Search

CN-122022522-A - Intelligent decision and recommendation method based on enterprise knowledge graph

CN122022522ACN 122022522 ACN122022522 ACN 122022522ACN-122022522-A

Abstract

The application provides an intelligent decision and recommendation method based on an enterprise knowledge graph. According to the method, through constructing the enterprise knowledge graph covering multidimensional entities and association relations of enterprise clients, products, supply chains, finances, staff and the like, semantic integration and association are carried out on heterogeneous data scattered in each business system in an enterprise to form a unified knowledge resource pool, the process can comprehensively and accurately reflect the internal association among the entities in the enterprise operation process, a complete and reliable data base is provided for subsequent decision analysis and recommendation, a multi-scene intelligent reasoning model is constructed based on the enterprise knowledge graph, and deep mining and reasoning analysis can be carried out on various data in the enterprise operation process by utilizing rich semantic association information in the knowledge graph.

Inventors

  • YAO JIEHUA

Assignees

  • 深圳市华思致远技术有限公司

Dates

Publication Date
20260512
Application Date
20260202

Claims (10)

  1. 1. An intelligent decision and recommendation method based on enterprise knowledge graph is characterized by comprising the following steps: s1, collecting multi-source heterogeneous data of an enterprise, performing cleaning, standardization and semantic labeling processing on the data, and extracting related entities, attributes and association relations among the entities of the enterprise to construct an enterprise knowledge graph, wherein the related entities of the enterprise comprise clients, products, suppliers, financial accounts, staff and business processes; S2, receiving a demand instruction input by a user, and analyzing semantic information of the demand instruction through a natural language processing technology by combining with service roles and historical behavior data of the user to determine a decision demand type or a recommendation demand type of the user, wherein the decision demand type comprises a market expansion decision, a supply chain optimization decision, a risk early warning decision and a financial planning decision; S3, extracting corresponding entity, attribute and association relation data from the enterprise knowledge graph according to the decision requirement type or the recommendation requirement type determined in the S2, and inputting the data into a preset intelligent reasoning model for reasoning analysis; S4, screening, sorting and optimizing the candidate decision suggestions or candidate recommended contents generated in the S3 by combining business rules, historical decision data and user feedback data of the enterprise to obtain final accurate decision suggestions or personalized recommended contents; And S5, presenting the accurate decision suggestion or personalized recommendation content obtained in the step S4 to a user in a visual mode, recording feedback information of the user on the result, and using the feedback information to optimize the enterprise knowledge graph and the intelligent reasoning model.
  2. 2. The intelligent decision and recommendation method based on enterprise knowledge graph as claimed in claim 1, wherein in S1, the step of constructing enterprise knowledge graph includes: S11, collecting enterprise internal business system data, external industry data and public data in a data interface calling mode, a web crawler mode and a database export mode; The enterprise internal business system data comprises client management system data, product management system data, supply chain management system data, financial management system data and human resource management system data, wherein the external industry data comprises industry dynamic data, competitor data and market trend data; s12, cleaning the collected multi-source heterogeneous data to remove repeated data, noise data and missing values, carrying out standardization processing on the cleaned data, unifying data formats and coding specifications, carrying out semantic annotation on unstructured data, and extracting key information; S13, extracting an entity, entity attributes and association relations among the entities from the preprocessed data by adopting a mode of combining a rule-based extraction method and a machine learning extraction method, wherein the entity attributes comprise basic information, state information and business related information of the entity, and the association relations comprise a cooperation relation, a supply-demand relation, a membership relation and a flow association relation; And S14, knowledge fusion and storage, namely performing conflict detection and resolution on the extracted knowledge, fusing the knowledge extracted by different data sources, storing the fused knowledge by adopting a graph database, constructing an enterprise knowledge graph, and establishing a knowledge index.
  3. 3. The intelligent decision and recommendation method based on the enterprise knowledge graph as claimed in claim 1, wherein in S2, the specific steps of determining the decision requirement type or recommendation requirement type of the user comprise word segmentation, part-of-speech tagging and dependency syntactic analysis of natural language requirement instructions input by the user, extracting requirement keywords, and semantic expansion of the requirement keywords by combining user business roles, historical operation records and requirement preferences in a user portrait database, so as to determine the core requirement of the user and divide the requirement type.
  4. 4. The intelligent decision-making and recommending method based on the enterprise knowledge graph as claimed in claim 1, wherein in the step S3, the specific steps of the intelligent reasoning model for carrying out reasoning analysis include determining a reasoning task and a reasoning rule according to the type of the requirement, retrieving a sub-graph related to the reasoning task from the enterprise knowledge graph, converting entity, attribute and association relation data in the sub-graph into vector representations identifiable by the model, inputting the vector representations into the intelligent reasoning model, carrying out deep feature mining and association reasoning through a graph neural network algorithm, and generating candidate decision-making suggestions or candidate recommendation contents, wherein the reasoning rule is preset according to business process specifications and decision standards of the enterprise.
