Real-time dynamic enterprise portrait construction method and system based on high-dimensional financial data

Abstract

The invention relates to the technical field of enterprise portraits, in particular to a real-time dynamic enterprise portrait construction method and system based on high-dimensional financial data, and the method comprises the steps: (1) multi-source data integration and feature extraction; (2) constructing a knowledge graph ontology model; (3) dynamic event driving and real-time analysis; (4) updating the event triggering risk assessment model; and (5) generating and visualizing an enterprise portrait. According to the scheme, the defects of a traditional financial evaluation model can be overcome, more comprehensive, real-time and accurate risk evaluation and strategic guidance can be provided for investors, supervision organizations and enterprise managers, and financial data analysis is promoted to be converted from a single static normal form to a multi-dimensional dynamic normal form.

Description

Technical Field

[0001] The present invention relates to the field of enterprise portrait technology, and in particular to a method and system for constructing real-time dynamic enterprise portraits based on high-dimensional financial data. Background Art

[0002] With the development of financial technology, corporate profiling technology has gradually become a core tool for investment research, risk management, and decision support. Traditional corporate profiling methods mainly rely on financial data and construct static analysis models by evaluating indicators such as profitability and growth.

[0003] Existing technologies use preset mechanisms to quantitatively evaluate and predict the financial status and financial risks of listed companies and display the results in graphical form, initially achieving the intuitive presentation of financial information. However, such methods still have significant limitations: First, the data dimension is single, mainly relying on structured data such as financial statements, ignoring the value of multi-source heterogeneous data such as industrial and commercial information, industrial chain connections, market sentiment, and market fluctuations; Second, static analysis models have difficulty capturing the real-time impact of dynamic events (such as policy adjustments, emergencies, and changes in industry trends) on corporate risks; Third, the lack of deep integration of high-dimensional data and semantic correlation mining leads to insufficient comprehensiveness and prediction accuracy of corporate portraits.

[0004] In the current financial environment, corporate risk is driven not only by financial indicators but also closely linked to non-quantitative factors such as supply chain stability, market sentiment, and the industry's competitive landscape. For example, public opinion events can trigger sharp stock price fluctuations, while industrial chain disruptions can lead to a sudden increase in operational risk. However, traditional models, due to their single data source and lack of dynamic response mechanisms, are unable to promptly identify such potential risks. Furthermore, existing methods often utilize independent analysis modules and fail to construct cross-domain data networks through technologies such as knowledge graphs, limiting the ability to collaboratively analyze multi-dimensional features.

[0005] Therefore, there is an urgent need for an enterprise portrait construction solution that can integrate multi-source high-dimensional data and realize dynamic and static combined analysis to solve the above technical problems. Summary of the Invention

[0006] The purpose of the present invention is to overcome the problems of the above-mentioned prior art and provide a real-time dynamic enterprise portrait construction method and system based on high-dimensional financial data to solve the shortcomings of traditional financial evaluation models and the technical problem of being unable to provide investors, regulators and corporate managers with more comprehensive, real-time and accurate risk assessment and strategic guidance.

[0007] The above objectives are achieved through the following technical solutions: A method for constructing a real-time dynamic enterprise profile based on high-dimensional financial data, comprising: Step (1) Multi-source data integration and feature extraction: collect industrial and commercial data, industrial chain data, public opinion data, market data and financial data, perform standardized field mapping and storage on structured data, and extract keywords, emotional states and event tags from unstructured data through natural language processing (NLP); Step (2) Knowledge graph ontology model construction: Design a financial data graph model that includes enterprise entities, event entities, industry entities and their relationships, use RDF triples to describe semantic associations and import them into the knowledge graph; Step (3) Dynamic event-driven and real-time analysis: Real-time capture of policy announcements, market conditions, public opinion information, and supply chain dynamic event data. Analyze event types and related entities through named entity recognition, sentiment analysis, and event classification, match them with nodes in the knowledge graph, and calculate event impact weights. Step (4) Event-triggered risk assessment model update: When the event impact weight exceeds the threshold, the multi-factor time series model is called to update the data, combined with GNN to simulate the risk transmission path, and the quantitative indicator analysis and non-quantitative analysis are integrated to generate the risk score; Step (5) Enterprise portrait generation and visualization: Generate an enterprise portrait map based on the above steps, display risk warnings and strategic recommendations in a graphical interface, and mark specific risk sources.

