Enterprise portrait automatic generation system and method based on multi-source data fusion

Abstract

The present invention discloses a system and method for automatically generating enterprise portraits based on multi-source data fusion, relating to the technical field of intelligent enterprise portrait generation. The method specifically includes the following steps: heterogeneous data collection; construction of a multidimensional graph; incremental updating; extraction of structural features of the multidimensional graph based on regularly updated multidimensional graphs to generate enterprise portraits; real-time acquisition of implicit features of the multidimensional graph; calculation and setting of dynamic tags based on the implicit features, supplementing the implicit data of the enterprise portraits through dynamic tags; and visual display of the generated enterprise portraits. This method utilizes an exponential decay model to update the association weights in the graph in real time, effectively resolving the static and time-sensitive nature of traditional enterprise portraits and significantly enhancing the system's ability to perceive and respond to dynamic changes in the enterprise.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent enterprise portrait generation, and specifically to a system and method for automatically generating enterprise portraits based on multi-source data fusion. Background Art

[0002] Intelligent enterprise portrait generation refers to the collection, integration, and modeling of multi-source heterogeneous data of an enterprise to build a multi-dimensional, structured, visual, and interactive digital enterprise portrait system, and realize dynamic updates and intelligent analysis;

[0003] With the rapid development of big data, artificial intelligence, and knowledge graph technologies, corporate profiling is playing an increasingly important role in scenarios such as financial risk management, industry chain analysis, and market insights. Traditional corporate profiling methods primarily rely on structured data, such as business registration information, financial statements, and transaction records. These methods are often static and single-dimensional, making it difficult to reflect the multidimensional characteristics and dynamic evolution of a company's actual operations. However, real-world business operations involve a large number of heterogeneous data sources, such as regulatory records, news and public opinion, patent applications, supply chain collaboration platform data, and bidding documents. These data types are diverse and complex in format. The large structural differences and semantic expressions between these multi-source data make it difficult to uniformly parse and establish effective associations. Existing systems often use isolated indicators to describe enterprise characteristics and lack a unified entity relationship modeling approach, making it difficult to reflect the complex network structure and influence between enterprises, such as supply and demand, shareholding, and technological dependence. Enterprise portraits are often constructed based on snapshots at a specific point in time, lacking the ability to track real-time data and incrementally update the graph, and are unable to reflect the evolving characteristics of enterprise relationships and potential changes in risks. Traditional portraits focus on displaying basic information, such as legal person, registered capital, and business scope, but fail to fully characterize and evaluate implicit characteristics such as the role of the enterprise in the industrial chain, associated risks, and disruption risks. Therefore, there is an urgent need for an enterprise portrait generation method that can integrate multi-source heterogeneous data, construct a multidimensional association graph, and support dynamic updates and implicit feature calculation. Summary of the Invention

[0004] The purpose of the present invention is to provide a system and method for automatically generating enterprise portraits based on multi-source data fusion to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for automatically generating enterprise portraits based on multi-source data fusion, the method specifically comprising the following steps:

[0006] Step S100: heterogeneous data collection, performing knowledge extraction and alignment on the collected heterogeneous data;

[0007] Step S200: Determine data entities in the heterogeneous data as nodes, determine knowledge extraction results as association relationships, calculate association weights of different association relationships, and construct a multidimensional graph based on the nodes, association relationships, and association weights of different association relationships;

[0008] Step S300: Setting an incremental update mechanism for the constructed multi-dimensional graph to regularly update nodes, relationships, and association weights of different relationships;

[0009] Step S400: Extracting structural features of the multidimensional graph based on the regularly updated multidimensional graph to generate a corporate profile, acquiring implicit features of the multidimensional graph in real time, calculating and setting dynamic tags based on the implicit features, and supplementing the implicit data of the corporate profile through dynamic tags;

[0010] Step S500: Visually display the generated enterprise portrait.

[0011] In step S100, heterogeneous data collection is performed, specifically:

[0012] Classify and collect heterogeneous data, including structured data, semi-structured data, and unstructured data;

[0013] The structured data includes business registrations, tax records, patent databases, and supply chain system logs;

[0014] The semi-structured data includes bidding documents and company annual reports;

[0015] The unstructured data includes news, public opinion and industry research reports.

