Personalized customer service method and system based on distributed intelligent knowledge management

Abstract

The invention discloses a personalized customer service method and system based on distributed intelligent knowledge management, and relates to the technical field of customer services, and the method comprises the steps: collecting customer information, verifying the identity of a customer, receiving customer request data, and carrying out the preprocessing of the customer request data, and generating a customer portrait; and performing multi-modal sentiment analysis and problem classification based on the preprocessed client request data, constructing a dynamic knowledge graph based on a classification result, and performing knowledge graph query. Through combination of a HowNet emotion dictionary, a capsule network dynamic routing mechanism and a Skip-gram model, customer emotions can be accurately captured, optimized emotion features can be generated, customer request features are modeled in combination with a graph neural network, knowledge graph nodes are dynamically updated and expanded, support is provided for quick response of complex problems, and the method has the advantages of being high in practicability and easy to popularize. The personalized service scheme is generated through the Seq2Seq model, the service content is dynamically adjusted according to the emotional state and the historical behavior habit of the customer, and the customer satisfaction and the service efficiency are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of customer service, and in particular to a method and system for personalized customer service based on distributed intelligent knowledge management. Background Art

[0002] With the rapid development of artificial intelligence (AI) and big data technology, personalized services are increasingly being used in various industries. In the field of customer service, personalized service methods can not only improve customer satisfaction, but also significantly improve service efficiency and resource utilization. Traditional customer service systems mostly adopt a static rule-driven approach to solve customer problems based on predefined processes and limited knowledge bases. However, with the diversification and complexity of customer needs, this approach has gradually shown its limitations. In recent years, intelligent customer service systems based on knowledge graphs, sentiment analysis, and distributed computing technologies have gradually become a research hotspot. Such systems achieve structured management of knowledge by introducing knowledge graphs, and dynamically understand customer emotions and needs by combining multimodal sentiment analysis, laying the foundation for personalized service responses. At the same time, with the maturity of blockchain technology, distributed identity authentication and data storage solutions provide guarantees for the security and consistency of customer information. Although relevant technologies have made great progress, existing intelligent customer service methods still have many shortcomings. For example, the accuracy of customer portrait generation is low, and it is difficult to fully mine the multimodal features of customer historical data and current requests. The construction and update speed of knowledge graphs is slow, and it is impossible to dynamically respond to new problem categories. The generation of service response content mostly adopts a template method, which lacks the ability to optimize customer emotional states. Therefore, current intelligent customer service technology urgently needs to be further optimized and improved in data processing, knowledge management, and personalized response. Summary of the invention

[0003] In view of the above problems existing in the existing personalized customer service method and system based on distributed intelligent knowledge management, the present invention is proposed.

[0004] Therefore, the present invention provides a personalized customer service method based on distributed intelligent knowledge management to solve the problem that the current intelligent customer service technology urgently needs to be further optimized and improved in data processing, knowledge management and personalized response.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] In a first aspect, the present invention provides a personalized customer service method based on distributed intelligent knowledge management, which comprises:

[0007] Collect customer information to verify customer identity, receive customer request data and generate customer portraits after pre-processing;

[0008] Perform multimodal sentiment analysis and question classification based on preprocessed customer request data, build a dynamic knowledge graph based on the classification results, and perform knowledge graph query;

[0009] Generate personalized service plans based on query results and collect customer feedback for service optimization;

[0010] The analytical data generated during the service process is stored in a distributed database.

[0011] As a preferred solution of the personalized customer service method based on distributed intelligent knowledge management described in the present invention, wherein: the collecting of customer information to verify customer identity, receiving customer request data and generating a customer portrait after pre-processing refers to collecting customer information that sends request data through APP, voice assistant and intelligent customer service robot, extracting customer identity identifier, calculating hash value through SHA256 algorithm, submitting hash value to blockchain smart contract for verification, if verification fails, returning prompt information of identity authentication failure, if verification succeeds, receiving request data submitted by the verified customer and collecting customer history data;

[0012] The request data includes voice data and text data;

[0013] The historical data includes the customer's service records and behavioral preferences;

[0014] Convert voice data into text data and integrate it with the collected text data. Standardize the integrated text data. Use the word segmentation tool to segment the text into word sequences to obtain the text sequence after word segmentation. Input the text sequence and historical service records into the pre-trained BERT model to obtain semantic vector features. One-hot encode the historical behavior preferences and convert the behavior preferences into vector form.

[0015] The semantic vector features of the current request data are spliced ​​with the features of the historical data to form a unified feature vector. The final feature vector after integration is classified into a behavior feature group, a service feature group, and a request feature group. A structured representation of the customer portrait is generated through feature grouping. The generated customer portrait is stored in a distributed database and synchronized among distributed nodes.

[0016] The behavior feature group includes customer behavior preferences, the service feature group includes customer historical service records, and the request feature group includes semantic vector features of the current request.

