Intelligent customer relationship knowledge graph construction method and system

Abstract

The invention discloses an intelligent customer relationship knowledge graph construction method and system. The method comprises the steps of obtaining customer information of an enterprise customer database; constructing and optimizing a customer relationship ontology based on the customer information; according to the optimized customer relationship ontology, the customer entities are mapped into knowledge graph nodes, the incidence relation between the customer entities is mapped into directed edges between the nodes, entity attribute information serves as knowledge graph node attributes to be stored, and an initial intelligent customer relationship knowledge graph is constructed; and performing data verification on the initial intelligent customer relationship knowledge graph, extracting deep features of customer information, and executing incremental updating on the initial intelligent customer relationship knowledge graph to determine a target intelligent customer relationship knowledge graph. The method can accurately reflect the complex association relationship between customers, and provides powerful support for customer relationship management of enterprises.

Description

Technical Field

[0001] This invention belongs to the field of knowledge graph construction technology, and in particular to a method and system for constructing an intelligent customer relationship knowledge graph. Background Technology

[0002] In today's digital economy, enterprise customer scale is constantly expanding, and customer information is characterized by its massive volume, multiple sources, and fragmentation. Customer relationship management has become a key link for enterprises to enhance their core competitiveness. The customer information stored in the enterprise customer database covers multiple dimensions such as basic customer information, transaction records, interaction behavior, and cooperation projects. This information not only includes independent customer entities (such as enterprise customers, contacts, and partner organizations), but also involves complex relationships between customer entities (such as business cooperation relationships, decision-making hierarchy relationships, and business dealings), as well as rich entity attribute information (such as customer industry category, transaction amount, credit rating, and cooperation period). Traditional customer relationship management (CRM) methods primarily rely on relational databases for data storage and retrieval. While this approach enables structured management of individual customer information, it has significant limitations: Firstly, it struggles to effectively uncover hidden relationships between customer entities. For instance, it cannot intuitively present the collaborative networks of different customer entities within the industry chain, or the logical connections between different business segments of the same customer, making it difficult for enterprises to fully grasp the overall picture of customer relationships. Secondly, data management lacks dynamism. When customer information is updated (such as through mergers and acquisitions, changes in contact persons, or adjustments to business scope), it cannot quickly synchronize and adjust data, leading to data lag and inconsistencies. This, in turn, affects the accuracy and timeliness of enterprise decision-making, becoming a critical technical problem that urgently needs to be solved in this field. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent customer relationship knowledge graph construction method and system to address the shortcomings of existing technologies. It proposes an intelligent customer relationship knowledge graph construction method that can accurately reflect the complex relationships between customers and provide strong support for enterprise customer relationship management.

[0004] One embodiment of this application provides a method for constructing an intelligent customer relationship knowledge graph, the method comprising: Obtain customer information from the enterprise customer database; the customer information includes customer entities, the relationships between customer entities, and entity attribute information. Based on the customer information, a customer relationship ontology is constructed and optimized. This ontology is constructed using an improved K-means clustering algorithm, which determines initial cluster centers by calculating intra-cluster object compactness, inter-cluster distance, and data density. According to the customer relationship ontology, customer entities and their relationships are categorized into different semantic categories, forming an optimized customer relationship ontology. The optimized customer relationship ontology is determined based on the semantics, relationships, and semantic completeness of the customer relationship ontology. Based on the optimized customer relationship ontology, customer entities are mapped to knowledge graph nodes, the relationships between customer entities are mapped to directed edges between nodes, and entity attribute information is stored as knowledge graph node attributes to construct an initial intelligent customer relationship knowledge graph. The initial intelligent customer relationship knowledge graph is validated, and deep features of customer information are extracted. Incremental updates are then performed on the initial intelligent customer relationship knowledge graph to determine the target intelligent customer relationship knowledge graph. Specifically, the initial intelligent customer relationship knowledge graph is validated using a combination of rule validation and data consistency validation. The rule validation, based on preset customer entity attribute constraint rules and relational logic rules, filters out data with abnormal attribute values ​​and contradictory relational logic. The data consistency validation compares the original data in the enterprise customer database with the customer information stored in the initial intelligent customer relationship knowledge graph to locate inconsistencies and marks and corrects abnormal data.

[0005] Optionally, obtaining customer information from the enterprise customer database includes: Identify customer entities in the enterprise's customer database; A hybrid algorithm combining a customer entity vectorization model and a bidirectional long short-term memory network is used to extract the relationships between customer entities to determine customer information; wherein, the relationships include business cooperation relationships, demand and supply relationships, and service interaction relationships.

[0006] Optionally, the process of extracting the relationships between customer entities using a hybrid algorithm combining a customer entity vectorization model and a bidirectional long short-term memory network to determine customer information includes: Construct an initial vectorized model of the customer entity; The context window consisting of n words before and after the customer data keywords is used as the input layer word sequence of the customer entity vectorization model. The input layer word sequence is represented by one-hot encoding, and the probability distribution of the predicted words is used as the output vector. The cross-entropy loss function is defined, and the parameters of the initial customer entity vectorization model are iteratively updated by the stochastic gradient descent method to obtain the trained customer entity vectorization model. The normalized customer entity data is input into the trained customer entity vectorization model to obtain word vector representations, which are then input into a bidirectional long short-term memory network to obtain the dependencies between customer entities. Based on the dependencies between the customer entities, the associations between the customer entities are extracted to determine customer information.

[0007] Optionally, the customer relationship ontology is determined using an improved K-means clustering algorithm, and the cluster centers are dynamically adjusted using the following formula: in, It indicates that at the (t+1)th iteration, the... The new cluster center of each cluster, Indicates the iteration number t when the t-th iteration is... The new cluster center of each cluster, Indicates the learning rate. This represents the first clustering step partitioned by the improved K-means clustering algorithm. A subset of data, Indicates the number of objects in the cluster. Represents customer data points Local density, Represents the inter-cluster repulsion coefficient. Cluster with cluster distance, Cluster with cluster distance, It indicates that at the t-th iteration, the... Cluster centers of each cluster It indicates that at the t-th iteration, the... Cluster centers of each cluster.

