A system and method for entering data assets into a table based on a multi-level knowledge graph

Abstract

The present invention proposes a system and method for entering data assets into a table based on a multi-level knowledge graph, which relates to the fields of data asset authentication and evaluation, knowledge graphs, and artificial intelligence technology. The improved JERE algorithm is used to automatically extract key entities and relationships from enterprise information system logs, enterprise documents, and external knowledge sources to construct a multi-level knowledge graph. Data resources are intelligently processed through clustering algorithms and other means to form data products, and accounting ledgers and recommended data asset combinations are automatically matched through AI algorithms. Accounting subjects are intelligently matched, combined with data product architecture and accounting ledgers, the value of data assets is intelligently evaluated according to preset calculation formulas, and an evaluation report is automatically generated. The present invention improves the efficiency of enterprise data asset management and decision-making support capabilities.

Description

Technical Field

[0001] The present invention relates to the fields of data asset authentication and evaluation, knowledge graphs and artificial intelligence technology, and specifically to a system and method for entering data assets into a table based on a multi-level knowledge graph. The present invention conducts in-depth analysis of log data in an enterprise information system, combines enterprise business processes and process documents, as well as external knowledge bases and case libraries, to construct a multi-level knowledge graph to achieve comprehensive and accurate identification and evaluation of enterprise data assets, and ultimately enters data assets into a table to generate a general ledger, providing strong support for enterprise data management and decision-making. Background Art

[0002] (Background Analysis) With the rapid development of information technology, the amount of enterprise data has shown explosive growth, and data has become an important asset of enterprises. However, how to effectively identify, evaluate and manage these data assets so that they can better serve the business development and decision support of enterprises is a major challenge facing enterprises. At present, enterprise data asset identification and management mainly rely on manual data analysis and annotation, which is not only inefficient, but also easily affected by subjective factors, resulting in low accuracy of data asset identification. In addition, traditional data asset identification methods often only focus on the basic attributes and characteristics of data, ignoring the correlation and business logic between data, and it is difficult to fully reflect the value and potential of data assets. As a structured semantic knowledge base, knowledge graphs can effectively reveal the correlation between data through the representation of entities, relationships and attributes.

[0003] (Patent comparison) CN111539770A discloses a data asset intelligent evaluation method and system. This method uses user-defined evaluation indicators, which are highly subjective and easily affected by human factors. Due to differences in the understanding and needs of data assets among different evaluators, large deviations may occur in the evaluation results. In addition, in the process of data asset confirmation and evaluation, manual work such as data cleaning, classification and labeling is required, which makes the entire evaluation process not very intelligent.

[0004] CN113988606A discloses a data asset value assessment method based on blockchain. It determines the relevant data assets by mining the mutual influence of new data assets, original data and other node data, and conducts preliminary assessment based on the original assessment results of the relevant data assets, and performs node consensus authentication to obtain the final assessment results of the new data assets, with the purpose of making the data asset assessment process open and transparent. However, this method mainly relies on data mining and blockchain authentication to determine the ownership of data assets, while ignoring the assessment of data quality, value connotation and other aspects, which to a certain extent affects the accuracy of the assessment results. Although blockchain technology provides a safe and reliable data storage and access mechanism, in actual operation, the valuation and identification of data assets are not directly related to the application of blockchain technology.

[0005] To sum up, a data asset entry solution based on a multi-level knowledge graph is needed.

[0006] Patent document 1: China Patent Gazette CN111539770A, 2020.08.14

[0007] Patent Document 2: China Patent Gazette CN 113988606A, 2022.01.28 Summary of the invention

[0008] The present invention aims to provide a system and method for entering data assets into a table based on a multi-level knowledge graph. There are five modules, namely, a data resource module, a data resource product module, a data product asset module, an intelligent report generation module and an algorithm configuration module. It is characterized in that the JERE algorithm improved by joint entity and relationship extraction is used to construct a three-layer knowledge graph, which can automatically identify data resources, and can intelligently process data resources through algorithms such as clustering to form data products. Finally, the AI ​​algorithm is applied to automatically match accounting detailed accounts for data assets and intelligently recommend the combination of data assets. Compared with the prior art, the advantage of the present invention is that firstly, the data resource module is used to generate a multi-level knowledge graph based on the underlying data, and the original data of the enterprise is accurately and effectively converted into data resources. Secondly, the data resource product module intelligently processes data resources into data products through specific algorithms, and records the data resources contained in the data products; in the data product asset module, the evaluation results can be given more objectively, which improves the applicability and professionalism of the method.

