Financial and economic data management method based on knowledge graph

Abstract

The invention discloses a knowledge graph-based financial data management method, and relates to the technical field of data management. The method comprises the following steps: carrying out original financial data acquisition from different data sources, and carrying out preprocessing and quality evaluation to obtain a plurality of standard financial data; performing knowledge extraction and modeling on a plurality of standard financial data by using a joint extraction model, and introducing a swarm intelligence optimization algorithm to perform knowledge fusion and treatment to obtain a financial knowledge graph; based on the financial knowledge graph, executing a preset automatic governance strategy, performing data blood relationship tracking and influence analysis, and generating an influence analysis report; and based on the financial knowledge graph, performing intelligent analysis according to the problem data to obtain an intelligent analysis report, and automatically generating a financial data asset map. The problems that in the prior art, data islands and semantic gaps exist, data quality is uneven, knowledge fusion and conflict resolution are difficult, the analysis dimension is single, and the data governance and blood relationship tracing ability is weak are solved.

Description

Technical Field

[0001] This invention relates to the field of data management technology, and in particular to a financial data management method based on knowledge graphs. Background Technology

[0002] With the rapid development of financial markets and the widespread application of information technology, financial data is experiencing explosive growth. These data come from a wide range of sources, including stock exchanges, listed company financial reports, news media, social media, and industry research reports. Their formats are also diverse, encompassing structured data (such as financial statements), semi-structured data (such as announcements in XML and JSON formats), and unstructured data (such as news texts and analyst comments).

[0003] In existing technologies, the management of massive amounts of heterogeneous financial data typically relies on traditional data warehouse or data lake technologies. However, these traditional methods have the following significant drawbacks: 1) Data silos and semantic gaps: Different data sources have different data models and definitions, making it difficult to effectively correlate and integrate them. For example, different data sources may have different definitions and statistical methods for "operating revenue," and traditional methods are unable to unify their semantics, making it difficult to deeply explore the value of the data.

[0004] 2) Inconsistent data quality: Due to the numerous data sources, data quality is difficult to guarantee, resulting in issues such as missing data, errors, inconsistencies, and poor timeliness. Existing technologies typically rely on manual data cleaning and quality assessment, which is inefficient and costly.

[0005] 3) Difficulties in knowledge integration and conflict resolution: When integrating information from different data sources, knowledge conflicts often arise. For example, different data sources may provide different values ​​for the same financial indicator for the same company. Existing technologies lack automated and intelligent conflict resolution mechanisms, leading to a decrease in the accuracy and reliability of the final data.

[0006] 4) Limited Analytical Dimensions: Traditional data analysis methods are mostly based on statistics and reports, making it difficult to handle complex correlation analysis and causal reasoning. Users cannot perform interactive, in-depth intelligent analysis using natural language, making it difficult to discover the complex relationships and potential risks hidden behind the data.

[0007] 5) Weak data governance and lineage tracing capabilities: When data quality issues are discovered, it is difficult to quickly locate the source of the problem and assess its impact on downstream applications. The lack of transparency in data lineage poses a significant challenge to data governance and risk management. Summary of the Invention

[0008] This invention provides a knowledge graph-based financial data management method that solves the problems of data silos and semantic gaps, inconsistent data quality, difficulties in knowledge integration and conflict resolution, single analysis dimensions, and weak data governance and lineage tracing capabilities in existing technologies.

[0009] In a first aspect, embodiments of the present invention provide a financial data management method based on knowledge graphs, the method comprising: Raw financial data is collected from different data sources, and the collected heterogeneous raw financial data is preprocessed and its quality is assessed to obtain the quality labels of the data sources and several homogeneous standard financial data from different data sources. A pre-built joint extraction model is used to extract and model knowledge from several standard financial data, and a swarm intelligence optimization algorithm is introduced for knowledge fusion and governance to obtain a financial knowledge graph. Based on financial knowledge graphs, a pre-set automated governance strategy is executed, and data lineage tracing and impact analysis are performed on data sources with low-quality labels, generating an impact analysis report. Based on a financial knowledge graph, the system performs intelligent analysis on user-input questions to generate intelligent analysis reports and automatically creates a financial data asset map.

[0010] The technical solution provided in this application has at least the following beneficial effects: This paper proposes a four-stage closed-loop management framework covering the entire data lifecycle, encompassing data acquisition, storage, processing, and analysis. This framework unifies fragmented management processes under the core asset of knowledge graphs, achieving a complete value chain loop from data to knowledge to value. Through knowledge graph technology, data from different sources and models are unified under a single semantic framework, completely breaking down data silos and enabling deep integration and mining of previously isolated data, significantly unlocking its potential value. Automated quality probing during the data acquisition stage, which labels data sources with quality tags and dynamically updates these tags during subsequent knowledge fusion and application, replaces the traditional, inefficient manual methods. This not only significantly improves the efficiency and accuracy of data quality assessment but also enables continuous monitoring and dynamic optimization of data source quality, providing a reliable data foundation for upper-layer applications. Addressing the conflict challenges in knowledge fusion, a particle swarm optimization algorithm combining chaotic initialization, Gaussian mutation, and reverse learning is introduced, capable of comprehensively evaluating the authority, timeliness, and other aspects of data sources. By intelligently seeking optimal conflict resolution strategies across multiple dimensions, including knowledge consistency, the accuracy and credibility of the financial knowledge graph are ensured, providing high-quality knowledge support for decision-making. Based on the financial knowledge graph, the ant colony algorithm can handle complex association analysis and causal reasoning tasks. Users can ask questions in natural language, which can not only answer simple queries but also automatically decompose complex tasks, search for high-confidence logical paths in the graph, and reveal hidden multi-hop relationships and potential risks behind the data. This breaks through the limitation of the single analytical dimension of traditional statistical reports and realizes interactive and in-depth intelligent analysis. By automatically constructing a full-link lineage graph from source to application for each data item (entity, relationship, and attribute) in the knowledge graph, the data lineage is made completely transparent. When data quality problems are found, the source of the problem can be quickly and accurately located based on the lineage graph, and its impact on downstream applications can be automatically analyzed to generate an impact analysis report. This transforms data governance from passive "post-event repair" to proactive "pre-event warning" and "in-event control," greatly improving risk management capabilities.

[0011] In one optional implementation, raw financial data is collected from different data sources, and the collected heterogeneous raw financial data is preprocessed and its quality assessed to obtain quality labels for the data sources and several homogeneous standard financial data from different data sources, including: For different data sources, deploy configurable data connectors to uniformly access and collect heterogeneous raw financial data, including structured, semi-structured, and unstructured data. When collecting heterogeneous raw financial data, metadata corresponding to the raw financial data and its data source is collected simultaneously. The collected heterogeneous raw financial data is preprocessed to obtain several homogeneous standard financial data from different data sources. The system performs automated quality checks on several homogeneous standard financial data sets, and based on the results, automatically generates quality tags for each data source and its corresponding standard financial data sets, and associates the quality tags with the corresponding metadata.