[0008] As an optimization of this method, in step (1), the structured data includes the enterprise registration information, shareholder structure, and abnormal operation records of the industrial and commercial data, the upstream and downstream enterprises and supply chain dependencies of the industrial chain data, and the balance sheet, income statement, and cash flow statement of the financial data; the unstructured data includes news texts, social media sentiment analysis results, and major event reports of the public opinion data.

[0009] As an optimization of this method, in step (2), the enterprise entity relationships include upstream and downstream relationships, shareholder relationships and external investment relationships, and the associated ontology includes enterprise members, financial reports, public opinion events, risk warnings and securities market modules, and multivariate data is imported into the graph according to the graph ontology model through RDF triples.

[0010] As an optimization of this method, in step (3), the named entity recognition adopts the BERT model, the sentiment analysis adopts the TextCNN model, and the event classification adopts the LSTM-based deep learning architecture; the event impact weight is calculated based on the spread heat, industry relevance and public opinion sentiment intensity.

[0011] As an optimization of this method, in step (4), the quantitative indicator analysis constructs a multi-factor time series model based on financial health, market volatility and industrial chain stability; the non-quantitative analysis is based on the public opinion sentiment index and the industrial chain risk transmission path simulated by the graph neural network (GNN), and generates low, medium and high risk levels through weighted fusion.

[0012] As an optimization of this method, the graph neural network (GNN) predicts the probability of the impact of upstream raw material price increases on downstream companies by aggregating the characteristics of enterprise nodes with supply relationships and equity-related edge weights, and combines propagation path analysis to identify the spread and intensity of negative public opinion.

[0013] As an optimization of this method, the graphical interface described in step (5) adopts a multi-layer overlay technology, with the bottom layer displaying the static relationship network of the enterprise and the dynamic layer displaying the real-time risk transmission path in the form of a heat map.

[0014] A real-time dynamic enterprise portrait construction system based on high-dimensional financial data, including: Multi-source data collection and preprocessing module: used to collect and process industrial and commercial, industrial chain, public opinion, market and financial data, perform standardized field mapping on structured data, and extract features from unstructured data through natural language processing (NLP); Knowledge graph ontology model building module: used to define enterprise entities, events and relationships, build static relationship networks, and import data through RDF triples; Dynamic event monitoring and analysis module: used to capture event data in real time, analyze event types and related entities, and calculate event impact weights; Event-driven risk assessment module: used to trigger the update of risk assessment models and generate dynamic risk scores by combining quantitative indicator analysis with non-quantitative analysis; Visualization and decision support module: used to display corporate portraits and risk transmission paths in a graphical interface and provide strategic recommendations.

[0015] As an optimization of this system, the dynamic event monitoring and analysis module includes a BERT model, a TextCNN model, and an LSTM model, which are used for named entity recognition, sentiment analysis, and event classification, respectively.

[0016] As an optimization of this system, the visualization and decision support module dynamically displays the static attributes, dynamic association paths and risk levels of enterprises through a graphical interface, and marks specific risk sources such as supply chain disruptions, public opinion events or policy adjustments.