[0017] As a preferred solution of the personalized customer service method based on distributed intelligent knowledge management described in the present invention, wherein: the multimodal sentiment analysis and question classification based on the pre-processed customer request data refers to introducing the HowNet sentiment dictionary, retrieving positive and negative words in the text, assigning initial sentiment weights W to the words, matching the text words with positive and negative seed words, and calculating the cosine similarity B between each word and the sentiment seed word;

[0018] Calculate the sentiment weight R of each word based on cosine similarity and seed word weight;

[0019] By fusing the word sentiment weight with the semantic vector, an optimized text feature T is generated;

[0020] Input the optimized text features into the main capsule layer of the capsule network to generate the main capsule feature P;

[0021] Use the dynamic routing mechanism of the capsule network to update the weights between the main capsule features and generate the optimized capsule feature Z;

[0022] Iteratively adjust the target word vector V(D) of positive and negative sentiment i ) direction;

[0023] Use the Skip-gram model to optimize the capsule feature Z in the direction;

[0024] Map the capsule feature Z to the sentiment vector space and decompose it into positive sentiment components and negative sentiment components;

[0025] Redistribute the weights of positive and negative components according to the optimization results of the Skip-gram model;

[0026] Use the attention mechanism to calculate the weight of each sentiment feature and generate the feature weight vector A;

[0027] The emotional direction feature is weighted according to the attention weight vector to generate the enhanced emotional feature A';

[0028] After normalizing the enhanced emotional features, the normalized enhanced emotional features are input into the fully connected layer of the convolutional neural network, the scores of each emotional category are calculated, and the score vector S is converted into a probability distribution using the Softmax function;

[0029] The maximum value of the probability distribution of the emotion category is taken as the customer's current emotion state, and the customer ID and the selected emotion state label as well as the probability distribution of all emotion categories are stored in a distributed database;

[0030] Combine text semantic features, emotional state features, and customer portrait features into a unified feature vector;

[0031] Initialize the feature vector of each customer request as the feature representation of the graph node, calculate the edge weights between nodes based on the similarity between requests, and construct the adjacency matrix A of the graph based on the edge weights to represent the connection relationship between nodes;

[0032] Use GNN to propagate features of each node in the graph, update the current node features through the information of adjacent nodes, and use Softmax to classify the node features in the last layer to output the problem category label.

[0033] As a preferred solution of the personalized customer service method based on distributed intelligent knowledge management described in the present invention, wherein: the construction of a dynamic knowledge graph based on the classification results and the knowledge graph query refer to mapping the current request to the corresponding node in the knowledge graph according to the classification label, and if there is no matching node in the knowledge graph, dynamically generating a new node and updating it to the distributed knowledge graph;

[0034] The new node is updated to all service nodes through the distributed database;

[0035] According to the question category label, the initial query node is determined. Based on the initial query node, multi-hop query is performed according to the relationship edges of the knowledge graph to expand the range of nodes related to the question. The SimRank algorithm is used to calculate the similarity between the expanded node and the initial node. The expanded query nodes are sorted in descending order according to the similarity. A threshold H is set, and knowledge nodes greater than the threshold H are screened out and integrated into a knowledge node set as the knowledge nodes most relevant to the customer's question.

[0036] As a preferred solution of the personalized customer service method based on distributed intelligent knowledge management described in the present invention, wherein: the generation of personalized service solutions based on query results refers to text parsing and preprocessing the knowledge point content in the knowledge node set, classifying the preprocessed knowledge point content into solution steps, precautions and related resources, structuring the classified content into a standard JSON data format, and inputting the structured JSON data into a Seq2Seq model to obtain a service response text;

[0037] The generated service response is optimized based on the customer's emotional state, and the service model that the customer is accustomed to is recommended based on the historical behavior in the customer portrait.

[0038] As a preferred solution of the personalized customer service method based on distributed intelligent knowledge management described in the present invention, wherein: the collection of customer feedback for service optimization refers to collecting customer quantitative data feedback through mobile applications, text messages and voice interactions after providing personalized services to users, aggregating the quantitative data, classifying services into high-scoring, medium-scoring and low-scoring, increasing the frequency of use of high-scoring knowledge points, and designing new solutions for low-scoring problems;

[0039] The quantitative data refers to setting a score to evaluate service quality.

[0040] As a preferred solution of the personalized customer service method based on distributed intelligent knowledge management described in the present invention, wherein: storing the analytical data generated during the service process into a distributed database refers to collecting feedback data, analytical data and service logs, slicing the data in chronological order and storing them in different distributed database tables according to data type, implementing access control and using a distributed file system to regularly back up data to off-site storage.

[0041] In a second aspect, the present invention provides a personalized customer service system based on distributed intelligent knowledge management, comprising:

[0042] Customer information collection module, used to collect customer request data, verify identity and generate customer portraits;

[0043] Sentiment analysis and question classification module, used to analyze customer emotional states and classify question types;

[0044] The knowledge graph construction module is used to dynamically construct the knowledge graph and perform multi-hop queries to expand the knowledge nodes related to the question;

[0045] The personalized service generation module is used to generate personalized service solutions based on the classification results and knowledge graph, and optimize them according to the emotional state;

[0046] Service feedback and optimization module, which is used to collect customer feedback data, analyze and optimize service strategies, and dynamically adjust knowledge graph node weights and service generation strategies;

[0047] The data storage and backup module is used to store data generated during the service process and regularly back it up to the distributed file system.

