Intelligent customer relationship knowledge graph construction method

Abstract

The invention relates to the technical field of knowledge graph construction, in particular to an intelligent customer relationship knowledge graph construction method, which comprises the following steps of: firstly, acquiring customer data from a multi-source data source; preprocessing the collected data, and protecting sensitive information by adopting a differential privacy technology; defining a mode of a knowledge graph, identifying an entity by adopting a BERT model, extracting a cross-domain relationship in combination with a graph neural network, and constructing a preliminary customer relationship knowledge graph; designing an incremental updating algorithm, and updating the knowledge graph in real time or regularly; based on the constructed customer relationship knowledge graph, a Cypher query language and a graph algorithm are applied to mine customer relationships; performing deep mining on the customer relationship by using a graph algorithm, and identifying key customers, potential customer groups and association relationships among the customers; in this way, the technical problems that in the prior art, when the knowledge graph is constructed, the data processing efficiency is low, relation extraction is inaccurate, and the knowledge graph is not updated in time are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of knowledge graph construction, and in particular to a method for constructing an intelligent customer relationship knowledge graph. Background Art

[0002] In the era of rapid development of big data and artificial intelligence, customer relationship management (CRM) has become a key area for enterprises to enhance their competitiveness. Traditional CRM systems mainly focus on data storage, query and simple statistical analysis, which makes it difficult to fully explore and utilize the potential value of customer data. With the diversification of data sources, including social media, online transaction records, customer service records, etc., enterprises are facing increasing challenges in data integration and analysis. Therefore, how to efficiently integrate multi-source customer data and build a comprehensive, accurate and dynamic customer relationship knowledge graph has become an important issue that needs to be solved in the current CRM field.

[0003] In response to the above problems, existing technologies have attempted to use knowledge graph technology to integrate and analyze customer data. However, the existing methods of constructing knowledge graphs have problems such as low data processing efficiency, inaccurate relationship extraction, and untimely knowledge graph updates. Summary of the invention

[0004] The purpose of the present invention is to provide a method for constructing an intelligent customer relationship knowledge graph, aiming to solve the technical problems of low data processing efficiency, inaccurate relationship extraction, and untimely knowledge graph updating when constructing knowledge graphs in the prior art.

[0005] To achieve the above purpose, the present invention adopts a method for constructing an intelligent customer relationship knowledge graph, comprising the following steps:

[0006] First, customer data is collected from multiple data sources based on API interfaces and crawler technology;

[0007] Clean, standardize and process the collected data using NLP, and use differential privacy technology to protect sensitive information;

[0008] Define the model of the knowledge graph, including entities and relationships, use the BERT model to identify entities, combine with the graph neural network (GNN) to extract cross-domain relationships (such as "customer A clicks on an ad → prefers product B → associates with corporate service C"), and build a preliminary customer relationship knowledge graph;

[0009] Design incremental update algorithms to update knowledge graphs in real time or periodically;

[0010] Based on the constructed customer relationship knowledge graph, use graph algorithms (such as PageRank, shortest path algorithm, clustering algorithm, etc.) to deeply mine customer relationships and identify key customers, potential customer groups, and relationships between customers;

[0011] Combining the XGBoost and LSTM models to predict customer purchase intent and service requirements.

[0012] Among them, when collecting customer data from multi-source data sources, it includes collecting user behavior data from advertising platforms (such as aggregated advertising SDKs), including advertising click-through rate, conversion rate, etc.;

[0013] Collecting health monitoring data and service request records from enterprise service systems;

[0014] Collecting unstructured data from social media platforms, where the unstructured data includes user comments and interaction behaviors.

[0015] Among them, when cleaning and standardizing the collected data:

[0016] First, remove duplicate data. The specific methods are as follows: identify and delete duplicate data records by comparing the unique identifiers or key fields of the data; use data cleaning tools or write scripts (such as the Pandas library in Python) for deduplication operations;

[0017] Then handle missing values. The specific methods are as follows: delete records containing missing values (suitable for cases with fewer missing values); fill in missing values using statistical methods (such as mean, median, mode, etc.); predict missing values using machine learning algorithms (such as K-nearest neighbor algorithm, decision tree, etc.);

[0018] Subsequently, correct incorrect data. The specific methods are as follows: identify and correct incorrect data by setting data verification rules (such as data type, value range, etc.); for outliers, use statistical methods (such as box plot analysis) or machine learning algorithms (such as isolation forest, DBSCAN, etc.) for detection and processing;

[0019] Then unify the data format, data unit, and data encoding of the data.

