A knowledge graph construction method, system, device and medium

Abstract

The application provides a knowledge graph construction method, system, device and medium, relates to the technical field of knowledge graph construction, and the method comprises the following steps: in response to a trigger event sent by a user end, acquiring basic information of an enterprise corresponding to the user end, combining the trigger event and the basic information, determining a plurality of characteristic data affecting the result of the trigger event, and constructing a sub-knowledge graph of each characteristic data; acquiring an influence index of each sub-knowledge graph on the result of the trigger event and a correlation coefficient between the sub-knowledge graphs; combining the influence index and the correlation coefficient to construct an initial knowledge graph; acquiring an abnormal conduction edge in the initial knowledge graph; processing the correlation coefficient and the influence index carried by the abnormal conduction edge to generate a target knowledge graph; and generating a strategic decision suggestion of the trigger event according to the target knowledge graph. The application has the technical effect that the accuracy of the strategic decision suggestion generated by the knowledge graph is improved.

Description

Technical Field

[0001] This application relates to the field of knowledge graph construction technology, specifically to a knowledge graph construction method, system, device, and medium. Background Technology

[0002] As the business environment becomes increasingly complex, enterprises need to comprehensively consider information from multiple dimensions, including policies and regulations, market competition, technological trends, and cultural differences, when facing major decision-making scenarios such as entering overseas markets, adjusting supply chains, and making strategic investments. Knowledge graphs, as a structured knowledge representation method, can integrate multi-source heterogeneous data into a semantic network composed of entities and relationships, providing knowledge support for enterprise decision-making. However, how to efficiently construct knowledge graphs covering multi-dimensional feature data for specific decision-making events, and how to output strategic decision-making recommendations with practical guidance based on these knowledge graphs, is a pressing technical problem that needs to be solved in the field of intelligent enterprise decision-making.

[0003] Currently, knowledge graphs are constructed by collecting multi-dimensional feature data related to decision-making events and generating decision-making suggestions based on them. Although the above methods can establish the relationship between feature data, the accuracy of the impact assessment of decision-making events by feature data from different sources varies during the knowledge graph construction process. This difference is not effectively reflected and processed in the graph structure, resulting in insufficient accuracy of the strategic decision-making suggestions generated based on the knowledge graph. Summary of the Invention

[0004] This application provides a knowledge graph construction method, system, device, and medium to improve the accuracy of strategic decision-making recommendations generated from knowledge graphs.

[0005] Firstly, this application provides a knowledge graph construction method, the method comprising: responding to a trigger event sent by a user terminal, obtaining basic information of the enterprise corresponding to the user terminal; combining the trigger event and the basic information to determine multiple feature data affecting the result of the trigger event, and constructing a sub-knowledge graph for each feature data; obtaining the influence index of each sub-knowledge graph on the result of the trigger event, and the correlation coefficient between each sub-knowledge graph; using the trigger event as the master node and each sub-knowledge graph as a child node, and combining the influence index and the correlation coefficient to construct an initial knowledge graph; wherein, the influence index characterizes the degree of influence of each sub-knowledge graph on the result of the trigger event, and the correlation coefficient characterizes the degree of mutual influence between the sub-knowledge graphs; in the initial knowledge graph, the master node and each child node are connected by connecting edges, and each connecting edge... Each child node carries a corresponding influence index, and the child nodes are connected by association edges, each association edge carrying a corresponding association coefficient. The reliability index of the feature data corresponding to each child node in the initial knowledge graph is obtained. The reliability index difference between the two child nodes connected by each association edge is calculated. Association edges whose propagation direction is from a child node with a low reliability index to a child node with a high reliability index, and whose difference exceeds a preset threshold, are marked as abnormal propagation edges. The association coefficient carried by the abnormal propagation edges is attenuated to generate a target association coefficient. The influence index carried by the corresponding connecting edge is adjusted according to the reliability index to generate a target influence index. The initial knowledge graph is updated by combining the target association coefficient and the target influence index to generate a target knowledge graph. Based on the target knowledge graph, a strategic decision-making suggestion for the triggered event is generated, and the strategic decision-making suggestion is sent to the user terminal.

[0006] By adopting the above technical solution, basic enterprise information is obtained by responding to trigger events, and multiple feature data affecting the outcome of the trigger events are identified. Sub-knowledge graphs of each feature data are constructed. Then, the trigger event is used as the master node, and each sub-knowledge graph is used as a child node. An initial knowledge graph is constructed by combining the influence index and the correlation coefficient, realizing a structured knowledge representation for specific trigger events. Furthermore, by obtaining the reliability index of the feature data corresponding to each child node, the reliability index difference between two child nodes connected by the related edge is calculated. Related edges with a transmission direction from a child node with a low reliability index to a child node with a high reliability index and a difference exceeding a preset threshold are marked as abnormal transmission edges. The correlation coefficient carried by the abnormal transmission edges is attenuated to generate a target correlation coefficient. At the same time, the influence index carried by the connecting edge is adjusted according to the reliability index to generate a target influence index. Thus, inappropriate influence transmission caused by reliability differences between different feature data is effectively suppressed in the knowledge graph structure. The updated target knowledge graph can more accurately reflect the real influence relationship of each feature data on the outcome of the trigger event. The strategic decision-making suggestions generated based on this target knowledge graph have higher accuracy and credibility, improving the accuracy of strategic decision-making suggestions generated by the knowledge graph.

[0007] Secondly, this application provides a knowledge graph construction system, the system comprising: a response module, a first acquisition module, a second acquisition module, a first generation module, and a second generation module; wherein, The response module is used to respond to a trigger event sent by the user terminal, obtain basic information of the enterprise corresponding to the user terminal, and determine multiple feature data affecting the result of the trigger event by combining the trigger event and the basic information, and construct sub-knowledge graphs for each feature data; the first acquisition module is used to acquire the influence index of each sub-knowledge graph on the result of the trigger event, and the correlation coefficient between each sub-knowledge graph; using the trigger event as the master node and each sub-knowledge graph as the child node, and combining the influence index and the correlation coefficient, an initial knowledge graph is constructed; wherein, the influence index represents the degree of influence of each sub-knowledge graph on the result of the trigger event, and the correlation coefficient represents the degree of mutual influence between the sub-knowledge graphs; in the initial knowledge graph, the master node and each child node are connected by connecting edges, each connecting edge carries the corresponding influence index, and the child nodes are connected by... The initial knowledge graph is constructed by connecting related edges, each carrying a corresponding correlation coefficient. The second acquisition module acquires the reliability index of the feature data corresponding to each child node in the initial knowledge graph, calculates the reliability index difference between the two child nodes connected by each related edge, and marks related edges whose propagation direction is from a child node with a low reliability index to a child node with a high reliability index and whose difference exceeds a preset threshold as abnormal propagation edges. The first generation module attenuates the correlation coefficient carried by the abnormal propagation edges to generate a target correlation coefficient; adjusts the influence index carried by the corresponding connecting edge according to the reliability index to generate a target influence index; and updates the initial knowledge graph by combining the target correlation coefficient and the target influence index to generate a target knowledge graph. The second generation module generates strategic decision-making suggestions for the triggered event based on the target knowledge graph and sends the strategic decision-making suggestions to the user terminal.

[0008] Thirdly, this application provides an electronic device that adopts the following technical solution: it includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes a computer program of any of the above-described knowledge graph construction methods.

[0009] Fourthly, this application provides a computer-readable storage medium that stores a computer program capable of being loaded by a processor and executing any of the above-mentioned knowledge graph construction methods.

[0010] In summary, this application includes at least one of the following beneficial technical effects: By responding to triggering events, basic enterprise information is obtained, and multiple feature data affecting the outcome of the triggering events are identified. Sub-knowledge graphs for each feature data are constructed. Then, using the triggering event as the master node and each sub-knowledge graph as a child node, an initial knowledge graph is constructed by combining the influence index and the correlation coefficient, realizing a structured knowledge representation for specific triggering events. Furthermore, by obtaining the reliability index of the feature data corresponding to each child node, the reliability index difference between two child nodes connected by the related edge is calculated. Related edges whose transmission direction is from the child node with a low reliability index to the child node with a high reliability index and whose difference exceeds a preset threshold are marked as abnormal transmission edges. The correlation coefficient carried by the abnormal transmission edges is attenuated to generate a target correlation coefficient. At the same time, the influence index carried by the connecting edge is adjusted according to the reliability index to generate a target influence index. Thus, inappropriate influence transmission caused by reliability differences between different feature data is effectively suppressed in the knowledge graph structure. The updated target knowledge graph can more accurately reflect the real influence relationship of each feature data on the outcome of the triggering event. The strategic decision-making suggestions generated based on this target knowledge graph have higher accuracy and credibility, improving the accuracy of strategic decision-making suggestions generated by the knowledge graph. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating a knowledge graph construction method provided in an embodiment of this application; Figure 2 This is a schematic diagram of a knowledge graph construction system architecture provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a knowledge graph construction system provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0012] Explanation of reference numerals in the attached figures: 1000, electronic device; 1001, processor; 1002, communication bus; 1003, user interface; 1004, network interface; 1005, memory. Detailed Implementation

[0013] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0014] In the description of the embodiments in this application, words such as "illustrative," "for example," or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "illustrative," "for example," or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Rather, the use of words such as "illustrative," "for example," or "for example" is intended to present the relevant concepts in a specific manner.

[0015] Figure 1 This is a flowchart illustrating a knowledge graph construction method provided in an embodiment of this application. Figure 1 As shown, the method includes S101-S105: S101, in response to the trigger event sent by the user terminal, obtains the basic information of the enterprise corresponding to the user terminal, combines the trigger event and the basic information, determines multiple feature data that affect the result of the trigger event, and constructs a sub-knowledge graph of each feature data.

[0016] In real-world strategic consulting scenarios, companies face critical decisions requiring comprehensive consideration of external information from multiple dimensions. Traditional manual analysis methods, however, are prone to biased decision recommendations due to insufficient coverage or omissions in dimensions. To address this issue, this step automatically acquires basic company information in response to triggering events. Based on this, it identifies multi-dimensional feature data and constructs sub-knowledge graphs for each dimension, thus providing a structured data foundation for subsequent knowledge integration and decision analysis.

