Method and system for judging industrial field of enterprise based on industrial chain

Abstract

The invention relates to the technical field of industrial chains, provides an industrial chain-based method and system for discriminating the industrial field of an enterprise, and solves the problems of low accuracy and poor adaptability of enterprise field discrimination in an industrial chain environment in the prior art. The method comprises the following steps: firstly, receiving multi-source data of a new enterprise and an associated enterprise, performing parallel processing through a block model to form an enterprise feature set, and mapping the enterprise feature set with an industry association rule base to generate an enterprise knowledge graph; extracting a new enterprise feature set, matching the new enterprise feature set with the map to obtain a candidate industrial chain node set, determining a new node adding position according to the topological connection density between nodes, and constructing a dynamic industrial chain; and finally, calculating the feature similarity between the new node and the field core node, and accurately judging the industry field to which the new enterprise belongs in combination with the business flow intensity. According to the invention, the accuracy and adaptability of enterprise field discrimination in an industrial chain environment are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial chain, and in particular to a method and system for determining the industrial field of an enterprise based on an industrial chain. BACKGROUND

[0002] In the context of industrial chain collaborative development, it is necessary to accurately determine the industrial field of an enterprise to support the formulation of industrial strategies and the optimization of supply chains. The existing technical requirements mainly involve three aspects: first, it is necessary to process multi-source heterogeneous enterprise data, including technical patents, product information, and supply chain relationships; second, it is necessary to establish a dynamically updated industrial knowledge system to reflect real-time changes in the industrial chain; and third, it is necessary to quantify the industrial correlation strength between enterprises to avoid subjective judgment bias.

[0003] The representative solution for this requirement at present is an industrial classification system based on multi-dimensional feature vector matching. This system first constructs a standard industrial feature library, extracts the technical direction and product type features of an enterprise using feature engineering methods, calculates the feature similarity between the enterprise and each industrial category using a vector space model, then combines the business association network between enterprises and uses network node centrality indicators to assist in judgment, and finally determines the industrial classification based on the weighted scoring results. This system uses a regular batch update mechanism to support the automated processing of large-scale enterprise data.

[0004] This solution has certain limitations in practical application. The update of the standard industrial feature library relies on manual maintenance, and the feature dimensions are relatively fixed; the vector space model has limited representation capabilities for technical product features, making it difficult to capture complex correlations in the industrial chain; and the network node centrality indicators mainly reflect static topological features, and the correspondence with the actual operating state of the industry is not sufficient. These factors affect the accuracy and practicality of industrial field determination. SUMMARY

[0005] The present application provides a method and system for determining the industrial field of an enterprise based on an industrial chain to solve the problem of low accuracy and poor adaptability in determining the field of an enterprise in an industrial chain environment in the prior art.

[0006] To solve the above technical problems, in a first aspect, the present application provides a method for determining the industrial field of an enterprise based on an industrial chain, comprising:

[0007] receiving the declaration data of a new enterprise and the multi-source data of a target enterprise and associated enterprises;

[0008] performing parallel processing of the multi-source data using a block model to form an enterprise feature set, mapping and associating the enterprise feature set with a pre-set industrial correlation rule library to generate an enterprise knowledge graph, the enterprise feature set including a technical feature set and a product feature set;

[0009] extract a new enterprise feature set from the new enterprise declaration data, match the new enterprise feature set with the enterprise knowledge graph, and generate a candidate industry chain node set according to a matching result;

[0010] calculate a topological connection density between each pair of nodes in the candidate industry chain node set, select a node pair with the largest topological connection density in the candidate industry chain node set as an anchor node pair, perform feature extraction on a technology association edge and a product circulation edge of the anchor node pair through a graph neural network to generate an enhanced edge vector, generate an offset coefficient of a new node addition position according to the enhanced edge vector, and add a new enterprise as a new enterprise node, add the new enterprise node at a coordinate corresponding to the offset coefficient as an addition position in the enterprise knowledge graph, and add a new enterprise node, predict a potential interaction relationship between the new enterprise node and the anchor node pair through a relationship reasoning model to generate a new technology association edge and a new product circulation edge, and generate a dynamic industry chain based on the new technology association edge and the new product circulation edge;

[0011] calculate a feature similarity between the new enterprise node and a field core node in the dynamic industry chain in real time, and determine an industry field to which the new enterprise belongs based on the feature similarity and a business flow intensity between the new enterprise node and an associated node in the candidate industry chain node set.

[0012] Optionally, the feature extraction on the technology association edge and the product circulation edge of the anchor node pair through the graph neural network to generate the enhanced edge vector, and the generation of the offset coefficient of the new node addition position according to the enhanced edge vector include:

[0013] input the technology association edge of the anchor node pair into a first edge processing channel of the graph neural network, and extract a technology association feature vector from the technology association edge of the anchor node pair through the first edge processing channel;

[0014] input the product circulation edge of the anchor node pair into a second edge processing channel of the graph neural network, and extract a product circulation feature vector from the product circulation edge of the anchor node pair through the second edge processing channel;

[0015] perform weighted splicing on the technology association feature vector and the product circulation feature vector through a cross-channel attention fusion module of the graph neural network to generate an enhanced edge vector;

[0016] input the enhanced edge vector into an industry chain space projection layer of the graph neural network to output a technology-intensive direction projection component and a circulation hub direction projection component;

[0017] calculate a direction fusion factor based on a ratio of the technology-intensive direction projection component and the logistics hub direction projection component;

[0018] According to the direction fusion factor, find the offset coefficient corresponding to the new node addition position in the preset offset amount mapping table.

[0019] Optionally, the cross-channel attention fusion module of the graph neural network performs weighted splicing on the technology association feature vector and the product logistics feature vector to generate an enhanced edge vector, including:

[0020] Based on the semantic diffusion path depth and the patent technology keyword coverage of the technology association feature vector, calculate a technology dynamic influence factor;

[0021] Based on the logistics track hop count and the spatiotemporal aggregation radius of the product logistics feature vector, calculate a logistics dynamic influence factor;

[0022] The technology dynamic influence factor and the logistics dynamic influence factor are normalized respectively to obtain a technology attention weight value and a logistics attention weight value;

[0023] Based on the technology attention weight value and the logistics attention weight value, the technology association feature vector and the product logistics feature vector are weighted and scaled respectively to generate a weighted technology feature block and a weighted logistics feature block;

[0024] The weighted technology feature block and the weighted logistics feature block are subjected to multi-scale feature fusion using a feature pyramid network to generate a fusion feature matrix;

[0025] The fusion feature matrix is subjected to dimensionality reduction compression processing to output an enhanced edge vector.

[0026] Optionally, the multi-scale feature fusion of the weighted technology feature block and the weighted logistics feature block using the feature pyramid network to generate the fusion feature matrix, includes:

[0027] A feature pyramid network including a first pyramid layer, a second pyramid layer, and a third pyramid layer is constructed;

[0028] The weighted technology feature block and the weighted logistics feature block are input into the feature pyramid network to output a technology multi-scale feature group including three scales and a logistics multi-scale feature group including three scales;

[0029] The technology multi-scale feature group and the logistics multi-scale feature group are subjected to cross-modal feature fusion, wherein the first pyramid layer adopts element-by-element addition fusion, the second pyramid layer adopts channel attention weighted fusion, and the third pyramid layer adopts spatial attention weighted fusion;

[0030] The multi-scale feature groups after fusion are unified to the original resolution through an upsampling operation to generate multi-scale fusion features;

[0031] Based on the industry type configuration parameters, the multi-scale fusion features are rearranged into a two-dimensional matrix structure according to a preset dimension ratio to form a fusion feature matrix.

[0032] Optionally, the multi-source data is processed in parallel through the block model to form an enterprise feature set, and the enterprise feature set is mapped and associated with a preset industry association rule library to generate an enterprise knowledge graph, including:

[0033] The multi-source data is divided into a reference technology feature block and a reference product feature block through the block model;

[0034] A plurality of patent technology keywords are extracted from the reference technology feature block to form a technology feature set, and a plurality of supply chain labels are extracted from the reference product feature block to form a product feature set;

[0035] A predefined industry chain technology attribute set and a product attribute set are extracted from the industry association rule library, wherein the industry chain technology attribute set contains corresponding key technology items of each industry field, and the product attribute set contains corresponding product circulation paths of each industry field;

[0036] The technology feature set is matched with the key technology items in a dependency relationship, and the semantic association degree of each patent technology keyword and the corresponding key technology item is calculated according to the matching result;

[0037] The product feature set is associated with the product circulation path in a circulation relationship, and the topological coincidence degree of each supply chain label and the corresponding product circulation path is calculated according to the association result;

[0038] When there is a key technology item with a semantic association degree greater than a preset association degree threshold and a product circulation path with a topological coincidence degree greater than a preset coincidence degree threshold, a technology association edge and a product circulation edge between an enterprise entity and an industry chain node are generated;

[0039] Based on the technology association edge and the product circulation edge, an enterprise knowledge graph is formed.

[0040] Optionally, the new enterprise feature set is matched with the enterprise knowledge graph according to the matching result to generate a candidate industry chain node set, including:

[0041] The to-be-matched technology feature set and the to-be-matched product feature set are separated from the new enterprise feature set;

[0042] input the to-be-matched technical feature set, the to-be-matched product feature set and the enterprise knowledge graph into a preset multi-modal matching network, perform dependency relationship strength calculation on the to-be-matched technical feature set and technical attributes of nodes in the enterprise knowledge graph through a technical matching sub-network of the multi-modal matching network;

[0043] perform circulation relationship coincidence degree calculation on the to-be-matched product feature set and product attributes of nodes in the enterprise knowledge graph through a product matching sub-network of the multi-modal matching network;

[0044] filter a plurality of candidate industrial chain nodes with dependency relationship strength greater than a preset strength threshold and circulation relationship coincidence degree greater than a preset circulation threshold from the enterprise knowledge graph;

[0045] aggregate all the candidate industrial chain nodes to form a candidate industrial chain node set.