[0012] In one alternative implementation, a pre-built joint extraction model is used to extract and model knowledge from several standard financial data sets, and a swarm intelligence optimization algorithm is introduced for knowledge fusion and governance to obtain a financial knowledge graph, including: Using a large-scale financial corpus, a general pre-trained language model is continuously pre-trained to obtain a Fin-BERT model that deeply understands financial contexts. Based on the Fin-BERT model, a joint extraction model is constructed. Using a joint extraction model, knowledge is extracted and modeled from several standard financial data to obtain several knowledge triples including entities, relations and attributes. Several knowledge triples are fused, and a particle swarm optimization algorithm is introduced to resolve the knowledge conflicts of conflicting knowledge triples, resulting in several fused knowledge triples. Based on several fused knowledge triples, a financial knowledge graph is constructed, stored in a graph database, and version management is implemented for updates to the financial knowledge graph.

[0013] In one optional implementation, several knowledge triples are fused, and a particle swarm optimization algorithm is introduced to resolve knowledge conflicts in the conflicting knowledge triples, resulting in several fused knowledge triples, including: Entity linking is performed on several knowledge triples, linking each entity to a unique entity ID in the knowledge base, and monitoring for conflicts during the entity linking process; If knowledge triples from different data sources conflict during the knowledge fusion process, a particle swarm optimization algorithm is introduced to find the corresponding optimal resolution strategy. According to the optimal resolution strategy, the knowledge triples that generate conflicts are resolved, and the other knowledge triples are further fused until all knowledge triples are fused, resulting in several fused knowledge triples.

[0014] In one alternative implementation, if knowledge triples from different data sources conflict during knowledge fusion, a particle swarm optimization algorithm is introduced to find the optimal resolution strategy, including: If knowledge triples from different data sources conflict during the knowledge fusion process, then the conflict description of the conflicting knowledge triples is extracted, and the conflict resolution problem of the conflict description is formalized into an optimization problem. Based on the conflict description and all possible resolution schemes, the solution space of the optimization problem is defined, the resolution schemes are encoded as the positions of particles in the particle swarm optimization algorithm, and the fitness function is designed. The chaotic sequence is generated using the Logistic mapping. The chaotic sequence is then mapped to the solution space to obtain the initial position of each particle in the initial particle swarm. Based on the fitness function, the velocity and individual optimal position of each particle, as well as the global optimal position of the initial particle swarm, are initialized. The position of each particle in the initial particle swarm is updated to obtain the updated position of each particle in the updated particle swarm. Then, Gaussian mutation is performed on some particles in the updated particle swarm with mutation probability to obtain the mutated position of the particles. If the global best position does not improve for several consecutive generations, reverse learning is triggered, and the dynamic reverse position of the particle corresponding to the current global best position is obtained. Based on the fitness function, obtain the fitness value of each particle's updated position, mutated position, and / or dynamic reverse position, and update the individual optimal position of each particle and the global optimal position of the particle swarm based on the fitness value. Repeat the particle swarm update steps until the number of iterations reaches the iteration threshold or the fitness value of the global optimum position meets the requirements, then stop the optimization and output the global optimum position; Decoding the globally optimal position yields the optimal resolution strategy for the knowledge triples that generate conflicts.

[0015] In one alternative implementation, the fitness function is formulated as follows:

[0016] In the formula, The position of the particle X The fitness value of the corresponding digestion scheme; Assess the authority of data sources related to quality labels; Rate the timeliness of information; Scoring based on knowledge consistency; The fitness weights are summed to 1.

[0017] In one optional implementation, based on a financial knowledge graph, a pre-defined automated governance strategy is executed, and data lineage tracing and impact analysis are performed on data sources with low-quality labels, generating an impact analysis report, including: According to the automated governance strategy, the relationships between all entities in the financial knowledge graph are traversed. If the relationship between different entities violates the preset business rules, the corresponding relationship is marked as abnormal. Based on the automated governance strategy, the quality label of each data source is dynamically calculated and updated according to the final adoption rate and correction frequency of knowledge triples from different data sources in the financial knowledge graph. Automatically construct a full-link lineage graph from the data source to the business applications in the financial knowledge graph for each knowledge triple in the financial knowledge graph; Based on the end-to-end lineage diagram, the downstream knowledge triples and business applications corresponding to the data sources with low-quality tags are retrieved. Then, based on all downstream knowledge triples and business applications, an impact analysis report on financial data management is generated using a preset template.

[0018] In one optional implementation, based on a financial knowledge graph, intelligent analysis is performed on the user-input question data to generate an intelligent analysis report and automatically generate a financial data asset map, including: Receive user-input question data based on the preset natural language query interface; If the question data is a simple query, the question data will be parsed into a graph query language, queried in the financial knowledge graph, and a corresponding structured answer will be generated. Based on the structured answer, a corresponding intelligent analysis report will be generated. If the problem data contains complex analysis tasks, then the complex analysis tasks are decomposed, and the task entities are located in the financial knowledge graph to obtain several task nodes, and then proceed to the next step. Based on several task nodes and related intermediate nodes, extract financial knowledge subgraphs related to complex analysis tasks from the financial knowledge graph; Using a pre-built node feature extraction model, extract the node feature vector of each node in the financial knowledge subgraph; Based on the node feature vectors and the relational edge weights of the financial knowledge subgraph, the ant colony algorithm is used to perform path search in the financial knowledge subgraph to obtain several high-confidence paths. Risk analysis is performed on high-confidence paths to obtain risk scores, and combined with structured answers to generate corresponding intelligent analysis reports; Based on the metadata of the data source, automatically generate a financial data asset map.

[0019] In one alternative implementation, based on the node feature vectors and the relational edge weights of the financial knowledge subgraph, an ant colony algorithm is used to perform path search in the financial knowledge subgraph, obtaining several high-confidence paths, including: Place several artificial ants at the starting node of the financial knowledge subgraph and initiate path search; Based on the node feature vector, the relation edge weights of the financial knowledge subgraph, and the pheromone, the transfer probability of the artificial ant at the current node choosing to move to the next hop neighbor node is obtained, and node transfer is performed based on the transfer probability. Repeat the node transfer steps until the current artificial ant reaches the target node of the financial knowledge subgraph, thus obtaining the search path of the artificial ant; Repeat the path search steps until all artificial ants have completed the path search. Based on the quality of the path search, update the corresponding pheromones and return to the starting node to perform iterative updates of the ant colony. Repeat the iterative update steps until the number of iterations reaches the iteration threshold, and obtain several search paths with pheromone concentrations higher than the concentration threshold, i.e., several high-confidence paths.

