Customer service intelligent processing method, device and equipment and storage medium

Abstract

The invention discloses a customer service intelligent processing method and device, equipment and a storage medium, belongs to the technical field of artificial intelligence, and is applied to optimization processing of financial business services. According to the method, the distributed data processing architecture is constructed, so that each node completes model training locally and uploads gradient parameters, and a global model is generated. And aligning the timestamp information and the updating frequency of each data source through a unified time reference. And standardizing the heterogeneous data into a unified feature vector by using feature mapping and dimension reduction technologies, and generating a customer portrait. Market policy changes and hot event information are collected in real time, and market state classification tags are generated in combination with natural language processing and event influence evaluation. Optimal model configuration adapting to different market states is obtained by dynamically adjusting model parameter weights and reward function evaluation. In a customer claim settlement processing process, customer portraits and market state information are fused, and intelligent decision-making of claim settlement risk grade assessment and personalized product recommendation is realized.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, specifically relating to a method, apparatus, device, and storage medium for intelligent customer service processing. Background Technology

[0002] As a crucial pillar of financial services, the insurance industry's level of intelligence directly impacts customer experience and operational efficiency, particularly in customer service, where intelligent processing can significantly reduce costs and improve response speed. However, current intelligent solutions in the insurance industry have significant shortcomings when dealing with complex business scenarios. Many methods struggle to achieve real-time consistency when handling multi-source, heterogeneous data, especially when customer privacy is involved, and data silos prevent effective information integration. Furthermore, existing solutions lack adaptability in dynamic market environments, making it difficult to quickly respond to changes in insurance product demands and regulatory requirements.

[0003] How can we achieve real-time alignment of multi-source data while protecting customer privacy and ensuring that the model can dynamically adapt to market fluctuations? First, the challenge of real-time data alignment stems from the diverse data sources involved in insurance operations, such as customer policy records, claims history, and real-time interaction data. These data vary in format and update frequency, making it difficult to guarantee consistency when processing complex queries. For example, when a customer submits a claim, the system needs to simultaneously integrate their historical policies, current health data, and real-time interaction records. However, due to asynchronous updates of data sources, information discrepancies often occur, affecting the accuracy of claims decisions. Second, the complexity of data alignment further exacerbates the difficulty of dynamic model adaptation. The insurance market is affected by policies, customer preferences, and external events, resulting in frequent fluctuations in demand. If the model cannot quickly adjust its predictive strategy, it may lead to recommended insurance products that do not meet the actual needs of customers, reducing customer satisfaction.

[0004] Therefore, how to resolve the contradiction between multi-source data consistency and market volatility adaptability through efficient data alignment mechanisms and dynamic adaptive models while ensuring privacy and security has become a key issue in improving the intelligence level of insurance customer service. Summary of the Invention

[0005] The purpose of this application is to propose a method, apparatus, computer device, and storage medium for intelligent customer service processing, in order to solve the technical problem of how to resolve the contradiction between multi-source data consistency and market volatility adaptability through an efficient data alignment mechanism and dynamic adaptive model while ensuring privacy and security.

[0006] To address the aforementioned technical problems, this application provides a method for intelligent customer service processing, employing the following technical solution:

[0007] A method for intelligent customer service processing includes:

[0008] A distributed data processing architecture is constructed, in which each data source node completes model training locally and uploads gradient parameters. The global model is obtained by aggregating the gradient parameters. The distributed data processing architecture includes several local models, with each data source node corresponding to a local model.

[0009] Based on the timestamp information and update frequency characteristics of various heterogeneous data sources, a time-synchronized multidimensional data matrix is ​​obtained by establishing a unified time benchmark.

[0010] Feature mapping is used to convert customer insurance records, claims history and interaction data in different formats in the multidimensional data matrix into standardized feature vectors, and unified customer profile data is generated based on the standardized feature vectors;

[0011] Real-time collection of market policy changes and hot topics information to determine the current market status classification labels;

[0012] Based on the current market status classification labels, the global model parameter weights are dynamically adjusted, and the prediction accuracy under different parameter configurations is evaluated through the reward function to obtain the optimal model configuration that adapts to the current market environment.

[0013] In response to customer claim requests, the system integrates customer profile data and current market status classification tags, uses an optimally configured global model to determine the claim risk level and product recommendation matching degree, and obtains personalized claim decision results and product recommendation solutions.

[0014] To address the aforementioned technical problems, this application also provides an apparatus for intelligent customer service processing, employing the following technical solution:

[0015] An apparatus for intelligent customer service processing includes:

[0016] The local training module is used to build a distributed data processing architecture. After each data source node completes model training locally, it uploads gradient parameters and obtains a global model through gradient parameter aggregation. The distributed data processing architecture includes several local models, with each data source node corresponding to one local model.

[0017] The unified benchmark module is used to obtain a time-synchronized multidimensional data matrix by establishing a unified time benchmark based on the timestamp information and update frequency characteristics of various heterogeneous data sources.

[0018] The customer profiling module is used to convert customer insurance records, claims history and interaction data in different formats in a multidimensional data matrix into standardized feature vectors using feature mapping, and to generate unified customer profile data based on the standardized feature vectors.

[0019] The market tagging module is used to collect information on market policy changes and hot topics in real time, and to determine the current market status classification tags.

[0020] The configuration optimization module is used to dynamically adjust the global model parameter weights based on the current market status classification labels, evaluate the prediction accuracy under different parameter configurations through the reward function, and obtain the optimal model configuration that adapts to the current market environment.

[0021] The personalized recommendation module is used to respond to customer claim requests. By integrating customer profile data and current market status classification tags, it uses an optimally configured global model to determine the claim risk level and product recommendation matching degree, and obtains personalized claim decision results and product recommendation solutions.

[0022] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution:

[0023] A computer device includes a memory and a processor, the memory storing computer-readable instructions, wherein the processor, when executing the computer-readable instructions, implements the steps of the intelligent customer service processing method as described in any of the preceding claims.

[0024] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the technical solution described below:

[0025] A computer-readable storage medium storing computer-readable instructions, which, when executed by a processor, implement the steps of the intelligent customer service processing method as described in any one of the preceding descriptions.