[0020] Among them, the specific method of using differential privacy technology to protect sensitive information is as follows:

[0021] When performing data cleaning, differential privacy statistical methods can be used to estimate the distribution of the data, so as to perform data cleaning without leaking sensitive information;

[0022] At the same time, when using the BERT model to identify entities and the graph neural network (GNN) to extract cross-domain relationships, differential privacy technology is also used to train the model.

[0023] Among them, when designing an incremental update algorithm to update the knowledge graph in real time or regularly:

[0024] By chunking new data, using the hash algorithm to filter duplicate data, adopting the BERT model and GNN to extract entities and relationships, and leveraging the upsert operation to seamlessly integrate new information into the existing knowledge graph, the real-time or periodic efficient update of the knowledge graph is achieved.

[0025] Among them, during the incremental update process, differential privacy technology is adopted to protect sensitive information; for example, when updating customer data, sensitive information can be perturbed and noise can be added to protect customer privacy.

[0026] Among them, when using graph algorithms to deeply mine customer relationships and identify key customers, potential customer groups, and the association relationships between customers:

[0027] Use graph algorithms to analyze the customer relationship knowledge graph, identify key customers through the PageRank algorithm, discover potential associations between customers through the shortest path algorithm, and divide potential customer groups through the clustering algorithm, so as to achieve the deep mining of customer relationships.

[0028] Among them, when combining the XGBoost and LSTM models to predict customer purchase intentions and service requirements:

[0029] After collecting customer data and extracting relevant features, train the XGBoost model to capture static and short-term behavior patterns, and the LSTM model to capture long-term behavior patterns and time series dependencies respectively. Finally, fuse the two for prediction, so as to accurately identify customer purchase intentions and service requirements.

[0030] Among them, before starting data collection, deeply evaluate potential data sources and select those that can provide high-quality and relevant data; at the same time, according to business requirements and data characteristics, optimize the usage strategy of API interfaces to improve the efficiency and accuracy of data collection.

[0031] Among them, when applying the XGBoost and LSTM models to predict customer purchase intentions and service requirements, conduct model evaluation, including the calculation of indicators such as accuracy and recall; according to the evaluation results, optimize the model to improve the accuracy and reliability of prediction;

[0032] The specific method is as follows: when applying the XGBoost and LSTM models to predict customer purchase intentions and service requirements, evaluate the model performance by calculating accuracy and recall indicators, and adjust the model parameters, optimize feature selection and scaling, and adopt cross-validation and early stopping strategy methods for optimization to improve the accuracy and reliability of prediction.

[0033] An intelligent customer relationship knowledge graph construction method of the present invention, when specifically used, first collects customer data from multi-source data sources based on API interfaces and web crawler technology; cleans, standardizes, and performs NLP processing on the collected data, and at the same time uses differential privacy technology to protect sensitive information; defines the schema of the knowledge graph, including entities and relationships, uses the BERT model to identify entities, combines graph neural networks to extract cross-domain relationships, and constructs a preliminary customer relationship knowledge graph; designs an incremental update algorithm to update the knowledge graph in real time or regularly; based on the constructed customer relationship knowledge graph, applies the Cypher query language and graph algorithms to mine customer relationships; uses graph algorithms to deeply mine customer relationships, identifies key customers, potential customer groups, and the association relationships between customers; combines the XGBoost and LSTM models to predict customer purchase intentions and service requirements, thereby solving the technical problems of low data processing efficiency, inaccurate relationship extraction, and untimely knowledge graph update existing in the prior art when constructing a knowledge graph. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0035] Figure 1 It is a flowchart of the intelligent customer relationship knowledge graph construction method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] The following will describe in detail the embodiments of the present invention. The examples of the embodiments are shown in the drawings. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present invention and should not be construed as a limitation of the present invention.