[0046] Optionally, the feature similarity between a new enterprise node in the dynamic industrial chain and a field core node is calculated in real time, and based on the feature similarity, the business flow strength between the new enterprise node and an associated node in the candidate industrial chain node set is combined to determine the industrial field to which the new enterprise belongs, including:

[0047] calculating technical feature similarity between a technical feature set of the new enterprise node and a technical attribute set of the field core node;

[0048] calculating product feature similarity between a product feature set of the new enterprise node and a product attribute set of the field core node;

[0049] calculating a business flow density value between the new enterprise node and the associated node in the candidate industrial chain node set;

[0050] performing first weighted fusion of the technical feature similarity and the business flow density value by using an attention mechanism, and performing second weighted fusion of the product feature similarity and the business flow density value;

[0051] when the first weighted fusion result is greater than a preset first threshold and the second weighted fusion result is greater than a preset second threshold, determining that the new enterprise belongs to an industrial field corresponding to the field core node.

[0052] In a second aspect, the present application provides an enterprise industrial field determination system based on an industrial chain, including:

[0053] a receiving module configured to receive declaration data of a new enterprise, and multi-source data of a target enterprise and associated enterprises;

[0054] The association module is configured to perform parallel processing on the multi-source data by using a block model, form an enterprise feature set, map and associate the enterprise feature set with a preset industry association rule library, and generate an enterprise knowledge graph, wherein the enterprise feature set includes a technology feature set and a product feature set;

[0055] The extraction module is configured to extract a new enterprise feature set from the new enterprise declaration data, match the new enterprise feature set with the enterprise knowledge graph, and generate a candidate industry chain node set according to a matching result;

[0056] The selection module is configured to calculate a topological connection density between each pair of nodes in the candidate industry chain node set, select a pair of nodes with the largest topological connection density in the candidate industry chain node set as an anchor node pair, perform feature extraction on a technology association edge and a product circulation edge of the anchor node pair by using a graph neural network to generate an enhanced edge vector, generate an offset coefficient of a new node addition position according to the enhanced edge vector, add a new enterprise as a new enterprise node, add the new enterprise node at the addition position corresponding to the offset coefficient in the enterprise knowledge graph, predict a potential interaction relationship between the new enterprise node and the anchor node pair by using a relationship reasoning model to generate a new technology association edge and a new product circulation edge, and generate a dynamic industry chain based on the new technology association edge and the new product circulation edge.

[0057] The calculation module is configured to calculate a feature similarity between the new enterprise node and a field core node in the dynamic industry chain in real time, and determine an industry field to which the new enterprise belongs based on the feature similarity and a business flow intensity between the new enterprise node and an associated node in the candidate industry chain node set.

[0058] In a third aspect, the present application provides an electronic device, comprising:

[0059] A memory configured to store a computer program.

[0060] A processor configured to execute the computer program to implement the steps of the method for determining an industry field to which an enterprise belongs based on an industry chain according to the first aspect.

[0061] In a fourth aspect, the present application provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executable by a processor to implement the steps of the method for determining an industry field to which an enterprise belongs based on an industry chain according to the first aspect.

[0062] In the present application, a method for determining the industrial field of an enterprise based on an industrial chain is provided, which comprises: receiving declaration data of a new enterprise, and multi-source data of a target enterprise and associated enterprises; performing parallel processing on the multi-source data through a block model to form an enterprise feature set, mapping and associating the enterprise feature set with a preset industrial association rule library to generate an enterprise knowledge graph, the enterprise feature set including a technology feature set and a product feature set; extracting a new enterprise feature set from the new enterprise declaration data, matching the new enterprise feature set with the enterprise knowledge graph, and generating a candidate industrial chain node set according to the matching result; calculating the topological connection density between each pair of nodes in the candidate industrial chain node set, selecting the pair of nodes with the maximum topological connection density in the candidate industrial chain node set as an anchor node pair, extracting features of the technology association edges and product circulation edges of the anchor node pair through a graph neural network to generate an enhanced edge vector, generating an offset coefficient of the new node addition position according to the enhanced edge vector, and adding the new enterprise as a new enterprise node, adding the new enterprise node at the addition position corresponding to the offset coefficient in the enterprise knowledge graph, predicting the potential interaction relationship between the new enterprise node and the anchor node pair using a relationship reasoning model, generating a new technology association edge and a new product circulation edge, and generating a dynamic industrial chain based on the new technology association edge and the new product circulation edge; and calculating the feature similarity between the new enterprise node and the field core node in the dynamic industrial chain in real time, and determining the industrial field of the new enterprise based on the feature similarity and the business flow intensity between the new enterprise node and the associated nodes in the candidate industrial chain node set.

[0063] The technical scheme provided by the present application has the following beneficial effects:

[0064] The present application realizes the unified access of multi-source heterogeneous data, provides a complete data basis for subsequent processing, improves the data processing efficiency, ensures the effective extraction of technology features and product features, constructs a structured industrial knowledge system, intuitively displays the technology association and product circulation relationship between enterprises, quickly filters out the industrial chain nodes with the highest association degree with the new enterprise, narrows down the determination range, realizes the accurate positioning of the new enterprise in the industrial chain, and ensures the rationality of the industrial chain structure. The static features and dynamic business data are combined to improve the accuracy and reliability of the determination.

[0065] Further, the present application further extracts a technology association feature vector and a product circulation feature vector through double-channel processing of the graph neural network, generates an enhanced edge vector using a cross-channel attention fusion module, obtains projection components of the technology-intensive direction and the circulation hub direction through industrial chain space projection, calculates a direction fusion factor, and queries an offset mapping table to obtain an offset coefficient of the new node addition position.

[0066] And the method realizes the deep extraction and intelligent fusion of the two-dimensional characteristics of technology and circulation through the technology of graph neural network, and obtains accurate offset coefficients through spatial projection and direction factor calculation, ensuring that the addition position of the new node in the industrial chain meets the characteristics of technology-intensive and adapts to the distribution of circulation hubs, effectively improving the accuracy and rationality of the new enterprise embedded in the industrial chain. BRIEF DESCRIPTION OF DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0068] Figure 1 The flow chart of the enterprise belonging to the industrial field discrimination method based on the industrial chain provided by the embodiments of the present application;

[0069] Figure 2 The specific implementation schematic diagram of the enterprise belonging to the industrial field discrimination method based on the industrial chain provided by the embodiments of the present application;

[0070] Figure 3 The structural schematic diagram of the enterprise belonging to the industrial field discrimination system based on the industrial chain provided by the embodiments of the present application. DETAILED DESCRIPTION

[0071] In the current industrial chain enterprise field discrimination technology, although the industrial classification system based on multi-dimensional feature matching can realize automatic processing, there are still obvious limitations. The update of the standard industrial feature library depends on manual maintenance, which is difficult to adapt to the rapidly changing industrial environment; the vector space model has limited representation ability for technical product features, and cannot fully capture the complex correlations in the industrial chain; the network node centrality index mainly reflects the static topological characteristics, which has a gap with the dynamic characteristics of the actual industrial operation state. These problems lead to the challenges of the existing methods in the accuracy and practicability of the industrial field discrimination.

[0072] To address the aforementioned issues, this application proposes a method for identifying the industry sector of a company based on its industry chain. This method uses a block model to process multi-source company data in parallel, constructing a company knowledge graph that integrates technological and product characteristics. It employs topological connection density analysis to determine the optimal position of a new company within the industry chain and dynamically updates the industry chain structure by combining graph neural networks and relational reasoning models. Finally, it achieves accurate identification through a comprehensive calculation of feature similarity and business flow intensity. This method overcomes the limitations of traditional static feature matching. Through dynamic industry chain construction and real-time business flow analysis, it effectively solves the problems of lagging industry feature updates, insufficient capture of complex relationships, and lack of dynamic adaptability in existing technologies, thus improving the accuracy and practicality of industry sector identification.

[0073] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0074] The core of this application is to provide a method for determining the industry sector to which a company belongs based on the industrial chain. A flowchart of one specific implementation method is shown below. Figure 1 As shown, the method includes:

[0075] Step 101: Receive the application data of the new enterprise, as well as multi-source data of the target enterprise and related enterprises.

[0076] In step 101, the data submitted by new enterprises refers to the basic information submitted by newly registered or unclassified enterprises, including enterprise name, registered address, business scope, patent list, product catalog, etc. The target enterprise refers to the core enterprise requiring industry sector identification. Related enterprises refer to enterprises that have business dealings, investment relationships, or technological cooperation with the target enterprise. Multi-source data refers to heterogeneous data from different channels, including business registration information, patent applications, supply chain records, financial reports, etc.

[0077] In this embodiment of the application, the application data submitted by the new enterprise is received through the data interface, and multi-source data of the target enterprise and related enterprises are obtained from public databases or enterprise cooperation platforms, including business registration information, patent texts, product lists and supply chain records. All data are stored in a temporary database after being standardized in format, providing a unified data source for subsequent processing.

[0078] For example, taking the new energy vehicle industry chain as an example, the declared data of a new battery enterprise D is received, including the enterprise name, the registered place, three lithium battery patents and two motor supply contracts; at the same time, the business information, five patents and the annual product list of the target enterprise battery manufacturer A and the supply chain data and the cooperation enterprise list of the associated enterprise motor supplier B are obtained, and all the data are stored in a temporary database after format conversion.

[0079] Step 102: parallel processing of the multi-source data by the block model, forming an enterprise feature set, mapping and associating the enterprise feature set with a preset industry association rule base to generate an enterprise knowledge graph, the enterprise feature set including a technology feature set and a product feature set.

[0080] In step 102, the block model represents a data parallel processing architecture, which divides data into independent modules for parallel processing according to types. The enterprise feature set includes a technology feature set (patent keywords, technology field) and a product feature set (product type, supply chain label). The industry association rule base represents a predefined industry chain rule set, including technology dependency relationship and product circulation path. The enterprise knowledge graph represents graph structure data, with nodes representing enterprises or industry fields and edges representing technology association or product circulation relationship.

