Intelligent supply chain risk prediction and decision optimization method based on knowledge graph

Abstract

The invention discloses an intelligent supply chain risk prediction and decision optimization method based on a knowledge graph, and belongs to the field of supply chain management. The intelligent supply chain risk prediction and decision optimization method based on the knowledge graph comprises supply chain knowledge graph construction, supply chain risk prediction model construction based on the knowledge graph and supply chain decision optimization based on the knowledge graph. The supply chain knowledge graph construction comprises data acquisition and preprocessing, knowledge graph construction and a dynamic updating mechanism. The construction of the supply chain risk prediction model based on the knowledge graph comprises feature engineering, model construction and model training and optimization. According to the method, the data integration and analysis capability can be improved, the risk early warning efficiency of the supply chain is improved, global optimization decision is realized, the quick response capability is enhanced, and support is provided for stable and efficient operation of the supply chain.

Description

Technical Field

[0001] The present invention belongs to the field of supply chain management, and specifically relates to an intelligent supply chain risk prediction and decision optimization method based on knowledge graph. Background Art

[0002] With the deepening of economic globalization, modern supply chain networks are becoming increasingly complex, characterized by a diverse set of participants, encompassing the entire process from raw material supply and product manufacturing to logistics and distribution, and finally, retail sales. Each link in the supply chain is strongly coupled and dynamically interconnected, forming a complex interactive network. Supply chain risks can have serious impacts on business operations, such as delivery delays, increased costs, and damaged reputations. Given the complexity and vulnerability of supply chain systems, establishing precise risk warning mechanisms and intelligent decision-making support systems is of great practical value.

[0003] Traditional approaches face significant challenges in addressing modern supply chain risks:

[0004] (1) Insufficient data processing capabilities: The supply chain contains a large amount of dispersed and heterogeneous data, including internal operational data (including production plans, inventory levels, equipment status, etc.), external environmental data (covering raw material market price fluctuations, terminal demand changes, industry competition situation, etc.), and partner data (such as supplier delivery on-time rate, logistics distribution information, etc.). Traditional methods have difficulty in comprehensively and effectively integrating and analyzing this massive and complex data, resulting in an inability to accurately grasp the overall status of the supply chain.

[0005] (2) Low risk prediction accuracy: Traditional prediction models are often based on simple statistical analysis or a single data source, ignoring the complex relationships between various links in the supply chain. For example, they only focus on the supplier's historical delivery data to predict supply risk, without considering the impact of factors such as the supplier's upstream raw material supply and industry policy changes on its delivery capacity. This leads to significant deviations in risk assessment results, making it difficult to provide timely and forward-looking early warning information, which weakens the company's risk response capabilities and decision-making timeliness.

[0006] (3) Decision-making lacks a global perspective: When making supply chain decisions, traditional methods usually start from a single enterprise or a single link, without fully considering the impact of the decision on the entire supply chain. For example, a company chooses to change suppliers to reduce costs, but does not consider the compatibility of the new supplier with the existing supply chain and the potential risks to subsequent production and distribution links. This results in a significant reduction in the collaborative efficiency of various links in the supply chain, undermining the overall stability of the system, and ultimately leading to a significant increase in the vulnerability and risk exposure of the supply chain network.

[0007] (4) Slow response: When risk events occur in the supply chain, traditional methods are unable to quickly obtain relevant information and make response decisions. Due to the delay in information transmission and the cumbersome decision-making process, companies often cannot take timely measures to reduce risk losses, causing risks to spread throughout the supply chain and cause greater impact.

[0008] Knowledge graph technology has achieved significant breakthroughs in recent years. Leveraging its powerful semantic representation capabilities and relational reasoning advantages, it has demonstrated tremendous technical potential and practical value across numerous application areas. Knowledge graphs can describe the relationships between entities in a structured manner, providing a powerful tool for processing complex supply chain data. However, research on applying knowledge graphs to supply chain risk prediction and decision optimization is still in its infancy. Current solutions still have significant limitations in key metrics such as data integration, risk identification accuracy, and decision support effectiveness, making it difficult to meet the demands for accurate prediction and intelligent decision-making in complex supply chain environments. Summary of the Invention

[0009] The purpose of this invention is to address the above-mentioned shortcomings and propose an intelligent supply chain risk prediction and decision optimization method based on knowledge graph. It focuses on three core links: supply chain knowledge graph construction, risk prediction model development and decision optimization method design, to achieve intelligent supply chain risk prediction and decision optimization based on knowledge graph.