  5. 5. The intelligent decision-making and recommending method based on the enterprise knowledge graph as set forth in claim 1, wherein in S4, the optimizing step includes constructing an optimizing scoring model, scoring candidate decision-making suggestions or candidate recommended contents from three dimensions of relevance, feasibility and effectiveness, wherein relevance refers to matching degree of the candidate decision-making suggestions with user requirements, feasibility refers to implementation degree of the candidate decision-making suggestions under existing resources and service conditions of enterprises, effectiveness refers to economic benefits or service improving effects of the candidate contents for the enterprises, and sorting the candidate contents according to the scoring result, and screening out contents with scores higher than a preset threshold value as final accurate decision-making suggestions or personalized recommended contents.
  6. 6. The intelligent decision-making and recommending method based on the enterprise knowledge graph as claimed in claim 1, wherein in S5, the visualization mode is in the first type comprising three forms of graph display, graph visualization and text description; and the feedback information is input into a knowledge graph updating module and a model optimizing module, the knowledge graph updating module supplements or corrects the entity, the attribute and the association relation in the enterprise knowledge graph according to the feedback information, and the model optimizing module adjusts the parameters of the intelligent reasoning model according to the feedback information.
  7. 7. The intelligent decision and recommendation method based on the enterprise knowledge graph according to claim 1, further comprising a dynamic updating step of monitoring changes of enterprise multisource data in real time, and automatically triggering knowledge extraction and knowledge fusion flows to dynamically update the enterprise knowledge graph when new, modified or deleted data occurs.
  8. 8. The intelligent decision and recommendation method based on the enterprise knowledge graph according to claim 1, wherein in S3, the specific construction method of the intelligent reasoning model based on the graph neural network algorithm comprises the following steps: S31, designing a network structure comprising an input layer, a hidden layer and an output layer based on a graph convolution neural network GCN framework, wherein the input layer is used for receiving vector representation of an enterprise knowledge graph subgraph, the hidden layer is provided with 3-5 layers of graph convolution layers, a ReLU activation function is adopted for carrying out characteristic nonlinear conversion, and the output layer adopts a Softmax activation function to output candidate decision suggestions or probability distribution of candidate recommended contents; S32, initializing weight parameters of the graph convolution layer by adopting an Xavier initialization method, initializing bias parameters to 0, setting learning rate to 0.01-0.05, and setting iteration times to 100-200 rounds; S33, selecting enterprise historical decision data and knowledge graph associated data to construct a training data set, taking a cross entropy loss function as an optimization target, adopting a random gradient descent algorithm to carry out iterative optimization on model parameters, calculating the accuracy of a verification set after each iteration, and stopping training when the accuracy of the verification set is continuously 5 times without improvement; And S34, evaluating the trained model by adopting a test set, wherein evaluation indexes comprise accuracy, recall rate and F1 value, determining that the model is constructed when the evaluation indexes are not lower than a preset threshold, and if the evaluation indexes do not reach the preset threshold, returning to the step S31 to adjust the network structure or the step S32 to optimize the initialization parameters, and carrying out training optimization again.
  9. 9. The intelligent decision-making and recommending method based on the enterprise knowledge graph as claimed in claim 1, wherein in S3, the inference rule is preset based on the business process specification and decision criteria of the enterprise, and the specific setting mode and content include the following aspects: setting a basis, namely taking business process management files formally released in enterprises, decision approval specifications, industry supervision requirements and historical successful decision cases as core basis, and combining the characteristics of different decision requirement types to extract quantifiable and executable reasoning logic; the business process association rule is used for defining entity association requirements among different business process links; a decision threshold rule, namely setting a threshold range for the quantitative decision index; risk avoidance rules, namely setting risk exclusion rules in a decision process based on enterprise risk management and control standards; A priority ordering rule, namely setting the priority ordering rule of the decision scheme according to the strategic target weight of the enterprise; and (3) rule management, namely constructing an inference rule base to perform centralized management on preset rules, wherein the rule base supports dynamic modification, addition or deletion according to enterprise business process specification updating, decision standard adjustment and industry requirement change.
  10. 10. The intelligent decision and recommendation method based on enterprise knowledge graph according to claim 9, wherein the decision threshold rule comprises a specific quantitative calculation formula for accurately calculating a decision index threshold, and the specific formula is as follows: The expected return on investment ROI calculation formula is roi= (expected net profit/(total project investment) ×100%, wherein expected net profit = expected annual business income-expected annual operation cost-expected annual tax, expected annual operation cost = fixed cost + unit change cost×expected annual sales; The calculation formula of the investment recovery period comprises the following steps of static investment recovery period = project total investment amount/(annual average net cash flow), wherein annual average net cash flow = annual average business income-annual average payment cost-annual average tax, and if residual value recovery exists, residual value amount is needed to be superposed in the final period net cash flow.

