Industrial chain key hub identification method based on social network analysis

Abstract

The invention relates to the technical field of industrial chain analysis, and discloses an industrial chain key hub identification method based on social network analysis, comprising the following steps: collecting multi-source heterogeneous data and constructing an industrial chain node incidence matrix, the data including enterprise industrial and commercial information, transaction records, patent cooperation data and social interaction records; based on a directed weighted heterogeneous network model, mapping industrial chain nodes into network vertexes, mapping association among the nodes into directed edges with weights, and constructing an industrial chain network; calculating a node global influence score by using an improved PageRank algorithm, and identifying a community core node in combination with a Louvain community discovery algorithm; and fusing the global influence score and a community core node result, and determining an industrial chain key hub. The industrial chain key hub identification method based on social network analysis aims to accurately identify key hub nodes by constructing an industrial chain network model and combining a network topology structure and dynamic interaction data.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial chain analysis, and specifically to a method for identifying key hubs in an industrial chain based on social network analysis. Background Art

[0002] In the existing technology system, key hubs in the industrial chain play an extremely important role. Such key hubs usually refer to nodes that occupy a special position in the industrial chain network, such as midstream platform companies that are like bridges across communities. They are also closely connected to multiple upstream suppliers and downstream distributors. From the perspective of collaborative operation of the industrial chain, these key hubs can efficiently integrate upstream and downstream resources of the industrial chain, greatly improving the operating efficiency of the entire industrial chain. In supply chain management, they can accurately allocate resources based on upstream and downstream supply and demand conditions, reduce inventory costs, and ensure smooth flow of materials; in terms of the diffusion of technological innovation, hub companies can quickly transmit new technologies to all links in the industrial chain, accelerate the overall technological upgrading process of the industry, and thereby promote the development and transformation of the entire industrial chain.

[0003] However, existing technologies have significant flaws and shortcomings in their understanding and analysis of key hubs in industrial chains. In most cases, existing methods simply simplify the industrial chain into a linear process of "production-circulation-consumption" or view it as a collection of isolated nodes, completely ignoring the fact that the industrial chain is actually a complex and dynamic network structure. Taking talent evaluation models as an example, traditional models focus solely on individual capability indicators while failing to consider the role of the individual's enterprise in industrial chain collaboration. This prevents talent evaluation from being closely integrated with the overall needs of the industrial chain. In customized production solutions, they also focus solely on the one-way "brand-user" link, ignoring key network connections such as technological cooperation and supply chain sharing between enterprises. This linear thinking makes it difficult to accurately capture the important role of key hub nodes in the network. Furthermore, existing technologies lack quantitative analysis of topological characteristics such as node degree and betweenness centrality. Node degree intuitively reflects the number of connections a node has with other nodes, while betweenness centrality reflects the node's control over information dissemination and resource circulation. Without quantitative analysis of these key topological characteristics, it is difficult to gain a deep understanding of the core position of key hub nodes in the industrial chain network and to predict the potential crippling impact of their failure on the local or even global industrial chain. Summary of the Invention

[0004] The purpose of this invention is to propose an industrial chain key hub identification method based on social network analysis by constructing an industrial chain network model, combining network topology structure and dynamic interaction data, and accurately identifying key hub nodes.

[0005] The technical solution of the present invention to solve the above technical problems is as follows:

[0006] The method for identifying key hubs in the industrial chain based on social network analysis includes the following steps:

[0007] S10. Collect multi-source heterogeneous data and construct an industry chain node association matrix, wherein the data includes business information of enterprises, transaction records, patent cooperation data, and social interaction records;

[0008] S20. Based on the directed weighted heterogeneous network model, the industry chain nodes are mapped as network vertices, and the associations between nodes are mapped as weighted directed edges to construct the industry chain network;

[0009] S30. Use the improved PageRank algorithm to calculate the global influence score of the node and combine it with the Louvain community discovery algorithm to identify the core nodes of the community;

[0010] S40: Integrate the global influence score and the community core node results to determine the key hubs of the industrial chain;

[0011] S50. Establish a dynamic verification mechanism to periodically update network data and verify the stability of hub nodes.

[0012] On the basis of the above technical solution, the present invention can also be improved as follows.