[0010] The present invention specifically adopts the following technical solutions:

[0011] Intelligent supply chain risk prediction and decision optimization methods based on knowledge graphs, including: supply chain knowledge graph construction, supply chain risk prediction model construction based on knowledge graphs, and supply chain decision optimization based on knowledge graphs;

[0012] The construction of supply chain knowledge graph includes data collection and preprocessing, knowledge graph construction and dynamic update mechanism;

[0013] The construction of supply chain risk prediction models based on knowledge graphs includes feature engineering, model construction, and model training and optimization;

[0014] The specific process of special engineering is to use the topological structure and semantic attributes of the knowledge graph to extract multi-level risk prediction features, including entity attribute features, relationship features, and graph structure features. The Node2Vec graph embedding algorithm is used to capture the structural features of the knowledge graph through a biased random walk strategy, encoding nodes and relationships into dense low-dimensional vector representations. Given a node pair (u, v), where v is the central node and u is its neighbor node, the goal of Node2Vec is to maximize the following objective function:

[0015]

[0016] Where V is the set of all nodes in the graph, N(v) is the set of neighbor nodes of node v, which is generated by biased random walk and controlled by parameters o and q, and h v , h u : The low-dimensional vectors of nodes v and u represent the embedding vector, σSigmoid function, which maps the inner product to the probability interval [0, 1][0,1].

[0017] The specific process of model construction is as follows: using graph neural network (GNN) as the basic framework, through a multi-layer message passing mechanism, deeply mining the semantic associations and potential patterns of relationships between entities in the knowledge graph to achieve representation learning of complex knowledge networks; introducing an attention mechanism into the GNN model to dynamically adjust the weights of different nodes and relationships in risk prediction, highlighting the parts that have a significant impact on risk prediction; combining time series analysis methods to model the temporal characteristics of supply chain data and capture the changing trends of risks; and fusing GNN and time series models to construct a risk prediction model that can simultaneously consider graph structural features and temporal characteristics.

[0018] Model training and optimization utilize a k-fold cross-validation strategy, ensuring the stability and reliability of model evaluation results through multiple partitioning and validation of the dataset. To mitigate the risk of overfitting, a combination of L1 / L2 regularization constraints and Dropout random loss techniques are employed to enhance the model's generalization capabilities. The optimization process utilizes the Adaptive Moment Estimation (Adam) algorithm, which dynamically adjusts the learning rate and momentum parameters to achieve efficient convergence of the loss function, thereby obtaining the optimal model parameter configuration and significantly improving prediction accuracy and robustness.

[0019] Supply chain decision optimization based on knowledge graph includes decision model construction, reinforcement learning algorithm application and multi-objective conflict resolution;

[0020] The specific process of building the decision model is as follows: Based on the information in the knowledge graph, a supply chain decision optimization model is constructed; risk prediction results are used as constraints, and multiple objectives such as cost, delivery time, and service level are considered to establish a multi-objective optimization model; mathematical programming methods are used to formally describe the multi-objective optimization model, clarifying the decision variables, objective functions, and constraints; and the model is further refined and improved based on the actual business rules and restrictions of the supply chain.

[0021] The specific process of applying the reinforcement learning algorithm is as follows: using the reinforcement learning algorithm to optimize decision-making in the knowledge graph environment; defining the state space, action space and reward function; and continuously optimizing the decision-making strategy through continuous interaction with the environment to maximize the cumulative reward.

[0022] The specific process of multi-objective conflict resolution is as follows: construct a multi-objective optimization decision-making framework, adopt the weighted summation method and ε-constraint method multi-objective planning technology, and establish a comprehensive utility function model; by dynamically adjusting the target weight coefficient and constraint threshold, generate a Pareto optimal solution set, and provide enterprises with a set of decision-making solutions that take into account multiple dimensions.

[0023] Preferably, the specific process of data collection and preprocessing in the construction of the supply chain knowledge graph is as follows:

[0024] Identify the primary sources of supply chain data, including the company's internal ERP system, logistics management system, and customer relationship management system, as well as external market data platforms and industry reports. Design appropriate data collection interfaces and strategies based on the data characteristics of different data sources to ensure the comprehensiveness and timeliness of the data. Clean the collected data to remove duplicate, erroneous, and missing data. Use data interpolation and filling methods to handle missing values, and apply data denoising algorithms to remove noisy data. Standardize the data according to unified data standards and semantic specifications.