[0081] In the embodiments of the present application, the block model divides the multi-source data into a technology data block and a product data block, the technology data block extracts patent keywords to form a technology feature set, and the product data block extracts supply chain labels to form a product feature set; the technology feature set is matched with the technology items in the industry association rule base to calculate the semantic association degree, and the product feature set is matched with the product circulation path to calculate the topological coincidence degree; when the semantic association degree and the topological coincidence degree both exceed the threshold value, the technology association edge and the product circulation edge between the enterprise node and the industry node are generated in the enterprise knowledge graph.

[0082] For example, the technology data block of the battery manufacturer A extracts patent keywords such as "lithium ion battery" and "energy density", and the product data block extracts supply chain labels such as "Yangtze River Delta distribution" and "48-hour delivery"; the technology feature set matches the semantic association degree 0.92 with "battery energy density > 300 Wh / kg" in the rule base, and the product feature set matches the topological coincidence degree 0.85 with the "battery-motor-vehicle manufacturer" path; since both values exceed the threshold value 0.8, the technology association edge and the product circulation edge between A and the new energy vehicle industry node are generated in the knowledge graph.

[0083] Step 103: extracting a new enterprise feature set from the new enterprise declared data, matching the new enterprise feature set with the enterprise knowledge graph, and generating a candidate industry chain node set according to the matching result.

[0084] In step 103, the new enterprise feature set represents the technical features and product features extracted from the new enterprise data. The candidate industry chain node set represents a set of industry nodes with high matching degrees in the enterprise knowledge graph.

[0085] In the embodiments of the present application, the technical keywords and supply chain labels are extracted from the new enterprise declaration data to form a new enterprise feature set. The cosine similarity between the technical keywords and the technical attributes of the nodes in the knowledge graph is calculated to obtain the technical matching degree, and the Jaccard similarity between the supply chain labels and the product attributes of the nodes is calculated to obtain the product matching degree. Nodes with both technical matching degrees and product matching degrees exceeding the threshold are screened and aggregated into a candidate industry chain node set.

[0086] For example, a new battery enterprise D extracts technical features "high-nickel positive electrode" and "silicon-carbon negative electrode" and product features "South China warehouse" and "72-hour delivery". The technical matching degree between D and the battery manufacturer A node is 0.88, the product matching degree is 0.70, the technical matching degree between D and the vehicle manufacturer C node is 0.86, and the product matching degree is 0.78. The threshold is set to 0.8, and the dual matching degrees of the C node both exceed the threshold, so the C node is selected into the candidate node set.

[0087] Step 104: Calculate the topological connection density between each pair of nodes in the candidate industry chain node set, select the node pair with the maximum topological connection density in the candidate industry chain node set as the anchor node pair, perform feature extraction on the technical association edges and product circulation edges of the anchor node pair through the graph neural network, generate an enhanced edge vector, generate an offset coefficient of the new node addition position according to the enhanced edge vector, and add the new enterprise as a new enterprise node. In the enterprise knowledge graph, the coordinates corresponding to the offset coefficient are used as the addition position, and the new enterprise node is added. The potential interaction relationship between the new enterprise node and the anchor node pair is predicted by using a relationship reasoning model, new technical association edges and new product circulation edges are generated, and a dynamic industry chain is generated based on the new technical association edges and the new product circulation edges.

[0088] In step 104, the topological connection density represents the comprehensive strength index of the technical association and product circulation relationship between nodes. The anchor node pair is the node pair with the largest topological connection density in the candidate node set, which is used as a reference benchmark for the addition of new enterprise nodes. The selection basis is the strength of the technical and product two-dimensional association between nodes. The enhanced edge vector is a technical and product feature fusion vector extracted by a graph neural network. The offset coefficient is a position adjustment parameter calculated according to the enhanced edge vector, which is used to determine the addition position of the new node relative to the anchor node pair. The addition position is a coordinate point determined according to the offset coefficient in the enterprise knowledge graph. The new enterprise node is a newly added node entity at the coordinate point. The relationship reasoning model is an algorithm model used to predict the potential relationship between nodes. The new technical association edge and the new product circulation edge are new edge relationships generated according to the prediction results. The dynamic industry chain represents a real-time updated industry chain structure, including newly added enterprise nodes and their associated relationships.

[0089] In the embodiments of the present application, the topological connection density between each pair of nodes in the candidate node set is calculated, and the node pair with the largest density is selected as the anchor node pair. The enhanced edge vector is generated by extracting the technical association edge and product circulation edge features of the anchor node pair through a graph neural network, and the offset coefficient of the addition position of the new node is calculated. The new enterprise node is added at the position corresponding to the offset coefficient in the knowledge graph, the interaction relationship between the new node and the anchor node pair is predicted using the relationship reasoning model to generate the new technical association edge and the new product circulation edge, and the knowledge graph is updated to form a dynamic industry chain.

[0090] For example, the topological connection density of candidate node set battery manufacturer A and vehicle manufacturer C is the highest (number of common patents x annual transaction volume = 8 x 100,000 = 800,000), and is selected as the anchor node pair; the graph neural network extracts the edge features (patent diffusion path depth 3) and product flow edge features (logistics jump number 4) to generate an enhanced edge vector to calculate the offset coefficient +0.3; add enterprise D node at the offset position of 30% of the connection between A and C, and the relationship reasoning model predicts the technical association edge weight value of D and A as 0.8 and the product flow edge weight value of C as 0.6 to generate a dynamic industry chain. The following is a specific example: in the new energy vehicle industry chain scene, the candidate industry chain node set includes battery manufacturer A, motor supplier B and vehicle manufacturer C. When calculating the topological connection density between each pair of nodes, the number of common patents between battery manufacturer A and motor supplier B is 6, the annual transaction volume is 100,000, the topological connection density is 6 x 100,000 = 600,000, the number of common patents between motor supplier B and vehicle manufacturer C is 4, the annual transaction volume is 80,000, the topological connection density is 4 x 80,000 = 320,000, and the number of common patents between battery manufacturer A and vehicle manufacturer C is 8, the annual transaction volume is 100,000, the topological connection density is 8 x 100,000 = 800,000, so battery manufacturer A and vehicle manufacturer C with the highest topological connection density are selected as the anchor node pair; the graph neural network processes the technical association edge of the anchor node pair, extracts the patent technology keyword coverage of 0.7 and the semantic diffusion path depth of 3, calculates the technical dynamic influence factor of 0.7 x 3 = 2.1, processes the product flow edge to extract the logistics track jump number of 4 and the space-time aggregation radius of 200 kilometers, calculates the flow dynamic influence factor of 4 ÷ 200 = 0.02, inputs the two factors into the normalization layer to obtain the technical attention weight value of about 0.89, and the flow attention weight value of 0.11. The weighted technical feature block is obtained by weighting the technical association feature vector by 0.89, the weighted flow feature block is obtained by weighting the product flow feature vector by 0.11, the enhanced edge vector is obtained by splicing and dimensionality reduction according to the ratio of 7 to 3. The technical dense direction projection component 0.85 and the flow hub direction projection component 0.15 are obtained by inputting the vector into the industry chain space projection layer, the direction fusion factor is calculated as 0.85 ÷ 0.15 ≈ 5.67, the offset coefficient is determined as +0.3 according to the offset amount mapping table; add new battery enterprise D node at the offset position of 30% of the connection between battery manufacturer A and vehicle manufacturer C; use the relationship reasoning model to analyze the potential interaction between enterprise D and the anchor node pair, predict the technical association edge weight value with battery manufacturer A as 0.8, indicating the technical cooperation probability, and the product flow edge weight value with vehicle manufacturer C as 0.6, indicating the product supply probability; based on the generated new technical association edge and new product flow edge, update the enterprise knowledge graph to form a dynamic industry chain, and complete the precise embedding of the new enterprise in the industry chain.

[0091] ​Step 105: Real-time calculation of feature similarity between the new enterprise node and the field core node in the dynamic industry chain, based on the feature similarity, combined with the business flow intensity between the new enterprise node and the associated nodes in the candidate industry chain node set, to determine the industry field to which the new enterprise belongs.

[0092] In step 105, the field core node represents the center node in the knowledge graph representing a specific industry field. The business flow intensity represents the density of business transactions between the new enterprise and the associated nodes in the candidate node set.

[0093] In the embodiments of the present application, the technical feature similarity between the new enterprise node technical feature set and the field core node technical attribute set is calculated, as well as the product feature similarity between the product feature set and the product attribute set. The business flow density value between the new enterprise and the associated nodes in the candidate node set is calculated. The technical feature similarity and the business flow density value are weighted and fused, and the product feature similarity and the business flow density value are weighted and fused. When both fusion results exceed the threshold value, it is determined that the new enterprise belongs to the industry field.

[0094] For example, the technical feature similarity between enterprise D and field core node A is 0.85 (cosine similarity), and the product feature similarity is 0.75 (Jaccard similarity). The business flow density value between D and candidate node C is 0.8 (quarterly order volume / industry benchmark). The weighted fusion value of technical similarity and business flow density is 0.83 (weights 0.6 and 0.4), and the weighted fusion value of product similarity and business flow density is 0.775 (weights 0.5 and 0.5). Both values exceed the threshold value of 0.7, and it is determined that D belongs to the new energy vehicle industry field.

[0095] The method constructs an enterprise knowledge graph through parallel processing of multiple source data, realizes accurate positioning of new enterprises in the industry chain through topological connection density analysis, dynamically determines the industry field based on feature similarity and business flow intensity, effectively improves the completeness of the industry chain structure and the accuracy of the industry field determination, and adapts to the rapidly changing industrial environment.