[0048] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the personalized customer service method based on distributed intelligent knowledge management as described in the first aspect of the present invention is implemented.

[0049] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the personalized customer service method based on distributed intelligent knowledge management as described in the first aspect of the present invention.

[0050] The beneficial effects of the present invention are as follows: the present invention significantly improves the accuracy of sentiment analysis through the combination of HowNet sentiment dictionary, capsule network dynamic routing mechanism and Skip-gram model, can accurately capture customer emotions and generate optimized sentiment features, and combines graph neural network to model customer request features, dynamically update and expand knowledge graph nodes, ensure the real-time and relevance of knowledge query, provide support for rapid response to complex problems, generate personalized service solutions through Seq2Seq model, dynamically adjust service content according to customer emotional state and historical behavior habits, and significantly improve customer satisfaction and service efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0052] Figure 1 This is a flow chart of the personalized customer service method based on distributed intelligent knowledge management in Example 1.

[0053] Figure 2 This is a structural diagram of the personalized customer service system based on distributed intelligent knowledge management in Example 1. DETAILED DESCRIPTION

[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0055] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0056] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0057] Example 1, reference Figure 1 and Figure 2, which is the first embodiment of the present invention, and which provides a personalized customer service method based on distributed intelligent knowledge management. The personalized customer service method based on distributed intelligent knowledge management includes the following steps:

[0058] S1. Collect customer information to verify customer identity, receive customer request data and generate customer profile after pre-processing;

[0059] Specifically, collecting customer information to verify customer identity, receiving customer request data and generating customer portraits after pre-processing means collecting customer information that sends request data through APP, voice assistants and intelligent customer service robots, extracting customer identity identifiers, calculating hash values ​​through the SHA256 algorithm, and submitting hash values ​​to blockchain smart contracts for verification. If the verification fails, a prompt message indicating that the identity verification failed is returned. If the verification succeeds, the request data submitted by the verified customer is received and customer historical data is collected.

[0060] By integrating customer information collected by APP, voice assistant and intelligent customer service robot, encrypting customer identity identifiers using SHA256 algorithm, and verifying customer identity with the help of blockchain smart contracts, this method effectively solves the risks of information leakage and forgery in traditional identity authentication methods, ensuring the security and uniqueness of customer data. In addition, by utilizing the immutability of blockchain technology, cross-system consistency verification is achieved, improving system security and customer trust.

[0061] The request data includes voice data and text data;

[0062] The historical data includes the customer's service records and behavioral preferences;

[0063] Convert voice data into text data and integrate it with the collected text data. Standardize the integrated text data. Use the word segmentation tool Jieba to segment the text into word sequences to obtain text sequences after word segmentation. Input the text sequences and historical service records into the pre-trained BERT model to obtain semantic vector features. One-hot encode the historical behavior preferences and convert the behavior preferences into vector form.

[0064] The voice data is converted into text data and integrated with the original text data. The integrated text is segmented by a word segmentation tool to generate a text sequence. The semantic vector features are then extracted using the BERT model. With this method, the system can accurately extract the core semantic information of the customer's current request. At the same time, the historical behavior preferences are One-Hot encoded and concatenated with the current request data into a unified feature vector, which can effectively integrate historical behavior with current needs to form a comprehensive customer feature representation. This process improves the standardization of data processing and the integrity of information expression.

[0065] The semantic vector features of the current request data are spliced ​​with the features of the historical data to form a unified feature vector. The final feature vector after integration is classified into a behavior feature group, a service feature group, and a request feature group. A structured representation of the customer portrait is generated through feature grouping. The generated customer portrait is stored in a distributed database and synchronized among distributed nodes.

[0066] The integrated feature vectors are divided into behavioral feature groups, service feature groups, and request feature groups, and structured customer portraits are generated through feature grouping. This approach not only improves the hierarchical expression capabilities of customer portraits, but also enables efficient organization and storage of customer information of different dimensions. By synchronizing customer portrait data through a distributed database, the system can share customer portraits in real time among multiple service nodes, providing data support for subsequent intelligent services and personalized recommendations.

[0067] The behavior feature group analyzes customers' click frequency, purchase habits and other preference information to accurately locate customers' interests and improve the accuracy of personalized services. The service feature group analyzes customers' historical service records to quickly locate the types of questions that customers may repeat and improve service response efficiency. The request feature group extracts the semantic features of the current request and cooperates with other feature groups to generate a comprehensive portrait to ensure that the service content can meet the immediate needs of customers.

[0068] The behavior feature group includes customer behavior preferences (such as click frequency, purchase habits), the service feature group includes customer historical service records, and the request feature group includes the semantic vector features of the current request.