[0016] Perform knowledge extraction and alignment on the collected heterogeneous data, specifically:

[0017] For structured data, directly extract data entities;

[0018] For semi-structured data, data entities are extracted through manual format parsing and keyword matching;

[0019] For unstructured data, a method based on the BiLSTM-CRF model is used to extract enterprise, product, and legal person data entities;

[0020] Knowledge extraction based on semantic relations of dependency syntax;

[0021] Unify time and location in heterogeneous data from different sources into a standardized format.

[0022] Optionally, the method for unifying time and place includes time-space correlation alignment, specifically:

[0023] Enterprise-Geographic Binding:

[0024] Linking businesses to GeoHash through registered address entity data;

[0025] Event-time binding:

[0026] Associate data entities of supply chain events with standardized timestamps (UTC time);

[0027] In step S200, data entities in heterogeneous data are determined as nodes, knowledge extraction results are determined as association relationships, association weights of different association relationships are calculated, and a multidimensional graph is constructed based on the nodes, association relationships, and association weights of different association relationships. Specifically,

[0028] Step S201: Determine data entities as nodes, including enterprises, products, and legal persons;

[0029] Step S202: Determine the knowledge extraction results as association relationships, where the association relationships include enterprise-production-product, enterprise-supply-enterprise, enterprise-holding-legal person, product-dependency-product, and geographic location-aggregation-enterprise;

[0030] Preferably, the node is represented as follows:

[0031] Enterprise: business registration number, industry classification code;

[0032] Product: HS code, technical complexity level;

[0033] Transaction: transaction amount, frequency, timestamp;

[0034] Investment: shareholding ratio, investment amount;

[0035] Legal person: depth of shareholding chain, control coefficient;

[0036] Geographical location: longitude and latitude, economic location entropy;

[0037] Step S203: Calculate the association weights of different association relationships by fitting the functional relationship with historical data;

[0038] The correlation weight calculation of the transaction relationship and the fitted functional relationship are represented as follows:

[0039] ; Among them, w ij represents the association weight from node i to node j, where i and j represent different node identifiers; a ij represents the annual transaction volume of node i to node j; a total represents the total transaction volume of the industry; z represents the annual transaction frequency;

[0040] Among them, the investment relationship expresses the association weight through the shareholding ratio or investment amount;

[0041] For legal person relationships, product supply relationships, etc., it is only necessary to indicate their association relationships for easy viewing, without calculating the association weights;

[0042] Among them, according to the actual situation and the update cycle, the time period of transaction volume is determined, and the total industry transaction volume and transaction frequency in the said time period are adjusted accordingly;

[0043] Step S204: Based on the nodes, association relationships, and association weights of different association relationships, the nodes are connected through the association relationships, and the association weights are marked to form a multi-dimensional graph.

[0044] In step S300, an incremental update mechanism is set for the constructed multi-dimensional graph to regularly update nodes, relationships, and association weights of different relationships, specifically:

[0045] Regularly update nodes, relationships, and the association weights of different relationships;

[0046] In the continuous update cycle, the exponential decay model is used to update the association weight in real time. The specific steps are as follows:

[0047] Step S301: Query the current association weight w t−1 and the last update time t last ;

[0048] Step S302: Real-time acquisition of the update period time interval Δt=t current −t last ; where t current Indicates the current time;

[0049] Step S303: Use an exponential decay model, superimpose the newly added weights, and apply the decay formula to update the weights in real time. Specifically,

[0050] ;

[0051] Among them, w t represents the updated association weight; w t−1 represents the current association weight; e represents the natural constant; λ represents the attenuation coefficient; Δt represents the time interval of the update cycle; Δw new Indicates the newly added weight;

[0052] Optionally, the newly added weight Δw new By the method in step S203, the updated parameters are calculated;

[0053] Optionally, the newly added weight Δw new It can also be represented by a normalized value of the transaction amount within the time interval of the update cycle; the normalization method adopts the Max-Min method.