[0025] The specific process of knowledge graph construction in supply chain knowledge graph construction is as follows:

[0026] Define the ontology model of the knowledge graph, including entity types, relationship types, and attributes. Based on a hybrid information extraction framework, use the BERT-BiLSTM-CRF model to accurately identify multiple entity types, and combine rule templates with deep learning models to accurately extract semantic relationships. Through knowledge fusion technologies such as entity disambiguation and relationship verification, ensure the high quality and reliability of the constructed knowledge graph, providing a solid knowledge foundation for subsequent risk prediction and decision optimization. Store entities and relationships according to the structure of the ontology model, and select an appropriate graph database for storage and management to facilitate subsequent query and reasoning.

[0027] The specific process of the dynamic update mechanism in the construction of the supply chain knowledge graph is as follows:

[0028] Design an event-driven update trigger mechanism to monitor key events in the supply chain in real time, such as supplier production interruptions, logistics delays, and the addition of new customers; when an event occurs, promptly trigger the update process of the knowledge graph; use an incremental learning algorithm to process newly generated data and only update the affected parts of the knowledge graph, reducing the update workload and the impact on system performance; build a dynamic update mechanism for the knowledge graph, and through continuous data mining and pattern recognition, automatically discover new entity nodes and relationship edges from incremental data to achieve iterative optimization and semantic expansion of the knowledge graph.

[0029] The present invention has the following beneficial effects:

[0030] (1) Improved risk prediction accuracy: By integrating multi-source data to build a knowledge graph and combining it with machine learning algorithms for analysis, the complex relationships between various links in the supply chain are fully considered, improving the accuracy of supply chain risk warnings. By building a forward-looking risk identification mechanism, it provides companies with a more adequate time window for risk response. Compared with traditional methods, the improved risk prediction accuracy significantly reduces operational losses and financial risks caused by delayed risk discovery, and enhances the company's risk resistance capabilities.

[0031] (2) Optimizing supply chain decisions: Based on the global perspective and decision optimization model of the knowledge graph, various factors can be comprehensively considered when making supply chain decisions to maximize the overall benefits of the supply chain. This effectively avoids the systemic risks that may be caused by local decisions. By establishing a global optimization mechanism, efficient coordination of all links in the supply chain is achieved, thereby significantly improving the market responsiveness and competitive advantage of the overall supply chain. For example, in supplier selection and procurement decisions, the optimization model can find the optimal balance between cost and supply stability, reducing procurement costs while ensuring supply reliability.

[0032] (3) Enhanced risk response capabilities: When a risk occurs, relevant information can be quickly obtained with the help of the knowledge graph, and effective risk response strategies can be formulated and implemented in a timely manner. The decision-making execution tracking mechanism ensures that risk response measures can be effectively implemented, greatly shortening the risk response time and reducing risk losses. The risk resistance of the supply chain system is significantly enhanced. By establishing a multi-level risk prevention and control mechanism, the robustness and business continuity of the supply chain network in complex environments are ensured.

[0033] (4) Improve supply chain management efficiency: System integration and visualization functions enable centralized management and intuitive display of supply chain data, making it easier for companies to monitor and manage their supply chains in real time. The visual interactive interface allows users to quickly locate key information, and through intuitive data exploration and real-time analysis functions, it significantly improves the timeliness and reliability of the decision-making process. In addition, the system's automated processing mechanism effectively replaces traditional manual data operations, significantly reducing data processing costs and minimizing the risk of human error, thereby achieving overall optimization of supply chain operational efficiency.

[0034] (5) Promoting digital transformation of the supply chain: This technical solution provides core technical support for the digital reconstruction and intelligent upgrade of the entire supply chain process through an innovative methodological system. By constructing a knowledge graph, various data in the supply chain are converted into structured knowledge, providing a foundation for enterprises to apply advanced technologies such as artificial intelligence and big data. It promotes the transformation of supply chain management from traditional experience-driven to data-driven and intelligent decision-making, and enhances the competitiveness of enterprises in the digital age. DETAILED DESCRIPTION

[0035] The specific implementation of the present invention will be further described below with reference to specific embodiments:

[0036] The intelligent supply chain risk prediction and decision optimization method based on knowledge graph includes: supply chain knowledge graph construction, supply chain risk prediction model construction based on knowledge graph and supply chain decision optimization based on knowledge graph.