[0069] By collecting, processing and storing customer information, we can form a structured customer portrait, which significantly improves the security of customer information management and the efficiency of data processing. The introduction of SHA256 algorithm and blockchain technology effectively ensures the security of customer identity authentication and solves the risk of data forgery and leakage in traditional systems. Through the integration and standardization of voice and text data, combined with the semantic vector extraction capability of the BERT model, it provides strong support for the high-quality generation of customer portraits. The application of distributed databases not only improves the storage efficiency of customer portrait data, but also ensures the consistency and availability of data through multi-node synchronization.

[0070] S2. Perform multimodal sentiment analysis and question classification based on the preprocessed customer request data, build a dynamic knowledge graph based on the classification results, and perform knowledge graph query;

[0071] Specifically, multimodal sentiment analysis and question classification based on preprocessed customer request data refers to introducing the HowNet sentiment dictionary, retrieving positive and negative words in the text, assigning initial sentiment weights W to the words, matching the text words with positive and negative seed words, and calculating the cosine similarity B between each word and the sentiment seed word;

[0072] Calculate the sentiment weight R of each word based on cosine similarity and seed word weight:

[0073]

[0074] In the formula, O j is the sentiment weight of sentiment seed word j, indicating the contribution intensity of seed word j to positive or negative sentiment, and N is the total number of sentiment seed words;

[0075] By fusing the word sentiment weight with the semantic vector, an optimized text feature T is generated:

[0076]

[0077] In the formula, V(D i ) is the word D i The semantic vector of is obtained through the BERT model, where n is the total number of words in the current request text and i is the index of the word being processed in the current text;

[0078] Input the optimized text features into the main capsule layer of the capsule network to generate the main capsule features P:

[0079] P = Squash (W p ·T+b p ),

[0080]

[0081] Where W p is the weight matrix for linear transformation, b p is the bias vector used to adjust the initial value of each capsule, Squash() is a nonlinear activation function, and x is the vector input;

[0082] Use the dynamic routing mechanism of the capsule network to update the weights between the main capsule features and generate the optimized capsule feature Z;

[0083] By introducing the HowNet sentiment dictionary, the words in the text are matched with positive and negative seed words, and the sentiment weight is calculated based on the cosine similarity and the seed word weight. The fusion of the word sentiment weight and the semantic vector optimizes the text features, making it more accurately reflect the sentiment tendency while retaining the semantic information. Combined with the dynamic routing mechanism of the capsule network, the correlation between features is further strengthened, effectively improving the accuracy and robustness of sentiment analysis. This multimodal analysis method can fully capture the emotional state of customers and lay a solid foundation for subsequent personalized service responses.

[0084] Iteratively adjust the target word vector V(D) of positive and negative sentiment i ) Direction:

[0085] For positive emotions:

[0086] Z1=V1(D i )+λ·h s ,

[0087] Where Z1 is the adjusted positive sentiment vector of the target word, and λ is the sentiment direction adjustment step, which is used to control the sentiment auxiliary vector h s The influence of V1(D i ) is the target word vector of positive sentiment;

[0088] For negative emotions:

[0089] Z2=V2(D i )-λ·h s ,

[0090] Where Z2 is the negative sentiment vector of the adjusted target word, V2(D i ) is the target word vector of negative sentiment;

[0091] Use the Skip-gram model to calculate the losses of Z1 and Z2 respectively:

[0092]

[0093] In the formula, L(D i ; c) is word D i The direction optimization loss, L, in the sentiment context c Diu is an indicator, indicating the target word D i Whether the sentiment is the same as the context word u (1 for the same, 0 for different), V(D i ) is the target word D i Vector, h sis the sentiment auxiliary vector, which is used to strengthen the sentiment direction of the word, σ is the Sigmoid function, which is used to calculate the similarity between the word and the sentiment auxiliary vector, T is the transposition operation, NEG(D i ) is the same as the target word D i A set of irrelevant negatively sampled words;

[0094] The maximum number of iterations is set by experience, and the iteration is stopped after reaching the maximum number of iterations to obtain the optimized positive sentiment vector Z'1 and negative sentiment vector Z'2 of the target word;

[0095] The emotional features generated by the capsule network are optimized by the Skip-gram model, which solves the problem of insufficient distinction between positive and negative emotional components in traditional sentiment analysis methods. i ; c) The model significantly improves the expressiveness of word vectors in the sentiment auxiliary vector space, making the sentiment features more interpretable in the multi-dimensional vector space. This optimization process not only improves the accuracy of sentiment classification, but also enhances the model's ability to identify extreme emotions (such as anger or disappointment);

[0096] Redistribute the weights of positive and negative components according to the optimization results of the Skip-gram model:

[0097] Z'=α·Z'1+β·Z'2,

[0098]

[0099] In the formula, Z' is the optimized emotional direction feature, α is the weight factor of the positive emotional component, β is the weight factor of the negative emotional component, V1(D i ) is the target word vector of positive sentiment, σ is the Sigmoid function, n is the total number of words in the current request text, and i is the word index of the current iteration, indicating the position of the word being processed;

[0100] Use the attention mechanism to calculate the weight of each sentiment feature and generate the feature weight vector A:

[0101] A=Softmax(W·Z'),

[0102] Where W is the attention weight matrix, and Z' is the optimized sentiment direction feature;