[0009] The data resource module includes three sub-modules: system log knowledge graph sub-module, enterprise document knowledge graph sub-module, and external knowledge graph sub-module.

[0010] The system log knowledge graph submodule is based on the improved JERE algorithm, which automatically receives and stores operation logs and transaction data from the core information systems of the enterprise (such as ERP, CRM, etc.), and then extracts key data objects (such as customers, orders, products) and data flows from the logs. Based on the extracted entities and relationships, the system log knowledge graph displays the internal structure and basic associations of data assets, and assigns weights to each entity and relationship, providing guidance for the subsequent construction of the knowledge graph; when processing sensitive data, encryption or desensitization measures are adopted, strict data access log records are implemented, and access and operations on sensitive data are monitored;

[0011] The enterprise document knowledge graph submodule analyzes the enterprise's standard operating procedures (SOP), processes and other documents based on the improved JERE algorithm. Based on the system log knowledge graph, we can accurately identify key entities and their relationships. This information is the basis for building the enterprise document knowledge graph. Compared with directly using algorithms to mine enterprise documents, the system log knowledge graph can more realistically map the user's actual operation data and processes, effectively reduce subjective judgment bias, and enhance the expression of business logic. On this basis, we map business knowledge and rules into the system log knowledge graph and assign weights to the extracted business processes and rules, which not only enriches the semantic information of data assets, but also ensures the accuracy and practicality of the graph.

[0012] The external knowledge graph submodule uses technologies such as APIs and web crawlers to collect external knowledge such as industry reports, market analysis, and public data sets based on the weights of each entity and relationship given by the system log knowledge graph and the enterprise document knowledge graph, integrate external knowledge with the first two layers of knowledge graphs, and build an external knowledge graph through knowledge fusion and reasoning technology to expand the industry background and cognitive boundaries of data assets. When introducing external knowledge, ensure that best practices and industry standards for privacy protection are included;

[0013] Through entity alignment and relationship fusion algorithms, redundancy and contradictions in knowledge graphs at different levels are eliminated to build a unified comprehensive knowledge graph. Data resources with high value, low risk and strong availability are identified and marked, and corresponding privacy protection measures are recommended. The multi-layer knowledge graph fusion algorithm is as follows:

[0014] In the process of multi-layer knowledge graph fusion, not only the weight of nodes (entities) but also the weight of edges (relationships) should be considered, while ensuring that information from different sources is reasonably integrated.

[0015] Definition: Graph extracted from enterprise logs, where is a set of entity tags, is a set of relations, each edge Concatenate two entity tags with weights . set up Graphs, structures and Similar. is an external knowledge graph, which also contains a set of entity tags and relationship set ,side With weight . set up It is a mapping function used to match entity labels from different sources to ensure consistency during fusion.

[0016] Node weight fusion: For each entity label , the node weight after fusion The weighted sum can be calculated based on the frequency or importance of each source, and the formula is:

[0017]

[0018] in, They are exist importance score in (e.g., number of occurrences or a centrality measure calculated based on some algorithm), while is the corresponding weight factor, reflecting the trust or importance of different sources.

[0019] Edge weight fusion: For each pair of entity labels The relationship between Direct relationships and indirect relationships through mapping should be considered, and the formula is expressed as:

[0020]

[0021] in, It is in the picture middle and The original weight of the relationship between and It is a mapping application to ensure the consistency of entity tags, and is the weight factor of different graphs. Through this method, a multi-level knowledge graph can be constructed through the extracted entities and relationships.