[0017] The present invention provides a method and system for constructing real-time dynamic enterprise portraits based on high-dimensional financial data. By integrating multi-dimensional data such as industrial and commercial registration, industrial chain relationships, public opinion, and market conditions, a financial data ontology model based on a knowledge graph is constructed, which can systematically characterize the static attributes and dynamic associations of enterprises. At the same time, a dynamic event-driven mechanism is introduced to analyze news, policies, market transactions and other event information in real time, trigger the update of risk prediction models, and combine quantitative indicators with unstructured data analysis to achieve intelligent upgrades of risk warning and decision support. This solution can not only make up for the shortcomings of traditional financial evaluation models, but also provide investors, regulators and corporate managers with more comprehensive, real-time and accurate risk assessment and strategic guidance, and promote the paradigm shift of financial data analysis from "single static" to "multi-dimensional dynamic". BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a module diagram of a real-time dynamic enterprise portrait construction system based on high-dimensional financial data according to the present invention; Figure 2 This is a flowchart of multi-source data integration and feature extraction in a method for constructing a real-time dynamic enterprise portrait based on high-dimensional financial data according to the present invention; Figure 3 This is a schematic diagram of the knowledge graph ontology model in the method for constructing a real-time dynamic enterprise portrait based on high-dimensional financial data described in the present invention; Figure 4 This is a flow chart of dynamic event driving and real-time analysis in a method for constructing a real-time dynamic enterprise portrait based on high-dimensional financial data as described in the present invention; Figure 5 This is a flow chart of the event-triggered risk assessment model in the method for constructing a real-time dynamic enterprise portrait based on high-dimensional financial data described in the present invention; Figure 6 This is a schematic diagram of the enterprise portrait map in the real-time dynamic enterprise portrait construction method based on high-dimensional financial data described in the present invention. DETAILED DESCRIPTION

[0019] The present invention will be further described in detail below with reference to the accompanying drawings and examples. The described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort shall fall within the scope of protection of the present invention.

[0020] like Figure 1 As shown in the figure, this solution provides a real-time dynamic enterprise portrait construction system based on high-dimensional financial data, including a multi-source data acquisition and preprocessing module, a knowledge graph ontology model construction module, a dynamic event monitoring and analysis module, an event-driven risk assessment module, and a visualization and decision support module, among which: The multi-source data collection and preprocessing module is used to collect and process industrial and commercial, industrial chain, public opinion, market and financial data, perform standardized field mapping on structured data, and extract features from unstructured data through natural language processing (NLP). In other words, it is mainly responsible for collecting structured and unstructured data such as industrial and commercial, industrial chain, public opinion, market and financial data, and performing cleaning, normalization and feature extraction. The knowledge graph ontology model construction module is used to define enterprise entities, events and relationships, build a static relationship network, and import data through RDF triples. In other words, it is mainly responsible for designing a high-dimensional financial data graph model based on ontology, defining enterprise entities, events, relationships and attributes, and building a static relationship network. The dynamic event monitoring and analysis module is used to capture event data in real time, analyze event types and related entities, and calculate event impact weights; that is, it is mainly responsible for capturing news, policies, market transactions and other event data in real time, analyzing event types, impact scope and related entities; The event-driven risk assessment module is used to trigger the update of the risk assessment model and generate a dynamic risk score by combining quantitative indicator analysis and non-quantitative analysis. In other words, it is mainly responsible for combining quantitative indicators and non-quantitative analysis to trigger the update of the risk prediction model through events and generate a dynamic enterprise profile. The visualization and decision support module is used to display the enterprise portrait and risk transmission path in a graphical interface and provide strategic recommendations; that is, it is mainly responsible for displaying the enterprise portrait in a graphical interface and providing risk warnings and strategic recommendations.