[0096] To further improve the accuracy of the new node addition position, in some embodiments, step 104: the technology association edge and the product flow edge of the anchor node pair are input into the first edge processing channel of the graph neural network, and the technology association feature vector is extracted from the technology association edge of the anchor node pair through the first edge processing channel. Figure 2 As shown in the figure, it includes:

[0097] Step 201: input the technology association edge of the anchor node pair into the first edge processing channel of the graph neural network, and extract the technology association feature vector from the technology association edge of the anchor node pair through the first edge processing channel.

[0098] In step 201, the technology association feature vector is a feature set representing the technology association characteristics extracted from the technology association edge of the anchor node pair, including dimensions such as patent technology keyword coverage and semantic diffusion path depth.

[0099] In the embodiment of the present application, the technology association edge of the anchor node pair is input into the first edge processing channel of the graph neural network, the patent technology keyword distribution features on the edge are extracted through the convolutional neural network layer, and then the technology diffusion path features are analyzed through the recurrent neural network layer, and finally the technology association feature vector containing multi-dimensional technology characteristics is output.

[0100] Step 202: input the product circulation edge of the anchor node pair into the second edge processing channel of the graph neural network, and through the second edge processing channel, extract the product circulation feature vector from the product circulation edge of the anchor node pair.

[0101] In step 202, the product circulation feature vector is a feature set representing the product circulation characteristics extracted from the product circulation edge, including dimensions such as logistics track jump number and spatiotemporal aggregation radius.

[0102] In the embodiment of the present application, the product circulation edge of the anchor node pair is input into the second edge processing channel of the graph neural network, the logistics track features on the edge are extracted through the graph attention network layer, and then the product circulation range features are analyzed through the spatiotemporal encoding layer, and finally the product circulation feature vector containing multi-dimensional circulation characteristics is output.

[0103] Step 203: through the cross-channel attention fusion module of the graph neural network, the technology association feature vector and the product circulation feature vector are weighted and spliced to generate an enhanced edge vector.

[0104] In the embodiment of the present application, the cross-channel attention fusion module is used to calculate the attention weights of the technology association feature vector and the product circulation feature vector, and then the two feature vectors are weighted and summed according to the weights, and the weighted feature vector is spliced to generate an enhanced edge vector.

[0105] Step 204: input the enhanced edge vector into the industry chain space projection layer of the graph neural network, and output the technology-intensive direction projection component and the circulation hub direction projection component.

[0106] In step 204, the technology-intensive direction projection component and the circulation hub direction projection component are the decomposition features of the enhanced edge vector in the industry chain space, respectively representing the technology aggregation degree and the logistics concentration degree.

[0107] In the embodiment of the present application, the enhanced edge vector is input into the industry chain space projection layer, the projection values of the vector in the technology-intensive direction and the circulation hub direction are calculated through the fully connected neural network, and the feature components of the two directions are output.

[0108] Step 205: Calculate a direction fusion factor based on the ratio of the technology-intensive direction projection component and the logistics hub direction projection component.

[0109] In step 205, the direction fusion factor is the ratio of the technology-intensive direction projection component and the logistics hub direction projection component, reflecting the relative importance of the two dimensions of technology and logistics.

[0110] In the embodiments of the present application, the ratio of the technology-intensive direction projection component and the logistics hub direction projection component is calculated to obtain the direction fusion factor, which is used to determine the direction tendency of adding a new node.

[0111] Step 206: According to the direction fusion factor, find the offset coefficient corresponding to the new node addition position in the preset offset mapping table.

[0112] In the embodiments of the present application, according to the calculated direction fusion factor, the corresponding offset coefficient is found in the preset offset mapping table, which defines the position offset corresponding to different factor ranges.

[0113] The following is a specific example:

[0114] In the new energy automobile industry chain scene, taking battery manufacturer A and vehicle manufacturer C as anchor node pairs, input the technology association edge into the first edge processing channel of the graph neural network to extract the technology association feature vector containing the patent technology keyword coverage 0.7 and the semantic diffusion path depth 3, wherein the coverage is calculated by counting the proportion of repeated keywords in the common patents of A and C, and the path depth is determined by analyzing the number of levels through which the technology is transmitted from A to C. Input the product flow edge into the second edge processing channel to extract the product flow feature vector containing the logistics track hop number 4 and the space-time aggregation radius 200 kilometers, wherein the hop number is obtained by counting the number of warehouse transportation rings that the product goes through from A to C, and the aggregation radius is determined by calculating the average distance of all distribution paths from A to C. Calculate the technology attention weight and the logistics attention weight through the cross-channel attention fusion module, specifically using the formula Calculate the technology weight, wherein represents the technology attention weight, is the technology dynamic influence factor 2.1 x (0.7 x 3), is the logistics dynamic influence factor 0.02 (4 ÷ 200), and , The technology correlation feature vector is weighted by 0.89 to obtain a weighted technology feature block, the product circulation feature vector is weighted by 0.11 to obtain a weighted circulation feature block, and after splicing according to a 7:3 ratio, a 128-dimensional enhanced edge vector is output after dimension reduction. The vector is input into an industry chain space projection layer, and a technology-intensive direction projection component 0.85 and a circulation hub direction projection component 0.15 are calculated through a fully connected network. A direction fusion factor wherein represents the direction fusion factor, is the technology-intensive direction projection component, is the circulation hub direction projection component. According to a preset offset amount mapping table, when R is greater than 5, the offset coefficient +0.3 is determined, and the new node is offset by 30% of the distance in the technology-intensive direction. Finally, the offset coefficient +0.3 is used for new node addition position calculation.

[0115] In the embodiments of the present application, an enhanced edge vector is generated through double-channel feature extraction and attention fusion, and an accurate offset coefficient is calculated through space projection and direction fusion, so as to realize accurate quantization of the new node addition position and improve the accuracy and adaptability of dynamic industry chain construction.

[0116] In order to further improve the accuracy of feature fusion, in some embodiments, step 203: the cross-channel attention fusion module of the graph neural network is used to weight and splice the technology correlation feature vector and the product circulation feature vector to generate an enhanced edge vector, comprising:

[0117] Step 301: calculating a technology dynamic influence factor based on the semantic diffusion path depth and the patent technology keyword coverage of the technology correlation feature vector.

[0118] In step 301, the semantic diffusion path depth refers to the number of transmission levels of the patent technology in the industry chain, which is obtained by counting the number of times of transmission between enterprises from R&D to application, for example, the path depth of the battery technology from material R&D to cell design to system integration to vehicle application is 4. The patent technology keyword coverage refers to the distribution ratio of a certain patent keyword in the associated enterprises, which is obtained by counting the percentage of the number of enterprises using the keyword in the total number of related enterprises in the industry chain, for example, the "high-nickel positive electrode" keyword is used in 8 of 10 battery enterprises, so the coverage is 80%. The technology dynamic influence factor is a quantitative index for measuring the importance of the technology feature, which is obtained by multiplying the semantic diffusion path depth and the patent technology keyword coverage in the technology correlation feature vector.

[0119] In the embodiments of the present application, the semantic diffusion path depth and the patent technology keyword coverage are obtained from the technology correlation feature vector, and the technology dynamic influence factor is calculated by multiplying the two parameters. The larger the factor value is, the more important the technology feature is.

[0120] Step 302: Calculate the circulation dynamic influence factor based on the logistics track hop number and the space-time aggregation radius of the product circulation feature vector.

[0121] In step 302, the logistics track hop number refers to the number of intermediate circulation links of a product from production to consumption, which is obtained by counting the number of nodes such as storage, transportation, and quality inspection recorded by supply chain documents. For example, a battery needs to pass through 3 warehouses and be transported twice from factory to vehicle assembly, so the hop number is 5. The space-time aggregation radius refers to the coverage range of the product distribution network, which is obtained by calculating the average distance from the regional storage center to each distribution point. For example, the average distance from the East China storage center of a certain battery enterprise to all 4S stores is 150 kilometers, so the aggregation radius is 150 kilometers. The circulation dynamic influence factor is a quantitative indicator for measuring the importance of product circulation features, which is obtained by dividing the logistics track hop number by the space-time aggregation radius in the product circulation feature vector.

[0122] In the embodiments of the present application, the logistics track hop number and the space-time aggregation radius are obtained from the product circulation feature vector, and the circulation dynamic influence factor is calculated by dividing the logistics track hop number by the space-time aggregation radius. The larger the factor value, the more important the circulation feature.

[0123] Step 303: Normalize the technical dynamic influence factor and the circulation dynamic influence factor respectively to obtain the technical attention weight value and the circulation attention weight value.

[0124] In step 303, the technical attention weight value and the circulation attention weight value are weight coefficients obtained by normalization processing, which are used to measure the relative importance of technical and circulation features.

[0125] In the embodiments of the present application, the technical dynamic influence factor and the circulation dynamic influence factor are input into the normalization processing layer, and two weight values are calculated by the softmax function to ensure that the sum of the two weight values is a fixed value.

[0126] Step 304: Based on the technical attention weight value and the circulation attention weight value, the technical correlation feature vector and the product circulation feature vector are weighted and scaled respectively to generate a weighted technical feature block and a weighted circulation feature block.

[0127] In step 304, the weighted technical feature block and the weighted circulation feature block are feature blocks obtained by adjusting the weight of the original feature vector respectively.

[0128] In the embodiments of the present application, the technical attention weight value is used for element-wise multiplication operation on the technical correlation feature vector to generate the weighted technical feature block, and the circulation attention weight value is used for element-wise multiplication operation on the product circulation feature vector to generate the weighted circulation feature block.

[0129] Step 305: performing multi-scale feature fusion on the weighted technical feature block and the weighted flow feature block by using a feature pyramid network to generate a fusion feature matrix.

[0130] In step 305, the fusion feature matrix is a composite feature representation obtained by splicing the two weighted feature blocks according to a preset proportion.

[0131] In the embodiments of the present application, the weighted technical feature block and the weighted flow feature block are spliced according to a preset dimension proportion, the technical feature block occupies a major proportion, and the flow feature block occupies a minor proportion, thereby forming the fusion feature matrix.

