A knowledge graph-based semantic association and logical rule inference method

Abstract

The application relates to the technical field of knowledge graphs, and discloses a reasoning method for semantic association and logical rules based on a knowledge graph, which comprises the following steps: acquiring a first data set across functional departments, and constructing a cross-department knowledge graph; extracting index nodes from the cross-department knowledge graph, identifying semantic conflict indexes, generating an index semantic conflict list, and constructing a cross-department knowledge graph after semantic mapping based on the index semantic conflict list; acquiring a second data set across functional departments, obtaining a fused cross-department knowledge graph and a semantic association degree matrix based on the second data set and the cross-department knowledge graph after semantic mapping; dynamically adjusting the weights of each node and edge of the fused cross-department knowledge graph according to the semantic association degree matrix, and generating a cross-department collaborative knowledge graph; and generating cross-department collaborative decision-making suggestions based on the cross-department collaborative knowledge graph; and the application significantly improves the efficiency and quality of cross-department collaborative decision-making.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of knowledge graph, more particularly, the present application relates to a reasoning method based on semantic association and logical rules of knowledge graph. BACKGROUND

[0002] In recent years, with the rapid development of industrial internet and artificial intelligence, intelligent decision support systems based on knowledge graph have been widely applied in the industrial field. In particular, in the field of intelligent factory manufacturing, cross-departmental collaborative optimization and decision-making face great challenges. Some existing research works focus on the construction of knowledge graph, trying to solve the problem of cross-departmental knowledge fusion and application.

[0003] Chinese patent application with publication number CN114610898A proposes a supply chain operation knowledge graph construction method and system. The method first acquires data in the supply chain operation process, and performs knowledge extraction and fusion to construct a supply chain operation knowledge graph. Based on the knowledge graph, knowledge semantic retrieval and knowledge precise pushing services are realized. The method constructs a knowledge graph that integrates multiple data sources, supports fast query and semantic retrieval of data in different fields, and to some extent meets the user's demand for cross-stage knowledge association retrieval. However, this method does not address the semantic conflict problem that exists in the cross-departmental knowledge fusion process, which may lead to bias in the reasoning and decision-making process. In addition, this method lacks a dynamic weight adjustment mechanism, making it difficult to adapt to complex and changing cross-departmental collaborative optimization scenarios.

[0004] Chinese patent CN118069856A discloses a knowledge graph construction method and application method. The method acquires multi-source heterogeneous data in the target field, and constructs a knowledge graph containing entities and relationships through data cleaning, integration, semantic recognition and other steps. This method introduces a semantic recognition model that can extract entities and relationships from unstructured data, expanding the information sources of the knowledge graph. However, this method mainly focuses on the construction of knowledge graph in a single field, lacking a semantic mapping and conflict resolution mechanism for cross-departmental knowledge. In practical applications, different departmental knowledge graphs may have semantic differences and conflicts, affecting the accuracy of collaborative reasoning.

[0005] In summary, the existing knowledge graph construction methods lack cross-departmental semantic mapping and alignment mechanisms, making it difficult to effectively solve the semantic conflict problem between different departments, affecting the efficiency and accuracy of cross-departmental collaborative decision-making. The reasoning and decision-making process does not fully consider the semantic association between different departments, failing to achieve cross-departmental collaborative optimization. SUMMARY

[0006] In order to overcome the above-mentioned defects of the prior art, the present application provides a reasoning method based on semantic association and logical rules of knowledge graph.

[0007] To achieve the above object, the present application provides the following technical solutions:

[0008] A reasoning method based on semantic association and logical rules of a knowledge graph, comprising:

[0009] A first data set across functional departments is obtained, a structured data table is constructed according to the first data set, and a cross-department knowledge graph is constructed based on the structured data table;

[0010] An index node is extracted from the cross-department knowledge graph, a semantic conflict index is identified, and an index semantic conflict list is generated; based on the index semantic conflict list, a cross-department knowledge graph after semantic mapping is constructed; a second data set across functional departments is obtained, and a fused cross-department knowledge graph and a semantic association degree matrix are obtained based on the second data set and the cross-department knowledge graph after semantic mapping; the weights of each node and edge of the fused cross-department knowledge graph are extracted, the weights of each node and edge of the fused cross-department knowledge graph are dynamically adjusted according to the semantic association degree matrix, and a cross-department collaborative knowledge graph is generated;

[0011] Based on the cross-department collaborative knowledge graph, a cross-department collaborative decision-making suggestion is generated.

[0012] Further, the first data set includes index definition documents, business process description documents, and historical decision records across functional departments; the functional departments include a production department, a quality management department, and a supply chain department;

[0013] The structured data table is constructed according to the first data set, comprising:

[0014] An index name A, a first relationship type, and an index name B are extracted from the index definition document to form an index information triple; wherein the index name A is the subject of the index information triple, the first relationship type is the predicate of the index information triple, and the index name B is the object of the index information triple;

[0015] A process node A', a second relationship type, and a process node B' are extracted from the business process description document to form a process triple; wherein the process node A' is the subject of the process triple, the second relationship type is the predicate of the process triple, and the process node B' is the object of the process triple;

[0016] A decision event, a third relationship type, a decision problem or scheme are extracted from the historical decision record to form a decision information triple; wherein the decision event is the subject of the decision information triple, the third relationship type is the predicate of the decision information triple, and the decision problem or scheme is the object of the decision information triple;

[0017] Based on the index information triplets, the process triplets and the decision information triplets, a structured data table is constructed, which includes triplet subjects, triplet predicates and triplet objects.

[0018] Further, the structured data table is used to construct a cross-department knowledge graph, which includes:

[0019] The triplet subjects and the triplet objects are mapped to nodes in the cross-department knowledge graph, and the triplet predicates are mapped to directed edges between the nodes; the nodes in the cross-department knowledge graph include index nodes, process nodes and decision nodes;

[0020] Attribute information in the first data set is extracted, and attributes are added to the nodes according to the attribute information to generate node attributes; the node attributes include index node attribute information, process node attribute information and decision node attribute information; the index node attribute information includes index name, index definition and responsible department;

[0021] The nodes in the cross-department knowledge graph are semantically labeled, and initial node weights are added to the nodes;

[0022] The directed edges in the cross-department knowledge graph are semantically labeled, and initial edge weights are added to the directed edges; and the cross-department knowledge graph is optimized.

[0023] Further, the optimization of the cross-department knowledge graph includes:

[0024] The semantic similarity SI between the nodes in the cross-department knowledge graph is calculated, and the nodes with a semantic similarity SI greater than a preset first similarity threshold θ1 are defined as synonymous concept nodes;

[0025] Synonymous relationship edges are added between the synonymous concept nodes;

[0026] Superordinate and subordinate concept nodes in the cross-department knowledge graph are identified, and belonging relationship edges are added between the superordinate and subordinate concept nodes.

[0027] Further, the generation of the index semantic conflict list includes:

[0028] All index nodes in the cross-department knowledge graph are searched, and index node attribute information is extracted;

[0029] The index names and the index definitions in the index node attribute information are vectorized to form index vectors of the indexes corresponding to the index nodes;

[0030] The similarity S1 between the index vectors of the indexes is calculated, a second similarity threshold θ2 is set, and for index pairs with a similarity S1 higher than θ2 but different responsible departments, semantic conflict indexes are marked and added to the index semantic conflict list.

[0031] Further, the constructed semantic mapping cross-department knowledge graph comprises:

[0032] For each index pair in the index semantic conflict list, a preset semantic mapping rule library is queried to match a mapping rule; the semantic mapping rule is in the form of IF-THEN;

[0033] According to the matched mapping rule, a mapping relationship edge and a mapping attribute between the two semantic conflict indexes in the index pair are added to form a cross-department semantic mapping layer;

[0034] The cross-department semantic mapping layer is integrated into the cross-department knowledge graph to form a semantic mapping cross-department knowledge graph.

[0035] Further, the second data set comprises real-time business parameters and real-time environmental data of cross-functional departments;

[0036] The obtained fused cross-department knowledge graph and semantic correlation degree matrix comprise:

[0037] According to the real-time business parameters of the cross-functional departments, a real-time business feature vector is formed;

[0038] According to the real-time environmental data of the cross-functional departments, a real-time environmental feature vector is formed;

[0039] The real-time business feature vector and the real-time environmental feature vector are fused in the semantic mapping cross-department knowledge graph to form a fused cross-department knowledge graph;

[0040] Based on the fused cross-department knowledge graph, the semantic correlation degree between departments is calculated to generate an N×N semantic correlation degree matrix, N being the number of departments.

[0041] Further, the generation of the cross-department collaborative knowledge graph comprises:

[0042] The semantic correlation degree between departments is extracted from the semantic correlation degree matrix, and for department pairs with a semantic correlation degree higher than a preset correlation degree threshold θ3, a first weight adjustment coefficient k1 is used to increase the weight of related nodes and edges, the first weight adjustment coefficient k1 being in a positive correlation with the semantic correlation degree;

[0043] For department pairs with a semantic correlation degree less than or equal to the preset correlation degree threshold θ3, a second weight adjustment coefficient k2 is used to reduce the weight of related nodes and edges, the second weight adjustment coefficient k2 being in an inverse correlation with the semantic correlation degree.

[0044] Further, the generation of the cross-department collaborative knowledge graph based on the cross-department collaborative knowledge graph comprises:

[0045] Collect operation time sequence data of nodes in the cross-department collaborative knowledge graph, and obtain time sequence dependent features between nodes according to the operation time sequence data and a pre-constructed time sequence dependent feature model;

[0046] According to the nodes and edges of the cross-department collaborative knowledge graph and a pre-constructed semantic interaction feature model, semantic interaction features between nodes are obtained.

[0047] The time sequence dependent features and the semantic interaction features are fused to form a fused feature vector.

[0048] A logic rule library is constructed, and a logic rule reasoning engine is established.

[0049] The fused feature vector is input into the logic rule reasoning engine for reasoning, and cross-department collaborative decision suggestions are generated.

[0050] A reasoning system based on semantic association and logic rules of a knowledge graph, which is used to implement the reasoning method based on semantic association and logic rules of a knowledge graph, comprises:

[0051] A graph construction module is configured to obtain a first data set of cross-functional departments, construct a structured data table according to the first data set, and construct a cross-department knowledge graph based on the structured data table.

[0052] A graph optimization module is configured to extract index nodes from the cross-department knowledge graph, identify semantic conflict indexes, and generate an index semantic conflict list; based on the index semantic conflict list, a cross-department knowledge graph after semantic mapping is constructed; a second data set of cross-functional departments is obtained, and a fused cross-department knowledge graph and a semantic association degree matrix are obtained based on the second data set and the cross-department knowledge graph after semantic mapping; the weights of each node and edge of the fused cross-department knowledge graph are extracted, the weights of each node and edge of the fused cross-department knowledge graph are dynamically adjusted according to the semantic association degree matrix, and a cross-department collaborative knowledge graph is generated.

[0053] A decision reasoning module is configured to generate cross-department collaborative decision suggestions based on the cross-department collaborative knowledge graph.

[0054] Compared with the prior art, the present application has the following advantages:

[0055] The application realizes comprehensive integration and correlation analysis of data by constructing a cross-department knowledge graph, fusing data of key business departments such as production, quality and supply chain, and breaking down data barriers between departments to provide a unified semantic basis for cross-department collaboration. The semantic mapping is constructed by identifying semantic conflicts of indicators, solving the problem of inconsistent definition and understanding of the same business indicators by different departments, eliminating semantic ambiguity, and ensuring the consistency and comparability of cross-department data to provide a reliable basis for collaborative decision-making. Real-time business and environmental data are introduced to dynamically optimize the knowledge graph, so that the knowledge graph can reflect the business operation status in real time, capture business changes, and quantify the correlation between different departments to timely reveal collaborative risks and optimization opportunities. The time-dependent features and semantic interaction features are fused to form cross-department collaborative decision-making suggestions by using logical reasoning. The field knowledge and data value are fully tapped to make the decision-making suggestions more comprehensive and reasonable, and the knowledge graph learning and reasoning ability is given to realize the closed loop from data to decision-making. Compared with the traditional cross-department collaboration method, the method provided by the application can significantly improve the collaboration efficiency, timely discover and solve the conflicts between departments, optimize resource allocation, and shorten the decision-making cycle. The collaboration quality is also guaranteed, and the decision-making is more scientific, transparent and interpretable, avoiding subjective bias and blind spots. The innovative application of knowledge graph and logical reasoning in the application provides a new idea and method for efficient collaboration in complex organizational environments, and has a wide application prospect. It is not only suitable for internal collaboration of enterprises, but also can be extended to inter-enterprise and even industry chain collaboration to help improve organizational resilience and competitiveness. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, a brief introduction will be given below to the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0057] Figure 1 The principle flowchart of the semantic association and logical rule reasoning method based on the knowledge graph in the present application;

[0058] Figure 2 The method flowchart for constructing a structured data table in the semantic association and logical rule reasoning method based on the knowledge graph in the present application;

[0059] Figure 3 The method flowchart for constructing a cross-department knowledge graph in the semantic association and logical rule reasoning method based on the knowledge graph in the present application;

[0060] Figure 4A method flowchart for optimizing a cross-department knowledge graph in a semantic association and logical rule reasoning method based on a knowledge graph according to the present application is shown in FIG. 1.