[0013] Furthermore, the method for processing multi-source heterogeneous data in step S10 includes:

[0014] The IQR algorithm is used to remove outliers in transaction records. The formula is:

[0015] Lower bound = Q1 - 1.5 × IQR, upper bound = Q3 + 1.5 × IQR

[0016] Among them, Q1 and Q3 are quartiles, IQR = Q3-Q1;

[0017] Use entropy weight method to determine transaction frequency f ij 、Number of cooperation c ij , social interaction times ij The weight coefficients w1, w2, w3 are as follows:

[0018]

[0019] Among them, a ij =w1f ij +w2c ij +w3s ij ;

[0020] Construct the adjacency matrix A = [a ij ] n×n , where a ij =w1f ij +w2cij +w3s ij .

[0021] Furthermore, the specific steps of constructing a directed weighted heterogeneous network in step S20 include:

[0022] Define the node type set T = {upstream, midstream, downstream}, for each node v i Assign type tag t i ∈T;

[0023] Introduce the type constraint matrix C = [c kl ] 3×3 , where c kl Represents the association weight correction coefficient of type k node to type l node, satisfying c 上游，中游 ≥1.2c 下游，中游 ;

[0024] Modify the type constraints of the adjacency matrix: And normalize to get the transition probability matrix P = [p ij ] n×n ,in

[0025] Furthermore, the execution process of the improved PageRank algorithm in step S30 includes:

[0026] Initialize influence vector Among them 1 n is an n-dimensional all-1 vector;

[0027] The iterative update formula is: (t+1) =αRr (t) +(1-α)b, where the damping coefficient α = 0.85 and the bias vector b assigns the upstream node Other nodes are assigned

[0028] When || r (t+1) -r (t) ||2<10 -5 The iteration is terminated when , and the node influence score vector r is obtained.

[0029] Furthermore, the optimization method of the Louvain community discovery algorithm in step S30 includes:

[0030] Define a modularity function with type constraints:

[0031] Among them, c i is the node community label, δ(t i ,t j) is the same type indicator function, where the value is 1 if the type is the same, otherwise it is 0;

[0032] Adopt a hierarchical merging strategy: merge nodes of the same type first, then merge nodes of different types. Calculate the modularity increment Q during merging, and only retain merge operations with ΔQ > 0.

[0033] The core nodes in the community are identified using weighted degree centrality: Take C′ in each community D (v i )The largest node serves as the community hub.

[0034] Furthermore, the fusion strategy of the key hub in step S40 includes:

[0035] Filter the top 20% of nodes in influence score as the global hub set H G , select the top 10% nodes of each community’s weighted degree centrality as the community hub set H C ;

[0036] Calculate the comprehensive score of candidate hubs:

[0037] Arrange in descending order of comprehensive scores and select the top 10% of nodes as the key hub set H of the industrial chain * .

[0038] Furthermore, the dynamic verification mechanism in step S50 includes:

[0039] Collect the latest data every quarter, recalculate the adjacency matrix A, and update the type constraint matrix C;

[0040] The Jaccard coefficient is used to verify the stability of the hub set: J is required to be ≥ 0.75, otherwise algorithm parameter tuning will be triggered;

[0041] Combined with industry reports, the actual influence of hub nodes is verified. Verification indicators include market share and supply chain response speed, and the accuracy rate must reach more than 80%.

[0042] Furthermore, the method for quantifying the social interaction data in step S10 includes:

[0043] Crawl the interactive texts of corporate executives on industry platforms and extract keyword weights using the TextRank algorithm: in,

[0044] w k As the keyword, Co-occur(w k ,v i ,v j ) is the keyword in v iWith v j the number of co-occurrences in the interaction;

[0045] Normalize the text interaction weight to the interval [0,1] as the number of social interactions s ij quantized value of .

[0046] Furthermore, the edge weight update rules of the directed weighted heterogeneous network include:

[0047] For transaction frequency f ij Using exponential decay model: f′ ij =f ij ×e -λΔt , where λ = 0.05 / month, Δt is the number of months since the transaction;

[0048] Applying technology similarity weighting to patent cooperation data:

[0049] Among them, p i 、p j Node v i 、v j A collection of patent technology fields.