[0103] The emotional direction feature is weighted according to the attention weight vector to generate the enhanced emotional feature A':

[0104] A′=A⊙Z′,

[0105] After normalizing the enhanced emotional features, the normalized enhanced emotional features are input into the fully connected layer of the convolutional neural network, the scores of each emotional category are calculated, and the score vector S is converted into a probability distribution using the Softmax function;

[0106] The optimized sentiment features are weighted using the attention mechanism to generate enhanced sentiment features. The attention mechanism effectively highlights key sentiment features by dynamically allocating weights while reducing noise interference. Subsequently, the enhanced features are normalized and input into the fully connected layer of the convolutional neural network to further extract the deep features of the sentiment category. The scores are converted into probability distributions using the Softmax function to ultimately determine the customer's sentiment state. This process not only improves the accuracy of sentiment classification, but also provides support for the dynamic management of sentiment states.

[0107] The maximum value of the probability distribution of the emotion category is taken as the customer's current emotion state, and the customer ID and the selected emotion state label as well as the probability distribution of all emotion categories are stored in a distributed database;

[0108] The emotion categories include satisfaction, anger, anxiety, and disappointment;

[0109] Combine text semantic features, emotional state features, and customer portrait features into a unified feature vector;

[0110] Initialize the feature vector of each customer request as the feature representation of the graph node, calculate the edge weights between nodes based on the similarity between requests, and construct the adjacency matrix A of the graph based on the edge weights to represent the connection relationship between nodes;

[0111] Use GNN to propagate features of each node in the graph, update the current node features through the information of adjacent nodes, and use Softmax to classify the node features in the last layer to output the problem category label;

[0112] The text semantic features, emotional state features and customer portrait features are spliced ​​into a unified feature vector, and the graph node features are initialized for each customer request. By constructing an adjacency matrix A based on feature similarity, the model can effectively characterize the correlation between requests. By using the feature propagation capability of the graph neural network (GNN), the system continuously optimizes the node feature representation in multi-layer iterations, and finally classifies the customer problem category with high precision. Compared with traditional classification methods, this step significantly improves the efficiency and accuracy of problem classification by considering the contextual relationship between requests.

[0113] The problem category labels include network-related issues, rate consultation, device support, account and payment issues, complaints and feedback, etc.;

[0114] The network-related issues include network failure, slow network speed, and router configuration.

[0115] Through innovative multimodal sentiment analysis, optimized feature weight allocation, accurate question classification and dynamic knowledge management, breakthrough improvements in intelligent customer service have been achieved, the accuracy of sentiment recognition and question classification has been significantly improved, and customer experience has been optimized. At the same time, distributed storage has ensured data security and consistency, which can widely improve service efficiency and satisfaction, and provide a new technical framework for intelligent customer service systems.

[0116] Furthermore, constructing a dynamic knowledge graph based on the classification results and performing knowledge graph query means mapping the current request to the corresponding node in the knowledge graph according to the classification label, for example:

[0117] If the classification result is "network problem", it is mapped to the "network maintenance" knowledge node;

[0118] If the classification result is "rate consultation", it is mapped to the "rate management" knowledge node;

[0119] If there is no matching node in the knowledge graph, a new node is dynamically generated and updated to the distributed knowledge graph. For example, if the current request is classified as "device compatibility issue" but there is no corresponding node in the knowledge graph, the system will create a new node "device compatibility";

[0120] The new node is updated to all service nodes through the distributed database;

[0121] By mapping the classification labels of customer requests to the corresponding nodes in the knowledge graph, the request and knowledge content are accurately matched. This mapping method combines the contextual information of the classification results and the structural advantages of the knowledge graph, and can quickly locate the core content of the problem category. If there is no matching node in the knowledge graph, the system can dynamically generate a new node and update it to the distributed knowledge graph. This function significantly improves the adaptability and scalability of the knowledge graph and effectively solves the lag of the static knowledge graph in responding to new problems.

[0122] According to the problem category label, the initial query node is determined. For example, if the problem classification result points to "network connection problem", the query starts from the "network optimization" node. Based on the initial query node, multi-hop query is performed according to the relationship edge of the knowledge graph to expand the range of nodes related to the problem. In the knowledge graph, the initial query node is determined according to the problem category label, and multi-hop query is performed through its relationship edge to expand the range of nodes related to the problem. This process can deeply mine the potential related information in the knowledge graph, such as the upstream and downstream dependencies of the problem or the collaborative solution path with other problems. Compared with the method that relies only on a single node query, multi-hop query can capture more related knowledge and improve the system's responsiveness to complex problems. The SimRank algorithm is used to calculate the similarity between the extended node and the initial node. The SimRank algorithm provides a scientific basis for the screening and sorting of the extended query nodes by calculating the structural similarity between the nodes. In knowledge graph query, different nodes may have complex relationship networks. Through the similarity calculation of SimRank, the system can dynamically evaluate the importance of extended nodes, give priority to nodes with high similarity to the initial query nodes, sort the extended query nodes in descending order according to similarity, set the threshold H through statistical analysis of historical request data, filter out knowledge nodes greater than the threshold H and integrate them into a knowledge node set as the knowledge nodes most relevant to customer questions. By setting the threshold H, the system can filter out the node set most relevant to customer requests based on the similarity score of the extended nodes. This process ensures the efficiency and practicality of the query results, avoids response delays or information redundancy caused by too large a node set, and passes the filtered and sorted knowledge node set to the personalized service generation module as the knowledge node most relevant to customer questions. These node sets contain specific steps, precautions and related resource links to solve the problem, providing structured high-quality knowledge input for service generation.