[0054] In step S400, the structural features of the multi-dimensional atlas are extracted based on the regularly updated multi-dimensional atlas, specifically:

[0055] The structural features of the multidimensional graph include node type, association relationship type and node feature vector;

[0056] The node types include enterprise, product and legal person;

[0057] The types of relationships include supply chain relationships (supply, procurement), investment relationships (holding, equity participation) and technology dependency relationships (patent citations);

[0058] The node feature vector includes:

[0059] Enterprise node: registered capital, number of patents and revenue growth rate data.

[0060] Product node: technical complexity (classified by HS code) and annual production data.

[0061] Legal person node: data on shareholding ratio and number of affiliated companies.

[0062] Generate an enterprise portrait through the structural characteristics of the multi-dimensional map, showing explicit information such as the enterprise's legal person, products, relationships with other enterprises, and technological dependence;

[0063] Acquire the implicit features of the multi-dimensional graph in real time, calculate and set dynamic tags based on the implicit features, and supplement the implicit data of the enterprise portrait through dynamic tags;

[0064] The implicit features of a multidimensional graph include the paths between nodes and the association weights between nodes;

[0065] Dynamic labels include a unilateral fracture risk label;

[0066] The unilateral fracture risk label can be used to show the stability of the enterprise in the later cooperation in real time;

[0067] The specific steps for calculating the unilateral fracture risk are as follows:

[0068] Step S401: Obtain the shortest paths of all node pairs and calculate the centrality of the nodes, specifically:

[0069] C i =Σ s≠i≠t [(σ st_i ) / (σ st )];

[0070] Among them, C i represents the node centrality of node i; s, i and t represent different node identifiers; σ st_irepresents the total number of shortest paths from node s to node t passing through node i; σ st represents the total number of shortest paths from node s to node t;

[0071] Step S402: Find all paths from node i to node j and obtain all paths that do not pass through edge e. ij The path, and through the association weights between the nodes in different paths, the weight of each path is obtained, and the nodes that do not pass through the edge e ij The ratio of the sum of the weights of the paths to the sum of the weights of all paths is used as the edge redundancy estimate;

[0072] The specific characteristics are:

[0073] ;

[0074] Among them, ρ ij Represents edge e ij Redundancy estimation; p represents the path identifier; It means starting from node i and ending at node j without passing through edge e. ij Path; represents all paths starting from node i and ending at node j; represents the node identifier in path p; w uv represents the association weight between node u and node v;

[0075] Step S403: Calculate the risk of a single edge break based on the centrality and edge redundancy of the node.

[0076] The calculation of the unilateral fracture risk is specifically characterized as follows:

[0077] R ij =w ij (t)×(C i +C j )×(1-ρ ij );

[0078] Among them, R ij Represents edge e ij Unilateral fracture risk; ij (t) represents the real-time association weight between node i and node j; C j represents the node centrality of node j;

[0079] The enterprise portrait generated by the visual display is as follows:

[0080] Display the equity structure through a tree model;

[0081] Display the business scope and search keywords of the enterprise through word cloud;

[0082] Update dynamic tags in real time.

[0083] An automated enterprise portrait generation system based on multi-source data fusion, comprising a data acquisition module, a data processing module, an entity relationship modeling module, a graph incremental update module, a feature extraction and portrait generation module, and a visualization display module;

[0084] The data collection module is used to collect heterogeneous enterprise-related data and collect them by categories, including structured data, semi-structured data and unstructured data;

[0085] The data processing module is used to perform knowledge extraction and semantic alignment on the collected heterogeneous data, and extract data entity and relationship information;

[0086] The entity relationship modeling module is used to determine data entities in heterogeneous data as nodes, determine knowledge extraction results as association relationships, calculate association weights of different association relationships, and construct a multidimensional graph based on the nodes, association relationships and association weights of different association relationships;

[0087] The graph incremental update module is used to regularly update nodes, associations, and association weights of different associations; in continuous update cycles, the association weights are updated in real time using an exponential decay model;

[0088] The feature extraction and portrait generation module is used to extract the structural features of the multidimensional map based on the regularly updated multidimensional map, generate the enterprise portrait, obtain the implicit features of the multidimensional map in real time, calculate and set dynamic tags based on the implicit features, and supplement the implicit data of the enterprise portrait through dynamic tags;

[0089] The visualization display module is used to display the enterprise portrait and the implicit data calculated based on the implicit features of the multi-dimensional map in different ways.