[0017] Specifically, when a user sends a trigger event to the server, the server first receives and parses it. The trigger event refers to a specific business event initiated by the user through the user's terminal that requires strategic analysis. For example, a company user inputting a consultation request to "assess the feasibility of a new energy vehicle export strategy in the Southeast Asian market" constitutes a trigger event. After receiving the trigger event, the server obtains the basic information of the company corresponding to the user's terminal. Basic information refers to structured descriptive data directly related to the company that initiated the trigger event, including the company's industry, operating scale, main business scope, main market distribution, and financial overview. For example, the basic information obtained by the server might be: "A new energy vehicle manufacturing company with annual revenue of 20 billion yuan, mainly producing pure electric passenger vehicles, with overseas business accounting for approximately 15%, primarily concentrated in the European market." The significance of obtaining basic information lies in the fact that even when facing the same type of trigger event, different companies will have significantly different analytical dimensions to focus on due to differences in their industry and business characteristics. Basic information can serve as a personalized anchor point for subsequent analysis.

[0018] After acquiring the triggering event and basic information, the server combines the two to determine multiple feature data that influence the outcome of the triggering event. Feature data refers to the set of data collected from different analytical dimensions that affects the final decision result of the triggering event. The server first uses natural language processing technology to semantically analyze the triggering event, extracting the event type and event keywords. The event type refers to the business category to which the triggering event belongs in a preset classification system, and the event keywords are the core indicative words extracted from the semantics of the triggering event. In the example above, the parsed event type is "overseas market expansion assessment," and the event keywords are "Southeast Asia," "new energy vehicles," and "export," etc. Subsequently, the server combines the event type, event keywords, and basic information to match multiple target feature dimensions from a preset feature dimension library. The feature dimension library is a pre-built knowledge base that stores the mapping relationship between different event types and feature dimensions. Target feature dimensions are analytical dimensions that are highly relevant to the current triggering event, determined through matching. In this example, these include industry environment dimension, market competition dimension, policy and regulation dimension, and supply chain dimension. The reason for analyzing from multiple dimensions is that the influencing factors of strategic decisions are multi-source and heterogeneous, and data from a single dimension cannot fully reflect the complex environment involved in the triggering event.

[0019] After determining the target feature dimensions, the server combines each target feature dimension with basic information to collect corresponding feature data from preset data sources. These preset data sources are multi-source data acquisition channels pre-configured by the server, including industry databases, policy information platforms, and market research report databases. During the collection process, basic information is used as a data filtering condition in the collection logic. For example, for the market competition dimension, feature data such as market share distribution and major brand rankings are collected using "pure electric passenger vehicles" and "Southeast Asia" as search conditions; for the policy and regulation dimension, feature data such as import tariff rates for new energy vehicles and localization production incentive policies in Southeast Asian countries are collected.

[0020] After completing the feature data collection, the server performs entity recognition and relation extraction on each feature data. Entity recognition refers to automatically identifying named entities with specific meanings from the feature data using named entity recognition technology, while relation extraction refers to automatically extracting the semantic relationships between named entities from the feature data. For example, for feature data in the policy and regulation dimension, the identified named entities include "Thailand," "import tariff rate 30%," "local assembly," and "tax rate halved to 15%," etc., and the extracted semantic relationships include "Thailand—implementation—new energy vehicle consumption tax reduction policy" and "local assembly—eligible—tax rate halved to 15%." The server constructs sub-knowledge graphs corresponding to each feature data, using each named entity as a graph node and each semantic relationship as a graph edge. A sub-knowledge graph is a local knowledge graph built around a single feature dimension, representing the network of relationships between entities in that dimension in the form of a graph structure. By constructing sub-knowledge graphs for each feature data, on the one hand, the knowledge relationships within each dimension are clearly expressed in the form of structured graphs, which facilitates subsequent deep semantic analysis based on graph computing. On the other hand, each sub-knowledge graph maintains the independence of the dimension level, providing a clear analytical unit for evaluating the influence of each dimension and the relationship between dimensions when integrating knowledge graphs in the future, thus avoiding the semantic ambiguity that may be caused by mixing all data to form a graph.

[0021] Based on the above embodiments, as an optional implementation, in S101, combining the triggering event and basic information, multiple feature data affecting the result of the triggering event are determined, and a sub-knowledge graph of each feature data is constructed, specifically including S11-S15: S11 performs semantic parsing on the triggering event, extracting the event type and event keywords.

[0022] After receiving the trigger event "Assess the strategic feasibility of Chinese new energy vehicle companies entering the Southeast Asian market," the server uses natural language processing (NLP) technology to perform syntactic analysis and semantic understanding of the text. Semantic parsing refers to the process of extracting structured information from the text content of the trigger event. Through semantic parsing, the server identifies the event type as "overseas market entry assessment." This event type is a general identifier of the business scenario category to which the trigger event belongs, representing the macro-level decision-making direction of the trigger event. Simultaneously, the server extracts the event keywords "new energy vehicles," "Southeast Asian market," and "strategic feasibility." These event keywords are key words or phrases extracted from the trigger event text that reflect the core content of the event, used to refine the matching scope of feature dimensions in subsequent steps. Through semantic parsing, the server transforms the natural language form of the trigger event into a structured event type and event keywords, providing a clear retrieval basis for accurate matching of target feature dimensions in subsequent steps.

[0023] S12, combining event type, event keywords and basic information, matches multiple target feature dimensions related to the triggering event from a preset feature dimension library; wherein, the feature dimension library pre-stores the mapping relationship between various event types and feature dimensions, and the target feature dimensions include at least two of the following: industry environment dimension, market competition dimension, financial status dimension, policy and regulation dimension and supply chain dimension.

[0024] The server combines the event type "overseas market entry assessment," event keywords "new energy vehicles," "Southeast Asian market," and "strategic feasibility," with basic information such as the company's main business being the R&D and sales of pure electric passenger vehicles, to match target feature dimensions from a pre-defined feature dimension library. This feature dimension library is a pre-built structured database stored on the server, containing mapping relationships between various event types and feature dimensions; that is, it predefines the combination of analytical dimensions to focus on for different types of events. Based on the event type, the server retrieves five candidate dimensions from the feature dimension library: industry environment, market competition, financial status, policy and regulations, and supply chain. Then, it performs relevance filtering based on event keywords and basic information, ultimately determining the target feature dimensions as four dimensions: policy and regulations, market competition, industry environment, and supply chain. The target feature dimensions are the set of feature dimensions most relevant to the currently triggering event, matched from the feature dimension library, including at least two of the following: industry environment, market competition, financial status, policy and regulations, and supply chain. The financial status dimension was not included due to its low relevance to the event keywords. By matching dimensions, the server identified the key analytical dimensions that needed to be focused on, providing a clear scope for the targeted collection of subsequent feature data.

[0025] S13, combining the feature dimensions and basic information of each target, collect the feature data corresponding to each target feature dimension from the preset data source.

[0026] Based on the specific content requirements of each target feature dimension and the enterprise attributes and areas of interest in the basic information, the server collects feature data corresponding to each target feature dimension from preset data sources. The preset data sources are a pre-configured set of external data interfaces covering various types of information sources, including government open data platforms, industry research databases, market research report platforms, and internal enterprise data interfaces. For the policy and regulatory dimension, the server collects feature data on tariff policies, consumption tax reduction policies, and localization production incentive policies for new energy vehicles in countries such as Thailand, Indonesia, and Malaysia from government open data platforms and international trade policy databases in Southeast Asian countries. For the market competition dimension, the server collects feature data on the market share distribution of pure electric passenger vehicles in major Southeast Asian countries, major competing brands, and their product layouts from market research report platforms. For the industry environment dimension, the server collects feature data on the overall development stage of the new energy vehicle industry in Southeast Asia, the progress of charging infrastructure construction, and consumer acceptance from industry research databases. For the supply chain dimension, the server collects feature data on the current distribution of component suppliers, the supply capacity of power battery raw materials, and logistics channel costs from internal enterprise data interfaces and third-party logistics evaluation platforms. Through targeted data collection, the server obtained the raw information from various dimensions that supports the subsequent construction of the knowledge graph.

[0027] S14, perform entity recognition and relation extraction on each feature data to generate an entity set and a relation set corresponding to each feature data; wherein, the entity set includes multiple named entities identified from the feature data, and the relation set includes the semantic relationships between the named entities.

[0028] The server uses named entity recognition (NER) and relation extraction (RE) technologies to process feature data for each target feature dimension. Entity recognition refers to the process of identifying named entities with specific meanings from unstructured feature data text, while relation extraction refers to the process of identifying semantic relationships between named entities from feature data text. After processing, an entity set and a relation set are generated for each feature data dimension. The entity set is the collection of all named entities identified from the feature data under a single target feature dimension. Named entities include information units with clear referential meaning, such as country names, policy names, company names, product categories, and indicator values. The relation set is the collection of semantic relationships between named entities, where semantic relationship refers to the type of association between two named entities at the semantic level. Taking the policy and regulation dimension as an example, the generated entity set includes named entities such as "Thailand," "new energy vehicle consumption tax reduction policy," "local assembly," "tax rate 15%," "Indonesia," and "imported vehicle tariff 50%." The relation set includes semantic relationships such as "Thailand—implementation—new energy vehicle consumption tax reduction policy," "local assembly—eligible—tax rate 15%," and "Indonesia—levy—imported vehicle tariff 50%." Taking the market competition dimension as an example, the generated entity set includes named entities such as "Southeast Asian pure electric passenger vehicle market," "Chinese brand A," "18% market share in Thailand," "Japanese brand B," and "hybrid vehicle layout." The relation set includes semantic relationships such as "Chinese brand A—occupies—18% market share in Thailand" and "Japanese brand B—accelerates layout—hybrid vehicle layout." Through entity recognition and relation extraction, the server completes the transformation of unstructured feature data into structured knowledge.

[0029] S15 uses named entities in the entity set as graph nodes and semantic relations in the relation set as graph edges to construct sub-knowledge graphs corresponding to each feature data.