[0021] In addition, this embodiment also provides a method for constructing a real-time dynamic enterprise portrait based on high-dimensional financial data: Step (1) Multi-source data integration and feature extraction: collect industrial and commercial data, industrial chain data, public opinion data, market data and financial data, perform standardized field mapping and storage on structured data, and extract keywords, emotional states and event tags from unstructured data through natural language processing (NLP); Step (2) Knowledge graph ontology model construction: Design a financial data graph model that includes enterprise entities, event entities, industry entities and their relationships, use RDF triples to describe semantic associations and import them into the knowledge graph; Step (3) Dynamic event-driven and real-time analysis: Real-time capture of policy announcements, market conditions, public opinion information, and supply chain dynamic event data. Analyze event types and related entities through named entity recognition, sentiment analysis, and event classification, match them with nodes in the knowledge graph, and calculate event impact weights. Step (4) Event-triggered risk assessment model update: When the event impact weight exceeds the threshold, the multi-factor time series model is called to update the data, combined with GNN to simulate the risk transmission path, and the quantitative indicator analysis and non-quantitative analysis are integrated to generate the risk score; Step (5) Enterprise portrait generation and visualization: Generate an enterprise portrait map based on the above steps, display risk warnings and strategic recommendations in a graphical interface, and mark specific risk sources.

[0022] like Figure 2 As shown, the multi-source data integration and feature extraction in step (1) of this embodiment mainly includes extracting and classifying different data sources, such as industrial and commercial data, industrial chain data, public opinion data, market data, and financial data, and storing them in a local temporary space. Among them: Industrial and commercial data mainly includes but is not limited to enterprise registration information, shareholder structure, abnormal operation records and other enterprise industrial and commercial data; Industry chain data mainly includes but is not limited to upstream and downstream enterprises, supply chain dependencies, industry classifications and other industry data; Public opinion data mainly includes but is not limited to news, social media sentiment analysis, major event reports, etc. Market data mainly includes but is not limited to stock price fluctuations, trading volume, market indices and other asset trading data; Financial data mainly includes but is not limited to balance sheet, income statement, cash flow statement and other financial report data.

[0023] After obtaining these data, they will be preliminarily processed.

[0024] Structured data such as industrial and commercial data, financial data, and industrial chain data will be stored in local temporary space using standardized field mapping.

[0025] For unstructured data such as public opinion data represented by text, keywords, emotional states and event labels will be extracted through natural language processing (NLP) and stored in a local temporary space for subsequent entry into the map.

[0026] like Figure 3 As shown, the construction of the knowledge graph ontology model in step (2) of this embodiment mainly includes building a dedicated graph ontology model for the usage scenario of the system of the present invention, and the graph ontology model is mainly developed around the enterprise.

[0027] There are upstream and downstream relationships, shareholder relationships, and overseas investment relationships between enterprises.

[0028] Ontologies related to enterprises include main members (individuals), financial reports, public opinion, enterprise forecasts, risk warnings, etc.

[0029] For listed companies, the relationship map also involves a securities market module, which is described by the securities ontology.

[0030] In addition to the company's own data, the ontology model also involves information about the industry to which the company belongs, as well as policy research reports related to the industry.

[0031] After completing the construction of the graph ontology model, the multivariate data extracted in step 1) can be imported into the graph according to the graph ontology model using RDF triples (subject-predicate-object) to describe semantic associations.

[0032] like Figure 4 As shown in Figure 3, the dynamic event-driven and real-time analysis in step (3) of this embodiment is the core link in building a real-time dynamic profile of the enterprise. It aims to achieve immediate response and accurate prediction of enterprise risks by capturing and analyzing multi-source heterogeneous event data in real time, combined with the dynamic update mechanism of the knowledge graph. This process relies on a variety of data sources, including policy announcements, market conditions, public opinion information, supply chain dynamics, and corporate behavior.

[0033] The event analysis part uses the BERT model to perform named entity recognition on key entities in texts such as news (such as company names and industry terms) to obtain descriptive objects. Sentiment analysis models (such as TextCNN) are used to determine the positive or negative tendencies of public opinion. Event classification models (such as LSTM-based deep learning architectures) classify events into preset categories, such as "policy adjustments" or "supply chain adjustments."