[0020] Secondly, embodiments of the present invention provide a knowledge graph-based financial data management device for implementing a knowledge graph-based financial data management method. The device includes: The financial data acquisition unit is used to collect raw financial data from different data sources, and to preprocess and assess the quality of the collected heterogeneous raw financial data to obtain the quality labels of the data sources and several homogeneous standard financial data from different data sources. The knowledge graph construction unit is used to extract and model knowledge from several standard financial data using a pre-built joint extraction model, and introduces a swarm intelligence optimization algorithm for knowledge fusion and governance to obtain a financial knowledge graph. The influence analysis unit is used to input the spatial distribution map of coating adhesion as the initial condition into a high-fidelity digital twin pre-constructed based on digital twin technology for coupled simulation and lifetime analysis, and obtain the remaining lifetime distribution map of the coating. The intelligent analysis unit is used to perform intelligent analysis based on financial knowledge graphs and user-input question data, generate intelligent analysis reports, and automatically generate financial data asset maps.

[0021] A third aspect of this invention provides an electronic device, which includes: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by at least one processor, such that the at least one processor can perform the method proposed in the first aspect of the present invention.

[0022] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect of the present invention. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention; Figure 2 This is a flowchart illustrating the steps of a knowledge graph-based financial data management method provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the functional units of a knowledge graph-based financial data management device provided in an embodiment of the present invention. Detailed Implementation

[0024] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0025] The present invention will be further described below with reference to the accompanying drawings.

[0026] Reference Figure 1 , Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention.

[0027] like Figure 1 As shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0028] Those skilled in the art will understand that Figure 1The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0029] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and an electronic program for a knowledge graph-based financial data management device.

[0030] exist Figure 1 In the illustrated electronic device, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be installed in the electronic device. The electronic device calls the electronic program of the knowledge graph-based financial data management device stored in the memory 1005 through the processor 1001, and executes the knowledge graph-based financial data management method provided in the embodiment of the present invention.

[0031] Reference Figure 2 The present invention provides a knowledge graph-based financial data management method, the method comprising: S201: Collect raw financial data from different data sources, preprocess and assess the quality of the collected heterogeneous raw financial data, and obtain the quality labels of the data sources and several homogeneous standard financial data from different data sources. S202: Use a pre-built joint extraction model to extract and model knowledge from several standard financial data, and introduce a swarm intelligence optimization algorithm for knowledge fusion and governance to obtain a financial knowledge graph; S203: Based on the financial knowledge graph, execute the preset automated governance strategy, and perform data lineage tracing and impact analysis on the data sources of low-quality tags, generating an impact analysis report; S204: Based on a financial knowledge graph, it performs intelligent analysis based on user-input question data, generates an intelligent analysis report, and automatically generates a financial data asset map.

[0032] The technical solution provided in this application has at least the following beneficial effects: This paper proposes a four-stage closed-loop management framework covering the entire data lifecycle, encompassing data acquisition, storage, processing, and analysis. This framework unifies fragmented management processes under the core asset of knowledge graphs, achieving a complete value chain loop from data to knowledge to value. Through knowledge graph technology, data from different sources and models are unified under a single semantic framework, completely breaking down data silos and enabling deep integration and mining of previously isolated data, significantly unlocking its potential value. Automated quality probing during the data acquisition stage, which labels data sources with quality tags and dynamically updates these tags during subsequent knowledge fusion and application, replaces the traditional, inefficient manual methods. This not only significantly improves the efficiency and accuracy of data quality assessment but also enables continuous monitoring and dynamic optimization of data source quality, providing a reliable data foundation for upper-layer applications. Addressing the conflict challenges in knowledge fusion, a particle swarm optimization algorithm combining chaotic initialization, Gaussian mutation, and reverse learning is introduced, capable of comprehensively evaluating the authority, timeliness, and other aspects of data sources. By intelligently seeking optimal conflict resolution strategies across multiple dimensions, including knowledge consistency, the accuracy and credibility of the financial knowledge graph are ensured, providing high-quality knowledge support for decision-making. Based on the financial knowledge graph, the ant colony algorithm can handle complex association analysis and causal reasoning tasks. Users can ask questions in natural language, which can not only answer simple queries but also automatically decompose complex tasks, search for high-confidence logical paths in the graph, and reveal hidden multi-hop relationships and potential risks behind the data. This breaks through the limitation of the single analytical dimension of traditional statistical reports and realizes interactive and in-depth intelligent analysis. By automatically constructing a full-link lineage graph from source to application for each data item (entity, relationship, and attribute) in the knowledge graph, the data lineage is made completely transparent. When data quality problems are found, the source of the problem can be quickly and accurately located based on the lineage graph, and its impact on downstream applications can be automatically analyzed to generate an impact analysis report. This transforms data governance from passive "post-event repair" to proactive "pre-event warning" and "in-event control," greatly improving risk management capabilities.