[0061] Figure 5 A method flowchart for generating an index semantic conflict list in a semantic association and logical rule reasoning method based on a knowledge graph according to the present application is shown in FIG. 2.

[0062] Figure 6 A method flowchart for constructing a cross-department knowledge graph after semantic mapping in a semantic association and logical rule reasoning method based on a knowledge graph according to the present application is shown in FIG. 3.

[0063] Figure 7 A method flowchart for obtaining a fused cross-department knowledge graph and semantic association degree matrix in a semantic association and logical rule reasoning method based on a knowledge graph according to the present application is shown in FIG. 4.

[0064] Figure 8 A functional module diagram of a semantic association and logical rule reasoning system based on a knowledge graph according to the present application is shown in FIG. 5. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0066] Embodiment 1

[0067] Referring to FIG. 1, the present embodiment provides a semantic association and logical rule reasoning method based on a knowledge graph, which comprises: Figure 1

[0068] Step S1000, a first data set of cross-functional departments is obtained, a structured data table is constructed according to the first data set, and a cross-department knowledge graph is constructed based on the structured data table;

[0069] Further, step S1000 comprises:

[0070] Step S1100, a first data set of cross-functional departments is obtained, the first data set comprising index definition documents, business process description documents and historical decision records of cross-functional departments; the cross-functional departments comprising a production department, a quality management department and a supply chain department;

[0071] ​Specifically, cross-functional departments refer to departments within an organization that perform different functions but are interconnected and influence each other in business operations. For example, the production department is responsible for product manufacturing, the quality management department is responsible for quality control, and the supply chain department is responsible for raw material procurement and finished product delivery. They work closely together in production and operation. Obtaining data from these departments requires submitting data applications to relevant departments.

[0072] A data application is submitted to the production department to obtain definition documents for production-related indicators such as output and pass rate, business process description documents such as production planning process and process flow, and historical production decision meeting minutes. Output refers to the number of qualified products produced within a certain period, which is an important indicator of production capacity. The pass rate refers to the percentage of qualified products produced within a certain period of total output, reflecting the product quality level. The production planning process specifies the steps and requirements for developing production plans, and the process flow describes the operation specifications for each production process in detail. Historical production decision meeting records major decisions made in production management, such as capacity adjustment and process improvement. By obtaining these data, a comprehensive understanding of production operation and management measures can be achieved. This helps identify risks and bottlenecks in production and optimize resource allocation.

[0073] A data application is submitted to the quality management department to obtain definition documents for quality-related indicators such as repair rate and customer complaint rate, quality management process documents such as quality inspection process and quality audit process, and quality accident handling decision records. Repair rate refers to the percentage of products requiring repair in total output, reflecting quality problems in the production process. Customer complaint rate refers to the number of customer complaints per sales volume, reflecting product use quality. The quality inspection process specifies the quality inspection methods and standards for raw materials, semi-finished products, and finished products, and the quality audit process is used to identify non-conformities in the quality management system. Quality accident handling records measures taken to address major quality issues, such as product recall and compensation. Obtaining data from the quality management department can help identify key factors affecting product quality and improve quality control mechanisms.

[0074] A data application is submitted to the supply chain department to obtain definition documents for delivery-related indicators such as delivery on-time rate and inventory turnover rate, supply chain management process documents such as procurement process and logistics management process, and supplier selection decision records. Delivery on-time rate refers to the percentage of orders delivered on time out of total orders, and inventory turnover rate refers to the ratio of product sales cost to average inventory balance within a certain period, reflecting inventory management efficiency. The procurement process includes management provisions for supplier selection, procurement planning, and order execution, while the logistics management process involves transportation, warehousing, and distribution operations. Supplier selection records reflect factors considered in supplier management decisions, such as quality, cost, and delivery time. Supply chain data helps optimize procurement strategies and inventory control, and improves overall supply chain performance.

[0075] The obtained department data sets are subjected to integrity check, missing or inconsistent data items are identified, a data quality report is formed and sent to the relevant department for supplement and correction. Data missing can be caused by non-registration, omission, etc., and data inconsistency can be caused by understanding deviation or statistical caliber change. Timely identification and correction of data quality problems can provide a reliable basis for subsequent data application.

[0076] In step S1200, a structured data table is constructed according to the first data set.

[0077] Further, as shown in Figure 2 , step S1200 includes:

[0078] In step S1210, an index name A, a first relationship type and an index name B are extracted from the index definition document to form an index information triple; the index name A is the subject of the index information triple, the first relationship type is the predicate of the index information triple, and the index name B is the object of the index information triple; the first relationship type includes a calculation relationship, an influence relationship and a composition relationship.

[0079] In step S1220, a process node A', a second relationship type and a process node B' are extracted from the business process description document to form a process triple; the process node A' is the subject of the process triple, the second relationship type is the predicate of the process triple, and the process node B' is the object of the process triple; the second relationship type includes a time sequence relationship and a logical relationship.

[0080] In step S1230, a decision event, a third relationship type, a decision problem or a scheme are extracted from the historical decision record to form a decision information triple; the decision event is the subject of the decision information triple, the third relationship type is the predicate of the decision information triple, and the decision problem or the scheme is the object of the decision information triple; the third relationship type includes the relationship between the decision event and the decision problem, and the relationship between the decision event and the decision scheme.

[0081] In step S1240, a structured data table is constructed based on the index information triple, the process triple and the decision information triple, and the structured data table includes the triple subject, the triple predicate and the triple object.

[0082] Specifically, the index information triple represents that there is a certain semantic association between two indexes, such as one index can be calculated by another index, or one index affects the value of another index, or one index is composed of multiple subordinate indexes. Taking "< yield, calculated from, qualified product quantity >" as an example, the yield of this index can be calculated by the qualified product quantity, and there is a calculation relationship between the two. Building an index information triple is beneficial to depicting the semantic relationship between indexes and laying the foundation for building an index system. The index system is built on the index information triple, which quantitatively depicts the performance of the enterprise in production, quality, delivery, etc., and supports performance evaluation and improvement.

[0083] The business process description document is usually in the form of natural language text, describing the execution steps, sequence, etc. of business activities. Using natural language processing technology, through syntactic dependency analysis, process triplets can be extracted, including process node A', second relationship type, and process node B'. Here, the process node refers to the activity in the business process, and the second relationship type depicts the temporal dependency or logical relationship between activities. For example, "< workpiece cleaning, successor, workpiece assembly >", workpiece assembly is arranged after workpiece cleaning, and there is a time sequence between the two. For example, "< part processing, parallel, part detection >" describes that two activities can be performed simultaneously, and there is no dependency constraint. The process triple reflects the temporal logic of business activities and is an important basis for process optimization and monitoring. By building a process path network based on process triple, and combining with time parameters, bottleneck processes can be analyzed and resource allocation can be optimized. When the actual process execution deviates from the temporal logic set by the process triple, timely warning is helpful for process control.

[0084] The historical decision record carries the experience of the enterprise in dealing with production, quality, supply, etc. challenges, and contains valuable management wisdom. By extracting decision events, decision problems, decision schemes and their associations from the decision record, decision information triplets can be formed. For example, "< decision event 1, for, high rework rate >", it reveals that in "decision event 1", the problem of "high rework rate" needs to be solved. Further, "< decision event 1, adopt, increase quality inspection frequency >", it reveals that to deal with the problem of "high rework rate", the measure taken in "decision event 1" is "increasing quality inspection frequency". The decision information triple connects the decision scenario, the decision problem, and the solution, presenting a kind of normalized experience knowledge. The decision information triple can be used for decision knowledge retrieval, recommendation, and reuse, and can also be used as the basis for case reasoning. When facing similar decision scenarios, reference schemes can be quickly matched according to the decision information triple to assist decision makers in judgment.

[0085] The indicators, processes, and decision data scattered in different departments are extracted into uniform triplets, and a structured data table is constructed with subject, predicate, and object as fields. Each triplet forms a row record of the data table. Taking an indicator information triplet as an example, a triplet “<yield, calculated from, number of qualified products>” can form a row record of the data table, in which “yield” is the subject, “calculated from” is the predicate, and “number of qualified products” is the object. Based on the triplets, the structured data table is constructed, and the heterogeneity of business data among departments is eliminated in terms of data organization form, and the uniform structure of triplets subject-predicate-object is obtained. This uniformity is beneficial to subsequent indexing, querying, and integrated analysis of the fields of the data table. At the same time, the triplet structure has clear semantics and is easy for manual interpretation. Relying on the structured data table, the influence transmission among key indicators can be found, the optimization space in the process path can be explored, and the formation mechanism of the decision scheme can be traced. The triplet structure makes it possible to uniformly integrate cross-department business data into the analysis field and understand business synergy.

[0086] In step S1300, a cross-department knowledge graph is constructed based on the structured data table.

[0087] Further, as shown in Figure 3 step S1300 includes:

[0088] In step S1310, the triplet subject and the triplet object are mapped to nodes in the cross-department knowledge graph, and the triplet predicate is mapped to a directed edge between the nodes; the nodes in the cross-department knowledge graph include indicator nodes, process nodes, and decision nodes.

[0089] Specifically, constructing the cross-department knowledge graph is to convert the triplet information in the structured data table into a graph data structure. The subject and the object of the triplet are mapped to the nodes of the graph, and the predicate is mapped to the directed edge between the nodes. For example:

[0090] <yield, calculation formula, number of qualified products>

[0091] The subject “yield” and the object “number of qualified products” are both indicator nodes, and the predicate “calculation formula” is the relationship type between them. This indicates that in the knowledge graph, there is a semantic association of “calculation formula” between the “yield” node and the “number of qualified products” node. The indicator nodes represent the performance indicators that the enterprise is concerned about, and the semantic network between the indicators is established through the directed edge, which is conducive to understanding the indicator system.

[0092] <workpiece cleaning, successor, workpiece assembly>

[0093] <workpiece assembly, successor, quality detection>

[0094] Here, the subject and object are both process nodes, and the predicate is "successor", indicating the order of the process. In the knowledge graph, a path is formed from the "workpiece cleaning" node to the "workpiece assembly" node, and from the "workpiece assembly" node to the "quality inspection" node. The process nodes and their timing relationships reflect the execution logic of business activities and are an important basis for process optimization and monitoring.

[0095] <Decision Event 1, for, high return rate>

[0096] <Decision Event 1, adopt, increase quality inspection frequency>

[0097] These two triples share the subject "Decision Event 1" and correspond to the same decision node in the knowledge graph. There is a "for" relationship between the "Decision Event 1" node and the "high return rate" node, and an "adopt" relationship between the "Decision Event 1" node and the "increase quality inspection frequency" node. The decision node records the decision scenario and response measures, which helps to reuse decision knowledge.

[0098] Mapping structured data to a knowledge graph allows indicators, processes, and decision information scattered in different business areas to be linked in the form of graph data. The node types of the graph identify different business elements, and the direction of the edges reflects business semantics, providing a unified knowledge base for cross-departmental collaborative optimization. When building a knowledge graph, it is necessary to merge nodes with the same name. If different triple subjects or objects refer to the same business entity, they should be merged into the same node in the graph instead of creating multiple nodes. Merging nodes with the same name can improve the simplicity and connectivity of knowledge representation, reducing redundancy and ambiguity.

[0099] In step S1320, attribute information in the first data set is extracted, and attributes are added to the nodes according to the attribute information to generate node attributes; the node attributes include indicator node attribute information, process node attribute information, and decision node attribute information; the indicator node attribute information includes indicator name, indicator definition, and responsible department;

[0100] Specifically, while building the knowledge graph nodes, it is necessary to enrich the attribute information of the nodes to make them carry more semantic connotations. The attributes of different types of nodes have different focuses:

[0101] The attribute information of the indicator node includes: indicator name, indicator definition, responsible department, etc. The indicator name is the identification of the indicator, the indicator definition explains the meaning and calculation formula of the indicator, and the responsible department clarifies the assessment attribution of the indicator. For example, the attributes of the node "output" may be <Name, Output>, <Definition, The number of qualified products produced in a certain period>, and <Responsible Department, Production Department>. The addition of indicator attributes makes the node meaning clear and can answer questions such as "What is this indicator? How is it calculated? Who is responsible?"