[0030] The server creates named entities from the entity set corresponding to each target feature dimension as graph nodes in the sub-knowledge graph, and creates directed graph edges connecting the corresponding graph nodes from the semantic relations in the relation set. The direction and label of the graph edges correspond to the direction and type of the semantic relations. In the example above, the sub-knowledge graph for the policy and regulation dimension uses "Thailand," "new energy vehicle consumption tax reduction policy," "localized assembly," and "tax rate of 15%" as graph nodes, and semantic relations such as "implementation" and "eligible" as graph edges, forming a structured knowledge network reflecting the policy and regulatory system of new energy vehicles in Southeast Asian countries. The sub-knowledge graph for the market competition dimension uses "Southeast Asian pure electric passenger vehicle market," "Chinese brand A," and "Japanese brand B" as graph nodes, and semantic relations such as "occupy" and "accelerate layout" as graph edges, forming a structured knowledge network reflecting the competitive landscape of the Southeast Asian new energy vehicle market. The sub-knowledge graphs for the industry environment dimension and the supply chain dimension are constructed in the same way. Through graph construction, the structured knowledge of each target feature dimension is organized into a graph structure that facilitates subsequent calculation and reasoning, laying the foundation for the overall construction of the initial knowledge graph in S101.

[0031] S102, obtain the influence index of each sub-knowledge graph on the result of the triggering event, and the correlation coefficient between each sub-knowledge graph; using the triggering event as the main node and each sub-knowledge graph as the child node, combine the influence index and the correlation coefficient to construct an initial knowledge graph; wherein, the influence index represents the degree of influence of each sub-knowledge graph on the result of the triggering event, and the correlation coefficient represents the degree of mutual influence between the sub-knowledge graphs; in the initial knowledge graph, the main node and each child node are connected by connecting edges, each connecting edge carrying the corresponding influence index, and each child node is connected by association edges, each association edge carrying the corresponding correlation coefficient.

[0032] Specifically, the server first obtains the influence index of each sub-knowledge graph on the outcome of the triggered event. The influence index is a value between 0 and 1, representing the degree of influence of each sub-knowledge graph on the outcome of the triggered event; a higher value indicates a stronger influence of the feature dimension corresponding to that sub-knowledge graph on the decision result of the triggered event. The server calculates the influence index by performing graph feature analysis on each sub-knowledge graph, specifically evaluating factors such as the node size, relation density, and semantic relevance between entities in the sub-knowledge graph and the keywords of the triggered event. The influence index of each sub-knowledge graph is obtained by weighting and summing these factors. Using the example from S101, for the triggering event "assessing the feasibility of exporting new energy vehicles to the Southeast Asian market," the sub-knowledge graph of the policy and regulation dimension has a high semantic relevance to the triggering event keywords because it contains entities and relationships directly related to export decisions, such as tariff rates and market access policies. Therefore, the calculated impact index is 0.85. The sub-knowledge graph of the market competition dimension contains information such as market share and competitive landscape, with an impact index of 0.78. The sub-knowledge graph of the industry environment dimension reflects the overall industry trend, with an impact index of 0.65. The sub-knowledge graph of the supply chain dimension involves logistics and production support information, with an impact index of 0.60.

[0033] While calculating the influence index, the server also obtains the correlation coefficients between each sub-knowledge graph. The correlation coefficient is a numerical value characterizing the degree of mutual influence between any two sub-knowledge graphs, also ranging from 0 to 1. A higher value indicates a stronger mutual influence between the feature dimensions corresponding to the two sub-knowledge graphs. The server calculates the correlation coefficient by analyzing whether there are co-existing entities or semantically related entities between different sub-knowledge graphs. Specifically, it detects whether two sub-knowledge graphs contain the same named entities or semantically related named entity pairs, and combines the proportion of co-existing entities with the strength of semantic association to arrive at the correlation coefficient.

[0034] In the above example, the sub-knowledge graph of the policy and regulation dimension contains the entity "localized assembly," while the sub-knowledge graph of the supply chain dimension also involves the capacity layout and component supply information corresponding to "localized assembly." The two have many co-existing entities, so the correlation coefficient is 0.72. The policy and regulation dimension and the market competition dimension also have a certain semantic relationship because tariff policies indirectly affect product pricing and competitiveness, with a correlation coefficient of 0.55. The direct relationship between the industry environment dimension and the supply chain dimension is weak, with a correlation coefficient of 0.30.

[0035] After obtaining all influence indices and correlation coefficients, the server constructs an initial knowledge graph, using the triggering event as the master node and each sub-knowledge graph as its child nodes. The master node is the core node located at the center of the initial knowledge graph, representing the triggering event to be analyzed. The child nodes are feature-dimension nodes distributed around the master node, each carrying the complete graph structure of its corresponding sub-knowledge graph. The initial knowledge graph is a weighted knowledge graph with a two-layer structure. The upper layer is a macroscopic relationship network between the master node and each child node, as well as among the child nodes themselves; the lower layer is a microscopic entity relationship network of the sub-knowledge graphs within each child node.

[0036] In the initial knowledge graph, the main node and each child node are connected by connecting edges. These connecting edges are directed edges from the main node to the child node, and each edge carries the influence index of the corresponding sub-knowledge graph as its weight, representing the strength of the child node's influence on the triggering event's outcome. The child nodes are connected by association edges. These association edges are undirected edges between child nodes, and each association edge carries the corresponding association coefficient as its weight, representing the strength of the mutual influence between the feature dimensions represented by the two child nodes. In the example above, the constructed initial knowledge graph uses "Assessing the Feasibility of New Energy Vehicle Export Strategy in the Southeast Asian Market" as the main node. The main node is connected to four child nodes—Policy and Regulations, Market Competition, Industry Environment, and Supply Chain—by ​​connecting edges carrying influence indices of 0.85, 0.78, 0.65, and 0.60, respectively. The Policy and Regulations sub-node is connected to the Supply Chain sub-node by an association edge carrying an association coefficient of 0.72, and the Policy and Regulations sub-node is connected to the Market Competition sub-node by an association edge carrying an association coefficient of 0.55. The remaining child nodes are also connected by their respective association coefficients.

[0037] Based on the above embodiments, as an optional implementation, in S102, obtaining the influence index of each sub-knowledge graph on the triggering event result specifically includes S21-S23: S21, perform graph embedding processing on each sub-knowledge graph to generate graph embedding vectors corresponding to each sub-knowledge graph; wherein, the graph embedding vectors are used to represent the overall semantic features of the corresponding sub-knowledge graph.

[0038] The server utilizes a graph neural network to perform graph embedding processing on each sub-knowledge graph. This graph embedding process refers to the server aggregating and compressing the features of all graph nodes and edges in the sub-knowledge graph, mapping the structural and semantic information of the entire sub-knowledge graph into a dense vector of fixed dimension. The graph embedding vector, obtained after graph embedding processing, is a vector representation of the overall semantic features of the corresponding sub-knowledge graph. Its position and direction in the vector space reflect the core knowledge content and semantic theme carried by the sub-knowledge graph. During the graph embedding process, the graph neural network first performs initial vector encoding on the named entities of each graph node in the sub-knowledge graph. Then, through a message passing mechanism, each graph node aggregates the semantic information of its neighboring nodes along the graph edges. After multiple iterative aggregations, the vector representation of each graph node has incorporated its local structural features and contextual semantic information within the sub-knowledge graph. Finally, a global pooling operation is performed on the vector representations of all graph nodes to generate a graph embedding vector representing the entire sub-knowledge graph. Following the previous example, the server performs graph embedding processing on the sub-knowledge graph of the policy and regulation dimension. The structural and semantic information of graph nodes such as "Thailand," "new energy vehicle consumption tax reduction policy," "local assembly," and "15% tax rate," as well as graph edges such as "implementation" and "eligible," are aggregated and encoded into a graph embedding vector. This vector points to the semantic region in the semantic space related to policies and regulations and new energy vehicle trade. Similarly, the sub-knowledge graphs of the market competition, industry environment, and supply chain dimensions are each processed by graph embedding to generate their corresponding graph embedding vectors. Through graph embedding, each sub-knowledge graph is transformed from a graph structure into a vector form, laying the foundation for subsequent similarity calculations with triggering events in the same vector space.

[0039] S22, Semantically encode the triggering event to generate the event semantic vector corresponding to the triggering event.

[0040] The server uses a pre-trained language model to semantically encode the triggering event. This semantic encoding refers to the server's process of performing deep semantic understanding and vectorization of the textual content of the triggering event. The event semantic vector is a vector representation of the overall semantic content of the triggering event obtained after semantic encoding. Its dimension is consistent with the graph embedding vector, ensuring that both exist in the same vector space, thus enabling similarity calculation. In the example above, the server semantically encodes the triggering event "assessing the strategic feasibility of Chinese new energy vehicle companies entering the Southeast Asian market." The generated event semantic vector points in the semantic space towards semantic elements including market entry assessment, the new energy vehicle industry, and the Southeast Asian region. Through semantic encoding, the triggering event in natural language form is transformed into a vector representation in the same space as the graph embedding vector, providing a comparable numerical basis for subsequent semantic similarity calculations between the two.

[0041] S23, calculate the semantic similarity score between the embedding vector of each graph and the semantic vector of the event, and combine the semantic similarity score with the preset influence weight mapping table to determine the influence index of each sub-knowledge graph on the triggering event result.