Description

Intelligent decision and recommendation method based on enterprise knowledge graph Technical Field The invention relates to the technical field of enterprise management, in particular to an intelligent decision and recommendation method based on an enterprise knowledge graph. Background In the digital economic age, the operation and development of enterprises is highly dependent on data-driven decision patterns. With the expansion of business scale and diversification of business scenes of enterprises, various data (such as client data, product data, supply chain data, financial data and the like) accumulated in the enterprises show explosive growth, the data types are more complex, and structured data, semi-structured data and unstructured data are covered; The traditional enterprise decision-making and recommending method mainly has the defects that the traditional method is mainly based on single data source or isolated data system for analysis, cross-department and cross-business data association fusion cannot be realized, and the overall situation of enterprise operation is difficult to comprehensively reflect. Therefore, the intelligent decision and recommendation method based on the enterprise knowledge graph is improved. Disclosure of Invention The invention provides an intelligent decision and recommendation method based on an enterprise knowledge graph, which comprises the following steps: s1, collecting multi-source heterogeneous data of an enterprise, performing cleaning, standardization and semantic labeling processing on the data, and extracting related entities, attributes and association relations among the entities of the enterprise to construct an enterprise knowledge graph, wherein the related entities of the enterprise comprise clients, products, suppliers, financial accounts, staff and business processes; S2, receiving a demand instruction input by a user, and analyzing semantic information of the demand instruction through a natural language processing technology by combining with service roles and historical behavior data of the user to determine a decision demand type or a recommendation demand type of the user, wherein the decision demand type comprises a market expansion decision, a supply chain optimization decision, a risk early warning decision and a financial planning decision; S3, extracting corresponding entity, attribute and association relation data from the enterprise knowledge graph according to the decision requirement type or the recommendation requirement type determined in the S2, and inputting the data into a preset intelligent reasoning model for reasoning analysis; S4, screening, sorting and optimizing the candidate decision suggestions or candidate recommended contents generated in the S3 by combining business rules, historical decision data and user feedback data of the enterprise to obtain final accurate decision suggestions or personalized recommended contents; And S5, presenting the accurate decision suggestion or personalized recommendation content obtained in the step S4 to a user in a visual mode, recording feedback information of the user on the result, and using the feedback information to optimize the enterprise knowledge graph and the intelligent reasoning model. Compared with the prior art, the invention has the beneficial effects that: In the scheme of the application: According to the application, through constructing the enterprise knowledge graph covering multidimensional entities and association relations of enterprise clients, products, supply chains, finances, staff and the like, the heterogeneous data scattered in each business system in the enterprise are semantically integrated and associated to form a unified knowledge resource pool, the process can comprehensively and accurately reflect the internal association among the entities in the enterprise operation process, a complete and reliable data basis is provided for subsequent decision analysis and recommendation, a multi-scenario intelligent reasoning model is constructed based on the enterprise knowledge graph, the abundant semantic association information in the knowledge graph can be utilized to carry out deep mining and reasoning analysis on various data in the enterprise operation process, compared with the traditional decision mode relying on artificial experience, the method can reduce the interference of subjective factors, rapidly identify potential rules and risks in the data, and provide prospective and scientific decision suggestions for an enterprise management layer. Drawings FIG. 1 is a flow chart of an intelligent decision and recommendation method based on enterprise knowledge graph provided by the application; FIG. 2 is a flow chart of the method for constructing the enterprise knowledge graph; fig. 3 is a specific construction flow of the intelligent reasoning model based on the graph neural network algorithm. Detailed Description In order that those skilled in the art w