[0123] By constructing a dynamic knowledge graph and combining the SimRank algorithm and threshold screening mechanism, accurate docking between customer requests and knowledge content is achieved. The construction of a dynamic knowledge graph enables the system to quickly adapt to new problem scenarios, and multi-hop query and similarity calculation optimize the depth and relevance of knowledge expansion. The integration of threshold screening and node sets ensures the efficiency and practicality of query results, and ultimately provides high-quality knowledge input for personalized service generation. The present invention not only improves the response speed and accuracy of customer service, but also enhances the dynamic expansion capability of the knowledge graph.

[0124] S3. Generate personalized service plans based on query results and collect customer feedback for service optimization;

[0125] Specifically, generating personalized service solutions based on query results refers to text parsing and preprocessing the knowledge point content in the knowledge node set, classifying the preprocessed knowledge point content into solution steps, precautions and related resources, and text parsing and preprocessing the knowledge node content to convert the unstructured knowledge point content into standardized classification information, including solution steps, precautions and related resources. This step significantly improves the structuring degree of knowledge points, which is convenient for efficient calling of subsequent models. At the same time, through classification optimization, the solution steps can clearly guide customers to perform specific operations, the precautions provide necessary risk warnings, and the related resources provide customers with additional technical support. The classified content is structured into a standard JSON data format. Organizing the classified knowledge point content into a standardized JSON data format is conducive to efficient data transmission and interpretation. The structured characteristics of the JSON format ensure the compatibility and flexibility of the data, and facilitate seamless docking with the Seq2Seq model. This standardized operation reduces the system's processing cost for unstructured data and improves the accuracy of service response generation. The structured JSON data is input into the Seq2Seq model to obtain the service response text;

[0126] The working principle of the Seq2Seq model is to encode the input content and extract key semantic information, and generate a service response based on the encoding result. The example output is:

[0127] If the client request is "slow network speed" and the knowledge graph nodes include "router restart method" and "bandwidth test tool", the response may be:

[0128] "Based on your request, we recommend that you try the following steps first:

[0129] 1. Check the router connection;

[0130] 2. Use the SpeedTest tool to test bandwidth;

[0131] 3. If the problem is not solved, please contact our technical support. ";

[0132] The Seq2Seq model can dynamically generate natural language service responses based on the input knowledge point set and customer requests. It extracts the semantic features of knowledge nodes through the Encoder and generates clear and accurate solution texts in combination with customer request content in the Decoder. Compared with the template generation method, the Seq2Seq model has powerful context understanding capabilities and flexible language generation capabilities, which significantly improves the personalization and pertinence of service responses.

[0133] Optimize the generated service response based on the customer's emotional state. For example, if the customer is satisfied, directly push the standard solution without adding additional information. If the customer is anxious, simplify the service description and add soothing language (such as "We understand your confusion, don't worry, we will help you solve the problem"). If the customer is angry, provide a clear and brief solution with additional quick upgrade options (such as directly contacting advanced technical support) and give priority to recommending the service mode that the customer is accustomed to based on the historical behavior in the customer portrait. For example, if the customer has selected "manual customer service support" many times in the past, give priority to including the manual support option in the generated service plan;

[0134] One of the core innovations of this invention is to dynamically optimize the generated service response according to the customer's emotional state. When the customer is anxious or angry, the system adds simplified step descriptions and soothing language to the service response to alleviate the customer's dissatisfaction; when the customer is satisfied, the system pushes standard solutions to avoid excessive interference. This emotion-driven response mechanism improves the comfort of the customer experience and service satisfaction.

[0135] Through the classification and analysis of knowledge nodes, the system can quickly respond to diverse customer needs, generate structured service solutions, and achieve dynamic personalized service responses through the Seq2Seq model. Compared with the traditional static template method, the flexibility and personalization of services are greatly improved. With the help of the optimization strategy driven by emotional state, the system can adjust the tone and content of service responses according to different emotional states, thereby improving customer satisfaction and reducing customer complaints. The combination of JSON data format and Seq2Seq model significantly reduces the complexity of unstructured data processing and improves the efficiency of service response generation. At the same time, the recommendation mechanism based on customer portraits reduces the time for customers to make repeated choices and improves the speed of service delivery. Through the accumulation of customer feedback and historical behavior data, the system continuously optimizes the content of knowledge nodes and service strategies, forming an efficient closed-loop self-optimization mechanism to provide customers with a continuously improved service experience.