[0042] The server calculates the semantic similarity score between each graph embedding vector and the event semantic vector using the cosine similarity method. The semantic similarity score measures the directional proximity between a single graph embedding vector and the event semantic vector in vector space, ranging from -1 to 1. A higher score indicates a closer similarity between the semantic content of the corresponding sub-knowledge graph and the semantic content of the triggering event, meaning a stronger correlation between that feature dimension and the triggering event. In the example above, the server calculates a semantic similarity score of 0.82 between the graph embedding vector for the policy and regulations dimension and the event semantic vector, 0.67 for the market competition dimension, 0.53 for the industry environment dimension, and 0.49 for the supply chain dimension. After obtaining the semantic similarity scores, the server uses a pre-defined influence weight mapping table to determine the influence index of each sub-knowledge graph on the triggering event result. The influence weight mapping table is a pre-defined segmented mapping rule table stored on the server, defining the correspondence between semantic similarity scores in different intervals and influence index values, used to convert continuous semantic similarity scores into influence indices that meet subsequent calculation requirements. The server maps the semantic similarity scores to corresponding influence indices using an influence weight mapping table. The resulting influence indices for the policy and regulations sub-knowledge graph on the triggering event are: 0.78 for the policy and regulations dimension, 0.59 for the market competition dimension, 0.44 for the industry environment dimension, and 0.42 for the supply chain dimension. The policy and regulations dimension has the highest influence index, indicating that in the current context of the triggering event, the new energy vehicle policies and regulations of Southeast Asian countries have the greatest impact on the feasibility assessment of market entry strategies. This aligns with the core concerns of cross-border market entry in the triggering event. Through semantic similarity calculation and influence weight mapping, the server quantifies the semantic association between each sub-knowledge graph and the triggering event into specific influence indices, providing a quantitative basis for assigning influence indices to the connecting edges in the initial knowledge graph and for constructing the target knowledge graph.

[0043] Based on the above embodiments, as an optional implementation, in S102, obtaining the correlation coefficients between each sub-knowledge graph specifically includes S24-S25: S24, calculate the entity overlap and relation overlap between any two sub-knowledge graphs respectively; where the entity overlap is the ratio between the number of named entities shared by the two sub-knowledge graphs and the total number of named entities in the two sub-knowledge graphs, and the relation overlap is the ratio between the number of semantic relations shared by the two sub-knowledge graphs and the total number of semantic relations in the two sub-knowledge graphs.

[0044] In S102, in addition to obtaining the influence index of each sub-knowledge graph on the trigger event result, the server also needs to obtain the correlation coefficient between each sub-knowledge graph to quantify the knowledge correlation strength between different feature dimensions. Each sub-knowledge graph carries structured knowledge under different target feature dimensions. When two sub-knowledge graphs have the same named entities or semantic relationships, it indicates that the corresponding two feature dimensions have overlap and correlation at the knowledge level. This correlation needs to be characterized by quantitative calculation. Therefore, the server calculates the entity overlap and relation overlap between any two sub-knowledge graphs. The entity overlap is the ratio between the number of named entities shared by the two sub-knowledge graphs and the total number of named entities in the two sub-knowledge graphs, used to measure the degree of knowledge overlap between the two sub-knowledge graphs at the entity level. The relation overlap is the ratio between the number of semantic relationships shared by the two sub-knowledge graphs and the total number of semantic relationships in the two sub-knowledge graphs, used to measure the degree of knowledge overlap between the two sub-knowledge graphs at the relation level. Using the previous example, the server compares the sub-knowledge graphs of the policy and regulations dimension and the supply chain dimension. The policy and regulations sub-knowledge graph contains 12 named entities, including "Thailand," "new energy vehicle consumption tax reduction policy," "local assembly," and "tax rate of 15%." The supply chain sub-knowledge graph contains 10 named entities, including "power battery raw materials," "Thailand," "component suppliers," and "logistics channel costs." The two sub-knowledge graphs share two named entities: "Thailand" and "local assembly." Therefore, the entity overlap is 2 divided by 22, which is approximately 0.09. The policy and regulations sub-knowledge graph contains 15 semantic relations, and the supply chain sub-knowledge graph contains 12 semantic relations. The two share one semantic relation: "local assembly—involves—Thailand." Therefore, the relation overlap is 1 divided by 27, which is approximately 0.04. Similarly, the server calculates the entity overlap and relation overlap between any other two sub-knowledge graphs.

[0045] S25, perform weighted fusion of entity overlap and relation overlap to generate the association coefficient between each sub-knowledge graph; wherein, the weight of the weighted fusion is determined based on the event type of the triggering event.

[0046] After obtaining the entity overlap and relation overlap between each sub-knowledge graph, the server performs a weighted fusion of the two to generate a unified association coefficient. This weighted fusion refers to the process where the server assigns different fusion weights to entity overlap and relation overlap, and then sums the weighted values ​​to obtain a single value as the association coefficient. The weights of this weighted fusion are determined based on the event type of the triggering event. This is because the contribution of entity-level overlap and relation-level overlap to the association strength between feature dimensions differs under different event types. In the example above, the event type of the triggering event is "overseas market entry assessment." This type of event involves cross-border policy comparison and multi-market competition pattern analysis. It is common for different feature dimensions to share the same entities (such as the same country name or the same market entity), while sharing the same semantic relationships often implies a deeper level of knowledge overlap. Therefore, the server determines the fusion weight for entity overlap to be 0.4 and the fusion weight for relation overlap to be 0.6 based on this event type. That is, under this event type, the contribution of relation-level overlap to the association strength is higher than that of entity-level overlap. Taking the calculation of the correlation coefficient between the policy and regulatory dimension and the supply chain dimension as an example, the server multiplies the entity overlap (0.09) by the fusion weight (0.4) to obtain 0.036, and multiplies the relationship overlap (0.04) by the fusion weight (0.6) to obtain 0.024. The sum of the two results in a correlation coefficient of 0.06. This correlation coefficient is relatively low, indicating that the degree of overlap between the policy and regulatory dimension and the supply chain dimension at the knowledge level is weak.

[0047] Similarly, after weighting and fusing the market competition dimension and the industry environment dimension, the server obtained a correlation coefficient of 0.45, which is relatively high, indicating that there is a strong correlation between the two in terms of knowledge content in the analysis of the Southeast Asian new energy vehicle market.

[0048] Based on the above embodiments, as an optional implementation, in S102, using the triggering event as the main node and each sub-knowledge graph as the sub-node, and combining the influence index and the correlation coefficient, the initial knowledge graph is constructed, specifically including S26-S29: S26, encapsulate the triggering event into nodes to generate the main node corresponding to the triggering event; abstract and encapsulate each sub-knowledge graph and treat each sub-knowledge graph as a child node.

[0049] After obtaining the influence index of each sub-knowledge graph on the trigger event result and the correlation coefficient between the sub-knowledge graphs, the server needs to organize the trigger event and each sub-knowledge graph into a unified graph structure to construct the initial knowledge graph. This first requires completing the node-level construction. The server performs node-based encapsulation on the trigger event, generating the corresponding master node. This node-based encapsulation refers to the process by which the server encapsulates the text content, event type, event keywords, and event semantic vector of the trigger event into a graph node data structure. The master node is the core node representing the trigger event in the initial knowledge graph, located at the center of the graph structure, and serves as the hub connecting all sub-nodes. Simultaneously, the server performs abstract encapsulation on each sub-knowledge graph, treating each sub-knowledge graph as a child node. This abstract encapsulation refers to the process by which the server encapsulates the complete graph structure of the sub-knowledge graph and its corresponding graph embedding vector and target feature dimension identifier into a graph node data structure. The encapsulated child node exists as an independent node in the initial knowledge graph, while still retaining the complete sub-knowledge graph structure internally for subsequent reasoning. Using the previous example, the server generates the main node corresponding to the trigger event "Assess the strategic feasibility of Chinese new energy vehicle companies entering the Southeast Asian market", as well as four sub-nodes corresponding to the four sub-knowledge graphs of policy and regulation dimension, market competition dimension, industry environment dimension and supply chain dimension.

[0050] S27, normalize each influence index to generate a normalized influence index for each child node; establish connection edges between the main node and each child node, and use each normalized influence index as the edge weight of the corresponding connection edge; wherein, the direction of each connection edge is from the main node to the corresponding child node.

[0051] After node construction is complete, the server normalizes each influence index, generating a normalized influence index for each child node. This normalization process involves scaling the influence indices of each sub-knowledge graph proportionally according to a constraint that their sum is equal to one. This ensures that the relative magnitudes of the influence indices remain unchanged while meeting the normative requirements of weight allocation. The normalized influence index is used to characterize the proportion of importance of each child node relative to other child nodes in the initial knowledge graph. In the example above, the influence indices for the four dimensions of policy and regulations, market competition, industry environment, and supply chain are 0.78, 0.59, 0.44, and 0.42, respectively. After normalization, their corresponding normalized influence indices are approximately 0.35, 0.26, 0.20, and 0.19. The server then establishes connections between the main node and each child node, using the normalized influence index as the edge weights for the corresponding connections. The connecting edges are directed edges in the initial knowledge graph that connect the main node and the child node. Their direction is from the main node to the corresponding child node, representing the analytical direction relationship of the triggering event on each feature dimension. The edge weight is the corresponding normalized influence index. The larger the value, the more significant the influence of the feature dimension on the result of the triggering event.

[0052] S28. Traverse any combination of two child nodes. For each combination, determine whether the corresponding correlation coefficient is greater than the preset correlation threshold. If it is greater than the preset correlation threshold, establish a correlation edge between the two child nodes and use the correlation coefficient as the edge weight of the correlation edge. Based on the relationship between the normalized influence indices of the two child nodes, determine the propagation direction of the correlation edge as from the child node with the higher normalized influence index to the child node with the lower normalized influence index.

[0053] After establishing a connection edge, the server iterates through any combination of two child nodes and determines whether the corresponding correlation coefficient for each combination is greater than a preset correlation threshold. The preset correlation threshold is a pre-defined numerical limit used to filter effective correlations. Two child nodes with a correlation coefficient lower than this threshold are considered to have no significant knowledge correlation, and no correlation edge is established, thus avoiding noise interference from weak correlations in subsequent reasoning. In the example above, the preset correlation threshold is set to 0.1. The server iterates through six combinations of child nodes: policy and regulations and market competition, policy and regulations and industry environment, policy and regulations and supply chain, market competition and industry environment, market competition and supply chain, and industry environment and supply chain. The correlation coefficient between policy and regulations and supply chain is 0.06, which is less than the preset correlation threshold of 0.1, so no correlation edge is established. The correlation coefficient between market competition and industry environment is 0.45, which is greater than the preset correlation threshold, so the server establishes a correlation edge between them, using the correlation coefficient of 0.45 as the edge weight. The correlation edge is a directed edge connecting two child nodes in the initial knowledge graph, and its edge weight is the corresponding correlation coefficient, used to characterize the strength of the knowledge correlation between the two feature dimensions. The propagation direction of the associated edges is determined based on the relationship between the normalized influence indices of the corresponding two child nodes. The direction is from the child node with the higher normalized influence index to the child node with the lower normalized influence index. This is because feature dimensions with higher influence indices are more dominant in the analysis, and their knowledge content is more likely to have an analytical transmission effect on dimensions with lower influence indices. In the example above, the normalized influence index of the market competition dimension is 0.26, which is higher than the 0.20 of the industry environment dimension. Therefore, the propagation direction of this associated edge is from the market competition child node to the industry environment child node. Similarly, the server establishes associated edges and determines the propagation direction for other combinations of child nodes that meet the threshold conditions.