[0026] Compared with the prior art, the embodiments of this application have the following main advantages:

[0027] This application discloses a method, apparatus, device, and storage medium for intelligent customer service processing, belonging to the field of artificial intelligence technology, and applied to the optimization of financial business services. This application constructs a distributed data processing architecture, enabling each node to complete model training locally and upload gradient parameters, generating a global model through federated aggregation. A unified time benchmark aligns the timestamp information and update frequency of each data source. Feature mapping and dimensionality reduction techniques are used to standardize heterogeneous data into a unified feature vector, generating customer profiles. Real-time collection of market policy changes and hot event information, combined with natural language processing and event impact assessment, generates market status classification labels. Optimal model configurations adapted to different market conditions are obtained by dynamically adjusting model parameter weights and reward function evaluation. In the customer claims processing process, customer profiles and market status information are integrated to achieve intelligent decision-making for claims risk level assessment and personalized product recommendations, improving claims processing efficiency and customer satisfaction, while ensuring business interpretability and compliance, providing insurance companies with accurate, efficient, and intelligent customer service capabilities. Attached Figure Description

[0028] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 An exemplary system architecture diagram is shown, in which this application can be applied;

[0030] Figure 2 A flowchart illustrating one embodiment of a method for intelligent customer service processing according to this application is shown;

[0031] Figure 3 It shows Figure 2 A flowchart of an embodiment of step S203;

[0032] Figure 4 A schematic diagram of the structure of an embodiment of a device for intelligent customer service processing according to this application is shown;

[0033] Figure 5 It shows Figure 4 A schematic diagram of a structure of an embodiment of the customer profiling module 403;

[0034] Figure 6 A schematic diagram of the structure of one embodiment of a computer device according to this application is shown. Detailed Implementation

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0036] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0037] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0038] like Figure 1 As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables.

[0039] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0040] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.

[0041] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0042] It should be noted that the intelligent customer service processing method provided in this application embodiment is generally executed by a server / terminal device, and correspondingly, the intelligent customer service processing device is generally located in the server / terminal device.

[0043] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative; the system can have any number of terminal devices, networks, and servers depending on implementation needs.

[0044] Continue to refer to Figure 2 A flowchart illustrating an embodiment of a method for intelligent customer service processing according to this application is shown. The method for intelligent customer service processing includes the following steps:

[0045] S201. Construct a distributed data processing architecture, where each data source node completes model training locally and then uploads gradient parameters. The global model is obtained by aggregating the gradient parameters. The distributed data processing architecture includes several local models, with each data source node corresponding to a local model.

[0046] Specifically, the distributed data processing architecture can be designed using a federated learning (FL) framework. This ensures that different data source nodes (such as various insurance branches, partner hospitals, and third-party data platforms) complete model training locally without directly uploading raw data, thereby protecting customer privacy and meeting data compliance requirements. Each data source node corresponds to a local model, typically a lightweight deep learning model (such as MobileNet or Lightweight Transformer) or a traditional machine learning model (such as XGBoost or LightGBM) can be selected to reduce edge computing pressure.

[0047] After training, nodes only upload model gradients or weight updates (gradient compression techniques such as Top-K sparsing and quantization encoding can be used to reduce transmission bandwidth consumption), and a secure aggregation protocol ensures that the training data of a single node cannot be back-inferred during parameter aggregation. Global model aggregation can use the FedAvg algorithm or its variants, and differential privacy mechanisms can be introduced during the aggregation process to add noise to the gradients to enhance security.

[0048] Furthermore, to address the issue of uneven data distribution across different nodes (Non-IID), model parameters can be weighted and adjusted before aggregation, with weights allocated based on node data volume, sample quality, and data freshness. The entire distributed architecture can be deployed in container orchestration environments such as Kubernetes, achieving efficient distributed training scheduling and fault-tolerant recovery through parameter servers or decentralized aggregation (such as Ring-AllReduce).

[0049] S202, based on the timestamp information and update frequency characteristics of various heterogeneous data sources, a time-synchronized multidimensional data matrix is ​​obtained by establishing a unified time benchmark;

[0050] Specifically, this involves achieving multi-source alignment and constructing a unified time benchmark for time series data. Different data sources (such as insurance systems, claims systems, customer service interaction systems, and market sentiment monitoring platforms) have different sampling frequencies (minute-level, hour-level, day-level) and inconsistent timestamp formats (Unix time, ISO 8601, local time zone). Therefore, it is necessary to first standardize the format of all timestamps and convert them to a unified time zone benchmark (such as UTC+0). Then, a unified sampling granularity can be determined based on the update frequency characteristics of each data source, for example, using hours as the basic unit, and time synchronization of data at different frequencies can be achieved through interpolation (linear interpolation, spline interpolation) or resampling methods. To minimize disruption to the time series trend, local weighted regression (LOESS) or Kalman filtering can be used for smoothing during the interpolation process. Simultaneously, for missing data points, time series imputation methods (such as nearest neighbor imputation based on similar time periods or autoregressive model prediction) can be used for completion. The final multidimensional data matrix can be arranged in chronological order, with each row representing all data features on a unified time slice. In addition, if the data contains spikes from sudden events, these spike signals should be preserved during alignment to avoid information loss due to over-smoothing.

[0051] S203 uses feature mapping to convert customer insurance records, claims history and interaction data in different formats in the multidimensional data matrix into standardized feature vectors, and generates unified customer profile data based on the standardized feature vectors;

[0052] Specifically, this stage requires designing a robust feature mapping and encoding scheme to uniformly convert multi-source heterogeneous data (including structured data, semi-structured data, and unstructured text data) into standardized feature vectors. For customer insurance records, fields such as policy type, sum insured, payment period, and coverage can be one-hot encoded, numerically standardized, or embedded for model processing. Claim history data can extract statistical features such as the number of claims, claim amount, rejection rate, and claim period, and the impact of outdated records can be reduced using a time decay factor. Interactive data (such as customer consultation records and complaint tickets) can use Natural Language Processing (NLP) techniques to extract keywords, sentiment, and intent classification labels, and map them into vectorized representations (such as embedding in pre-trained models like BERT / ERNIE). All features need to be standardized after mapping (e.g., Z-Score standardization, Min-Max normalization) to avoid bias in model training caused by features of different dimensions. Feature fusion can be achieved through methods such as concatenation and multi-modal fusion to generate unified customer profile data, which can include multi-dimensional information such as demographic features, behavioral features, risk features, and preference features.