[0090] Compared with the prior art, the present invention has the following beneficial effects:

[0091] 1. This method achieves accurate extraction and semantic relationship alignment of core data entities such as enterprises, legal persons, and products by categorizing and collecting structured, semi-structured, and unstructured data and processing them in layers, combined with natural language processing technologies such as the BiLSTM-CRF model and dependency parsing, effectively improving the comprehensiveness and accuracy of data fusion.

[0092] 2. This method models entity relationships as nodes and edges, introduces multiple semantic relationships such as supply, holding, and technology dependence, and combines them with quantitative indicators such as transaction weight and investment ratio to form a semantically rich, clearly structured, and scalable multi-dimensional graph, providing comprehensive structural support for enterprise portraits.

[0093] 3. This method uses an exponential decay model to update the association weights in the graph in real time, effectively solving the problem of static and time-sensitive traditional enterprise portraits, and significantly enhancing the system's ability to perceive and respond to dynamic changes in the enterprise;

[0094] 4. This method uses the structural characteristics of the graph to depict explicit information, while mining implicit structural characteristics such as path centrality and edge redundancy, further calculating dynamic labels to enhance risk warning and intelligent analysis capabilities;

[0095] 5. This invention uses tree diagrams, word clouds, and other methods to visualize legal person structures, business scopes, and dynamic tags, allowing users to intuitively understand the operational characteristics of an enterprise and its position in the industry chain from a multi-dimensional perspective, thereby enhancing the practical value of enterprise portraits and their ability to support business decision-making.

[0096] 6. This invention uses the standardized means of GeoHash geocoding and UTC timestamps to achieve unified expression of enterprise, event, and temporal and spatial information, providing basic support for cross-regional and cross-period data alignment and event reasoning. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1 Schematic diagram of the process of the method for automatically generating enterprise portraits based on multi-source data fusion according to the present invention;

[0098] Figure 2 This is a structural diagram of the automatic enterprise portrait generation system based on multi-source data fusion of the present invention. DETAILED DESCRIPTION

[0099] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0100] Example: Figure 1-Figure 2 As shown, the present invention provides a technical solution, a method for automatically generating enterprise portraits based on multi-source data fusion, and the method specifically includes the following steps:

[0101] Step S100: heterogeneous data collection, performing knowledge extraction and alignment on the collected heterogeneous data;

[0102] Step S200: Determine data entities in heterogeneous data as nodes, determine knowledge extraction results as association relationships, calculate association weights of different association relationships, and construct a multidimensional graph based on the nodes, association relationships, and association weights of different association relationships;

[0103] Step S300: Setting an incremental update mechanism for the constructed multi-dimensional graph to regularly update nodes, relationships, and association weights of different relationships;

[0104] Step S400: Extracting structural features of the multidimensional graph based on the regularly updated multidimensional graph to generate a corporate profile, acquiring implicit features of the multidimensional graph in real time, calculating and setting dynamic tags based on the implicit features, and supplementing the implicit data of the corporate profile through dynamic tags;

[0105] Step S500: Visually display the generated enterprise portrait.

[0106] In step S100, heterogeneous data collection is performed, specifically:

[0107] Classify and collect heterogeneous data, including structured data, semi-structured data, and unstructured data;

[0108] Structured data includes business registries, tax records, patent databases, and supply chain system logs;

[0109] Semi-structured data includes bidding documents and company annual reports;

[0110] Unstructured data includes news, public opinion, and industry research reports.

[0111] Perform knowledge extraction and alignment on the collected heterogeneous data, specifically:

[0112] For structured data, directly extract data entities;

[0113] For semi-structured data, data entities are extracted through manual format parsing and keyword matching;

[0114] For unstructured data, a method based on the BiLSTM-CRF model is used to extract enterprise, product, and legal person data entities;

[0115] Knowledge extraction based on semantic relations of dependency syntax;

[0116] S1. Generate a dependency tree using the spaCy tool, extract dependency paths, and determine the main path.