[0132] Step 306: performing dimension reduction compression processing on the fusion feature matrix to output an enhanced edge vector.

[0133] In the embodiments of the present application, the fusion feature matrix is subjected to principal component analysis dimension reduction processing, the most important feature components are extracted, and a fixed-dimension enhanced edge vector is output.

[0134] The following is a specific example:

[0135] In the new energy automobile industry chain scenario, the technical correlation feature vector based on the battery manufacturer A and the vehicle manufacturer C contains a patent technology keyword coverage of 0.7 and a semantic diffusion path depth of 3, and the calculation of the technical dynamic influence factor is 0.7x3=2.1, wherein the coverage is obtained by counting the proportion of repeated keywords in the patents shared by A and C, and the path depth is determined by analyzing the research and development, production, and testing three levels through which the technology is transferred from A to C. Based on the product flow feature vector, the logistics track hop number is 4 and the space-time aggregation radius is 200 kilometers, and the calculation of the flow dynamic influence factor is 4÷200=0.02, wherein the hop number is obtained by counting the four links of factory storage, trunk transportation, regional distribution, and terminal distribution that the product goes through from A to C, and the aggregation radius is determined by calculating the average distance of 200 kilometers of all distribution paths from A to C. The technical dynamic influence factor 2.1 and the flow dynamic influence factor 0.02 are normalized, and the formula is used to calculate the technical attention weight value, and the flow attention weight value is obtained as ​The technical correlation feature vector is weighted and scaled using a technical attention weight value 0.89, each dimension value in the vector is multiplied by 0.89 to generate a weighted technical feature block, and the product circulation feature vector is weighted and scaled using a circulation attention weight value 0.11 to generate a weighted circulation feature block. The weighted technical feature block and the weighted circulation feature block are spliced into a fusion feature matrix according to a preset dimension ratio of 7 to 3, wherein the technical features occupy 70% of the dimensions and the circulation features occupy 30% of the dimensions. The fusion feature matrix is subjected to principal component analysis dimension reduction processing, and the first 128 main feature components are extracted, and a 128-dimensional enhanced edge vector is output, which comprehensively retains the core information of the technical and circulation features.

[0136] In the embodiments of the present application, through dynamic influence factor calculation and attention weight distribution, intelligent fusion of technical and circulation features is realized, and then through proportional splicing and dimension reduction compression, an enhanced edge vector is generated, which effectively improves the accuracy and calculation efficiency of feature representation and provides a reliable feature basis for subsequent offset coefficient calculation.

[0137] In order to further improve the effect of multi-scale feature fusion, in some embodiments, step 305: the multi-scale feature fusion of the weighted technical feature block and the weighted circulation feature block is performed by using a feature pyramid network to generate a fusion feature matrix, comprising:

[0138] Step 401: constructing a feature pyramid network comprising a first pyramid layer, a second pyramid layer and a third pyramid layer.

[0139] In step 401, the feature pyramid network is a multi-scale feature extraction architecture, the first pyramid layer is a feature layer that maintains the original resolution, the second pyramid layer is a feature layer that has been down-sampled once, and the third pyramid layer is a feature layer that has been down-sampled twice.

[0140] In the embodiments of the present application, a feature pyramid network structure comprising three pyramid layers is constructed, the first pyramid layer directly processes the original resolution feature, the second pyramid layer reduces the feature size by half through convolution and down-sampling operation, and the third pyramid layer further reduces the feature size to one quarter of the original size through convolution and down-sampling operation based on the second layer.

[0141] Step 402: inputting the weighted technical feature block and the weighted circulation feature block into the feature pyramid network to output a technical multi-scale feature group comprising three scales and a circulation multi-scale feature group comprising three scales.

[0142] In step 402, the technical multi-scale feature group is a set of technical features of three different scales obtained after the weighted technical feature block is processed by the feature pyramid network, and the circulation multi-scale feature group is a set of circulation features of three different scales obtained after the weighted circulation feature block is processed by the feature pyramid network.

[0143] In the embodiment of the present application, the weighted technical feature block is input into the feature pyramid network, and through convolution and down-sampling operations at each level, a technical multi-scale feature group containing original scale, half scale and quarter scale is output. At the same time, the weighted circulation feature block is input into the same network structure, and a corresponding circulation multi-scale feature group of three scales is output.

[0144] In step 403, the cross-modal feature fusion is a process of combining the technical multi-scale feature group and the circulation multi-scale feature group at different scales. The element-wise addition is to directly add the feature values at the corresponding positions. The channel attention weighting is to perform weighted fusion according to the importance of the feature channels. The spatial attention weighting is to perform weighted fusion according to the importance of the feature spatial positions.

[0145] In step 403, the cross-modal feature fusion is a process of combining the technical multi-scale feature group and the circulation multi-scale feature group at different scales. The element-wise addition is to directly add the feature values at the corresponding positions. The channel attention weighting is to perform weighted fusion according to the importance of the feature channels. The spatial attention weighting is to perform weighted fusion according to the importance of the feature spatial positions.

[0146] In the embodiment of the present application, the technical multi-scale feature group and the circulation multi-scale feature group of the first pyramid layer are fused by element-wise addition. The feature groups of the second pyramid layer are weighted fused after calculating the weights of each channel by the channel attention mechanism. The feature groups of the third pyramid layer are weighted fused after calculating the weights of each position by the spatial attention mechanism.

[0147] In step 404, the multi-scale fusion feature is a comprehensive feature representation obtained after the multi-scale fusion features of different scales are unified to the original resolution.

[0148] In step 404, the multi-scale fusion feature is a comprehensive feature representation obtained after the multi-scale fusion features of different scales are unified to the original resolution.

[0149] In the embodiment of the present application, the features after the fusion of the three pyramid layers are respectively subjected to up-sampling operations, so that all the features are restored to the original resolution size, and then spliced to generate a multi-scale fusion feature containing multi-scale information.

[0150] In step 405, based on the industry type configuration parameter, the multi-scale fusion feature is rearranged into a two-dimensional matrix structure according to a preset dimension ratio, to form a fusion feature matrix.

[0151] In step 405, the industry type configuration parameter is a dimension ratio parameter set according to different industry characteristics.

[0152] In the embodiment of the present application, the corresponding configuration parameter is selected according to the industry type, the multi-scale fusion feature is reorganized and arranged according to the preset technical feature and circulation feature dimension ratio, and a standard two-dimensional matrix structure is formed as a fusion feature matrix.

[0153] The following is a specific example:

[0154] In the new energy vehicle industry chain scene, the weighted technical feature block obtained based on the foregoing embodiment contains four dimension data [0.623, 0.623, 0.267, 2.67] and the weighted circulation feature block contains three dimension data [1.1, 0.44, 22], a feature pyramid network containing a first pyramid layer maintaining the original resolution 4x3, a second pyramid layer down-sampling to 2x2, and a third pyramid layer down-sampling to 1x1 is constructed; the weighted technical feature block and the weighted circulation feature block are input into the feature pyramid network, and feature extraction is performed through convolution operation with a convolution kernel size of 3x3, the first pyramid layer outputs a technical multi-scale feature group [0.62, 0.62, 0.27, 2.67] and a circulation multi-scale feature group [1.10, 0.44, 22.00], the second pyramid layer outputs a down-sampled feature group through maximum pooling with a step of 2, and the third pyramid layer continues to output a coarsest granularity feature group; cross-modal feature fusion is performed on the three pyramid layers, the first pyramid layer adopts an element-by-element addition formula for calculation, wherein represents the element of the first row and the first column of the fusion feature matrix, represents the element of the first row and the first column of the technical multi-scale feature group, represents the element of the first row and the first column of the circulation multi-scale feature group, and the calculation obtains fusion values such as [1.72, 1.06, 22.27], the second pyramid layer adopts a channel attention weighted fusion formula , wherein represents the fusion feature of the i-th channel, represents the attention weight of the i-th channel of the technical feature, represents the value of the i-th channel of the technical feature, represents the attention weight of the i-th channel of the circulation feature, represents the value of the i-th channel of the circulation feature, ​​​​The values of the channels are calculated by a softmax function to obtain attention weights, and a third pyramid layer adopts a spatial attention weighting fusion formula wherein represents a position of a fusion feature, represents an attention weight of a technical feature at the position , represents a value of the technical feature at the position , represents an attention weight of a circulation feature at the position , represents a value of the circulation feature at the position ; the feature groups after the fusion of the three pyramid layers are uniformly restored to the original resolution of 4x3 by a bilinear upsampling operation to generate multi-scale fusion features containing multi-scale information; based on the characteristics that the new energy vehicle industry belongs to a technology-intensive industry, the technical feature dimension ratio is configured as 0.7 and the circulation feature dimension ratio is configured as 0.3, the multi-scale fusion features are rearranged into a two-dimensional matrix structure of 4 rows and 3 columns according to the ratio, wherein the technical feature related data occupies the first 2 rows of the matrix and the circulation feature related data occupies the last 2 rows of the matrix, to form a final fusion feature matrix for subsequent processing.

[0155] In the embodiments of the present application, multi-scale feature extraction is realized through a feature pyramid network, different levels of features are processed through a differentiated fusion strategy, and a fusion feature matrix is generated through industry adaptive dimension configuration, which effectively improves the comprehensiveness and industry adaptability of feature fusion.

[0156] In order to further improve the accuracy and efficiency of the construction of the enterprise knowledge graph, in some embodiments, step 102: the parallel processing of the multi-source data by the block model to form an enterprise feature set, and the mapping and association of the enterprise feature set with a preset industry association rule library to generate an enterprise knowledge graph, includes:

[0157] Step 501: dividing the multi-source data into a benchmark technical feature block and a benchmark product feature block through a block model.

[0158] In step 501, the benchmark technical feature block and the benchmark product feature block are data units obtained by classifying and processing the multi-source data through a block model, and respectively contain technical related data and product related data.