[0054] S29. Construct an initial knowledge graph based on the master node, each child node, each connecting edge, and each associated edge.

[0055] S103, obtain the reliability index of the feature data corresponding to each child node in the initial knowledge graph, calculate the reliability index difference between the two child nodes connected by each associated edge, and mark the associated edge whose transmission direction is from the child node with low reliability index to the child node with high reliability index and whose difference exceeds the preset threshold as an abnormal transmission edge.

[0056] After constructing the initial knowledge graph, cross-dimensional relationships have been established between the child nodes through connecting edges. The correlation coefficient on the connecting edge represents the degree of mutual influence between the two child nodes. However, in practical applications, the reliability of the feature data corresponding to each child node often varies. For example, policy and regulatory data from authoritative government platforms are usually more reliable than predictive data from third-party market research institutions. When there is a significant difference in reliability between the two child nodes connected by a connecting edge, if the child node with lower reliability transmits its influence to the child node with higher reliability through the connecting edge, it may cause low-quality data to interfere with the analysis results of high-quality data, thereby reducing the accuracy of the final decision-making recommendations. Therefore, this step evaluates the reliability index of the feature data of each child node, identifies and marks connecting edges that may cause abnormal transmission, and provides a basis for subsequent targeted optimization of the initial knowledge graph.

[0057] Specifically, the server first obtains the reliability index of the feature data corresponding to each child node in the initial knowledge graph. The reliability index is a numerical indicator used to quantify the trustworthiness of feature data; its value ranges from 0 to 1, with higher values ​​indicating stronger reliability. The server comprehensively evaluates the feature data corresponding to each child node from three aspects: data source authority, data timeliness, and data completeness to obtain the reliability index. Data source authority refers to the credibility of the source of the feature data within its respective field; for example, data from official government channels is more authoritative than data from commercial consulting firms. Data timeliness refers to the proximity of the content reflected by the feature data to the current time; the more recent the data, the higher its timeliness. Data completeness refers to the sufficiency of the feature data in terms of coverage and information granularity.

[0058] The server scores each of the three aspects separately and then sums the scores in a weighted manner to obtain the reliability index for each sub-node. Continuing with the previous example, the feature data for the policy and regulation dimension sub-node mainly comes from publicly released tariff policy documents from Southeast Asian governments, ensuring high data authority and clear content, resulting in a reliability index of 0.92. The feature data for the market competition dimension sub-node comes from reports from multiple market research institutions, indicating relatively high data authority, although there are some discrepancies between different sources, resulting in a reliability index of 0.75. The feature data for the industry environment dimension sub-node comes partly from annual reports of industry associations and partly from media analysis articles, resulting in a reliability index of 0.68. The feature data for the supply chain dimension sub-node comes from internal supplier information and third-party logistics assessment reports, resulting in a reliability index of 0.70.

[0059] After obtaining the reliability index of each child node, the server calculates the reliability index difference between the two child nodes connected by each associated edge. The reliability index difference is the absolute value of the difference between the reliability indices of two child nodes connected by the same associated edge, reflecting the magnitude of the difference in data reliability between the two child nodes. In the example above, the reliability index difference between the associated edge of the policy and regulations child node and the supply chain child node is 0.92 minus 0.70 equals 0.22; the reliability index difference between the associated edge of the policy and regulations child node and the market competition child node is 0.92 minus 0.75 equals 0.17; and the reliability index difference between the associated edge of the industry environment child node and the supply chain child node is 0.70 minus 0.68 equals 0.02.

[0060] Subsequently, the server performs anomaly propagation judgment on each associated edge. In the initial knowledge graph, each associated edge is undirected, but due to the existence of association coefficients, a propagation effect occurs between child nodes, meaning that the analysis result of one child node may affect the analysis result of another child node along the associated edge. The propagation direction refers to the actual direction of the influence transmission along the associated edge. The server marks associated edges whose propagation direction is from a child node with a low reliability index to a child node with a high reliability index, and whose reliability index difference exceeds a preset threshold, as abnormal propagation edges. The preset threshold is a pre-set critical value used to judge whether the reliability difference constitutes an abnormal risk; in this embodiment, the preset threshold is set to 0.15. An abnormal propagation edge refers to an associated edge that has the risk of improperly propagating low-reliability data to high-reliability data. In the above example, the connection edge between the policy and regulation sub-node and the supply chain sub-node includes a transmission direction from the supply chain sub-node (reliability index 0.70) to the policy and regulation sub-node (reliability index 0.92). The reliability index difference is 0.22, which exceeds the preset threshold of 0.15. Therefore, this connection edge is marked as an abnormal transmission edge. The connection edge between the policy and regulation sub-node and the market competition sub-node has a reliability index difference of 0.17 corresponding to the transmission direction from the market competition sub-node (reliability index 0.75) to the policy and regulation sub-node (reliability index 0.92), which also exceeds the preset threshold. It is also marked as an abnormal transmission edge. However, the connection edge between the industry environment sub-node and the supply chain sub-node has a reliability index difference of only 0.02, which does not exceed the preset threshold. Therefore, it is not marked as an abnormal transmission edge.

[0061] S104, attenuate the correlation coefficient carried by the abnormal propagation edge to generate the target correlation coefficient; adjust the influence index carried by the corresponding connection edge according to the reliability index to generate the target influence index; combine the target correlation coefficient and the target influence index to update the initial knowledge graph and generate the target knowledge graph.

[0062] Specifically, the server first performs attenuation processing on the correlation coefficients carried by each anomalous propagation edge to generate a target correlation coefficient. Attenuation processing refers to the process of reducing the original correlation coefficient of the edge according to a certain attenuation rule, based on the reliability index difference between the two child nodes connected by the anomalous propagation edge. The target correlation coefficient is the new correlation coefficient obtained after attenuation processing, used to replace the original correlation coefficient on the anomalous propagation edge. The attenuation rule used by the server is: the target correlation coefficient equals the original correlation coefficient multiplied by an attenuation factor. The attenuation factor equals 1 minus the normalized value of the ratio of the reliability index difference to a preset threshold. The value of the attenuation factor is limited to between 0 and 1; the larger the reliability index difference, the smaller the attenuation factor, and the lower the corresponding target correlation coefficient.

[0063] Continuing with the previous example, the abnormal transmission edge between the policy and regulation sub-node and the supply chain sub-node has an original correlation coefficient of 0.72 and a reliability index difference of 0.22. Based on the attenuation rule, the attenuation factor is calculated to be 0.63, therefore the target correlation coefficient is 0.72 multiplied by 0.63, which equals 0.45. Similarly, the abnormal transmission edge between the policy and regulation sub-node and the market competition sub-node has an original correlation coefficient of 0.55 and a reliability index difference of 0.17. The calculated attenuation factor is 0.79, and the target correlation coefficient is 0.55 multiplied by 0.79, which equals 0.43. For correlation edges not marked as abnormal transmission edges, their correlation coefficients remain unchanged and are directly used as the target correlation coefficient. For example, the correlation edge between the industry environment sub-node and the supply chain sub-node is not marked as an abnormal transmission edge, and its target correlation coefficient remains the original correlation coefficient of 0.30. Through this attenuation process, the correlation strength of abnormal transmission edges is reasonably weakened, preserving the objectively existing correlation between dimensions while effectively reducing the risk of low-reliability data excessively interfering with high-reliability data through correlation paths.

[0064] After generating the target correlation coefficient, the server further adjusts the influence index carried by the corresponding connection edges based on the reliability index of each child node to generate the target influence index. The target influence index is a new influence index obtained by weighting the reliability index, and it replaces the original influence index on the connection edges between the master node and each child node. The adjustment method is: the target influence index equals the original influence index multiplied by the reliability index of the corresponding child node. The logic behind this adjustment is that the original influence index only reflects the semantic correlation between each sub-knowledge graph and the triggering event, without considering the reliability of the feature data itself. Introducing the reliability index into the calculation of the influence index allows dimensions with higher data quality to receive relatively greater analytical weight in the final decision.

[0065] In the above examples, the original impact index of the policy and regulation sub-node is 0.85, the reliability index is 0.92, and the target impact index is 0.85 multiplied by 0.92, which equals 0.78; the original impact index of the market competition sub-node is 0.78, the reliability index is 0.75, and the target impact index is 0.78 multiplied by 0.75, which equals 0.59; the original impact index of the industry environment sub-node is 0.65, the reliability index is 0.68, and the target impact index is 0.65 multiplied by 0.68, which equals 0.44; and the original impact index of the supply chain sub-node is 0.60, the reliability index is 0.70, and the target impact index is 0.60 multiplied by 0.70, which equals 0.42.

[0066] After generating all target correlation coefficients and target influence indices, the server writes them into the corresponding correlation edges and connection edges in the initial knowledge graph, replacing the original correlation coefficients and influence indices, thereby updating the initial knowledge graph and generating the target knowledge graph. The target knowledge graph is the final knowledge graph after reliability dimension optimization. Its overall structure is consistent with the initial knowledge graph, still using triggering events as main nodes and each sub-knowledge graph as child nodes, but the weights on each connection edge and correlation edge have been corrected according to data reliability. In the example above, the target influence index of the connection edge between the main node and the policy and regulation sub-node in the target knowledge graph is 0.78, still the highest among all child nodes, indicating that the dominant position of this dimension in decision-making is maintained; while the target correlation coefficient of the connection edge between the policy and regulation sub-node and the supply chain sub-node has decreased from the original 0.72 to 0.45, effectively suppressing the excessive transmission of the analysis results of the policy and regulation dimension by the less reliable data in the supply chain dimension.