[0022] The data resource productization module will intelligently classify systematized data resources based on a variety of clustering algorithms and community discovery algorithms. These algorithms can automatically identify patterns and structures in the data and group data resources according to their inherent connections and similarities, thereby achieving efficient data management and operations. Through this process, data resources are given clear category labels, which provides a solid foundation for subsequent data operations. Data resources of each category can be processed and analyzed in a targeted manner according to their characteristics and needs. For example, similar data can be aggregated together for batch processing or analysis to improve efficiency and accuracy.

[0023] The data product assetization module uses AI algorithms to intelligently match accounting ledgers with clearly categorized data products based on underlying data to assess asset value, and uses external knowledge to make reasonable future value estimates for data assets.

[0024] In addition, the data product assetization module develops an intelligent combination algorithm to evaluate the value and risk of data assets from multiple dimensions such as business, technology, and economy based on the attributes, correlations, and business needs of data assets, analyze the acquisition, processing, storage, and transmission capabilities of data assets, and determine the actual availability of data assets; the system receives, acquires, and stores data assets that need to be valued; the source data is processed according to the evaluation indicators of preset data dimensions to calculate the value of data assets, and machine learning and data analysis techniques are used to intelligently recommend the optimal data asset combination based on the company's business goals and financial status.

[0025] The intelligent report generation module is designed to meet different user roles and business needs, covering multiple areas such as management decision-making, investor relations, and compliance reporting. It integrates financial data and business indicators to provide professional advice on the optimal layout of data assets in the balance sheet, income statement, and cash flow statement. Based on the evaluation results, the module generates a detailed data asset evaluation report, providing a solid scientific basis for the company's data management and decision-making. When evaluating data assets, the module specifically includes privacy risk assessment to ensure that the use of data assets is in full compliance with privacy regulations and avoids the risk of violations. The report lists privacy protection measures and legality recommendations in detail to help companies use data safely and in compliance. In addition, the report is presented in the form of an interactive dashboard, allowing users to simulate dynamic data asset combinations. This intuitive display method allows users to instantly observe the potential impact of different data asset combinations on financial statements, so as to make more informed decisions.

[0026] The algorithm configuration module is an important component of the present invention. It logically and structuredly expresses and stores key information such as algorithms, calculation formulas, parameters, output weights, and scoring indicators in the form of multi-tuples. This information is carefully designed and stored in the database to facilitate efficient data management and call. The design of this module allows it to be seamlessly integrated with various algorithms. Whether it is traditional statistical analysis methods or modern machine learning techniques, they can be integrated and applied through the algorithm configuration module. This flexibility enables the module to adapt to different business scenarios and needs, thereby achieving a highly automated and intelligent calculation process. The automation characteristics of the algorithm configuration module mean that it can automatically perform complex calculation tasks without human intervention. This not only improves the calculation efficiency, but also reduces the possibility of human errors. The intelligent computing power is reflected in the module's ability to dynamically adjust the algorithm parameters and optimize the calculation process according to the characteristics of the input data to achieve the best calculation effect. In addition, the algorithm configuration module also has strong scalability. With the development of business and the advancement of technology, new algorithms and computing needs continue to emerge. The design of the module allows it to easily add new algorithms and parameters to adapt to these changes and ensure the long-term effectiveness and competitiveness of the system.

[0027] The present invention proposes a system and method for entering data assets into tables based on a multi-level knowledge graph, which realizes efficient management and accurate evaluation of data assets through five key modules: data resourceization, data resource productization, data product assetization, intelligent report generation and algorithm configuration. Using the knowledge graph constructed by the improved JERE algorithm, this method not only integrates system logs, corporate documents and external knowledge, but also provides dynamic management and real-time monitoring capabilities of data assets. The triple representation of the algorithm configuration module further enhances the flexibility and efficiency of automated and intelligent computing. The implementation of the present invention significantly improves the accuracy of data asset identification and evaluation, and provides innovative solutions for the optimized management and decision support of enterprise data assets. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 The diagram is a schematic diagram of the method and data flow in the present invention, showing the flow path of data input, processing steps and output results of the core algorithm of the present invention;

[0029] Figure 2 This is a schematic diagram of the operation flow of the present invention, which lists in detail the specific steps when executing the method of the present invention. Implementation

[0030] In a specific embodiment of the present invention, a system and method for entering data assets into a table based on a multi-level knowledge graph is constructed to comprehensively evaluate the data assets of an enterprise and generate a customized evaluation report.