[0102] The attribute information of the process node includes: process name, process description, department to which the process belongs, etc. The process name identifies a business activity, the process description explains the input, conversion, output, etc. of the activity, and the department to which the process belongs defines the division of responsibilities for process management. For example, the attributes of the node "workpiece assembly" can be <name, workpiece assembly>, <description, combine the workpiece parts that have been machined into a whole piece according to the drawing requirements>, and <department, production workshop>. The process attributes make the content of the process node more rich, and enhance the information basis for process analysis and improvement.

[0103] The attribute information of the decision node includes: decision event name, decision problem, decision scheme, responsible department, etc. The decision event name identifies a decision behavior, the decision problem explains the challenge faced, the decision scheme summarizes the response measures, and the responsible department defines the decision participants. For example, the attributes of the "decision event 1" node can be <event name, production meeting in the first quarter of 2022>, <decision problem, excessive rework rate>, <decision scheme, strengthen first inspection and adjust process parameters>, and <responsible department, quality department, production department>. The decision attributes provide more comprehensive context information, enhancing the relevance of decision tracing and review.

[0104] Node attributes are an important part of the knowledge graph, and are a semantic description of the content of the node. The more attributes a node has, the more knowledge it contains. Extracting attribute information from business data and organizing it uniformly in the knowledge graph allows scattered and unstructured business knowledge to be integrated and structured, thereby serving unified semantic association analysis and comprehensive application.

[0105] In step S1330, the nodes in the cross-department knowledge graph are labeled with semantic types, and initial node weights are added to the nodes.

[0106] Specifically, the nodes in the knowledge graph can belong to different semantic types. By labeling the type of a node, its role in the business semantic network can be clearly defined. Common semantic types include indicators, processes, decisions, etc. and custom types can also be defined according to business needs. During graph construction, the semantic type of each node needs to be judged and labeled. For example, the node "production" is labeled as <node type, indicator>, the node "workpiece assembly" is labeled as <node type, process step>, and the node "decision event 1" is labeled as <node type, decision>. Through semantic type labeling, the knowledge graph presents a global classification view, making knowledge organization more organized. Graph indexing and querying based on types will also be more efficient.

[0107] In addition to type labeling, initial weights also need to be added to the nodes. Node weights reflect the importance of the node, which can be based on the attribute values of the node, such as the weight of the indicator node, which can be derived from the weight of the indicator in performance evaluation, the weight of the process node, which can be derived from the proportion of working hours of the process step, and the weight of the decision node, which can be derived from the influence evaluation of the decision. The initial weight assignment can be based on objective data or a combination of expert experience and judgment. By weighting the nodes, the knowledge graph forms a distribution map of key knowledge, guiding the focus of analysis. Subsequently, the connection relationship between nodes can be used to iteratively optimize the weight values of the nodes using algorithms such as PageRank, to mine key nodes in the graph.

[0108] Step S1340, semantic type labeling is performed on the directed edges in the cross-department knowledge graph, and initial edge weights are added to the directed edges;

[0109] Specifically, the directed edges between nodes in the knowledge graph represent different relationship types, and semantic labeling of relationship types helps to clarify the semantic relationship between business elements. For example, there may be mathematical operation relationships such as <add> and <divide> between indicator nodes, there may be time sequence dependency relationships such as <before> and <successor> between process nodes, and there may be influence relationships such as <improve> and <restrict> between indicators and processes. By predefining these semantic relationship types and labeling the relationship types of each edge in the graph, business knowledge can be organized in a more flexible and expressive way. This provides a foundation for complex semantic association analysis, such as multi-hop relationship queries (such as "a <influence> b <influence> c") and pattern matching (such as "process 1 <before> process 2 <before> process 3").

[0110] At the same time of semantic type labeling, weights are also needed to be added to the directed edges, representing the relationship strength. The initial edge weights can be assigned according to the confidence, frequency, and expert score of the relationship, such as the data dependency frequency of two indicators and the business relevance of two process steps. Weighted edges enable the knowledge graph to represent the strength of business association, enhancing its application flexibility. For example, in the risk analysis scenario, high-weight node associations represent possible transmission paths of risk influence; in the knowledge recommendation scenario, high-weight edges correspond to more relevant node combinations. Subsequently, various connection paths between nodes can be used to evaluate the structured weight and semantic weight of the edges, and to dynamically optimize the weight values of the edges.

[0111] Step S1350, the cross-department knowledge graph is optimized.

[0112] Further, as shown in Figure 4 Step S1350 includes:

[0113] Step S1351: Calculate the semantic similarity SI between nodes in the cross-departmental knowledge graph, and define nodes with semantic similarity SI greater than the preset first similarity threshold θ1 as synonymous concept nodes.

[0114] The calculation of semantic similarity SI between nodes in a cross-departmental knowledge graph includes:

[0115]

[0116] in:

[0117] Represents a node and nodes The semantic similarity between them, with values ​​ranging from 1 to 2. A larger value indicates that the nodes are more semantically similar.

[0118] and Representing nodes respectively and nodes The embedded vector representation can be learned through a knowledge graph embedding model (such as TransE).

[0119] The cosine similarity between two embedding vectors is defined as:

[0120]

[0121] and Representing nodes respectively and nodes The set of attributes.

[0122] The Jaccard similarity coefficient between two attribute sets is defined as follows:

[0123]

[0124] Represents a node and Between the first Characteristic function of relation It is used to capture the structural information of nodes in the graph. This represents the number of features relating nodes.

[0125] In cross-departmental knowledge graphs, there may be various structural relationships between different nodes, such as:

[0126] Shortest path length: The shortest path between nodes reflects their distance in the graph;

[0127] Common neighbors: The number of direct neighbors shared by nodes can indicate their similarity;

[0128] Path diversity: The number of paths connecting nodes through different types of edges;

[0129] Information flow weight: The weight or influence of information passing between nodes through a specific path.

[0130] Each relationship can be defined as a feature function , where identifies the number of features.

[0131] Role:

[0132] Capture the structural information of nodes: The size determines the types and granularity of relationship features, which can describe the connection between nodes from multiple angles.

[0133] Trade-off between complexity and accuracy: Increasing can improve the accuracy of similarity calculation, but will increase the computational complexity. Therefore, in practical applications, we need to choose the appropriate number of features according to the scene.

[0134] is the weight coefficient of the th relationship feature.

[0135] are the weight factors of embedding similarity, attribute similarity and relationship features, respectively, satisfying Their values can be set according to experience or optimized by parameter search methods (such as grid search).

[0136] This formula considers the semantic representation of nodes, attribute similarity and structural relationship in knowledge graph, and more comprehensively describes the semantic similarity between nodes. The setting of parameters needs to balance the calculation efficiency and effect. With the improvement of node embedding vector similarity, the increase of attribute overlap and the enhancement of relationship features, the semantic similarity SI between nodes will also be improved accordingly. This similarity formula can help identify semantically equivalent or highly related nodes in cross-department knowledge graphs, providing a basis for subsequent mapping relationships, eliminating redundancy, and integrating knowledge, and improving the semantic consistency and compactness of knowledge graphs.

[0137] Step S1352, add synonymous relationship edges between synonymous concept nodes;

[0138] Step S1353, identify the upper and lower concept nodes in the cross-department knowledge graph, and add belonging relationship edges between the upper and lower concept nodes.

[0139] Specifically, the semantic similarity is calculated to identify the synonymous concept nodes in the knowledge graph. The semantic similarity comprehensively considers the multi-faceted features of the nodes, including:

[0140] (1) The embedding vector representation of the node. Using a knowledge graph embedding model such as TransE, TransR, etc., a low-dimensional dense vector representation of each node can be learned. The similarity of the vectors (such as cosine similarity) can depict the semantic similarity of the nodes in the embedding space. For example, the two nodes of "product" and "commodity" have similar neighbor nodes and connection relationships in the graph, so the learned embedding vectors are also relatively close.

[0141] (2) The attribute features of the node. Directly compare the attribute sets of two nodes. If the attribute overlap degree is high (such as Jaccard similarity coefficient), it indicates that the semantic similarity of the two nodes is strong. For example, the two indicators of "shipment quantity" and "sales quantity" may have highly consistent attributes such as definition, calculation formula, etc.

[0142] (3) The structural features of the node in the graph. The structure relationship of the node in the knowledge graph can also represent its semantics. For example, two nodes with multiple common neighbor nodes have a high degree of semantic association. For example, if there are many connection paths between two nodes (such as "A produces B, B transports C, and A sells C"), their relevance can be inferred. "Webpage" and "article" may play similar structural roles in the knowledge graph.

[0143] When calculating the semantic similarity, the above factors need to be weighed comprehensively. A weighted average method can be used to obtain the final similarity score of the node pair. In order to automatically identify synonymous concepts, a threshold needs to be set. Node pairs above the threshold can be defined as synonymous concepts. The threshold setting needs to balance the accuracy and recall rate, which can be calibrated through experimental statistics and expert evaluation, etc.

[0144] Identifying synonymous concepts in the knowledge graph can merge the expression differences of business concepts and improve the standardization of knowledge organization. For example, "user" and "customer" may have different expressions in different business departments, but they refer to the same concept in semantics and need to be unified. Building a cross-department knowledge graph, especially when dealing with the name ambiguity problem of cross-referenced concepts between departments, the semantic similarity-based synonymous concept alignment is an effective method.

[0145] After identifying the synonymous concept nodes in step S1351, it is necessary to explicitly add the "synonymous relationship" edge between these nodes to form a special edge type. This explicit modeling helps the association analysis and query of synonymous concepts, so that the knowledge graph can conveniently answer questions such as "which concepts are synonymous". Integrating synonymous relationships into the structure of the knowledge graph in the form of edges makes it a knowledge element that can be reasoned. For example, through the connection of synonymous relationship edges, the nodes "article-synonymous->webpage", "webpage-synonymous->document", and "document-synonymous->material" form a "concept family", representing their semantic relevance.

[0146] The unified representation of synonymous concepts facilitates the simplification of the knowledge graph and reduces potential redundancy and inconsistency. By adding synonymous relationship edges, "equivalence classes" of concepts can be flexibly established without disrupting the original graph structure, preserving the diversity of concept representation while achieving logical unification. Synonymous relationship edges can assist knowledge reasoning, such as extended retrieval based on synonymous concepts (e.g., when searching for "article", "webpage", "document", etc. that are synonymous with it), making the knowledge graph better adapt to users' diverse language usage habits.

[0147] In addition to adding synonymous edges, redundant nodes need to be merged during knowledge graph optimization. For nodes with very high semantic similarity (e.g., similarity equal to 1), from the perspective of simplifying the graph, only one node can be retained, and its neighbor edges can be merged to simplify knowledge representation. At the same time, the "synonymous concept" attribute needs to be added to the merged node to list all synonymous concept names, ensuring that the diversity of concept representation information is not lost. For example, if "product" and "commodity" are identified as completely synonymous, only the "product" node can be retained, and the edges pointing to "commodity" can be connected to "product", and the <synonymous concept, "commodity"> attribute can be added to the "product" node.

[0148] When merging synonymous concept nodes, weight updates need to be considered. The weights of synonymous concept nodes can be accumulated to form the new weight of the merged node. The merging of weights allows the importance of synonymous concept nodes to be preserved in the simplified graph. For example, after merging the "user" and "customer" nodes, the weight of the new node is the sum of the weights of the two nodes, reflecting the overall weight of the "user / customer" concept in the entire graph.

[0149] Hyponym and hypernym concepts reflect the hierarchical semantic relationships of concepts in the knowledge graph. By identifying the hyponym and hypernym associations of nodes, a concept hierarchy can be constructed, which can improve the systematicness of knowledge organization and enhance knowledge reasoning capabilities. For example, the two nodes "yield" and "production indicator" represent an upper-lower relationship in concept, which is "yield is a kind of production indicator".

[0150] The identification of hyponym and hypernym concepts can comprehensively utilize the following strategies:

[0151] (1) Utilize ontology knowledge base. Map the hierarchical relationship solidified in the domain ontology knowledge base to the corresponding nodes of the knowledge graph, and obtain a batch of hierarchical relationship seeds in advance.

