the summary view is reduced to a preset target number.

[0147] This invention provides a network representation system based on a multidimensional heterogeneous resource summary view. Addressing the shortcomings of existing technologies that only model node attributes, lacking a unified capacity to characterize multidimensional heterogeneous resources, and particularly failing to effectively integrate the relationships between link and node attributes, resulting in insufficient accuracy in representing the original network structure and attribute features, and unsuitability for efficient analysis and processing in large-scale complex network environments, this invention offers a network representation method based on a multidimensional heterogeneous resource summary view. This method acquires information on network topology, devices, and link states, constructs a global original view, and initializes each node and link as a supernode and superedge, respectively, forming an initial summary view. Candidate supernode pairs are generated, and the combined merging benefit, balancing storage compression gains and path quality distortion, is calculated. Under attribute purity constraints, supernode merging is iteratively performed until the summary view size meets a preset target. This invention reduces the network state size while maintaining effective representation of multidimensional resource features and possesses good dynamic adaptability, making it suitable for efficient modeling and analysis in large-scale complex network environments.

[0148] Based on the same inventive concept as the foregoing embodiments, this embodiment of the invention also provides an electronic device, including a memory and a processor. The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize a network representation method based on a multidimensional heterogeneous resource summary view as proposed in the above embodiments.

[0149] Finally, it should be noted that the above specific embodiments are merely representative examples of the present invention. Obviously, the present invention is not limited to the above specific embodiments and many variations are possible. Any simple modifications, equivalent changes, and alterations made to the above specific embodiments based on the technical essence of the present invention should be considered within the protection scope of the present invention.

Claims

1. A network representation method based on a multidimensional heterogeneous resource summary view, characterized in that, include: Step 1: Collect information on network topology, devices, and link status to construct a global original view of multidimensional heterogeneous resources; Step 2: Initialize each node and each link in the global original view as a super node and a super edge, respectively, and construct an initial summary view based on all super nodes and super edges; Step 3: Based on the initial summary view, a candidate supernode pair set is generated using a wedge sampling-based candidate generation strategy; Step 3 includes: Calculate the degree of each supernode in the initial summary view, and calculate the wedge number of the central supernode and the total wedge number of the entire graph based on the degree; Based on the preset sampling budget, the number of samples is allocated according to the proportion of the wedge number of each central supernode to the total number of wedges in the entire graph. The number of samples is denoted as [sample number]. ; Based on the neighbor set of each central supernode, extract Each pair of supernodes is counted, and the occurrence count of each pair of supernodes is accumulated. Based on the occurrence count, candidate merge objects corresponding to each supernode are determined, and the candidate merge objects are sorted and filtered according to the occurrence count from high to low. Based on the candidate merged objects after sorting and filtering, a two-way mutual selection verification is performed, and the supernode pairs that pass the two-way mutual selection verification are included in the candidate supernode pair set. Step 4: Based on the preset attribute purity constraint, the candidate supernode pair set is filtered to obtain candidate supernode pairs whose attribute purity is greater than or equal to the set attribute purity threshold; the new supernode is the supernode formed by merging the two supernodes in the supernode pair. Step 5: Calculate the combined benefit of each candidate supernode pair after screening; the combined benefit is the weighted sum of storage compression benefit and normalized path quality distortion penalty; Step 6: Merge the two supernodes in the candidate supernode pair with the highest overall benefit, update the node mapping relationship, aggregate superedge attributes based on the number of nodes, and determine whether to retain the superedge. Iterate through steps 3 to 6 until the number of supernodes in the summary view is reduced to the preset target number; aggregate superedge attributes based on the number of nodes, including: Based on any neighboring supernode of the merged new supernode, obtain the three-dimensional path quality vectors between the two merged supernodes and their neighboring supernodes before the merger. The number of original nodes contained in each of the two merged supernodes is used as the weight to perform a weighted average on the two 3D path quality vectors, resulting in the 3D path quality vector between the new supernode and its neighboring supernodes.

2. The network representation method based on a multidimensional heterogeneous resource summary view according to claim 1, characterized in that, The purity of the attribute is defined as the ratio of the number of nodes corresponding to the device type with the highest proportion within the supernode to the total number of nodes within the supernode.

3. The network representation method based on a multidimensional heterogeneous resource summary view according to claim 1, characterized in that, The expression for the storage compression benefit is: ， in, Indicates the benefits of storage compression. Indicates a pair of supernodes. Indicates the new local storage overhead. This indicates the local storage overhead.

4. The network representation method based on a multidimensional heterogeneous resource summary view according to claim 1, characterized in that, The expression for the normalized path quality distortion penalty is: ， in, This represents the penalty for normalized path quality distortion. This indicates the local maximum path quality difference. Indicates a pair of supernodes. Represents the set of candidate supernode pairs. It represents the smallest positive number that is preset.

5. A network representation method based on a multidimensional heterogeneous resource summary view according to claim 1, characterized in that, Determining whether to retain the superedge includes: Calculate the total cost of creating a hyperedge, and calculate the direct storage cost of not creating a hyperedge; If the total overhead is less than or equal to the direct storage overhead, then the hyperedge is retained; If the total overhead is greater than the direct storage overhead, the hyperedge is not retained.

6. The network representation method based on a multidimensional heterogeneous resource summary view according to claim 1, characterized in that, The expression for updating the node mapping relationship is: ， in, Indicates the new supernode. Both represent supernodes. Represents a set of supernode pairs. Indicates from Delete and , This indicates an update to the assignment relationship.

7. A network representation system based on a multidimensional heterogeneous resource summary view, characterized in that, A network representation method based on a multidimensional heterogeneous resource summary view as described in any one of claims 1-6 includes: The first processing module is used to collect information on network topology, devices and link status, and to build a global original view of multi-dimensional heterogeneous resources. The second processing module is used to initialize each node and each link in the global original view as a super node and a super edge, respectively, and to construct an initial summary view based on all super nodes and super edges; The third processing module is used to generate a set of candidate supernode pairs based on the initial summary view and a wedge sampling-based candidate generation strategy. The fourth processing module is used to filter the set of candidate supernode pairs based on a preset attribute purity constraint, and obtain candidate supernode pairs whose attribute purity is greater than or equal to a set attribute purity threshold; the new supernode is the supernode formed by merging two supernodes in the supernode pair. The fifth processing module is used to calculate the combined benefit of each candidate supernode pair after screening; the combined benefit is a weighted sum of storage compression benefit and normalized path quality distortion penalty; The sixth processing module is used to merge the two supernodes in the candidate supernode pair with the highest overall benefit, update the node mapping relationship, aggregate the superedge attributes according to the number of nodes, and determine whether the superedge should be retained. Iteratively execute the third processing module to the sixth processing module until the number of supernodes in the summary view is reduced to the preset target number.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the network representation method based on a multidimensional heterogeneous resource summary view as described in any one of claims 1 to 6.

Images