[0037] Please refer to Figure 1 , Figure 1 It is a flowchart of the intelligent customer relationship knowledge graph construction method of the present invention.

[0038] The present invention provides an intelligent customer relationship knowledge graph construction method, including the following steps:

[0039] S1. First, collect customer data from multi-source data sources based on API interfaces and web crawler technology;

[0040] For this specific embodiment, when collecting customer data from multi-source data sources, it includes collecting user behavior data from advertising platforms (such as aggregated advertising SDK), including advertising click-through rate, conversion rate, etc.;

[0041] Collect health monitoring data and service request records from the enterprise service system;

[0042] Collect unstructured data from social media platforms, where the unstructured data includes user comments and interaction behaviors.

[0043] S2. Clean, standardize, and perform NLP processing on the collected data, and at the same time use differential privacy technology to protect sensitive information;

[0044] For this specific embodiment, when cleaning and standardizing the collected data:

[0045] First, remove duplicate data. The specific method is as follows: Identify and delete duplicate data records by comparing the unique identifiers or key fields of the data; Use data cleaning tools or write scripts (such as the Pandas library in Python) for deduplication operations;

[0046] Then handle missing values. The specific method is as follows: Delete records containing missing values (suitable for cases with fewer missing values); Use statistical methods (such as mean, median, mode, etc.) to fill in missing values; Use machine learning algorithms (such as K-nearest neighbor algorithm, decision tree, etc.) to predict missing values;

[0047] Subsequently, correct incorrect data. The specific method is as follows: Identify and correct incorrect data by setting data verification rules (such as data type, value range, etc.); For outliers, use statistical methods (such as box plot analysis) or machine learning algorithms (such as isolation forest, DBSCAN, etc.) for detection and processing;

[0048] Then unify the data format, data unit, and data encoding of the data.

[0049] When performing data cleaning, statistical methods of differential privacy can be used to estimate the distribution of the data, so as to perform data cleaning without revealing sensitive information;

[0050] At the same time, when using the BERT model to identify entities and the graph neural network (GNN) to extract cross-domain relationships, differential privacy technology is also used to train the model.

[0051] S3. Define the schema of the knowledge graph, including entities and relationships, use the BERT model to identify entities, combine the graph neural network (GNN) to extract cross-domain relationships (such as "Customer A clicks through an advertisement → prefers product B → is associated with enterprise service C"), and construct a preliminary customer relationship knowledge graph;

[0052] For this specific embodiment, when constructing the structure of the knowledge graph, entities (such as customers, products, services, etc.) and the relationships between them are clearly defined. To accurately capture these entities, the BERT model is adopted for efficient entity recognition. Further, the present invention combines the powerful capabilities of graph neural networks (GNNs) to deeply analyze and extract complex cross-domain relationships, such as the behavior chain of customers ("Customer A shows a preference for a specific product B through an advertisement click behavior → and then is associated with the demand for enterprise service C"). Based on this detailed information, a preliminary customer relationship knowledge graph is successfully constructed, which not only covers rich entity information but also deeply reveals the multi-dimensional associations between customers.

[0053] S4. Design an incremental update algorithm to update the knowledge graph in real time or periodically;

[0054] For this specific embodiment, by processing new data in chunks, using the hash algorithm to filter duplicate data, adopting the BERT model and GNN to extract entities and relationships, and using the upsert operation to seamlessly integrate new information into the existing knowledge graph, the real-time or periodic and efficient update of the knowledge graph is achieved;

[0055] In the incremental update process, differential privacy technology is adopted to protect sensitive information; for example, when updating customer data, sensitive information can be perturbed and noise can be added to protect customer privacy.

[0056] S5. Based on the constructed customer relationship knowledge graph, use graph algorithms (such as PageRank, shortest path algorithm, clustering algorithm, etc.) to deeply mine customer relationships, and identify key customers, potential customer groups, and the association relationships between customers;

[0057] For this specific embodiment, use graph algorithms to analyze the customer relationship knowledge graph, identify key customers through the PageRank algorithm, discover potential associations between customers through the shortest path algorithm, and divide potential customer groups through the clustering algorithm, so as to achieve the deep mining of customer relationships.