[0152] (2) Graph mining. Analyze the structural patterns of the knowledge graph. If there are many ISA (“is a”) type edges between two nodes, they tend to be hierarchical relationships, such as “output-ISA->production indicator” “qualified rate-ISA->production indicator”. The symbol “->” represents a “relationship” or “pointing”.

[0153] (3) Distributed semantics. Compare the word embedding vectors of the corresponding words of two nodes. If the upper word embedding and the lower word embedding exist a certain vector difference, such as upper word embedding-lower word embedding ≈ constant vector, then the hierarchical relationship can be inferred, such as vec(“production indicator”)-vec(“output”) ≈ vec(“quality indicator”)-vec(“rework rate”).

[0154] (4) Pattern matching. Match the hierarchical relationship expression templates such as “E is a F” and “F includes E” in large-scale text corpus, and count the co-occurrence frequency to discover the hierarchical relationship of concepts.

[0155] By comprehensively using multiple strategies, the hierarchical semantic relationship contained in the knowledge graph can be more comprehensively mined. To reflect the hierarchical organization of concepts, “belongs to” relationship edges need to be added to the identified hierarchical concept nodes to form a directed semantic hierarchy of concepts. The “belongs to” edge points from the lower concept to the upper concept, expressing the semantics of “lower belongs to upper” and “upper contains lower”. For example, “output-belong to->production indicator” “production indicator-belong to->performance indicator”, forming the hierarchical semantic chain of “output < production indicator < performance indicator”.

[0156] The hierarchical concept hierarchy formed based on the “belongs to” relationship edge enables the knowledge graph to have the ability of concept abstraction and concept refinement. Through the upper concept, abstraction can be achieved, such as “output” abstracted as “production indicator”; through the lower concept, specific can be achieved, such as “production indicator” refined as “output”. In query and reasoning, the semantic extension and connotation ability of the concept makes the knowledge graph more intelligent and closer to human understanding. For example, when asking “How did the production indicator perform in the first quarter of 2022?”, the knowledge graph can aggregate the statistical data of lower concepts such as “output” and “qualified rate” through the “belongs to” edge to form the overall performance of “production indicator”.

[0157] The weights of the upper and lower concepts need to be reconsidered on the basis of the edges. From the semantic point of view, the "belongs to" edge expresses a "belongs to" relationship, and the weight of the upper concept should be higher than the sum of the weights of the direct lower concepts. Therefore, the weights of the upper and lower nodes can be updated by using an additive model. For example, if node E "belongs to" node F, the weight of F is equal to the sum of the original weight of F and the weight of E. This weight propagation based on the "belongs to" edge can objectively reflect the cumulative importance of the concept in the semantic network. For example, the sum of the weights of "output" and "qualified rate" is transferred to the upper concept "production indicator", so that the overall weight of "production indicator" exceeds that of any lower concept. The weight distribution in the concept hierarchy helps to prioritize knowledge reasoning, such as starting with the upper concept with a high weight to analyze the problem.

[0158] In summary, the three sub-steps of step S1350 optimize the knowledge graph from different aspects, enriching the semantic association and organization of the knowledge graph. Synonymous relationships reveal the equivalence of concepts, simplifying the graph; hierarchical relationships reveal the hierarchy of concepts, systematically organizing them; and the attributes and weights of nodes and edges further characterize the semantics of the elements. The integration of the three optimization aspects makes the cross-department knowledge graph a highly semantic business knowledge representation, laying a solid foundation for intelligent analysis and decision-making.

[0159] Step S2000, extracting indicator nodes from the cross-department knowledge graph, identifying semantic conflict indicators, generating an indicator semantic conflict list; based on the indicator semantic conflict list, constructing a semantic-mapped cross-department knowledge graph; obtaining a second data set of cross-functional departments, based on the second data set and the semantic-mapped cross-department knowledge graph, obtaining a fused cross-department knowledge graph and a semantic correlation matrix; extracting the weights of each node and edge of the fused cross-department knowledge graph, dynamically adjusting the weights of each node and edge of the fused cross-department knowledge graph according to the semantic correlation matrix, generating a cross-department collaborative knowledge graph;

[0160] Further, step S2000 includes:

[0161] Step S2100, extracting indicator nodes from the cross-department knowledge graph, identifying semantic conflict indicators through semantic similarity calculation, and generating an indicator semantic conflict list;

[0162] Further, as shown in Figure 5 Step S2100 includes:

[0163] Step S2110, retrieving all indicator nodes in the cross-department knowledge graph and extracting attribute information of the indicator nodes;

[0164] Step S2120, vectorizing the indicator names and definitions in the attribute information of the indicator nodes to form the indicator vectors corresponding to the indicators of the indicator nodes;

[0165] Step S2130, calculate the similarity S1 between the index vectors of each index, set a second similarity threshold θ2, for the index pair with similarity S1 higher than θ2 but different responsibility departments, mark as semantic conflict index, and add to the index semantic conflict list.

[0166] Specifically, the index nodes in the cross-department knowledge graph carry the performance index information used by each department, including index name, index definition, responsible department and other attributes. The index name embodies the popular expression of the index, the index definition explains the connotation of the index from the business perspective, and the responsible department clarifies the object of the index. These attribute information is an important basis for carrying out index semantic comparison and identifying semantic conflicts. Therefore, the first step is to traverse the graph to find all nodes of type "index" and extract the attribute information of the nodes. This process can be achieved by using the query language of the graph database. Taking the graph database Neo4j as an example, Cypher statements can be used for retrieval and extraction. Cypher is a declarative graph query language with simple structure similar to SQL.

[0167] Quick retrieval and extraction of index node information in the knowledge graph through the graph database query fully utilizes the advantage of the structured representation of business knowledge by the graph. The node-edge-attribute model of the graph structure intuitively organizes the index information scattered in different business fields, and the Cypher query language expresses the retrieval logic in the form of a graph, without the need to split complex connection operations, greatly simplifying the extraction process of cross-department index information. At the same time, the graph database also supports rich indexing for nodes, edges and attributes, thereby accelerating the query response of large-scale graph.

[0168] Transforming unstructured text data into numerical form that computers can understand and process is an important foundation of natural language processing. Word vector is a widely used text representation learning method that maps words to a multidimensional real number space to form a word embedding. In the word vector space, semantically similar words are closer in geometry. Representing indicator information as a word vector can quantitatively characterize the semantic features of the indicator name and definition, and then realize the comparison of indicator semantics through vector operations. Google's word vector model (Word2Vec), Stanford's global vector model (GloVe), etc. can be used to vectorize the indicator names and definitions in the indicator information list. Vectorizing indicator information and embedding unstructured indicator text into a structured vector space is a key step in realizing semantic correlation analysis of indicators. By converting complex language expressions into standardized mathematical forms through word vectors, semantic similarity can be measured by vector distance, and semantic operations can be realized through vector algebra. At the same time, pre-trained word vector models trained on massive text corpora capture a wide range of semantic information, making the semantic representation of indicator vectors more comprehensive and accurate.

[0169] In step S2120, the attribute information of each indicator node, such as the name and definition, has been vectorized to form corresponding indicator vectors. The indicator vectors depict the semantic features of the indicators in multidimensional real numbers. Intuitively, in the vector space, the closer the geometric distance between two indicator vectors, the more similar the semantics of the indicators they represent. Therefore, by calculating the similarity between indicator vectors, the semantic proximity of different indicators can be determined.

[0170] By constructing a cross-department knowledge graph and applying word vectors, similarity thresholds and other technical means based on it, potential semantic conflict indicators within the enterprise are automatically discovered. These conflicts reflect the differences between different departments in data standards and are an important hidden danger affecting data quality and data value. Traditional manual sorting is difficult to identify such problems comprehensively, and the knowledge graph combined with semantic analysis technology provides an intelligent and efficient solution. For the identified semantic conflict indicators, a problem list can be formed, and data governance can be carried out to eliminate differences and ultimately form a unified enterprise-level indicator standard through communication and coordination to unify the standard and standardize the definition. This is of great significance to improving data application value and supporting scientific decision-making.

[0171] The calculation of the similarity S1 between the indicator vectors of each indicator includes:

[0172]

[0173] Wherein:

[0174] representation of indicators and indicators The similarity between the indicator vectors, with a value range of The greater the value, the more similar the semantics of the indicators.

[0175] and represent the vector representation of indicators and indicators on the first semantic dimension. Here we assume that the indicator vectors consist of semantic dimensions.

[0176] represents the weight of the first semantic dimension, reflecting the importance of this dimension to the similarity calculation. It can be learned through expert knowledge or data-driven methods.

[0177] represents the cosine similarity between the two vectors on the first dimension, used to measure the consistency of the vector direction.

[0178] is a regularization parameter that controls the influence of vector length difference on similarity. The greater the value, the greater the punishment of vector length difference. It can be set according to experience or optimized through parameter search.

[0179] represents the Euclidean distance squared between indicators and indicators on the first semantic dimension, measuring the numerical closeness of the vectors.

[0180] represents the Jaccard similarity coefficient between the attribute sets and indicators , the calculation formula is:

[0181]

[0182] It measures the similarity of two indicators at the attribute level.

[0183] is a control parameter of the steepness of the Sigmoid function, is a control parameter of the center position of the Sigmoid function, and​​ The contribution of attribute similarity to overall similarity is jointly determined. and It can be set according to business understanding or obtained through parameter learning.

[0184] The formula comprehensively considers the direction similarity, length difference and attribute matching degree of the index vector, and fully describes the semantic correlation between indexes.

[0185] Introducing dimension weight , the influence of different semantic dimensions on similarity calculation can be flexibly adjusted.

[0186] Using exponential function Smooth the vector length difference, reduce the influence of outliers.

[0187] Using Sigmoid function Normalize attribute similarity and control its contribution to overall similarity through parameters and .

[0188] When the vector direction of two indexes is more consistent, the length difference is smaller, and the attribute matching degree is higher, their similarity will be greater. Conversely, the similarity will decrease.

[0189] By adjusting the parameters , the similarity calculation formula can be optimized to better adapt to specific fields and application scenarios. This helps to accurately identify semantic conflict indicators and generate high-quality index semantic conflict lists, providing reliable basis for subsequent semantic mapping and knowledge fusion. At the same time, the calculation results of the similarity formula can also be used as the weight of the edge between index nodes, enriching the semantic representation ability of the knowledge graph. In summary, the formula starts from multiple angles such as vector semantics, numerical characteristics and attribute matching, and designs an index similarity measurement method with strong robustness and good interpretability.

[0190] Step S2200, traversing the index semantic conflict list, generating a unified semantic representation for each semantic conflict indicator, and constructing a semantic mapped cross-department knowledge graph;

[0191] Further, as Figure 6 shown, step S2200 includes:

[0192] Step S2210, for each index pair in the index semantic conflict list, query the pre-set semantic mapping rule library to match the mapping rule; the semantic mapping rule is represented in the form of IF-THEN;

[0193] Step S2220, according to the matched mapping rule, adding the mapping relationship edge and mapping attribute between the two semantic conflict indicators in the indicator pair to form a cross-department semantic mapping layer;

[0194] Step S2230, integrating the cross-department semantic mapping layer into the cross-department knowledge graph to form a semantic mapping cross-department knowledge graph.

[0195] Specifically, semantic conflict refers to the use of different terms by different business departments to express the same or similar concepts, or the use of the same term but referring to different concept connotations. This semantic conflict phenomenon is common in cross-department collaboration and becomes a barrier to business interconnection. For example, the production department uses "output" to represent product output, while the finance department uses "output" to represent value, which has differences in semantic reference. For another example, the "pass rate" of the production department and the "first inspection pass rate" of the quality department have similar terms, but the former has a denominator of production number and the latter has a denominator of inspection number, which have inconsistent calculation scopes. Semantic conflict leads to difficulties in directly interfacing and transferring cross-department information, affecting collaboration efficiency.

[0196] To eliminate semantic gaps, each pair of semantic conflict indicators identified in the semantic conflict list needs to be traversed, and a unified standard semantic representation is given to them through semantic mapping. There are two ways to unify semantic representation: one is to select one term from the conflict indicator pair as the standard expression, and the other is to create a new standard term. The choice depends on the semantic similarity of the conflict indicators. The result of semantic mapping is to replace the original multiple synonymous indicators with a unified term, and to establish a mapping association between indicators. This process outputs the semantic mapping knowledge graph. In the graph, indicators of the same concept are integrated into a unified term node, eliminating semantic redundancy; indicators of different concepts, even if the terms are similar, are distinguished through attribute definition, eliminating semantic confusion; for indicators with similar semantics, semantic tracking is maintained through mapping relationship edges.