[0117] S2. Manually pre-formulate dependency template matching rules, including controlling relationship dependency template matching rules and supply relationship dependency template matching rules;

[0118] S3. Extract the verbs ("invest", "supply"), entity types, and path lengths in the dependency path; obtain the bag-of-words context features ("purchase", "holding"); and perform BERT-based sequence classification, using the dependency path context as input:

[0119] S4, taking the paths between the data entities involved in the trunk path as the knowledge extraction result;

[0120] Unify time and location in heterogeneous data from different sources into a standardized format.

[0121] The method for unifying time and place includes time-space correlation alignment, specifically:

[0122] Enterprise-Geographic Binding:

[0123] Linking businesses to GeoHash through registered address entity data;

[0124] Event-time binding:

[0125] Associate data entities of supply chain events with standardized timestamps (UTC time);

[0126] In step S200, the data entities in the heterogeneous data are determined as nodes, the knowledge extraction results are determined as association relationships, the association weights of different association relationships are calculated, and a multidimensional graph is constructed based on the nodes, association relationships, and association weights of different association relationships. Specifically,

[0127] Step S201: Determine data entities as nodes, which include enterprises, products, and legal persons;

[0128] Step S202: Determine the knowledge extraction results as association relationships, including enterprise-production-product, enterprise-supply-enterprise, enterprise-holding-legal person, product-dependency-product, and geographic location-aggregation-enterprise;

[0129] The node representation is specifically:

[0130] Enterprise: business registration number, industry classification code;

[0131] Product: HS code, technical complexity level;

[0132] Transaction: transaction amount, frequency, timestamp;

[0133] Investment: shareholding ratio, investment amount;

[0134] Legal person: depth of shareholding chain, control coefficient;

[0135] Geographical location: longitude and latitude, economic location entropy;

[0136] Step S203: Calculate the association weights of different association relationships by fitting the functional relationship with historical data;

[0137] The correlation weight calculation of the transaction relationship and the fitted functional relationship are represented as follows:

[0138] ; Among them, w ij represents the association weight from node i to node j, where i and j represent different node identifiers; a ij represents the annual transaction volume of node i to node j; a total represents the total transaction volume of the industry; z represents the annual transaction frequency;

[0139] Among them, the investment relationship expresses the association weight through the shareholding ratio or investment amount;

[0140] For legal person relationships, product supply relationships, etc., it is only necessary to indicate their association relationships for easy viewing, without calculating the association weights;

[0141] Among them, according to the actual situation and the update cycle, the time period of transaction volume is determined, and the total industry transaction volume and transaction frequency in the said time period are adjusted accordingly;

[0142] Optionally, the location entropy index is calculated as:

[0143] LQ i =(E ir / E r ) / (E i / E); where LQ i represents the location entropy index of node i; E ir represents the flow of people in area r where node i is located; E r represents the total flow of people in area r; E i represents the total number of employees in the industry corresponding to node i; E represents the total number of employees in the country;

[0144] Step S204: Based on the nodes, association relationships and association weights of different association relationships, the nodes are connected through the association relationships and the association weights are marked to form a multi-dimensional graph.

[0145] In step S300, an incremental update mechanism is set for the constructed multi-dimensional graph to regularly update nodes, relationships, and association weights of different relationships, specifically:

[0146] Regularly update nodes, relationships, and the association weights of different relationships;

[0147] In the continuous update cycle, the exponential decay model is used to update the association weight in real time. The specific steps are as follows:

[0148] Step S301: Query the current association weight w t−1 and the last update time t last ;

[0149] Step S302: Real-time acquisition of the update period time interval Δt=t current −t last; where t current Indicates the current time;

[0150] Step S303: Use an exponential decay model, superimpose the newly added weights, and apply the decay formula to update the weights in real time. Specifically,

[0151] ;

[0152] Among them, w t represents the updated association weight; w t−1 represents the current association weight; e represents the natural constant; λ represents the attenuation coefficient; Δt represents the time interval of the update cycle; Δw new Indicates the newly added weight;

[0153] New weight Δw new By the method in step S203, the updated parameters are calculated;

[0154] The newly added weight Δw new It can also be represented by a normalized value of the transaction amount within the time interval of the update cycle; the normalization method adopts the Max-Min method.