[0050] Furthermore, the termination conditions of the Louvain algorithm include:

[0051] The number of communities remains stable to Between, n is the total number of nodes;

[0052] The modularity increment is less than 0.01 for three consecutive iterations;

[0053] The proportion of cross-type communities is less than 15%. Otherwise, the cross-type communities are split. The splitting rules are as follows:

[0054] Split(C)={C t |t∈T,C t ={v i ∈C|t i =t}}

[0055] Among them, C is the cross-type community C t Subcommunities after splitting by type

[0056] Compared with the prior art, the technical solution of this application has the following beneficial technical effects:

[0057] The present invention collects multi-source heterogeneous data and constructs an industrial chain node association matrix. Through various data such as enterprise business information and transaction records, it comprehensively captures the complex connections between enterprises and restores the dynamic and complex network structure of the industrial chain. Based on a directed weighted heterogeneous network model, the industrial chain is constructed, and weights and directional attributes are assigned to the industrial chain. This accurately reflects the strength and direction of relationships between nodes and more realistically reflects the actual circulation of resources and information in the industrial chain. The improved PageRank algorithm is used to calculate the global influence score of nodes and the Louvain community discovery algorithm is used to identify community core nodes. This not only quantifies the influence of nodes in the entire industrial chain network, but also identifies core nodes within each community, accurately capturing the key role of "hub nodes" in the industrial chain network. The global influence score and community core node results are integrated to determine the key hubs of the industrial chain. Comprehensively considering both global and local dimensions, the key hubs of the industrial chain can be more comprehensively and accurately located. A dynamic verification mechanism is established to periodically update network data and verify the stability of hub nodes. This allows the identification of key hubs to adapt to the dynamic changes of the industrial chain, continuously ensuring the accuracy of key hub identification and providing strong support for the stable operation and risk prevention of the industrial chain. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a flow chart of the method for identifying key hubs in an industrial chain based on social network analysis according to the present invention. DETAILED DESCRIPTION

[0059] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0060] The method for identifying key hubs in an industrial chain based on social network analysis of the present invention comprises the following steps:

[0061] S10. Collect multi-source heterogeneous data and construct an industry chain node association matrix. The data includes business information, transaction records, patent cooperation data, and social interaction records.

[0062] S20. Based on the directed weighted heterogeneous network model, the industry chain nodes are mapped as network vertices, and the associations between nodes are mapped as weighted directed edges to construct the industry chain network;

[0063] S30. Use the improved PageRank algorithm to calculate the global influence score of the node and combine it with the Louvain community discovery algorithm to identify the core nodes of the community;

[0064] S40: Integrate the global influence score and the community core node results to determine the key hubs of the industrial chain;

[0065] S50. Establish a dynamic verification mechanism to periodically update network data and verify the stability of hub nodes.

[0066] In a preferred embodiment, the present invention can be further configured as follows: the method for processing multi-source heterogeneous data in step S10 includes:

[0067] The IQR algorithm is used to remove outliers in transaction records. The formula is:

[0068] Lower bound = Q1 - 1.5 × IQR, upper bound = Q3 + 1.5 × IQR

[0069] Among them, Q1 and Q3 are quartiles, IQR = Q3-Q1;

[0070] Use entropy weight method to determine transaction frequency f ij 、Number of cooperation c ij , social interaction times ij The weight coefficients w1, w2, w3 are as follows:

[0071]

[0072] Among them, a ij =w1f ij +w2c ij +w3s ij ;

[0073] Construct the adjacency matrix A = [a ij ] n×n , where a ij =w1f ij +w2c ij +w3s ij The IQR algorithm removes outliers from transaction data, preventing extreme values ​​from interfering with the model and improving data reliability. The entropy weight method objectively calculates the weight coefficients of transaction frequency, number of collaborations, and social interactions, breaking through the limitations of existing technologies that rely on subjective assignments and achieving scientific integration of multi-source data. The constructed adjacency matrix integrates material flow, technology flow, and information flow, allowing the connections between industry chain nodes to more comprehensively reflect the actual collaborative relationships and laying the foundation for subsequent network modeling.