[0159] In the embodiments of the present application, the multi-source data is divided into a technical feature block and a product feature block according to the data type through a block model, the technical feature block contains patent text and technical document data, and the product feature block contains supply chain record and product information data, realizing the classified parallel processing of data.

[0160] Step 502: extracting a plurality of patent technology keywords from the reference technology feature block to form a technology feature set, and extracting a plurality of supply chain labels from the reference product feature block to form a product feature set.

[0161] In step 502, the technology feature set is a set of patent technology keywords extracted from the reference technology feature block. The product feature set is a set of supply chain labels extracted from the reference product feature block.

[0162] In the embodiments of the present application, the patent technology keywords are extracted from the technology feature block by text mining technology to form the technology feature set, and the supply chain labels are extracted from the product feature block by pattern recognition to form the product feature set, providing a feature basis for subsequent association matching.

[0163] Step 503: extracting a pre-defined industry chain technology attribute set and a product attribute set from the industry association rule base, wherein the industry chain technology attribute set contains key technology items corresponding to each industry field, and the product attribute set contains product circulation paths corresponding to each industry field.

[0164] In step 503, the industry chain technology attribute set is a set of key technology items corresponding to each industry field in the industry association rule base. The product attribute set is a set of pre-defined product circulation paths.

[0165] In the embodiments of the present application, the pre-defined technology attribute set and the product attribute set are extracted from the industry association rule base. The technology attribute set contains technology standards and requirements of each industry field, and the product attribute set contains typical product circulation and supply chain paths.

[0166] Step 504: matching the technology feature set with the key technology items in a dependency relationship, and calculating the semantic association degree of each patent technology keyword and the corresponding key technology item according to the matching result.

[0167] In step 504, the semantic association degree is the semantic similarity between the patent technology keyword and the key technology item, which is calculated by natural language processing technology.

[0168] In the embodiments of the present application, each patent technology keyword in the technology feature set is semantically matched with a key technology item in the industry chain technology attribute set. The semantic similarity is calculated using a word vector model to obtain the association degree of each keyword and the corresponding item.

[0169] Step 505: associating the product feature set with the product circulation path in a circulation relationship, and calculating the topological coincidence degree of each supply chain label and the corresponding product circulation path according to the association result.

[0170] In step 505, the topological coincidence degree is the matching degree between the supply chain label and the product circulation path, which is calculated by the path similarity algorithm.

[0171] In the embodiment of the present application, each supply chain label in the product feature set is matched with the product circulation path in the product attribute set, the coincidence degree of the business process described by the label and the standard path is calculated, and the topological coincidence degree is obtained.

[0172] Step 506: When there is a key technology item with a semantic correlation degree greater than a preset correlation degree threshold and a product circulation path with a topological coincidence degree greater than a preset coincidence degree threshold, a technology correlation edge and a product circulation edge between the enterprise entity and the industry chain node are generated.

[0173] In step 506, the enterprise entity refers to a data node representing a specific enterprise in the enterprise knowledge graph, which contains the technology feature set and the product feature set of the enterprise, and is connected to the industry chain node through the technology correlation edge and the product circulation edge. The industry chain node refers to a data node representing a specific industry field in the enterprise knowledge graph, which contains a pre-defined industry chain technology attribute set (key technology item) and a product attribute set (typical circulation path), used to represent the technology and product features of the industry field.

[0174] In the embodiment of the present application, when the semantic correlation degree between the key word in the technology feature set and the technology item exceeds the set threshold, and the topological coincidence degree between the label in the product feature set and the product path exceeds the set threshold, the technology correlation edge and the product circulation edge are established between the enterprise entity and the industry chain node.

[0175] Step 507: Forming an enterprise knowledge graph based on the technology correlation edge and the product circulation edge.

[0176] In the embodiment of the present application, based on the generated technology correlation edge and product circulation edge, a graph structure containing all enterprise entities and industry chain nodes is constructed to form a complete enterprise knowledge graph.

[0177] The following is a specific example:

[0178] In the new energy vehicle industry chain scene, for the multi-source data processing of battery manufacturer A, first, the data is divided into benchmark technology feature blocks containing five patent texts and benchmark product feature blocks containing annual product lists and supply chain records through block modeling. From the technology feature blocks, patent technology keywords such as lithium ion battery and energy density are extracted to form a technology feature set, and from the product feature blocks, supply chain labels such as Yangtze River Delta distribution and 48-hour delivery are extracted to form a product feature set. From the industry association rule base, the industry chain technology attribute set in the new energy vehicle field is extracted, including key technology items such as battery energy density greater than or equal to 300 watt-hours per kilogram, and the product attribute set includes product circulation paths such as battery motor and vehicle factory. The energy density keyword in the technology feature set is matched with the battery energy density greater than or equal to 300 watt-hours per kilogram item in the technology attribute set in the dependency relationship, and the semantic correlation degree is calculated by a word vector model, with the formula being where u and v represent the word vectors of the keyword and the item respectively, and the semantic correlation degree 0.92 is calculated. The Yangtze River Delta distribution label in the product feature set is associated with the battery motor and vehicle factory path in the product attribute set in the circulation relationship, and the topological overlap degree is calculated, with the formula being , and the topological overlap degree 0.85 is obtained. The correlation threshold is set to 0.8, the overlap threshold is set to 0.8, the semantic correlation degree 0.92 is greater than the threshold, and the topological overlap degree 0.85 is greater than the threshold, so the technology association edge and the product circulation edge are generated between the battery manufacturer A enterprise entity and the new energy vehicle industry chain node. Based on these edge relationships, an enterprise knowledge graph is constructed, in which nodes represent enterprises and industry fields, edges represent technology association and product circulation relationships, and a complete knowledge graph structure is formed to provide a basis for subsequent analysis.

[0179] In the embodiments of the present application, parallel processing and feature extraction of multi-source data are realized through block modeling, and an enterprise knowledge graph is generated by mapping association of the industry rule base, effectively improving the accuracy and completeness of the knowledge graph construction, and providing a reliable knowledge base for subsequent industry chain analysis.

[0180] To further improve the accuracy of matching new enterprises with industry chain nodes, in some embodiments, step 103: the matching of the new enterprise feature set with the enterprise knowledge graph generates a candidate industry chain node set according to the matching result, comprising:

[0181] Step 601: separating a to-be-matched technology feature set and a to-be-matched product feature set from the new enterprise feature set.

[0182] In step 601, the to-be-matched technology feature set and the to-be-matched product feature set are two types of feature sets separated from the new enterprise feature set, containing technology-related features and product-related features respectively.

[0183] In the embodiments of the present application, the technical features and product features are separated from the new enterprise feature set, the technical features include patent technology keywords and other technical attributes, and the product features include supply chain labels and other product attributes, thereby providing classified feature inputs for subsequent matching.

[0184] Step 602: input the to-be-matched technical feature set, the to-be-matched product feature set, and the enterprise knowledge graph into a preset multi-modal matching network, and perform dependency relationship strength calculation on the to-be-matched technical feature set and the technical attributes of the nodes in the enterprise knowledge graph through a technical matching subnetwork of the multi-modal matching network.

[0185] In step 602, the dependency relationship strength is the correlation degree between the to-be-matched technical feature set and the technical attributes of the nodes in the enterprise knowledge graph, which is obtained through semantic similarity calculation.

[0186] In the embodiments of the present application, the similarity between each technical feature in the to-be-matched technical feature set and the technical attributes of the nodes in the knowledge graph is calculated through the technical matching subnetwork of the multi-modal matching network, and the correlation strength value of each feature and the technical attributes of the nodes is obtained.

[0187] Step 603: perform circulation relationship coincidence degree calculation on the to-be-matched product feature set and the product attributes of the nodes in the enterprise knowledge graph through a product matching subnetwork of the multi-modal matching network.

[0188] In step 603, the circulation relationship coincidence degree is the matching degree between the to-be-matched product feature set and the product attributes of the nodes in the enterprise knowledge graph, which is obtained through path similarity calculation.

[0189] In the embodiments of the present application, the matching degree between each product feature in the to-be-matched product feature set and the product attributes of the nodes in the knowledge graph is calculated through the product matching subnetwork of the multi-modal matching network, and the coincidence degree of each feature and the product attributes of the nodes is obtained.

[0190] Step 604: filter a plurality of candidate industrial chain nodes with a dependency relationship strength greater than a preset strength threshold and a circulation relationship coincidence degree greater than a preset circulation threshold from the enterprise knowledge graph.

[0191] In step 604, the candidate industrial chain node is a node in the knowledge graph that meets both the technical matching and product matching requirements, representing an industrial node with a high correlation with the new enterprise.

[0192] In the embodiments of the present application, nodes with a technical matching strength exceeding a set threshold and a product matching coincidence degree exceeding a set threshold are filtered from the enterprise knowledge graph, and these nodes form a candidate node set.

[0193] Step 605: aggregate all candidate industry chain nodes to form a candidate industry chain node set.

[0194] In the embodiments of the present application, all screened candidate industry chain nodes are aggregated to form a complete node set, providing a basis for subsequent new enterprise industry chain positioning.