[0053] S204: Real-time collection of market policy changes and hot events information to determine the current market status classification label;

[0054] Specifically, a real-time information collection and market status classification system should be established. Information sources may include policy announcement platforms (such as the website of the State Financial Supervision and Administration Bureau), news media APIs, social media monitoring (Weibo, Twitter), industry analysis reports, etc. The system needs to have data crawling, information extraction, and event detection capabilities. Data crawling can be achieved using web crawlers or subscription-based APIs; information extraction can use Named Entity Recognition (NER) to extract key fields such as the event subject, product category involved, and geographical scope; event detection can identify emerging hot topics based on text clustering (such as K-Means, HDBSCAN) or topic models (such as LDA). To address the quality differences between different information sources, an information credibility scoring mechanism (based on factors such as source authority, publication time, and content completeness) can be introduced. In terms of market status classification, classification labels can be automatically generated by constructing a rule-based labeling system (such as "policy tightening," "price war," "new product launch") or training a supervised learning classification model (using historical market events and corresponding market reaction data as a training set). This label can be combined with time-series market indicators (such as the insurance industry premium growth rate and the number of competitor products) to form a market state vector, which can be used to dynamically adjust model strategies.

[0055] S205: Based on the current market status classification labels, dynamically adjust the global model parameter weights, evaluate the prediction accuracy under different parameter configurations through the reward function, and obtain the optimal model configuration that adapts to the current market environment.

[0056] Specifically, the system needs to implement dynamic model parameter adjustment and optimal configuration search. First, the current market state label is mapped to a set of parameter adjustment rules (e.g., weighting the importance of specific features, adjusting prediction thresholds), and fine-tuned in the weight space of the global model. Dynamic adjustment can be achieved through a reinforcement learning (RL) framework, treating the market state as the environment state, parameter configuration as an action, and prediction accuracy as a reward. Specifically, policy gradient-based methods (such as PPO, A3C) or value-based algorithms (such as DQN) can be used to search for the optimal parameter combination. During the search process, a multi-armed bandit strategy can be used to quickly explore different parameter combinations, and Bayesian optimization or genetic algorithms can be used for global optimization. To reduce the risk of trial and error in a real production environment, the effect of parameter adjustment can be simulated in a backtesting environment using historical market data before deploying the best-performing configuration to the real-time system. Simultaneously, to cope with sudden market changes, trigger thresholds (such as market volatility index, news popularity index) can be set, and parameter recalculation can be initiated immediately when the threshold is exceeded.

[0057] S206 responds to customer claim requests by integrating customer profile data and current market status classification tags, using an optimally configured global model to determine the claim risk level and product recommendation matching degree, and obtains personalized claim decision results and product recommendation solutions.

[0058] Specifically, in the final execution phase, the system integrates the previously constructed customer profile data with current market status classification labels, inputting them into the optimally configured global model to complete claims risk assessment and personalized recommendations. The risk assessment module outputs a claims risk level (e.g., low, medium, high), based on factors such as the customer's historical claims records, the risk coefficient of the current product, and industry fraud model output. To improve interpretability, interpretive algorithms such as SHAP or LIME can be called after the model output to generate a feature contribution report, allowing reviewers to understand the reasons for the judgment. The product recommendation module utilizes preference features from the customer profile and market status information, invoking collaborative filtering, knowledge graph recommendation, or deep matching networks (such as DSSM and DIN) to provide customized insurance product portfolio suggestions to the customer. During execution, the system also needs to achieve multi-objective optimization, i.e., improving the conversion rate of recommended products while ensuring claims processing efficiency, which may involve using a weighted reward function to balance different objectives. Furthermore, the entire decision-making process can be recorded in a traceable log or stored on the blockchain.

[0059] Furthermore, a distributed data processing architecture is constructed, in which each data source node completes model training locally and uploads gradient parameters, and the global model is obtained through gradient parameter aggregation. This process specifically includes:

[0060] Build a matching distributed data processing architecture based on the attributes of each data source node;

[0061] Local data is obtained from various data source nodes, and local model training is performed using a preset machine learning algorithm to obtain model gradient parameters.

[0062] The model gradient parameters are uploaded from each data source node to the central server via an encrypted channel;

[0063] The model gradient parameters are aggregated using a federated averaging algorithm on the central server to obtain the global model parameters;

[0064] The preset initial global model is updated based on the global model parameters, and the updated global model is generated.

[0065] In this embodiment, the construction of the distributed data processing architecture prioritizes factors such as the computing power, data type, and network bandwidth of each data source node to match appropriate computing and communication strategies. When each node performs model training locally, it can select a deep learning model (such as LSTM or Transformer) or a traditional machine learning model (such as XGBoost or RandomForest) based on the business scenario, and combine gradient compression (such as Top-K sparsity and quantization encoding) to reduce communication overhead. To ensure data privacy, differential privacy mechanisms can be introduced to add noise to gradient parameters before uploading, and the parameters are transmitted to the central server via secure encrypted channels such as TLS / SSL. The central server uses a federated averaging (FedAvg) algorithm during aggregation, and applies weighted aggregation (such as sample size and data quality weights) based on the data distribution characteristics of the nodes when necessary to improve the model's generalization ability. The aggregated global parameters are synchronized back to each node to replace or fine-tune the local model, achieving continuous iterative optimization, thereby completing efficient and privacy-preserving global model training without directly transmitting the original data.

[0066] Through the above steps, while ensuring data security and privacy, collaborative modeling of multi-source data is achieved, effectively improving the generalization performance and adaptability of the global model.

[0067] Furthermore, based on the timestamp information and update frequency characteristics of various heterogeneous data sources, the steps to obtain a time-synchronized multidimensional data matrix by establishing a unified time benchmark specifically include:

[0068] Analyze the timestamp information of various heterogeneous data sources to determine the data generation and update times;

[0069] Based on the data generation time and update time, determine the data update frequency, and divide the data into different update cycles according to the update frequency;

[0070] Select a common time base and convert the timestamps of various heterogeneous data sources to the common time base;

[0071] For each update cycle, data from different data sources are arranged in chronological order based on the converted timestamps to construct a time-synchronized multidimensional data matrix.

[0072] In this embodiment, the timestamp formats of various heterogeneous data sources (such as policy systems, claims systems, customer service record platforms, and third-party market data interfaces) are first uniformly parsed, supporting multiple formats including Unix timestamps, ISO 8601 standard formats, and localized time representations. The system compares the generation time and update time of each data source, extracts its actual data refresh frequency, and divides it into different update cycles such as minutes, hours, or days. Then, a unified common time base (such as UTC+0) is selected as the alignment standard, and all timestamps are converted to a unified representation under this base to avoid data mismatch caused by time zone differences. When constructing a time-synchronized multidimensional data matrix, for data with low update frequencies, time interpolation and missing data imputation (such as KNN imputation and time series prediction) can be used to fill in blank periods, while for high-frequency data, window aggregation or downsampling is performed to match the common cycle. Finally, each row of the matrix corresponds to a common time slice, integrating various features from different data sources in chronological order.