[0074] The IQR algorithm identifies outliers through the interquartile range. In specific applications, a sliding window can be set to update the threshold on a quarterly or annual basis to adapt to the dynamic characteristics of industrial chain data. j Reflects the degree of information disorder of the indicator, weight w kIt is inversely proportional to the entropy value, that is, the greater the fluctuation of the indicator (the smaller the information entropy), the greater the contribution to distinguishing node associations. For example, if the entropy value of social interaction data is lower than that of transaction data, it means that its dispersion is high and it can more sensitively reflect the implicit cooperation between enterprises, so it has a higher weight. ij It not only quantifies the strength of the association, but also records the timeliness of the data through the timestamp field, providing a basis for subsequent dynamic weight updates.

[0075] In a preferred embodiment, the present invention can be further configured as follows: the specific steps of constructing a directed weighted heterogeneous network in step S20 include:

[0076] Define the node type set T = {upstream, midstream, downstream}, for each node v i Assign type tag t i ∈T;

[0077] Introduce the type constraint matrix C = [c kl ] 3×3 , where c kl Represents the association weight correction coefficient of type k node to type l node, satisfying c 上游，中游 ≥1.2c 下游，中游 ;

[0078] Modify the type constraints of the adjacency matrix: And normalize to get the transition probability matrix P = [p ij ] n×n ,in Introduce node type labels and type constraint matrices to clearly distinguish the roles of upstream, midstream, and downstream nodes. 上游，中游 ≥1.2c 下游，中游 , strengthen the basic role of upstream supply in midstream production, make the network model more in line with the actual structure of the industrial chain "upstream supports midstream, midstream serves downstream", and avoid hub identification deviation caused by treating different types of nodes equally.

[0079] The specific values ​​of the type constraint matrix C can be set based on industry standards. For example, in the electronics manufacturing industry, the weight correction coefficient of upstream chip suppliers on midstream assembly companies can be set to 1.5, and the coefficient of downstream distributors on midstream companies can be set to 1.0, highlighting the strong dependence of the core supply chain. The normalized transition probability matrix P can be used to calculate the "type-sensitive PageRank score" of nodes. Specifically, the influence transmission path of upstream nodes is preferentially weighted, and the hub status of midstream nodes must meet both high intra-type correlation and cross-type transmission capabilities, thereby accurately identifying key nodes that combine industry attributes and network status.

[0080] In a preferred embodiment, the present invention can be further configured as follows: the execution process of the improved PageRank algorithm in step S30 includes:

[0081] Initialize influence vector Among them 1 n is an n-dimensional all-1 vector;

[0082] The iterative update formula is: (t+1) =αPr (t) +(1-α)b, where the damping coefficient α = 0.85 and the bias vector b assigns the upstream node Other nodes are assigned

[0083] When || r (t+1) -r (t) ||2<10 -5 The iteration is terminated when , resulting in a node influence score vector r. Using the bias vector b, upstream nodes are assigned a higher initial score (e.g., 2 / n). This overcomes the flaw in prior art where traditional algorithms fail to account for node heterogeneity, highlighting the influence of upstream hubs such as key raw material suppliers and core technology providers. The L2 norm convergence condition is introduced during the iteration process to ensure the stability of the score calculation and enable targeted identification of "foundational support hubs" within the industry chain.

[0084] The design principle of the bias vector is based on the risk transmission characteristics of the industrial chain - the failure of the upstream node may lead to the paralysis of the entire chain, so its importance needs to be identified in advance. For example, in the new energy industry chain, lithium mining companies as upstream nodes can have their bias value increased by an additional 0.1 to reflect their resource monopoly. The damping coefficient α=0.85 in the iterative formula balances the node's own influence and the network propagation effect. When an upstream node has a high-weight association with multiple midstream companies, its score will be adjusted by Pr (t) The items accumulate quickly, and are eventually reflected in the convergence results as influence values ​​significantly higher than those of downstream nodes.

[0085] In a preferred embodiment, the present invention can be further configured as follows: the optimization method of the Louvain community discovery algorithm in step S30 includes:

[0086] Define a modularity function with type constraints:

[0087] Among them, c i is the node community label, δ(t i ,t j ) is the same type indicator function, where the value is 1 if the type is the same, otherwise it is 0;

[0088] A hierarchical merging strategy is adopted: first merge nodes of the same type, then merge nodes of different types. When merging, the modularity increment q is calculated, and only the merge operations with ΔQ>0 are retained.