[0033] In one optional implementation, raw financial data is collected from different data sources, and the collected heterogeneous raw financial data is preprocessed and its quality assessed to obtain quality labels for the data sources and several homogeneous standard financial data from different data sources, including: S2011: Deploy configurable data connectors for different data sources to uniformly access and collect heterogeneous raw financial data, including structured, semi-structured, and unstructured data. In practice, a modular data access layer needs to be built, and corresponding data connectors need to be deployed for different types of data sources. For structured data sources (such as relational databases like Oracle, SQL Server, and MySQL), configure a JDBC / ODBC-based connector and write configurable SQL query scripts to retrieve data periodically or on demand. For semi-structured data sources (such as financial data service providers offering RESTful APIs, and publicly available government data in XML / JSON format), configure the API connector to automatically retrieve data by setting request parameters, authentication methods, and parsing rules; For unstructured data sources (such as financial news websites, listed company announcement PDFs, and social media), configure a web crawler connector. This connector allows users to define crawling rules (such as URL lists, page parsing rules, XPath / CSS selectors, and pagination logic) through a visual interface or configuration file, and includes anti-crawler strategies (such as dynamic User-Agent and proxy IP pool rotation). All connectors support plug-in deployment and can be used for task orchestration, monitoring and alerting through a unified scheduling system (such as Apache Airflow) to achieve automation and scalability of data collection. S2012: When collecting heterogeneous raw financial data, simultaneously collect the metadata of the raw financial data and its data source. It is worth noting that metadata is the cornerstone of data governance. During the data collection process, the following metadata is automatically captured and stored: Technical metadata: data source type, connection information, data table / file structure, field name, field type, field length, collection timestamp, data file size, number of records; Business metadata: business name of the data source, data owner, data update frequency, data security level, business terminology table (such as the corresponding field of "operating revenue" in different data sources); Operation metadata: Collect task execution logs, success / failure status, and error messages; These metadata entries, along with the original data itself, are stored in the metadata database, forming the data's "identity card" and providing basic information for subsequent data quality assessment, lineage tracing, and asset mapping. S2013: Preprocess the collected heterogeneous raw financial data to obtain several homogeneous standard financial data from different data sources. In this embodiment, preprocessing is a multi-stage data cleaning and transformation process aimed at transforming heterogeneous data into high-quality data that conforms to a unified standard. Data cleaning: Missing value handling: Different strategies are adopted based on the importance and missing rate of the field. Records with missing key fields (such as company ID) are deleted directly; for numerical fields (such as net profit), the mean, median or industry average can be used to fill them; for text fields (such as company profile), they can be marked as "unknown". Outlier handling: Identify numerical outliers using statistical methods (such as Z-score) and combine them with business rules (such as the price-to-earnings ratio cannot be negative) to make judgments, and correct, delete or mark the confirmed outliers; Duplicate value handling: Deduplication is performed based on key field combinations (such as company ID + reporting period), retaining the latest or most complete data version; Data transformation: Standardize the format: unify all date formats to the ISO 8601 standard (YYYY-MM-DD); convert all currency values ​​to a unified functional currency (such as RMB) and record the original currency and exchange rate; standardize percentage data to decimal form; Data standardization: Convert text data for uppercase and lowercase, full-width / half-width, and traditional / simplified characters; Data integration: Based on entity alignment technology, data from different data sources that point to the same entity are associated; for example, by using various identifiers such as the company's full name, abbreviation, and stock code, combined with fuzzy matching algorithms, "XXX Co., Ltd." from data source A and "6***19.SH" from data source B are associated with the same unified entity ID; S2014: Perform automated quality checks on several homogeneous standard financial data, and based on the results of the checks, automatically generate quality tags for each data source and its corresponding standard financial data, and associate the quality tags with the corresponding metadata; In this embodiment, automated quality inspection is performed through a series of predefined quality rule engines: Integrity probing: Calculate the non-null rate and record integrity of key fields (such as company_id, report_date); Consistency probe: Check for consistency across tables and across sources; for example, check whether "Total Assets" is approximately equal to "Liabilities + Owner's Equity"; check whether the industry classification of the same company is consistent across different data sources; Accuracy verification: By comparing with gold datasets (such as authoritative data released by the government) or by verifying based on business logic (such as profit margins should be within a reasonable range); Timeliness assessment: Calculate the delay between data collection time and data business time (such as the financial report release date) to assess the freshness of the data; Based on the exploration results, a weighted scoring model is used to generate a comprehensive quality score (0-100) for each data source and its generated dataset, and map it to a quality label (e.g., >90 is "high", 70-90 is "medium", <70 is "low"). This quality label is stored as an important metadata and directly affects subsequent knowledge fusion decisions.

[0034] In one alternative implementation, a pre-built joint extraction model is used to extract and model knowledge from several standard financial data sets, and a swarm intelligence optimization algorithm is introduced for knowledge fusion and governance to obtain a financial knowledge graph, including: S2021: Using a large-scale financial corpus, a general pre-trained language model is continuously pre-trained to obtain a Financial-Bidirectional Encoder Representations from Transformers (Fin-BERT) model that deeply understands financial contexts, and a joint extraction model is built based on the Fin-BERT model. In this embodiment, firstly, a large-scale financial corpus is constructed or connected, containing several listed company annual reports, announcements, securities research reports, financial news, etc. Then, a powerful general-purpose pre-trained model (such as BERT-wwm-ext) is selected and continuously pre-trained on this financial corpus. The training tasks include masking language model and next sentence prediction. The goal is to enable the model to learn financial domain-specific vocabulary (such as "goodwill impairment" and "interbank lending"), syntactic structure, and semantic relationships, and finally obtain the Fin-BERT model. Based on the Fin-BERT model, a joint extraction model based on sequence labeling is constructed. This model unifies entity recognition and relationship extraction tasks into a single labeling problem. For example, a labeling system is designed, such as B-Company (the beginning of a company entity), I-Company (the interior of a company entity), B-Invest (the beginning of an investment relationship), I-Invest (the interior of an investment relationship), etc. The model outputs a label sequence for each token in the input text. By decoding this sequence, entities and their relationships can be extracted simultaneously. This approach avoids the error propagation problem of pipelined methods and improves the extraction accuracy. S2022: Using a joint extraction model, knowledge is extracted and modeled from several standard financial data to obtain several knowledge triples including entities, relations and attributes; In this embodiment, preprocessed standard financial data (especially the text content, such as the "major events" section of company announcements and news articles) is input into the joint extraction model. The model will output a large number of knowledge triples in the form of (head entity, relation, tail entity). S2023: Perform knowledge fusion on several knowledge triples, and introduce particle swarm optimization algorithm to resolve knowledge conflicts in the knowledge triples that generate conflicts, to obtain several fused knowledge triples. In this embodiment, knowledge fusion is the core of ensuring the quality of the knowledge graph, and this step includes two sub-processes: Entity alignment and linking: Merging entities from different sources that point to the same real-world object in a knowledge triple. For example, linking entities with different expressions such as "Guizhou XX", "X Shares", and "6***19" to a unique entity ID in the knowledge base (such as Q12345). This is usually calculated based on the similarity of entity names, the similarity of contextual information (such as industry and main business), and the similarity of entity attributes. Knowledge conflict resolution: After linking entities, check whether there are conflicting values ​​for the same attribute of the same entity; for example, for entity Q12345, its net profit in 2022 has a value of "62.716 billion yuan" in data source A and a value of "62.615 billion yuan" in data source B; at this time, the particle swarm optimization algorithm is started for intelligent resolution. S2024: Based on several fused knowledge triples, construct a financial knowledge graph, store the financial knowledge graph in the construction graph database, and manage the version updates of the financial knowledge graph; In this embodiment, the fused knowledge triples are imported into a graph database (such as Neo4j or JanusGraph). Each entity is a node in the graph, each relation is a directed edge in the graph, and attributes are attributes of nodes or edges. The native storage and query capabilities of graph databases (such as Cypher or Gremlin languages) are very suitable for handling complex relational queries. At the same time, a version control mechanism is established, and each addition, deletion, or modification operation on the graph generates a new version snapshot, which supports advanced applications such as data backtracking, auditing, and A / B testing.