[0031] The system and method include a data resource module, a data resource product module, a data product asset module, an intelligent report generation module and an algorithm configuration module. The specific implementation is as follows:

[0032] The data resource module: First, based on the JERE algorithm, the module uses the bidirectional long short-term memory network technology (Long Short-Term Memory, LSTM) to extract key data objects from enterprise documents, system logs and data crawled from the Internet, and thus constructs the system log knowledge graph, enterprise document knowledge graph and external knowledge graph;

[0033] The JERE algorithm is an algorithm specifically used for joint extraction of entities and relations. The core idea is to integrate the two tasks of entity recognition (ER) and relation classification (RC) into a unified model. This integrated model improves the learning efficiency and prediction accuracy of the model by sharing encoding layers and parameters.

[0034] Improved JERE model:

[0035] ① Realize the integration of entity recognition and relationship classification: The present invention integrates entity recognition and relationship classification tasks into a unified model, and improves the learning efficiency and prediction accuracy of the model by sharing encoding layers and parameters.

[0036] ②Introduction of external knowledge: The present invention introduces an external knowledge graph in the process of entity recognition and relationship classification, and uses external knowledge to enhance the model's understanding of entities and relationships, thereby improving the accuracy and comprehensiveness of the extraction.

[0037] ③ Realize weighted fusion of entities and relationships: When constructing a knowledge graph, the present invention not only considers the frequency of occurrence of entities and relationships, but also their importance. By weighted fusion of information from different sources, a more accurate and reliable knowledge graph is constructed.

[0038] Specific mathematical formula:

[0039] In the entity recognition step, the improved JERE algorithm is used to identify entities in the text through the following mathematical model:

[0040]

[0041] in, Represents the set of recognized entities. Represents the set of all possible entities. Represents the text entered. Represents the parameters of the entity recognition model. Indicates the number of entities in the text. Indicates that in a given text and model parameters In the case of Probability of occurrence. It means to find the function with the largest value. .

[0042] In the relationship classification step, the improved JERE algorithm is used to classify the relationship between entity pairs through the following mathematical model:

[0043]

[0044] in, Represents the set of identified relations. Represents the set of all possible relationships. Indicates Entity pairs. Represents the parameters of the relation classification model. Represents the number of entity pairs in the text. Indicates that in a given entity ,text and model parameters In the case of Probability of occurrence. It means to find the function with the largest value. .

[0045] In the weight fusion step, weights are assigned to each identified entity and relationship and fused using the following formula:

[0046]

[0047]

[0048] in, Representing Entities The weight of . Representing relationships The weight of . Represents a collection of text. Indicated in text Medium Entity The weight of . Indicated in text China Relations The weight of .

[0049] In the clustering enhancement step, the entities are grouped using a clustering algorithm and the weights of entities and relationships are enhanced through the following mathematical model:

[0050] in, Indicates clusters, represents the set of all possible clusters, Represents a collection of entities. Represents the parameters of the clustering model. Indicates that in a given cluster and model parameters In the case of The probability of belonging to this cluster. It means to find the function with the largest value. .

[0051] The knowledge graph construction steps combine weight fusion and clustering results to construct a multi-level knowledge graph, which is expressed as:

[0052]

[0053] in, Represents the constructed knowledge graph, represents the set of recognized entities, represents the set of identified relations, represents the weight set of the entity, A set of weights representing a relationship.

[0054] In the process of entity and relationship extraction based on the improved JERE model of the present invention, the key mathematical expressions involved mainly revolve around the deep learning framework, especially the neural network-based method. The improved JERE model usually includes components such as word embedding layer, encoding layer (such as bidirectional LSTM), entity recognition layer, and relationship classification layer. The following is the relevant mathematical expression process:

[0055] Input definition: With Long text samples , where each text It can be expressed as A sequence of words , and each word From a predefined vocabulary Our goal is to extract all entities and the relationships between them from these texts.