[0067] Based on the above embodiments, as an optional implementation, in S104, the correlation coefficient carried by the abnormal propagation edge is attenuated to generate a target correlation coefficient; the influence index carried by the corresponding connection edge is adjusted according to the reliability index to generate the target influence index, specifically including S41-S44: S41. For each abnormal propagation edge, obtain the reliability index difference between the two corresponding child nodes; calculate the attenuation factor corresponding to each abnormal propagation edge based on the reliability index difference and the preset threshold; wherein, the attenuation factor is the ratio of the preset threshold to the reliability index difference, and the reliability index difference is inversely proportional to the attenuation factor.

[0068] In S103, the server has identified anomalous propagation edges in the initial knowledge graph. In S104, the association coefficients carried by these anomalous propagation edges need to be attenuated. The server first obtains the reliability index difference between the two child nodes corresponding to each anomalous propagation edge. The reliability index difference refers to the absolute difference between the reliability indices of the two child nodes connected by the anomalous propagation edge, used to quantify the degree of deviation between the two feature dimensions at the data reliability level. The larger the reliability index difference, the more significant the difference in reliability between the child nodes at both ends of the anomalous propagation edge, the lower the credibility of the propagation relationship, and the stronger the attenuation required. Subsequently, the server calculates the attenuation factor corresponding to each anomalous propagation edge based on the reliability index difference and a preset threshold. The preset threshold is a pre-set upper limit for the reliability index difference, serving as a reference benchmark for calculating the attenuation factor. The attenuation factor is the ratio of the preset threshold to the reliability index difference; the reliability index difference and the attenuation factor are inversely proportional, meaning the larger the reliability index difference, the smaller the attenuation factor, and the stronger the subsequent attenuation of the association coefficient. Using the previous example, assuming that the connection edge between the market competition sub-node and the industry environment sub-node is identified as an abnormal propagation edge, the reliability index of the market competition sub-node is 0.85, and the reliability index of the industry environment sub-node is 0.52, then the reliability index difference is 0.33. The preset threshold is set to 0.2, and the attenuation factor is 0.2 divided by 0.33, which is approximately 0.61.

[0069] S42, multiply the correlation coefficient carried by each abnormal propagation edge by the corresponding attenuation factor to generate the target correlation coefficient of each abnormal propagation edge.

[0070] After obtaining the attenuation factor corresponding to each anomalous transmission edge, the server multiplies the correlation coefficient carried by each anomalous transmission edge with the corresponding attenuation factor to generate the target correlation coefficient for each anomalous transmission edge. The target correlation coefficient is the attenuated correlation coefficient, used to replace the original correlation coefficient of the anomalous transmission edge in the initial knowledge graph. Its value is lower than the original correlation coefficient, reflecting the actual effective knowledge correlation strength between the two feature dimensions after considering reliability bias. In the example above, the original correlation coefficient of the anomalous transmission edge between the market competition sub-node and the industry environment sub-node is 0.45, the attenuation factor is 0.61, and multiplying them yields a target correlation coefficient of approximately 0.27. The target correlation coefficient is significantly lower than the original correlation coefficient, indicating that because the data reliability of the industry environment dimension is significantly lower than that of the market competition dimension, the knowledge transmission strength between the two is correspondingly weakened, thereby inhibiting the diffusion of uncertain information from the low-reliability dimension to the high-reliability dimension through the correlation edge.

[0071] S43. For each connection edge in the initial knowledge graph, obtain the reliability index of the corresponding child node; calculate the reliability adjustment coefficient of each child node based on the reliability index and the preset reliability benchmark value; wherein, the reliability adjustment coefficient is the ratio of the reliability index to the reliability benchmark value; when the reliability index is greater than or equal to the reliability benchmark value, the reliability adjustment coefficient is 1, and when the reliability index is less than the reliability benchmark value, the reliability adjustment coefficient is the ratio of the reliability index to the reliability benchmark value.

[0072] After attenuating the abnormal propagation edges, the server further adjusts the reliability of the influence index carried by each connection edge in the initial knowledge graph. The server obtains the reliability index of the corresponding child node for each connection edge and calculates the reliability adjustment coefficient for each child node based on the reliability index and a preset reliability benchmark value. The reliability benchmark value is a pre-set reliability index reference standard used to determine whether the data reliability of the child node reaches a qualified level. The reliability adjustment coefficient is a scaling factor used to adjust the influence index on the connection edge. Its value logic is as follows: when the reliability index is greater than or equal to the reliability benchmark value, it indicates that the data reliability of the child node has reached a qualified level, and the reliability adjustment coefficient is 1, meaning the influence index of the corresponding connection edge is not reduced; when the reliability index is less than the reliability benchmark value, the reliability adjustment coefficient is the ratio of the reliability index to the reliability benchmark value, and the value is less than 1, meaning the influence index of the corresponding connection edge will be reduced proportionally. In the above example, the preset reliability benchmark value is 0.7. The reliability index of the policy and regulation sub-node is 0.90, which is greater than 0.7, and its reliability adjustment coefficient is 1. The reliability index of the market competition sub-node is 0.85, which is greater than 0.7, and its reliability adjustment coefficient is 1. The reliability index of the industry environment sub-node is 0.52, which is less than 0.7, and its reliability adjustment coefficient is 0.52 divided by 0.7, which is approximately 0.74. The reliability index of the supply chain sub-node is 0.61, which is less than 0.7, and its reliability adjustment coefficient is 0.61 divided by 0.7, which is approximately 0.87.

[0073] S44. Multiply the impact index carried by each connecting edge by the reliability adjustment coefficient of the corresponding child node to generate the target impact index corresponding to each connecting edge.

[0074] S105: Based on the target knowledge graph, generate strategic decision-making suggestions for the triggering event and send the strategic decision-making suggestions to the user terminal.

[0075] Specifically, the server first prioritizes each sub-node based on the target influence index carried by each connection edge in the target knowledge graph, determining the analytical weight of each feature dimension in decision-making reasoning. Sub-nodes with higher target influence indices are assigned higher analytical priority during the reasoning process, and the knowledge carried by their internal sub-knowledge graphs accounts for a larger proportion in the final recommendation. Continuing with the previous example, the policy and regulations sub-node has the highest target influence index of 0.78 among all sub-nodes, and is therefore identified as the primary analytical dimension; the market competition sub-node has a target influence index of 0.59, making it a secondary analytical dimension; and the industry environment and supply chain sub-nodes have target influence indices of 0.44 and 0.42, respectively, serving as auxiliary analytical dimensions.

[0076] After determining the analysis priorities, the server performs knowledge reasoning on the sub-knowledge graphs within each sub-node. Knowledge reasoning refers to the process by which the server extracts key conclusions based on entities and semantic relationships in the sub-knowledge graphs through graph traversal and path analysis. Starting from the sub-node with the highest target influence index, the server sequentially traverses the core entities and key relationship paths in each sub-knowledge graph, extracting core conclusions under each dimension. In the example above, the server reasons on the sub-knowledge graph of the policy and regulations sub-node, extracting the core conclusion "Thailand's current policy environment provides significant tax incentives for the localization of new energy vehicle production" along relationship paths such as "Thailand—implementation—new energy vehicle consumption tax reduction policy" and "local assembly—eligible—tax rate halved to 15%"; and reasons on the sub-knowledge graph of the market competition sub-node, extracting the core conclusion "The current competitive landscape of the Southeast Asian pure electric passenger vehicle market is not yet solidified, and Chinese brands have already occupied a certain first-mover advantage but face competitive pressure from the accelerated expansion of Japanese brands."

[0077] After completing independent reasoning for each sub-node, the server further utilizes the target correlation coefficient on the associated edges to perform cross-dimensional joint reasoning. Joint reasoning refers to the process by which the server identifies the synergistic or restrictive relationships between conclusions of various dimensions based on the target correlation coefficient, and performs cross-analysis on the core conclusions of multiple dimensions to generate a comprehensive judgment. The higher the target correlation coefficient, the stronger the cross-analysis correlation is assigned to the core conclusions between two sub-nodes. In the example above, the target correlation coefficient between the policy and regulation sub-node and the supply chain sub-node is 0.45. Based on this, the server performs cross-analysis on the "tax incentives for localized production" conclusion of the policy and regulation dimension and the capacity layout information of the supply chain dimension, and arrives at the joint reasoning conclusion that "although Thailand's tax incentive policy is conducive to localized production layout, the current supply chain support capabilities of enterprises in Southeast Asia are not yet perfect, and the feasibility of full localization in the short term is limited." Since the associated edge has been attenuated, the impact of relatively unreliable data in the supply chain dimension on this joint conclusion has been controlled within a reasonable range, avoiding excessive interference from the highly uncertain supply chain assessment on the clear conclusion of the policy and regulation dimension.

[0078] After completing independent reasoning across dimensions and joint reasoning across dimensions, the server weights and integrates all reasoning conclusions according to the analytical weights corresponding to the target impact index, generating strategic decision-making recommendations for the triggering event. These strategic decision-making recommendations are structured, user-oriented decision-making guidance solutions formed by the server based on all reasoning results from the target knowledge graph. They include strategic direction suggestions, key risk warnings, and implementation path suggestions. In the example above, the server-generated strategic decision-making recommendation is: "It is recommended that the company prioritize entering the Thai and Indonesian markets through complete vehicle exports, utilizing the current tariff preference window to establish brand awareness and channel networks; in the medium term, it can explore localized assembly models in Thailand to further reduce tax costs, but it needs to first improve its Southeast Asian supply chain support system; at the same time, it needs to closely monitor the dynamics of Japanese brands' new energy transformation in the Southeast Asian market and adjust its competitive strategies accordingly." This strategic decision-making recommendation prioritizes the conclusions from the policy and regulatory dimensions and the market competition dimension in terms of content weighting, reflecting the core role of dimensions with higher target impact indices in recommendation generation. It also incorporates auxiliary judgments from the industry environment and supply chain dimensions to ensure the comprehensiveness of the recommendation content.