[0035] In one optional implementation, several knowledge triples are fused, and a particle swarm optimization algorithm is introduced to resolve knowledge conflicts in the conflicting knowledge triples, resulting in several fused knowledge triples, including: S20231: Link entities of several knowledge triples, link each entity to a unique entity ID in the knowledge base, and monitor conflicts during the entity linking process; In this embodiment, when performing entity linking, the similarity score of each candidate entity pair is recorded. When multiple candidate entity scores are similar and exceed the threshold, they are marked as "link ambiguity", requiring manual intervention or a more advanced disambiguation strategy. More importantly, when the same attribute values ​​of two linked unique entity IDs differ, a "knowledge conflict" event is automatically generated, and the details of the conflict are recorded. S20232: If knowledge triples from different data sources conflict during the knowledge fusion process, a particle swarm optimization algorithm is introduced to find the corresponding optimal resolution strategy. In this embodiment, when a knowledge conflict event is detected, for example, when attribute P of entity E has a conflicting set of values. ,in, For data source The conflict value, l For data source indication quantity, n Given the total number of data sources, instead of using a simple "voting" or "get the latest" strategy, an optimization task is initiated with the goal of finding the optimal solution from all possible solutions. S20233: According to the optimal resolution strategy, knowledge conflict resolution is performed on the knowledge triples that generate conflicts, and knowledge fusion is continued on other knowledge triples until all knowledge triples have completed knowledge fusion, resulting in several fused knowledge triples. In this embodiment, if the optimal strategy given by the particle swarm optimization algorithm is "adopt the value of data source A",... In a knowledge graph, an entity... E Attributes P Set as At the same time, a proofance information is attached to this attribute to record the basis, time, algorithm version, etc. of this resolution decision to enhance transparency and traceability. Then, the next knowledge conflict is processed until all conflicts are resolved.

[0036] In one alternative implementation, if knowledge triples from different data sources conflict during knowledge fusion, a particle swarm optimization algorithm is introduced to find the optimal resolution strategy, including: S202321: If knowledge triples from different data sources conflict during the knowledge fusion process, extract the conflict description of the conflicting knowledge triples and formalize the conflict resolution problem of the conflict description into an optimization problem; In this embodiment, the conflict description is a structured data object that includes at least: conflict entity ID, conflict attributes, and a list of conflict values. The problem can be formalized as: maximizing a fitness function; S202322: Based on the conflict description and all possible resolution schemes, define the solution space of the optimization problem, encode the resolution schemes as the positions of particles in the particle swarm optimization algorithm, and design the fitness function; In this embodiment, the solution space is defined as follows: the solution space contains all possible solutions. For the above-mentioned conflict, the solution can be: {adopt A, adopt A} ..., Adopt ..., Manual review}; Each scheme is encoded as the position of a particle. For example, one D dimensional vector If the solution is "adopted" ", can then be designed as ,in, j For data source indication; The formula for the fitness function is:

[0037] In the formula, The position of the particle X The fitness value of the corresponding digestion scheme; The authority score for the data source associated with the quality label is used to assess the authority of the knowledge source on which the solution relies; the lower the quality label, the lower the score. The timeliness of information is scored to assess the freshness of the knowledge upon which the solution relies; the closer the information is to the present, the higher the score. The knowledge consistency score assesses the degree to which a solution matches existing knowledge in the knowledge graph. A solution that is highly consistent with existing reliable knowledge is more likely to be correct and will receive a higher score. The fitness weights sum to 1; S202323: Use Logistic mapping to generate chaotic sequences, map the chaotic sequences to the solution space, obtain the initial position of each particle in the initial particle swarm, and initialize the velocity and individual optimal position of each particle, as well as the global optimal position of the initial particle swarm, based on the fitness function. The formula is:

[0038] In the formula, For the first i+ 1. i One chaotic variable; The stability coefficient is typically 4. This sequence is ergodic and random, which ensures that the initial particle swarm is uniformly distributed in the solution space, avoiding getting trapped in local optima, and is superior to traditional random initialization.

[0039] In the formula, The first in the particle swarm i One particle; For the first i One chaotic variable; These are the upper and lower bounds of the search space; i For particle indication; S202324: Update the position of each particle in the initial particle swarm to obtain the updated position of each particle in the updated particle swarm, and perform Gaussian mutation on some particles in the updated particle swarm with mutation probability to obtain the mutated position of the particle. The formula is:

[0040] In the formula, Number of iterations t+ 1 of i The search direction for each particle; Inertial weight; Number of iterations t The globally optimal position; Number of iterations t The i The position of each particle; For individual learning factors and social learning factors, both are random numbers between [0,1]. t This represents the current iteration number; A random number uniformly distributed between [0, 1]. Number of iterations t The i The updated particle's own optimal position found so far;

[0041] In the formula, Number of iterations t+ 1 of i The updated position of each particle; Number of iterations t The i The position of each particle; Number of iterations t+ 1 of i The search direction for each particle; With a small mutation probability A Gaussian mutation operation is performed on a subset of particles, using the following formula:

[0042] In the formula, Number of iterations t+ 1 of iThe mutated position of each particle; Gaussian noise helps the algorithm escape local optima in the later stages, increasing population diversity. Standard deviation; S202325: If the global optimal position has not improved for several consecutive generations, reverse learning is triggered, and the dynamic reverse position of the particle corresponding to the current global optimal position is obtained. In this embodiment, a stagnation threshold is set. gen_stagnant ,if Fitness values ​​are continuous gen_ stagnant If there is no improvement over a given number of iterations, it is assumed that the algorithm may be trapped in a local optimum. Reverse learning, the formula is:

[0043] In the formula, Number of iterations t+ 1 of i The dynamic reverse position of each particle; Number of iterations t+ 1 of i The reverse solution of each particle; The fitness function;

[0044] In the formula, Number of iterations t+ The reverse center point of 1, determined by the dynamic boundary of the current search space; It is a dynamic inverse factor that increases with iteration (intensified inverse exploration in later stages). S202326: Based on the fitness function, obtain the fitness value of each particle's updated position, mutated position, and / or dynamic reverse position, and update the individual optimal position of each particle and the updated global optimal position of the particle swarm according to the fitness value. S202327: Repeat the particle swarm update steps until the number of iterations reaches the iteration threshold or the fitness value of the global optimum position meets the requirements, then stop the optimization and output the global optimum position; S202328: Decode the global optimal position to obtain the optimal resolution strategy for the knowledge triplet that generates the conflict; In this embodiment, the particle position vector representing the optimal solution is... Decode back to the specific resolution scheme; for example, if If the solution is decoded as "adopt the value of data source A", then this solution will be used as the final decision, and knowledge conflict resolution will be performed.