[0073] The above steps ensure accurate alignment of multi-source heterogeneous data in the time dimension, improving the integrity and usability of data fusion.

[0074] Further, please refer to Figure 3 The process involves using feature mapping to convert customer insurance records, claims history, and interaction data in different formats from a multidimensional data matrix into standardized feature vectors, and then generating unified customer profile data based on these standardized feature vectors. Specifically, this includes:

[0075] S301, through a preset feature mapping method, maps customer insurance records, claims history and interaction data of different formats in the multidimensional data matrix into numerical features, and generates an initial feature set;

[0076] S302, Principal component analysis algorithm is used to reduce the dimensionality of the initial feature set to obtain a dimensionality-reduced feature set;

[0077] S303: Based on the dimensionality reduction feature set, using a preset vector generation rule, the dimensionality reduction feature set is transformed into a standardized feature vector through linear transformation, generating a feature representation in a unified format, and thus obtaining a standardized feature vector.

[0078] S304, group the standardized feature vectors to determine customer behavior pattern categories;

[0079] S305, based on customer behavior pattern categories, adopts a profile data structure to integrate standardized feature vectors into unified customer profile data.

[0080] In this embodiment, firstly, in S301, the multidimensional data matrix is ​​structured using a preset feature mapping method. Customer insurance records of different formats (e.g., policy type, coverage period, payment method), claims history (e.g., number of claims, average claim amount, rejection rate, claims cycle), and interactive data (e.g., customer service call records, online consultation content, complaint tags) are quantified using appropriate encoding strategies. For categorical fields, One-Hot encoding, label encoding, or embedding vectors are used; for numerical fields, normalization or Z-Score standardization is used; for textual fields, semantic vectors are extracted using natural language processing models (e.g., BERT, Word2Vec). Subsequently, in S302, Principal Component Analysis (PCA) is used to reduce the dimensionality of the initial feature set to reduce redundant features and improve computational efficiency, while retaining principal components with a cumulative variance contribution rate of over 95%. In S303, based on preset vector generation rules, linear transformations (such as matrix multiplication and affine transformations) are performed on the dimensionality-reduced feature set to generate standardized feature vectors, thereby achieving a unified feature representation format. In S304, the standardized feature vectors are clustered according to behavioral patterns (such as high-frequency claims, long-term stable, and price-sensitive), using algorithms such as K-Means and DBSCAN. Finally, in S305, the clustering results are integrated with the standardized feature vectors to generate a complete customer profile according to a preset customer profile data structure (such as basic information layer, behavioral feature layer, risk feature layer, and preference feature layer).

[0081] Taking a large insurance company as an example, the company possesses heterogeneous customer data from multiple sources, including the underwriting system, claims system, and customer service center. Customer A has held three different types of insurance policies in the past five years, including a 10-year life insurance policy, a critical illness insurance policy, and a car insurance policy, with payment methods covering both annual and monthly payments. In terms of claims records, the car insurance policy has made two small claims in the past two years (average claim amount of 8,000 yuan), while there are no claims records for the life insurance and critical illness insurance policies. In terms of customer service interaction data, Customer A has consulted online three times in the past year about the details of health insurance terms and expressed concerns about the timeliness of claims processing during a phone call.

[0082] The system first executes S301, performing One-Hot encoding on categorical fields (such as policy type and payment method) and Z-Score standardization on numerical fields (such as coverage period and claim amount) in the insurance application records. Call and text content in the interaction data are converted into semantic vectors using the BERT model and added to the feature set. In S302, PCA dimensionality reduction retains over 95% of the main components contributing to the cumulative variance, reducing feature redundancy. In S303, the dimensionality-reduced feature set is transformed linearly to generate standardized feature vectors, ensuring consistent format across different data sources. In S304, a clustering algorithm categorizes customer A into "low-frequency claims + high health insurance focus" and groups them with other customers exhibiting similar behavioral patterns. In S305, the system integrates customer A's standardized feature vector and behavioral pattern labels into the customer profile structure to form a hierarchical profile: the basic information layer records demographic data such as gender, age, and place of residence; the behavioral feature layer reflects the policy holding structure and customer service interaction frequency; the risk feature layer marks its low-risk claims history; and the preference feature layer shows its significant attention to health insurance.

[0083] Through the above steps, unified standardization and structured profiling of multi-source heterogeneous customer data were achieved, significantly improving the accuracy and consistency of model processing.

[0084] Furthermore, the steps of collecting real-time information on market policy changes and trending events to determine the current market status classification labels specifically include:

[0085] Data on market policy changes and hot topics are obtained from preset information sources and stored as structured time series data;

[0086] Natural language processing technology is used to analyze the content of structured time series data, extract keywords and sentiment, and obtain the event impact assessment results;

[0087] Based on the event impact assessment results and the preset label generation rules, a current market status classification label is generated.

[0088] In this embodiment, the system first continuously acquires market policy changes and hot topic data from preset information sources (including announcements from the State Financial Regulatory Commission, industry association publishing platforms, authoritative news media APIs, social media monitoring interfaces, and third-party market analysis data services), and converts this data into structured time-series data to ensure that information from different sources is aligned in the time dimension. Subsequently, natural language processing techniques (including word segmentation, named entity recognition, and dependency parsing) are used to parse the event content, extracting key information such as relevant policy clauses, product types, and affected customer groups. Simultaneously, a sentiment analysis model (such as BERT + sentiment classifier) ​​is used to determine the market sentiment tendency (positive, neutral, negative) of the event. Based on this, the impact assessment value of the event is calculated by combining its scope, duration, and volatility. Finally, according to preset label generation rules (such as "policy tightening," "industry benefit," "intensified competition," and "risk warning"), the event is mapped to market state classification labels that can be used by the model.

[0089] Through the above steps, automated monitoring of market information and generation of structured labels can be achieved, enabling the model to quickly perceive and adapt to changes in the external environment.

[0090] Furthermore, based on the current market state classification labels, the global model parameter weights are dynamically adjusted, and the prediction accuracy under different parameter configurations is evaluated through a reward function to obtain the optimal model configuration adapted to the current market environment. This process specifically includes:

[0091] Based on the current market status classification labels, market data is classified using a clustering algorithm to obtain a feature representation of the market status;

[0092] Based on the feature representation, the gradient descent algorithm is used to adjust the parameter weights of the global model to obtain the updated parameter configuration;

[0093] The updated parameter configuration is evaluated using a reward function, the prediction accuracy is calculated, and the performance score of the current configuration is calculated.