[0058] S6. Combine the XGBoost and LSTM models to predict customer purchase intentions and service requirements.

[0059] For this specific embodiment, after collecting customer data and extracting relevant features, train the XGBoost model to capture static and short-term behavior patterns, and the LSTM model to capture long-term behavior patterns and time series dependencies respectively, and finally fuse the two for prediction, so as to accurately identify customer purchase intentions and service requirements.

[0060] Before starting data collection, conduct an in-depth evaluation of potential data sources and select those that can provide high-quality and relevant data. At the same time, optimize the usage strategy of the API interface according to business requirements and data characteristics to improve the efficiency and accuracy of data collection.

[0061] Among them, when applying the XGBoost and LSTM models to predict customer purchase intentions and service requirements, conduct model evaluation, including calculating metrics such as accuracy and recall rate. According to the evaluation results, optimize the model to improve the accuracy and reliability of the prediction.

[0062] The specific methods are as follows: When applying the XGBoost and LSTM models to predict customer purchase intentions and service requirements, evaluate the model performance by calculating accuracy and recall rate metrics, and adjust the model parameters, optimize feature selection and scaling, and adopt cross-validation and early stopping strategy methods for optimization to improve the accuracy and reliability of the prediction.

[0063] At the same time, in the process of multi-source data collection, adopt federated learning technology so that different data sources can jointly train the model without sharing the original data, further protecting data privacy and security.

[0064] Furthermore, in view of the complexity of customer relationships, design a more efficient graph neural network structure, such as a graph neural network introducing an attention mechanism, to better capture the complex relationships and important features among customers.

[0065] The present invention also provides the application of enhanced graph algorithms: In addition to the existing PageRank, shortest path algorithm, and clustering algorithm, introduce more advanced graph algorithms, such as graph attention network (GAT) and graph convolutional network (GCN) in graph neural networks, to further improve the accuracy and depth of customer relationship analysis.

[0066] On the basis of combining the XGBoost and LSTM models, introduce more advanced deep learning models, such as the Transformer model, to better process time series data and long sequence dependencies and improve the accuracy of prediction.

[0067] During the construction and update process of the knowledge graph, add a user feedback mechanism so that users can evaluate and provide feedback on the construction results of the knowledge graph, further optimizing the quality and accuracy of the knowledge graph.

[0068] Apply the method of the present invention to more business scenarios, such as customer churn prediction, customer value assessment, customer segmentation, etc., to further improve its practicality and universality.

[0069] The present invention has the following beneficial effects:

[0070] Through highly integrated API interfaces and advanced web crawler technologies, this solution demonstrates powerful data acquisition capabilities, enabling it to quickly and efficiently gather customer data from diverse data sources. This strategy not only significantly shortens the data collection cycle, allowing enterprises to grasp market trends and customer information faster, but also ensures the comprehensiveness and timeliness of the data.

[0071] After data collection is completed, cutting-edge data cleaning, standardization, and natural language processing (NLP) technologies are further employed to deeply preprocess the raw data. The data cleaning step effectively removes redundant, incorrect, or inconsistent information, laying a solid foundation for subsequent analysis. Data standardization ensures seamless integration of data from different sources, enhancing data interoperability and comparability. Meanwhile, the introduction of NLP technology enables us to understand and analyze the deep meaning in text data, such as customer reviews and social media posts, thus uncovering more valuable customer insights.

[0072] Enhance the accuracy of relationship extraction: When constructing the customer relationship knowledge graph, this solution uses the BERT model for entity recognition and combines it with graph neural networks for cross-domain relationship extraction. The powerful capabilities of the BERT model in the field of natural language processing and the advantages of graph neural networks in processing complex relationship data significantly improve the accuracy of relationship extraction. This helps to build a more accurate and comprehensive customer relationship knowledge graph.