[0034] After event parsing is complete, the system accurately associates the event with entities in the knowledge graph created in step 2). For example, a news item about a "raw material supplier being shut down" is linked to the corresponding supplier node in the knowledge graph through entity matching technology, further tracing its downstream manufacturers and industry chain dependencies. The system then calculates the event's impact weight, taking into account multiple metrics such as the event's popularity, industry relevance, and the intensity of public sentiment.

[0035] When the impact weight of an event exceeds a preset threshold, the system automatically invokes a risk assessment model, combining quantitative indicators (such as corporate financial health and market volatility) with non-quantitative data (such as public sentiment and supply chain dependencies) to generate a dynamic risk score. During this process, graph neural networks (GNNs) are used to simulate the transmission path of risk within the supply chain, predicting the potential impact of a disruption in a particular link on upstream and downstream companies.

[0036] like Figure 5 As shown, the event in step (4) of this embodiment triggers the update of the risk assessment model. This model is the core module for realizing dynamic risk warning in the construction of enterprise portraits. Its core lies in adjusting the enterprise risk score and generating visual results through the dual-track drive of dynamic quantitative and non-quantitative analysis.

[0037] When the system detects a high-weight event identified in step (3) (e.g., an impact coefficient exceeding a threshold of 0.7), it immediately triggers a full recalculation of the risk assessment model. This process includes: 1) Update multi-factor models with real-time data (e.g., latest financial reports, daily stock price volatility); 2) Retrain the multi-factor time series model to incorporate the latest event sequence; 3) Adjust non-quantitative risk weights based on the latest public opinion sentiment index and GNN transmission prediction results.

[0038] Ultimately, the risk level is divided into low risk, medium risk and high risk through weighted fusion of quantitative and non-quantitative scores, and the specific risk sources are marked.

[0039] At the quantitative analysis level, the system first constructs a multi-factor model, integrating multi-dimensional data such as financial indicators, market conditions, and industrial chain stability.

[0040] At the non-quantitative analysis level, the system generates a dynamic sentiment index based on public opinion sentiment analysis, uses natural language processing technology to classify the sentiment polarity of news and social media texts (positive / neutral / negative), and combines communication path analysis (such as forwarding levels and the influence of key opinion leaders) to identify the spread and intensity of negative public opinion.

[0041] Specifically, GNNs aggregate node (e.g., enterprise) features and edge (e.g., supply relationships, equity connections) weights to predict the impact of a risk event at a specific link (e.g., the probability of an upstream raw material price increase impacting a downstream enterprise). like Figure 6 As shown, the enterprise portrait generation and visualization in step (5) of this embodiment is based on the final results obtained from the above steps (1) to (4), and a corporate portrait map based on high-dimensional financial data is constructed.

[0042] The above description is only for explaining the embodiments of the present invention and is not intended to limit the present invention. For those skilled in the art, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A real-time dynamic enterprise portrait construction method based on high-dimensional financial data, characterized by: include: Step (1) Multi-source data integration and feature extraction: collect industrial and commercial data, industrial chain data, public opinion data, market data and financial data, perform standardized field mapping and storage on structured data, and extract keywords, emotional states and event tags from unstructured data through natural language processing; Step (2) Knowledge graph ontology model construction: Design a financial data graph model that includes enterprise entities, event entities, industry entities and their relationships, use RDF triples to describe semantic associations and import them into the knowledge graph; Step (3) Dynamic event-driven and real-time analysis: Real-time capture of policy announcements, market conditions, public opinion information, and supply chain dynamic event data. Analyze event types and related entities through named entity recognition, sentiment analysis, and event classification, match them with nodes in the knowledge graph, and calculate event impact weights. Step (4) Event-triggered risk assessment model update: When the event impact weight exceeds the threshold, the multi-factor time series model is called to update the data, combined with GNN to simulate the risk transmission path, and the quantitative indicator analysis and non-quantitative analysis are integrated to generate the risk score; Step (5) Enterprise portrait generation and visualization: Generate an enterprise portrait map based on the above steps, display risk warnings and strategic recommendations in a graphical interface, and mark specific risk sources.