[0136] Furthermore, collecting customer feedback for service optimization means collecting quantitative data feedback from customers through mobile applications, text messages, and voice interactions after providing personalized services to users, aggregating the quantitative data, and classifying services into high scores (8-10), medium scores (5-7), and low scores (1-4), increasing the frequency of use of high-scoring knowledge points, and designing new solutions for low-scoring problems.

[0137] The quantitative data refers to setting a score (1-10 points) to evaluate the service quality.

[0138] By collecting customer feedback data through multiple channels and classifying and aggregating quantitative data, a scientific basis is provided for service optimization. The steps of dynamically adjusting the weight of knowledge points and designing targeted solutions not only effectively improve the accuracy and response efficiency of services, but also enable the service system to have the ability of self-learning and continuous improvement. Distributed data storage further ensures the security and persistence of data, and provides comprehensive support for subsequent strategy optimization. This solution can greatly improve the intelligence level of personalized customer service.

[0139] S4. Store the analytical data generated during the service process into a distributed database;

[0140] Specifically, storing the analytical data generated during the service process in a distributed database means collecting feedback data, analytical data, and service logs, sharding the data in chronological order and storing them in different distributed database tables according to data type, implementing access control, and using a distributed file system (HDFS) to regularly back up data to off-site storage.

[0141] Through the distributed storage and management of feedback data, analysis data and service logs, the deficiencies of traditional centralized storage in scalability, security and efficiency are solved. The application of distributed databases not only improves the efficiency of data storage and query, but also provides solid data support for service optimization. Through access control and off-site backup mechanisms, the security and reliability of data are guaranteed. At the same time, the classified storage and real-time synchronization design of multi-type data make data analysis more efficient and cross-departmental collaboration smoother. In general, the present invention realizes the intelligence and efficiency of data management, and provides important technical support for the continuous optimization of personalized customer service.

[0142] This embodiment also provides a personalized customer service system based on distributed intelligent knowledge management, including:

[0143] Customer information collection module, used to collect customer request data, verify identity and generate customer portraits;

[0144] Sentiment analysis and question classification module, used to analyze customer emotional states and classify question types;

[0145] The knowledge graph construction module is used to dynamically construct the knowledge graph and perform multi-hop queries to expand the knowledge nodes related to the question;

[0146] The personalized service generation module is used to generate personalized service solutions based on the classification results and knowledge graph, and optimize them according to the emotional state;

[0147] Service feedback and optimization module, which is used to collect customer feedback data, analyze and optimize service strategies, and dynamically adjust knowledge graph node weights and service generation strategies;

[0148] The data storage and backup module is used to store data generated during the service process and regularly back it up to the distributed file system.

[0149] This embodiment also provides a computer device, which is suitable for the personalized customer service method based on distributed intelligent knowledge management, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the personalized customer service method based on distributed intelligent knowledge management proposed in the above embodiment.

[0150] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0151] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the personalized customer service method based on distributed intelligent knowledge management proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination of them, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0152] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A personalized customer service method based on distributed intelligent knowledge management, characterized in that: include: Collect customer information to verify customer identity, receive customer request data and generate customer portraits after pre-processing; Perform multimodal sentiment analysis and question classification based on preprocessed customer request data, build a dynamic knowledge graph based on the classification results, and perform knowledge graph query; Generate personalized service plans based on query results and collect customer feedback for service optimization; The analytical data generated during the service process is stored in a distributed database.

2. The personalized customer service method based on distributed intelligent knowledge management as claimed in claim 1, characterized in that: The collecting of customer information to verify customer identity, receiving customer request data and generating a customer portrait after pre-processing means collecting customer information that sends request data through APP, voice assistant and intelligent customer service robot, extracting customer identity identifier, calculating hash value through SHA256 algorithm, submitting hash value to blockchain smart contract for verification, if verification fails, returning a prompt message of identity authentication failure, if verification succeeds, receiving request data submitted by the verified customer and collecting customer historical data; The request data includes voice data and text data; The historical data includes the customer's service records and behavioral preferences; Convert voice data into text data and integrate it with the collected text data. Standardize the integrated text data. Use the word segmentation tool to segment the text into word sequences to obtain the text sequence after word segmentation. Input the text sequence and historical service records into the pre-trained BERT model to obtain semantic vector features. One-hot encode the historical behavior preferences and convert the behavior preferences into vector form. The semantic vector features of the current request data are spliced ​​with the features of the historical data to form a unified feature vector. The final feature vector after integration is classified into a behavior feature group, a service feature group, and a request feature group. A structured representation of the customer portrait is generated through feature grouping. The generated customer portrait is stored in a distributed database and synchronized among distributed nodes. The behavior feature group includes customer behavior preferences, the service feature group includes customer historical service records, and the request feature group includes semantic vector features of the current request.