[0008] Optionally, the step of performing data verification on the initial intelligent customer relationship knowledge graph, extracting deep features of customer information, and performing incremental updates on the initial intelligent customer relationship knowledge graph to determine the target intelligent customer relationship knowledge graph includes: A deep feature extraction model for customer information, combining residual neural networks and attention mechanisms, is constructed to extract deep features from customer information in order to perform incremental updates on the initial intelligent customer relationship knowledge graph and determine the target intelligent customer relationship knowledge graph.

[0009] Optionally, the construction of a deep feature extraction model for customer information combining a residual neural network and an attention mechanism, performing deep feature extraction on customer information to perform incremental updates to the initial intelligent customer relationship knowledge graph and determine the target intelligent customer relationship knowledge graph, includes: After normalizing the customer entity attribute information and relationship data, the data is used as the input data for the deep feature extraction model of customer information that combines residual neural network and attention mechanism; By using the embedding vector corresponding to the input data generated by the Siamese network, multi-dimensional basic features are extracted, and then the attention mechanism is used to assign weights to the initial features of the customer in order to output the deep feature vector of the customer. Based on the customer's deep feature vector and combined with the new customer information in the enterprise customer database, an incremental update is performed on the initial intelligent customer relationship knowledge graph; wherein, the new customer information is mapped into new knowledge graph nodes, directed edges and node attributes according to the optimized customer relationship ontology and added to the initial intelligent customer relationship knowledge graph. The incrementally updated customer relationship knowledge graph is evaluated. If the evaluation result meets the preset threshold, the current incrementally updated customer relationship knowledge graph is determined as the target intelligent customer relationship knowledge graph. The evaluation indicators include node completeness, relationship accuracy, and feature relevance.

[0010] Optionally, the method further includes: If the evaluation result does not meet the preset threshold, the steps of performing data verification on the initial intelligent customer relationship knowledge graph, extracting deep features of customer information, and performing incremental updates on the initial intelligent customer relationship knowledge graph are repeated.

[0011] Another embodiment of this application provides an intelligent customer relationship knowledge graph construction system, the system comprising: The acquisition module is used to acquire customer information from the enterprise customer database; the customer information includes customer entities, the relationships between customer entities, and entity attribute information. An optimization module is used to construct and optimize a customer relationship ontology based on the customer information. The customer relationship ontology is constructed using an improved K-means clustering algorithm, which determines initial cluster centers by calculating intra-cluster object compactness, inter-cluster distance, and data density. Based on the customer relationship ontology, customer entities and their relationships are categorized into different semantic categories to form an optimized customer relationship ontology. The optimized customer relationship ontology is determined based on the semantics, relationships, and semantic completeness of the customer relationship ontology. The construction module is used to map customer entities to knowledge graph nodes based on the optimized customer relationship ontology, map the relationships between customer entities to directed edges between nodes, and store entity attribute information as knowledge graph node attributes to build an initial intelligent customer relationship knowledge graph. The determination module is used to perform data verification on the initial intelligent customer relationship knowledge graph, extract deep features of customer information, and perform incremental updates on the initial intelligent customer relationship knowledge graph to determine the target intelligent customer relationship knowledge graph. Specifically, the initial intelligent customer relationship knowledge graph is verified using a combination of rule verification and data consistency verification. The rule verification is based on preset customer entity attribute constraint rules and association logic rules to filter out data with abnormal attribute values ​​and contradictory relationship logic. The data consistency verification compares the original data in the enterprise customer database with the customer information stored in the initial intelligent customer relationship knowledge graph to locate inconsistencies and mark and correct abnormal data.

[0012] Another embodiment of this application provides a storage medium storing a computer program, wherein the computer program is configured to implement the method described in any of the above-described embodiments when running.

[0013] Another embodiment of this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to implement the method described in any of the above embodiments.

[0014] Compared with existing technologies, this invention first obtains customer information from an enterprise customer database; then, based on this customer information, it constructs and optimizes a customer relationship ontology; according to the optimized customer relationship ontology, customer entities are mapped to knowledge graph nodes, the relationships between customer entities are mapped to directed edges between nodes, and entity attribute information is stored as knowledge graph node attributes, thus constructing an initial intelligent customer relationship knowledge graph; the initial intelligent customer relationship knowledge graph is then validated, and deep features of customer information are extracted, followed by incremental updates to determine the target intelligent customer relationship knowledge graph. This accurately reflects the complex relationships between customers, providing strong support for enterprise customer relationship management. Attached Figure Description

[0015] Figure 1 A hardware structure block diagram of a computer terminal for an intelligent customer relationship knowledge graph construction method provided in an embodiment of the present invention; Figure 2 A flowchart illustrating an intelligent customer relationship knowledge graph construction method provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an intelligent customer relationship knowledge graph construction system provided in an embodiment of the present invention. Detailed Implementation

[0016] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0017] This invention first provides a method for constructing an intelligent customer relationship knowledge graph, which can be applied to electronic devices, such as computer terminals, specifically ordinary computers, tablets, etc.

[0018] The following detailed explanation uses a computer terminal as an example. Figure 1 This is a hardware structure block diagram of a computer terminal for a method of constructing an intelligent customer relationship knowledge graph, as provided in an embodiment of the present invention. Figure 1 As shown, the computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0019] Non-volatile storage media can store operating systems and computer programs. These computer programs include program instructions that, when executed, cause the processor to perform any intelligent customer relationship knowledge graph construction method.

[0020] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0021] Internal memory provides an environment for the execution of computer programs stored in non-volatile storage media. When these computer programs are executed by a processor, the processor can execute any intelligent customer relationship knowledge graph construction method.

[0022] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0023] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0024] See Figure 2 , Figure 2 A flowchart illustrating an intelligent customer relationship knowledge graph construction method provided in this embodiment of the invention may include the following steps: S201: Obtain customer information from the enterprise customer database; the customer information includes customer entities, relationships between customer entities, and entity attribute information.

[0025] Specifically, enterprise customer databases typically store massive amounts of unstructured or semi-structured data (such as transaction records, communication logs, and contract documents), making it difficult to directly extract the relationships and attribute information between customer entities using traditional methods. This invention utilizes Natural Language Processing (NLP), graph database technology, and machine learning algorithms to achieve automated extraction and structured integration of customer information.