[0089] The core nodes in the community are identified using weighted degree centrality: Take C′ in each community D (v i ) serves as the community hub. A modularity function with type constraints ties community divisions to node types, prioritizing the merging of nodes of the same type. A hierarchical merging strategy prioritizes communities of the same type before evaluating cross-type associations. This avoids the chaotic community structure caused by the mixing of cross-industry nodes, ensuring that the identified community hubs better align with the actual collaboration models within industry chain segments. For example, core enterprises in a midstream manufacturing community must possess both high internal collaboration density and type consistency.

[0090] The same type indicator function δ(t i ,t j ) When calculating modularity, only when the node types are the same are the association weights taken into account to ensure the uniformity of business attributes within the community. For example, in the automotive industry chain, the midstream parts manufacturing community and the vehicle assembly community will be divided into independent communities due to their type differences, and their respective hubs are parts suppliers and vehicle manufacturers with high weighted centrality. Weighted centrality C′ D (v i ) further filters cross-type associations and only accumulates the weights of nodes of the same type, so that the community hub is more focused on core enterprises in the sub-sectors. For example, the hub of the battery manufacturing community needs to have the highest patent cooperation and production collaboration weights among enterprises of the same type.

[0091] In a preferred embodiment, the present invention can be further configured as follows: the fusion strategy of the key hub in step S40 includes:

[0092] Filter the top 20% of nodes in influence score as the global hub set H G , select the top 10% nodes of each community’s weighted degree centrality as the community hub set H C ;

[0093] Calculate the comprehensive score of candidate hubs:

[0094] Arrange in descending order of comprehensive scores and select the top 10% of nodes as the key hub set H of the industrial chain * , integrating global influence and community core nodes, avoiding the one-sidedness of a single indicator (such as relying only on scale or only on local cooperation), and the comprehensive score S (v i ) linearly combines the standardized influence score with the weighted degree centrality to identify the "global hub" with strong cross-chain radiation capabilities, while retaining the "community hub" in each sub-field, to achieve a complete portrayal of the "pyramid hub system" of the industrial chain.

[0095] The screening ratio (global top 20%, community top 10%) can be calibrated by industry cases. For example, in highly concentrated monopoly industries (such as chip manufacturing), the global hub ratio can be lowered to 10% to highlight the leading companies. The standardization of the comprehensive score adopts extreme value normalization, i.e., r i / max(r) reflects the node's influence ranking in the entire chain. The top 10% of nodes are ultimately selected as key hubs, ensuring both identification accuracy and the principle that "core nodes are a minority" in the industry chain. For example, in an industry chain consisting of 500 companies, approximately 50 hub companies are ultimately identified for key monitoring.

[0096] In a preferred embodiment, the present invention can be further configured as follows: the dynamic verification mechanism in step S50 includes:

[0097] Collect the latest data every quarter, recalculate the adjacency matrix A, and update the type constraint matrix C;

[0098] The Jaccard coefficient is used to verify the stability of the hub set: J is required to be ≥ 0.75, otherwise algorithm parameter tuning will be triggered;

[0099] The actual influence of hub nodes is verified by combining industry reports. Verification indicators include market share and supply chain response speed, with an accuracy rate of over 80%. Periodic data updates and Jaccard coefficient verification address the "model static" flaw in background technologies, ensuring that hub identification results are dynamically adjusted as the industry chain evolves. By introducing industry report verification (such as market share and supply chain response speed), the algorithm results are combined with actual industry status, avoiding the limitations of a "pure data-driven" approach and enhancing the practical guidance of hub identification.

[0100] When calculating the Jaccard coefficient, if the intersection ratio of the new and old hub sets falls below 0.75, the parameter tuning process is triggered, including recalibrating the entropy weight method weights and adjusting the type constraint matrix coefficients. Industry report verification indicators can be quantified as follows: the combined market share of hub enterprises must account for more than 60% of the entire supply chain, and the supply chain response speed (order delivery cycle) must be more than 30% faster than that of non-hub enterprises. The dynamic update cycle is set to quarterly, synchronized with corporate financial report disclosures and industry policy adjustment cycles, to ensure that the model promptly reflects industry changes. For example, after adjustments to new energy subsidy policies, the hub status of relevant enterprises can be reassessed through quarterly data updates.