[0197] Through data analysis and expert experience summary, some common indicator semantic mapping rules can be induced. These rules are included in the semantic mapping rule library, which is used to guide semantic conflict resolution. The rule library organizes rules in the form of IF-THEN, the former part (IF) judges the semantic relationship type of the indicator pair, and the latter part (THEN) gives the corresponding mapping strategy. The semantic relationship type can be determined based on the extension overlap degree and the connotation similarity of the concept, mainly including: equivalence relationship, implication relationship, intersection relationship, etc. For example:

[0198] IF the extension of indicator G is completely overlapped with that of indicator H THEN select the term with high word frequency as the standard expression;

[0199] This rule is applicable to the case of term expression difference, but the reference object is completely consistent. The disambiguation strategy is to select the more extensive indicator term.

[0200] IF the extension of indicator G has a containment relationship with the extension of indicator H THEN select the indicator term with the larger extension as the standard expression.

[0201] This rule is applicable to the case where one indicator concept covers another indicator concept. The disambiguation strategy is to select the term with a broader concept extension.

[0202] IF the semantic similarity of the intension of indicator G is higher than the threshold value compared with the intension of indicator H THEN construct a new standard indicator term with a superordinate concept, and establish a "synonymous" relationship edge between indicators G / H and the newly constructed indicator.

[0203] This rule is applicable to the case where the concept intension is similar, but it is difficult to directly express it with the original term of G / H. The disambiguation strategy is to newly construct a superordinate concept indicator that covers the intension of G / H.

[0204] The semantic mapping rule library is not fixed and can be dynamically updated according to business changes. In the rule matching process, for semantic relationship types that cannot be covered, new mapping rules are generated and added to the rule library. The supplement and improvement of rules make the processing paradigm of semantic mapping more standardized and the degree of automation continuously improved.

[0205] Apply the semantic mapping rule to the indicator pair, that is, establish a mapping relationship edge with attributes between the conflicting indicators. The mapping relationship type can be "equivalent", "containment", "intersection", etc., consistent with the semantic relationship type matched in the rule library. The attributes of the mapping edge record the extraction rules, confidence, and other meta-information of the mapping relationship. Aggregate the mapping relationship edges in each conflicting indicator pair to form a semantic mapping layer. The semantic mapping layer is a mesh-like semantic association network between conflicting indicators, and the mapping relationship in it clearly depicts the semantic similarity and difference of each indicator. Visualize the semantic mapping layer to intuitively display the semantic context of cross-departmental terms.

[0206] For indicator pairs that cannot be matched to mapping rules, it means that the semantic relationship they contain exceeds the expression ability of the existing rule library, and new rules need to be expanded. Experts analyze the semantics of these indicator pairs, determine their mapping path, and summarize new mapping rules. New rules are incorporated into the rule library and applied to subsequent mapping practice. The semantic mapping rule library is continuously enriched and improved in use to adapt to the semantic association needs of different fields.

[0207] The mapping relationship edges and their attributes in the semantic mapping layer need to be integrated into the cross-departmental knowledge graph to replace the original semantic conflict nodes and relationships in the graph. The fusion process needs to coordinate the semantic representation of the original knowledge graph and the semantic mapping layer:

[0208] Traverse the mapping relationship edge in the semantic mapping layer, and for the two end nodes of the edge, add the corresponding mapping relationship edge in the cross-department knowledge graph. At the same time, the attributes of the mapping edge are also synchronized to the knowledge graph. For example, there is an "equivalent" relationship edge connecting "yield" and "pass rate" in the mapping layer, and the corresponding "equivalent" edge needs to be added in the graph.

[0209] Then, the original homonymy index nodes and related edges in the knowledge graph are deleted, and the index nodes with unified semantic representation are retained. The edges connected with "yield" and "pass rate" in the graph are deleted, and only the "pass rate" node is retained as the standard expression of the semantic concept.

[0210] For the case where there is a mapping relationship in the semantic mapping layer but there is no corresponding node in the knowledge graph, these nodes and edges are directly added to the knowledge graph. For example, "output" and "output value" form a "cross" relationship in the mapping layer, but the "output value" node is missing in the graph, so the "output value" node and the "cross" mapping edge are added.

[0211] Semantic mapping standardizes the index terms in the cross-department knowledge graph, and connects related indexes through mapping relationship edges, which not only eliminates redundancy and ambiguity, but also maintains the richness and continuity of semantics. The mapped knowledge graph contains unified semantics and the association between original semantics, making the knowledge organization more standardized and flexible. Semantic mapping alleviates the communication barriers between departments and lays the foundation for building a unified semantic system. Under the premise of semantic consistency, integrating the business knowledge of each department can achieve a global understanding of business activities and optimize decision-making. Taking "pass rate" as an example, through semantic mapping, the difference between it and "first inspection pass rate" is clarified. The former reflects the quality of the final product, and the latter focuses on the production process control. The two form a progressive relationship. When the final product pass rate decreases, the first inspection pass rate of the upstream link can be traced to find the source process that introduces defects. After semantic unification, the knowledge association extends the quality management from end detection to source control, realizing cross-department quality collaboration.

[0212] Step S2300, a second data set of cross-functional departments is obtained, the second data set comprising real-time business parameters and real-time environmental data of the cross-functional departments; based on the second data set and the semantic mapped cross-department knowledge graph, a fused cross-department knowledge graph and a semantic correlation degree matrix are obtained;

[0213] Further, as shown in Figure 7 , step S2300 comprises:

[0214] Step S2310, forming a real-time business feature vector according to the real-time business parameters of the cross-functional departments;

[0215] Step S2320, forming a real-time environmental feature vector according to the real-time environmental data of the cross-functional departments;

[0216] Step S2330, fusing the real-time business feature vector and the real-time environment feature vector in the semantic-mapped cross-department knowledge graph to form a fused cross-department knowledge graph;

[0217] Step S2340, calculating the semantic correlation degree between departments based on the fused cross-department knowledge graph to generate an N*N semantic correlation degree matrix, N being the number of departments;

[0218] Step S2350, normalizing the semantic correlation degree matrix.

[0219] Specifically, the second data set refers to the data reflecting the latest business operation status of each department continuously collected after the construction of the knowledge graph. Unlike the historical data used in the construction phase, the second data set emphasizes real-time and aims to provide the latest business perspective for the knowledge graph. Real-time business parameters can include order quantity, production plan execution progress, equipment working condition parameters, quality detection results, etc., which depict the instant performance of the business process. Real-time environment data can include upstream and downstream supply and demand dynamics, industry development trend, macroeconomic trend, etc., which reflect the changes in the external ecological environment of the enterprise. To obtain the second data set, real-time interface with the information system of the business department needs to be established, and environment data needs to be collected from external channels. During the data collection process, attention should be paid to the timeliness, integrity and accuracy of the data. By integrating the second data set into the semantic-mapped cross-department knowledge graph, the knowledge graph is closely combined with the actual business, thus becoming an important basis for the implementation of business decisions.

[0220] The real-time business feature vector is a quantitative representation of the operation status of each department. Each element in the vector corresponds to a business parameter, and the parameter value reflects the real-time performance of the business element. To construct the real-time business feature vector, first, the dimension of the vector needs to be determined, i.e., how many business parameters to include. Usually, the selected parameters are screened based on their business importance and data availability. Then, the real-time data interface of the business department is accessed to obtain the current value of the selected parameters. The real-time data interface here can be a production execution system (MES), a supply chain management system (SCM), a warehouse management system (WMS), etc. Interface access needs to handle issues such as identity authentication and permission control to ensure data security. For heterogeneous data returned by different interfaces, cleaning, conversion and unified representation are needed to form standardized parameter values. Finally, the real-time values of each parameter are assembled into a vector according to the agreed order, i.e., the real-time business feature vector. When the business parameters change, the real-time business feature vector is also updated. Taking the production department as an example, the real-time business feature vector may include elements such as "current capacity utilization", "current shift qualification rate" and "equipment running rate". Through the real-time business feature vector, scattered department business data can be aggregated into a unified feature representation, providing a basis for cross-department knowledge correlation.

[0221] The business operation status across departments is quantified as a real-time business feature vector, establishing a unified data representation for semantic correlation analysis. Compared to the original business parameters, the real-time business feature vector has a fixed structure and consistent semantics, reducing the difficulty of cross-department data correlation. Through real-time feature extraction, business data between departments that were originally difficult to compare directly are mapped to the same vector space, providing computational capabilities at the semantic level. This feature-based representation simplifies the complexity of the original data and highlights the key semantic relationships between business elements. This lays the data foundation for subsequent cross-department business comparison, impact analysis, and coordinated decision-making.

[0222] The real-time environmental feature vector aims to characterize the macro environment in which the enterprise operates. Environmental factors such as industry policies and market trends often have a global impact on various business departments. By sensing environmental changes, it helps enterprises respond and innovate in a timely manner. To construct a real-time environmental feature vector, relevant environmental factors need to be defined. This requires analysis from industry attributes, industry chain position, and competitive situation. The selection of environmental factors should consider both importance and availability, and prefer factors that have a significant impact on each department and relatively easy to obtain data. Then, access various external data sources to obtain real-time values of each environmental factor. External data sources can be industry associations, third-party research institutions, and upstream and downstream enterprises in the industry chain. Establish a long-term data sharing mechanism with data sources and clarify the scope and frequency of data use. At the same time, handle data outliers, identify and correct unreasonable extreme values. Finally, process real-time environmental data through feature engineering to extract a fixed-dimensional real-time environmental feature vector. Taking a home appliance manufacturing enterprise as an example, real-time environmental features may include "real estate sentiment index", "consumer confidence index", and "key raw material price index". The real-time environmental feature vector enables enterprise managers to view internal management from an open perspective and actively adapt to external changes.

[0223] By introducing the real-time environmental feature vector, the knowledge graph gains the ability to perceive the macro environment, helping enterprises maintain strategic focus in a complex and changing external situation. Since environmental factors often have cross-departmental impacts, a single department cannot fully grasp the overall situation. The real-time environmental feature vector provides a "nerve ending" for cross-departmental collaboration, enabling departments to form linkages based on consistent environmental awareness. When the real-time environmental feature vector indicates significant changes in the external situation, the knowledge graph can quickly calculate the impact on each business element, guiding relevant departments to adjust their decisions in a timely manner. This transition from passive adaptation to active change enhances the overall environmental resilience of the enterprise.

[0224] The fused cross-department knowledge graph is based on the knowledge graph after semantic mapping, and is formed by superimposing real-time business and environmental characteristics. This process embodies the semantic enhancement from static to dynamic and from concept to instance. In terms of technical implementation, the graph embedding technology is usually used to map the real-time feature vector into a new node in the knowledge graph, and the new semantic association edge is generated based on the correlation algorithm between features. For example, the "current capacity utilization rate" element in the real-time business feature vector can be mapped to the "current capacity utilization rate" node in the knowledge graph, and an association edge is established with the "production" node. The element value in the real-time feature vector is the attribute of the newly generated node. Through the continuous update of the attribute value, the knowledge graph can immediately reflect the latest business status. In the fusion process, attention should be paid to the logical coordination of new nodes, new edges and the original knowledge graph to avoid semantic conflicts. In addition, with the continuous writing of real-time data, the size of the fused knowledge graph will continue to increase. Optimization measures such as graph partitioning and hierarchical storage should be taken to ensure the efficiency of query and reasoning. In summary, the fused cross-department knowledge graph fully combines the originally conceptually abstract domain knowledge with real-time business data, and gives the knowledge graph the ability to analyze the present and guide the future.

[0225] The fused cross-department knowledge graph becomes the information hub connecting real-time business and environment, making knowledge-driven collaborative decision-making more closely related to business reality. Through continuous fusion of real-time features, the knowledge graph's grasp of the overall business is more three-dimensional and dynamic, and the timeliness and pertinence of collaborative optimization are significantly improved. On the one hand, the integration of real-time business features enables the knowledge graph to accurately depict the strengths and weaknesses of current business operations, identify "benefit coupling points" for department collaboration, and propose targeted optimization measures. On the other hand, after the integration of real-time environmental features, the knowledge graph's judgment of the "opportunity-threat" situation for each department is more forward-looking. This shift from passive response to proactive exploration is conducive to coordinating internal and external resources to achieve a turnaround.