[0026] For example, raw customer information can be obtained by accessing enterprise customer databases (such as MySQL, Oracle, MongoDB) or third-party data interfaces (such as CRM system APIs). For instance, customer transaction records, service orders, contact information, and other table data can be exported from the CRM. Then, duplicate data can be removed, such as multiple duplicate records for the same customer, missing values ​​can be filled (e.g., filling missing annual income values ​​with the average), and the format can be standardized, such as using a uniform date format of YYYY-MM-DD.

[0027] Customer entity recognition can employ pre-trained language models (such as BERT and RoBERTa) combined with Conditional Random Fields (CRF) for Named Entity Recognition (NER) to extract entities such as customer names, contact persons, and product names. For example, from the text "Customer A purchased product X in 2025," the entities "Customer A" (customer entity) and "Product X" (product entity) can be identified. Then, entities with the same name are merged by calculating entity similarity (such as Jaccard similarity and word vector cosine similarity). Relationships between customer entities can be obtained by defining relationship templates, such as regular expressions or keyword matching, to extract explicit relationships. For example, if Customer A and Customer B signed contract C, the relationship "cooperation (Customer A → Contract C → Customer B)" can be extracted. Relationships between customer entities can also be obtained through graph embedding relationship reasoning, i.e., using graph neural networks (GNNs) to learn the embeddings of entity graphs such as customers, products, and suppliers to infer implicit relationships. For example, an initial entity graph can be constructed, node representations can be updated using GNNs, and finally, cluster analysis can be used to discover potential relationships. For example, if it is found that customers A, B, and C all purchase product X at overlapping times, it can be inferred that they have a competitive relationship or an industry collaboration relationship.

[0028] Among them, the entity attribute information can be extracted directly from the customer database fields as explicit attributes, or from the customer database fields as implicit attributes through NLP technology, such as customer satisfaction attributes and customer integrity attributes. Finally, the customer information can be integrated into structured data in a unified format for easy subsequent processing.

[0029] In one optional implementation, obtaining customer information from the enterprise customer database may include: 1. Identify customer entities in the enterprise's customer database.

[0030] Specifically, customer entities, such as company names, contact persons, and product names, are accurately extracted from the aforementioned unstructured / semi-structured data.

[0031] For example, an improved entity recognition model can be used to define a customer entity confidence score: in, This is a score that measures the probability that candidate customer entity e is a real customer entity. It typically ranges from [0,1], representing the entity confidence score. This represents the existing, verified set of customer entities of the enterprise. is the similarity score between candidate customer entity e and the most similar entity in the known customer entity database EB, representing the similarity between the candidate entity and the known customer entity database. This represents the normalized value of the semantic frequency of a customer entity; a higher frequency generally corresponds to a higher confidence level. This is used to measure the degree of matching between the context of customer entity e and the typical context of customer entities, that is, to represent the entity context consistency feature, and , , The sum is 1.

[0032] 2. By using a hybrid algorithm combining a customer entity vectorization model and a bidirectional long short-term memory network, the relationships between customer entities are extracted to determine customer information; wherein, the relationships include business cooperation relationships, demand and supply relationships, and service interaction relationships.

[0033] This study employs a hybrid algorithm combining a customer entity vectorization model with a bidirectional long short-term memory (LSTM) network to mine implicit relationships between customer entities. The process begins by initializing the customer entity vectorization model, such as the Word2Vec model. This model uses the context windows of n words preceding and following customer data keywords as the input word sequence, employing one-hot encoding. The cross-entropy loss function is optimized using stochastic gradient descent to train the word vector model. Normalized customer entity data is then input into the trained word vector model, and the resulting word vector representations are fed into the bidirectional LSM network to capture temporal dependencies between customer entities, such as the supply and demand relationship in Company A's purchase of raw materials from Supplier B. Based on these dependencies, three types of relationships—business cooperation, supply and demand, and service interaction—are extracted to form a customer information set.

[0034] In one optional implementation, the step of extracting the relationships between customer entities using a hybrid algorithm combining a customer entity vectorization model and a bidirectional long short-term memory network to determine customer information may include: Step 1: Construct the initial customer entity vectorization model.

[0035] For example, a vectorized model capable of capturing the semantic features of customer entities is initialized. The model vectors can be set to 256 dimensions and initialized using Xavier. Finally, 50-100 negative samples are set to accelerate training and improve the learning effect of low-frequency words. In this application, the model architecture can choose a Skip-gram model, which predicts context words from the current word, making it more suitable for capturing co-occurrence relationships between customer entities. For example, if customer A and product X frequently co-occur in the contract text, their vector similarity is high.

[0036] Step 2: Use the context window consisting of n words before and after the customer data keywords as the input layer word sequence of the customer entity vectorization model; wherein, the input layer word sequence is represented by one-hot encoding, the probability distribution of the predicted words is used as the output vector, the cross-entropy loss function is defined, and the parameters of the initial customer entity vectorization model are iteratively updated by the stochastic gradient descent method to obtain the trained customer entity vectorization model.

[0037] For example, the input layer of a customer entity vectorization model can be the input layers of customer data before and after keywords. Context window of a word Perform one-hot encoding, generating dimension . input matrix ,in, For the vocabulary size, the client entity vectorization model optimizes word vector representations by predicting word probability distributions and defines a cross-entropy loss function: in, Represents the cross-entropy loss function. Represents the number of training samples. Indicates a given headword When predicting its context words The probability of occurrence This represents a given customer data center term. These are context terms representing customer data.

[0038] Furthermore, to optimize the customer entity vectorization model, a set of customer sample pairs can also be provided. and a customer tag set The following contrast loss function is used: in, This represents the contrastive loss function. The basic unit representing a customer sample Indicates the weight of the customer sample. express , Distance measurement between Indicates margin.

[0039] This contrastive loss function aims to increase the distance between two samples in a pair of customer samples labeled 0, and to decrease the distance between two samples in a pair of customer samples labeled 1. Here, This represents the trainable parameters of the customer entity vectorization model. This is achieved by introducing margins. For customer sample pairs with label 0, they can only affect the loss function if their distance is less than the margin. The model parameters are updated as follows: : in, It's the learning rate. It updates in one step. Become ,based on The customer entity vectorization model can map customer entity vectors to low-dimensional embedding vectors.

[0040] Step 3: Input the normalized customer entity data into the trained customer entity vectorization model to obtain word vector representations, and then input them into a bidirectional long short-term memory network to obtain the dependencies between customer entities.