[0155] In step S400, the structural features of the multi-dimensional atlas are extracted based on the regularly updated multi-dimensional atlas, specifically:

[0156] The structural features of the multidimensional graph include node type, relationship type and node feature vector;

[0157] Node types include enterprise, product, and legal person;

[0158] The types of related relationships include supply chain relationships (supply, procurement), investment relationships (holding, equity participation) and technology dependency relationships (patent citations);

[0159] The node feature vector includes:

[0160] Enterprise node: registered capital, number of patents and revenue growth rate data.

[0161] Product node: technical complexity (classified by HS code) and annual production data.

[0162] Legal person node: data on shareholding ratio and number of affiliated companies.

[0163] Generate enterprise portraits through the structural characteristics of multi-dimensional graphs, showing explicit information such as the enterprise's legal person, products, relationships with other enterprises, and technological dependencies;

[0164] Acquire the implicit features of the multi-dimensional graph in real time, calculate and set dynamic tags based on the implicit features, and supplement the implicit data of the enterprise portrait through dynamic tags;

[0165] The structural features of the multidimensional graph include node type, relationship type and node feature vector;

[0166] Dynamic labels include a unilateral fracture risk label;

[0167] The unilateral fracture risk label can be used to show the stability of the enterprise in the later cooperation in real time;

[0168] The specific steps for calculating the risk of unilateral fracture are as follows:

[0169] Step S401: Obtain the shortest paths of all node pairs and calculate the centrality of the nodes, specifically:

[0170] C i =Σ s≠i≠t [(σ st_i ) / (σ st )];

[0171] Among them, C i represents the node centrality of node i; s, i and t represent different node identifiers; σ st_i represents the total number of shortest paths from node s to node t passing through node i; σ st represents the total number of shortest paths from node s to node t;

[0172] Step S402: Find all paths from node i to node j and obtain all paths that do not pass through edge e. ij The path, and through the association weights between the nodes in different paths, the weight of each path is obtained, and the nodes that do not pass through the edge e ij The ratio of the sum of the weights of the paths to the sum of the weights of all paths is used as the edge redundancy estimate;

[0173] The specific characteristics are:

[0174] ;

[0175] Among them, ρ ij Represents edge e ij Redundancy estimation; p represents the path identifier; It means starting from node i and ending at node j without passing through edge e. ij Path; represents all paths starting from node i and ending at node j; represents the node identifier in path p; w uv represents the association weight between node u and node v;

[0176] Step S403: Calculate the risk of a single edge break based on the node centrality and edge redundancy.

[0177] The calculation of the unilateral fracture risk is specifically characterized as follows:

[0178] R ij =w ij (t)×(C i +C j )×(1-ρ ij );

[0179] Among them, R ij Represents edge e ij Unilateral fracture risk; ij (t) represents the real-time association weight between node i and node j; C j represents the node centrality of node j;

[0180] The enterprise portrait generated by the visual display is as follows:

[0181] Display the equity structure through a tree model;

[0182] Display the business scope and search keywords of the enterprise through word cloud;

[0183] Update dynamic tags in real time.

[0184] like Figure 2 As shown, an enterprise portrait automatic generation system based on multi-source data fusion includes a data acquisition module, a data processing module, an entity relationship modeling module, a graph incremental update module, a feature extraction and portrait generation module, and a visualization display module;

[0185] The data collection module is used to collect heterogeneous enterprise-related data and classify it, including structured data, semi-structured data, and unstructured data;

[0186] The data processing module is used to perform knowledge extraction and semantic alignment on the collected heterogeneous data, and extract data entity and relationship information;

[0187] The entity relationship modeling module is used to identify data entities in heterogeneous data as nodes, determine knowledge extraction results as association relationships, calculate the association weights of different association relationships, and construct a multidimensional graph based on nodes, association relationships, and association weights of different association relationships;

[0188] The graph incremental update module is used to regularly update nodes, relationships, and the association weights of different relationships. During the continuous update cycle, the association weights are updated in real time using an exponential decay model.