[0037] The construction of supply chain knowledge graph includes data collection and preprocessing, knowledge graph construction and dynamic update mechanism.

[0038] The specific process of data collection and preprocessing is as follows:

[0039] Identify the primary sources of supply chain data, including internal ERP systems, logistics management systems, and customer relationship management systems, as well as external market data platforms and industry reports. Design appropriate data collection interfaces and strategies based on the characteristics of each data source to ensure comprehensiveness and timeliness. Clean the collected data to remove duplicate, erroneous, and missing data. Use methods such as data interpolation and infilling to address missing values, and employ data denoising algorithms to remove noisy data. Standardize the data according to unified data standards and semantic specifications, such as standardizing product names and supplier names.

[0040] The specific process of knowledge graph construction is as follows:

[0041] Define the ontology model of the knowledge graph, including entity types (such as suppliers, products, logistics nodes, customers, etc.), relationship types (such as supply relationships, transportation relationships, demand relationships, etc.), and attributes (such as product prices, supplier ratings, logistics time, etc.). Based on a hybrid information extraction framework, the BERT-BiLSTM-CRF model is used to accurately identify multiple types of entities, and semantic relationships are accurately extracted by combining rule templates with deep learning models. Through knowledge fusion technologies such as entity disambiguation and relationship verification, the constructed knowledge graph is ensured to be of high quality and reliability, providing a solid knowledge foundation for subsequent risk prediction and decision optimization. Store entities and relationships according to the structure of the ontology model, and select an appropriate graph database (such as Neo4j) for storage and management to facilitate subsequent queries and reasoning.

[0042] The specific process of the dynamic update mechanism is as follows:

[0043] Design an event-driven update trigger mechanism to monitor key supply chain events in real time, such as supplier production disruptions, logistics delays, and new customer onboarding. When an event occurs, promptly trigger the knowledge graph update process. Employ an incremental learning algorithm to process newly generated data and update only the affected portions of the knowledge graph, reducing the update workload and the impact on system performance. Build a dynamic knowledge graph update mechanism that automatically discovers new entity nodes and relationship edges from incremental data through continuous data mining and pattern recognition, enabling iterative optimization and semantic expansion of the knowledge graph. This process utilizes an incremental learning algorithm to ensure the timeliness and completeness of the knowledge graph.

[0044] The construction of supply chain risk prediction model based on knowledge graph includes feature engineering, model construction, and model training and optimization.

[0045] The specific process of special engineering is to use the topological structure and semantic attributes of the knowledge graph to extract multi-level risk prediction features. These include attribute features of entities (such as the credit rating of suppliers, market demand fluctuations of products, etc.), relationship features (such as the stability of supply relationships, the reliability of logistics relationships, etc.) and graph structure features (such as the degree of nodes, shortest paths, etc.). The Node2Vec graph embedding algorithm is used to capture the structural features of the knowledge graph through a biased random walk strategy, and encode nodes and relationships into dense low-dimensional vector representations. These vectorized features effectively retain the semantic information of the graph structure and provide high-quality feature input for subsequent risk prediction models. Given a node pair (u, v), where v is the central node and u is its neighbor node, the goal of Node2Vec is to maximize the following objective function:

[0046]

[0047] Where V is the set of all nodes in the graph, N(v) is the set of neighbor nodes of node v, which is generated by biased random walk and controlled by parameters p and q, and h v , h u : The low-dimensional vectors of nodes v and u represent the embedding vectors, and the σSigmoid function maps the inner product to the probability interval [0, 1][0,1];

[0048] This formula is the objective function of Node2Vec, which aims to maximize the similarity between the embedding vectors of node v and its neighbor node u. By optimizing this objective, the model is able to encode topological relationships (such as homogeneity and structurality) in the graph structure into low-dimensional vectors.

[0049] Filter key features based on domain knowledge, build effective feature combinations, and improve model efficiency and prediction accuracy.