[0041] For example, the normalized customer entity data output from the customer entity vectorization model. Input into a bidirectional long short-term memory network, via forward pass and backward By concatenating the hidden states, we obtain the dependency vector: in, These represent the forward and backward hidden states of the Long Short-Term Memory network at position t, respectively. This represents the dependency matrix between customer entities.

[0042] Based on dependency vectors Then, the Softmax function is used to classify business cooperation relationships, supply and demand relationships, and other types of relationships: in, Represents a given dependency vector When predicting its relation type. The probability, , These are trainable parameters.

[0043] The above method generates a set of customer entities. Dependency matrix between its client entities , where k is the number of customer relationship types.

[0044] Step 4: Based on the dependencies between the customer entities, extract the associations between the customer entities to determine customer information.

[0045] S202: Based on the customer information, construct and optimize the customer relationship ontology; wherein the customer relationship ontology is constructed using an improved K-means clustering algorithm, and the improved K-means clustering algorithm determines the initial cluster centers by calculating the compactness of objects within clusters, the distance between clusters, and the data density; according to the customer relationship ontology, customer entities and the relationships between customer entities are divided into different semantic categories to form an optimized customer relationship ontology; the optimized customer relationship ontology is determined based on the semantics, relationships, and semantic completeness of the customer relationship ontology.

[0046] Specifically, based on the customer information, a customer relationship ontology is constructed and optimized; the customer relationship ontology is determined using an improved K-means clustering algorithm and the cluster centers are dynamically adjusted using the following formula: in, It indicates that at the (t+1)th iteration, the... The new cluster center of each cluster, Indicates the iteration number t when the t-th iteration is... The new cluster center of each cluster, Indicates the learning rate. This represents the first clustering step partitioned by the improved K-means clustering algorithm. A subset of data, Indicates the number of objects in the cluster. Represents customer data points Local density, Represents the inter-cluster repulsion coefficient. Cluster with cluster distance, Cluster with cluster distance, It indicates that at the t-th iteration, the... Cluster centers of each cluster It indicates that at the t-th iteration, the... Cluster centers of each cluster.

[0047] For example, a customer relationship ontology can be constructed using an improved K-means clustering algorithm based on customer information. The improved K-means clustering algorithm determines initial cluster centers by calculating intra-cluster object compactness, inter-cluster distance, and data density. Then, based on the customer relationship ontology, customer entities and their relationships are divided into different semantic categories to form an optimized customer relationship ontology. The optimized customer relationship ontology is determined based on the semantics, relationships, and semantic completeness of the customer relationship ontology.

[0048] Specifically, the collected customer information is first cleaned to remove duplicate, erroneous, or incomplete data, and then standardized for subsequent analysis. Next, the number of clusters, K, is set, along with parameters for calculating intra-cluster object compactness, inter-cluster distance, and data density. For each potential cluster center, the average distance to its surrounding objects is calculated, and regions with low compactness are selected as candidate cluster centers to avoid local optima. The selected cluster centers are ensured to be sufficiently far apart to reduce cluster overlap. Regions with high data density are preferentially selected as cluster centers to improve the representativeness and stability of the clusters. Considering these three factors, the final initial cluster centers can be determined through a weighted summation or voting mechanism. Alternatively, each customer object can be assigned to the nearest cluster center, and the center point of each cluster can be recalculated. This process is repeated until the cluster centers no longer change or a preset number of iterations is reached to obtain the initial cluster centers.

[0049] Based on the clustering results, customer objects with similar characteristics are grouped into the same category, and customer entities such as customers, products, transactions, and their attributes are defined. Simultaneously, based on the interaction history between customer objects, the relationships between customer entities are constructed, forming a preliminary customer relationship ontology.

[0050] Optimizing the customer relationship ontology involves analyzing the semantic features of customer entities and their relationships based on the initially constructed ontology. Natural language processing techniques, such as part-of-speech tagging and named entity recognition, are used to extract key semantic information. Based on this information, customer entities and their relationships are categorized into different semantic categories, such as "high-value customers," "potential churned customers," and "frequently traded products." Each semantic category is then checked to ensure completeness and accuracy by covering all relevant customer entities and relationships. Finally, association rule mining or centrality analysis in graph theory is used to evaluate the strength of relationships between different semantic categories and identify key association paths.

[0051] Based on the evaluation results, semantic categories are merged, split, or adjusted to optimize the structure of the customer relationship ontology. Simultaneously, the definitions of customer entities and relationships are updated to ensure they align with business needs. After these optimization steps, the final optimized customer relationship ontology is formed. This ontology not only contains rich customer information and relationships but also improves the understandability and operability of customer relationships through semantic category division, providing strong support for subsequent customer relationship management and decision support.

[0052] For example, when using an improved K-means clustering algorithm to construct a customer relationship ontology, the intra-cluster compactness can first be defined. Inter-cluster distance and data density As an initial metric for selecting cluster centers, intra-cluster compactness can be represented as follows: in, Cluster All points to its center The smaller the mean square distance, the more concentrated the data points within the cluster. This represents the i-th subset of data partitioned by the K-Means algorithm. Indicates the center of the i-th cluster. This indicates the number of objects in the cluster.

[0053] Inter-cluster distance It can be represented as follows: in, Used to measure the degree of separation between two cluster centers These represent the centers of the i-th and j-th clusters, respectively.

[0054] Data density It can be represented as follows: in, Let x represent the set of the k data points that are closest to point x. , The feature vector representing a single customer entity. This refers to the preset parameters that control the range of density calculation.

[0055] Then select those that satisfy the criteria. , and The points are taken as the initial cluster centers, where, This represents the preset threshold. The dynamic cluster center adjustment mechanism allows cluster centers to move towards higher-density regions while maintaining inter-cluster distances, thus solving the problem of traditional K-means' sensitivity to initial values.

[0056] Finally, based on the customer relationship ontology, customer entities and the relationships between them are divided into different semantic categories to form an optimized customer relationship ontology. The semantic completeness can be calculated in the following way: in, It is the number of entity categories. It refers to the number of relation types. This indicates the degree of association between the i-th type of entity and the j-th type of relationship, and its value is 0 or 1.