[0189] The feature extraction and portrait generation module is used to extract the structural features of the multi-dimensional map based on the regularly updated multi-dimensional map, generate the enterprise portrait, obtain the implicit features of the multi-dimensional map in real time, calculate and set dynamic tags based on the implicit features, and supplement the implicit data of the enterprise portrait through dynamic tags;

[0190] The visualization display module is used to display enterprise portraits and implicit data calculated based on the implicit features of the multi-dimensional map in different ways.

[0191] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A method for automatically generating enterprise portraits based on multi-source data fusion, characterized by: The method for automatically generating enterprise portraits specifically includes the following steps: Step S100: heterogeneous data collection, performing knowledge extraction and alignment on the collected heterogeneous data; Step S200: Determine data entities in the heterogeneous data as nodes, determine knowledge extraction results as association relationships, calculate association weights of different association relationships, and construct a multidimensional graph based on the nodes, association relationships, and association weights of different association relationships; Step S300: Set up an incremental update mechanism for the constructed multi-dimensional graph to regularly update nodes, relationships, and association weights of different relationships; specifically: Regularly update nodes, relationships, and the association weights of different relationships; In the continuous update cycle, the exponential decay model is used to update the association weight in real time. The specific steps are as follows: Step S301: Query the current association weight w t−1 and the last update time t last ; Step S302: Real-time acquisition of the update period time interval Δt=t current −t last ; where t current Indicates the current time; Step S303: Use an exponential decay model, superimpose the newly added weights, and apply the decay formula to update the weights in real time. Specifically, ； Among them, w t represents the updated association weight; w t−1 represents the current association weight; e represents the natural constant; λ represents the attenuation coefficient; Δt represents the time interval of the update cycle; Δw new Indicates the newly added weight; Step S400: Extracting structural features of the multidimensional graph based on the regularly updated multidimensional graph to generate a corporate profile, obtaining implicit features of the multidimensional graph in real time, calculating and setting dynamic tags based on the implicit features, and supplementing the implicit data of the corporate profile through dynamic tags; the implicit features of the multidimensional graph include paths between nodes in the multidimensional graph and association weights between nodes; The dynamic label includes a unilateral fracture risk label; The specific steps for calculating the unilateral fracture risk are as follows: Step S401: Obtain the shortest paths of all node pairs and calculate the centrality of the nodes, specifically: C i =S s≠i≠t [(s st_i ) / (s st )]; Among them, C i represents the node centrality of node i; s, i and t represent different node identifiers; σ st_i represents the total number of shortest paths from node s to node t passing through node i; σ st represents the total number of shortest paths from node s to node t; Step S402: Find all paths from node i to node j and obtain all paths that do not pass through edge e. ij The path, and through the association weights between the nodes in different paths, the weight of each path is obtained, and the nodes that do not pass through the edge e ij The ratio of the sum of the weights of the paths to the sum of the weights of all paths is used as the edge redundancy estimate; The specific characteristics are: ； Among them, ρ ij Represents edge e ij Redundancy estimation; p represents the path identifier; It means starting from node i and ending at node j without passing through edge e. ij Path; represents all paths starting from node i and ending at node j; represents the node identifier in path p; w uv represents the association weight between node u and node v; Step S403: Calculate the risk of a single edge break based on the centrality and edge redundancy of the node; specifically, the risk is: R ij =w ij (t)×(C i +C j )×(1-ρ ij ); Among them, R ij Represents edge e ij Unilateral fracture risk; ij (t) represents the real-time association weight between node i and node j; C j represents the node centrality of node j; Step S403: Calculate the risk of a single edge break based on the centrality and edge redundancy of the node; The unilateral breakage risk is calculated based on the centrality of nodes and edge redundancy estimation in the multidimensional graph; Step S500: Visually display the generated enterprise portrait.

2. The method for automatically generating enterprise portraits based on multi-source data fusion according to claim 1, characterized in that: In step S100, heterogeneous data collection is performed, specifically: Classify and collect heterogeneous data, including structured data, semi-structured data, and unstructured data; The structured data includes business registrations, tax records, patent databases, and supply chain system logs; The semi-structured data includes bidding documents and company annual reports; The unstructured data includes news, public opinion and industry research reports.