[0045] In one optional implementation, based on a financial knowledge graph, a pre-defined automated governance strategy is executed, and data lineage tracing and impact analysis are performed on data sources with low-quality labels, generating an impact analysis report, including: S2031: According to the automated governance strategy, traverse the relationships of all entities in the financial knowledge graph. If the relationship between different entities violates the preset business rules, the corresponding relationship will be marked as abnormal. S2032: Based on the automated governance strategy, dynamically calculate and update the quality label of each data source according to the final adoption rate and correction frequency of knowledge triples from different data sources in the financial knowledge graph. In this embodiment, the final adoption rate is: the percentage of knowledge triples provided by a data source that are ultimately adopted into the knowledge graph (i.e., win in conflict resolution). Correction frequency: Statistics on how much of the knowledge provided by a data source is found to be incorrect and corrected in subsequent governance or updates; Based on these two metrics, the quality score for each data source is recalculated periodically (e.g., monthly), using the following formula:

[0046] In the formula, The current reputation score corresponding to the updated quality label; The historical reputation score corresponding to the quality label before the update; The final adoption rate; The frequency being corrected; This is for quality weights; the updated quality labels will be rewritten to the metadata database. S2033: Automatically construct a full-link lineage graph from the data source to the business application of the financial knowledge graph for each knowledge triple in the financial knowledge graph; In this embodiment, data lineage is a core governance capability. Detailed operation logs are recorded at each stage of data processing (collection, preprocessing, extraction, and fusion). Based on these logs, a lineage graph can be constructed for any data item in the knowledge graph (such as the net profit attribute of entity Q12345). This graph is a directed acyclic graph, where nodes represent the various stages and intermediate products of data processing, and edges represent data transformation and flow. For example: Data source A table → Preprocessing script 1 → Extract triples → Fusion decision → Knowledge graph attribute → Business application C; S2034: Based on the full-link lineage diagram, retrieve the downstream knowledge triples and business applications corresponding to the data sources with low-quality tags, and generate an impact analysis report on financial data management using a preset template based on all downstream knowledge triples and business applications. In this embodiment, when the quality label of a data source (such as data source A) is downgraded to "low", an impact analysis task is triggered: Location: In the lineage graph, find all paths that originate from data source A; Search: Along these paths, find all downstream knowledge triples (i.e., triples that use data from data source A) and the final business applications (such as "financial risk warning model" and "portfolio analysis dashboard" that use these triples). Report generation: Using a preset Markdown or HTML template, the retrieved information is automatically populated to generate a structured impact analysis report. The report includes: the source of the problem, details of the quality problem, a list of affected business applications, the number of affected knowledge items, a description of potential risks, and recommended remedial measures (such as recalibrating the data source, manually reviewing downstream applications, etc.).

[0047] In one optional implementation, based on a financial knowledge graph, intelligent analysis is performed on the user-input question data to generate an intelligent analysis report and automatically generate a financial data asset map, including: S2041: Receive user-inputted question data according to the preset natural language query interface; In this embodiment, a user-friendly web interface or API is provided, allowing users to ask questions in natural language; for example: "Please analyze for me, what risks might Company A's acquisition of Company B bring to Company A's supply chain?" S2042: If the question data is a simple query, then the question data will be parsed into a graph query language, queried in the financial knowledge graph, and a corresponding structured answer will be generated. Based on the structured answer, a corresponding intelligent analysis report will be generated. In this embodiment, for simple fact queries, such as "Who is the chairman of Company A?", the query intent is identified as "query attribute", the entity is "Company A", and the attribute is "chairman" through intent recognition and entity linking. Then, a graph query statement is automatically generated (e.g., Cypher:MATCH (c:Company {name:'Company A'})-[:HAS_CHAIRMAN]→ (p:Person)RETURN p.name), the query is executed, and the results are presented in the form of cards or lists. S2043: If the problem data contains complex analysis tasks, then the complex analysis tasks are decomposed, and the task entities are located in the financial knowledge graph to obtain several task nodes, and then proceed to the next step. In this embodiment, complex analysis tasks are broken down into a series of sub-tasks: Locating the "acquisition event": (Company A) - [:ACQUIRE] → (Company B); Locating "Company A's suppliers": (s:Company)-[:SUPPLIES_TO]→(Company A); Locate "Company B's business": (Company B) - [:HAS_BUSINESS] → (b:Business); Find the connections: Is there a competitive or substitutive relationship between (s) and (b)? Locate these task entities and relationships in the knowledge graph to obtain a set of starting nodes and target nodes; S2044: Based on several task nodes and related intermediate nodes, extract the financial knowledge subgraphs related to complex analysis tasks from the financial knowledge graph; In this embodiment, a K-hop neighbor expansion (e.g., K=2 or 3) is performed with the task node as the center to extract a subgraph containing all relevant nodes and edges; this subgraph contains all the potential information needed to answer the complex question and is the basis for subsequent in-depth analysis; S2045: Use a pre-built node feature extraction model to extract the node feature vector of each node in the financial knowledge subgraph; In this embodiment, node features are crucial for path search. A Graph Neural Network (GNN), such as Graph Sampling and Aggregation (GraphSAGE) or Graph Attention Network (GAT), is employed to learn the embedded representations of nodes. The GNN generates a low-dimensional, dense feature vector for each node by aggregating its own attributes and information from its neighbors. This vector captures the node's structural role and semantic information within the graph. S2046: Based on the node feature vectors and the relational edge weights of the financial knowledge subgraph, the ant colony algorithm is used to perform path search in the financial knowledge subgraph to obtain several high-confidence paths. S2047: Perform risk analysis on high-confidence paths, obtain risk scores, and generate corresponding intelligent analysis reports by combining structured answers; In this embodiment, a risk assessment is performed on each found high-confidence path: Risk factor identification: Identify whether there are negative entities (such as "litigation" or "regulatory penalties") or negative relationships (such as "termination of cooperation" or "competitive relationship") along the path. Risk score: A comprehensive risk score is calculated based on the number of negative nodes on the path, the strength of the relationship (edge ​​weight), and the authority of the nodes. Finally, all high-confidence paths, their corresponding risk scores, and key evidence along the paths (such as original news articles and financial report excerpts) are integrated to generate a visually appealing and logically clear intelligent analysis report. S2048: Automatically generate financial data asset maps based on metadata from data sources; In this embodiment, the data asset map is a visualized data catalog. Based on the metadata collected step by step, a multi-level, interactive map can be constructed. Macro view: Displays the distribution of data assets by business domain (such as finance, marketing, public opinion) or data source type; Mid-level view: Clicking on a business domain allows you to drill down to see the specific dataset, data table, and its basic information (data volume, update time, quality labels). Micro view: Clicking on a data table will show its field details, data lineage (linked to the lineage graph of the knowledge graph), and which applications use it; Users can quickly discover, understand, and apply for the data assets they need through operations such as searching, filtering, and drill-down.