[0057] Furthermore, based on the clustering results Semantic completeness can also be defined. Assess the quality of the substance ,satisfy: in, This represents the overall assessment score for the quality of the customer's product. , , These are the weighting coefficients. This indicates the coverage rate of customer entities and their relationships. Indicates semantic consistency within the cluster. Indicates redundancy.

[0058] S203: Based on the optimized customer relationship ontology, customer entities are mapped to knowledge graph nodes, the relationships between customer entities are mapped to directed edges between nodes, and entity attribute information is stored as knowledge graph node attributes to construct an initial intelligent customer relationship knowledge graph.

[0059] For example, based on the optimized customer relationship ontology, mapping customer entities to knowledge graph nodes can be achieved through a heterogeneous data fusion model, namely: in, Represents the customer entity vector. Represents a vector of association relationships. Represents the attribute feature vector. , , To fuse the weight matrix, For bias terms, This is the ReLU activation function.

[0060] The aforementioned heterogeneous data fusion model achieves a joint representation of entities, relationships, and attributes, providing a foundation for subsequent reasoning. Subsequently, customer entities can also be... Mapped to graph nodes Relationship Mapped to directed edges Attribute feature vector Stored as node characteristics Construct an initial intelligent customer relationship knowledge graph .

[0061] The mapping relationship between customer entities can also be determined by constructing a hybrid association strength calculation model: in, This represents the measure of customer entities output by the hybrid association strength calculation model. and The overall score of the mapping relationship between them. Represents the cosine similarity of customer entity vectors. This represents the dependency score output by the bidirectional long short-term memory network model. Indicates from The relation transition probability, and , , All are weighting coefficients. , , The sum of the three is 1.

[0062] The above hybrid association strength calculation model can effectively identify three types of core mapping relationships: business cooperation mapping relationship, with weights biased towards relationship transfer probability; demand and supply mapping relationship, with weights biased towards the dependency score of the bidirectional LSTM output; and service interaction mapping relationship, with weights biased towards the cosine similarity of customer entity vectors.

[0063] S204: Perform data verification on the initial intelligent customer relationship knowledge graph, extract deep features of customer information, and perform incremental updates on the initial intelligent customer relationship knowledge graph to determine the target intelligent customer relationship knowledge graph; wherein, the initial intelligent customer relationship knowledge graph is verified by combining rule verification and data consistency verification; the rule verification is based on preset customer entity attribute constraint rules and association relationship logic rules to filter out data with abnormal attribute values ​​and contradictory relationship logic; the data consistency verification compares the original data in the enterprise customer database with the customer information stored in the initial intelligent customer relationship knowledge graph to locate data inconsistencies and mark and correct abnormal data.

[0064] Specifically, the step of performing data verification on the initial intelligent customer relationship knowledge graph, extracting deep features of customer information, and performing incremental updates on the initial intelligent customer relationship knowledge graph to determine the target intelligent customer relationship knowledge graph may include: 1) The initial intelligent customer relationship knowledge graph is validated by combining rule validation and data consistency validation. Rule validation is based on preset customer entity attribute constraint rules and relationship logic rules to filter out data with abnormal attribute values ​​and contradictory relationship logic. Data consistency validation compares the original data in the enterprise customer database with the customer information stored in the initial intelligent customer relationship knowledge graph to locate data inconsistencies and mark and correct abnormal data.

[0065] For example, data validation rules can be defined by defining attribute constraint rules. such as "customer age" [0,100] and relational logic rules Directional rules, such as "supplier → customer," can be used to filter out outlier data. Consistency verification can be performed by calculating the map data. Differences from enterprise customer database (DB): in, Represents an indicator function that satisfies ,in, This represents the attribute value difference function. This indicates the preset minimum value.

[0066] A deep feature extraction model for customer information, combining residual neural networks and attention mechanisms, is constructed to extract deep features from customer information in order to perform incremental updates on the initial intelligent customer relationship knowledge graph and determine the target intelligent customer relationship knowledge graph.

[0067] Specifically, the construction of a deep feature extraction model for customer information that combines residual neural networks and attention mechanisms, to perform deep feature extraction on customer information to perform incremental updates to the initial intelligent customer relationship knowledge graph and determine the target intelligent customer relationship knowledge graph, may include: First, the customer entity attribute information and relationship data are normalized and then used as input data for the deep feature extraction model of customer information that combines residual neural network and attention mechanism; Second, by using the embedding vector corresponding to the input data generated by the Siamese network, multi-dimensional basic features are extracted, and then the attention mechanism is used to assign weights to the initial features of the customer in order to output the deep feature vector of the customer. Third, based on the customer's deep feature vector and combined with the new customer information in the enterprise customer database, perform incremental updates to the initial intelligent customer relationship knowledge graph; wherein, the new customer information is mapped to new knowledge graph nodes, directed edges and node attributes according to the optimized customer relationship ontology and added to the initial intelligent customer relationship knowledge graph. Fourth, evaluate the incrementally updated customer relationship knowledge graph. If the evaluation result meets the preset threshold, the current incrementally updated customer relationship knowledge graph is determined as the target intelligent customer relationship knowledge graph. The evaluation indicators include node completeness, relationship accuracy, and feature relevance.

[0068] Specifically, a deep feature extraction model for customer information, combining a residual neural network and an attention mechanism, is first constructed. This model aims to extract deep features from customer entity attribute information and relationship data to support incremental updates of the intelligent customer relationship knowledge graph. Then, the customer entity attribute information and relationship data are normalized to eliminate dimensional differences between different features, improving the model's training efficiency and accuracy. The normalized customer information is used as input data for the model combining the residual neural network and the attention mechanism. A Siamese network structure is used to encode the input data, generating corresponding embedding vectors. These embedding vectors capture the multi-dimensional basic features of customer information, providing a foundation for subsequent feature extraction. After extracting the multi-dimensional basic features, an attention mechanism is introduced to assign weights to the initial customer features. The attention mechanism dynamically adjusts the weights based on the importance of the features, making the model focus more on features that have a key impact on customer relationship analysis, thereby outputting more accurate deep customer feature vectors. Finally, based on the extracted deep customer feature vectors and combined with the newly added customer information in the enterprise customer database, the new customer information is mapped to new knowledge graph nodes, directed edges, and node attributes according to the optimized customer relationship ontology, and added to the initial intelligent customer relationship knowledge graph. An incremental update operation is then performed, integrating the new nodes, directed edges, and node attributes into the initial knowledge graph to form the updated customer relationship knowledge graph. The incrementally updated customer relationship knowledge graph is evaluated, with evaluation indicators including node completeness (whether the knowledge graph contains all relevant customer entities and attributes), relationship accuracy (whether the relationships in the knowledge graph accurately reflect the actual relationships between customers), and feature correlation (whether the features in the knowledge graph can be effectively correlated and analyzed). If the evaluation results meet a preset threshold, the current incrementally updated customer relationship knowledge graph is determined as the target intelligent customer relationship knowledge graph. This target knowledge graph can more accurately reflect the complex relationships between customers, providing strong support for the enterprise's customer relationship management.