3. The method for automatically generating enterprise portraits based on multi-source data fusion according to claim 2, characterized in that: Perform knowledge extraction and alignment on the collected heterogeneous data, specifically: For unstructured data, a method based on the BiLSTM-CRF model is used to extract enterprise, product, and legal person data entities; Knowledge extraction based on semantic relations of dependency syntax; Unify time and location in heterogeneous data from different sources into a standardized format.

4. The method for automatically generating enterprise portraits based on multi-source data fusion according to claim 3 is characterized by: In step S200, data entities in heterogeneous data are determined as nodes, knowledge extraction results are determined as association relationships, association weights of different association relationships are calculated, and a multidimensional graph is constructed based on the nodes, association relationships, and association weights of different association relationships. Specifically, Step S201: Determine data entities as nodes, including enterprises, products, and legal persons; Step S202: Determine the knowledge extraction results as association relationships, where the association relationships include enterprise-production-product, enterprise-supply-enterprise, enterprise-holding-legal person, product-dependency-product, and geographic location-aggregation-enterprise; Step S203: Calculate the association weights of different association relationships by fitting the functional relationship with historical data; The correlation weight calculation of the transaction relationship and the fitted functional relationship are represented as follows: ; Among them, w ij represents the association weight from node i to node j, where i and j represent different node identifiers; a ij represents the annual transaction volume of node i to node j; a total represents the total transaction volume of the industry; z represents the annual transaction frequency; Step S204: Based on the nodes, association relationships, and association weights of different association relationships, the nodes are connected through the association relationships, and the association weights are marked to form a multi-dimensional graph.

5. The method for automatically generating enterprise portraits based on multi-source data fusion according to claim 4 is characterized by: In step S400, the structural features of the multi-dimensional atlas are extracted based on the regularly updated multi-dimensional atlas, specifically: The structural features of the multidimensional graph include node type, association relationship type and node feature vector; The node types include enterprise, product and legal person; The types of association relationships include supply chain relationships, investment relationships, and technology dependency relationships; The node feature vector includes: Enterprise node: registered capital, number of patents and revenue growth rate data; Product nodes: technical complexity and annual production data; Legal person node: data on shareholding ratio and number of affiliated companies.

6. The method for automatically generating enterprise portraits based on multi-source data fusion according to claim 5, characterized in that: The enterprise portrait generated by the visual display is as follows: Display the equity structure through a tree model; Display the business scope and search keywords of the enterprise through word cloud; Update dynamic tags in real time.

7. A system for automatically generating enterprise portraits based on multi-source data fusion, applied to the method for automatically generating enterprise portraits based on multi-source data fusion according to any one of claims 1 to 6, characterized in that: The enterprise portrait automatic generation system includes a data acquisition module, a data processing module, an entity relationship modeling module, a graph incremental update module, a feature extraction and portrait generation module and a visualization display module; The data collection module is used to collect heterogeneous enterprise-related data and collect them by categories, including structured data, semi-structured data and unstructured data; The data processing module is used to perform knowledge extraction and semantic alignment on the collected heterogeneous data, and extract data entity and relationship information; The entity relationship modeling module is used to determine data entities in heterogeneous data as nodes, determine knowledge extraction results as association relationships, calculate association weights of different association relationships, and construct a multidimensional graph based on the nodes, association relationships and association weights of different association relationships; The graph incremental update module is used to regularly update nodes, association relationships and association weights of different association relationships; In continuous update cycles, the association weights are updated in real time using an exponential decay model; The feature extraction and portrait generation module is used to extract the structural features of the multidimensional map based on the regularly updated multidimensional map, generate the enterprise portrait, obtain the implicit features of the multidimensional map in real time, calculate and set dynamic tags based on the implicit features, and supplement the implicit data of the enterprise portrait through dynamic tags; The visualization display module is used to display the enterprise portrait and the implicit data calculated based on the implicit features of the multi-dimensional map in different ways.

Images