[0056] Preprocessing and word embedding: Each word Transformed into a fixed-dimensional vector through the word embedding layer ,here is the dimension of the word embedding. If you use pre-trained word embeddings such as Word2Vec or GloVe, then ,in is the word embedding matrix.

[0057] Encoding layer: These embedding vector sequences are processed through an encoder (e.g., bidirectional LSTM) to capture the contextual dependencies between words. Assume that the encoder output is ,in Expressing words In the context of is the dimension of the output of the encoding layer.

[0058] Entity Recognition: For each location , the entity recognition task can be achieved by a classifier based on Output the probability distribution of each position belonging to different entity categories ,in Indicates The entity tag of each word, such as B (beginning), I (internal), O (non-entity) in BIO annotation.

[0059] Relation classification: After entity recognition, the relationship between entity pairs needs to be determined. This usually involves selecting an entity pair and predicting the type of relationship between them based on the contextual representation of the two entities (which may be some combination of the encoder outputs of their start and end positions). Let the recognized entity pair be , then the output of the relation classification task is the relation type The probability distribution of .

[0060] Output: The final output of the model is for each text A series of entity label sequences and relationship label sequences. Specifically: divided into entity output and relationship output. Entity output: ,in The model predicts text, The most likely entity label for each position. Relation output: , lists all predicted relationship types and their corresponding entity pairs. The loss function of the entity recognition task is , which quantifies the difference between the model's predicted entity labels and the true labels. The loss function for the relation extraction task is , measures the accuracy of relationship prediction. Joint loss It may be the weighted sum of the two, and its formula is expressed as:

[0061]

[0062] The data resource module: The second step is to carry out multi-layer knowledge graph fusion. In this process, not only the weight of the node (entity label) but also the weight of the edge (relationship) should be considered, while ensuring that information from different sources is reasonably integrated.

[0063] Definition: Graph extracted from enterprise logs, where is a set of entity tags, is a set of relations, each edge Concatenate two entity tags with weights . set up Graphs, structures and Similar. is an external knowledge graph, which also contains a set of entity tags and relationship set ,side With weight . set up It is a mapping function used to match entity labels from different sources to ensure consistency during fusion.

[0064] Node weight fusion: For each entity label , the node weight after fusion The weighted sum can be calculated based on the frequency or importance of each source, and the formula is:

[0065]

[0066] in, They are exist importance score in (e.g., number of occurrences or a centrality measure calculated based on some algorithm), while is the corresponding weight factor, reflecting the trust or importance of different sources.

[0067] Edge weight fusion: For each pair of entity labels The relationship between Direct relationships and indirect relationships through mapping should be considered, and the formula is expressed as:

[0068]

[0069] in, It is in the picture middle and The original weight of the relationship between and is a mapping application to ensure the consistency of entity tags, and is the weight factor of different graphs. Through this method, a multi-level knowledge graph can be constructed through the extracted entities and relationships.

[0070] The data resource productization module:

[0071] Based on a variety of clustering algorithms and community discovery algorithms, systematic data resources are intelligently classified.

[0072] Before clustering and community discovery, data resources need to be cleaned and preprocessed. This includes removing noise data, filling missing values, standardizing data formats, etc. to ensure data quality and consistency. These algorithms can automatically identify patterns and structures in the data, group data resources according to internal connections and similarities, and extract useful features from the original data to represent the essential attributes of the data while retaining the most important information. Select a suitable clustering algorithm to apply to the preprocessed data set, divide the data into different categories or clusters based on the similarity or distance between data points, and assign clear category labels to the data resources based on the clustering results. This enables efficient data management and operation.

[0073] Through this process, data resources are given clear category labels, which provides a solid foundation for subsequent data operations. Data resources of each category can be processed and analyzed in a targeted manner according to their characteristics and needs. For example, similar data can be aggregated together for batch processing or analysis to improve efficiency and accuracy.

[0074] In addition, intelligent classification also makes it possible to automatically generate data products. The system can automatically select appropriate processing procedures and parameters according to the categories and attributes of the data to generate data products that meet the requirements. These data products not only have a unified format and standard, but also can objectively reflect the essential characteristics and value of the data. In order to further improve the flexibility and adaptability of data operations, we have also designed a dynamic adjustment mechanism. The system can automatically adjust the parameters of the clustering algorithm and community discovery algorithm according to data changes and user needs to optimize the effect of classification. This enables the system to adapt to different data environments and business scenarios and provide more accurate and personalized services.