[0069] In one optional implementation, if the evaluation result does not meet the preset threshold, the steps of performing data verification on the initial intelligent customer relationship knowledge graph, extracting deep features of customer information, and performing incremental updates on the initial intelligent customer relationship knowledge graph are re-executed.

[0070] For example, the input to a deep feature extraction model for customer information that combines residual neural networks and attention mechanisms is normalized data. Basic features are extracted through ResNet blocks. ,Right now: in, This represents the customer-based feature vector extracted through ResNet blocks. This represents the first-layer weight matrix, used to perform a linear transformation on the input features. This represents the second-layer weight matrix, used for further transformation of the gated features. This represents the first-level bias vector, used to adjust the results of the linear transformation. This represents the second-layer bias vector, used to further adjust the transformation result. The ReLU activation function is used. This represents the Sigmoid function.

[0071] Attention weights can be assigned in the following way: in, This represents a dynamic query vector that adaptively adjusts according to the feature distribution of customer i. In customer categories The logarithmic probability values ​​are then input into the cross-entropy function: The optimization strategy for the deep feature extraction model of customer information, which follows the combination of residual neural networks and attention mechanisms, is updated in the following ways: : in, It is the meta-learning rate. This indicates the number of meta-tasks in each iteration, which improves the computational power of the deep feature extraction model for customer information that combines residual neural networks with attention mechanisms.

[0072] Specifically, incremental updates to the initial intelligent customer relationship knowledge graph can be performed using an update triggering mechanism based on feature drift detection. in, This represents a score that measures the overall drift of the characteristic distribution compared to the previous time t-1. This represents the dimension of the customer's deep feature vector. This represents the total dimension of the deep feature vectors. This represents the mean of the d-th dimension of the deep features of all customers at time t. This represents the mean of the d-th dimension of the deep features of all customers at time t-1. This represents the standard deviation of the d-th dimension of the deep features of all customers at time t-1, used to normalize the amount of change. Let d represent the mean of the d-th feature. The standard deviation of the d-th dimension feature can be set when... Incremental update steps are triggered at certain times.

[0073] Alternatively, the association between new information and the existing graph can be determined by calculating similarity based on deep feature vectors: in, This represents the deep feature vector of a newly added customer. This represents the deep feature vector of existing customer i.

[0074] This invention achieves a technological breakthrough in building customer relationship knowledge graphs through the aforementioned method. Specifically, the knowledge graph update response time is reduced to the second level, and the data verification accuracy reaches 97.6%, significantly reducing the impact of erroneous data on decision-making. This application is particularly suitable for large enterprises with complex customer networks and large data volumes, providing accurate relationship insights for business scenarios such as customer segmentation, cross-selling, and risk warning.

[0075] As can be seen, this invention first obtains customer information from an enterprise customer database; then, based on this customer information, it constructs and optimizes a customer relationship ontology; according to the optimized customer relationship ontology, customer entities are mapped to knowledge graph nodes, the relationships between customer entities are mapped to directed edges between nodes, and entity attribute information is stored as knowledge graph node attributes, thus constructing an initial intelligent customer relationship knowledge graph; the initial intelligent customer relationship knowledge graph is then validated, and deep features of customer information are extracted, followed by incremental updates to determine the target intelligent customer relationship knowledge graph. This accurately reflects the complex relationships between customers, providing strong support for enterprise customer relationship management.

[0076] Another embodiment of this application provides an intelligent customer relationship knowledge graph construction system, such as... Figure 3 The diagram shown illustrates the structure of an intelligent customer relationship knowledge graph construction system, which includes: The module 301 is used to obtain customer information from the enterprise customer database; the customer information includes customer entities, the relationships between customer entities, and entity attribute information. The optimization module 302 is used to construct and optimize a customer relationship ontology based on the customer information. The customer relationship ontology is constructed using an improved K-means clustering algorithm, which determines initial cluster centers by calculating intra-cluster object compactness, inter-cluster distance, and data density. Based on the customer relationship ontology, customer entities and their relationships are divided into different semantic categories to form an optimized customer relationship ontology. The optimized customer relationship ontology is determined based on the semantics, relationships, and semantic completeness of the customer relationship ontology. The construction module 303 is used to map customer entities to knowledge graph nodes based on the optimized customer relationship ontology, map the relationships between customer entities to directed edges between nodes, and store entity attribute information as knowledge graph node attributes to construct an initial intelligent customer relationship knowledge graph. The determination module 304 is used to perform data verification on the initial intelligent customer relationship knowledge graph, extract deep features of customer information, and perform incremental updates on the initial intelligent customer relationship knowledge graph to determine the target intelligent customer relationship knowledge graph. Specifically, the initial intelligent customer relationship knowledge graph is verified by combining rule verification and data consistency verification. The rule verification is based on preset customer entity attribute constraint rules and association logic rules to filter out data with abnormal attribute values ​​and contradictory relationship logic. The data consistency verification compares the original data in the enterprise customer database with the customer information stored in the initial intelligent customer relationship knowledge graph to locate inconsistencies and mark and correct abnormal data.

[0077] Compared with existing technologies, this invention first obtains customer information from an enterprise customer database; then, based on this customer information, it constructs and optimizes a customer relationship ontology; according to the optimized customer relationship ontology, customer entities are mapped to knowledge graph nodes, the relationships between customer entities are mapped to directed edges between nodes, and entity attribute information is stored as knowledge graph node attributes, thus constructing an initial intelligent customer relationship knowledge graph; the initial intelligent customer relationship knowledge graph is then validated, and deep features of customer information are extracted, followed by incremental updates to determine the target intelligent customer relationship knowledge graph. This accurately reflects the complex relationships between customers, providing strong support for enterprise customer relationship management.