[0101] In a preferred embodiment, the present invention may be further configured as follows: the method for quantifying social interaction data in step S10 includes:

[0102] Crawl the interactive texts of corporate executives on industry platforms and extract keyword weights using the TextRank algorithm: in,

[0103] w k As the keyword, Co-occur(w k ,v i ,v j ) is the keyword in v i With v j the number of co-occurrences in the interaction;

[0104] Normalize the text interaction weight to the interval [0,1] as the number of social interactions s ij The quantitative value of the interaction text is extracted through the TextRank algorithm, which converts the implicit social interaction of corporate executives into quantifiable correlation strength, making up for the defect of the background technology that "only focuses on explicit transactions and ignores information flow and trust relationships". The normalized number of social interactions s ij Combined with transaction and cooperation data, the industrial chain network model can more comprehensively reflect the complex relationship of "hard cooperation + soft connection" between enterprises.

[0105] The TextRank algorithm calculates keyword weights by constructing a word co-occurrence network. Similar to the iterative process of PageRank, the weight of each word is iteratively updated along with the weights of its neighboring words. For example, in an industry forum, if a company executive frequently mentions keywords such as "joint technical research" and "standard setting," other companies interacting with them can calculate the strength of the association based on the number of co-occurrences, reflecting potential technical cooperation intentions. Normalization uses Min-Max scaling to map text weights to [0,1], making data from different modalities (transaction amounts in the tens of thousands of yuan, social interactions in the single digits) comparable. Ultimately, the contribution of each type of association is evenly reflected in the adjacency matrix.

[0106] In a preferred embodiment, the present invention can be further configured as follows: the edge weight update rule of the directed weighted heterogeneous network includes:

[0107] For transaction frequency f ij Using exponential decay model: f′ ij =f ij ×e -λΔt , where λ = 0.05 / month, Δt is the number of months since the transaction;

[0108] Applying technology similarity weighting to patent cooperation data:

[0109] Among them, p i 、p j Node v i 、v j The patent technology field collection, the exponential decay model f′ of transaction frequency ij =fij ×e -λΔt Reduce the impact of historical data and highlight the timeliness of recent cooperation; weighted Sim(p i , p j ) Quantify the overlap of technical fields, solve the problem of "insufficient description of data timeliness and technical correlation" in background technology, and make the network model closer to the characteristics of "dynamic collaboration + technological evolution" of the industrial chain.

[0110] The half-life of the exponential decay model is about 20 months (λ = 0.05 / month), which means that the transaction weight two years ago decays to about 13.5%, which is applicable to the cooperation cycle of most traditional manufacturing industries. The Jaccard coefficient is used to calculate the technical similarity. If the intersection of the patent fields of two companies includes key technologies such as "lithium battery positive electrode materials" and "solid-state batteries", then Sim(p i , p j ) value close to 1 indicates close technical collaboration and the potential for forming an R&D hub. Edge weight updates can be scheduled as scheduled tasks. At the beginning of each month, the attenuated transaction frequency and technical similarity are automatically calculated to generate an updated adjacency matrix, ensuring that the network model reflects the collaborative enthusiasm and technological trends of the industry chain in real time.

[0111] In a preferred embodiment, the present invention can be further configured as follows: the termination condition of the Louvain algorithm includes:

[0112] The number of communities remains stable to Between, n is the total number of nodes;

[0113] The modularity increment is less than 0.01 for three consecutive iterations;

[0114] The proportion of cross-type communities is less than 15%. Otherwise, the cross-type communities are split. The splitting rules are as follows:

[0115] Split(C)={C t |t∈T,C t ={v i ∈C|t i =t}}

[0116] Among them, C is the cross-type community C t For sub-communities after splitting by type, the number of communities is limited to to This prevents communities from being too small (fragmentation) or too large (loss of meaningful segmentation), resolving the flaw of "irrational community segmentation granularity" in previous technologies. Cross-type community splitting rules enforce the division of subcommunities by type, ensuring consistent node attributes within each community. This aligns with the actual structure of the industry chain, where "upstream and downstream division of labor is clearly defined," and improves the accuracy of community hub identification.