[0075] The data product assetization module:

[0076] The function of the data product assetization module is crucial. It significantly improves the efficiency and accuracy of data asset valuation through intelligent means. The input data of this module includes data products generated by clustering algorithms and the company's financial ledger. Data products are obtained by clustering analysis of raw data resources, and they are closely related to the company's information flow and logistics. The financial ledger is directly connected to the company's capital flow. By intelligently matching the two, the module can significantly reduce the workload of identification and matching that previously required manual work.

[0077] The core logic of the data product assetization module is to accurately match data products with corresponding accounting details. This process is not only based on the characteristics and value of data products, but also takes into account the company's accounting policies and relevant information in the external knowledge base. After the match is successful, the module uses advanced calculation models to evaluate the value of each data asset to ensure the accuracy and rationality of the evaluation results.

[0078] In addition, the module's intelligent recommendation function can recommend the most appropriate accounting details based on the evaluation results of data assets, helping enterprises to comply with regulations and reflect the true value of data assets in accounting treatment. This recommendation process is dynamic and can be adjusted according to the specific business needs of the enterprise and changes in the market environment.

[0079] Furthermore, the data product assetization module can also recommend a combination of multiple data assets based on the business needs of the enterprise. This combination recommendation is designed to optimize the performance indicators of the enterprise and help the enterprise achieve the optimal allocation of resources and maximize value. Through in-depth analysis and intelligent combination of different data assets, enterprises can respond to market changes more flexibly and improve the scientificity and foresight of decision-making.

[0080] The intelligent report generation module:

[0081] In the implementation of the present invention, the intelligent report generation module plays a vital role. This module is designed to meet different user roles and business needs. It not only serves multiple fields such as management decision-making, investor relations, and compliance reporting, but also integrates financial data and business indicators to provide professional suggestions for the optimal layout of data assets in the balance sheet, income statement, and cash flow statement.

[0082] The core function of the module is to generate an assessment report containing key information based on the normalized assessment score of each data asset. These reports list in detail the scores of each data asset in various dimensions, the risk-adjusted scores, the weighted scores, and the final normalized assessment scores, providing decision support for corporate management and data asset review experts. While generating the assessment report, the module specifically includes a privacy risk assessment to ensure that the use of data assets is in full compliance with privacy regulations and avoids the risk of violations.

[0083] The report lists privacy protection measures and legal recommendations in detail to help companies use data safely and in compliance with regulations. In addition, the intelligent report generation module presents reports in the form of interactive dashboards, allowing users to simulate dynamic data asset portfolios. This intuitive display method allows users to instantly observe the potential impact of different data asset portfolios on financial statements, so as to make more informed decisions.

[0084] The algorithm configuration module:

[0085] The algorithm configuration module of the present invention is the core of realizing the intelligent entry process of data assets. It realizes efficient data management and call by logically and structuredly storing key information such as algorithms, calculation formulas, parameters and scoring indicators in the form of multi-tuples in the central database. The module is flexibly designed and can be seamlessly integrated with various traditional and modern algorithms. It supports the automated execution of complex computing tasks without human intervention. It also has intelligent parameter adjustment capabilities and dynamically optimizes the calculation process according to the characteristics of input data. In addition, the module has strong scalability and can easily add new algorithms and parameters to adapt to business development and technological progress, ensuring the long-term effectiveness and competitiveness of the system. The user-friendly interface allows users to configure parameters, start tasks and monitor the calculation process. The module also includes a security mechanism to protect the integrity and privacy of data.

[0086] Through the above-mentioned embodiments, enterprises can fully understand the value and risks of their data assets while protecting the privacy and security of the data, thereby making more efficient and accurate data asset assessments.