[0048] In one alternative implementation, based on the node feature vectors and the relational edge weights of the financial knowledge subgraph, an ant colony algorithm is used to perform path search in the financial knowledge subgraph, obtaining several high-confidence paths, including: S20471: Place several artificial ants at the starting node of the financial knowledge subgraph and start the path search; In this embodiment, in a complex analysis task, one or more starting nodes (such as "Company A") and one or more target nodes (such as "Company A's suppliers") are determined; at all starting nodes, M artificial ants are initialized to prepare for path exploration. S20472: Based on the node feature vector, the relation edge weights of the financial knowledge subgraph, and the pheromone, obtain the transition probability of the artificial ant of the current node choosing to move to the next hop neighbor node, and perform node transfer based on the transition probability; The formula is:

[0049] In the formula, For the node Artificial ants Choose to move to the next hop neighbor node. The transition probability; For path The pheromone concentration is initially the same for all edges. As the search progresses, the edges that appear on successful paths will receive more pheromone. For path The heuristic information is the weight of the relationship edges. For example, in the "supply chain risk" scenario, the weight of relationship edges such as "supply", "procurement" and "holding" will be increased, while the weight of "industry peers" and "collaborative R&D" will be lower. node vector The cosine similarity between nodes guides the artificial ants to move towards semantically more relevant nodes, even if there are no direct strong relation edges between them. For node indication; The transfer parameters control the importance of pheromones, heuristic information, and cosine similarity, respectively. For ant indicators; S20473: Repeat the node transfer steps until the current artificial ant reaches the target node of the financial knowledge subgraph, and obtain the search path of the artificial ant; In this embodiment, the artificial ant moves step by step according to the above-mentioned transfer probability until it reaches any target node. At this point, a path search is completed, and the sequence of nodes traversed is recorded, which is a path. S20474: Repeat the path search steps until all artificial ants have completed the path search. Based on the quality of the path search, update the corresponding pheromone and return to the starting node to perform iterative updates of the ant colony. In this embodiment, pheromone updates begin after all M artificial ants have completed their path search: Pheromone evaporation: All pheromones on the edges evaporate at a certain rate, simulating the evaporation process in nature. The formula is:

[0050] In the formula, Number of iterations t+ 1, t pheromones; Volatilization rate; Pheromone release: Each artificial ant releases pheromones along every edge of its path; the amount released is proportional to the path quality; path quality can be defined as the reciprocal of the path length, or the reciprocal of the risk score of the nodes on the path (if a low-risk path is being sought), using the following formula:

[0051] In the formula, Number of iterations t+ 1 artificial ant In the path The amount of pheromones released; For path Path quality; The pheromone update formula is:

[0052] In the formula, Number of iterations t+ 1,t in path The updated pheromones; after the update is complete, all ants return to the starting point, ready for the next iteration; S20475: Repeat the iterative update steps until the number of iterations reaches the iteration threshold, and obtain several search paths with pheromone concentrations higher than the concentration threshold, i.e., several high-confidence paths. In this embodiment, those paths that accumulate a large amount of pheromones during the iteration process are considered to be the optimal or suboptimal paths connecting the starting point and the target. By setting a pheromone concentration threshold, several high-confidence paths can be screened out. These paths represent the most likely logical reasoning chains in the knowledge graph, providing the core basis for the final analysis report.

[0053] This invention also provides a knowledge graph-based financial data management device, referring to... Figure 3 The diagram illustrates a functional unit diagram of a knowledge graph-based financial data management device 300 according to the present invention. The device may include the following units: The financial data acquisition unit 301 is used to acquire raw financial data from different data sources, and to preprocess and assess the quality of the acquired heterogeneous raw financial data to obtain the quality labels of the data sources and several homogeneous standard financial data from different data sources. The knowledge graph construction unit 302 is used to extract and model knowledge from several standard financial data using a pre-built joint extraction model, and to introduce a swarm intelligence optimization algorithm for knowledge fusion and governance to obtain a financial knowledge graph. The influence analysis unit 303 is used to input the spatial distribution map of coating adhesion force as the initial condition into a high-fidelity digital twin pre-constructed based on digital twin technology for coupled simulation and lifetime analysis, so as to obtain the distribution map of the remaining lifetime of the coating. The intelligent analysis unit is used to perform intelligent analysis based on financial knowledge graphs and user-input question data, generate an intelligent analysis report 304, and automatically generate a financial data asset map.

[0054] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus. Memory, used to store computer programs; When a processor executes a program stored in memory, it implements a knowledge graph-based financial data management method of the present invention.

[0055] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EI) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory can include Random Access Memory (RAM), or non-volatile memory, such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0056] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0057] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a knowledge graph-based financial data management method according to embodiments of the present invention.

[0058] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable hardware devices (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0059] The embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (apparatus), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0060] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0061] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0062] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. "And / or" indicates that either one or both can be chosen. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the element.