[0226] The semantic correlation degree measures the mutual influence strength of each department's business in the fused knowledge graph. The higher the correlation degree, the greater the coupling degree between departments, and the need to strengthen collaborative control. Calculating semantic correlation degree is usually based on similarity algorithms for graph data, such as random walk algorithm, node neighbor algorithm, etc. These algorithms take into account the structure of the knowledge graph (such as the number of shared nodes) and semantic (such as the weight of the edge) factors, and can comprehensively evaluate the correlation degree between departments. To improve computational efficiency, a correlation degree threshold is usually set in the knowledge graph to filter out redundant calculations with very weak correlation degree. The semantic correlation degree matrix is N x N dimensional, where N is the number of departments within the enterprise. The (i', j') element in the matrix represents the semantic correlation degree of the i'th department to the j'th department. Since correlation is asymmetric, the matrix is often not a symmetric matrix. The semantic correlation degree matrix dynamically depicts the interwoven influence network of each department's business within the enterprise, and reveals the business dependency relationship of "one wins together, one loses together". Relying on the correlation matrix, we can see through the weak links of business collaboration and provide decision-making reference for systemically promoting collaborative change.

[0227] By dynamically calculating the semantic correlation degree, the implicit involvement of business between departments is continuously revealed, the "information cocoon" is broken, and a collaborative atmosphere is created. From the perspective of correlation, each department is deeply aware of the limitations of "fighting alone", and actively seeks collaborative breakthroughs. On the one hand, high correlation degree indicates great potential for collaborative efficiency. Departments should focus on breaking through key links to achieve complementary advantages and mutual promotion. On the other hand, departments with high negative correlation degree are "collaborative minefields", whose decision-making errors will affect the whole, and need to be carefully managed. In addition, the correlation degree matrix can also be used for department performance evaluation, incorporating collaborative performance into the evaluation system, forming a community of shared honor and disgrace, and stimulating collaborative motivation.

[0228] Normalization is a mathematical transformation that maps the original correlation degree to the interval [0, 1] to form a unified and intuitive measure. Common normalization methods include linear normalization, logarithmic normalization, exponential normalization, etc. The appropriate normalization function should be selected according to the value range distribution of the correlation degree. The normalized semantic correlation degree matrix is convenient for horizontal comparison. Matrix elements close to 1 represent strong correlation between departments, implying the inherent demand for "mutual support"; matrix elements close to 0 mean weak business connections between departments, indicating the risk of "going it alone". Through correlation clustering analysis, the departments within an enterprise can be divided into several "coordination circles", and the coordination effect within the circle is strong, which is the "coordination template" for key breakthroughs. Cross-circle coordination is weak and needs to focus on building bridges and links for department collaboration. The normalized semantic correlation degree matrix intuitively presents the interest connection degree between departments, becoming a "configuration map" to guide the optimization of enterprise organization and process mechanism. According to the "coordination circle" layout of enterprise architecture, supporting coordination incentive mechanism can maximize the synergy effect and improve the overall competitiveness of the enterprise. Through correlation matrix normalization, the interest correlation between departments is intuitively depicted, simplifying the difficulty of coordination management. Traditional department division is mainly function-oriented, while the normalized semantic correlation degree matrix is based on business essence, outlining the implicit departmental interest community and providing new ideas for enterprise organization remodeling.

[0229] Step S2300 keeps the knowledge graph synchronized with business practice by continuously introducing real-time business and environmental data, reducing the time lag caused by using historical data in the knowledge graph construction phase. This "living knowledge graph" enhances the ability of the knowledge graph to analyze and trace the source of business problems and predict trends by dynamically mapping the business panorama. At the same time, embedding real-time data into a unified semantic system allows business intelligence from different departments to be integrated and associated at the semantic level, triggering linkage. This provides a data and knowledge foundation for cross-departmental collaboration and optimization.

[0230] Step S2400 extracts the weights of each node and edge of the fused cross-departmental knowledge graph, dynamically adjusts the weights of each node and edge of the fused cross-departmental knowledge graph according to the semantic correlation degree matrix, and generates a cross-departmental collaboration knowledge graph.

[0231] Further, step S2400 includes:

[0232] Step S2410 extracts the semantic correlation degree between departments from the semantic correlation degree matrix. For department pairs with a semantic correlation degree higher than a preset correlation degree threshold θ3, the first weight adjustment coefficient k1 is used to increase the weight of related nodes and edges, and the first weight adjustment coefficient k1 is in a positive correlation with the semantic correlation degree.

[0233] Step S2420, for the department pair with a semantic correlation degree less than or equal to a preset correlation degree threshold θ3, the weight of the related nodes and edges thereof is reduced by a second weight adjustment coefficient k2, and the second weight adjustment coefficient k2 is inversely proportional to the semantic correlation degree.

[0234] Specifically, on the basis of the fused cross-department knowledge graph, the weight of the nodes and edges in the graph is adjusted according to the closeness of the business association between different departments, so as to form a knowledge graph that can reflect the cross-department collaboration relationship. The weight of a node reflects the importance of the node in cross-department collaboration, and the weight of an edge reflects the strength of the collaborative association between the connected nodes. The basis for weight adjustment is the semantic correlation degree matrix, which quantitatively characterizes the semantic relevance between different departments based on business data. The weight adjustment includes two directions: weight increase and weight decrease.

[0235] The semantic correlation degree matrix quantitatively characterizes the degree of relevance between different departments in the semantic layer of business data, and is an important basis for weight adjustment. θ3 is a threshold for judging whether the semantic correlation between departments is significant, which needs to be set by business experts based on the analysis of the distribution of semantic correlation degrees, such as 0.8. For department pairs with a semantic correlation degree higher than the threshold θ3 (such as 0.8), such as the semantic correlation degree of departments E' and F' is 0.9, which means that they are closely related in terms of index association, process integration, decision-making collaboration, etc., and there is a large collaborative optimization space, so it is necessary to highlight their position in the collaborative knowledge graph and increase the weight of the related nodes and edges. k1 is the first weight adjustment coefficient, and its value increases with the increase of the semantic correlation degree, that is, k1 is proportional to the semantic correlation degree. For example, if k1 = semantic correlation degree, the k1 of departments E' and F' is 0.9, and if the original weight of a node between them is 1, the new weight after adjustment = 1 x (1 + k1) = 1 x (1 + 0.9) = 1.9, which is increased by 90%. After k1 adjustment, the connection of high semantic correlation departments in the collaborative knowledge graph will be more prominent, and the corresponding indicators, processes, decision-making problems, etc. will become the focus of collaborative management. Weight increase makes resource allocation tilt towards areas with high collaboration needs, and strengthens cross-department collaboration.

[0236] For departments with a semantic correlation lower than the threshold value θ3, such as departments C' and D' with a semantic correlation of 0.5, it indicates that the business-level collaborative correlation between the two is relatively weak, and the collaborative appeal is not the most urgent. In order to focus resources on key collaborative fields, it is necessary to reduce the weight of related nodes and edges between C' and D', and weaken their influence in the collaborative knowledge graph. k2 is a second weight adjustment coefficient, and its value increases as the semantic correlation decreases, that is, k2 is inversely proportional to the semantic correlation. For example, k2 = 1-semantic correlation, k2 of departments C' and D' = 1-0.5 = 0.5, and if the original weight of a node between them is 1, the new weight after adjustment = 1 x (1-k2) = 1 x (1-0.5) = 0.5, which is reduced by 50%. After k2 adjustment, the connection strength of low semantic correlation departments will be reduced in the collaborative knowledge graph, avoiding excessive dispersion of collaborative resources. The application of k2 aims to identify relatively weak collaborative fields, while focusing on high semantic correlation directions, and moderately reducing attention and investment in marginal collaborative fields.

[0237] In summary, the weight adjustment in the generation process of the cross-department collaborative knowledge graph embodies the differentiated collaborative resource allocation idea based on semantic correlation. High semantic correlation reveals the key areas and focus points of cross-department collaboration, which need to be given priority and resource tilt; while the collaborative needs of low semantic correlation fields are relatively secondary, and resource investment should be correspondingly reduced. This differentiated weight adjustment helps collaborative management to focus on key points and improve resource utilization efficiency. The quantitative semantic correlation provides an objective basis for weight adjustment, and k1 and k2 further realize the adaptation of weight adjustment amplitude to semantic correlation, and comprehensively consider the balance between strengthening collaboration and focusing on key points. It can be predicted that the cross-department collaborative knowledge graph will provide a global perspective and landing starting point for collaborative optimization, and become an important tool for connecting scattered businesses and stimulating collaborative vitality. Facing the future, with the continuous increase of business complexity, cross-department collaboration will become the norm, and it is urgent to activate organizational vitality and drive collaborative upgrading through collaborative knowledge graph.

[0238] Step S3000, based on the cross-department collaborative knowledge graph, generates cross-department collaborative decision suggestions.

[0239] Further, step S3000 includes:

[0240] Step S3100, collect operation time series data of nodes in the cross-department collaborative knowledge graph, and obtain time sequence dependent features between nodes according to the operation time series data and the pre-constructed time sequence dependent feature model;

[0241] Specifically, the operation records of the process nodes are extracted from the business systems and logs of each department to obtain the time series of each node. Step S3100 depicts the time sequence dependency relationship between the process nodes by analyzing the operation time series generated by the business activities. The operation time series data records the start and end times of each business activity. By aligning the operation logs of different departments, a globally unified time series is established.

[0242] In the preprocessing stage, the time format needs to be unified, and missing values and outliers need to be processed to ensure the integrity and accuracy of the time series. Common missing value processing methods include deleting missing records, nearest neighbor filling, interpolation, etc. Outliers can be identified by methods such as box plots and replaced or removed.

[0243] The pre-constructed time sequence dependency feature model can be a statistical model or a machine learning model, which is used to extract the time sequence dependency features between nodes from the time series data. Taking a statistical model as an example, the time delay distribution between nodes can be calculated, i.e. the time interval distribution between the end time of node G' and the start time of node H'. The concentration of the time delay distribution reflects the time sequence dependency strength between nodes, and the dispersion of the delay distribution reflects the time sequence dependency stability between nodes.

[0244] For example, for two consecutive production process nodes, the time delay distribution is concentrated in 1-2 hours, indicating that there is a close time sequence dependency between the two processes, and the end time of the previous process basically determines the start time of the next process, and the two processes are highly synchronized. If the time delay distribution is dispersed, such as some batches are delayed by 1 hour and some are delayed by 5 hours, it indicates that there is a problem with the coordination of the two processes, and the time sequence dependency is unstable.

[0245] Machine learning models such as RNN, LSTM, etc. can consider the long-term dependency of time series and mine deep patterns of time sequence relationships between nodes. By training the model to predict the operation time series of a node, and taking the error between the predicted value and the true value as the measure of time sequence dependency strength. The smaller the prediction error, the greater the influence of the operation time of the node on other nodes, and the stronger the time sequence dependency.

[0246] After obtaining the time sequence dependency features between nodes, a time sequence dependency feature vector can be constructed for each node. For example, node The time sequence dependency strength of nodes , , is 0.8, 0.5, and 0.1 respectively, and the time sequence dependency feature vector of node can be represented as [0.8, 0.5, 0.1]. The time sequence dependency feature vector depicts the dependency relationship of the node under global time sequence constraints, and is an important input for subsequent fusion analysis.

[0247] Step S3100 mines the time sequence dependency between nodes, which is crucial for optimizing the process and controlling the rhythm. There may be implicit constraints in apparently parallel business activities. Revealing the time sequence dependency helps to find the bottleneck of the process, optimize task scheduling and resource allocation. By introducing a pre-built model, time sequence features can be efficiently and accurately extracted from large-scale time sequence data. Artificial analysis of time series is time-consuming and difficult to find complex patterns. By building a statistical or machine learning model, time sequence features can be automatically analyzed and adapted to different business scenarios. The obtained time sequence dependency features can quantitatively represent the synchronization degree between nodes, providing early warning indicators for business process monitoring. By tracking the changes in time sequence dependency strength, abnormal delays or advances of process nodes can be detected in a timely manner, and the rhythm and resource scheduling can be optimized.

[0248] Step S3200 obtains semantic interaction features between nodes according to the nodes and edges of the cross-department collaborative knowledge graph and the pre-built semantic interaction feature model.

[0249] Specifically, step S3200 learns the semantic interaction patterns between nodes through deep exploration of the knowledge graph structured data. In the knowledge graph, nodes represent specific business elements (such as indicators, processes, decisions, etc.), and edges represent the association between business elements. By analyzing the semantic interaction of nodes and edges, key nodes and their influence paths can be found.

[0250] Semantic interaction features describe the influence strength and mode of a node on other nodes. A node transmits its influence to adjacent nodes through edges in the graph, and then indirectly influences more distant nodes. The strength of the influence is related to the shortest path length between nodes and the semantic of the connecting edge.