Claims

1. A data asset entry system based on a multi-level knowledge graph, characterized in that: include: Data resource module, data resource product module, data product asset module, intelligent report generation module and algorithm configuration module; The data resource module uses the improved JERE algorithm to automatically extract key entities and relationships from the log data in the enterprise information system, build a system log knowledge graph, and assign weights to entities and relationships, parse enterprise documents, identify key entities and their relationships, build an enterprise document knowledge graph, collect external knowledge through API and web crawler technology, build an external knowledge graph, and jointly form a multi-layer knowledge graph, and input the identified data resources into the data resource productization module; the improved JERE model: integrates entity recognition and relationship classification tasks into a unified model, sharing encoding layers and parameters; introduces external knowledge graphs in the process of entity recognition and relationship classification, and uses external knowledge to enhance the model's understanding of entities and relationships; when building the knowledge graph, not only the frequency of occurrence of entities and relationships is considered, but also their importance, and information from different sources is weighted and integrated; Specific mathematical formula: In the entity recognition step, the improved JERE algorithm is used to identify entities in the text through the following mathematical model: in, represents the identified entity set, E represents all possible entity sets, t represents the input text, θ E represents the parameters of the entity recognition model, N represents the number of entities in the text, and P(e i |t;θ E ) means that given the text t and model parameters θ E In the case of entity e i Probability of occurrence, argmax E It means to find E that maximizes the function value; In the relationship classification step, the improved JERE algorithm is used to classify the relationship between entity pairs through the following mathematical model: in, represents the set of relations that identify R, R represents the set of all possible relations, e j represents the jth entity pair, θ R represents the parameters of the relational classification model, M represents the number of entity pairs in the text, and P(r j |e j θ R ) indicates that in a given entity pair e j , text t and model parameters θ R In the case of j Probability of occurrence, argmax R It means to find R that maximizes the function value; In the weight fusion step, weights are assigned to each identified entity and relationship and fused using the following formula: Among them, W entity(e) represents the weight of entity e, W relation(r) represents the weight of relation r, T represents the text set, w t，e represents the weight of entity e in text t, w t，r Represents the weight of relation r in text t; In the clustering enhancement step, the entities are grouped using a clustering algorithm and the weights of entities and relationships are enhanced through the following mathematical model: Among them, C k represents the Kth cluster, C represents the set of all possible clusters, E represents the entity set, θ C represents the parameters of the clustering model, P(e|C;θ C ) means that given the cluster C and model parameters θ C In the case of , the probability that entity e belongs to the cluster, argmax C It means to find C that maximizes the function value; The knowledge graph construction steps combine weight fusion and clustering results to construct a multi-level knowledge graph, which is expressed as: Among them, KG represents the constructed knowledge graph. represents the set of recognized entities, represents the set of identified relations, W entity Represents the weight set of the entity, W relation A set of weights representing the relationship; The multi-layer knowledge graph includes: the knowledge graphs constructed by the system log knowledge graph submodule, the enterprise document knowledge graph submodule, and the external knowledge graph submodule, which together form the basis of the multi-layer knowledge graph. The redundancy and contradiction in the knowledge graphs at different levels are eliminated through entity alignment and relationship fusion algorithms to construct a unified comprehensive knowledge graph. The data resource productization module is linked to the data resourceization module, and intelligently classifies the systematized data resources based on a variety of clustering algorithms and community discovery algorithms to form data products, and inputs the processed data products into the data product assetization module; The data product assetization module further processes and identifies data assets, and inputs the processed data asset information into the intelligent report generation module; The intelligent report generation module is used to generate an evaluation report of data assets; The algorithm configuration module is connected to the data product assetization module, applies the Al algorithm to automatically match accounting ledgers for data assets, intelligently recommends combinations of data assets, and supports algorithm configuration and optimization of the entire system.

2. The system according to claim 1, characterized in that The data products include data products obtained by clustering analysis of original data resources and financial detailed accounts directly related to the capital flow of the enterprise.

3. The system according to claim 1, characterized in that The algorithm configuration includes: storing the key information such as algorithms, calculation formulas, parameters and scoring indicators in a central database in a logical and structured manner in the form of multiple groups to achieve data management and call.

4. A method for entering data assets into a table based on a multi-level knowledge graph, characterized in that: Applicable to the system for entering data assets into tables based on a multi-level knowledge graph according to claim 1.

Images