[0073] Achieve real-time updates of the knowledge graph: The incremental update algorithm designed in this solution can update the knowledge graph in real-time or periodically, ensuring that the information in the graph always remains consistent with the actual situation. This feature is crucial for capturing market dynamics and responding to customer needs in a timely manner, effectively solving the problem of untimely updates of knowledge graphs in existing technologies. By processing new data in chunks, using the hash algorithm to filter duplicate data, employing the BERT model and GNN to extract entities and relationships, and leveraging the upsert operation to seamlessly integrate new information into the existing knowledge graph, real-time or periodic and efficient updates of the knowledge graph are achieved. Meanwhile, during the incremental update process, differential privacy technology is used to protect sensitive information, ensuring the privacy and security of customer data.

[0074] Deeply mine customer relationships: Based on the constructed customer relationship knowledge graph, this solution uses the Cypher query language and graph algorithms to deeply mine customer relationships. By identifying key customers, potential customer groups, and the relationships between customers, it provides valuable market insights and decision-making support for enterprises. This not only helps enterprises optimize their customer management strategies but also enhances customer satisfaction and loyalty.

[0075] Accurately predict customers' purchase intentions and service needs: By combining the XGBoost and LSTM models, this solution can accurately predict customers' purchase intentions and service needs. These prediction results provide enterprises with forward-looking market forecasts and customer demand analyses, helping enterprises to layout the market in advance, optimize product and service strategies, and thus gain a competitive advantage.

[0076] This invention uses differential privacy technology to protect sensitive information. Differential privacy technology is used in the processes of data cleaning, standardization, and model training to ensure the privacy and security of customer data.

[0077] The constructed knowledge graph not only covers rich entity information but also deeply reveals the multi-dimensional associations among customers, providing a detailed information basis for subsequent customer relationship analysis.

[0078] Combine the XGBoost and LSTM models to predict customers' purchase intentions and service needs. After collecting customer data and extracting relevant features, train the XGBoost model to capture static and short-term behavior patterns, and the LSTM model to capture long-term behavior patterns and time series dependencies. Finally, fuse the two for prediction, so as to accurately identify customers' purchase intentions and service needs.

[0079] The prediction results can provide enterprises with accurate marketing and personalized service suggestions, helping enterprises to better meet customer needs and improve customer satisfaction and loyalty.

[0080] When using a method for constructing an intelligent customer relationship knowledge graph of this invention, in specific use, first collect customer data from multi-source data sources based on API interfaces and web crawler technology; clean, standardize, and perform NLP processing on the collected data, and at the same time use differential privacy technology to protect sensitive information; define the schema of the knowledge graph, including entities and relationships, use the BERT model to identify entities, combine graph neural networks to extract cross-domain relationships, and construct a preliminary customer relationship knowledge graph; design an incremental update algorithm to update the knowledge graph in real-time or regularly; based on the constructed customer relationship knowledge graph, apply the Cypher query language and graph algorithms to mine customer relationships; use graph algorithms to deeply mine customer relationships, identify key customers, potential customer groups, and the association relationships among customers; combine the XGBoost and LSTM models to predict customers' purchase intentions and service needs. In this way, the technical problems of low data processing efficiency, inaccurate relationship extraction, and untimely knowledge graph update existing in the prior art when constructing a knowledge graph are solved.

[0081] The above-disclosed is only a preferred embodiment of the present invention. Of course, it cannot be used to limit the scope of the rights of the present invention. Those of ordinary skill in the art can understand all or part of the processes of implementing the above embodiments, and the equivalent changes made according to the claims of the present invention still fall within the scope covered by the invention.

Claims

1. A method for constructing an intelligent customer relationship knowledge graph, characterized in that: The steps include: First, customer data is collected from multiple data sources based on API interfaces and crawler technology; Clean, standardize and process the collected data using NLP, and use differential privacy technology to protect sensitive information; Define the knowledge graph model, including entities and relationships, use the BERT model to identify entities, combine graph neural networks to extract cross-domain relationships, and build a preliminary customer relationship knowledge graph; Design incremental update algorithms to update knowledge graphs in real time or periodically; Based on the constructed customer relationship knowledge graph, we use graph algorithms to conduct in-depth mining of customer relationships and identify key customers, potential customer groups, and relationships between customers. Combine XGBoost and LSTM models to predict customer purchase intention and service needs.