[0251] After obtaining the node and edge data of the knowledge graph, the first step is to perform representation learning on the nodes and edges. Through models such as Skip-gram, CBOW, and DeepWalk, nodes can be mapped to a low-dimensional dense vector space, making semantically similar nodes closer in the vector space. The representation learning of edges can use models such as TransE, which realizes the mapping from the head entity vector to the tail entity vector through translation transformation, encoding the directional semantics of the edge.

[0252] Based on the vector representation of nodes and edges, a pre-trained semantic interaction feature model is used to learn the semantic interaction features of nodes. The semantic interaction feature model uses Graph Attention Network (GAT). Compared with traditional graph convolution networks, GAT introduces an attention mechanism for the aggregation of neighboring nodes, making key neighbors obtain higher weights, while the influence of non-key neighbors is weakened. This conforms to the actual characteristics of business element interaction.

[0253] Each node in GAT pays attention to the importance of its neighbor nodes, which is proportional to the attention weight. The attention weight is obtained by calculating the similarity of the nodes. The higher the similarity, the stronger the semantic correlation between the nodes, and the higher the weight obtained when aggregating the information of the neighbor nodes. GAT fuses the original features of the nodes with the information transmitted by the neighbor nodes at each layer, and finally obtains the semantic interaction feature representation of a node through multi-layer iteration and update.

[0254] By analyzing the semantic interaction features between nodes, the association patterns of business elements can be found. For example, when the quality management indicators are abnormal, the nodes with high semantic interaction intensity with the quality indicators may be upstream factors such as supplier change and equipment failure, revealing the root cause of the quality problem. While the downstream nodes with high semantic interaction intensity with the quality indicators may have customer satisfaction and return rate, indicating that the impact of the quality problem has spread. Around the semantic interaction features of the quality abnormal nodes, an influence network is outlined, which helps traceability analysis and decision making.

[0255] The graph attention network can adaptively focus on the importance of different neighbor nodes and extract key features of semantic interaction. Business elements are complex and not all associations are effective. Through the attention mechanism, higher weights are given to key nodes and key paths, so that the decision focuses on the main contradictions. The multi-layer graph attention network can explore high-order semantic associations between nodes and reveal the implicit dependencies of business elements. Nodes with direct associations can be found in the shallow layer of the graph, while nodes with indirect associations need more layers of information transmission to be mined. This helps to systematically understand business problems and develop fundamental solutions. Semantic interaction features can be used to predict the consequences of changes in node states, assisting risk assessment. Understanding the influence transmission path between nodes can predict the scope of abnormality of a certain business indicator, take measures in advance, and achieve precise governance.

[0256] Step S3300, fuse the time-dependent features and semantic interaction features, combine the logical rules for reasoning, and generate cross-department collaborative decision-making suggestions.

[0257] Further, step S3300 includes:

[0258] Step S3310, fuse the time-dependent features and semantic interaction features to form a fusion feature vector;

[0259] Step S3320, build a logical rule library and establish a logical rule reasoning engine;

[0260] Step S3330, input the fusion feature vector into the logical rule reasoning engine for reasoning, and generate cross-department collaborative decision-making suggestions.

[0261] Specifically, feature fusion is the process of integrating node features extracted from two different perspectives into a unified representation. Temporal dependency features depict the association patterns of nodes with other nodes in the time dimension, such as time-lag correlation, Granger causality, etc. Semantic interaction features depict the importance of nodes in the semantic network of the knowledge graph and their semantic similarity with other nodes. These two types of features describe the role of nodes in the knowledge graph from the perspectives of temporal dependence and semantic association, respectively. Fusing them into a vector can form a more comprehensive feature description of the node.

[0262] Common methods of feature fusion include feature concatenation and feature weighted summation. Feature concatenation is to connect two feature vectors end to end to form a longer vector. The advantage of feature vector concatenation is to retain the complete information of the original features, but the disadvantage is that the dimension of the fused features is high. When the original feature dimension is already high, concatenation will further exacerbate the dimension disaster. Another fusion method is feature weighted summation, which can map features from different sources to the same dimension by weighted summation, forming a "mix" in semantics. The selection of weight coefficients can be based on the discriminative ability of the features, and the features that have a greater impact on the prediction results are given higher weights. The dimension of the fused features generated by feature weighted summation is the same as that of the original features, which to some extent alleviates the dimension explosion problem. However, part of the original feature information may be lost in the fusion process.

[0263] The fused feature vector expresses the joint semantics of temporal dependency features and semantic interaction features, and can better support subsequent tasks. Through feature fusion, heterogeneous features can be compared in the same semantic space, and the internal relationship between cross-department business data can be found. This feature fusion is a common technique in data mining and machine learning, which can be applied to clustering, classification, association rule mining, etc. In this scenario, by extracting and fusing different features of the knowledge graph nodes, conditions are created for mining cross-department collaboration. The fused features can be input into the prediction model to find out the indicators and processes that are more likely to have cross-influence, identify key objects that need cross-department coordination, and discover feature patterns that promote or lead to non-coordination. Accordingly, the levers for inter-department communication and cooperation and the entry points for joint optimization can be formed.

[0264] Logical rules are IF-THEN form statements used to depict the causal dependency relationship between variables. Logical rule reasoning triggers rules when the conditions are met by matching the rule antecedent conditions, and draws corresponding conclusions. Building a logical rule base is to formalize the business domain knowledge as a series of rules by summarizing the business domain knowledge. These rules can come from the experience of domain experts, business constraints, management specifications, etc., and include causal relationship judgments in production, quality, supply, etc. For example:

[0265] IF [Defect rate rises] AND [Raw material quality declines] THEN [Incoming material quality inspection fails];

[0266] IF [Key process capability index > 1.33] THEN [Process meets production requirements];

[0267] IF [Raw material price rises] AND [Order volume declines] THEN [Reduce raw material procurement];

[0268] In the rules, the antecedent (after IF and before THEN) represents the trigger condition, and the consequent (after THEN) represents the reasoning result. Converting business knowledge into logical rules can make it organized and applied in a way that computers can understand.

[0269] On this basis, a logical rule reasoning engine needs to be established to realize the automatic triggering and matching of rules. The reasoning engine matches the input data with the antecedents of the rules in the rule base. When it finds that a rule is satisfied, it triggers the rule and outputs the reasoning result defined in the consequent. The implementation of the reasoning engine consists of three steps: matching, conflict resolution, and execution. Matching is to check whether the input data can satisfy the antecedent condition of one or more rules. When multiple rule antecedents are satisfied, conflicts may arise. Conflict resolution strategies need to be used to determine which rule to execute, such as based on rule priority or based on rule specificity (rules with more specific conditions are preferred). After determining the rule, the rule is executed, and its reasoning result is returned.

[0270] For example, the input data shows that the characteristics of node meet "quality index declines" and "production speed increases", which match the rule "IF [quality index declines] AND [production speed increases] THEN [there may be quality control problems]". The rule is triggered and executed, and the reasoning result "there may be quality control problems" is output.

[0271] The advantage of logical rule reasoning is that it is interpretable, and the reasoning result can be traced back to the triggered rule. This makes the reasoning process transparent to business personnel and easy to interpret. At the same time, the rule base can be continuously updated and expanded, adding new business knowledge to improve system adaptability. The disadvantage is that it requires domain experts to participate in knowledge acquisition and rule construction, which is costly. And the coverage of the rule base is limited by the breadth of expert knowledge, and the generalization ability to unknown scenarios is not strong. It needs to be combined with other machine learning methods to overcome its limitations.

[0272] Node fusion features can be used as input for logical rule reasoning. The reasoning engine matches node features with rule antecedents to identify feature combinations that meet pre-set business logic and then derives the corresponding reasoning result. This process can be applied to abnormal node identification, bottleneck mining, opportunity discovery, etc.

[0273] Take the identification of abnormal nodes as an example, the following rules are designed:

[0274] IF [quality index decline amplitude > 10%] AND [process parameter change] THEN [process change leads to quality abnormality];

[0275] The quality index decline amplitude of a certain node is 15%, and its process parameter has changed. This feature combination hits the above rule. The reasoning engine judges that the node has "process change leading to quality abnormality" accordingly. The system generates a warning message and pushes it to the quality and production departments, prompting attention to this node, analyzing the impact of process change, and taking corrective measures.

[0276] Intelligent decision-making assistance based on logical rule reasoning has three positive effects:

[0277] 1. Triggered decision-making. Passive waiting is not suitable for dynamic business environment. Rule triggering based on real-time data flow can realize decision-making closed loop, making management from passive response to active prevention.

[0278] 2. Collaborative decision-making. Departmental walls and information silos lead to limited decision-making perspective. Cross-departmental data fusion and rule reasoning help clarify decision-making dependencies and consolidate consistent action.

[0279] 3. Traceable decision-making. Lack of effective mechanism to trace the reasons for decision-making is not conducive to continuous optimization. Rule reasoning can provide decision-making basis, making the decision-making process interpretable and traceable, and providing support for review and learning.

[0280] For example, a certain enterprise uses this solution to optimize cross-departmental production collaboration. By integrating production, quality, and supply data to build a knowledge graph, it extracts the timing and semantic features of key nodes and builds a rule base. The system monitors a material supply node whose features meet the rule "IF [supplier performance assessment less than 80 points] AND [material quality fluctuation large] AND [arrival cycle extended] THEN [replace alternative supplier]". To avoid the spread of supply risks to the production link, the reasoning engine timely outputs the decision-making suggestion of "replace alternative supplier", and analyzes the possible affected production task nodes based on the knowledge graph. The production management department adjusts the production plan accordingly, and the quality department focuses on tracking related batches of products, forming a collaborative response among multiple departments.

[0281] This case reflects a typical path of integrating features and rule reasoning to support cross-departmental collaborative decision-making, i.e., breaking down data flow, integrating business rules, and connecting departmental actions. This solution effectively improves the timeliness, effectiveness, and interpretability of cross-departmental collaboration, helping enterprises strengthen digital operation and improve risk response capability. Subsequently, feature extraction and rule setting can be dynamically optimized based on rectification effect data, so that the collaborative mechanism can continuously evolve.

[0282] Embodiment 2

[0283] The embodiment provides a reasoning system based on semantic association and logical rules of a knowledge graph, as shown in Figure 8 The embodiment provides a reasoning system based on semantic association and logical rules of a knowledge graph, as shown in

[0284] The graph construction module is used for obtaining a first data set across functional departments, constructing a structured data table according to the first data set, and constructing a cross-department knowledge graph based on the structured data table.

[0285] The graph optimization module is used for extracting an index node from the cross-department knowledge graph, identifying a semantic conflict index, and generating an index semantic conflict list; based on the index semantic conflict list, a semantic-mapped cross-department knowledge graph is constructed; a second data set across functional departments is obtained, and based on the second data set and the semantic-mapped cross-department knowledge graph, a fused cross-department knowledge graph and a semantic association degree matrix are obtained; weights of each node and edge of the fused cross-department knowledge graph are extracted, the weights of each node and edge of the fused cross-department knowledge graph are dynamically adjusted according to the semantic association degree matrix, and a cross-department collaborative knowledge graph is generated.

[0286] The decision reasoning module is used for generating a cross-department collaborative decision suggestion based on the cross-department collaborative knowledge graph.

[0287] In the graph construction module, the first data set includes index definition documents, business process description documents and historical decision records across functional departments; and the functional departments include a production department, a quality management department and a supply chain department.

[0288] In the graph construction module, the construction of the structured data table according to the first data set includes:

[0289] In step S1210, an index name A, a first relationship type and an index name B are extracted from the index definition document to form an index information triple; the index name A is the subject of the index information triple, the first relationship type is the predicate of the index information triple, and the index name B is the object of the index information triple; and the first relationship type includes a calculation relationship, an influence relationship and a composition relationship.

[0290] In step S1220, a process node A', a second relationship type and a process node B' are extracted from the business process description document to form a process triple; the process node A' is the subject of the process triple, the second relationship type is the predicate of the process triple, and the process node B' is the object of the process triple; and the second relationship type includes a time sequence relationship and a logical relationship.

[0291] Step S1230, extracting the decision event, the third relationship type, the decision question or the scheme from the historical decision record to form a decision information triple; wherein the decision event is the subject of the decision information triple, the third relationship type is the predicate of the decision information triple, and the decision question or the scheme is the object of the decision information triple; the third relationship type includes the relationship between the decision event and the decision question and the relationship between the decision event and the decision scheme;

[0292] Step S1240, constructing a structured data table based on the index information triple, the process triple and the decision information triple, wherein the structured data table includes the triple subject, the triple predicate and the triple object.