[0117] The setting of community number threshold is based on network science theory. It is approximately the square root of the number of nodes and is applicable to most medium-sized industrial chains (e.g., when n=100, the number of communities is about 10-20). The criterion for determining that the proportion of cross-type communities is less than 15% can be achieved by statistically analyzing the type distribution of each community. For example, if the upstream nodes in a community account for 40%, the midstream nodes account for 50%, and the downstream nodes account for 10%, it is determined to be a cross-type community and a split is triggered. When splitting, the nodes are assigned to subcommunities C according to the type label. t For example, the upstream nodes in the original community are divided into "upstream supply communities", the midstream nodes are divided into "manufacturing core communities", and the downstream nodes are merged into the existing downstream communities, ensuring that the type purity of each sub-community is higher than 90%, so that the subsequent community hubs can truly represent the core collaborative capabilities of this type of node.

[0118] The overall working principle of this method is as follows: first, collect multi-source heterogeneous data such as corporate business information, transaction records, patent cooperation data, and social interaction records, remove outliers in transaction records through the IQR algorithm, and use the entropy weight method to calculate the transaction frequency f ij 、Number of cooperation c ij , social interaction times ij The weight coefficients w1, w2, w3 are used to construct the adjacency matrix A = [a ij ] n×n , where a ij =w1f ij +w2c ij +w3s ij , complete data preprocessing. Next, define the node type set T = {upstream, midstream, downstream}, assign type labels to nodes, and introduce the type constraint matrix C (satisfying c 上游，中游 ≥1.2c 下游，中游 ) Modify the adjacency matrix to obtain the transition probability matrix P and construct a directed weighted heterogeneous network. Then, calculate the global influence score of the node by improving the PageRank algorithm: Initialize the influence vector Using the iterative formula r (t+1) =αPr (t) +(1-α)b(α=0.85, the bias vector b assigns the upstream node Other nodes are assigned ) Update the score until ||r (t+1) -r (t) ||2<10 -5 When it is terminated, the influence score is r; at the same time, the Louvain community discovery algorithm with type constraints is used to find the community through the modularity function Identify communities for type-identity indicator functions and hierarchical merging strategies, and calculate weighted degree centrality Determine the core nodes of the community. Then, select the nodes with the top 20% influence scores as the global hub set H G , the top 10% nodes of each community’s weighted degree centrality are taken as the community hub set H C , through comprehensive score Fusion results, select the top 10% of nodes as the key hub set H of the industrial chain * Finally, the adjacency matrix A and the type constraint matrix C are updated quarterly using the Jaccard coefficient J≥0.75 is required to verify the stability of the hub set, and combined with industry reports, indicators include market share, supply chain response speed, etc. The accuracy rate must reach more than 80% to verify the actual influence and realize dynamic and accurate identification of key hubs in the industrial chain.

[0119] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0120] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. The method for identifying key hubs in the industrial chain based on social network analysis is characterized by: The following steps are involved: S10. Collect multi-source heterogeneous data and construct an industry chain node association matrix, wherein the data includes business information of enterprises, transaction records, patent cooperation data, and social interaction records; S20. Based on the directed weighted heterogeneous network model, the industry chain nodes are mapped as network vertices, and the associations between nodes are mapped as weighted directed edges to construct the industry chain network; S30. Use the improved PageRank algorithm to calculate the global influence score of the node and combine it with the Louvain community discovery algorithm to identify the core nodes of the community; S40: Integrate the global influence score and the community core node results to determine the key hubs of the industrial chain; S50. Establish a dynamic verification mechanism to periodically update network data and verify the stability of hub nodes.

2. The method for identifying key hubs in an industrial chain based on social network analysis according to claim 1, characterized in that: The method for processing multi-source heterogeneous data in step S10 includes: The IQR algorithm is used to remove outliers in transaction records. The formula is: Lower bound = Q1 - 1.5 × IQR, upper bound = Q3 + 1.5 × IQR Among them, Q1 and Q3 are quartiles, IQR = Q3-Q1; Use entropy weight method to determine transaction frequency f ij 、Number of cooperation c ij , social interaction times ij The weight coefficients w1, w2, w3 are as follows: Among them, a ij =w1f ij +w2c ij +w3s ij ; Construct the adjacency matrix A = [a ij ] n×n , where a ij =w1f ij +w2c ij +w3s ij .