[0195] The following is a specific example:

[0196] In the new energy vehicle industry chain scenario, for the feature set processing of new battery enterprise D, first separate the to-be-matched technology feature set from its feature set, which contains two patent technology keywords of high-nickel positive electrode and silicon-carbon negative electrode, and the to-be-matched product feature set contains two supply chain labels of South China warehouse and 72-hour delivery. Input the two feature sets and the enterprise knowledge graph into the multi-modal matching network, the technology matching sub-network calculates the dependency relationship strength between the to-be-matched technology feature set and the node technology attribute, wherein the cosine similarity calculation formula of high-nickel positive electrode and the node technology attribute of battery manufacturer A is , wherein u and v represent the word vectors of keywords and node technology attributes respectively, and the dependency relationship strength is calculated as 0.88, and the dependency relationship strength of silicon-carbon negative electrode and the node technology attribute of A is 0.92. The product matching sub-network calculates the circulation relationship coincidence degree between the to-be-matched product feature set and the node product attribute, wherein the Jaccard similarity calculation formula of South China warehouse and the node product attribute of A is , wherein represents the circulation relationship coincidence degree, and A and B represent the label feature and the node product attribute feature set respectively, and the circulation relationship coincidence degree is calculated as 0.65, and the circulation relationship coincidence degree of 72-hour delivery and the node product attribute of A is 0.70. Set the strength threshold to 0.8 and the circulation threshold to 0.75, although the A node has indicators exceeding the threshold in technology matching, but the product matching does not meet the standard. Continue to match the whole vehicle manufacturer C node, and calculate the dependency relationship strength of high-nickel positive electrode and the node technology attribute of C as 0.86, the dependency relationship strength of silicon-carbon negative electrode and the node technology attribute of C as 0.82, the circulation relationship coincidence degree of South China warehouse and the node product attribute of C as 0.78, and the circulation relationship coincidence degree of 72-hour delivery and the node product attribute of C as 0.75. Since the C node has indicators exceeding the threshold in both technology matching and product matching, the C node is selected into the candidate industry chain node set. Finally, all nodes meeting the conditions are aggregated to form a candidate node set, providing a basis for subsequent industry chain positioning.

[0197] In the embodiments of the present application, the multi-modal matching network is used to realize the feature matching of technology and product in two dimensions, screen out the industry chain nodes with the highest correlation with the new enterprise, effectively improve the accuracy and completeness of the candidate node set, and lay a solid foundation for subsequent industry chain positioning.

[0198] To further improve the accuracy of industry field discrimination, in some embodiments, step 105: the real-time computing the feature similarity between the new enterprise node in the dynamic industry chain and the field core node, based on the feature similarity, combining the business flow intensity between the new enterprise node and the associated nodes in the candidate industry chain node set, to discriminate the industry field of the new enterprise, including:

[0199] Step 701: Calculate the technical feature similarity between the technical feature set of the new enterprise node and the technical attribute set of the field core node.

[0200] In step 701, the technical feature similarity is the matching degree of the new enterprise node and the field core node in the technical feature, reflecting the similarity of the technical direction of the two.

[0201] In the embodiments of the present application, the similarity between the technical feature set of the new enterprise node and the technical attribute set of the field core node is calculated, and the technical matching degree of the two is quantified by comparing the distribution characteristics of the patent technology keywords using the similarity algorithm.

[0202] Step 702: Calculate the product feature similarity between the product feature set of the new enterprise node and the product attribute set of the field core node.

[0203] In step 702, the product feature similarity is the matching degree of the new enterprise node and the field core node in the product feature, reflecting the similarity of the product characteristics of the two.

[0204] In the embodiments of the present application, the similarity between the product feature set of the new enterprise node and the product attribute set of the field core node is calculated, and the product matching degree of the two is quantified by comparing the matching of the supply chain label and the product characteristics using the similarity algorithm.

[0205] Step 703: Calculate the business flow density value between the new enterprise node and the associated nodes in the candidate industry chain node set.

[0206] In step 703, the business flow density value is the business transaction intensity between the new enterprise node and the associated nodes in the candidate industry chain node set, reflecting the actual business association degree.

[0207] In the embodiments of the present application, the business transaction data between the new enterprise node and each associated node in the candidate node set is counted, including order quantity, transaction frequency and other indicators, and the business flow density value is obtained by quantitative calculation.

[0208] Step 704: Use the attention mechanism to first weight and fuse the technical feature similarity and the business flow density value, and second weight and fuse the product feature similarity and the business flow density value.

[0209] In step 704, the first weighted fusion and the second weighted fusion are processes of organically combining the feature similarity with the business flow density through an attention mechanism.

[0210] In the embodiments of the present application, the attention mechanism is used to assign weights to the technical feature similarity and the business flow density value, and to obtain a first fusion result through weighted fusion, and to assign weights to the product feature similarity and the business flow density value, and to obtain a second fusion result through weighted fusion.

[0211] Step 705: When the first weighted fusion result is greater than a preset first threshold and the second weighted fusion result is greater than a preset second threshold, it is determined that the new enterprise belongs to the industrial field corresponding to the field core node.

[0212] In step 705, the first threshold and the second threshold are preset discrimination criteria for the final determination of the industrial field.

[0213] In the embodiments of the present application, the first weighted fusion result is compared with the first threshold, and the second weighted fusion result is compared with the second threshold, and when both fusion results exceed the corresponding threshold, it is determined that the new enterprise belongs to the industrial field corresponding to the field core node.

[0214] The following is a specific example:

[0215] In the new energy automobile industry chain scene, the technical feature set of the new battery enterprise D node contains two patent technology keywords of high nickel positive electrode and silicon carbon negative electrode, and the technical attribute set of the field core node battery manufacturer A contains two keywords of lithium ion battery and energy density. The technical feature similarity between the two is calculated by using the cosine similarity formula , wherein represents the technical feature similarity, represents the number of feature dimensions, represents the first component of the technical feature vector of the new enterprise node, represents the first component of the technical attribute vector of the field core node, ​The technical feature similarity is calculated to be 0.85. The product feature set of enterprise D contains two supply chain tags of South China warehouse and 72-hour delivery, and the product attribute set of node A contains two tags of Yangtze River Delta delivery and 48-hour delivery. The product feature similarity between the two sets is calculated to be 0.75 by using the Jaccard similarity formula. The business flow density value of enterprise D and the associated node of the candidate node set, whole vehicle manufacturer C, is calculated to be 0.8 by dividing the quarterly order quantity of 0.5 million by the industry benchmark value of 6.25 million. The weighted fusion is performed by using the attention mechanism, the technical feature similarity is assigned a weight of 0.6, the business flow density value is assigned a weight of 0.4, the first weighted fusion result is 0.85*0.6+0.8*0.4=0.83, the product feature similarity is assigned a weight of 0.5, the business flow density value is assigned a weight of 0.5, and the second weighted fusion result is 0.75*0.5+0.8*0.5=0.775. The first threshold is set to 0.8, the second threshold is set to 0.7, the first weighted fusion result 0.83 is greater than the first threshold, and the second weighted fusion result 0.775 is greater than the second threshold, so it is determined that the new battery enterprise D belongs to the new energy automobile industry field corresponding to the field core node battery manufacturer A.

[0216] In the embodiments of the present application, by comprehensively calculating the feature similarity and the business flow density and using the attention mechanism for weighted fusion, the technical and product dual dimensions and the actual business are considered in multiple ways, which effectively improves the accuracy and reliability of the industry field judgment, and ensures that the judgment result reflects both the technical product features and the actual business association.

[0217] Figure 3 The structure diagram of an enterprise belonging to an industry field judgment system based on an industry chain provided in the embodiments of the present application is described in the specific implementation manner part:

[0218] The receiving module 31 is configured to receive the declaration data of a new enterprise, and multi-source data of a target enterprise and associated enterprises.

[0219] The association module 32 is configured to perform parallel processing on the multi-source data by using a block model, form an enterprise feature set, map and associate the enterprise feature set with a preset industry association rule library, and generate an enterprise knowledge graph. The enterprise feature set includes a technical feature set and a product feature set.

[0220] The extraction module 33 is configured to extract a new enterprise feature set from the new enterprise declaration data, match the new enterprise feature set with the enterprise knowledge graph, and generate a candidate industry chain node set according to the matching result.

[0221] The selection module 34 is configured to calculate a topological connection density between each pair of nodes in the candidate industrial chain node set, select a pair of nodes with the maximum topological connection density from the candidate industrial chain node set as an anchor node pair, perform feature extraction on a technology association edge and a product circulation edge of the anchor node pair through a graph neural network to generate an enhanced edge vector, generate an offset coefficient of a new node addition position according to the enhanced edge vector, add a new enterprise as a new enterprise node, add the new enterprise node at the addition position corresponding to the offset coefficient in the enterprise knowledge graph, and predict a potential interaction relationship between the new enterprise node and the anchor node pair through a relationship reasoning model to generate a new technology association edge and a new product circulation edge, and generate a dynamic industrial chain based on the new technology association edge and the new product circulation edge.

[0222] The calculation module 35 is configured to calculate a feature similarity between the new enterprise node and a domain core node in the dynamic industrial chain in real time, and determine an industrial domain of the new enterprise based on the feature similarity and a business flow intensity between the new enterprise node and an associated node in the candidate industrial chain node set.

[0223] The industrial chain-based enterprise industrial domain determination system according to the embodiments of the present application is used to implement the industrial chain-based enterprise industrial domain determination method described above, and therefore the specific embodiments of the industrial chain-based enterprise industrial domain determination system can be seen from the foregoing embodiments of the industrial chain-based enterprise industrial domain determination method, and the specific embodiments can be referred to the descriptions of the corresponding embodiments, which will not be repeated here.

[0224] The present application also provides an electronic device, which comprises a memory for storing a computer program and a processor for executing the computer program to implement the steps of the industrial chain-based enterprise industrial domain determination method described above.

[0225] The present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the industrial chain-based enterprise industrial domain determination method described above.

[0226] In an exemplary embodiment, the computer readable storage medium described above can include but is not limited to a U disk, a read-only memory, a random access memory, a mobile hard disk, a magnetic disk or an optical disk, and various media that can store computer programs.

[0227] The embodiments of the present application also provide a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps of the industrial chain-based enterprise industrial domain determination method described above.

[0228] Those skilled in the art will further appreciate that the units and algorithms described in connection with the examples disclosed herein can be embodied in electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various examples disclosed herein have been described generally in terms of their functionality, without reference to the corresponding structure. Those skilled in the art will appreciate that the described functionality can be implemented in either hardware or software, or any combination thereof, depending on the particular application and design constraints. It will also be appreciated that the described functionality can be implemented using different amounts of hardware and software depending on the application and design constraints. For example, where long battery life is important, more functionality can be implemented in hardware, while where short battery life is not important, more functionality can be implemented in software.