[0094] If the performance score is lower than the preset threshold, the current market status classification label will be adjusted based on environmental change detection, and a new feature representation will be obtained.

[0095] Based on the new feature representation, the parameter weights of the global model are iterated again until the model reaches the optimal configuration, thus obtaining the optimal model configuration that adapts to the market environment.

[0096] In this embodiment, the system first classifies recent market data based on the current market state classification label and uses clustering algorithms (such as K-Means and Gaussian Mixture Model 1) to group the data and extract multi-dimensional vector representations reflecting market characteristics. These features may include price volatility, policy sales trends, and the frequency of competing product launches. Subsequently, gradient descent optimization algorithms (such as Adam and RMSProp) are used to fine-tune the global model's parameter space, adjusting the weight coefficients of each feature to adapt to the current market state and generating a new parameter configuration. To quantify the configuration effect, the system defines a reward function that comprehensively considers prediction accuracy, stability, and overfitting risk to calculate a performance score. When the performance score falls below a preset threshold, the system triggers environmental change detection (based on sliding window statistics, market volatility index, and sentiment trend analysis), updates the market state label, re-acquires feature representations, and performs gradient optimization and evaluation again. This process can be iterated multiple times in a simulated backtesting environment until the model reaches its optimal performance configuration in the current market context, and finally, it is deployed to the real-time business system.

[0097] When the performance score is higher than or equal to a preset threshold, the system considers the current global model parameter configuration to meet or exceed the business's requirements for prediction accuracy and stability. Therefore, it marks this configuration as a candidate optimal configuration and triggers the model solidification and version management process. Specifically, the system stores this parameter configuration along with the corresponding market state feature representation in the model repository, recording the configuration generation time, market environment label, performance score, and related evaluation metrics. This allows for direct retrieval under similar market conditions in the future, reducing the overhead of retraining. Simultaneously, this optimal configuration is pushed to the online inference service node, replacing the existing running model for real-time business prediction and decision execution. To prevent model performance degradation over time, the system also sets up periodic review tasks to re-verify the applicability of the configuration when the market environment changes significantly or when a set period is reached.

[0098] By following the steps above, model parameters can be quickly and adaptively adjusted when the market environment changes, maintaining the stability of predictive performance and business decisions.

[0099] Furthermore, in response to customer claim requests, the process involves integrating customer profile data and current market status classification tags, using an optimally configured global model to determine the claim risk level and product recommendation matching degree, and obtaining personalized claim decision results and product recommendation solutions. This process specifically includes:

[0100] Receive claims requests submitted by customers and extract key information from the requests, including customer identification, claim type, and claim amount;

[0101] By integrating key information with customer profile data, claim application information that includes customers' historical behavior, preferences, and current status can be obtained;

[0102] Based on the current market status classification labels, select suitable claims risk assessment models and product recommendation models from the optimally configured global models;

[0103] Input the claim application information into the selected claim risk assessment model, determine the risk level, and output the claim risk level assessment result;

[0104] Input the claim application information into the selected product recommendation model, and combine it with customer profile data and current market status classification tags to generate an insurance product recommendation plan;

[0105] The final personalized claims decision and product recommendation are formed by combining the comprehensive claims risk level assessment results with the insurance product recommendation scheme.

[0106] In this embodiment, the system first receives a claim request submitted by the customer and extracts key information, including customer identification, claim type, and claim amount, by parsing structured fields and unstructured text information. For text-based materials (such as accident descriptions and medical record summaries), natural language processing technologies (NER, keyword extraction, and sentiment analysis) can be used for semantic parsing. Subsequently, this key information is integrated with customer profile data, which includes information such as historical insurance records, claim frequency, interaction behavior patterns, and risk preferences. This information is then combined with real-time collected market status classification tags to form complete claim application information. Next, based on the market status tags, the system automatically selects the most suitable claim risk assessment model and product recommendation model for the current environment from the optimally configured global model. The claim risk assessment model may be based on gradient boosting trees or deep neural networks, while the product recommendation model may employ collaborative filtering, knowledge graph recommendation, or deep matching networks. The claim risk assessment model analyzes the input information and outputs the customer's risk level (e.g., low, medium, high), along with the weights of the main risk factors. The product recommendation model combines the customer profile with market trends to generate personalized insurance product combination suggestions. Finally, the system integrates the risk level assessment results with the product recommendation plan to form a final claims decision and recommendation plan that can be directly executed by business personnel or automated systems, and can generate interpretable reports for compliance review and customer communication.

[0107] In this embodiment, the fusion process is mainly carried out through two methods: feature level and semantic level. First, key information in the claim request (such as claim type, claim amount, accident time, etc.) is mapped to numerical or categorical features consistent with the corresponding dimensions in the customer profile. For example, the claim amount is normalized into a standardized range, and the claim type is encoded into a one-hot vector. Then, these features are concatenated with feature vectors from the customer profile data, such as historical insurance records, claim frequency, interaction behavior patterns, and risk preferences, in a unified format to form an extended feature vector. Furthermore, combined with real-time market status classification labels, key features are assigned dynamic weights through weighted fusion or attention mechanisms, so that the fused claim information not only includes long-term customer behavior characteristics but also reflects the impact of the current market environment.

[0108] For example, a customer submits a car insurance claim for a rear-end collision, requesting a claim amount of 12,000 yuan. The system first integrates the claim information with the customer's profile data, which includes the customer's insurance records over the past three years, two claims, an average claim amount of 8,000 yuan, and a preference for low-risk policies. Simultaneously, it combines current market status classification tags (such as recent frequent traffic accidents and the launch of premium discount policies) to generate a fused feature vector. Subsequently, the claim risk assessment model determines that this claim is of medium risk, highlighting the main risk factors as accident frequency and vehicle age. The product recommendation model, based on the customer profile and market tags, generates recommended options for extended car insurance coverage and a fast-track accident claims process. The system integrates the risk level with the recommended options to form a final decision, automatically generating an interpretable report, such as "Due to the customer's relatively old vehicle and medium accident frequency, the claim risk assessment is medium; we recommend extended coverage and fast-track claims services," facilitating review by sales staff and communication with the customer.

[0109] Through the above steps, integrated and personalized processing of claims approval and product recommendation is achieved, improving decision-making accuracy and customer satisfaction.