[0078] This invention also provides a storage medium storing a computer program, wherein the computer program is configured to implement the steps in the above method embodiments when running.

[0079] Specifically, in this embodiment, the storage medium can be configured to store a computer program for performing the following steps: S201: Obtain customer information from the enterprise customer database; the customer information includes customer entities, the relationships between customer entities, and entity attribute information; S202: Based on the customer information, construct and optimize the customer relationship ontology; wherein the customer relationship ontology is constructed using an improved K-means clustering algorithm, which determines the initial cluster centers by calculating the compactness of objects within clusters, the distance between clusters, and the data density; according to the customer relationship ontology, customer entities and the relationships between customer entities are divided into different semantic categories to form an optimized customer relationship ontology; the optimized customer relationship ontology is determined based on the semantics, relationships, and semantic completeness of the customer relationship ontology. S203: Based on the optimized customer relationship ontology, customer entities are mapped to knowledge graph nodes, the relationships between customer entities are mapped to directed edges between nodes, and entity attribute information is stored as knowledge graph node attributes to construct an initial intelligent customer relationship knowledge graph. S204: Perform data verification on the initial intelligent customer relationship knowledge graph, extract deep features of customer information, and perform incremental updates on the initial intelligent customer relationship knowledge graph to determine the target intelligent customer relationship knowledge graph; wherein, the initial intelligent customer relationship knowledge graph is verified by combining rule verification and data consistency verification; the rule verification is based on preset customer entity attribute constraint rules and association relationship logic rules to filter out data with abnormal attribute values ​​and contradictory relationship logic; the data consistency verification compares the original data in the enterprise customer database with the customer information stored in the initial intelligent customer relationship knowledge graph to locate data inconsistencies and mark and correct abnormal data.

[0080] Specifically, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0081] Compared with existing technologies, this invention first obtains customer information from an enterprise customer database; then, based on this customer information, it constructs and optimizes a customer relationship ontology; according to the optimized customer relationship ontology, customer entities are mapped to knowledge graph nodes, the relationships between customer entities are mapped to directed edges between nodes, and entity attribute information is stored as knowledge graph node attributes, thus constructing an initial intelligent customer relationship knowledge graph; the initial intelligent customer relationship knowledge graph is then validated, and deep features of customer information are extracted, followed by incremental updates to determine the target intelligent customer relationship knowledge graph. This accurately reflects the complex relationships between customers, providing strong support for enterprise customer relationship management.

[0082] This invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps described in the method embodiments above.

[0083] Specifically, the aforementioned electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the aforementioned processor, and the input / output device is connected to the aforementioned processor.

[0084] Specifically, in this embodiment, the processor can be configured to perform the following steps via a computer program: S201: Obtain customer information from the enterprise customer database; the customer information includes customer entities, the relationships between customer entities, and entity attribute information; S202: Based on the customer information, construct and optimize the customer relationship ontology; wherein the customer relationship ontology is constructed using an improved K-means clustering algorithm, which determines the initial cluster centers by calculating the compactness of objects within clusters, the distance between clusters, and the data density; according to the customer relationship ontology, customer entities and the relationships between customer entities are divided into different semantic categories to form an optimized customer relationship ontology; the optimized customer relationship ontology is determined based on the semantics, relationships, and semantic completeness of the customer relationship ontology. S203: Based on the optimized customer relationship ontology, customer entities are mapped to knowledge graph nodes, the relationships between customer entities are mapped to directed edges between nodes, and entity attribute information is stored as knowledge graph node attributes to construct an initial intelligent customer relationship knowledge graph. S204: Perform data verification on the initial intelligent customer relationship knowledge graph, extract deep features of customer information, and perform incremental updates on the initial intelligent customer relationship knowledge graph to determine the target intelligent customer relationship knowledge graph; wherein, the initial intelligent customer relationship knowledge graph is verified by combining rule verification and data consistency verification; the rule verification is based on preset customer entity attribute constraint rules and association relationship logic rules to filter out data with abnormal attribute values ​​and contradictory relationship logic; the data consistency verification compares the original data in the enterprise customer database with the customer information stored in the initial intelligent customer relationship knowledge graph to locate data inconsistencies and mark and correct abnormal data.

[0085] Compared with existing technologies, this invention first obtains customer information from an enterprise customer database; then, based on this customer information, it constructs and optimizes a customer relationship ontology; according to the optimized customer relationship ontology, customer entities are mapped to knowledge graph nodes, the relationships between customer entities are mapped to directed edges between nodes, and entity attribute information is stored as knowledge graph node attributes, thus constructing an initial intelligent customer relationship knowledge graph; the initial intelligent customer relationship knowledge graph is then validated, and deep features of customer information are extracted, followed by incremental updates to determine the target intelligent customer relationship knowledge graph. This accurately reflects the complex relationships between customers, providing strong support for enterprise customer relationship management.

[0086] 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 the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. 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 the present invention.

[0087] 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.

[0088] In the several embodiments provided by this invention, 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 the units described above 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 coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0089] The units described above 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.

[0090] Furthermore, the functional units in the various embodiments of the present invention 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.

[0091] If the aforementioned integrated units are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this invention, 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 invention. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0092] The embodiments of the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for constructing an intelligent customer relationship knowledge graph, characterized in that, The method includes: Obtain customer information from the enterprise customer database; the customer information includes customer entities, the relationships between customer entities, and entity attribute information. Based on the customer information, a customer relationship ontology is constructed and optimized. This ontology is constructed using an improved K-means clustering algorithm, which determines initial cluster centers by calculating intra-cluster object compactness, inter-cluster distance, and data density. According to the customer relationship ontology, customer entities and their relationships are categorized into different semantic categories, forming an optimized customer relationship ontology. The optimized customer relationship ontology is determined based on the semantics, relationships, and semantic completeness of the customer relationship ontology. Based on the optimized customer relationship ontology, customer entities are mapped to knowledge graph nodes, the relationships between customer entities are mapped to directed edges between nodes, and entity attribute information is stored as knowledge graph node attributes to construct an initial intelligent customer relationship knowledge graph. The initial intelligent customer relationship knowledge graph is validated, and deep features of customer information are extracted. Incremental updates are then performed on the initial intelligent customer relationship knowledge graph to determine the target intelligent customer relationship knowledge graph. Specifically, the initial intelligent customer relationship knowledge graph is validated using a combination of rule validation and data consistency validation. The rule validation, based on preset customer entity attribute constraint rules and relational logic rules, filters out data with abnormal attribute values ​​and contradictory relational logic. The data consistency validation compares the original data in the enterprise customer database with the customer information stored in the initial intelligent customer relationship knowledge graph to locate inconsistencies and marks and corrects abnormal data.