2. The method for constructing an intelligent customer relationship knowledge graph according to claim 1, characterized in that: When collecting customer data from multiple data sources, including collecting user behavior data from advertising platforms, including ad click-through rates, conversion rates, etc.; Collect health monitoring data and service request records from enterprise service systems; Collect unstructured data from social media platforms, including user comments and interactive behaviors.

3. The method for constructing an intelligent customer relationship knowledge graph according to claim 2, characterized in that: When cleaning and standardizing the collected data: First, remove duplicate data by: identifying and deleting duplicate data records by comparing the unique identifiers or key fields of the data; using data cleaning tools or writing scripts to perform deduplication operations; Then, missing values ​​are processed by: deleting records containing missing values; using statistical methods to fill missing values; using machine learning algorithms to predict missing values; The erroneous data is then corrected in the following ways: by setting data verification rules, erroneous data is identified and corrected; for outliers, statistical methods or machine learning algorithms are used to detect and process them; Then the data format, data unit and data encoding are unified.

4. The method for constructing an intelligent customer relationship knowledge graph according to claim 3, characterized in that: The specific ways to protect sensitive information using differential privacy technology are as follows: When performing data cleaning, statistical methods of differential privacy can be used to estimate the distribution of data, thereby performing data cleaning without leaking sensitive information; At the same time, when using the BERT model to identify entities and graph neural networks to extract cross-domain relationships, differential privacy technology is also used to train the model.

5. The method for constructing an intelligent customer relationship knowledge graph according to claim 4, characterized in that: When designing an incremental update algorithm to update the knowledge graph in real time or periodically: By processing new data in blocks, using the hash algorithm to filter duplicate data, using the BERT model and GNN to extract entities and relationships, and using the upsert operation to seamlessly integrate new information into the existing knowledge graph, real-time or regular and efficient updating of the knowledge graph can be achieved.

6. The method for constructing an intelligent customer relationship knowledge graph according to claim 5, characterized in that: During the incremental update process, differential privacy technology is used to protect sensitive information.

7. The method for constructing an intelligent customer relationship knowledge graph according to claim 6, characterized in that: Based on the constructed customer relationship knowledge graph, we use graph algorithms to deeply mine customer relationships and identify key customers, potential customer groups, and relationships between customers: Graph algorithms are used to analyze customer relationship knowledge graphs, PageRank algorithms are used to identify key customers, shortest path algorithms are used to discover potential connections between customers, and clustering algorithms are used to divide potential customer groups, thereby achieving in-depth mining of customer relationships.

8. The method for constructing an intelligent customer relationship knowledge graph according to claim 7, characterized in that: When combining XGBoost and LSTM models to predict customer purchase intentions and service needs: After extracting relevant features from the collected customer data, we train the XGBoost model to capture static and short-term behavior patterns, and the LSTM model to capture long-term behavior patterns and time series dependencies. Finally, we combine the two for prediction, thereby accurately identifying customers’ purchasing intentions and service needs.

9. The method for constructing an intelligent customer relationship knowledge graph according to claim 8, characterized in that: Before starting data collection, conduct an in-depth evaluation of potential data sources and select those that can provide high-quality, highly relevant data; at the same time, optimize the use strategy of the API interface based on business needs and data characteristics to improve the efficiency and accuracy of data collection.

10. The method for constructing an intelligent customer relationship knowledge graph according to claim 9, characterized in that: When applying XGBoost and LSTM models to predict customer purchase intentions and service needs, model evaluation is performed, including calculations of indicators such as accuracy and recall. Based on the evaluation results, the model is tuned to improve the accuracy and reliability of the prediction. The specific method is as follows: When applying XGBoost and LSTM models to predict customer purchase intentions and service needs, the model performance is evaluated by calculating the accuracy and recall rate indicators, and the model parameters are adjusted according to the evaluation results, feature selection and scaling are optimized, and cross-validation and early stopping strategy methods are used for tuning to improve the accuracy and reliability of the prediction.

Images