[0079] Finally, the server sends the generated strategic decision-making recommendations to the user's end. Upon receiving the recommendations, the user presents them to the user for strategic decision-making reference. Through this process, the multi-dimensional knowledge stored in the target knowledge graph in a graph structure is transformed into strategic recommendations that users can directly understand and adopt. The entire process forms a complete closed loop, from receiving triggering events, collecting multi-dimensional feature data and constructing sub-knowledge graphs, quantifying influence indices and correlation coefficients, assessing reliability and suppressing abnormal transmission, to the final knowledge reasoning and recommendation generation. This ensures the comprehensive advantages of the strategic decision-making recommendations in terms of multi-dimensional coverage, controllable data quality, and cross-dimensional collaborative analysis.

[0080] Based on the target knowledge graph, generate strategic decision-making suggestions for triggering events, including: The algorithm iterates through the target knowledge graph, traversing the target influence index corresponding to each child node and the target correlation coefficient between each child node. For each child node, it performs a weighted sum of the corresponding target influence index and the indirect influence contribution value of other child nodes transmitted through the target correlation coefficient to calculate the comprehensive influence weight of each child node on the trigger event result. The indirect influence contribution value is the sum of the products of the target influence index and the corresponding target correlation coefficient of other child nodes connected to the current child node through the correlation edge. The algorithm sorts the child nodes in descending order according to their comprehensive influence weights and selects the child nodes with the first preset number of positions as key influence nodes. It extracts the key entities and key relationships in the sub-knowledge graph corresponding to each key influence node. Combining the event type of the trigger event, the comprehensive influence weight of each key influence node, key entities, and key relationships, it matches the decision suggestion template corresponding to the event type from the preset decision template library, fills the key entities, key relationships, and comprehensive influence weights into the decision suggestion template, and generates strategic decision suggestions corresponding to the trigger event.

[0081] After the target knowledge graph is constructed, the server iterates through the target influence index corresponding to each sub-node in the target knowledge graph and the target correlation coefficient between each sub-node as the data input for subsequent calculations.

[0082] For each child node, the server calculates the comprehensive impact weight of each child node on the trigger event outcome by weighted summing of its corresponding target impact index and the indirect impact contribution value of other child nodes transmitted through the target correlation coefficient. The indirect impact contribution value is the sum of the products of the target impact index and the corresponding target correlation coefficient of other child nodes connected to the current child node through the correlation edge. The comprehensive impact weight is the sum of the child node's own target impact index and indirect impact contribution value, used to characterize the overall impact of this feature dimension on the trigger event outcome after considering both direct and cross-dimensional transmission effects. Using the previous example, the target impact index of the policy and regulations child node is 0.35, its indirect impact contribution value received through the correlation edge is approximately 0.07, and its comprehensive impact weight is approximately 0.42; similarly, the server calculates the comprehensive impact weight for each of the remaining child nodes.

[0083] The server then sorts the child nodes in descending order based on their overall influence weight, selecting the top-ranked child nodes (within a predetermined number of positions) as key influence nodes. These key influence nodes are the child nodes with the highest overall influence weight, representing the core feature dimension that has the most significant impact on the outcome of the triggered event. In the example above, the predetermined number is 2, and the policy and regulation child node and the market competition child node are selected as key influence nodes. The server further extracts key entities and key relationships from the sub-knowledge graph corresponding to each key influence node. The key entity is the named entity with the highest connectivity or the largest semantic weight in the sub-knowledge graph, and the key relationship is the core semantic relationship connecting the key entities.

[0084] Finally, the server combines the event type of the triggering event, the comprehensive impact weight of each key influencing node, key entities, and key relationships to match a decision suggestion template corresponding to the event type from a pre-set decision template library. The server then fills the template with the aforementioned information to generate strategic decision suggestions. The decision template library is a collection of structured templates designed for different event types and pre-stored on the server. Each template has reserved spaces for key entities, key relationships, and comprehensive impact weights. In the example above, the server matches a decision suggestion template of the "Overseas Market Entry Assessment" type, fills the template with key entities, key relationships, and comprehensive impact weights from the policy, regulation, and market competition dimensions, and generates strategic decision suggestions. For example, it suggests prioritizing Thailand's new energy vehicle consumption tax reduction policy and entering the market through localized assembly to obtain tax benefits, thus achieving a complete closed loop from knowledge graph to decision output.

[0085] like Figure 2 As shown, Figure 2This is a schematic diagram of a knowledge graph construction system architecture provided in an embodiment of this application; it includes a user terminal and a server. The user terminal sends a trigger event to the server. After receiving the trigger event, the server sequentially completes the construction of the knowledge graph and the generation of strategic decision-making suggestions through five core processing modules. The server first obtains the basic information of the enterprise corresponding to the user terminal through the feature data determination and sub-knowledge graph construction module in S101. Combining the trigger event and the basic information, it determines multiple feature data that affect the result of the trigger event, and performs entity recognition and relationship extraction on each feature data to construct the sub-knowledge graph corresponding to each feature data.

[0086] Subsequently, the server uses the S102 initial knowledge graph construction module to obtain the influence index of each sub-knowledge graph on the result of the triggering event and the correlation coefficient between each sub-knowledge graph. Taking the triggering event as the main node and each sub-knowledge graph as the child node, the main node and each child node are connected by connecting edges carrying the influence index, and each child node is connected by connecting edges carrying the correlation coefficient, thus generating the initial knowledge graph.

[0087] Next, the server uses the S103 abnormal propagation edge identification module to obtain the reliability index of the feature data corresponding to each child node, calculates the reliability index difference between the two child nodes connected by each associated edge, and marks the associated edge whose propagation direction is from the child node with low reliability index to the child node with high reliability index and whose difference exceeds the preset threshold as an abnormal propagation edge.

[0088] The server then optimizes and adjusts the initial knowledge graph through the S104 target knowledge graph generation module. This module contains two parallel processing sub-modules. The abnormal propagation edge attenuation processing sub-module calculates the attenuation factor based on the reliability index difference and a preset threshold, and multiplies the correlation coefficient carried by the abnormal propagation edge with the attenuation factor to generate the target correlation coefficient. The connection edge influence index reliability adjustment sub-module calculates the reliability adjustment coefficient based on the reliability index and the reliability benchmark value, and multiplies the influence index carried by each connection edge with the corresponding reliability adjustment coefficient to generate the target influence index. The processing results of the two sub-modules are combined to update the initial knowledge graph and generate the target knowledge graph.

[0089] Finally, the server uses the S105 strategic decision-making suggestion generation module to calculate the comprehensive influence weight of each sub-node based on the target influence index of each sub-node in the target knowledge graph and the target correlation coefficient between each sub-node. The sub-nodes ranked in descending order of comprehensive influence weight are selected as key influence nodes. The key entities and key relationships in the corresponding sub-knowledge graph of each key influence node are extracted. The server then matches the corresponding decision suggestion template from the preset decision template library with the event type of the triggering event, comprehensive influence weight, key entities and key relationships to generate strategic decision suggestions corresponding to the triggering event. Finally, the strategic decision suggestions are sent back to the user terminal.

[0090] Based on the above method, this application also discloses a knowledge graph construction system, such as... Figure 3 As shown, Figure 3 This is a schematic diagram of the structure of a knowledge graph construction system provided in an embodiment of this application. The system includes: a response module, a first acquisition module, a second acquisition module, a first generation module, and a second generation module; wherein, The response module is used to respond to trigger events sent by the user terminal, obtain the basic information of the enterprise corresponding to the user terminal, and combine the trigger event and the basic information to determine multiple feature data affecting the result of the trigger event, and construct sub-knowledge graphs for each feature data. The first acquisition module is used to acquire the influence index of each sub-knowledge graph on the result of the trigger event, and the correlation coefficient between each sub-knowledge graph. Using the trigger event as the master node and each sub-knowledge graph as the child node, the initial knowledge graph is constructed by combining the influence index and the correlation coefficient. Among them, the influence index represents the degree of influence of each sub-knowledge graph on the result of the trigger event, and the correlation coefficient represents the degree of mutual influence between the sub-knowledge graphs. In the initial knowledge graph, the master node and each child node are connected by connecting edges, each connecting edge carrying the corresponding influence index, and each child node is connected by association edges. The first module carries the corresponding association coefficients for each associated edge. The second module is used to acquire the reliability index of the feature data corresponding to each child node in the initial knowledge graph, calculate the reliability index difference between the two child nodes connected by each associated edge, and mark the associated edges whose transmission direction is from the child node with low reliability index to the child node with high reliability index and whose difference exceeds a preset threshold as abnormal transmission edges. The first generation module is used to attenuate the association coefficients carried by the abnormal transmission edges to generate target association coefficients. The second generation module is used to adjust the influence index carried by the corresponding connecting edge according to the reliability index to generate target influence index. The first generation module combines the target association coefficient and the target influence index to update the initial knowledge graph and generate the target knowledge graph. The third generation module is used to generate strategic decision-making suggestions for triggering events based on the target knowledge graph and send the strategic decision-making suggestions to the user terminal.

[0091] It should be noted that the system provided in the above embodiments is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0092] Please see Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4As shown, the electronic device 1000 may include: at least one processor 1001, at least one network interface 1004, a user interface 1003, a memory 1005, and at least one communication bus 1002.

[0093] The communication bus 1002 is used to realize the connection and communication between these components.

[0094] The user interface 1003 may include a display screen and a camera. Optionally, the user interface 1003 may also include a standard wired interface and a wireless interface.

[0095] The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0096] The processor 1001 may include one or more processing cores. The processor 1001 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 1005, and by calling data stored in the memory 1005. Optionally, the processor 1001 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 1001 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content to be displayed on the screen; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 1001 and may be implemented as a separate chip.

[0097] The memory 1005 may include random access memory (RAM) or read-only memory. Optionally, the memory 1005 may include a non-transitory computer-readable storage medium. The memory 1005 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 1005 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 1005 may also be at least one storage device located remotely from the aforementioned processor 1001. Figure 4 As shown, the memory 1005, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for a knowledge graph construction method.

[0098] exist Figure 4 In the electronic device 1000 shown, the user interface 1003 is mainly used to provide an input interface for the user and obtain the user input data; while the processor 1001 can be used to call an application program stored in the memory 1005 that provides a knowledge graph construction method. When executed by one or more processors, the electronic device performs one or more of the methods described in the above embodiments.