[0110] In the above embodiments, this application discloses a method for intelligent customer service processing, belonging to the field of artificial intelligence technology, and applied to the optimization of financial business services. This application constructs a distributed data processing architecture, enabling each node to complete model training locally and upload gradient parameters, generating a global model through federated aggregation. A unified time benchmark is used to align the timestamp information and update frequency of each data source. Feature mapping and dimensionality reduction techniques are used to standardize heterogeneous data into a unified feature vector, generating customer profiles. Real-time collection of market policy changes and hot event information, combined with natural language processing and event impact assessment, generates market status classification labels. By dynamically adjusting model parameter weights and reward function evaluation, the optimal model configuration adapted to different market states is obtained. During customer claims processing, customer profiles and market status information are integrated to achieve intelligent decision-making for claims risk level assessment and personalized product recommendations, improving claims processing efficiency and customer satisfaction, while ensuring business interpretability and compliance, providing insurance companies with accurate, efficient, and intelligent customer service capabilities.

[0111] In this embodiment, the method for intelligent customer service processing runs on electronic devices (e.g., Figure 1 The server shown can receive commands or acquire data via wired or wireless connections. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G connections, Wi-Fi connections, Bluetooth connections, Wi-Fi connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future wireless connection methods.

[0112] It should be emphasized that, to further ensure the privacy and security of the aforementioned customer privacy information, this information can also be stored in a blockchain node.

[0113] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0114] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0115] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0116] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0117] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0118] Further reference Figure 4 As a response to the above Figure 2 To implement the method shown, this application provides an embodiment of a device for intelligent customer service processing, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0119] like Figure 4 As shown, the intelligent customer service processing device 400 described in this embodiment includes:

[0120] The local training module 401 is used to build a distributed data processing architecture. After each data source node completes model training locally, it uploads gradient parameters and obtains a global model through gradient parameter aggregation. The distributed data processing architecture includes several local models, with each data source node corresponding to a local model.

[0121] The unified benchmark module 402 is used to obtain a time-synchronized multidimensional data matrix by establishing a unified time benchmark based on the timestamp information and update frequency characteristics of each heterogeneous data source.

[0122] The customer profiling module 403 is used to convert customer insurance records, claims history and interaction data in different formats in the multidimensional data matrix into standardized feature vectors using feature mapping, and to generate unified customer profile data based on the standardized feature vectors.

[0123] The market tag module 404 is used to collect information on market policy changes and hot events in real time, and to determine the current market status classification tag.

[0124] The configuration optimization module 405 is used to dynamically adjust the global model parameter weights according to the classification labels of the current market status, evaluate the prediction accuracy under different parameter configurations through the reward function, and obtain the optimal model configuration that adapts to the current market environment.

[0125] The personalized recommendation module 406 is used to respond to customer claim requests. By integrating customer profile data and current market status classification tags, it uses an optimally configured global model to determine the claim risk level and product recommendation matching degree, and obtains personalized claim decision results and product recommendation solutions.

[0126] Furthermore, the local training module 401 specifically includes:

[0127] Distributed architecture unit, used to build a matching distributed data processing architecture based on the attributes of each data source node;

[0128] The local model training unit is used to obtain local data from various data source nodes, combine it with a preset machine learning algorithm to train the local model, and obtain the model gradient parameters.

[0129] The gradient parameter upload unit is used to upload the model gradient parameters from various data source nodes to the central server through an encrypted channel.

[0130] The gradient parameter aggregation unit is used to aggregate the model gradient parameters on the central server using a federated averaging algorithm to obtain the global model parameters.

[0131] The model parameter update unit is used to update the preset initial global model based on the global model parameters and generate the updated global model.

[0132] Furthermore, the unified benchmark module 402 specifically includes:

[0133] The timestamp analysis unit is used to analyze the timestamp information of various heterogeneous data sources to determine the data generation and update times.

[0134] The data partitioning unit is used to determine the data update frequency based on the data generation time and update time, and to divide the data into different update cycles according to the update frequency;

[0135] The timestamp conversion unit is used to select a common time base and convert the timestamps of various heterogeneous data sources to the common time base;

[0136] The data sorting unit is used to sort data from different data sources in chronological order according to the converted timestamps for each update cycle, and to build a time-synchronized multidimensional data matrix.

[0137] Further, please refer to Figure 5 The customer profiling module 403 specifically includes:

[0138] The data mapping unit 501 is used to map customer insurance records, claims history and interaction data of different formats in the multidimensional data matrix into numerical features through a preset feature mapping method, and generate an initial feature set.

[0139] The dimension reduction processing unit 502 is used to perform dimension reduction processing on the initial feature set using the principal component analysis algorithm to obtain a dimension-reduced feature set.

[0140] The standardization processing unit 503 is used to convert the dimensionality-reduced feature set into a standardized feature vector by using a preset vector generation rule and a linear transformation according to the dimensionality-reduced feature set, thereby generating a feature representation in a unified format and obtaining a standardized feature vector.

[0141] Behavior pattern unit 504 is used to group standardized feature vectors to determine customer behavior pattern categories;

[0142] The profile building unit 505 is used to integrate standardized feature vectors into unified customer profile data based on customer behavior pattern categories and profile data structures.

[0143] Furthermore, the market label module 404 specifically includes:

[0144] The policy hotspot acquisition unit is used to acquire data on market policy changes and hot events from preset information sources and store them as structured time series data;

[0145] The content parsing unit is used to parse the content of structured time series data using natural language processing technology, extract keywords and sentiment, and obtain event impact assessment results;

[0146] The category label generation unit is used to generate category labels for the current market status based on the event impact assessment results and in accordance with preset label generation rules.

[0147] Furthermore, the configuration optimization module 405 specifically includes:

[0148] The data classification unit is used to classify market data based on the current market status classification label, and to obtain a feature representation of the market status through clustering algorithms;

[0149] The parameter weight adjustment unit is used to adjust the parameter weights of the global model based on the feature representation using the gradient descent algorithm, so as to obtain the updated parameter configuration.

[0150] The performance score calculation unit is used to evaluate the updated parameter configuration through the reward function, calculate the prediction accuracy, and calculate the performance score of the current configuration.

[0151] The classification label adjustment unit is used to adjust the current market status classification label based on environmental change detection and obtain a new feature representation if the performance score is lower than a preset threshold.

[0152] The parameter weight iteration unit is used to iterate the parameter weights of the global model based on the new feature representation until the model reaches the optimal configuration, thus obtaining the optimal model configuration that adapts to the market environment.