2. The method according to claim 1, characterized in that, The customer information obtained from the enterprise customer database includes: Identify customer entities in the enterprise's customer database; A hybrid algorithm combining a customer entity vectorization model and a bidirectional long short-term memory network is used to extract the relationships between customer entities to determine customer information; wherein, the relationships include business cooperation relationships, demand and supply relationships, and service interaction relationships.

3. The method according to claim 2, characterized in that, The hybrid algorithm, which combines a customer entity vectorization model with a bidirectional long short-term memory network, extracts the relationships between customer entities to determine customer information, including: Construct an initial vectorized model of the customer entity; The context window consisting of n words before and after the customer data keywords is used as the input layer word sequence of the customer entity vectorization model. The input layer word sequence is represented by one-hot encoding, and the probability distribution of the predicted words is used as the output vector. The cross-entropy loss function is defined, and the parameters of the initial customer entity vectorization model are iteratively updated by the stochastic gradient descent method to obtain the trained customer entity vectorization model. The normalized customer entity data is input into the trained customer entity vectorization model to obtain word vector representations, which are then input into a bidirectional long short-term memory network to obtain the dependencies between customer entities. Based on the dependencies between the customer entities, the associations between the customer entities are extracted to determine customer information.

4. The method according to claim 3, characterized in that, The customer relationship ontology is determined using an improved K-means clustering algorithm, and the cluster centers are dynamically adjusted using the following formula: in, It indicates that at the (t+1)th iteration, the... The new cluster center of each cluster, In the t-th iteration, the first... The new cluster center of each cluster, Indicates the learning rate. This represents the first clustering step partitioned by the improved K-means clustering algorithm. A subset of data, Indicates the number of objects in the cluster. Represents customer data points Local density, Represents the inter-cluster repulsion coefficient. Cluster with cluster distance, Cluster with cluster distance, It indicates that at the t-th iteration, the... Cluster centers of each cluster It indicates that at the t-th iteration, the... Cluster centers of each cluster.

5. The method according to claim 4, characterized in that, The process of performing data verification on the initial intelligent customer relationship knowledge graph, extracting deep features of customer information, and performing incremental updates on the initial intelligent customer relationship knowledge graph to determine the target intelligent customer relationship knowledge graph includes: A deep feature extraction model for customer information, combining residual neural networks and attention mechanisms, is constructed to extract deep features from customer information in order to perform incremental updates on the initial intelligent customer relationship knowledge graph and determine the target intelligent customer relationship knowledge graph.

6. The method according to claim 5, characterized in that, The aforementioned construction of a deep feature extraction model for customer information, combining residual neural networks and attention mechanisms, performs deep feature extraction on customer information to perform incremental updates to the initial intelligent customer relationship knowledge graph and determine the target intelligent customer relationship knowledge graph, including: After normalizing the customer entity attribute information and relationship data, the data is used as the input data for the deep feature extraction model of customer information that combines residual neural network and attention mechanism; By using the embedding vector corresponding to the input data generated by the Siamese network, multi-dimensional basic features are extracted, and then the attention mechanism is used to assign weights to the initial features of the customer in order to output the deep feature vector of the customer. Based on the customer's deep feature vector and combined with the new customer information in the enterprise customer database, an incremental update is performed on the initial intelligent customer relationship knowledge graph; wherein, the new customer information is mapped into new knowledge graph nodes, directed edges and node attributes according to the optimized customer relationship ontology and added to the initial intelligent customer relationship knowledge graph. The incrementally updated customer relationship knowledge graph is evaluated. If the evaluation result meets the preset threshold, the current incrementally updated customer relationship knowledge graph is determined as the target intelligent customer relationship knowledge graph. The evaluation indicators include node completeness, relationship accuracy, and feature relevance.

7. The method according to claim 6, characterized in that, The method further includes: If the evaluation result does not meet the preset threshold, the steps of performing data verification on the initial intelligent customer relationship knowledge graph, extracting deep features of customer information, and performing incremental updates on the initial intelligent customer relationship knowledge graph are repeated.

8. An intelligent customer relationship knowledge graph construction system, characterized in that, The system includes: The acquisition module is used to acquire customer information from the enterprise customer database; the customer information includes customer entities, the relationships between customer entities, and entity attribute information. An optimization module is used to construct and optimize a customer relationship ontology based on the customer information. The customer relationship ontology is constructed using an improved K-means clustering algorithm, which determines initial cluster centers by calculating intra-cluster object compactness, inter-cluster distance, and data density. Based on the customer relationship ontology, customer entities and their relationships are categorized into different semantic categories to form an optimized customer relationship ontology. The optimized customer relationship ontology is determined based on the semantics, relationships, and semantic completeness of the customer relationship ontology. The construction module is used to map customer entities to knowledge graph nodes based on the optimized customer relationship ontology, map the relationships between customer entities to directed edges between nodes, and store entity attribute information as knowledge graph node attributes to build an initial intelligent customer relationship knowledge graph. The determination module is used to perform data verification on the initial intelligent customer relationship knowledge graph, extract deep features of customer information, and perform incremental updates on the initial intelligent customer relationship knowledge graph to determine the target intelligent customer relationship knowledge graph. Specifically, the initial intelligent customer relationship knowledge graph is verified using a combination of rule verification and data consistency verification. The rule verification is based on preset customer entity attribute constraint rules and association logic rules to filter out data with abnormal attribute values ​​and contradictory relationship logic. The data consistency verification compares the original data in the enterprise customer database with the customer information stored in the initial intelligent customer relationship knowledge graph to locate inconsistencies and mark and correct abnormal data.

9. A storage medium, characterized in that, The storage medium stores a computer program, wherein the computer program is configured to implement the method of any one of claims 1 to 7 when it is run.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to implement the method of any one of claims 1 to 7.

Images