[0293] In the atlas construction module, the construction of the cross-department knowledge graph based on the structured data table includes:

[0294] Step S1310, mapping the triple subject and the triple object into nodes in the cross-department knowledge graph, and mapping the triple predicate into a directed edge between the nodes; the nodes in the cross-department knowledge graph include index nodes, process nodes and decision nodes;

[0295] Step S1320, extracting attribute information in the first data set, adding attributes to the nodes according to the attribute information, and generating node attributes; the node attributes include index node attribute information, process node attribute information and decision node attribute information; the index node attribute information includes index name, index definition and responsible department;

[0296] Step S1330, performing semantic type labeling on the nodes in the cross-department knowledge graph, and adding initial node weights to the nodes;

[0297] Step S1340, performing semantic type labeling on the directed edges in the cross-department knowledge graph, and adding initial edge weights to the directed edges;

[0298] Step S1350, optimizing the cross-department knowledge graph.

[0299] The step S1350 includes:

[0300] Step S1351, calculating the semantic similarity SI between the nodes in the cross-department knowledge graph, and defining the nodes with a semantic similarity SI greater than a preset first similarity threshold θ1 as synonymous concept nodes;

[0301] Step S1352, adding synonymous relationship edges between the synonymous concept nodes;

[0302] Step S1353, identifying superordinate and subordinate concept nodes in the cross-department knowledge graph, and adding belonging relationship edges between the superordinate and subordinate concept nodes.

[0303] In the atlas optimization module, the generating of the index semantic conflict list comprises:

[0304] In step S2110, all index nodes in the cross-department knowledge atlas are retrieved, and attribute information of the index nodes is extracted.

[0305] In step S2120, the index name and the index definition in the attribute information of the index nodes are vectorized to form an index vector corresponding to the index of the index node.

[0306] In step S2130, the similarity S1 between the index vectors of each index is calculated, a second similarity threshold θ2 is set, and for the index pair with the similarity S1 higher than θ2 but different responsibility departments, the index pair is marked as a semantic conflict index and added to the index semantic conflict list.

[0307] In the atlas optimization module, the constructing of the cross-department knowledge atlas after semantic mapping comprises:

[0308] In step S2210, for each index pair in the index semantic conflict list, a preset semantic mapping rule library is queried to match a mapping rule; the semantic mapping rule is in an IF-THEN form.

[0309] In step S2220, according to the matched mapping rule, a mapping relationship edge and a mapping attribute between the two semantic conflict indexes in the index pair are added to form a cross-department semantic mapping layer.

[0310] In step S2230, the cross-department semantic mapping layer is integrated into the cross-department knowledge atlas to form the cross-department knowledge atlas after semantic mapping.

[0311] In the atlas optimization module, the obtaining of the fused cross-department knowledge atlas and the semantic correlation degree matrix comprises:

[0312] In step S2310, a real-time business feature vector is formed according to real-time business parameters of the cross-functional departments.

[0313] In step S2320, a real-time environment feature vector is formed according to real-time environment data of the cross-functional departments.

[0314] In step S2330, the real-time business feature vector and the real-time environment feature vector are fused in the cross-department knowledge atlas after semantic mapping to form a fused cross-department knowledge atlas.

[0315] In step S2340, based on the fused cross-department knowledge atlas, the semantic correlation degree between the departments is calculated to generate an N×N semantic correlation degree matrix, N being the number of departments.

[0316] In step S2350, the semantic correlation degree matrix is normalized.

[0317] In the atlas optimization module, the generating the cross-department collaborative knowledge graph comprises:

[0318] In step S2410, the semantic correlation degrees between departments are extracted from the semantic correlation matrix, and for the department pair with a semantic correlation degree higher than the preset correlation threshold θ3, a first weight adjustment coefficient k1 is used to increase the weight of the related nodes and edges, and the first weight adjustment coefficient k1 is in a positive correlation with the semantic correlation degree;

[0319] In step S2420, for the department pair with a semantic correlation degree less than or equal to the preset correlation threshold θ3, a second weight adjustment coefficient k2 is used to reduce the weight of the related nodes and edges, and the second weight adjustment coefficient k2 is in an inverse correlation with the semantic correlation degree.

[0320] In the decision reasoning module, the generating the cross-department collaborative decision suggestion based on the cross-department collaborative knowledge graph comprises:

[0321] In step S3100, the operation time sequence data of the nodes in the cross-department collaborative knowledge graph are collected, and the time sequence dependent features between the nodes are obtained according to the operation time sequence data and the pre-constructed time sequence dependent feature model;

[0322] In step S3200, the semantic interaction features between the nodes are obtained according to the nodes and edges of the cross-department collaborative knowledge graph and the pre-constructed semantic interaction feature model;

[0323] In step S3300, the time sequence dependent features and the semantic interaction features are fused, and reasoning is performed in combination with the logic rules to generate the cross-department collaborative decision suggestion.

[0324] The step S3300 comprises:

[0325] In step S3310, the time sequence dependent features and the semantic interaction features are fused to form a fused feature vector;

[0326] In step S3320, a logic rule library is constructed, and a logic rule reasoning engine is established;

[0327] In step S3330, the fused feature vector is input into the logic rule reasoning engine for reasoning to generate the cross-department collaborative decision suggestion.

[0328] In addition, the parts of the above technical solutions in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive repetition.

[0329] The specific embodiments described above are further explained in connection with the purposes, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A knowledge graph-based semantic association and logical rule inference method, characterized in that, The method comprises: acquiring a first data set across departments, constructing a structured data table according to the first data set, and constructing a cross-department knowledge graph based on the structured data table; extracting index nodes from the cross-department knowledge graph, identifying semantic conflict indicators, and generating an index semantic conflict list; based on the index semantic conflict list, a semantic-mapped cross-department knowledge graph is constructed; a second data set across departments is acquired, and a fused cross-department knowledge graph and a semantic correlation matrix are obtained based on the second data set and the semantic-mapped cross-department knowledge graph; the weights of each node and edge of the fused cross-department knowledge graph are extracted, the weights of each node and edge of the fused cross-department knowledge graph are dynamically adjusted according to the semantic correlation matrix, and a cross-department collaborative knowledge graph is generated; based on the cross-department collaborative knowledge graph, cross-department collaborative decision suggestions are generated; the first data set includes index definition documents, business process description documents, and historical decision records across departments; the cross-department includes production departments, quality management departments, and supply chain departments; constructing a structured data table according to the first data set comprises: extracting index name A, first relationship type, and index name B from the index definition document to form an index information triple; wherein index name A is the subject of the index information triple, the first relationship type is the predicate of the index information triple, and index name B is the object of the index information triple; extracting process node A', second relationship type, and process node B' from the business process description document to form a process triple; wherein process node A' is the subject of the process triple, the second relationship type is the predicate of the process triple, and process node B' is the object of the process triple; extracting decision events, third relationship types, and decision problems or solutions from historical decision records to form decision information triples; wherein the decision events are the subjects of the decision information triples, the third relationship types are the predicates of the decision information triples, and the decision problems or solutions are the objects of the decision information triples; based on the index information triple, the process triple, and the decision information triple, a structured data table is constructed, which includes a triple subject, a triple predicate, and a triple object; constructing a cross-department knowledge graph based on the structured data table comprises: mapping the triple subject and the triple object into nodes in the cross-department knowledge graph, and mapping the triple predicate into directed edges between the nodes; the nodes in the cross-department knowledge graph include index nodes, process nodes, and decision nodes; extracting attribute information in the first data set, adding attributes to the nodes according to the attribute information, and generating node attributes; the node attributes include index node attribute information, process node attribute information, and decision node attribute information; the index node attribute information includes index name, index definition, and responsible department; performing semantic type labeling on the nodes in the cross-department knowledge graph, and adding initial node weights to the nodes; performing semantic type labeling on the directed edges in the cross-department knowledge graph, and adding initial edge weights to the directed edges; and optimizing the cross-department knowledge graph.

2. The knowledge graph-based semantic association and logical rule inference method according to claim 1, characterized in that, the optimization of the cross-department knowledge graph comprises: Calculate semantic similarity SI between nodes in the cross-department knowledge graph, and define nodes with semantic similarity SI greater than a preset first similarity threshold θ1 as synonymous concept nodes. Add synonymous relationship edges between synonymous concept nodes. Identify superordinate and subordinate concept nodes in the cross-department knowledge graph, and add belonging relationship edges between superordinate and subordinate concept nodes.

3. The knowledge graph-based semantic association and logical rule reasoning method according to claim 2, characterized in that, The generation of the index semantic conflict list includes: Retrieving all index nodes in the cross-department knowledge graph, and extracting index node attribute information; Vectorizing the index name and index definition in the index node attribute information to form an index vector corresponding to the index of the index node; Calculate the similarity S1 between the index vectors of each index, and set a second similarity threshold θ2. For index pairs with similarity S1 higher than θ2 but different responsibility departments, mark them as semantic conflict indexes and add them to the index semantic conflict list.

4. The knowledge graph-based semantic association and logical rule reasoning method according to claim 3, characterized in that, The construction of the semantic mapping cross-department knowledge graph includes: For each index pair in the index semantic conflict list, query the preset semantic mapping rule library to match the mapping rules; the semantic mapping rules are represented in the IF-THEN form; According to the matched mapping rules, add mapping relationship edges and mapping attributes between the two semantic conflict indexes in the index pair to form a cross-department semantic mapping layer; Integrate the cross-department semantic mapping layer into the cross-department knowledge graph to form the semantic mapping cross-department knowledge graph.

5. The knowledge graph-based semantic association and logical rule reasoning method according to claim 4, characterized in that, The second data set includes real-time business parameters and real-time environmental data of cross-functional departments; The obtained fused cross-department knowledge graph and semantic correlation matrix includes: Form real-time business feature vectors according to real-time business parameters of cross-functional departments; Form real-time environmental feature vectors according to real-time environmental data of cross-functional departments; Fuse real-time business feature vectors and real-time environmental feature vectors in the semantic mapping cross-department knowledge graph to form a fused cross-department knowledge graph; Based on the fused cross-department knowledge graph, calculate the semantic correlation between departments to generate an N×N semantic correlation matrix, where N is the number of departments.

6. The knowledge graph-based semantic association and logical rule inference method of claim 5, wherein, The generation of the cross-department collaborative knowledge graph includes: Extract the semantic correlation between departments from the semantic correlation matrix. For department pairs with semantic correlation higher than a preset correlation threshold θ3, use a first weight adjustment coefficient k1 to increase the weight of related nodes and edges, and the first weight adjustment coefficient k1 is in a positive correlation with the semantic correlation. For department pairs with semantic correlation less than or equal to the preset correlation threshold θ3, use a second weight adjustment coefficient k2 to reduce the weight of related nodes and edges, and the second weight adjustment coefficient k2 is in an inverse correlation with the semantic correlation.

7. The knowledge graph-based semantic association and logical rule reasoning method according to claim 6, characterized in that, Based on the cross-department collaborative knowledge graph, the generation of cross-department collaborative decision suggestions includes: Collect operation time series data of nodes in the cross-department collaborative knowledge graph, and obtain time series dependent features between nodes according to the operation time series data and a pre-constructed time series dependent feature model; Obtain semantic interaction features between nodes according to the nodes and edges of the cross-department collaborative knowledge graph and a pre-constructed semantic interaction feature model; Fuse the time series dependent features and the semantic interaction features to form a fused feature vector; Construct a logic rule base and establish a logic rule reasoning engine; Input the fusion feature vector into the logic rule reasoning engine, perform reasoning, and generate cross-departmental collaborative decision suggestions.

8. A knowledge graph based semantic association and logical rule reasoning system for implementing the knowledge graph based semantic association and logical rule reasoning method of any one of claims 1-7, characterized in that, The system comprises: A graph construction module, configured to acquire a first data set across functional departments, construct a structured data table according to the first data set, and construct a cross-departmental knowledge graph based on the structured data table; A graph optimization module, configured to extract index nodes from the cross-departmental knowledge graph, identify semantic conflict indexes, and generate an index semantic conflict list; construct a cross-departmental knowledge graph after semantic mapping based on the index semantic conflict list; acquire a second data set across functional departments, obtain a fused cross-departmental knowledge graph and a semantic correlation degree matrix based on the second data set and the cross-departmental knowledge graph after semantic mapping; extract weights of each node and edge of the fused cross-departmental knowledge graph, dynamically adjust the weights of each node and edge of the fused cross-departmental knowledge graph according to the semantic correlation degree matrix, and generate a cross-departmental collaborative knowledge graph; A decision reasoning module, configured to generate cross-departmental collaborative decision suggestions based on the cross-departmental collaborative knowledge graph.

Images