[0063] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1.A knowledge graph-based financial data management method, characterized in that, The method comprises: Collecting original financial data from different data sources, and preprocessing and quality evaluating the collected heterogeneous original financial data to obtain quality labels of the data sources and several standard financial data from different data sources; Using a pre-constructed joint extraction model to extract and model knowledge from the several standard financial data, and introducing a swarm intelligence optimization algorithm for knowledge fusion and management to obtain a financial knowledge graph; Based on the financial knowledge graph, a preset automatic management strategy is executed, and data blood relationship tracking and influence analysis are performed on the data sources with low quality labels to generate an influence analysis report; Based on the financial knowledge graph, intelligent analysis is performed according to user input question data to obtain an intelligent analysis report, and a financial data asset map is automatically generated. 2.The financial data management method based on a knowledge graph according to claim 1, characterized in that, Collecting original financial data from different data sources, and preprocessing and quality evaluating the collected heterogeneous original financial data to obtain quality labels of the data sources and several standard financial data from different data sources, comprising: Deploying configurable data connectors for different data sources to uniformly access and collect structured, semi-structured and unstructured heterogeneous original financial data; Synchronously collecting original financial data and metadata corresponding to the data sources when collecting the heterogeneous original financial data; Preprocessing the collected heterogeneous original financial data to obtain several standard financial data from different data sources; Automatically quality investigating the several standard financial data, and automatically generating quality labels for each data source and the corresponding several standard financial data according to the obtained investigation results, and associating the quality labels with the corresponding metadata. 3.The financial data management method based on the knowledge graph according to claim 2, characterized in that, Using a pre-constructed joint extraction model to extract and model knowledge from the several standard financial data, and introducing a swarm intelligence optimization algorithm for knowledge fusion and management to obtain a financial knowledge graph, comprising: Using large-scale financial corpus to continuously pre-train a general pre-training language model to obtain a Fin-BERT model with deep understanding of financial context, and constructing a joint extraction model based on the Fin-BERT model; Using the joint extraction model to extract and model knowledge from the several standard financial data to obtain several knowledge triples including entities, relationships and attributes; Fusing the several knowledge triples, and introducing a particle swarm optimization algorithm to resolve conflicts in the knowledge triples to obtain several fused knowledge triples; Constructing a financial knowledge graph based on the several fused knowledge triples, storing the financial knowledge graph in a constructed graph database, and version managing updates of the financial knowledge graph. 4.The financial data management method based on a knowledge graph according to claim 3, characterized in that, Fusing the several knowledge triples, and introducing a particle swarm optimization algorithm to resolve conflicts in the knowledge triples to obtain several fused knowledge triples, comprising: Entity linking the several knowledge triples to unique entity IDs in a knowledge base, and monitoring conflicts in the entity linking process; If the knowledge triples from different data sources conflict during knowledge fusion, introducing a particle swarm optimization algorithm to find the corresponding optimal resolution strategy; According to the optimal resolution strategy, the knowledge conflict resolution of the knowledge triplets in conflict is performed, and the knowledge fusion of other knowledge triplets is continuously performed until the knowledge fusion of all knowledge triplets is completed, and a plurality of fused knowledge triplets are obtained. 5.The financial data management method based on a knowledge graph according to claim 4, characterized in that, If the knowledge triplets from different data sources generate conflicts in the knowledge fusion process, a particle swarm optimization algorithm is introduced to find the corresponding optimal resolution strategy, including: If the knowledge triplets from different data sources generate conflicts in the knowledge fusion process, the conflict description of the knowledge triplets in conflict is extracted, and the conflict resolution problem of the conflict description is formalized as an optimization problem; According to the conflict description and all possible resolution schemes thereof, the solution space of the optimization problem is defined, the resolution schemes are coded as the positions of particles of the particle swarm optimization algorithm, and the fitness function is designed; A chaotic sequence is generated using a Logistic mapping, the chaotic sequence is mapped to the solution space to obtain the initial position of each particle in the initial particle swarm, and the speed and individual optimal position of each particle and the global optimal position of the initial particle swarm are initialized based on the fitness function; The position of each particle in the initial particle swarm is updated to obtain the updated position of each particle in the updated particle swarm, and a part of particles in the updated particle swarm are subjected to Gaussian mutation with a mutation probability to obtain the mutated position of the particles; If the global optimal position does not improve for a plurality of generations, reverse learning is triggered, and the dynamic reverse position of the particle corresponding to the current global optimal position is obtained; The fitness value of the updated position, the mutated position and / or the dynamic reverse position of each particle is obtained based on the fitness function, and the individual optimal position of each particle and the global optimal position of the updated particle swarm are updated according to the fitness value; The particle swarm updating step is repeated until the iteration number reaches a threshold or the fitness value of the global optimal position meets the requirement, and then the optimization stops, and the global optimal position is output. The global optimal position is decoded to obtain the optimal resolution strategy of the knowledge triplets in conflict. 6.The financial data management method based on a knowledge graph according to claim 5, characterized in that, The formula of the fitness function is: wherein is the position of the particle X a fitness value of the corresponding resolution scheme; is a data source authority score related to the quality label; is an information timeliness score; is a knowledge consistency score; is a fitness weight, and is 1. 7.The knowledge graph-based financial data management method of claim 6, wherein, Based on the financial knowledge graph, a preset automatic management strategy is executed, and data blood relationship tracing and influence analysis are performed on the data source with a low quality label to generate an influence analysis report, including: According to the automatic management strategy, the relationships of all entities of the financial knowledge graph are traversed, and if the relationships between different entities violate the preset business rules, the corresponding relationships are marked as abnormal; According to the automatic management strategy, the final adoption rate and the correction frequency of the knowledge triplets from different data sources in the financial knowledge graph are used to dynamically calculate and update the quality label of each data source; A full-link blood relationship graph from the data source to the business application of each data item of the knowledge triplet in the financial knowledge graph is automatically constructed; According to the full-link blood relationship graph, the downstream knowledge triplets and business applications corresponding to the data source with a low quality label are retrieved, and an influence analysis report of financial data management is generated using a preset template according to all downstream knowledge triplets and business applications. 8.The financial data management method based on a knowledge graph according to claim 7, characterized in that, Based on the financial knowledge graph, according to the user input question data, intelligent analysis is carried out to obtain an intelligent analysis report, and a financial data asset map is automatically generated, including: According to the preset natural language query interface, the user input question data is received; If the question data is a simple query, the question data is parsed into a graph query language, and a corresponding structured answer is generated in the financial knowledge graph, and an intelligent analysis report corresponding to the structured answer is generated; If the question data contains a complex analysis task, the complex analysis task is decomposed, the task entity is positioned in the financial knowledge graph, a plurality of task nodes are obtained, and the next step is entered; According to the plurality of task nodes and the related intermediate nodes, the financial knowledge subgraph related to the complex analysis task in the financial knowledge graph is extracted; Using a pre-constructed node feature extraction model, a node feature vector of each node in the financial knowledge subgraph is extracted; According to the node feature vector and the relationship edge weight of the financial knowledge subgraph, an ant colony algorithm is used to search the path in the financial knowledge subgraph, and a plurality of high-confidence paths are obtained; Risk analysis is performed on the high-confidence paths to obtain a risk score, and an intelligent analysis report corresponding to the structured answer is generated; According to the metadata of the data source, a financial data asset map is automatically generated. 9.The financial data management method based on the knowledge graph according to claim 8, characterized in that, According to the node feature vector and the relationship edge weight of the financial knowledge subgraph, an ant colony algorithm is used to search the path in the financial knowledge subgraph, and a plurality of high-confidence paths are obtained, including: Placing a plurality of artificial ants on the starting node in the financial knowledge subgraph, and starting the path search; According to the node feature vector, the relationship edge weight of the financial knowledge subgraph and the pheromone, the transition probability of the artificial ant moving to the next hop neighbor node is obtained, and the node transition is performed based on the transition probability; Repeat the node transition step until the current artificial ant reaches the target node of the financial knowledge subgraph to obtain the search path of the artificial ant; Repeat the path search step until all artificial ants complete the path search, update the corresponding pheromone according to the quality of the path search, and return to the starting node for iterative updating of the ant colony; Repeat the iterative updating step until the number of iterations reaches the iteration threshold value, and obtain a plurality of search paths with a pheromone concentration higher than the concentration threshold value, i.e. a plurality of high-confidence paths. 10.A financial data management device based on a knowledge graph, used to implement the financial data management method based on a knowledge graph according to any one of claims 1-9, characterized in that, The device comprises: A financial data acquisition unit for acquiring raw financial data from different data sources and preprocessing and quality evaluating the acquired heterogeneous raw financial data to obtain quality labels of the data sources and a plurality of standard financial data from different data sources; A knowledge graph construction unit for using a pre-constructed joint extraction model to extract knowledge from the plurality of standard financial data and model the knowledge, and introducing a swarm intelligence optimization algorithm to fuse and manage the knowledge to obtain a financial knowledge graph; An impact analysis unit for inputting the coating adhesion spatial distribution map as an initial condition into a high-fidelity digital twin body pre-constructed based on digital twinning technology for coupling simulation and life analysis to obtain a coating residual life distribution map; The intelligent analysis unit is configured to perform intelligent analysis based on the financial knowledge graph according to the question data input by the user, obtain an intelligent analysis report, and automatically generate a financial data asset map.

Images