[0050] The model construction process involves adopting graph neural networks (GNNs) as the foundational framework, such as the Graph Convolutional Network (GCN) and the Graph Attention Network (GAT). This leverages GNNs' strengths in processing graph-structured data. Through a multi-layer message-passing mechanism, it deeply explores the underlying patterns of semantic associations and relationships between entities in the knowledge graph, enabling representation learning of complex knowledge networks. An attention mechanism is introduced into the GNN model to dynamically adjust the weights of different nodes and relationships in risk prediction, highlighting those with significant impact. Time series analysis methods, such as LSTM and GRU, are combined to model the temporal characteristics of supply chain data and capture changing risk trends. The GNN and time series models are then integrated to construct a risk prediction model that simultaneously considers both graph structural and temporal characteristics.

[0051] Model training and optimization utilize a k-fold cross-validation strategy, ensuring the stability and reliability of model evaluation results through multiple partitioning and validation of the dataset. To mitigate the risk of overfitting, a combination of L1 / L2 regularization constraints and Dropout random loss techniques are employed to enhance the model's generalization capabilities. The optimization process utilizes the Adaptive Moment Estimation (Adam) algorithm, which dynamically adjusts the learning rate and momentum parameters to achieve efficient convergence of the loss function, thereby obtaining the optimal model parameter configuration and significantly improving prediction accuracy and robustness.

[0052] Supply chain decision optimization based on knowledge graph includes decision model construction, reinforcement learning algorithm application and multi-objective conflict resolution;

[0053] The specific process of constructing a decision model is as follows: Based on the information in the knowledge graph, a supply chain decision optimization model is constructed. Risk prediction results are used as constraints, and a multi-objective optimization model is established, taking into account multiple objectives such as cost, delivery time, and service level. For example, the goal may be to minimize the overall cost of the supply chain while ensuring that service level requirements are met within the specified delivery time. Mathematical programming methods (such as linear programming and integer programming) are used to formally describe the multi-objective optimization model, clarifying the decision variables, objective function, and constraints. The model is further refined and improved based on the actual business rules and constraints of the supply chain.

[0054] The specific process of applying reinforcement learning algorithms involves applying them (such as Q-learning and Deep QNetwork (DQN)) to optimize decisions within a knowledge graph environment. This involves defining a state space (e.g., the current state of the supply chain, including inventory levels, order status, and risk profile), an action space (e.g., purchasing decisions, production plan adjustments, and logistics and distribution plan selection), and a reward function (which assigns rewards based on the outcome of the decision, such as positive rewards for cost reductions and negative rewards for delivery delays). By continuously interacting with the environment, the decision-making strategy is continuously optimized to maximize the cumulative reward.

[0055] The specific process for resolving multi-objective conflicts involves constructing a multi-objective optimization decision-making framework, employing multi-objective planning techniques such as the weighted summation method and the ε-constraint method, and establishing a comprehensive utility function model. By dynamically adjusting objective weight coefficients and constraint thresholds, a Pareto-optimal solution set is generated, providing enterprises with a set of decision-making options that balance multiple dimensions. This mechanism supports enterprises in selecting the optimal resource allocation solution from the non-dominated solution set based on their strategic orientation and actual constraints, achieving dynamic balancing and coordinated optimization among multiple objectives.

[0056] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the scope of protection of the present invention.