3. The method for identifying key hubs in an industrial chain based on social network analysis according to claim 2, characterized in that: The specific steps of constructing the directed weighted heterogeneous network in step S20 include: Define the node type set T = {upstream, midstream, downstream}, for each node v i Assign type tag t i ∈T; Introduce the type constraint matrix C = [c kl ] 3×3 , where c kl Represents the association weight correction coefficient of type k node to type l node, satisfying c 上游，中游 ≥1.2c 下游，中游 ; Modify the type constraints of the adjacency matrix: And normalize to get the transition probability matrix P = [p ij ] n×n ,in 4. The method for identifying key hubs in an industrial chain based on social network analysis according to claim 3 is characterized in that: The execution process of the improved PageRank algorithm in step S30 includes: Initialize influence vector Among them 1 n is an n-dimensional all-1 vector; The iterative update formula is: (t+1) =αPr (t) +(1-α)b, where the damping coefficient α = 0.85 and the bias vector b assigns the upstream node Other nodes are assigned When || r (t+1) -r (t) ||2<10 -5 The iteration is terminated when , and the node influence score vector r is obtained.

5. The method for identifying key hubs in an industrial chain based on social network analysis according to claim 4 is characterized in that: The optimization method of the Louvain community discovery algorithm in step S30 includes: Define a modularity function with type constraints: Among them, c i is the node community label, δ(t i ,t j ) is the same type indicator function, where the value is 1 if the type is the same, otherwise it is 0; Adopt a hierarchical merging strategy: merge nodes of the same type first, then merge nodes of different types. Calculate the modularity increment Q during merging, and only retain merge operations with ΔQ > 0. The core nodes in the community are identified using weighted degree centrality: Take C′ in each community D (v i )The largest node serves as the community hub.

6. The method for identifying key hubs in an industrial chain based on social network analysis according to claim 5, characterized in that: The fusion strategy of the key hub in step S40 includes: Filter the top 20% of the nodes with the highest influence scores as the global hub set H G , select the top 10% nodes of each community’s weighted degree centrality as the community hub set H C ; Calculate the comprehensive score of candidate hubs: Arrange in descending order of comprehensive scores and select the top 10% of nodes as the key hub set H of the industrial chain * .

7. The method for identifying key hubs in an industrial chain based on social network analysis according to claim 1, characterized in that: The dynamic verification mechanism in step S50 includes: Collect the latest data every quarter, recalculate the adjacency matrix A, and update the type constraint matrix C; The Jaccard coefficient is used to verify the stability of the hub set: J is required to be ≥ 0.75, otherwise algorithm parameter tuning will be triggered; Combined with industry reports, the actual influence of hub nodes is verified. Verification indicators include market share and supply chain response speed, and the accuracy rate must reach more than 80%.

8. The method for identifying key hubs in an industrial chain based on social network analysis according to claim 1, characterized in that: The method for quantifying the social interaction data in step S10 includes: Crawl the interactive texts of corporate executives on industry platforms and extract keyword weights using the TextRank algorithm: in, w k As the keyword, Co-occur(w k ,v i ,v j ) is the keyword in v i With v j the number of co-occurrences in the interaction; Normalize the text interaction weight to the interval [0,1] as the number of social interactions s ij quantized value of .

9. The method for identifying key hubs in an industrial chain based on social network analysis according to claim 3, characterized in that: The edge weight update rules of the directed weighted heterogeneous network include: For transaction frequency f ij Using exponential decay model: f′ ij =f ij ×e -λΔt , where λ = 0.05 / month, Δt is the number of months since the transaction; Applying technology similarity weighting to patent cooperation data: Among them, p i 、p j Node v i 、v j A collection of patent technology fields.

10. The method for identifying key hubs in an industrial chain based on social network analysis according to claim 5, characterized in that: The termination conditions of the Louvain algorithm include: The number of communities remains stable to Between, n is the total number of nodes; The modularity increment is less than 0.01 for three consecutive iterations; The proportion of cross-type communities is less than 15%. Otherwise, the cross-type communities are split. The splitting rules are as follows: Split(C)={C t ∣t∈T,C t ={v i ∈C∣t i =t}} Among them, C is the cross-type community C t It is the subcommunity after splitting by type.

Images