[0099] An electronic device readable storage medium stores instructions that, when executed by one or more processors, cause the electronic device to perform one or more of the methods described in the above embodiments.

[0100] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0101] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0102] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some service interfaces; indirect couplings or communication connections between devices or units may be electrical or other forms.

[0103] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0104] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0105] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0106] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Other embodiments of this disclosure will be readily apparent to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for constructing a knowledge graph, characterized in that, The method includes: In response to a trigger event sent by a user, the system obtains the basic information of the enterprise corresponding to the user, combines the trigger event and the basic information to determine multiple feature data that affect the result of the trigger event, and constructs a sub-knowledge graph for each feature data. Obtain the influence index of each sub-knowledge graph on the result of the triggering event, and the correlation coefficient between each sub-knowledge graph; construct an initial knowledge graph by using the triggering event as the main node and each sub-knowledge graph as a child node, combining the influence index and the correlation coefficient; wherein, the influence index represents the degree of influence of each sub-knowledge graph on the result of the triggering event, and the correlation coefficient represents the degree of mutual influence between the sub-knowledge graphs; in the initial knowledge graph, the main node and each child node are connected by connecting edges, each connecting edge carrying the corresponding influence index, and each child node is connected by association edges, each association edge carrying the corresponding correlation coefficient; Obtain the reliability index of the feature data corresponding to each child node in the initial knowledge graph, calculate the reliability index difference between the two child nodes connected by each associated edge, and mark the associated edge whose transmission direction is from the child node with low reliability index to the child node with high reliability index and whose difference exceeds a preset threshold as an abnormal transmission edge. The correlation coefficient carried by the abnormal propagation edge is attenuated to generate a target correlation coefficient; the influence index carried by the corresponding connection edge is adjusted according to the reliability index to generate a target influence index; the initial knowledge graph is updated by combining the target correlation coefficient and the target influence index to generate a target knowledge graph. Based on the target knowledge graph, strategic decision-making suggestions for the triggering event are generated, and the strategic decision-making suggestions are sent to the user terminal.

2. The knowledge graph construction method according to claim 1, characterized in that, The step of combining the triggering event and the basic information to determine multiple feature data affecting the outcome of the triggering event, and constructing a sub-knowledge graph for each feature data, includes: The triggering event is semantically parsed to extract its event type and keywords; Combining the event type, the event keywords, and the basic information, multiple target feature dimensions related to the triggering event are matched from a preset feature dimension library; wherein, the feature dimension library pre-stores mapping relationships between various event types and feature dimensions, and the target feature dimensions include at least two of the following: industry environment dimension, market competition dimension, financial status dimension, policy and regulation dimension, and supply chain dimension; Combining the target feature dimensions and the basic information, feature data corresponding to each target feature dimension is collected from a preset data source; Entity recognition and relation extraction are performed on each of the feature data to generate an entity set and a relation set corresponding to each feature data; wherein, the entity set includes multiple named entities identified from the feature data, and the relation set includes the semantic relationships between the named entities; Using named entities in the entity set as graph nodes and semantic relations in the relation set as graph edges, a sub-knowledge graph corresponding to each feature data is constructed.

3. The knowledge graph construction method according to claim 1, characterized in that, The step of obtaining the influence index of each of the sub-knowledge graphs on the result of the triggering event includes: Each of the sub-knowledge graphs is subjected to graph embedding processing to generate a graph embedding vector corresponding to each sub-knowledge graph; wherein, the graph embedding vector is used to represent the overall semantic features of the corresponding sub-knowledge graph; The triggering event is semantically encoded to generate an event semantic vector corresponding to the triggering event; Calculate the semantic similarity score between each graph embedding vector and the event semantic vector, and combine the semantic similarity score with a preset influence weight mapping table to determine the influence index of each sub-knowledge graph on the trigger event result.

4. The knowledge graph construction method according to claim 1, characterized in that, The step of obtaining the correlation coefficients between the sub-knowledge graphs includes: Calculate the entity overlap and relation overlap between any two of the sub-knowledge graphs respectively; wherein, the entity overlap is the ratio between the number of named entities shared by the two sub-knowledge graphs and the total number of named entities in the two sub-knowledge graphs, and the relation overlap is the ratio between the number of semantic relations shared by the two sub-knowledge graphs and the total number of semantic relations in the two sub-knowledge graphs. The entity overlap and the relationship overlap are weighted and fused to generate the association coefficient between each of the sub-knowledge graphs; wherein the weight of the weighted fusion is determined based on the event type of the triggering event.

5. The knowledge graph construction method according to claim 1, characterized in that, The process of constructing an initial knowledge graph, using the triggering event as the main node and each of the sub-knowledge graphs as sub-nodes, and combining the influence index and the correlation coefficient, includes: The triggering event is encapsulated into nodes to generate a main node corresponding to the triggering event; each of the sub-knowledge graphs is abstracted and encapsulated, and each of the sub-knowledge graphs is used as a child node. The influence indices are normalized to generate normalized influence indices for each child node; connection edges are established between the main node and each child node, and the normalized influence indices are used as the edge weights of the corresponding connection edges; wherein the direction of each connection edge is from the main node to the corresponding child node. Traverse any combination of two child nodes, and for each combination, determine whether the corresponding correlation coefficient is greater than a preset correlation threshold; if it is greater than the preset correlation threshold, establish a correlation edge between the two child nodes, and use the correlation coefficient as the edge weight of the correlation edge; based on the relationship between the normalized influence indices of the two child nodes, determine the propagation direction of the correlation edge as from the child node with the higher normalized influence index to the child node with the lower normalized influence index. An initial knowledge graph is constructed based on the master node, each of the child nodes, each of the connecting edges, and each of the associated edges.

6. The knowledge graph construction method according to claim 1, characterized in that, The process of attenuating the correlation coefficient carried by the abnormal transmission edge to generate a target correlation coefficient; and adjusting the impact index carried by the corresponding connection edge according to the reliability index to generate a target impact index, includes: For each of the abnormal propagation edges, the reliability index difference between the corresponding two child nodes is obtained; based on the reliability index difference and the preset threshold, the attenuation factor corresponding to each of the abnormal propagation edges is calculated; wherein, the attenuation factor is the ratio of the preset threshold to the reliability index difference, and the reliability index difference is inversely proportional to the attenuation factor; The correlation coefficient carried by each of the abnormal propagation edges is multiplied by the corresponding attenuation factor to generate the target correlation coefficient of each of the abnormal propagation edges. For each connection edge in the initial knowledge graph, the reliability index of the corresponding child node is obtained; based on the reliability index and a preset reliability benchmark value, the reliability adjustment coefficient corresponding to each child node is calculated; wherein, the reliability adjustment coefficient is the ratio of the reliability index to the reliability benchmark value; when the reliability index is greater than or equal to the reliability benchmark value, the reliability adjustment coefficient is 1; when the reliability index is less than the reliability benchmark value, the reliability adjustment coefficient is the ratio of the reliability index to the reliability benchmark value. The impact index carried by each of the connecting edges is multiplied by the reliability adjustment coefficient of the corresponding child node to generate the target impact index corresponding to each of the connecting edges.

7. The knowledge graph construction method according to claim 1, characterized in that, The step of generating strategic decision-making suggestions for the triggering event based on the target knowledge graph includes: Traverse the target influence index corresponding to each sub-node in the target knowledge graph and the target correlation coefficient between each sub-node; For each of the aforementioned child nodes, the corresponding target influence index is weighted and summed with the indirect influence contribution value of other child nodes transmitted through the target correlation coefficient to calculate the comprehensive influence weight of each of the aforementioned child nodes on the result of the triggering event; wherein, the indirect influence contribution value is the sum of the products of the target influence index and the corresponding target correlation coefficient of other child nodes connected to the current child node through the correlation edge; The sub-nodes are sorted in descending order according to their comprehensive influence weights, and the sub-nodes at the top of the sorted list are selected as key influence nodes. Extract the key entities and key relationships from the sub-knowledge graphs corresponding to each of the key influencing nodes; Combining the event type of the triggering event, the comprehensive influence weight of each key influencing node, the key entity, and the key relationship, a decision suggestion template corresponding to the event type is matched from a preset decision template library. The key entity, the key relationship, and the comprehensive influence weight are then filled into the decision suggestion template to generate a strategic decision suggestion corresponding to the triggering event.

8. A knowledge graph construction system, characterized in that, The system includes: a response module, a first acquisition module, a second acquisition module, a first generation module, and a second generation module; wherein, The response module is used to respond to a trigger event sent by the user terminal, obtain the basic information of the enterprise corresponding to the user terminal, combine the trigger event and the basic information, determine multiple feature data that affect the result of the trigger event, and construct a sub-knowledge graph of each feature data. The first acquisition module is used to acquire the influence index of each of the sub-knowledge graphs on the result of the triggering event, and the correlation coefficient between the sub-knowledge graphs; using the triggering event as the main node and each of the sub-knowledge graphs as child nodes, and combining the influence index and the correlation coefficient, an initial knowledge graph is constructed; wherein, the influence index represents the degree of influence of each of the sub-knowledge graphs on the result of the triggering event, and the correlation coefficient represents the degree of mutual influence between the sub-knowledge graphs; in the initial knowledge graph, the main node and each of the child nodes are connected by connecting edges, each connecting edge carrying the corresponding influence index, and each of the child nodes is connected by correlation edges, each correlation edge carrying the corresponding correlation coefficient; The second acquisition module is used to acquire the reliability index of the feature data corresponding to each child node in the initial knowledge graph, calculate the reliability index difference between the two child nodes connected by each associated edge, and mark the associated edge whose transmission direction is from the child node with low reliability index to the child node with high reliability index and whose difference exceeds a preset threshold as an abnormal transmission edge. The first generation module is used to attenuate the correlation coefficient carried by the abnormal propagation edge to generate a target correlation coefficient; adjust the influence index carried by the corresponding connection edge according to the reliability index to generate a target influence index; and update the initial knowledge graph by combining the target correlation coefficient and the target influence index to generate a target knowledge graph. The second generation module is used to generate strategic decision-making suggestions for the triggering event based on the target knowledge graph, and send the strategic decision-making suggestions to the user terminal.

9. An electronic device, characterized in that, The device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1-7.

Images