Claims

1. An intelligent supply chain risk prediction and decision optimization method based on knowledge graph, characterized by: include: Supply chain knowledge graph construction, supply chain risk prediction model construction based on knowledge graph, and supply chain decision optimization based on knowledge graph; The construction of supply chain knowledge graph includes data collection and preprocessing, knowledge graph construction and dynamic update mechanism; The construction of supply chain risk prediction models based on knowledge graphs includes feature engineering, model construction, and model training and optimization; The specific process of special engineering is to extract multi-level risk prediction features by using the topological structure and semantic attributes of the knowledge graph; Including entity attribute features, relationship features and graph structure features; The Node2Vec graph embedding algorithm captures the structural features of the knowledge graph through a biased random walk strategy, encoding nodes and relationships into dense low-dimensional vector representations. Given a node pair (u, v), where v is the central node and u is its neighbor node, the goal of Node2Vec is to maximize the following objective function: Where V is the set of all nodes in the graph, N(v) is the set of neighbor nodes of node v, which is generated by biased random walk and controlled by parameters p and q, and h v , h u : The low-dimensional vectors of nodes v and u represent the embedding vectors, and the σSigmoid function maps the inner product to the probability interval [0, 1][0,1]; Screen key features based on domain knowledge and build effective feature combinations; The specific process of model construction is as follows: using graph neural network (GNN) as the basic framework, through a multi-layer message passing mechanism, deeply mining the semantic associations and potential patterns of relationships between entities in the knowledge graph to achieve representation learning of complex knowledge networks; introducing an attention mechanism into the GNN model to dynamically adjust the weights of different nodes and relationships in risk prediction, highlighting the parts that have a significant impact on risk prediction; combining time series analysis methods to model the temporal characteristics of supply chain data and capture the changing trends of risks; and fusing GNN and time series models to construct a risk prediction model that can simultaneously consider graph structural features and temporal characteristics. Model training and optimization utilize a k-fold cross-validation strategy, ensuring the stability and reliability of model evaluation results through multiple partitioning and validation of the dataset. To mitigate the risk of overfitting, L1 / L2 regularization constraints and Dropout random loss techniques are integrated to enhance the model's generalization capabilities. The optimization process utilizes the adaptive moment estimation Adam algorithm, which dynamically adjusts the learning rate and momentum parameters to achieve efficient convergence of the loss function, thereby obtaining the optimal model parameter configuration. Supply chain decision optimization based on knowledge graph includes decision model construction, reinforcement learning algorithm application and multi-objective conflict resolution; The specific process of building the decision model is as follows: Based on the information in the knowledge graph, a supply chain decision optimization model is constructed; risk prediction results are used as constraints, and multiple objectives such as cost, delivery time, and service level are considered to establish a multi-objective optimization model; mathematical programming methods are used to formally describe the multi-objective optimization model, clarifying the decision variables, objective functions, and constraints; and the model is further refined and improved based on the actual business rules and restrictions of the supply chain. The specific process of applying the reinforcement learning algorithm is as follows: using the reinforcement learning algorithm to optimize decision-making in the knowledge graph environment; defining the state space, action space and reward function; and continuously optimizing the decision-making strategy through continuous interaction with the environment to maximize the cumulative reward. The specific process of multi-objective conflict resolution is as follows: construct a multi-objective optimization decision-making framework, adopt the weighted summation method and ε-constraint method multi-objective planning technology, and establish a comprehensive utility function model; by dynamically adjusting the target weight coefficient and constraint threshold, generate a Pareto optimal solution set, and provide enterprises with a set of decision-making solutions that take into account multiple dimensions.

2. The method for intelligent supply chain risk prediction and decision optimization based on knowledge graph according to claim 1, characterized in that: The specific process of data collection and preprocessing in the construction of the supply chain knowledge graph is as follows: Identify the primary sources of supply chain data, including the company's internal ERP system, logistics management system, and customer relationship management system, as well as external market data platforms and industry reports. Design appropriate data collection interfaces and strategies based on the data characteristics of different data sources to ensure the comprehensiveness and timeliness of the data. Clean the collected data to remove duplicate, erroneous, and missing data. Use data interpolation and filling methods to handle missing values, and apply data denoising algorithms to remove noisy data. Standardize the data according to unified data standards and semantic specifications. The specific process of knowledge graph construction in supply chain knowledge graph construction is as follows: Define the ontology model of the knowledge graph, including entity types, relationship types, and attributes. Based on a hybrid information extraction framework, use the BERT-BiLSTM-CRF model to accurately identify multiple entity types, and combine rule templates with deep learning models to accurately extract semantic relationships. Through knowledge fusion technologies such as entity disambiguation and relationship verification, ensure the high quality and reliability of the constructed knowledge graph, providing a solid knowledge foundation for subsequent risk prediction and decision optimization. Store entities and relationships according to the structure of the ontology model, and select an appropriate graph database for storage and management to facilitate subsequent query and reasoning. The specific process of the dynamic update mechanism in the construction of the supply chain knowledge graph is as follows: Design an event-driven update trigger mechanism to monitor key events in the supply chain in real time, such as supplier production disruptions, logistics delays, and new customer additions. When an event occurs, it triggers the knowledge graph update process in a timely manner. An incremental learning algorithm is used to process newly generated data, and only the affected parts of the knowledge graph are updated, reducing the update workload and the impact on system performance; a dynamic update mechanism for the knowledge graph is built, and through continuous data mining and pattern recognition, new entity nodes and relationship edges are automatically discovered from incremental data, realizing iterative optimization and semantic expansion of the knowledge graph.

Images