2. A method for constructing a real-time dynamic enterprise portrait based on high-dimensional financial data according to claim 1, characterized in that: In step (1), the structured data includes the business registration information, shareholder structure, and abnormal operation records of the industrial and commercial data, the upstream and downstream enterprises and supply chain dependencies of the industrial chain data, and the balance sheet, income statement, and cash flow statement of the financial data; the unstructured data includes news texts, social media sentiment analysis results, and major event reports of the public opinion data.

3. The method for constructing a real-time dynamic enterprise portrait based on high-dimensional financial data according to claim 1, characterized in that: In step (2), the enterprise entity relationships include upstream and downstream relationships, shareholder relationships and external investment relationships, and the associated ontology includes enterprise members, financial reports, public opinion events, risk warnings and securities market modules, and multivariate data is imported into the graph according to the graph ontology model through RDF triples.

4. A method for constructing a real-time dynamic enterprise portrait based on high-dimensional financial data according to claim 1, characterized in that: In step (3), the named entity recognition adopts the BERT model, the sentiment analysis adopts the TextCNN model, and the event classification adopts the LSTM-based deep learning architecture; the event impact weight is calculated based on the spread heat, industry relevance and public opinion sentiment intensity.

5. The method for constructing a real-time dynamic enterprise portrait based on high-dimensional financial data according to claim 1, characterized in that: In step (4), the quantitative indicator analysis constructs a multi-factor time series model based on financial health, market volatility and industrial chain stability; the non-quantitative analysis is based on the public opinion sentiment index and the industrial chain risk transmission path simulated by the graph neural network, and generates low, medium and high risk levels through weighted fusion.

6. A method for constructing a real-time dynamic corporate portrait based on high-dimensional financial data according to claim 5, characterized in that: The graph neural network (GNN) predicts the probability of the impact of upstream raw material price increases on downstream companies by aggregating the characteristics of enterprise nodes with supply relationships and equity-related edge weights, and combines transmission path analysis to identify the spread and intensity of negative public opinion.

7. The method for constructing a real-time dynamic enterprise portrait based on high-dimensional financial data according to claim 1, characterized in that: The graphical interface described in step (5) adopts a multi-layer overlay technology, with the bottom layer displaying the static relationship network of the enterprise and the dynamic layer displaying the real-time risk transmission path in the form of a heat map.

8. A real-time dynamic enterprise portrait construction system based on high-dimensional financial data, characterized by: include: Multi-source data collection and preprocessing module: used to collect and process industrial and commercial, industrial chain, public opinion, market and financial data, perform standardized field mapping on structured data, and extract features from unstructured data through natural language processing (NLP); Knowledge graph ontology model building module: used to define enterprise entities, events and relationships, build static relationship networks, and import data through RDF triples; Dynamic event monitoring and analysis module: used to capture event data in real time, analyze event types and related entities, and calculate event impact weights; Event-driven risk assessment module: used to trigger the update of risk assessment models and generate dynamic risk scores by combining quantitative indicator analysis with non-quantitative analysis; Visualization and decision support module: used to display corporate portraits and risk transmission paths in a graphical interface and provide strategic recommendations.

9. A real-time dynamic enterprise portrait construction system based on high-dimensional financial data according to claim 8, characterized in that: The dynamic event monitoring and analysis module includes a BERT model, a TextCNN model, and an LSTM model, which are used for named entity recognition, sentiment analysis, and event classification, respectively.

10. A real-time dynamic enterprise portrait construction system based on high-dimensional financial data according to claim 8, characterized in that: The visualization and decision support module dynamically displays the static attributes, dynamic association paths and risk levels of the enterprise through a graphical interface, and marks specific risk sources such as supply chain disruptions, public opinion events or policy adjustments.

Images