[0229] The above provides a kind of based on industry chain's enterprise belonging to industry field discrimination method, system, electronic equipment and storage medium of enterprise, which are provided in the application.The principle and implementation mode of the present application are described in this paper by applying specific examples.The above example is only used to help understand the method and its core idea of the present application.It should be pointed out that for ordinary skilled in the art, without departing from the principle of the present application, some improvements and modifications can be made to the present application, and these improvements and modifications also fall within the protection scope of the present application.

Claims

1. An industry chain-based method for determining the industry field to which an enterprise belongs, characterized by comprising: The method comprises the following steps: receiving declaration data of a new enterprise and multi-source data of a target enterprise and associated enterprises; performing parallel processing on the multi-source data through a block model to form an enterprise feature set, mapping and associating the enterprise feature set with a preset industry association rule library to generate an enterprise knowledge graph, wherein the enterprise feature set comprises a technology feature set and a product feature set; extracting a new enterprise feature set from the declaration data of the new enterprise, matching the new enterprise feature set with the enterprise knowledge graph, and generating a candidate industry chain node set according to a matching result; calculating a topological connection density between each pair of nodes in the candidate industry chain node set, selecting a node pair with the largest topological connection density from the candidate industry chain node set as an anchor node pair, performing feature extraction on a technology association edge and a product flow edge of the anchor node pair through a graph neural network to generate an enhanced edge vector, generating an offset coefficient of a new node addition position according to the enhanced edge vector, and adding a new enterprise as a new enterprise node at the addition position corresponding to the offset coefficient in the enterprise knowledge graph, and predicting a potential interaction relationship between the new enterprise node and the anchor node pair through a relationship reasoning model to generate a new technology association edge and a new product flow edge, and generating a dynamic industry chain based on the new technology association edge and the new product flow edge; calculating a feature similarity between the new enterprise node and a field core node in the dynamic industry chain in real time, and determining an industry field to which the new enterprise belongs based on the feature similarity and a business flow intensity between the new enterprise node and an associated node in the candidate industry chain node set.

2. The method according to claim 1, wherein The method comprises the following steps: inputting the technology association edge of the anchor node pair into a first edge processing channel of the graph neural network, and extracting a technology association feature vector from the technology association edge of the anchor node pair through the first edge processing channel; inputting the product flow edge of the anchor node pair into a second edge processing channel of the graph neural network, and extracting a product flow feature vector from the product flow edge of the anchor node pair through the second edge processing channel; performing weighted splicing on the technology association feature vector and the product flow feature vector through a cross-channel attention fusion module of the graph neural network to generate an enhanced edge vector; inputting the enhanced edge vector into an industry chain space projection layer of the graph neural network to output a technology-intensive direction projection component and a flow hub direction projection component; calculating a direction fusion factor based on a ratio of the technology-intensive direction projection component to the flow hub direction projection component; looking up an offset coefficient of a new node addition position in a preset offset mapping table according to the direction fusion factor.

3. The method according to claim 2, wherein The method comprises the following steps: a technology dynamic influence factor is calculated based on the semantic diffusion path depth of the technology correlation feature vector and the patent technology keyword coverage; a circulation dynamic influence factor is calculated based on the logistics track hop count of the product circulation feature vector and the space-time aggregation radius; the technology dynamic influence factor and the circulation dynamic influence factor are normalized respectively to obtain a technology attention weight value and a circulation attention weight value; the technology correlation feature vector and the product circulation feature vector are weighted and scaled respectively based on the technology attention weight value and the circulation attention weight value to generate a weighted technology feature block and a weighted circulation feature block; a fusion feature matrix is generated by performing multi-scale feature fusion on the weighted technology feature block and the weighted circulation feature block using a feature pyramid network; dimensionality reduction compression processing is performed on the fusion feature matrix to output an enhanced edge vector.

4. The method according to claim 3, wherein the multi-scale feature fusion on the weighted technology feature block and the weighted circulation feature block using the feature pyramid network to generate the fusion feature matrix comprises: a feature pyramid network including a first pyramid layer, a second pyramid layer, and a third pyramid layer is constructed; the weighted technology feature block and the weighted circulation feature block are input into the feature pyramid network to output a technology multi-scale feature group including three scales and a circulation multi-scale feature group including three scales; cross-modal feature fusion is performed on the technology multi-scale feature group and the circulation multi-scale feature group, wherein the first pyramid layer adopts element-by-element addition fusion, the second pyramid layer adopts channel attention weighted fusion, and the third pyramid layer adopts spatial attention weighted fusion; the multi-scale fused feature groups are unified to the original resolution through upsampling operation to generate multi-scale fusion features; based on the industry type configuration parameters, the multi-scale fusion features are rearranged into a two-dimensional matrix structure according to a preset dimension ratio to form a fusion feature matrix.

5. The method according to claim 1, wherein the parallel processing of the multi-source data through the block model to form an enterprise feature set, the mapping and association of the enterprise feature set with a preset industry correlation rule library to generate an enterprise knowledge graph, comprises: the multi-source data is divided into a baseline technology feature block and a baseline product feature block through the block model; a plurality of patent technology keywords are extracted from the baseline technology feature block to form a technology feature set, and a plurality of supply chain tags are extracted from the baseline product feature block to form a product feature set; predefined industry chain technology attribute sets and product attribute sets are extracted from the industry correlation rule library, wherein the industry chain technology attribute sets include corresponding key technology items of each industry field, and the product attribute sets include corresponding product circulation paths of each industry field; the technology feature set is matched with the key technology items in a dependency relationship, and the semantic correlation degree of each patent technology keyword and the corresponding key technology item is calculated according to the matching result; the product feature set is associated with the product circulation paths in a circulation relationship, and the topological coincidence degree of each supply chain tag and the corresponding product circulation path is calculated according to the association result; When the key technology entry with the semantic correlation degree greater than the preset correlation degree threshold and the product circulation path with the topology coincidence degree greater than the preset coincidence degree threshold exist, a technology correlation edge and a product circulation edge between the enterprise entity and the industry chain node are generated; Based on the technology correlation edge and the product circulation edge, an enterprise knowledge graph is formed.

6. The method according to claim 1, wherein The matching of the new enterprise feature set and the enterprise knowledge graph includes: Separating a to-be-matched technology feature set and a to-be-matched product feature set from the new enterprise feature set; Inputting the to-be-matched technology feature set, the to-be-matched product feature set, and the enterprise knowledge graph into a preset multi-modal matching network, performing dependency relationship strength calculation on the to-be-matched technology feature set and the technology attribute of the node in the enterprise knowledge graph by a technology matching subnetwork of the multi-modal matching network; Performing circulation relationship coincidence degree calculation on the to-be-matched product feature set and the product attribute of the node in the enterprise knowledge graph by a product matching subnetwork of the multi-modal matching network; Filtering a plurality of candidate industry chain nodes with the dependency relationship strength greater than a preset strength threshold and the circulation relationship coincidence degree greater than a preset circulation threshold from the enterprise knowledge graph; Aggregating all the candidate industry chain nodes to form a candidate industry chain node set.

7. The method according to claim 1, wherein the method is characterized by: The real-time calculation of the feature similarity between the new enterprise node and the field core node in the dynamic industry chain includes: Calculating the technology feature similarity between the technology feature set of the new enterprise node and the technology attribute set of the field core node; Calculating the product feature similarity between the product feature set of the new enterprise node and the product attribute set of the field core node; Calculating the business flow density value between the new enterprise node and the associated nodes in the candidate industry chain node set; Performing first weighted fusion of the technology feature similarity and the business flow density value by using an attention mechanism, and performing second weighted fusion of the product feature similarity and the business flow density value; When the first weighted fusion result is greater than a preset first threshold and the second weighted fusion result is greater than a preset second threshold, it is determined that the new enterprise belongs to the industry field corresponding to the field core node.

8. An industry chain-based enterprise industry field determination system, characterized by comprising: The method includes: The receiving module is configured to receive the declaration data of the new enterprise and the multi-source data of the target enterprise and the associated enterprises; The association module is configured to perform parallel processing on the multi-source data by using a block model, form an enterprise feature set, map and associate the enterprise feature set with a preset industry association rule library, and generate an enterprise knowledge graph, wherein the enterprise feature set includes a technology feature set and a product feature set; The extraction module is configured to extract a new enterprise feature set from the new enterprise declaration data, match the new enterprise feature set with the enterprise knowledge graph, and generate a candidate industry chain node set according to a matching result. The selection module is configured to calculate a topological connection density between each pair of nodes in the candidate industrial chain node set, select a pair of nodes with the maximum topological connection density from the candidate industrial chain node set as an anchor node pair, perform feature extraction on a technology association edge and a product circulation edge of the anchor node pair through a graph neural network to generate an enhanced edge vector, generate an offset coefficient of a new node addition position according to the enhanced edge vector, add a new enterprise as a new enterprise node, add the new enterprise node at a coordinate corresponding to the offset coefficient as an addition position in the enterprise knowledge graph, and predict a potential interaction relationship between the new enterprise node and the anchor node pair through a relationship reasoning model to generate a new technology association edge and a new product circulation edge, and generate a dynamic industrial chain based on the new technology association edge and the new product circulation edge. The calculation module is configured to calculate a feature similarity between the new enterprise node and a domain core node in the dynamic industrial chain in real time, and determine an industrial domain of the new enterprise based on the feature similarity and a business flow intensity between the new enterprise node and an associated node in the candidate industrial chain node set.

9. An electronic device, comprising: The method comprises the following steps: a memory configured to store a computer program; a processor configured to execute the computer program to implement the steps of the industrial chain-based enterprise industrial domain determination method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the industrial chain-based enterprise industrial domain determination method according to any one of claims 1 to 7.

Images