[0153] Furthermore, the personalized recommendation module 406 specifically includes:

[0154] The request processing unit is used to receive claims requests submitted by customers and extract key information from the requests, including customer identification, claim type, and claim amount.

[0155] The data fusion unit is used to merge key information with customer profile data to obtain claim application information that includes the customer's historical behavior, preferences and current status;

[0156] The model matching unit is used to select the appropriate claims risk assessment model and product recommendation model from the optimally configured global model based on the classification labels of the current market status.

[0157] The risk level assessment unit is used to input claim application information into the selected claim risk assessment model, determine the risk level, and output the claim risk level assessment result;

[0158] The insurance product recommendation unit is used to input claim application information into the selected product recommendation model, and generate an insurance product recommendation scheme by combining customer profile data and current market status classification tags.

[0159] The personalized claims unit is used to combine the results of the claims risk level assessment with the insurance product recommendation plan to form the final personalized claims decision and product recommendation plan.

[0160] In the above embodiments, this application discloses an intelligent customer service processing device, belonging to the field of artificial intelligence technology, and applied to the optimization processing of financial business services. This application constructs a distributed data processing architecture, enabling each node to complete model training locally and upload gradient parameters, generating a global model through federated aggregation. A unified time benchmark aligns the timestamp information and update frequency of each data source. Using feature mapping and dimensionality reduction techniques, heterogeneous data is standardized into a unified feature vector to generate customer profiles. Real-time collection of market policy changes and hot event information, combined with natural language processing and event impact assessment, generates market status classification labels. By dynamically adjusting model parameter weights and reward function evaluation, the optimal model configuration adapted to different market states is obtained. During customer claims processing, the integration of customer profiles and market status information enables intelligent decision-making for claims risk level assessment and personalized product recommendations, improving claims processing efficiency and customer satisfaction, while ensuring business interpretability and compliance, providing insurance companies with accurate, efficient, and intelligent customer service capabilities.

[0161] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 6 , Figure 6 This is a basic structural block diagram of the computer device in this embodiment.

[0162] The computer device 6 includes a memory 61, a processor 62, and a network interface 63 that are interconnected via a system bus. It should be noted that only the computer device 6 with memory 61, processor 62, and network interface 63 is shown in the figure; however, it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0163] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0164] The memory 61 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 61 may be an internal storage unit of the computer device 6, such as the hard disk or memory of the computer device 6. In other embodiments, the memory 61 may also be an external storage device of the computer device 6, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 6. Of course, the memory 61 may also include both the internal storage unit and its external storage device of the computer device 6. In this embodiment, the memory 61 is typically used to store the operating system and various application software installed on the computer device 6, such as computer-readable instructions for intelligent customer service processing methods. In addition, the memory 61 can also be used to temporarily store various types of data that have been output or will be output.

[0165] In some embodiments, the processor 62 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 62 is typically used to control the overall operation of the computer device 6. In this embodiment, the processor 62 is used to execute computer-readable instructions stored in the memory 61 or to process data, such as computer-readable instructions for executing the intelligent customer service processing method.

[0166] The network interface 63 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 6 and other electronic devices.

[0167] This application also provides an embodiment, namely, a computer device including a memory and a processor. The memory stores computer-readable instructions, and the processor, when executing the computer-readable instructions, implements the steps of the above-described intelligent customer service processing method, that is, it implements:

[0168] A method for intelligent customer service processing includes:

[0169] A distributed data processing architecture is constructed, in which each data source node completes model training locally and uploads gradient parameters. The global model is obtained by aggregating the gradient parameters. The distributed data processing architecture includes several local models, with each data source node corresponding to a local model.

[0170] Based on the timestamp information and update frequency characteristics of various heterogeneous data sources, a time-synchronized multidimensional data matrix is ​​obtained by establishing a unified time benchmark.

[0171] Feature mapping is used to convert customer insurance records, claims history and interaction data in different formats in the multidimensional data matrix into standardized feature vectors, and unified customer profile data is generated based on the standardized feature vectors;

[0172] Real-time collection of market policy changes and hot topics information to determine the current market status classification labels;

[0173] Based on the current market status classification labels, the global model parameter weights are dynamically adjusted, and the prediction accuracy under different parameter configurations is evaluated through the reward function to obtain the optimal model configuration that adapts to the current market environment.

[0174] In response to customer claim requests, the system integrates customer profile data and current market status classification tags, uses an optimally configured global model to determine the claim risk level and product recommendation matching degree, and obtains personalized claim decision results and product recommendation solutions.

[0175] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the above-described intelligent customer service processing method, i.e., to implement:

[0176] A method for intelligent customer service processing includes:

[0177] A distributed data processing architecture is constructed, in which each data source node completes model training locally and uploads gradient parameters. The global model is obtained by aggregating the gradient parameters. The distributed data processing architecture includes several local models, with each data source node corresponding to a local model.

[0178] Based on the timestamp information and update frequency characteristics of various heterogeneous data sources, a time-synchronized multidimensional data matrix is ​​obtained by establishing a unified time benchmark.

[0179] Feature mapping is used to convert customer insurance records, claims history and interaction data in different formats in the multidimensional data matrix into standardized feature vectors, and unified customer profile data is generated based on the standardized feature vectors;

[0180] Real-time collection of market policy changes and hot topics information to determine the current market status classification labels;

[0181] Based on the current market status classification labels, the global model parameter weights are dynamically adjusted, and the prediction accuracy under different parameter configurations is evaluated through the reward function to obtain the optimal model configuration that adapts to the current market environment.

[0182] In response to customer claim requests, the system integrates customer profile data and current market status classification tags, uses an optimally configured global model to determine the claim risk level and product recommendation matching degree, and obtains personalized claim decision results and product recommendation solutions.

[0183] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0184] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0185] It should be noted that the software tools or components not belonging to this company that appear in the various embodiments of this application are merely illustrative examples and do not represent actual use.