3. The personalized customer service method based on distributed intelligent knowledge management as claimed in claim 2, characterized in that: The multimodal sentiment analysis and question classification based on the preprocessed customer request data refers to introducing the HowNet sentiment dictionary, retrieving positive and negative words in the text, assigning initial sentiment weights W to the words, matching the text words with positive and negative seed words, and calculating the cosine similarity B between each word and the sentiment seed word; Calculate the sentiment weight R of each word based on cosine similarity and seed word weight; By fusing the word sentiment weight with the semantic vector, an optimized text feature T is generated; Input the optimized text features into the main capsule layer of the capsule network to generate the main capsule feature P; Use the dynamic routing mechanism of the capsule network to update the weights between the main capsule features and generate the optimized capsule feature Z; Iteratively adjust the target word vector V(D) of positive and negative sentiment i ) direction; Use the Skip-gram model to optimize the capsule feature Z in the direction; Map the capsule feature Z to the sentiment vector space and decompose it into positive sentiment components and negative sentiment components; Redistribute the weights of positive and negative components according to the optimization results of the Skip-gram model; Use the attention mechanism to calculate the weight of each sentiment feature and generate the feature weight vector A; The emotional direction feature is weighted according to the attention weight vector to generate the enhanced emotional feature A'; After normalizing the enhanced emotional features, the normalized enhanced emotional features are input into the fully connected layer of the convolutional neural network, the scores of each emotional category are calculated, and the score vector S is converted into a probability distribution using the Softmax function; The maximum value of the probability distribution of the emotion category is taken as the customer's current emotion state, and the customer ID and the selected emotion state label as well as the probability distribution of all emotion categories are stored in a distributed database; Combine text semantic features, emotional state features, and customer portrait features into a unified feature vector; Initialize the feature vector of each customer request as the feature representation of the graph node, calculate the edge weights between nodes based on the similarity between requests, and construct the adjacency matrix A of the graph based on the edge weights to represent the connection relationship between nodes; Use GNN to propagate features of each node in the graph, update the current node features through the information of adjacent nodes, and use Softmax to classify the node features in the last layer to output the problem category label.

4. The personalized customer service method based on distributed intelligent knowledge management as claimed in claim 3, characterized in that: The construction of a dynamic knowledge graph based on the classification results and the query of the knowledge graph refers to mapping the current request to the corresponding node in the knowledge graph according to the classification label. If there is no matching node in the knowledge graph, a new node is dynamically generated and updated to the distributed knowledge graph; The new node is updated to all service nodes through the distributed database; According to the question category label, the initial query node is determined. Based on the initial query node, multi-hop query is performed according to the relationship edges of the knowledge graph to expand the range of nodes related to the question. The SimRank algorithm is used to calculate the similarity between the expanded node and the initial node. The expanded query nodes are sorted in descending order according to the similarity. A threshold H is set, and knowledge nodes greater than the threshold H are screened out and integrated into a knowledge node set as the knowledge nodes most relevant to the customer's question.

5. The personalized customer service method based on distributed intelligent knowledge management as claimed in claim 4, characterized in that: Generating a personalized service solution based on the query result refers to performing text parsing and preprocessing on the knowledge point content in the knowledge node set, classifying the preprocessed knowledge point content into solution steps, precautions and related resources, structuring the classified content into a standard JSON data format, and inputting the structured JSON data into a Seq2Seq model to obtain a service response text; The generated service response is optimized based on the customer's emotional state, and the service model that the customer is accustomed to is recommended based on the historical behavior in the customer portrait.

6. The personalized customer service method based on distributed intelligent knowledge management as claimed in claim 5, characterized in that: Collecting customer feedback for service optimization refers to collecting customer quantitative data feedback through mobile applications, text messages and voice interactions after providing personalized services to users, aggregating the quantitative data, classifying services into high-scoring, medium-scoring and low-scoring, increasing the frequency of use of high-scoring knowledge points, and designing new solutions for low-scoring problems; The quantitative data refers to setting a score to evaluate service quality.

7. The personalized customer service method based on distributed intelligent knowledge management as claimed in claim 6, characterized in that: Storing the analytical data generated during the service process into a distributed database refers to collecting feedback data, analytical data, and service logs, slicing the data in chronological order and storing them in different distributed database tables according to data type, implementing access control, and using a distributed file system to regularly back up data to off-site storage.

8. A personalized customer service system based on distributed intelligent knowledge management based on the personalized customer service method based on distributed intelligent knowledge management according to any one of claims 1 to 7, characterized in that: include: Customer information collection module, used to collect customer request data, verify identity and generate customer portraits; Sentiment analysis and question classification module, used to analyze customer emotional states and classify question types; The knowledge graph construction module is used to dynamically construct the knowledge graph and perform multi-hop queries to expand the knowledge nodes related to the question; The personalized service generation module is used to generate personalized service solutions based on the classification results and knowledge graph, and optimize them according to the emotional state; Service feedback and optimization module, which is used to collect customer feedback data, analyze and optimize service strategies, and dynamically adjust knowledge graph node weights and service generation strategies; The data storage and backup module is used to store data generated during the service process and regularly back it up to the distributed file system.

9. A computer device comprising: A memory and a processor; the memory stores a computer program, characterized in that: when the processor executes the computer program, the steps of the personalized customer service method based on distributed intelligent knowledge management described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the personalized customer service method based on distributed intelligent knowledge management described in any one of claims 1 to 7 are implemented.

Images