[0186] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A method for intelligent customer service processing, characterized in that, include: A distributed data processing architecture is constructed, in which each data source node completes model training locally and uploads gradient parameters, and obtains a global model through the aggregation of the gradient parameters. The distributed data processing architecture includes several local models, and each data source node corresponds to a local model. Based on the timestamp information and update frequency characteristics of various heterogeneous data sources, a time-synchronized multidimensional data matrix is ​​obtained by establishing a unified time benchmark. Feature mapping is used to convert customer insurance records, claims history and interaction data in different formats in the multidimensional data matrix into standardized feature vectors, and unified customer profile data is generated based on the standardized feature vectors; Real-time collection of market policy changes and hot topics information to determine the current market status classification labels; Based on the current market state classification labels, the global model parameter weights are dynamically adjusted, and the prediction accuracy under different parameter configurations is evaluated through the reward function to obtain the optimal model configuration that adapts to the current market environment. In response to customer claim requests, the system integrates the customer profile data and the current market status classification tags, uses an optimally configured global model to determine the claim risk level and product recommendation matching degree, and obtains personalized claim decision results and product recommendation solutions.

2. The method for intelligent customer service processing as described in claim 1, characterized in that, The step of constructing a distributed data processing architecture, in which each data source node uploads gradient parameters after completing model training locally, and obtains a global model through aggregation of the gradient parameters, specifically includes: Build a matching distributed data processing architecture based on the attributes of each data source node; Local data is obtained from various data source nodes, and local model training is performed using a preset machine learning algorithm to obtain model gradient parameters. The model gradient parameters are uploaded from each data source node to the central server via an encrypted channel; The model gradient parameters are aggregated using a federated averaging algorithm on a central server to obtain global model parameters; The preset initial global model is updated based on the global model parameters, and the updated global model is generated.

3. The method for intelligent customer service processing as described in claim 1, characterized in that, The step of obtaining a time-synchronized multidimensional data matrix by establishing a unified time base based on the timestamp information and update frequency characteristics of various heterogeneous data sources specifically includes: Analyze the timestamp information of various heterogeneous data sources to determine the data generation and update times; Based on the data generation time and update time, the data update frequency is determined, and the data is divided into different update cycles according to the update frequency. Select a common time base and convert the timestamps of various heterogeneous data sources to the common time base; For each update cycle, data from different data sources are arranged in chronological order based on the converted timestamps to construct a time-synchronized multidimensional data matrix.

4. The method for intelligent customer service processing as described in claim 1, characterized in that, The step of using feature mapping to convert customer insurance records, claims history, and interaction data in different formats from the multidimensional data matrix into standardized feature vectors, and generating unified customer profile data based on the standardized feature vectors, specifically includes: The customer insurance records, claims history and interaction data of different formats in the multidimensional data matrix are mapped into numerical features by a preset feature mapping method to generate an initial feature set. The initial feature set is reduced in dimensionality using principal component analysis to obtain a dimensionality-reduced feature set. Based on the dimensionality reduction feature set, a preset vector generation rule is used to transform the dimensionality reduction feature set into a standardized feature vector through linear transformation, generating a feature representation in a unified format, and thus obtaining the standardized feature vector. The standardized feature vectors are grouped to determine customer behavior pattern categories; Based on the customer behavior pattern categories, a profile data structure is used to integrate the standardized feature vectors into unified customer profile data.

5. The method for intelligent customer service processing as described in claim 1, characterized in that, The steps for collecting real-time information on market policy changes and trending events, and determining the current market status classification label, specifically include: Data on market policy changes and hot topics are obtained from preset information sources and stored as structured time series data; Natural language processing techniques are used to parse the content of the structured time series data, extract keywords and sentiment, and obtain the event impact assessment results; Based on the event impact assessment results and in conjunction with preset tag generation rules, the current market status classification tag is generated.

6. The method for intelligent customer service processing as described in claim 5, characterized in that, The step of dynamically adjusting the global model parameter weights based on the current market state classification labels, evaluating the prediction accuracy under different parameter configurations through a reward function, and obtaining the optimal model configuration adapted to the current market environment specifically includes: Based on the current market status classification labels, market data is classified using a clustering algorithm to obtain a feature representation of the market status; Based on the feature representation, the parameter weights of the global model are adjusted using the gradient descent algorithm to obtain the updated parameter configuration; The updated parameter configuration is evaluated using the reward function, the prediction accuracy is calculated, and the performance score of the current configuration is calculated. If the performance score is lower than a preset threshold, the current market status classification label is adjusted based on environmental change detection, and a new feature representation is obtained; Based on the new feature representation, the parameter weights of the global model are iterated again until the model reaches the optimal configuration, thus obtaining the optimal model configuration adapted to the market environment.

7. The method for intelligent customer service processing as described in claim 1, characterized in that, The steps of responding to a customer's claim request, by integrating the customer profile data and the current market status classification tags, using an optimally configured global model to determine the claim risk level and product recommendation matching degree, and obtaining a personalized claim decision result and product recommendation plan, specifically include: Receive a claim request submitted by a customer and extract key information from the request, including customer identification, claim type, and claim amount; By integrating the key information with the customer profile data, claim application information containing the customer's historical behavior, preferences, and current status is obtained; Based on the current market status classification labels, select the appropriate claims risk assessment model and product recommendation model from the optimally configured global model; The claim application information is input into the selected claim risk assessment model to determine the risk level and output the claim risk level assessment result. The claim application information is input into the selected product recommendation model, and combined with the customer profile data and the current market status classification tags, an insurance product recommendation scheme is generated; Based on the combined results of the claims risk assessment and the recommended insurance products, a final personalized claims decision and product recommendation are formed.

8. A device for intelligent customer service processing, characterized in that, include: The local training module is used to build a distributed data processing architecture. After each data source node completes model training locally, it uploads gradient parameters and obtains a global model by aggregating the gradient parameters. The distributed data processing architecture includes several local models, and each data source node corresponds to a local model. The unified benchmark module is used to obtain a time-synchronized multidimensional data matrix by establishing a unified time benchmark based on the timestamp information and update frequency characteristics of various heterogeneous data sources. The customer profiling module is used to convert customer insurance records, claims history and interaction data in different formats in the multidimensional data matrix into standardized feature vectors using feature mapping, and to generate unified customer profile data based on the standardized feature vectors. The market tagging module is used to collect information on market policy changes and hot topics in real time, and to determine the current market status classification tags. The configuration optimization module is used to dynamically adjust the global model parameter weights according to the current market state classification label, evaluate the prediction accuracy under different parameter configurations through the reward function, and obtain the optimal model configuration that adapts to the current market environment. The personalized recommendation module is used to respond to customer claim requests. By integrating the customer profile data and the current market status classification tags, it uses an optimally configured global model to determine the claim risk level and product recommendation matching degree, and obtains personalized claim decision results and product recommendation solutions.

9. A computer device, characterized in that, The system includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the method for intelligent customer service processing as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the intelligent customer service processing method as described in any one of claims 1 to 7.

Images